Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Association of Wealth With Longevity in US Adults at Midlife

Association of Wealth With Longevity in US Adults at Midlife Key Points Question Is net worth at midlife IMPORTANCE Wealthy adults tend to live longer than those with less wealth. However, a challenge associated with all-cause mortality? in this area of research has been the reduction of potential confounding by factors associated with Findings In this cohort study of 5414 the early environment and heritable traits, which could simultaneously affect socioeconomic participants in the Midlife in the United circumstances in adulthood and health across the life course. States study, those who had accumulated a higher net worth by OBJECTIVE To identify the association between net worth at midlife and subsequent all-cause midlife had significantly lower mortality mortality in individuals as well as within siblings and twin pairs. risk over the subsequent 24 years. In sibling and twin comparison models that DESIGN, SETTING, AND PARTICIPANTS This cohort study conducted a series of analyses using controlled for shared early life data from the Midlife in the United States (MIDUS) study, an ongoing national study of health and experiences and genetic influence, the aging. The sample included adults (unrelated individuals, full siblings, and dizygotic and monozygotic association between net worth and twins) aged 20 to 75 years, who participated in wave 1 of the MIDUS study, which occurred from 1994 longevity was similar in magnitude. to 1996. The analyses were conducted between November 16, 2019, and May 18, 2021. Meaning Net worth at midlife was EXPOSURES Self-reported net worth (total financial assets minus liabilities) at midlife (the middle associated with longevity among adults years of life). in the study, and this association is unlikely to be merely an artifact of early MAIN OUTCOMES AND MEASURES All-cause mortality was tracked over nearly 24 years of experiences or heritable traits shared follow-up, with a censor date of October 31, 2018. Survival models tested the association between by families. net worth and all-cause mortality. Discordant sibling and twin analyses compared longevity within siblings and twin pairs who, given their shared early experiences and genetic backgrounds, were Invited Commentary matched on these factors. Supplemental content RESULTS The full sample comprised 5414 participants, who had a mean (SD) age of 46.7 (12.7) years Author affiliations and article information are and included 2766 women (51.1%). Higher net worth was associated with lower mortality risk (hazard listed at the end of this article. ratio [HR], 0.95; 95% CI, 0.94-0.97; P < .001). Among siblings and twin pairs specifically (n = 2490), a similar within-family association was observed between higher net worth and lower mortality (HR, 0.94; 95% CI, 0.91-0.97; P = .001), suggesting that the sibling or twin with more wealth tended to live longer than their co-sibling or co-twin with less wealth. When separate estimates were performed for the subsamples of siblings (HR, 0.94; 95% CI, 0.90-0.97; P = .002), dizygotic twins (HR, 0.94; 95% CI, 0.86-1.02; P = .19), and monozygotic twins (HR, 0.95; 95% CI, 0.87-1.04; P = .34), the within-family estimates of the net worth–mortality association were similar, although the precision of estimates was reduced among twins. CONCLUSIONS AND RELEVANCE This cohort study found that wealth accumulation at midlife was associated with longevity in US adults. Discordant sibling analyses suggested that this association is unlikely to be simply an artifact of early experiences or heritable characteristics shared by families. JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 Open Access. This is an open access article distributed under the terms of the CC-BY License. JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 (Reprinted) July 23, 2021 1/12 JAMA Health Forum | Original Investigation Association of Wealth With Longevity in US Adults at Midlife Introduction 1-3 Socioeconomic disparities in life expectancy are substantial in size. Financial wealth or net worth, which is the value of an individual’s assets (such as savings, real estate, and vehicles) minus 4,5 4,6-9 liabilities, is directly associated with longevity and, in some studies, has been found to be more strongly associated with mortality than other indicators of socioeconomic status, such as 7 6 occupational prestige, educational attainment, and income. However, a challenge in this area of research has been eliminating or minimizing the potential for confounding by the early environment and heritable traits, either of which could simultaneously affect socioeconomic conditions in adulthood and health in the course of life. Discordant sibling designs allow for the identification and control of such confounders. Full siblings who were raised in the same family share much of their early rearing environment and are genetically related to one another. Thus, in sibling-comparison studies, factors that are shared 11,12 between siblings are controlled. Twin comparisons provide an even greater control of family-level early-life confounding and, in the case of monozygotic (MZ) twins, control for all heritable genetic 13,14 factors. Previous research found that discordance in occupational prestige was associated with 15 16 cardiovascular risk and overall mortality ; twins with lower-prestige jobs had worse health on both outcomes compared with their co-twins with higher-prestige jobs. This pattern suggests that socioeconomic disparities in health are affected by experiential factors in adulthood over and above any potential confounders that involve the siblings’ shared early environment and genetic 16-22 characteristics. In other discordant sibling and twin analyses, educational attainment and composite measures of adult socioeconomic position also have been associated with better adult 18,22,23 16,17,19-21 health outcomes and longevity. However, results from these and other studies that used different methods do suggest these associations may be partially explained by shared family- 17-19,21,22 10,18,23 level environmental factors or genetic predispositions. Comparatively little attention has been given to wealth disparities, a potentially important oversight because wealth inequality is far greater and growing at a faster rate than income inequality 24,25 in the United States. At the individual level, those with greater wealth are better able to access health-promoting resources (eg, medical care, safe places to exercise, and fresh foods). Wealthy individuals also have more protection from economic shocks, such as job loss, unexpected health care expenses, or other financial crises. The association between wealth and longevity has become more relevant in the past year because of the economic fallout from the COVID-19 pandemic, which 26 27 has disproportionately affected the financial security of low-income and older workers. In this cohort study, we used a discordant sibling design to conservatively estimate the association between wealth and longevity. Specifically, we aimed to identify the association between net worth at midlife (the middle years of life) and subsequent all-cause mortality in individuals as well as within siblings and twin pairs. We posed 2 research questions. First, was wealth accumulation at midlife associated with longevity over a nearly 24-year follow-up? Consistent with previous 4,6-9,28 work, we expected that higher wealth accumulation would be associated with increased longevity. Second, was the wealth-longevity association present over and above controls for family and heritable factors that could confound this association? That is, was there evidence of a within- family association between wealth and longevity among siblings and twins in the same family? Alternatively, was the wealth-longevity association primarily driven by common factors at the family level, presumably involving early experiences and/or heritable factors? Methods The analyses in this cohort study were conducted between November 16, 2019, and May 18, 2021, and were preregistered on the Open Science Framework on November 15, 2019. The preregistration included hypotheses, decision rules for inclusion, rationale for covariates, and statistical plan. Deviations from the original analysis plan are detailed in eMethods 3 in the JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 (Reprinted) July 23, 2021 2/12 JAMA Health Forum | Original Investigation Association of Wealth With Longevity in US Adults at Midlife Supplement. We followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline. The institutional review boards of the University of Wisconsin-Madison and Harvard Medical School approved the Midlife in the United States (MIDUS) study procedures. Participants in the MIDUS study provided oral informed consent. The institutional review board approval and informed consent for the MIDUS study extend to the present study, which used publicly available MIDUS data, including all mortality data up to 2016. Participants and Analysis Sample The data used were obtained from the MIDUS study, an ongoing national study of health and aging that began in 1994. In the present study, we used data from wave 1 of the MIDUS study (MIDUS 1), which were collected in 1994 to 1996 from 7108 adults aged 20 to 75 years. These participants were recruited through a nationally representative random-digit dialing sampling strategy that included subsamples of siblings and twins. Participants who provided oral informed consent completed a telephone-assisted survey and a mailed self-administered questionnaire. Mortality data through October 31, 2018, were collected by the MIDUS study team at the University of Wisconsin-Madison. Of the 7108 cases from MIDUS 1, 7017 were considered for inclusion in the current analysis. We selected cases with complete data on mortality status, net worth, and relevant covariates (n = 5414). These cases included 2675 individuals (designated here as singletons) who did not have a sibling or co-twin in the MIDUS study, 1282 nontwin full siblings, 864 dizygotic (DZ) twins, and 593 MZ twins. Twins or siblings who had complete data and who were in a pair with a co-sibling or co-twin who were missing data were included in the full analytic sample. However, they were excluded from subsequent discordant twin or sibling analyses, resulting in 1214 nontwin full siblings, 740 DZ twins, and 536 MZ twins in the discordant twin or sibling analyses. The eMethods 1 in the Supplement provides details on the eligibility criteria and missing data. Measures At MIDUS 1, participants responded to the following question: “Suppose you (and your spouse or partner) cashed in all your checking and savings accounts, stocks and bonds, real estate, sold your home, your vehicles, and all your valuable possessions. Then suppose you put that money toward paying off your mortgage and all your other loans, debts, and credit cards. Would you have any money left over after paying your debts or would you still owe money?” Participants reported how much that amount would be, using binned response categories that specified ranges of dollar amounts. Amount of money owed was truncated at a value of $0. Net worth that exceeded $1 million was truncated at that value. Covariates included parental educational level and participant age, self-reported race/ethnicity (analyzed here as non-White vs White), sex (female vs male), history of cancer or heart disease as diagnosed by a medical doctor, and status of ever smoking cigarettes regularly or consuming alcohol regularly (the eMethods 1 in the Supplement provides coding details). At MIDUS 1, participants self- reported on their race. Response options included White, Black and/or African American, Native American or Aleutian Islander, Asian or Pacific Islander, multiracial, or other race. Because of the low number of individuals who identified as a race other than White, we dichotomized race categories into non-White vs White. In addition, data on self-identification as Hispanic or Latino were not available at MIDUS 1. Mortality follow-up was completed by the MIDUS study team at the University of Wisconsin- Madison. Date of death was obtained from various sources, including relative responses, other informant reports, newspaper or online obituaries, and the National Death Index (using the 15th day of the month, rather than the exact day of death, to maintain participant confidentiality). Survival time was the number of years between the date when the MIDUS 1 self-administered questionnaires were returned to the MIDUS study team (1994 to 1996) and the date of death; if the participant was alive, the censor date was October 31, 2018, which was the date of the MIDUS study team’s latest mortality update. JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 (Reprinted) July 23, 2021 3/12 JAMA Health Forum | Original Investigation Association of Wealth With Longevity in US Adults at Midlife Statistical Analysis A series of Cox proportional hazards regression models for survival analyses were estimated using Stata, version 16 (StataCorp LLC). In model 1, we tested the individual-level association between wealth and longevity by conducting survival analysis among all cases who had complete data on the analysis variables and by using robust SEs to account for dependence among family members. Next, we estimated survival models only within the subsample of twins and nontwin siblings. In model 2, siblings and twins were treated as individuals, but a shared frailty was included to estimate dependence among family members. In model 3, we estimated the within-family association between wealth and longevity by calculating the family-level mean net worth (among families that had 2 members with complete data) and subsequently calculating the difference between each individual’s net worth and their family’s mean net worth. This between-within method is a common 33,34 approach to fixed-effects modeling. When applied in this way, the between-within method allowed us to compare siblings or twins in the same family to one another and thus to control for all 13,14,35 unmeasured shared family-level variables consistent with the discordant sibling design. In survival analysis, the between-within method has been shown to provide similar estimates to more common approaches for co-sibling or co-twin control (eg, conditional likelihood methods like stratified Cox regression) and has been observed to be optimal statistically. To further disambiguate environmental vs genetic influences, we tested whether within-family associations between net worth and longevity varied across the nontwin sibling, MZ twin, and DZ twin subsamples. We included a pair of 2-way interaction terms crossing the net worth mean deviation scores with dummy codes for DZ twins or MZ twins. A Wald test was used to measure the equality of the within-family net worth coefficient across siblings, DZ twins, and MZ twins. We estimated separate survival models for nontwin siblings, DZ twins, and MZ twins (models 4-6). A significant within-family association (2-sided P < .05) that was observed among nontwin siblings but not among DZ or MZ twin pairs would suggest residual confounding by early life factors because twins share a closer prenatal and postnatal environment than nontwin siblings. A within-family association that was observed among both siblings and DZ twin pairs but not among MZ twin pairs would suggest genetic confounding. Sensitivity analyses addressed the skewed distribution of net worth and tested the possibility of nonlinear associations between net worth and longevity. Sensitivity analyses also clarified the role of preexisting health problems and considered other model specifications. Primary models were reestimated as stratified Cox regressions using Stata (eMethods 2 in the Supplement) and as multilevel Cox regressions using Mplus (Muthén & Muthén) (eTables 3-5 in the Supplement). A 2-sided P < .05 was considered statistically significant. Results A total of 5414 participants from MIDUS 1 were included in the full analysis sample. These participants had a mean (SD) age of 46.7 (12.7) years; 2766 were women (51.1%), 2648 were men (48.9%), and 4927 (91.0%) self-identified as White individuals. Participants had a mean (SD) net worth of $122 153 ($209 537; median, $32 500). Of these participants, 675 (12.5%) had been diagnosed with a heart problem and 381 (7%) had a cancer diagnosis. In addition, 2790 participants (51.5%) reported ever having smoked cigarettes regularly and 2301 participants (42.5%) reported having used alcohol regularly. Table 1 displays the descriptive statistics, by subsamples of singletons, nontwin siblings, DZ twins, and MZ twins. By the censor date (October 31, 2018), 1010 individuals (18.7%) in the full sample had died. Fifty percent of siblings and twins were in families whose members differed by less than or equal to $87 500 in net worth at MIDUS 1 (interquartile range=$212 500). JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 (Reprinted) July 23, 2021 4/12 JAMA Health Forum | Original Investigation Association of Wealth With Longevity in US Adults at Midlife Cohort Analyses Survival models were run as gamma shared frailty Cox regressions, stratified Cox regressions, and multilevel Cox regressions. Regardless of the model used, the estimates were similar (eMethods 2 and 3 in the Supplement). Thus, results from the shared frailty Cox regressions are described herein. Results from the full sample (model 1; n = 5414; 1010 deaths) are presented in Table 2. After covariate adjustment, net worth was inversely associated with mortality (HR, 0.95; 95% CI, 0.94-0.97; P < .001) such that the hazard was 5% lower for every additional $50 000 of net worth accumulated at midlife. In the subsample of siblings and twins (model 2; n = 2490; 421 deaths), the net worth HR was similar (HR, 0.95; 95% CI, 0.93-0.97; P < .001). Between-Within Models Similar to models 1 and 2, the between-within model among siblings and twins (model 3) suggested that net worth was inversely associated with mortality within families (HR, 0.94; 95% CI, 0.91-0.97; P = .001). These patterns suggested that a sibling or twin who had accumulated more net worth by midlife tended to live longer than their co-siblings or co-twin with less net worth. Figure 1 depicts survival curves that were split at values above and below 0.5 of an SD from the mean deviation net worth score (mean of 0.00). In Figure 1, the net worth–mortality association is conditional on the shared frailty, thus the lines represent the estimated survival of 2 family members whose net worth differed by 0.5 SD or $139 000 and who were approximately 47 years of age at MIDUS 1 (with all other variables held at their means). A difference of $139 000 in net worth was associated with a 13% relative decrease in the probability of death nearly 24 years later, favoring the family member with a higher net worth; given the low base rate of mortality in the sample, this decrease translated into more than a 1% absolute difference in survival. The interactions between sibling type (ie, nontwin siblings, DZ twins, and MZ twins) and net worth deviation score were not significant, and the Wald Table 1. Descriptive Statistics of the Analysis Variables No. (%) Nontwin Full sample Singleton full sibling DZ twins MZ twins Variable (n = 5414) (n = 2675) (n = 1214) (n = 740) (n = 536) Net worth, mean (SD), $ 122 153.02 111 425.23 160 577.84 97 760.81 131 624.06 (209 537.49) (202 341.67) (240 643.26) (167 631.37) (217 415.75) Age, mean (SD), y 46.7 (12.7) 46.2 (13.0) 49.3 (12.4) 45.9 (12.1) 44.7 (11.7) Sex Female 2766 (51.1) 1267 (47.4) 667 (54.9) 409 (55.3) 282 (52.6) Male 2648 (48.9) 1408 (52.6) 547 (45.1) 331 (44.7) 254 (47.4) Race White 4927 (91) 2334 (87.3) 1161 (95.6) 706 (95.4) 503 (93.8) Black or African 255 (4.7) 173 (6.5) 23 (1.9) 23 (3.1) 20 (3.7) American Other 232 (4.3) 168 (6.3) 30 (2.5) 11 (1.5) 13 (2.4) Heart disease 675 (12.5) 315 (11.8) 172 (14.2) 104 (14.1) 58 (10.8) Cancer 381 (7.0) 165 (6.2) 126 (10.4) 43 (5.8) 26 (4.9) Regular cigarette smoking 2790 (51.5) 1464 (54.7) 587 (48.4) 362 (48.9) 242 (45.1) Regular alcohol use 2301 (42.5) 1204 (45.0) 508 (41.8) 301 (40.7) 186 (34.7) Parental educational level <High school diploma 1275 (23.6) 744 (27.8) 182 (15.0) 184 (24.9) 106 (19.8) High school diploma 1937 (35.8) 983 (36.7) 433 (35.7) 252 (34.1) 176 (32.8) Some college, 2-y 873 (16.1) 368 (13.8) 229 (18.9) 110 (14.9) 118 (22.0) associate’s degree, or vocational school 4-y Bachelor’s degree 822 (15.2) 371 (13.9) 215 (17.7) 126 (17.0) 80 (14.9) Abbreviations: DZ, dizygotic; MZ, monozygotic. or some graduate school Master’s or professional 507 (9.4) 209 (7.8) 155 (12.8) 68 (9.2) 56 (10.4) Other race included multiracial, Native American or degree Aleutian Islander, Asian or Pacific Islander, and Other. JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 (Reprinted) July 23, 2021 5/12 JAMA Health Forum | Original Investigation Association of Wealth With Longevity in US Adults at Midlife Table 2. Cox Proportional Hazards Regression for Survival Models a b Model HR (95% CI) P value Model 1: full sample (n = 5414) Age 1.10 (1.09-1.11) <.001 Female sex 0.82 (0.72-0.94) .005 Non-White race 1.13 (0.87-1.46) .35 Parental education 0.97 (0.95-0.99) .03 Heart disease 1.87 (1.61-2.17) <.001 Cancer 1.44 (1.22-1.70) <.001 Cigarette smoking 1.75 (1.53-2.01) <.001 Alcohol use 1.04 (0.91-1.19) .47 Net worth 0.95 (0.94-0.97) <.001 Model 2: siblings and twins (n = 2490) Age 1.10 (1.09-1.12) <.001 Female sex 0.76 (0.61-0.95) .01 Non-White race 0.85 (0.48-1.52) .60 Parental education 0.97 (0.93-1.01) .17 Heart disease 1.93 (1.54-2.43) <.001 Cancer 1.55 (1.17-2.06) .002 Cigarette smoking 2.08 (1.66-2.61) <.001 Alcohol use 0.92 (0.73-1.15) .47 Net worth 0.95 (0.93-0.97) <.001 Model 3: siblings and twins (n = 2490) Age 1.10 (1.09-1.11) <.001 Female sex 0.76 (0.61-0.95) .01 Non-White race 0.87 (0.49-1.54) .64 Parental education 0.97 (0.93-1.01) .15 Heart disease 1.94 (1.54-2.43) <.001 Cancer 1.56 (1.18-2.07) .002 Cigarette smoking 2.09 (1.66-2.62) <.001 Alcohol use 0.91 (0.73-1.14) .43 Net worth Between family 0.96 (0.93-0.99) .01 Within family 0.94 (0.91-0.97) .001 Model 4: nontwin siblings (n = 1214) Age 1.10 (1.08-1.12) <.001 Female sex 0.64 (0.47-0.86) .004 Non-White race 1.51 (0.72-3.14) .27 Parental education 1.00 (0.94-1.06) .88 Heart disease 2.09 (1.52-2.86) <.001 Cancer 1.92 (1.35-2.72) <.001 Cigarette smoking 2.21 (1.61-3.04) <.001 Alcohol use 0.73 (0.53-1.00) .05 Net worth Between family 0.98 (0.94-1.02) .38 Within family 0.94 (0.90-0.97) .002 Model 5: DZ twins (n = 740) Age 1.12 (1.09-1.14) <.001 Female sex 0.88 (0.59-1.31) .53 Non-White race 0.25 (0.06-1.02) .05 Parental education 0.96 (0.90-1.03) .37 Heart disease 1.72 (1.14-2.57) .009 Cancer 1.01 (0.57-1.79) .96 (continued) JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 (Reprinted) July 23, 2021 6/12 JAMA Health Forum | Original Investigation Association of Wealth With Longevity in US Adults at Midlife Table 2. Cox Proportional Hazards Regression for Survival Models (continued) a b Model HR (95% CI) P value Cigarette smoking 2.08 (1.38-3.14) <.001 Alcohol use 1.18 (0.80-1.75) .39 Net worth Between family 0.90 (0.84-0.97) .009 Within family 0.94 (0.86-1.02) .19 Model 6: MZ twins (n = 536) Age 1.11 (1.08-1.14) <.001 Female sex 1.02 (0.60-1.73) .93 Non-White race 1.17 (0.33-4.14) .79 Abbreviations: DZ, dizygotic; HR, hazard ratio; Parental education 0.91 (0.83-1.01) .10 MZ, monozygotic. Heart disease 2.37 (1.27-4.42) .006 The models are described in the Statistical Analysis Cancer 1.53 (0.61-3.81) .35 subsection of the Methods section. Cigarette smoking 2.05 (1.19-3.54) .009 The HR for net worth of both between-family and within-family associations reflects a difference of Alcohol use 1.01 (0.59-1.71) .97 $50 000. Net worth Non-White race included Black and/or African Between family 0.96 (0.89-1.02) .25 American, Native American or Aleutian Islander, Within family 0.95 (0.87-1.04) .34 Asian or Pacific Islander, multiracial, and other. Figure 1. Within-Family Association Between Net Worth and Longevity 1.00 0.98 +0.5 SD net worth deviation 0.96 The survival curves represent 2 family members whose net worth differed by approximately $139 000 at –0.5 SD net worth deviation Midlife in the United States (MIDUS) study wave 1; this 0.94 amount corresponded to within-family net worth deviation scores that are ±0.5 SD from the mean 0.92 deviation score of 0.00. The survival curves were adjusted for family-level mean net worth as well as participant age at wave 1 of the MIDUS study, race/ 0.90 0 5 10 15 20 25 ethnicity, sex, history of cancer or heart disease, health Time after MIDUS study wave 1, y behaviors, and parental educational level. test indicated that differences in the within-family HRs across sibling types were no greater than chance (χ (2) = 0.05; P =.97). Separate survival models were estimated for nontwin siblings (model 4), DZ twins (model 5), and MZ twins (model 6). Consistent with the Wald test, the within-family net worth estimates for each of the 3 subsamples were similar (nontwin siblings: HR, 0.94 [95% CI, 0.90-0.97; P = .002]; DZ twins: HR, 0.94 [95% CI, 0.86-1.02; P = .19]; MZ twins: HR, 0.95 [95% CI, 0.87-1.04; P = .34]). These HRs were also similar in magnitude to the net worth HR observed in the individual-level analyses (models 1 and 2). Among nontwin siblings, the P value for the within-family net worth estimate was significant at P < .05, although the P values were not significant in the DZ twin and MZ twin subsamples, likely because of the substantial decrease in the sample size and the number of deaths within the DZ twin (n = 118 decedents) and MZ twin (n = 79 decedents) subsamples. Figure 2 depicts the HRs and 95% CIs for the net worth estimate in the full sample (model 1) and sibling and twin subsample (model 2), the within-family comparison in the sibling and twin subsample (model 3), and the within-sibling and within-twin comparisons (models 4, 5, and 6). JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 (Reprinted) July 23, 2021 7/12 Survival JAMA Health Forum | Original Investigation Association of Wealth With Longevity in US Adults at Midlife Sensitivity Analyses The eMethods 2 and eTables 1-5 in the Supplement present the results of the sensitivity analyses conducted in the combined sibling and twin subsample. Briefly, these analyses found that the observed wealth-longevity association (1) was not an artifact of the right-skewed distribution of net worth, (2) had a roughly linear shape for most of the sample, and (3) was not explained by lifestyle practices and health conditions at MIDUS 1. For example, when net worth was recoded into deciles, the within-family association between net worth and mortality remained statistically significant (HR, 0.92; 95% CI, 0.87-0.96; P = .001). In addition, spline models (eTable 1 in the Supplement) suggested that the HR for the net worth–mortality association was similar for approximately 90% of participants in the full sample. Moreover, when restricting the full sample to only the siblings and twin pairs who were free of previous cancer or heart disease, the within-family association between net worth and mortality remained statistically significant (HR, 0.94; 95% CI, 0.90-0.98; P = .01). The eMethods 2 in the Supplement provides a complete description of all sensitivity analyses. Discussion In contrast to most previous studies that examined associations between socioeconomic position and longevity using between-person analytic approaches, the present cohort study examined within-family associations between wealth and longevity— a more conservative test of the hypothesis because it implicitly controlled for all factors (eg, early experience and heritable characteristics) shared by siblings and twins. Findings from the within-family analyses suggested that wealth accumulation at midlife may be associated with longevity among adults in the United States. We believe that investigations such as the kind we conducted are important because of the near impossibility of performing an experimental study of the wealth-longevity association. Findings from this study also converged with the results from other studies that used different methods. For example, an analysis found that mortality risk among retired stockholders increased in the years after wealth shocks (ie, acute reductions in wealth) owing to exogenous stock market fluctuations. Across models, the associations observed between net worth and longevity were modest: we observed a 1% absolute difference in the probability of survival after nearly 24 years between family members who differed by approximately $139 000 in net worth at midlife. However, the MIDUS sample was relatively young (mean age of approximately 70 years) with relatively low mortality (18.7% of the full sample had died by the censor date). Thus, follow-up analyses are needed to ascertain whether the magnitude of these associations would change as mortality increases. The associations between net worth and longevity in the DZ and MZ twin pairs were similar in magnitude to the associations observed in nontwin siblings. However, when the analyses were separated into subsamples of siblings, DZ twins, and MZ twins, the within-family net worth associations among DZ twins and MZ twins were not statistically significant. This finding was likely the result of lower power and reduced precision in the twin subsamples, as reflected in the wide CIs around the within-family estimates. However, it could also suggest the presence of confounding by shared environmental features. Other studies have discussed the difficulty in distinguishing null Figure 2. Hazard Ratios (HRs) and 95% CIs for the Net Worth Estimate Across Survival Models Model No. and description HR (95% CI) P value 1. Full sample 0.95 (0.94-0.97) <.001 2. Siblings and twins 0.95 (0.93-0.97) <.001 3. Siblings and twins (within family comparison) 0.94 (0.91-0.97) .001 4. Nontwin siblings (within-sibling comparison) 0.94 (0.90-0.97) .002 In each model, the HR reflects the decrease in hazard 5. DZ twins (within twin comparison) 0.94 (0.85-1.02) .19 associated with a $50 000 increase in net worth. The 6. MZ twins (within-twin comparison) 0.95 (0.87-1.04) .34 squares represent the HR estimates, and the lines represent the 95% CIs. DZ indicates dizygotic; 0.85 0.90 0.95 1.00 1.05 1.10 HR (95% CI) MZ, monozygotic. JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 (Reprinted) July 23, 2021 8/12 JAMA Health Forum | Original Investigation Association of Wealth With Longevity in US Adults at Midlife associations from false-negative results in sibling study designs when power varies widely between subsamples. The findings of this study suggest that the association between wealth at midlife and longevity is unlikely to be a simple artifact of environmental and heritable characteristics shared by siblings and twins. Moreover, the findings suggest that policies to support individuals’ ability to accrue wealth and to achieve financial security in adulthood could have considerable health benefits. These findings should be interpreted through a broader societal lens. The US ranks first in economic inequality among high-income nations. Over the past 30 years, the wealth gap between the high-income and low-income people in the US has widened through policies and practices that have diverted a substantial and increasing share of wealth from the lower- and middle-income groups to the affluent 24,38 group. Such redistribution may have implications for longevity patterns in the coming decades. Policies to reduce the wealth gap, if implemented, could be expected to generate substantial returns to public health. Limitations This study has several limitations. First, although the discordant sibling design, as we have modeled it, reduced confounding by shared environmental and heritable characteristics, it cannot elucidate whether wealth itself is a causal actor or a marker of other nonshared factors (eg, self-regulatory capacity and cognitive ability) and nonshared experiences (eg, social, educational, and professional trajectories) that covary with and/or contribute to both wealth and health. Thus, a causal interpretation of these findings is not warranted. We did consider some possible nonshared confounders in adulthood that were related to lifestyle and disease, however. Sensitivity analyses restricted to participants in good midlife health suggested that these variables had a modest role in the wealth-longevity association. Second, at MIDUS 1, a single self-reported questionnaire item was used to measure net worth, which may have introduced error into estimates of net worth. In addition, most participants in the MIDUS 1 sample self-identified as White individuals; thus, the estimates we reported may be less generalizable to underrepresented racial/ethnic groups. This limitation is particularly relevant for present purposes given the substantial and persistent racial disparities in household wealth in the US. Third, although we observed evidence of within-family associations between wealth and longevity, the mechanisms involved were not identified. Wealth is likely to be a distal factor in longevity, acting through more proximal mechanisms (eg, stress and lifestyle) to affect biological processes involved in disease as well as access to medical care and other health-promoting resources. This interpretation is consistent with the way other studies have conceptualized 40,41 socioeconomic status as a fundamental cause of health disparities. From this perspective, having wealth allows individuals numerous opportunities to invest resources into the many proximal factors that promote health and longevity. Conclusions This cohort study found an association between wealth at midlife and longevity, and this association is unlikely to be merely an artifact of environmental and heritable characteristics shared by families. Policies that support individuals' ability to accrue wealth and achieve financial security could have considerable health benefits. In addition, policies to reduce the wealth gap may generate substantial returns to public health. JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 (Reprinted) July 23, 2021 9/12 JAMA Health Forum | Original Investigation Association of Wealth With Longevity in US Adults at Midlife ARTICLE INFORMATION Accepted for Publication: May 24, 2021. Published: July 23, 2021. doi:10.1001/jamahealthforum.2021.1652 Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2021 Finegood ED et al. JAMA Health Forum. Corresponding Author: Eric D. Finegood, PhD, Institute for Policy Research, Northwestern University, 1801 Maple Ave, Ste 2450, Evanston, IL 60201 (eric.finegood@northwestern.edu). Author Affiliations: Institute for Policy Research, Northwestern University, Evanston, Illinois (Finegood, Freedman, Chen, Miller); Department of Psychology, Northwestern University, Evanston, Illinois (Finegood, Freedman, Chen, Mroczek, Miller); Department of Psychology, University of Illinois Urbana-Champaign, Urbana (Briley); Department of Psychology, West Virginia University, Morgantown (Turiano); Department of Psychological Sciences, Purdue University, West Lafayette, Indiana (South); Department of Psychology, University of Minnesota, Minneapolis (Krueger); Department of Medical Social Sciences, Northwestern University, Evanston, Illinois (Mroczek). Author Contributions: Dr Finegood had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Concept and design: Finegood, Turiano, Krueger, Mroczek, Miller. Acquisition, analysis, or interpretation of data: All authors. Drafting of the manuscript: Finegood, Turiano, Mroczek, Miller. Critical revision of the manuscript for important intellectual content: Finegood, Briley, Turiano, Freedman, South, Krueger, Chen, Mroczek. Statistical analysis: Finegood, Briley, Turiano, Freedman, Mroczek. Obtained funding: Krueger, Mroczek. Administrative, technical, or material support: Miller. Supervision: Chen, Mroczek, Miller. Conflict of Interest Disclosures: Dr Krueger reported receiving grants from the National Institute on Aging (NIA) during the conduct of the study and outside the submitted work. Dr Miller reported receiving grants from the National Institutes of Health (NIH) outside the submitted work. No other disclosures were reported. Funding/Support: Dr Finegood was supported by grant F32HL146005 from the National Heart, Lung, and Blood Institute of the NIH. Dr Briley was supported by a Jacobs Foundation Research Fellowship. Dr Turiano was supported by the West Virginia Prevention Research Center and Cooperative Agreement No. 1-U48-DP-005004 from the Centers for Disease Control and Prevention. Dr Freedman was supported by grant F32HD100076 from the Eunice Kennedy Shriver National Institute of Child Health and Human Development of the NIH. Dr Krueger was supported by grants R01AG053217 and U19AG051426 from the NIH. Dr Mroczek was supported by grants R01 AG018436, R01 AG067622, RF1 AG064006, and U19AG051426 from the NIA and grant P30 AG05998 from the Claude D. Pepper Older Adults Independence Center. Dr Miller was supported by grant R01 MD011749 from the National Institute on Minority Health and Health Disparities and grant P50 DA051361 from the National Institute on Drug Abuse. Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication. Disclaimer: The content is solely the responsibility of the authors and does not represent the official views of the funders. Additional Information: The statistical code used in this study is available upon request to the first author. MIDUS 1 data are publicly available at the Inter-university Consortium for Political and Social Research (https://www.icpsr. umich.edu/web/ICPSR/series/203). As of this writing, the 2018 mortality update data are not publicly available. REFERENCES 1. Marmot M. Social determinants of health inequalities. Lancet. 2005;365(9464):1099-1104. doi:10.1016/S0140-6736(05)71146-6 2. Adler NE, Rehkopf DHUS. U.S. disparities in health: descriptions, causes, and mechanisms. Annu Rev Public Health. 2008;29:235-252. doi:10.1146/annurev.publhealth.29.020907.090852 3. Chetty R, Stepner M, Abraham S, et al. The association between income and life expectancy in the United States, 2001-2014. JAMA. 2016;315(16):1750-1766. doi:10.1001/jama.2016.4226 JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 (Reprinted) July 23, 2021 10/12 JAMA Health Forum | Original Investigation Association of Wealth With Longevity in US Adults at Midlife 4. Pollack CE, Chideya S, Cubbin C, Williams B, Dekker M, Braveman P. Should health studies measure wealth? A systematic review. Am J Prev Med. 2007;33(3):250-264. doi:10.1016/j.amepre.2007.04.033 5. Wolfe JD, Baker EH, Scarinci IC. Wealth and obesity among US adults entering midlife. Obesity (Silver Spring). 2019;27(12):2067-2075. doi:10.1002/oby.22625 6. Demakakos P, Biddulph JP, Bobak M, Marmot MG. Wealth and mortality at older ages: a prospective cohort study. J Epidemiol Community Health. 2016;70(4):346-353. doi:10.1136/jech-2015-206173 7. Daly MC, Duncan GJ, McDonough P, Williams DR. Optimal indicators of socioeconomic status for health research. Am J Public Health. 2002;92(7):1151-1157. doi:10.2105/AJPH.92.7.1151 8. Hajat A, Kaufman JS, Rose KM, Siddiqi A, Thomas JC. Long-term effects of wealth on mortality and self-rated health status. Am J Epidemiol. 2011;173(2):192-200. doi:10.1093/aje/kwq348 9. Pool LR, Burgard SA, Needham BL, Elliott MR, Langa KM, Mendes de Leon CF. Association of a negative wealth shock with all-cause mortality in middle-aged and older adults in the United States. JAMA. 2018;319(13): 1341-1350. doi:10.1001/jama.2018.2055 10. Hill WD, Davies NM, Ritchie SJ, et al. Genome-wide analysis identifies molecular systems and 149 genetic loci associated with income. Nat Commun. 2019;10(1):5741. doi:10.1038/s41467-019-13585-5 11. Donovan SJ, Susser E. Commentary: advent of sibling designs. Int J Epidemiol. 2011;40(2):345-349. doi:10.1093/ije/dyr057 12. Lawlor DA, Mishra GD, eds. Family Matters: Designing, Analysing and Understanding Family Based Studies in Life Course Epidemiology. Oxford University Press; 2009. doi:10.1093/acprof:oso/9780199231034.001.0001 13. Carlin JB, Gurrin LC, Sterne JAC, Morley R, Dwyer T. Regression models for twin studies: a critical review. Int J Epidemiol. 2005;34(5):1089-1099. doi:10.1093/ije/dyi153 14. Turkheimer E, Harden KP. Behavior genetic research methods: testing quasi-causal hypotheses using multivariate twin data. In: Reis HT, Judd CM, eds. Handbook of Research Methods in Social and Personality Psychology. 2nd ed. Cambridge University Press; 2014:159-187. 15. Krieger N, Chen JT, Coull BA, Selby JV. Lifetime socioeconomic position and twins’ health: an analysis of 308 pairs of United States women twins. PLoS Med. 2005;2(7):e162. doi:10.1371/journal.pmed.0020162 16. Ericsson M, Pedersen NL, Johansson ALV, Fors S, Dahl Aslan AK. Life-course socioeconomic differences and social mobility in preventable and non-preventable mortality: a study of Swedish twins. Int J Epidemiol. 2019;48 (5):1701-1709. doi:10.1093/ije/dyz042 17. Søndergaard G, Mortensen LH, Nybo Andersen AM, et al. Does shared family background influence the impact of educational differences on early mortality? Am J Epidemiol. 2012;176(8):675-683. doi:10.1093/aje/kws230 18. Fujiwara T, Kawachi I. Is education causally related to better health? a twin fixed-effect study in the USA. Int J Epidemiol. 2009;38(5):1310-1322. doi:10.1093/ije/dyp226 19. Næss O, Hoff DA, Lawlor D, Mortensen LH. Education and adult cause-specific mortality—examining the impact of family factors shared by 871 367 Norwegian siblings. Int J Epidemiol. 2012;41(6):1683-1691. doi:10.1093/ije/dys143 20. Lundborg P, Lyttkens CH, Nystedt P. The effect of schooling on mortality: new evidence from 50,000 Swedish twins. Demography. 2016;53(4):1135-1168. doi:10.1007/s13524-016-0489-3 21. Madsen M, Andersen AM, Christensen K, Andersen PK, Osler M. Does educational status impact adult mortality in Denmark? A twin approach. Am J Epidemiol. 2010;172(2):225-234. doi:10.1093/aje/kwq072 22. Madsen M, Andersen PK, Gerster M, Andersen AM, Christensen K, Osler M. Are the educational differences in incidence of cardiovascular disease explained by underlying familial factors? A twin study. Soc Sci Med. 2014;118: 182-190. doi:10.1016/j.socscimed.2014.04.016 23. Osler M, McGue M, Christensen K. Socioeconomic position and twins’ health: a life-course analysis of 1266 pairs of middle-aged Danish twins. Int J Epidemiol. 2007;36(1):77-83. doi:10.1093/ije/dyl266 24. World Inequality Lab. World Inequality Report 2018. Accessed July 27, 2020. https://wir2018.wid.world 25. Menasce Horowitz J, Igielnik R, Kochhar R. Most Americans say there is too much economic inequality in the U.S., but fewer than half call it a top priority. January 9, 2020. Accessed October 14, 2020. https://www. pewsocialtrends.org/2020/01/09/most-americans-say-there-is-too-much-economic-inequality-in-the-u-s- but-fewer-than-half-call-it-a-top-priority/ 26. Powell JH. Recent Economic Developments and the Challenges Ahead: Speech at the National Association for Business Economics Virtual Annual Meeting. October 6, 2020. Accessed December 28, 2020. https://www.federalreserve.gov/newsevents/speech/powell20201006a.htm JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 (Reprinted) July 23, 2021 11/12 JAMA Health Forum | Original Investigation Association of Wealth With Longevity in US Adults at Midlife 27. Cook FL, Weiss E. COVID-19 is hitting older workers like a sledgehammer—here’s what Congress can do. The Hill. May 1, 2020. Accessed June 9, 2020. https://thehill.com/opinion/finance/495541-covid-19-is-hitting-older- workers-like-a-sledge-hammer-heres-what-congress 28. Schwandt H. Wealth shocks and health outcomes: evidence from stock market fluctuations. Am Econ J Appl Econ. 2018;10(4):349-377. doi:10.1257/app.20140499 29. Parsing genetic and environmental influences on longevity: A twin and sibling study in MIDUS. Open Science Framework. Registered November 15, 2019. Accessed June 17, 2021. https://osf.io/zyedp 30. von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP; STROBE Initiative. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. PLoS Med. 2007;4(10):e296. doi:10.1371/journal.pmed.0040296 31. Brim OG, Ryff CD, Kessler RC eds. How Healthy Are We? A National Study of Well-Being at Midlife. University of Chicago Press; 2004. 32. StataCorp. Stata Statistical Software: Release 16. StataCorp LLC; 2019. 33. Allison PD. Fixed Effects Regression Models. Sage Publications, Inc; 2009. doi:10.4135/9781412993869 34. Sjölander A, Lichtenstein P, Larsson H, Pawitan Y. Between-within models for survival analysis. Stat Med. 2013;32(18):3067-3076. doi:10.1002/sim.5767 35. McGue M, Osler M, Christensen K. Causal inference and observational research: the utility of twins. Perspect Psychol Sci. 2010;5(5):546-556. doi:10.1177/1745691610383511 36. Muthén LK, Muthén BO. Mplus User’s Guide. 8 ed. Muthén & Muthén; 1998-2017. 37. Gilman SE, Loucks EB. Another casualty of sibling fixed-effects analysis of education and health: an informative null, or null information? Soc Sci Med. 2014;118:191-193. doi:10.1016/j.socscimed.2014.06.029 38. Stiglitz JE. The Price of Inequality: How Today’s Divided Society Endangers Our Future. W W Norton & Co; 2012. 39. Sykes BL, Maroto M. A wealth of inequalities: mass incarceration, employment, and racial disparities in U.S. household wealth, 1996 to 2011. Russell Sage Found J Soc Sci. 2016;2(6):129-152. doi:10.7758/rsf.2016.2.6.07 40. Phelan JC, Link BG, Diez-Roux A, Kawachi I, Levin B. “Fundamental causes” of social inequalities in mortality: a test of the theory. J Health Soc Behav. 2004;45(3):265-285. doi:10.1177/002214650404500303 41. Link BG, Phelan J. Social conditions as fundamental causes of disease. J Health Soc Behav. 1995;Spec No:80-94. doi:10.2307/2626958 SUPPLEMENT. eMethods 1. Full Methods Section eMethods 2. Supplementary Sensitivity Analyses eMethods 3. Deviations From Original Preregistered Analysis Plan eTable 1. Spline Model eTable 2. BW Model Among Siblings and Twins in Groups/Pairs Where All Family Members Have$382,500 in Net Worth at M1 eTable 3. Multi-Level Cox Regression Analysis in Full Analysis Sample eTable 4. Multi-Level Cox Regression Analysis in Sample of Siblings and Twins eTable 5. Multi-Level Cox Regression Analysis Across Sibling Subsamples eReferences. JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 (Reprinted) July 23, 2021 12/12 Supplementary Online Content Finegood ED, Briley DA, Turiano NA, et al. Association of wealth with longevity in US adults at midlife. JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 eMethods 1. Full Methods Section eMethods 2. Supplementary Sensitivity Analyses eMethods 3. Deviations From Original Preregistered Analysis Plan eTable 1. Spline Model eTable 2. BW Model Among Siblings and Twins in Groups/Pairs Where All Family Members Have <= $382,500 in Net Worth at M1 eTable 3. Multi-Level Cox Regression Analysis in Full Analysis Sample eTable 4. Multi-Level Cox Regression Analysis in Sample of Siblings and Twins eTable 5. Multi-Level Cox Regression Analysis Across Sibling Subsamples eReferences. This supplementary material has been provided by the authors to give readers additional information about their work. © 2021 Finegood ED et al. JAMA Health Forum. eMethods 1. Full Methods Section This report adheres to the STROBE guidelines for cohort studies . The analyses presented here were conducted in 2020 2021 and preregistered on the Open Science Framework (https://osf.io/zyedp) on November 15, 2019. The pre registration includes hypotheses, decision rules for inclusion, rationale for covariates, and statistical plan. Deviations are detailed in eMethods 3 of this Supplement. Participants Data come from the Midlife in the United States (MIDUS) study, an ongoing national study of health and aging processes begun in 1994. In the current analysis, we utilized data from wave 1 of MIDUS (MIDUS I) collected from years 1994 1996. 7,108 adults aged 20 75 years participated at MIDUS I. They were recruited via a nationally representative random digit dialing sampling strategy, which included recruiting subsamples of siblings and twins. Those who provided oral informed consent to participate completed a telephone assisted survey as well as a mailed self administered questionnaire. The Institutional Review Boards at the University of Wisconsin and Harvard Medical School approved the MIDUS study procedures. Mortality data through October 2018 were collected by the MIDUS study team at the University of Wisconsin. Analysis sample Of the 7,108 cases participating at MIDUS 1, n = 3,715 were designated for purposes of the current analysis as ‘singletons’, meaning that they did not have a sibling or a co twin in MIDUS. 1,479 cases were non twin full siblings, meaning that they had at least one non twin full sibling in MIDUS. 1,914 were twins. Of these, n = 30 were missing zygosity information and © 2021 Finegood ED et al. JAMA Health Forum. were excluded from our analysis sample. Additionally, n = 61 did not have a co twin in MIDUS so were designated as singletons in our analysis. In families that had both a pair of MZ twins and a pair of DZ twins, only the MZ pair was retained in our analysis sample (19 DZ twins excluded). We chose to retain the MZ pair as opposed to the DZ pair because the number of MZ pairs was lower overall than the number of DZ pairs. In families that had either two pairs of MZ twins or two pairs of DZ twins, one twin pair was retained at random (42 cases excluded from analysis sample). Given all of these criteria, n = 7,017 cases were considered for inclusion in our analysis sample (3,776 singletons; 1,479 full siblings; 1,064 DZ twins; and 698 MZ twins). Survival analyses were conducted among those individuals with complete data on mortality status, net worth, and the covariates (n = 5,414 cases: 2,675 singletons, 1,282 non twin siblings, 864 DZ twins, 593 MZ twins). Twin or siblings who had complete data and who were in a pair with a co twin/sibling that was missing data were included in this analytic sample of 5,414 although they were excluded from subsequent discordant twin/sibling analyses, resulting in 1,214 non twin siblings, 740 DZs, and 536 MZs in the discordant twin/sibling analyses. Missing data. Survival analyses were conducted among those individuals with complete data on mortality status, net worth, and the covariates. The amount of missing data among the analysis variables collected in the phone interview was low (0% to 4% missing). Of the 7,017 cases, n = 6,240 (88.9%) had completed both the phone interview and the self administered questionnaire. Accordingly, missingness on the analysis variables collected in the self  administered questionnaire was higher (12% out of n = 7,017 were missing race/ethnicity data and 20.2% out of n = 7,017 were missing net worth data). In addition, 6 participants did not have mortality data as of October 31, 2018. Considering the missing data on all analysis © 2021 Finegood ED et al. JAMA Health Forum. variables, 5,414 cases had complete data (2,675 singletons, 1,282 non twin siblings, 864 DZ twins, 593 MZ twins). Compared to those included in the analysis sample (n = 5,414), those who were excluded from the analysis sample (n = 1,603) were more likely to be non white ( (1, n = 6171) = 28.96, p < .001), to have had a previous heart condition ( (1, n = 6997) = 4.62, p = .03), were older in age (t(2443.35) =  2.01, p = .04, and reported lower levels of parent education (t(2206.91) = 5.60, p < .001). There were no associations between inclusion status and sex, previous cancer diagnosis, smoking status, and regular alcohol consumption. Because our primary research question concerned within sibling/twin associations between net worth and mortality risk, these sociodemographic and health related factors related to inclusion status were considered less problematic. Measures Net worth. At MIDUS 1, participants responded to the following prompt: “Suppose you (and your spouse or partner) cashed in all your checking and savings accounts, stocks and bonds, real estate, sold your home, your vehicles, and all your valuable possessions. Then suppose you put that money toward paying off your mortgage and all your other loans, debts, and credit cards. Would you have any money left over after paying your debts or would you still owe money?” Participants reported how much that amount would be using binned response categories that specified ranges of dollars. Those who would still owe money were truncated at a value of $0. Individuals whose net worth exceeded $1,000,000 were truncated at that value. This truncation prior to MIDUS 1 data release was due to privacy and human subjects concerns. However, the MIDUS study does provide the percentage that had negative net worth, and this was 13%. On the other end, those with net worth over $1 million comprised only 2% of the © 2021 Finegood ED et al. JAMA Health Forum. sample. These bottom  and top truncations, like any truncation, reduce the size of correlation and regression coefficients thereby creating an underestimate. Such a conservative bias was deemed acceptable. Covariates. Survival models included several covariates selected a priori because of their known associations with wealth and health. These included participant age, self reported race/ethnicity (analyzed here as non white vs. white), sex (female vs. male), and self reported history of cancer or heart disease, as diagnosed by a medical doctor. We also controlled for ever having smoked regularly or consumed alcohol regularly. Highest level of parents’ education (an indicator of childhood socioeconomic status), was also included as a covariate. In the case that siblings/twins differed in their report of the highest level of parents’ education, the higher of the discrepant responses was used for all siblings/twins within the same family. All cause mortality. Mortality follow up was completed by the MIDUS study team at the University of Wisconsin, Madison. Date of death was obtained from various sources, including responses from relatives, other informant reports, newspaper or online obituaries, th and the National Death Index (using the 15 day of the month, rather than the exact day of death, to maintain participant confidentiality). Survival time was the number of years between the date when the MIDUS 1 self administered questionnaires were returned to the MIDUS study team (years 1995 96) and the date of death (otherwise, if alive, the censor date of Oct 31, 2018, which was the date of the MIDUS study team’s latest mortality update). Analysis plan A series of Cox survival analyses were estimated using Stata 16 . To test the individual  level association between wealth and longevity, we first conducted analyses among all cases © 2021 Finegood ED et al. JAMA Health Forum. that had complete data on analysis variables utilizing robust SEs to account for dependence among family members (Model 1). Next, we estimated survival models only within the sub sample of twins and non twin siblings. In Model 2, siblings and twins were treated “as individuals” but a shared frailty was included to model dependence among family members. In Model 3, we estimated the within  family association between wealth and longevity by calculating the family level average net worth (among families that had >=2 members with complete data) and subsequently calculating the difference between each individual’s net worth and their family average. When included alongside the family level mean, the hazard ratio (HR) for these mean deviation scores estimate the within family association between net worth and longevity. This “between within” 4,5 (BW) method is a common approach to fixed effects modeling . When applied in this way, it allows us to compare siblings/twins in the same family to one another, and thus to control for all unmeasured shared family level variables, consistent with the discordant sibling/twin 6–8 design . In survival analysis, the BW method has been shown to provide similar estimates to more common approaches for co twin/sibling control with survival data (e.g. conditional likelihood methods like stratified Cox regression) and has been observed to be optimal statistically . To further disambiguate environmental versus genetic influences, we tested whether within family associations between net worth and longevity varied across non twin sibling, MZ, and DZ subsamples. We did this by including a pair of two way interaction terms crossing the mean deviation scores of net worth with dummy codes for DZs or MZs; non twin full siblings modeled as reference group). A Wald test tested the equality of the within family net worth © 2021 Finegood ED et al. JAMA Health Forum. coefficient across siblings, DZ, and MZ subsamples. Lastly, we estimated separate survival models for non twin siblings, DZ pairs, and MZ pairs (Models 4 6). A significant within family association (p < 0.05) observed among non twin siblings  but not among DZ or MZ pairs  would suggest some residual confounding by early life factors since twins share a closer pre  and post  natal environment than non twin siblings who may be born years apart. A within family association observed both among siblings and DZ pairs but not among MZ pairs would suggest genetic confounding. Sensitivity analyses were undertaken to test the robustness of findings. These analyses addressed the skewed distribution of the net worth variable and tested the possibility of non linear associations between net worth and longevity. They also clarified the role of pre existing health problems and considered other model specifications. Primary models were also re estimated as stratified Cox regressions using STATA (see eMethods 2) and as multilevel Cox regressions using Mplus (see eMethods 3). © 2021 Finegood ED et al. JAMA Health Forum. eMethods 2. Supplementary Sensitivity Analyses Sensitivity analyses were undertaken in the combined sibling and twin subsample to test the robustness of the findings to different model specifications. First, to assess the potential for a non linear association between net worth and longevity, we estimated a spline model (eTable th th 1) including two knots: one at the 75 percentile of net worth ($125,000) and one at the 90 th percentile ($382,500). As shown in eTable 1, HRs for net worth for those below the 75 th th percentile (HR=.86, CI=.76 .97, p=.01) and between the 75  90 percentiles (HR=.90, CI=.82  .98, p=.01) were similar, and a subsequent test of the estimates confirmed that their difference th was no greater than chance (p=.66). The HR for net worth for those above the 90 percentile (individuals with >=$680,000; n=182, 45 decedents) was not statistically significant (HR=1.02, 95%CI=.97 1.08, p=.31) and a test of the estimates indicated that the difference between this th th HR and the HR for those between the 75  90 percentiles was statistically significant (p=.03). This indicates a possible diminished return on net worth at the very high end of the net worth distribution, though this reflects only approximately 7% of the sample. Collectively, the spline model indicates that among the large majority (93%) of siblings and twins (i.e. those whose net worth was <= $382,000), the association between net worth and survival was approximately linear. We subsequently reran Model 3 in a subsample of siblings and twins who were in family groups where all members had <= $382,500 in net worth (n=2,110; 321 deaths). As shown in eTable 2 of this supplement, the HR for net worth in this restricted sample (HR=0.89, CI=0.82  0.96, p=0.004) suggested a larger association between net worth and longevity among those © 2021 Finegood ED et al. JAMA Health Forum. with lower family level wealth. However, these results should be interpreted with a high level of caution due to the restricted sample size. In another sensitivity analysis, we recoded net worth into ordinal decile groups, given the large positive skew of the net worth distribution. In the combined sibling and twin subsample, between family (HR = 0.90, 95% CI = 0.85 0.94, p < .001) and within family (HR = 0.92, 95% CI = 0.87 0.96, p = .001) net worth estimates remained significant predictors of mortality. Next, to account for the possibility of residual confounding by health status (having a medical problem may both reduce one’s ability to accumulate wealth and increase mortality risk), analyses were re estimated among sibling/twin pairs who were free of previous cancer or heart disease. Among sibling groups with >2 members, only those siblings without heart disease and cancer were compared to one another. Results were largely similar in this restricted sample (n = 1,740; 196 deaths): HR = 0.95, 95% CI= 0.90  0.99, p = 0.04, and HR = 0.94, 95% between within CI = 0.90 0.98, p = 0.01). We also tested the possibility of a nonlinear age trend by including an age term (HR =1.00, 95% CI = 0.99 1.00, p = .65) and also an age*sex interaction term (HR = 1.01, 95% CI = 0.99 1.03, p = 0.14), neither of which was associated with mortality risk nor changed the interpretations of other model estimates. We also tested an interaction between the within  family net worth estimate and participant age at MIDUS 1. The rationale being that the within  family association between net worth and longevity may vary as a function of age. The interaction term was not statistically significant, HR = 1.00, 95% CI =0.99 1.00, p=0.37, © 2021 Finegood ED et al. JAMA Health Forum. suggesting that the within family association between net worth and longevity did not vary by age at MIDUS 1. Lastly, as a more conservative test of possible confounding by early experience, we restricted the analysis sample to only same sex sibling groups/twin pairs. Point estimates of between family (HR = .94, 95% CI = 0.90 0.98 p = .007) and within family (HR = .96, 95% CI = 0.92 1.01, p = .16) net worth estimates were consistent with estimates observed in Model 3 in the main text, although the p value for the within family estimate was not statistically significant, likely due to the substantial reduction in power and sample size in this restricted sample (N =2,490; n=421 dead; vs. N = 1,359; n=221 dead). Given the large loss in Model3 Same sex power due to the reduction in sample size and number of deaths, the result of this sensitivity analysis should be interpreted with a high level of caution. Stratified Cox Regression models. We re ran Models 3 6 as stratified Cox regressions, stratifying by Family ID. The within family net worth estimates were as follows: Model 3 HR = .94, 95%CI = .90 .97, p = .002; Model 4 HR = .93, 95%CI = .89 .97, p = .004; Model 5 HR = .95, 95%CI = .87 1.05, p = .37; Model 6 HR = .95, 95%CI = .84 1.07, p = .42. © 2021 Finegood ED et al. JAMA Health Forum. eMethods 3. Deviations From Original Preregistered Analysis Plan The analyses presented in the main text are consistent with the study rationale, hypotheses, and overall analytic plan outlined in our preregistration (https://osf.io/zyedp). We did, however, deviate from the original analytic plan in some minor ways—these deviations are outlined below. First, in the original analysis plan, we specified that survival analyses would be estimated as multi level Cox regression models. Instead, each survival model was estimated as a Cox model with shared frailty term to account for clustering. Both analytic approaches gave very similar results for all models—compare coefficients in Model 1 and Model 3 of Table 2 in the main text to estimates in eTables 3 and 4 (results from multilevel Cox regressions run in Mplus version 8 ). Second, for the sibling/twin comparison analysis (step 2 of the original analysis plan), we originally planned to run a two level mixture analysis using a Cox regression model specifying the KNOWNCLASS option in Mplus version 8 in order to estimate a multiple group analysis— allowing model coefficients to vary across subsamples of non twin siblings, DZ twins, MZ twins. Instead, in the main text, we present results from three separate survival models: one among non twin siblings, one among DZ twins, and one among MZ twins. With either analytic strategy, coefficients were very similar and provided the same interpretations—compare coefficients from Models 4, 5, and 6 in main text Table 2 to coefficients in eTable 5, which display results from the multilevel mixture model using Cox regression with KNOWNCLASS option in Mplus version 8). © 2021 Finegood ED et al. JAMA Health Forum. Third, as described in the main text, we undertook a conservative test of whether the within family net worth effect varied across full sibling, MZ twin, and DZ twin subsamples by including two way interaction terms between the within family net worth variable and dummy codes for cluster type and conducting a Wald test to test the equality of the within family estimates. This test of the equality of within family net worth effects across sibling subsamples was not described in the original analysis plan. We also originally proposed an exploratory test of the two way interaction between parent education and the within family net worth twin/sibling difference score. We decided to omit this test from the main text because we felt that the rationale for this test was tangential to our primary analyses, which were already many in number. It is also not an optimal test for answering questions related to social mobility, its stated purpose in the preregistration. In any case, we report results from this test here. Among the combined sample of non twin siblings, DZ twins, and MZ twins (n =2,490) there was no interaction between parents’ highest level of education and sibling/twin net worth difference score (HR = .99, 95% CI .98  1.00, p = .30). Indeed, when the sample is split into subsamples of those at or below (n = 1,333) and above (n = 1,157) the median on parent education, the within family net worth estimate was significant in both subsamples (at/below median: HR =.95, 95% CI .91 .99, p = .04; above median: within HR =.94, 95% CI .89 .98, p = .006). within We also specified that full information maximum likelihood estimation (FIML) would be used to handle missing data. In the preregistration, we specified the criteria by which siblings and twin pairs would be excluded from our analysis sample—resulting in an analytic sample of n=7,017. As described in eMethods 1 of the Supplement, of the n=7,017, n=6,240 had © 2021 Finegood ED et al. JAMA Health Forum. completed both the phone interview and the self administered questionnaire. Another n = 6 did not have mortality data, resulting in a possible analytic sample of n=6,234. Of the analysis variables, net worth had the most missingness: n=640 out of 6,234 (10.3%) missing net worth, likely because some participants were unwilling or unable to provide this information. Because our primary interest was to compare mortality risk within sibling groups/twin pairs who were discordant on net worth, only siblings groups/twin pairs in which discordance could be estimated (e.g. twin pairs in which both twins had non missing net worth data) were useful analytically. Thus, analyses were conducted only among cases that had complete data and FIML was not used. © 2021 Finegood ED et al. JAMA Health Forum. eTable 1. Spline Model HR 95% CI p Age 1.11 1.09 1.12 < .001 Female 0.74 0.60 0.93 .009 Non white 0.82 0.46 1.46 0.51 Parent education 0.97 0.93 1.01 0.18 Heart disease 1.96 1.56 2.47 < .001 Cancer 1.57 1.18 2.07 0.002 Smoking 2.07 1.65 2.60 < .001 Alcohol use 0.92 0.74 1.15 0.50 Net worth spline 1 0.86 0.76 0.97 0.01 Net worth spline 2 0.90 0.82 0.98 0.01 Net worth spline 3 1.02 0.97 1.08 0.31 Note: N=2,490, which includes 421 deaths. HR=hazard ratio, 95% CI= 95% confidence interval of the hazard ratio. © 2021 Finegood ED et al. JAMA Health Forum. eTable 2. BW Model Among Siblings and Twins in Groups/Pairs Where All Family Members Have <= $382,500 in Net Worth at M1 HR 95% CI p Age 1.11 1.09 1.12 < 0.001 Female 0.74 0.58 0.95 0.01 Non white 0.67 0.37 1.23 0.20 Parent education 0.97 0.92 1.02 0.27 Heart disease 2.04 1.58 2.63 < 0.001 Cancer 1.56 1.14 2.14 0.005 Smoking 1.87 1.45 2.41 < 0.001 Alcohol use 1.04 0.81 1.34 0.71 Net worth (between family) 0.84 0.78 0.90 < 0.001 Net worth (within family) 0.89 0.82 0.96 0.004 Note: N=2,110, which includes 321 deaths. HR=hazard ratio, 95% CI= 95% confidence interval of the hazard ratio. © 2021 Finegood ED et al. JAMA Health Forum. eTable 3. Multi-Level Cox Regression Analysis in Full Analysis Sample HR (95% CI) p value Age 1.10 (1.09 1.11) < 0.001 Female 0.82 (0.72 0.94) 0.005 Non white 1.13 (0.87 1.46) 0.35 Parent education 0.97 (0.95 0.99) 0.03 Heart disease 1.87 (1.61 2.17) < 0.001 Cancer 1.44 (1.22 1.70) < 0.001 Smoking 1.75 (1.53 2.00) < 0.001 Alcohol use 1.04 (0.91 1.19) 0.47 Net worth 0.95 (0.94 0.97) < 0.001 © 2021 Finegood ED et al. JAMA Health Forum. eTable 4. Multi-Level Cox Regression Analysis in Sample of Siblings and Twins HR (95% CI) p value Age 1.10 (1.09 1.11) < 0.001 Female 0.78 (0.64 0.96) 0.01 Non white 0.85 (0.48 1.50) 0.58 Parent education 0.97 (0.93 1.01) 0.25 Heart disease 1.85 (1.46 2.35) < 0.001 Cancer 1.53 (1.18 1.99) 0.001 Smoking 2.03 (1.63 2.52) < 0.001 Alcohol use 0.92 (0.74 1.14) 0.46 Net worth (between family) 0.96 (0.93 0.99) 0.01 Net worth (within family) 0.95 (0.92 0.98) 0.002 © 2021 Finegood ED et al. JAMA Health Forum. eTable 5. Multi-Level Cox Regression Analysis Across Sibling Subsamples Siblings (n = 1,214) DZ twins (n = 740) MZ twins (n = 536) HR (95% CI) p value HR (95% CI) p value HR (95% CI) p value Age 1.10 (1.08 1.12) < 0.001 1.12 (1.09 1.14) < 0.001 1.11 (1.08 1.15) < 0.001 Female 0.64 (0.47 0.87) 0.005 0.88 (0.60 1.29) 0.51 1.03 (0.58 1.83) 0.89 Non white 1.50 (0.66 3.42) 0.32 0.25 (0.05 1.09) 0.06 1.16 (0.31 4.29) 0.81 Parent edu. 1.00 (0.94 1.06) 0.87 0.96 (0.90 1.03) 0.38 0.91 (0.82 1.02) 0.11 Heart disease 2.07 (1.51 2.86) < 0.001 1.72 (1.09 2.69) 0.01 2.39 (1.10 5.21) 0.02 Cancer 1.92 (1.37 2.70) < 0.001 1.01 (0.57 1.77) 0.96 1.55 (0.58 4.14) 0.38 Smoking 2.21 (1.63 2.98) < 0.001 2.08 (1.39 3.11) < 0.001 2.11 (1.17 3.80) 0.01 Alcohol use 0.73 (0.52 1.01) 0.06 1.18 (0.81 1.72) 0.36 1.04 (0.61 1.76) 0.86 Net worth (between) 0.98 (0.93 1.02) 0.41 0.90 (0.84 0.97) 0.005 0.95 (0.89 1.03) 0.26 Net worth (within) 0.94 (0.90 0.98) 0.005 0.94 (0.86 1.02) 0.19 0.95 (0.90 1.01) 0.16 © 2021 Finegood ED et al. JAMA Health Forum. eReferences 1. von Elm E, Altman DG, Egger M, et al. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement: Guidelines for Reporting Observational Studies. PLoS Med. 2007;4(10):e296. doi:10.1371/journal.pmed.0040296 2. Brim OG, Ryff CD, Kessler RC, eds. How Healthy Are We? A National Study of Well Being at Midlife. Chicago, IL: University of Chicago Press; 2004. 3. StataCorp. Stata Statistical Software: Release 16. StataCorp LLC; 2019. 4. Allison PD. Fixed Effects Regression Models. Sage Publications, Inc.; 2009. 5. Sjölander A, Lichtenstein P, Larsson H, Pawitan Y. Between within models for survival analysis. Stat Med. 2013;32(18):3067 3076. doi:10.1002/sim.5767 6. Carlin JB, Gurrin LC, Sterne JAC, Morley R, Dwyer T. Regression models for twin studies: A critical review. Int J Epidemiol. 2005;34(5):1089 1099. doi:https://doi.org/10.1093/ije/dyi153 7. Turkheimer E, Harden KP. Behavior Genetic Research Methods: Testing Quasi Causal Hypotheses Using Multivariate Twin Data. In: Reis HT, Judd CM, eds. Handbook of Research Methods in Social and Personality Psychology. 2nd ed. Cambridge University Press; 2014:159 187. 8. McGue M, Osler M, Christensen K. Causal inference and observational research: The utility of twins. Perspect Psychol Sci. 2010;5(5):546 556. doi:10.1177/1745691610383511 9. Muthén LK, Muthén BO. Mplus User’s Guide. Eighth Ed. Los Angeles, CA: Muthén & Muthén. 1998 2017. © 2021 Finegood ED et al. JAMA Health Forum. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png JAMA Health Forum American Medical Association

Association of Wealth With Longevity in US Adults at Midlife

Download PDF
31 pages

Loading next page...
 
/lp/american-medical-association/association-of-wealth-with-longevity-in-us-adults-at-midlife-55d0JyeYDa

References (73)

Publisher
American Medical Association
Copyright
Copyright 2021 Finegood ED et al. JAMA Health Forum.
eISSN
2689-0186
DOI
10.1001/jamahealthforum.2021.1652
Publisher site
See Article on Publisher Site

Abstract

Key Points Question Is net worth at midlife IMPORTANCE Wealthy adults tend to live longer than those with less wealth. However, a challenge associated with all-cause mortality? in this area of research has been the reduction of potential confounding by factors associated with Findings In this cohort study of 5414 the early environment and heritable traits, which could simultaneously affect socioeconomic participants in the Midlife in the United circumstances in adulthood and health across the life course. States study, those who had accumulated a higher net worth by OBJECTIVE To identify the association between net worth at midlife and subsequent all-cause midlife had significantly lower mortality mortality in individuals as well as within siblings and twin pairs. risk over the subsequent 24 years. In sibling and twin comparison models that DESIGN, SETTING, AND PARTICIPANTS This cohort study conducted a series of analyses using controlled for shared early life data from the Midlife in the United States (MIDUS) study, an ongoing national study of health and experiences and genetic influence, the aging. The sample included adults (unrelated individuals, full siblings, and dizygotic and monozygotic association between net worth and twins) aged 20 to 75 years, who participated in wave 1 of the MIDUS study, which occurred from 1994 longevity was similar in magnitude. to 1996. The analyses were conducted between November 16, 2019, and May 18, 2021. Meaning Net worth at midlife was EXPOSURES Self-reported net worth (total financial assets minus liabilities) at midlife (the middle associated with longevity among adults years of life). in the study, and this association is unlikely to be merely an artifact of early MAIN OUTCOMES AND MEASURES All-cause mortality was tracked over nearly 24 years of experiences or heritable traits shared follow-up, with a censor date of October 31, 2018. Survival models tested the association between by families. net worth and all-cause mortality. Discordant sibling and twin analyses compared longevity within siblings and twin pairs who, given their shared early experiences and genetic backgrounds, were Invited Commentary matched on these factors. Supplemental content RESULTS The full sample comprised 5414 participants, who had a mean (SD) age of 46.7 (12.7) years Author affiliations and article information are and included 2766 women (51.1%). Higher net worth was associated with lower mortality risk (hazard listed at the end of this article. ratio [HR], 0.95; 95% CI, 0.94-0.97; P < .001). Among siblings and twin pairs specifically (n = 2490), a similar within-family association was observed between higher net worth and lower mortality (HR, 0.94; 95% CI, 0.91-0.97; P = .001), suggesting that the sibling or twin with more wealth tended to live longer than their co-sibling or co-twin with less wealth. When separate estimates were performed for the subsamples of siblings (HR, 0.94; 95% CI, 0.90-0.97; P = .002), dizygotic twins (HR, 0.94; 95% CI, 0.86-1.02; P = .19), and monozygotic twins (HR, 0.95; 95% CI, 0.87-1.04; P = .34), the within-family estimates of the net worth–mortality association were similar, although the precision of estimates was reduced among twins. CONCLUSIONS AND RELEVANCE This cohort study found that wealth accumulation at midlife was associated with longevity in US adults. Discordant sibling analyses suggested that this association is unlikely to be simply an artifact of early experiences or heritable characteristics shared by families. JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 Open Access. This is an open access article distributed under the terms of the CC-BY License. JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 (Reprinted) July 23, 2021 1/12 JAMA Health Forum | Original Investigation Association of Wealth With Longevity in US Adults at Midlife Introduction 1-3 Socioeconomic disparities in life expectancy are substantial in size. Financial wealth or net worth, which is the value of an individual’s assets (such as savings, real estate, and vehicles) minus 4,5 4,6-9 liabilities, is directly associated with longevity and, in some studies, has been found to be more strongly associated with mortality than other indicators of socioeconomic status, such as 7 6 occupational prestige, educational attainment, and income. However, a challenge in this area of research has been eliminating or minimizing the potential for confounding by the early environment and heritable traits, either of which could simultaneously affect socioeconomic conditions in adulthood and health in the course of life. Discordant sibling designs allow for the identification and control of such confounders. Full siblings who were raised in the same family share much of their early rearing environment and are genetically related to one another. Thus, in sibling-comparison studies, factors that are shared 11,12 between siblings are controlled. Twin comparisons provide an even greater control of family-level early-life confounding and, in the case of monozygotic (MZ) twins, control for all heritable genetic 13,14 factors. Previous research found that discordance in occupational prestige was associated with 15 16 cardiovascular risk and overall mortality ; twins with lower-prestige jobs had worse health on both outcomes compared with their co-twins with higher-prestige jobs. This pattern suggests that socioeconomic disparities in health are affected by experiential factors in adulthood over and above any potential confounders that involve the siblings’ shared early environment and genetic 16-22 characteristics. In other discordant sibling and twin analyses, educational attainment and composite measures of adult socioeconomic position also have been associated with better adult 18,22,23 16,17,19-21 health outcomes and longevity. However, results from these and other studies that used different methods do suggest these associations may be partially explained by shared family- 17-19,21,22 10,18,23 level environmental factors or genetic predispositions. Comparatively little attention has been given to wealth disparities, a potentially important oversight because wealth inequality is far greater and growing at a faster rate than income inequality 24,25 in the United States. At the individual level, those with greater wealth are better able to access health-promoting resources (eg, medical care, safe places to exercise, and fresh foods). Wealthy individuals also have more protection from economic shocks, such as job loss, unexpected health care expenses, or other financial crises. The association between wealth and longevity has become more relevant in the past year because of the economic fallout from the COVID-19 pandemic, which 26 27 has disproportionately affected the financial security of low-income and older workers. In this cohort study, we used a discordant sibling design to conservatively estimate the association between wealth and longevity. Specifically, we aimed to identify the association between net worth at midlife (the middle years of life) and subsequent all-cause mortality in individuals as well as within siblings and twin pairs. We posed 2 research questions. First, was wealth accumulation at midlife associated with longevity over a nearly 24-year follow-up? Consistent with previous 4,6-9,28 work, we expected that higher wealth accumulation would be associated with increased longevity. Second, was the wealth-longevity association present over and above controls for family and heritable factors that could confound this association? That is, was there evidence of a within- family association between wealth and longevity among siblings and twins in the same family? Alternatively, was the wealth-longevity association primarily driven by common factors at the family level, presumably involving early experiences and/or heritable factors? Methods The analyses in this cohort study were conducted between November 16, 2019, and May 18, 2021, and were preregistered on the Open Science Framework on November 15, 2019. The preregistration included hypotheses, decision rules for inclusion, rationale for covariates, and statistical plan. Deviations from the original analysis plan are detailed in eMethods 3 in the JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 (Reprinted) July 23, 2021 2/12 JAMA Health Forum | Original Investigation Association of Wealth With Longevity in US Adults at Midlife Supplement. We followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline. The institutional review boards of the University of Wisconsin-Madison and Harvard Medical School approved the Midlife in the United States (MIDUS) study procedures. Participants in the MIDUS study provided oral informed consent. The institutional review board approval and informed consent for the MIDUS study extend to the present study, which used publicly available MIDUS data, including all mortality data up to 2016. Participants and Analysis Sample The data used were obtained from the MIDUS study, an ongoing national study of health and aging that began in 1994. In the present study, we used data from wave 1 of the MIDUS study (MIDUS 1), which were collected in 1994 to 1996 from 7108 adults aged 20 to 75 years. These participants were recruited through a nationally representative random-digit dialing sampling strategy that included subsamples of siblings and twins. Participants who provided oral informed consent completed a telephone-assisted survey and a mailed self-administered questionnaire. Mortality data through October 31, 2018, were collected by the MIDUS study team at the University of Wisconsin-Madison. Of the 7108 cases from MIDUS 1, 7017 were considered for inclusion in the current analysis. We selected cases with complete data on mortality status, net worth, and relevant covariates (n = 5414). These cases included 2675 individuals (designated here as singletons) who did not have a sibling or co-twin in the MIDUS study, 1282 nontwin full siblings, 864 dizygotic (DZ) twins, and 593 MZ twins. Twins or siblings who had complete data and who were in a pair with a co-sibling or co-twin who were missing data were included in the full analytic sample. However, they were excluded from subsequent discordant twin or sibling analyses, resulting in 1214 nontwin full siblings, 740 DZ twins, and 536 MZ twins in the discordant twin or sibling analyses. The eMethods 1 in the Supplement provides details on the eligibility criteria and missing data. Measures At MIDUS 1, participants responded to the following question: “Suppose you (and your spouse or partner) cashed in all your checking and savings accounts, stocks and bonds, real estate, sold your home, your vehicles, and all your valuable possessions. Then suppose you put that money toward paying off your mortgage and all your other loans, debts, and credit cards. Would you have any money left over after paying your debts or would you still owe money?” Participants reported how much that amount would be, using binned response categories that specified ranges of dollar amounts. Amount of money owed was truncated at a value of $0. Net worth that exceeded $1 million was truncated at that value. Covariates included parental educational level and participant age, self-reported race/ethnicity (analyzed here as non-White vs White), sex (female vs male), history of cancer or heart disease as diagnosed by a medical doctor, and status of ever smoking cigarettes regularly or consuming alcohol regularly (the eMethods 1 in the Supplement provides coding details). At MIDUS 1, participants self- reported on their race. Response options included White, Black and/or African American, Native American or Aleutian Islander, Asian or Pacific Islander, multiracial, or other race. Because of the low number of individuals who identified as a race other than White, we dichotomized race categories into non-White vs White. In addition, data on self-identification as Hispanic or Latino were not available at MIDUS 1. Mortality follow-up was completed by the MIDUS study team at the University of Wisconsin- Madison. Date of death was obtained from various sources, including relative responses, other informant reports, newspaper or online obituaries, and the National Death Index (using the 15th day of the month, rather than the exact day of death, to maintain participant confidentiality). Survival time was the number of years between the date when the MIDUS 1 self-administered questionnaires were returned to the MIDUS study team (1994 to 1996) and the date of death; if the participant was alive, the censor date was October 31, 2018, which was the date of the MIDUS study team’s latest mortality update. JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 (Reprinted) July 23, 2021 3/12 JAMA Health Forum | Original Investigation Association of Wealth With Longevity in US Adults at Midlife Statistical Analysis A series of Cox proportional hazards regression models for survival analyses were estimated using Stata, version 16 (StataCorp LLC). In model 1, we tested the individual-level association between wealth and longevity by conducting survival analysis among all cases who had complete data on the analysis variables and by using robust SEs to account for dependence among family members. Next, we estimated survival models only within the subsample of twins and nontwin siblings. In model 2, siblings and twins were treated as individuals, but a shared frailty was included to estimate dependence among family members. In model 3, we estimated the within-family association between wealth and longevity by calculating the family-level mean net worth (among families that had 2 members with complete data) and subsequently calculating the difference between each individual’s net worth and their family’s mean net worth. This between-within method is a common 33,34 approach to fixed-effects modeling. When applied in this way, the between-within method allowed us to compare siblings or twins in the same family to one another and thus to control for all 13,14,35 unmeasured shared family-level variables consistent with the discordant sibling design. In survival analysis, the between-within method has been shown to provide similar estimates to more common approaches for co-sibling or co-twin control (eg, conditional likelihood methods like stratified Cox regression) and has been observed to be optimal statistically. To further disambiguate environmental vs genetic influences, we tested whether within-family associations between net worth and longevity varied across the nontwin sibling, MZ twin, and DZ twin subsamples. We included a pair of 2-way interaction terms crossing the net worth mean deviation scores with dummy codes for DZ twins or MZ twins. A Wald test was used to measure the equality of the within-family net worth coefficient across siblings, DZ twins, and MZ twins. We estimated separate survival models for nontwin siblings, DZ twins, and MZ twins (models 4-6). A significant within-family association (2-sided P < .05) that was observed among nontwin siblings but not among DZ or MZ twin pairs would suggest residual confounding by early life factors because twins share a closer prenatal and postnatal environment than nontwin siblings. A within-family association that was observed among both siblings and DZ twin pairs but not among MZ twin pairs would suggest genetic confounding. Sensitivity analyses addressed the skewed distribution of net worth and tested the possibility of nonlinear associations between net worth and longevity. Sensitivity analyses also clarified the role of preexisting health problems and considered other model specifications. Primary models were reestimated as stratified Cox regressions using Stata (eMethods 2 in the Supplement) and as multilevel Cox regressions using Mplus (Muthén & Muthén) (eTables 3-5 in the Supplement). A 2-sided P < .05 was considered statistically significant. Results A total of 5414 participants from MIDUS 1 were included in the full analysis sample. These participants had a mean (SD) age of 46.7 (12.7) years; 2766 were women (51.1%), 2648 were men (48.9%), and 4927 (91.0%) self-identified as White individuals. Participants had a mean (SD) net worth of $122 153 ($209 537; median, $32 500). Of these participants, 675 (12.5%) had been diagnosed with a heart problem and 381 (7%) had a cancer diagnosis. In addition, 2790 participants (51.5%) reported ever having smoked cigarettes regularly and 2301 participants (42.5%) reported having used alcohol regularly. Table 1 displays the descriptive statistics, by subsamples of singletons, nontwin siblings, DZ twins, and MZ twins. By the censor date (October 31, 2018), 1010 individuals (18.7%) in the full sample had died. Fifty percent of siblings and twins were in families whose members differed by less than or equal to $87 500 in net worth at MIDUS 1 (interquartile range=$212 500). JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 (Reprinted) July 23, 2021 4/12 JAMA Health Forum | Original Investigation Association of Wealth With Longevity in US Adults at Midlife Cohort Analyses Survival models were run as gamma shared frailty Cox regressions, stratified Cox regressions, and multilevel Cox regressions. Regardless of the model used, the estimates were similar (eMethods 2 and 3 in the Supplement). Thus, results from the shared frailty Cox regressions are described herein. Results from the full sample (model 1; n = 5414; 1010 deaths) are presented in Table 2. After covariate adjustment, net worth was inversely associated with mortality (HR, 0.95; 95% CI, 0.94-0.97; P < .001) such that the hazard was 5% lower for every additional $50 000 of net worth accumulated at midlife. In the subsample of siblings and twins (model 2; n = 2490; 421 deaths), the net worth HR was similar (HR, 0.95; 95% CI, 0.93-0.97; P < .001). Between-Within Models Similar to models 1 and 2, the between-within model among siblings and twins (model 3) suggested that net worth was inversely associated with mortality within families (HR, 0.94; 95% CI, 0.91-0.97; P = .001). These patterns suggested that a sibling or twin who had accumulated more net worth by midlife tended to live longer than their co-siblings or co-twin with less net worth. Figure 1 depicts survival curves that were split at values above and below 0.5 of an SD from the mean deviation net worth score (mean of 0.00). In Figure 1, the net worth–mortality association is conditional on the shared frailty, thus the lines represent the estimated survival of 2 family members whose net worth differed by 0.5 SD or $139 000 and who were approximately 47 years of age at MIDUS 1 (with all other variables held at their means). A difference of $139 000 in net worth was associated with a 13% relative decrease in the probability of death nearly 24 years later, favoring the family member with a higher net worth; given the low base rate of mortality in the sample, this decrease translated into more than a 1% absolute difference in survival. The interactions between sibling type (ie, nontwin siblings, DZ twins, and MZ twins) and net worth deviation score were not significant, and the Wald Table 1. Descriptive Statistics of the Analysis Variables No. (%) Nontwin Full sample Singleton full sibling DZ twins MZ twins Variable (n = 5414) (n = 2675) (n = 1214) (n = 740) (n = 536) Net worth, mean (SD), $ 122 153.02 111 425.23 160 577.84 97 760.81 131 624.06 (209 537.49) (202 341.67) (240 643.26) (167 631.37) (217 415.75) Age, mean (SD), y 46.7 (12.7) 46.2 (13.0) 49.3 (12.4) 45.9 (12.1) 44.7 (11.7) Sex Female 2766 (51.1) 1267 (47.4) 667 (54.9) 409 (55.3) 282 (52.6) Male 2648 (48.9) 1408 (52.6) 547 (45.1) 331 (44.7) 254 (47.4) Race White 4927 (91) 2334 (87.3) 1161 (95.6) 706 (95.4) 503 (93.8) Black or African 255 (4.7) 173 (6.5) 23 (1.9) 23 (3.1) 20 (3.7) American Other 232 (4.3) 168 (6.3) 30 (2.5) 11 (1.5) 13 (2.4) Heart disease 675 (12.5) 315 (11.8) 172 (14.2) 104 (14.1) 58 (10.8) Cancer 381 (7.0) 165 (6.2) 126 (10.4) 43 (5.8) 26 (4.9) Regular cigarette smoking 2790 (51.5) 1464 (54.7) 587 (48.4) 362 (48.9) 242 (45.1) Regular alcohol use 2301 (42.5) 1204 (45.0) 508 (41.8) 301 (40.7) 186 (34.7) Parental educational level <High school diploma 1275 (23.6) 744 (27.8) 182 (15.0) 184 (24.9) 106 (19.8) High school diploma 1937 (35.8) 983 (36.7) 433 (35.7) 252 (34.1) 176 (32.8) Some college, 2-y 873 (16.1) 368 (13.8) 229 (18.9) 110 (14.9) 118 (22.0) associate’s degree, or vocational school 4-y Bachelor’s degree 822 (15.2) 371 (13.9) 215 (17.7) 126 (17.0) 80 (14.9) Abbreviations: DZ, dizygotic; MZ, monozygotic. or some graduate school Master’s or professional 507 (9.4) 209 (7.8) 155 (12.8) 68 (9.2) 56 (10.4) Other race included multiracial, Native American or degree Aleutian Islander, Asian or Pacific Islander, and Other. JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 (Reprinted) July 23, 2021 5/12 JAMA Health Forum | Original Investigation Association of Wealth With Longevity in US Adults at Midlife Table 2. Cox Proportional Hazards Regression for Survival Models a b Model HR (95% CI) P value Model 1: full sample (n = 5414) Age 1.10 (1.09-1.11) <.001 Female sex 0.82 (0.72-0.94) .005 Non-White race 1.13 (0.87-1.46) .35 Parental education 0.97 (0.95-0.99) .03 Heart disease 1.87 (1.61-2.17) <.001 Cancer 1.44 (1.22-1.70) <.001 Cigarette smoking 1.75 (1.53-2.01) <.001 Alcohol use 1.04 (0.91-1.19) .47 Net worth 0.95 (0.94-0.97) <.001 Model 2: siblings and twins (n = 2490) Age 1.10 (1.09-1.12) <.001 Female sex 0.76 (0.61-0.95) .01 Non-White race 0.85 (0.48-1.52) .60 Parental education 0.97 (0.93-1.01) .17 Heart disease 1.93 (1.54-2.43) <.001 Cancer 1.55 (1.17-2.06) .002 Cigarette smoking 2.08 (1.66-2.61) <.001 Alcohol use 0.92 (0.73-1.15) .47 Net worth 0.95 (0.93-0.97) <.001 Model 3: siblings and twins (n = 2490) Age 1.10 (1.09-1.11) <.001 Female sex 0.76 (0.61-0.95) .01 Non-White race 0.87 (0.49-1.54) .64 Parental education 0.97 (0.93-1.01) .15 Heart disease 1.94 (1.54-2.43) <.001 Cancer 1.56 (1.18-2.07) .002 Cigarette smoking 2.09 (1.66-2.62) <.001 Alcohol use 0.91 (0.73-1.14) .43 Net worth Between family 0.96 (0.93-0.99) .01 Within family 0.94 (0.91-0.97) .001 Model 4: nontwin siblings (n = 1214) Age 1.10 (1.08-1.12) <.001 Female sex 0.64 (0.47-0.86) .004 Non-White race 1.51 (0.72-3.14) .27 Parental education 1.00 (0.94-1.06) .88 Heart disease 2.09 (1.52-2.86) <.001 Cancer 1.92 (1.35-2.72) <.001 Cigarette smoking 2.21 (1.61-3.04) <.001 Alcohol use 0.73 (0.53-1.00) .05 Net worth Between family 0.98 (0.94-1.02) .38 Within family 0.94 (0.90-0.97) .002 Model 5: DZ twins (n = 740) Age 1.12 (1.09-1.14) <.001 Female sex 0.88 (0.59-1.31) .53 Non-White race 0.25 (0.06-1.02) .05 Parental education 0.96 (0.90-1.03) .37 Heart disease 1.72 (1.14-2.57) .009 Cancer 1.01 (0.57-1.79) .96 (continued) JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 (Reprinted) July 23, 2021 6/12 JAMA Health Forum | Original Investigation Association of Wealth With Longevity in US Adults at Midlife Table 2. Cox Proportional Hazards Regression for Survival Models (continued) a b Model HR (95% CI) P value Cigarette smoking 2.08 (1.38-3.14) <.001 Alcohol use 1.18 (0.80-1.75) .39 Net worth Between family 0.90 (0.84-0.97) .009 Within family 0.94 (0.86-1.02) .19 Model 6: MZ twins (n = 536) Age 1.11 (1.08-1.14) <.001 Female sex 1.02 (0.60-1.73) .93 Non-White race 1.17 (0.33-4.14) .79 Abbreviations: DZ, dizygotic; HR, hazard ratio; Parental education 0.91 (0.83-1.01) .10 MZ, monozygotic. Heart disease 2.37 (1.27-4.42) .006 The models are described in the Statistical Analysis Cancer 1.53 (0.61-3.81) .35 subsection of the Methods section. Cigarette smoking 2.05 (1.19-3.54) .009 The HR for net worth of both between-family and within-family associations reflects a difference of Alcohol use 1.01 (0.59-1.71) .97 $50 000. Net worth Non-White race included Black and/or African Between family 0.96 (0.89-1.02) .25 American, Native American or Aleutian Islander, Within family 0.95 (0.87-1.04) .34 Asian or Pacific Islander, multiracial, and other. Figure 1. Within-Family Association Between Net Worth and Longevity 1.00 0.98 +0.5 SD net worth deviation 0.96 The survival curves represent 2 family members whose net worth differed by approximately $139 000 at –0.5 SD net worth deviation Midlife in the United States (MIDUS) study wave 1; this 0.94 amount corresponded to within-family net worth deviation scores that are ±0.5 SD from the mean 0.92 deviation score of 0.00. The survival curves were adjusted for family-level mean net worth as well as participant age at wave 1 of the MIDUS study, race/ 0.90 0 5 10 15 20 25 ethnicity, sex, history of cancer or heart disease, health Time after MIDUS study wave 1, y behaviors, and parental educational level. test indicated that differences in the within-family HRs across sibling types were no greater than chance (χ (2) = 0.05; P =.97). Separate survival models were estimated for nontwin siblings (model 4), DZ twins (model 5), and MZ twins (model 6). Consistent with the Wald test, the within-family net worth estimates for each of the 3 subsamples were similar (nontwin siblings: HR, 0.94 [95% CI, 0.90-0.97; P = .002]; DZ twins: HR, 0.94 [95% CI, 0.86-1.02; P = .19]; MZ twins: HR, 0.95 [95% CI, 0.87-1.04; P = .34]). These HRs were also similar in magnitude to the net worth HR observed in the individual-level analyses (models 1 and 2). Among nontwin siblings, the P value for the within-family net worth estimate was significant at P < .05, although the P values were not significant in the DZ twin and MZ twin subsamples, likely because of the substantial decrease in the sample size and the number of deaths within the DZ twin (n = 118 decedents) and MZ twin (n = 79 decedents) subsamples. Figure 2 depicts the HRs and 95% CIs for the net worth estimate in the full sample (model 1) and sibling and twin subsample (model 2), the within-family comparison in the sibling and twin subsample (model 3), and the within-sibling and within-twin comparisons (models 4, 5, and 6). JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 (Reprinted) July 23, 2021 7/12 Survival JAMA Health Forum | Original Investigation Association of Wealth With Longevity in US Adults at Midlife Sensitivity Analyses The eMethods 2 and eTables 1-5 in the Supplement present the results of the sensitivity analyses conducted in the combined sibling and twin subsample. Briefly, these analyses found that the observed wealth-longevity association (1) was not an artifact of the right-skewed distribution of net worth, (2) had a roughly linear shape for most of the sample, and (3) was not explained by lifestyle practices and health conditions at MIDUS 1. For example, when net worth was recoded into deciles, the within-family association between net worth and mortality remained statistically significant (HR, 0.92; 95% CI, 0.87-0.96; P = .001). In addition, spline models (eTable 1 in the Supplement) suggested that the HR for the net worth–mortality association was similar for approximately 90% of participants in the full sample. Moreover, when restricting the full sample to only the siblings and twin pairs who were free of previous cancer or heart disease, the within-family association between net worth and mortality remained statistically significant (HR, 0.94; 95% CI, 0.90-0.98; P = .01). The eMethods 2 in the Supplement provides a complete description of all sensitivity analyses. Discussion In contrast to most previous studies that examined associations between socioeconomic position and longevity using between-person analytic approaches, the present cohort study examined within-family associations between wealth and longevity— a more conservative test of the hypothesis because it implicitly controlled for all factors (eg, early experience and heritable characteristics) shared by siblings and twins. Findings from the within-family analyses suggested that wealth accumulation at midlife may be associated with longevity among adults in the United States. We believe that investigations such as the kind we conducted are important because of the near impossibility of performing an experimental study of the wealth-longevity association. Findings from this study also converged with the results from other studies that used different methods. For example, an analysis found that mortality risk among retired stockholders increased in the years after wealth shocks (ie, acute reductions in wealth) owing to exogenous stock market fluctuations. Across models, the associations observed between net worth and longevity were modest: we observed a 1% absolute difference in the probability of survival after nearly 24 years between family members who differed by approximately $139 000 in net worth at midlife. However, the MIDUS sample was relatively young (mean age of approximately 70 years) with relatively low mortality (18.7% of the full sample had died by the censor date). Thus, follow-up analyses are needed to ascertain whether the magnitude of these associations would change as mortality increases. The associations between net worth and longevity in the DZ and MZ twin pairs were similar in magnitude to the associations observed in nontwin siblings. However, when the analyses were separated into subsamples of siblings, DZ twins, and MZ twins, the within-family net worth associations among DZ twins and MZ twins were not statistically significant. This finding was likely the result of lower power and reduced precision in the twin subsamples, as reflected in the wide CIs around the within-family estimates. However, it could also suggest the presence of confounding by shared environmental features. Other studies have discussed the difficulty in distinguishing null Figure 2. Hazard Ratios (HRs) and 95% CIs for the Net Worth Estimate Across Survival Models Model No. and description HR (95% CI) P value 1. Full sample 0.95 (0.94-0.97) <.001 2. Siblings and twins 0.95 (0.93-0.97) <.001 3. Siblings and twins (within family comparison) 0.94 (0.91-0.97) .001 4. Nontwin siblings (within-sibling comparison) 0.94 (0.90-0.97) .002 In each model, the HR reflects the decrease in hazard 5. DZ twins (within twin comparison) 0.94 (0.85-1.02) .19 associated with a $50 000 increase in net worth. The 6. MZ twins (within-twin comparison) 0.95 (0.87-1.04) .34 squares represent the HR estimates, and the lines represent the 95% CIs. DZ indicates dizygotic; 0.85 0.90 0.95 1.00 1.05 1.10 HR (95% CI) MZ, monozygotic. JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 (Reprinted) July 23, 2021 8/12 JAMA Health Forum | Original Investigation Association of Wealth With Longevity in US Adults at Midlife associations from false-negative results in sibling study designs when power varies widely between subsamples. The findings of this study suggest that the association between wealth at midlife and longevity is unlikely to be a simple artifact of environmental and heritable characteristics shared by siblings and twins. Moreover, the findings suggest that policies to support individuals’ ability to accrue wealth and to achieve financial security in adulthood could have considerable health benefits. These findings should be interpreted through a broader societal lens. The US ranks first in economic inequality among high-income nations. Over the past 30 years, the wealth gap between the high-income and low-income people in the US has widened through policies and practices that have diverted a substantial and increasing share of wealth from the lower- and middle-income groups to the affluent 24,38 group. Such redistribution may have implications for longevity patterns in the coming decades. Policies to reduce the wealth gap, if implemented, could be expected to generate substantial returns to public health. Limitations This study has several limitations. First, although the discordant sibling design, as we have modeled it, reduced confounding by shared environmental and heritable characteristics, it cannot elucidate whether wealth itself is a causal actor or a marker of other nonshared factors (eg, self-regulatory capacity and cognitive ability) and nonshared experiences (eg, social, educational, and professional trajectories) that covary with and/or contribute to both wealth and health. Thus, a causal interpretation of these findings is not warranted. We did consider some possible nonshared confounders in adulthood that were related to lifestyle and disease, however. Sensitivity analyses restricted to participants in good midlife health suggested that these variables had a modest role in the wealth-longevity association. Second, at MIDUS 1, a single self-reported questionnaire item was used to measure net worth, which may have introduced error into estimates of net worth. In addition, most participants in the MIDUS 1 sample self-identified as White individuals; thus, the estimates we reported may be less generalizable to underrepresented racial/ethnic groups. This limitation is particularly relevant for present purposes given the substantial and persistent racial disparities in household wealth in the US. Third, although we observed evidence of within-family associations between wealth and longevity, the mechanisms involved were not identified. Wealth is likely to be a distal factor in longevity, acting through more proximal mechanisms (eg, stress and lifestyle) to affect biological processes involved in disease as well as access to medical care and other health-promoting resources. This interpretation is consistent with the way other studies have conceptualized 40,41 socioeconomic status as a fundamental cause of health disparities. From this perspective, having wealth allows individuals numerous opportunities to invest resources into the many proximal factors that promote health and longevity. Conclusions This cohort study found an association between wealth at midlife and longevity, and this association is unlikely to be merely an artifact of environmental and heritable characteristics shared by families. Policies that support individuals' ability to accrue wealth and achieve financial security could have considerable health benefits. In addition, policies to reduce the wealth gap may generate substantial returns to public health. JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 (Reprinted) July 23, 2021 9/12 JAMA Health Forum | Original Investigation Association of Wealth With Longevity in US Adults at Midlife ARTICLE INFORMATION Accepted for Publication: May 24, 2021. Published: July 23, 2021. doi:10.1001/jamahealthforum.2021.1652 Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2021 Finegood ED et al. JAMA Health Forum. Corresponding Author: Eric D. Finegood, PhD, Institute for Policy Research, Northwestern University, 1801 Maple Ave, Ste 2450, Evanston, IL 60201 (eric.finegood@northwestern.edu). Author Affiliations: Institute for Policy Research, Northwestern University, Evanston, Illinois (Finegood, Freedman, Chen, Miller); Department of Psychology, Northwestern University, Evanston, Illinois (Finegood, Freedman, Chen, Mroczek, Miller); Department of Psychology, University of Illinois Urbana-Champaign, Urbana (Briley); Department of Psychology, West Virginia University, Morgantown (Turiano); Department of Psychological Sciences, Purdue University, West Lafayette, Indiana (South); Department of Psychology, University of Minnesota, Minneapolis (Krueger); Department of Medical Social Sciences, Northwestern University, Evanston, Illinois (Mroczek). Author Contributions: Dr Finegood had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Concept and design: Finegood, Turiano, Krueger, Mroczek, Miller. Acquisition, analysis, or interpretation of data: All authors. Drafting of the manuscript: Finegood, Turiano, Mroczek, Miller. Critical revision of the manuscript for important intellectual content: Finegood, Briley, Turiano, Freedman, South, Krueger, Chen, Mroczek. Statistical analysis: Finegood, Briley, Turiano, Freedman, Mroczek. Obtained funding: Krueger, Mroczek. Administrative, technical, or material support: Miller. Supervision: Chen, Mroczek, Miller. Conflict of Interest Disclosures: Dr Krueger reported receiving grants from the National Institute on Aging (NIA) during the conduct of the study and outside the submitted work. Dr Miller reported receiving grants from the National Institutes of Health (NIH) outside the submitted work. No other disclosures were reported. Funding/Support: Dr Finegood was supported by grant F32HL146005 from the National Heart, Lung, and Blood Institute of the NIH. Dr Briley was supported by a Jacobs Foundation Research Fellowship. Dr Turiano was supported by the West Virginia Prevention Research Center and Cooperative Agreement No. 1-U48-DP-005004 from the Centers for Disease Control and Prevention. Dr Freedman was supported by grant F32HD100076 from the Eunice Kennedy Shriver National Institute of Child Health and Human Development of the NIH. Dr Krueger was supported by grants R01AG053217 and U19AG051426 from the NIH. Dr Mroczek was supported by grants R01 AG018436, R01 AG067622, RF1 AG064006, and U19AG051426 from the NIA and grant P30 AG05998 from the Claude D. Pepper Older Adults Independence Center. Dr Miller was supported by grant R01 MD011749 from the National Institute on Minority Health and Health Disparities and grant P50 DA051361 from the National Institute on Drug Abuse. Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication. Disclaimer: The content is solely the responsibility of the authors and does not represent the official views of the funders. Additional Information: The statistical code used in this study is available upon request to the first author. MIDUS 1 data are publicly available at the Inter-university Consortium for Political and Social Research (https://www.icpsr. umich.edu/web/ICPSR/series/203). As of this writing, the 2018 mortality update data are not publicly available. REFERENCES 1. Marmot M. Social determinants of health inequalities. Lancet. 2005;365(9464):1099-1104. doi:10.1016/S0140-6736(05)71146-6 2. Adler NE, Rehkopf DHUS. U.S. disparities in health: descriptions, causes, and mechanisms. Annu Rev Public Health. 2008;29:235-252. doi:10.1146/annurev.publhealth.29.020907.090852 3. Chetty R, Stepner M, Abraham S, et al. The association between income and life expectancy in the United States, 2001-2014. JAMA. 2016;315(16):1750-1766. doi:10.1001/jama.2016.4226 JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 (Reprinted) July 23, 2021 10/12 JAMA Health Forum | Original Investigation Association of Wealth With Longevity in US Adults at Midlife 4. Pollack CE, Chideya S, Cubbin C, Williams B, Dekker M, Braveman P. Should health studies measure wealth? A systematic review. Am J Prev Med. 2007;33(3):250-264. doi:10.1016/j.amepre.2007.04.033 5. Wolfe JD, Baker EH, Scarinci IC. Wealth and obesity among US adults entering midlife. Obesity (Silver Spring). 2019;27(12):2067-2075. doi:10.1002/oby.22625 6. Demakakos P, Biddulph JP, Bobak M, Marmot MG. Wealth and mortality at older ages: a prospective cohort study. J Epidemiol Community Health. 2016;70(4):346-353. doi:10.1136/jech-2015-206173 7. Daly MC, Duncan GJ, McDonough P, Williams DR. Optimal indicators of socioeconomic status for health research. Am J Public Health. 2002;92(7):1151-1157. doi:10.2105/AJPH.92.7.1151 8. Hajat A, Kaufman JS, Rose KM, Siddiqi A, Thomas JC. Long-term effects of wealth on mortality and self-rated health status. Am J Epidemiol. 2011;173(2):192-200. doi:10.1093/aje/kwq348 9. Pool LR, Burgard SA, Needham BL, Elliott MR, Langa KM, Mendes de Leon CF. Association of a negative wealth shock with all-cause mortality in middle-aged and older adults in the United States. JAMA. 2018;319(13): 1341-1350. doi:10.1001/jama.2018.2055 10. Hill WD, Davies NM, Ritchie SJ, et al. Genome-wide analysis identifies molecular systems and 149 genetic loci associated with income. Nat Commun. 2019;10(1):5741. doi:10.1038/s41467-019-13585-5 11. Donovan SJ, Susser E. Commentary: advent of sibling designs. Int J Epidemiol. 2011;40(2):345-349. doi:10.1093/ije/dyr057 12. Lawlor DA, Mishra GD, eds. Family Matters: Designing, Analysing and Understanding Family Based Studies in Life Course Epidemiology. Oxford University Press; 2009. doi:10.1093/acprof:oso/9780199231034.001.0001 13. Carlin JB, Gurrin LC, Sterne JAC, Morley R, Dwyer T. Regression models for twin studies: a critical review. Int J Epidemiol. 2005;34(5):1089-1099. doi:10.1093/ije/dyi153 14. Turkheimer E, Harden KP. Behavior genetic research methods: testing quasi-causal hypotheses using multivariate twin data. In: Reis HT, Judd CM, eds. Handbook of Research Methods in Social and Personality Psychology. 2nd ed. Cambridge University Press; 2014:159-187. 15. Krieger N, Chen JT, Coull BA, Selby JV. Lifetime socioeconomic position and twins’ health: an analysis of 308 pairs of United States women twins. PLoS Med. 2005;2(7):e162. doi:10.1371/journal.pmed.0020162 16. Ericsson M, Pedersen NL, Johansson ALV, Fors S, Dahl Aslan AK. Life-course socioeconomic differences and social mobility in preventable and non-preventable mortality: a study of Swedish twins. Int J Epidemiol. 2019;48 (5):1701-1709. doi:10.1093/ije/dyz042 17. Søndergaard G, Mortensen LH, Nybo Andersen AM, et al. Does shared family background influence the impact of educational differences on early mortality? Am J Epidemiol. 2012;176(8):675-683. doi:10.1093/aje/kws230 18. Fujiwara T, Kawachi I. Is education causally related to better health? a twin fixed-effect study in the USA. Int J Epidemiol. 2009;38(5):1310-1322. doi:10.1093/ije/dyp226 19. Næss O, Hoff DA, Lawlor D, Mortensen LH. Education and adult cause-specific mortality—examining the impact of family factors shared by 871 367 Norwegian siblings. Int J Epidemiol. 2012;41(6):1683-1691. doi:10.1093/ije/dys143 20. Lundborg P, Lyttkens CH, Nystedt P. The effect of schooling on mortality: new evidence from 50,000 Swedish twins. Demography. 2016;53(4):1135-1168. doi:10.1007/s13524-016-0489-3 21. Madsen M, Andersen AM, Christensen K, Andersen PK, Osler M. Does educational status impact adult mortality in Denmark? A twin approach. Am J Epidemiol. 2010;172(2):225-234. doi:10.1093/aje/kwq072 22. Madsen M, Andersen PK, Gerster M, Andersen AM, Christensen K, Osler M. Are the educational differences in incidence of cardiovascular disease explained by underlying familial factors? A twin study. Soc Sci Med. 2014;118: 182-190. doi:10.1016/j.socscimed.2014.04.016 23. Osler M, McGue M, Christensen K. Socioeconomic position and twins’ health: a life-course analysis of 1266 pairs of middle-aged Danish twins. Int J Epidemiol. 2007;36(1):77-83. doi:10.1093/ije/dyl266 24. World Inequality Lab. World Inequality Report 2018. Accessed July 27, 2020. https://wir2018.wid.world 25. Menasce Horowitz J, Igielnik R, Kochhar R. Most Americans say there is too much economic inequality in the U.S., but fewer than half call it a top priority. January 9, 2020. Accessed October 14, 2020. https://www. pewsocialtrends.org/2020/01/09/most-americans-say-there-is-too-much-economic-inequality-in-the-u-s- but-fewer-than-half-call-it-a-top-priority/ 26. Powell JH. Recent Economic Developments and the Challenges Ahead: Speech at the National Association for Business Economics Virtual Annual Meeting. October 6, 2020. Accessed December 28, 2020. https://www.federalreserve.gov/newsevents/speech/powell20201006a.htm JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 (Reprinted) July 23, 2021 11/12 JAMA Health Forum | Original Investigation Association of Wealth With Longevity in US Adults at Midlife 27. Cook FL, Weiss E. COVID-19 is hitting older workers like a sledgehammer—here’s what Congress can do. The Hill. May 1, 2020. Accessed June 9, 2020. https://thehill.com/opinion/finance/495541-covid-19-is-hitting-older- workers-like-a-sledge-hammer-heres-what-congress 28. Schwandt H. Wealth shocks and health outcomes: evidence from stock market fluctuations. Am Econ J Appl Econ. 2018;10(4):349-377. doi:10.1257/app.20140499 29. Parsing genetic and environmental influences on longevity: A twin and sibling study in MIDUS. Open Science Framework. Registered November 15, 2019. Accessed June 17, 2021. https://osf.io/zyedp 30. von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP; STROBE Initiative. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. PLoS Med. 2007;4(10):e296. doi:10.1371/journal.pmed.0040296 31. Brim OG, Ryff CD, Kessler RC eds. How Healthy Are We? A National Study of Well-Being at Midlife. University of Chicago Press; 2004. 32. StataCorp. Stata Statistical Software: Release 16. StataCorp LLC; 2019. 33. Allison PD. Fixed Effects Regression Models. Sage Publications, Inc; 2009. doi:10.4135/9781412993869 34. Sjölander A, Lichtenstein P, Larsson H, Pawitan Y. Between-within models for survival analysis. Stat Med. 2013;32(18):3067-3076. doi:10.1002/sim.5767 35. McGue M, Osler M, Christensen K. Causal inference and observational research: the utility of twins. Perspect Psychol Sci. 2010;5(5):546-556. doi:10.1177/1745691610383511 36. Muthén LK, Muthén BO. Mplus User’s Guide. 8 ed. Muthén & Muthén; 1998-2017. 37. Gilman SE, Loucks EB. Another casualty of sibling fixed-effects analysis of education and health: an informative null, or null information? Soc Sci Med. 2014;118:191-193. doi:10.1016/j.socscimed.2014.06.029 38. Stiglitz JE. The Price of Inequality: How Today’s Divided Society Endangers Our Future. W W Norton & Co; 2012. 39. Sykes BL, Maroto M. A wealth of inequalities: mass incarceration, employment, and racial disparities in U.S. household wealth, 1996 to 2011. Russell Sage Found J Soc Sci. 2016;2(6):129-152. doi:10.7758/rsf.2016.2.6.07 40. Phelan JC, Link BG, Diez-Roux A, Kawachi I, Levin B. “Fundamental causes” of social inequalities in mortality: a test of the theory. J Health Soc Behav. 2004;45(3):265-285. doi:10.1177/002214650404500303 41. Link BG, Phelan J. Social conditions as fundamental causes of disease. J Health Soc Behav. 1995;Spec No:80-94. doi:10.2307/2626958 SUPPLEMENT. eMethods 1. Full Methods Section eMethods 2. Supplementary Sensitivity Analyses eMethods 3. Deviations From Original Preregistered Analysis Plan eTable 1. Spline Model eTable 2. BW Model Among Siblings and Twins in Groups/Pairs Where All Family Members Have$382,500 in Net Worth at M1 eTable 3. Multi-Level Cox Regression Analysis in Full Analysis Sample eTable 4. Multi-Level Cox Regression Analysis in Sample of Siblings and Twins eTable 5. Multi-Level Cox Regression Analysis Across Sibling Subsamples eReferences. JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 (Reprinted) July 23, 2021 12/12 Supplementary Online Content Finegood ED, Briley DA, Turiano NA, et al. Association of wealth with longevity in US adults at midlife. JAMA Health Forum. 2021;2(7):e211652. doi:10.1001/jamahealthforum.2021.1652 eMethods 1. Full Methods Section eMethods 2. Supplementary Sensitivity Analyses eMethods 3. Deviations From Original Preregistered Analysis Plan eTable 1. Spline Model eTable 2. BW Model Among Siblings and Twins in Groups/Pairs Where All Family Members Have <= $382,500 in Net Worth at M1 eTable 3. Multi-Level Cox Regression Analysis in Full Analysis Sample eTable 4. Multi-Level Cox Regression Analysis in Sample of Siblings and Twins eTable 5. Multi-Level Cox Regression Analysis Across Sibling Subsamples eReferences. This supplementary material has been provided by the authors to give readers additional information about their work. © 2021 Finegood ED et al. JAMA Health Forum. eMethods 1. Full Methods Section This report adheres to the STROBE guidelines for cohort studies . The analyses presented here were conducted in 2020 2021 and preregistered on the Open Science Framework (https://osf.io/zyedp) on November 15, 2019. The pre registration includes hypotheses, decision rules for inclusion, rationale for covariates, and statistical plan. Deviations are detailed in eMethods 3 of this Supplement. Participants Data come from the Midlife in the United States (MIDUS) study, an ongoing national study of health and aging processes begun in 1994. In the current analysis, we utilized data from wave 1 of MIDUS (MIDUS I) collected from years 1994 1996. 7,108 adults aged 20 75 years participated at MIDUS I. They were recruited via a nationally representative random digit dialing sampling strategy, which included recruiting subsamples of siblings and twins. Those who provided oral informed consent to participate completed a telephone assisted survey as well as a mailed self administered questionnaire. The Institutional Review Boards at the University of Wisconsin and Harvard Medical School approved the MIDUS study procedures. Mortality data through October 2018 were collected by the MIDUS study team at the University of Wisconsin. Analysis sample Of the 7,108 cases participating at MIDUS 1, n = 3,715 were designated for purposes of the current analysis as ‘singletons’, meaning that they did not have a sibling or a co twin in MIDUS. 1,479 cases were non twin full siblings, meaning that they had at least one non twin full sibling in MIDUS. 1,914 were twins. Of these, n = 30 were missing zygosity information and © 2021 Finegood ED et al. JAMA Health Forum. were excluded from our analysis sample. Additionally, n = 61 did not have a co twin in MIDUS so were designated as singletons in our analysis. In families that had both a pair of MZ twins and a pair of DZ twins, only the MZ pair was retained in our analysis sample (19 DZ twins excluded). We chose to retain the MZ pair as opposed to the DZ pair because the number of MZ pairs was lower overall than the number of DZ pairs. In families that had either two pairs of MZ twins or two pairs of DZ twins, one twin pair was retained at random (42 cases excluded from analysis sample). Given all of these criteria, n = 7,017 cases were considered for inclusion in our analysis sample (3,776 singletons; 1,479 full siblings; 1,064 DZ twins; and 698 MZ twins). Survival analyses were conducted among those individuals with complete data on mortality status, net worth, and the covariates (n = 5,414 cases: 2,675 singletons, 1,282 non twin siblings, 864 DZ twins, 593 MZ twins). Twin or siblings who had complete data and who were in a pair with a co twin/sibling that was missing data were included in this analytic sample of 5,414 although they were excluded from subsequent discordant twin/sibling analyses, resulting in 1,214 non twin siblings, 740 DZs, and 536 MZs in the discordant twin/sibling analyses. Missing data. Survival analyses were conducted among those individuals with complete data on mortality status, net worth, and the covariates. The amount of missing data among the analysis variables collected in the phone interview was low (0% to 4% missing). Of the 7,017 cases, n = 6,240 (88.9%) had completed both the phone interview and the self administered questionnaire. Accordingly, missingness on the analysis variables collected in the self  administered questionnaire was higher (12% out of n = 7,017 were missing race/ethnicity data and 20.2% out of n = 7,017 were missing net worth data). In addition, 6 participants did not have mortality data as of October 31, 2018. Considering the missing data on all analysis © 2021 Finegood ED et al. JAMA Health Forum. variables, 5,414 cases had complete data (2,675 singletons, 1,282 non twin siblings, 864 DZ twins, 593 MZ twins). Compared to those included in the analysis sample (n = 5,414), those who were excluded from the analysis sample (n = 1,603) were more likely to be non white ( (1, n = 6171) = 28.96, p < .001), to have had a previous heart condition ( (1, n = 6997) = 4.62, p = .03), were older in age (t(2443.35) =  2.01, p = .04, and reported lower levels of parent education (t(2206.91) = 5.60, p < .001). There were no associations between inclusion status and sex, previous cancer diagnosis, smoking status, and regular alcohol consumption. Because our primary research question concerned within sibling/twin associations between net worth and mortality risk, these sociodemographic and health related factors related to inclusion status were considered less problematic. Measures Net worth. At MIDUS 1, participants responded to the following prompt: “Suppose you (and your spouse or partner) cashed in all your checking and savings accounts, stocks and bonds, real estate, sold your home, your vehicles, and all your valuable possessions. Then suppose you put that money toward paying off your mortgage and all your other loans, debts, and credit cards. Would you have any money left over after paying your debts or would you still owe money?” Participants reported how much that amount would be using binned response categories that specified ranges of dollars. Those who would still owe money were truncated at a value of $0. Individuals whose net worth exceeded $1,000,000 were truncated at that value. This truncation prior to MIDUS 1 data release was due to privacy and human subjects concerns. However, the MIDUS study does provide the percentage that had negative net worth, and this was 13%. On the other end, those with net worth over $1 million comprised only 2% of the © 2021 Finegood ED et al. JAMA Health Forum. sample. These bottom  and top truncations, like any truncation, reduce the size of correlation and regression coefficients thereby creating an underestimate. Such a conservative bias was deemed acceptable. Covariates. Survival models included several covariates selected a priori because of their known associations with wealth and health. These included participant age, self reported race/ethnicity (analyzed here as non white vs. white), sex (female vs. male), and self reported history of cancer or heart disease, as diagnosed by a medical doctor. We also controlled for ever having smoked regularly or consumed alcohol regularly. Highest level of parents’ education (an indicator of childhood socioeconomic status), was also included as a covariate. In the case that siblings/twins differed in their report of the highest level of parents’ education, the higher of the discrepant responses was used for all siblings/twins within the same family. All cause mortality. Mortality follow up was completed by the MIDUS study team at the University of Wisconsin, Madison. Date of death was obtained from various sources, including responses from relatives, other informant reports, newspaper or online obituaries, th and the National Death Index (using the 15 day of the month, rather than the exact day of death, to maintain participant confidentiality). Survival time was the number of years between the date when the MIDUS 1 self administered questionnaires were returned to the MIDUS study team (years 1995 96) and the date of death (otherwise, if alive, the censor date of Oct 31, 2018, which was the date of the MIDUS study team’s latest mortality update). Analysis plan A series of Cox survival analyses were estimated using Stata 16 . To test the individual  level association between wealth and longevity, we first conducted analyses among all cases © 2021 Finegood ED et al. JAMA Health Forum. that had complete data on analysis variables utilizing robust SEs to account for dependence among family members (Model 1). Next, we estimated survival models only within the sub sample of twins and non twin siblings. In Model 2, siblings and twins were treated “as individuals” but a shared frailty was included to model dependence among family members. In Model 3, we estimated the within  family association between wealth and longevity by calculating the family level average net worth (among families that had >=2 members with complete data) and subsequently calculating the difference between each individual’s net worth and their family average. When included alongside the family level mean, the hazard ratio (HR) for these mean deviation scores estimate the within family association between net worth and longevity. This “between within” 4,5 (BW) method is a common approach to fixed effects modeling . When applied in this way, it allows us to compare siblings/twins in the same family to one another, and thus to control for all unmeasured shared family level variables, consistent with the discordant sibling/twin 6–8 design . In survival analysis, the BW method has been shown to provide similar estimates to more common approaches for co twin/sibling control with survival data (e.g. conditional likelihood methods like stratified Cox regression) and has been observed to be optimal statistically . To further disambiguate environmental versus genetic influences, we tested whether within family associations between net worth and longevity varied across non twin sibling, MZ, and DZ subsamples. We did this by including a pair of two way interaction terms crossing the mean deviation scores of net worth with dummy codes for DZs or MZs; non twin full siblings modeled as reference group). A Wald test tested the equality of the within family net worth © 2021 Finegood ED et al. JAMA Health Forum. coefficient across siblings, DZ, and MZ subsamples. Lastly, we estimated separate survival models for non twin siblings, DZ pairs, and MZ pairs (Models 4 6). A significant within family association (p < 0.05) observed among non twin siblings  but not among DZ or MZ pairs  would suggest some residual confounding by early life factors since twins share a closer pre  and post  natal environment than non twin siblings who may be born years apart. A within family association observed both among siblings and DZ pairs but not among MZ pairs would suggest genetic confounding. Sensitivity analyses were undertaken to test the robustness of findings. These analyses addressed the skewed distribution of the net worth variable and tested the possibility of non linear associations between net worth and longevity. They also clarified the role of pre existing health problems and considered other model specifications. Primary models were also re estimated as stratified Cox regressions using STATA (see eMethods 2) and as multilevel Cox regressions using Mplus (see eMethods 3). © 2021 Finegood ED et al. JAMA Health Forum. eMethods 2. Supplementary Sensitivity Analyses Sensitivity analyses were undertaken in the combined sibling and twin subsample to test the robustness of the findings to different model specifications. First, to assess the potential for a non linear association between net worth and longevity, we estimated a spline model (eTable th th 1) including two knots: one at the 75 percentile of net worth ($125,000) and one at the 90 th percentile ($382,500). As shown in eTable 1, HRs for net worth for those below the 75 th th percentile (HR=.86, CI=.76 .97, p=.01) and between the 75  90 percentiles (HR=.90, CI=.82  .98, p=.01) were similar, and a subsequent test of the estimates confirmed that their difference th was no greater than chance (p=.66). The HR for net worth for those above the 90 percentile (individuals with >=$680,000; n=182, 45 decedents) was not statistically significant (HR=1.02, 95%CI=.97 1.08, p=.31) and a test of the estimates indicated that the difference between this th th HR and the HR for those between the 75  90 percentiles was statistically significant (p=.03). This indicates a possible diminished return on net worth at the very high end of the net worth distribution, though this reflects only approximately 7% of the sample. Collectively, the spline model indicates that among the large majority (93%) of siblings and twins (i.e. those whose net worth was <= $382,000), the association between net worth and survival was approximately linear. We subsequently reran Model 3 in a subsample of siblings and twins who were in family groups where all members had <= $382,500 in net worth (n=2,110; 321 deaths). As shown in eTable 2 of this supplement, the HR for net worth in this restricted sample (HR=0.89, CI=0.82  0.96, p=0.004) suggested a larger association between net worth and longevity among those © 2021 Finegood ED et al. JAMA Health Forum. with lower family level wealth. However, these results should be interpreted with a high level of caution due to the restricted sample size. In another sensitivity analysis, we recoded net worth into ordinal decile groups, given the large positive skew of the net worth distribution. In the combined sibling and twin subsample, between family (HR = 0.90, 95% CI = 0.85 0.94, p < .001) and within family (HR = 0.92, 95% CI = 0.87 0.96, p = .001) net worth estimates remained significant predictors of mortality. Next, to account for the possibility of residual confounding by health status (having a medical problem may both reduce one’s ability to accumulate wealth and increase mortality risk), analyses were re estimated among sibling/twin pairs who were free of previous cancer or heart disease. Among sibling groups with >2 members, only those siblings without heart disease and cancer were compared to one another. Results were largely similar in this restricted sample (n = 1,740; 196 deaths): HR = 0.95, 95% CI= 0.90  0.99, p = 0.04, and HR = 0.94, 95% between within CI = 0.90 0.98, p = 0.01). We also tested the possibility of a nonlinear age trend by including an age term (HR =1.00, 95% CI = 0.99 1.00, p = .65) and also an age*sex interaction term (HR = 1.01, 95% CI = 0.99 1.03, p = 0.14), neither of which was associated with mortality risk nor changed the interpretations of other model estimates. We also tested an interaction between the within  family net worth estimate and participant age at MIDUS 1. The rationale being that the within  family association between net worth and longevity may vary as a function of age. The interaction term was not statistically significant, HR = 1.00, 95% CI =0.99 1.00, p=0.37, © 2021 Finegood ED et al. JAMA Health Forum. suggesting that the within family association between net worth and longevity did not vary by age at MIDUS 1. Lastly, as a more conservative test of possible confounding by early experience, we restricted the analysis sample to only same sex sibling groups/twin pairs. Point estimates of between family (HR = .94, 95% CI = 0.90 0.98 p = .007) and within family (HR = .96, 95% CI = 0.92 1.01, p = .16) net worth estimates were consistent with estimates observed in Model 3 in the main text, although the p value for the within family estimate was not statistically significant, likely due to the substantial reduction in power and sample size in this restricted sample (N =2,490; n=421 dead; vs. N = 1,359; n=221 dead). Given the large loss in Model3 Same sex power due to the reduction in sample size and number of deaths, the result of this sensitivity analysis should be interpreted with a high level of caution. Stratified Cox Regression models. We re ran Models 3 6 as stratified Cox regressions, stratifying by Family ID. The within family net worth estimates were as follows: Model 3 HR = .94, 95%CI = .90 .97, p = .002; Model 4 HR = .93, 95%CI = .89 .97, p = .004; Model 5 HR = .95, 95%CI = .87 1.05, p = .37; Model 6 HR = .95, 95%CI = .84 1.07, p = .42. © 2021 Finegood ED et al. JAMA Health Forum. eMethods 3. Deviations From Original Preregistered Analysis Plan The analyses presented in the main text are consistent with the study rationale, hypotheses, and overall analytic plan outlined in our preregistration (https://osf.io/zyedp). We did, however, deviate from the original analytic plan in some minor ways—these deviations are outlined below. First, in the original analysis plan, we specified that survival analyses would be estimated as multi level Cox regression models. Instead, each survival model was estimated as a Cox model with shared frailty term to account for clustering. Both analytic approaches gave very similar results for all models—compare coefficients in Model 1 and Model 3 of Table 2 in the main text to estimates in eTables 3 and 4 (results from multilevel Cox regressions run in Mplus version 8 ). Second, for the sibling/twin comparison analysis (step 2 of the original analysis plan), we originally planned to run a two level mixture analysis using a Cox regression model specifying the KNOWNCLASS option in Mplus version 8 in order to estimate a multiple group analysis— allowing model coefficients to vary across subsamples of non twin siblings, DZ twins, MZ twins. Instead, in the main text, we present results from three separate survival models: one among non twin siblings, one among DZ twins, and one among MZ twins. With either analytic strategy, coefficients were very similar and provided the same interpretations—compare coefficients from Models 4, 5, and 6 in main text Table 2 to coefficients in eTable 5, which display results from the multilevel mixture model using Cox regression with KNOWNCLASS option in Mplus version 8). © 2021 Finegood ED et al. JAMA Health Forum. Third, as described in the main text, we undertook a conservative test of whether the within family net worth effect varied across full sibling, MZ twin, and DZ twin subsamples by including two way interaction terms between the within family net worth variable and dummy codes for cluster type and conducting a Wald test to test the equality of the within family estimates. This test of the equality of within family net worth effects across sibling subsamples was not described in the original analysis plan. We also originally proposed an exploratory test of the two way interaction between parent education and the within family net worth twin/sibling difference score. We decided to omit this test from the main text because we felt that the rationale for this test was tangential to our primary analyses, which were already many in number. It is also not an optimal test for answering questions related to social mobility, its stated purpose in the preregistration. In any case, we report results from this test here. Among the combined sample of non twin siblings, DZ twins, and MZ twins (n =2,490) there was no interaction between parents’ highest level of education and sibling/twin net worth difference score (HR = .99, 95% CI .98  1.00, p = .30). Indeed, when the sample is split into subsamples of those at or below (n = 1,333) and above (n = 1,157) the median on parent education, the within family net worth estimate was significant in both subsamples (at/below median: HR =.95, 95% CI .91 .99, p = .04; above median: within HR =.94, 95% CI .89 .98, p = .006). within We also specified that full information maximum likelihood estimation (FIML) would be used to handle missing data. In the preregistration, we specified the criteria by which siblings and twin pairs would be excluded from our analysis sample—resulting in an analytic sample of n=7,017. As described in eMethods 1 of the Supplement, of the n=7,017, n=6,240 had © 2021 Finegood ED et al. JAMA Health Forum. completed both the phone interview and the self administered questionnaire. Another n = 6 did not have mortality data, resulting in a possible analytic sample of n=6,234. Of the analysis variables, net worth had the most missingness: n=640 out of 6,234 (10.3%) missing net worth, likely because some participants were unwilling or unable to provide this information. Because our primary interest was to compare mortality risk within sibling groups/twin pairs who were discordant on net worth, only siblings groups/twin pairs in which discordance could be estimated (e.g. twin pairs in which both twins had non missing net worth data) were useful analytically. Thus, analyses were conducted only among cases that had complete data and FIML was not used. © 2021 Finegood ED et al. JAMA Health Forum. eTable 1. Spline Model HR 95% CI p Age 1.11 1.09 1.12 < .001 Female 0.74 0.60 0.93 .009 Non white 0.82 0.46 1.46 0.51 Parent education 0.97 0.93 1.01 0.18 Heart disease 1.96 1.56 2.47 < .001 Cancer 1.57 1.18 2.07 0.002 Smoking 2.07 1.65 2.60 < .001 Alcohol use 0.92 0.74 1.15 0.50 Net worth spline 1 0.86 0.76 0.97 0.01 Net worth spline 2 0.90 0.82 0.98 0.01 Net worth spline 3 1.02 0.97 1.08 0.31 Note: N=2,490, which includes 421 deaths. HR=hazard ratio, 95% CI= 95% confidence interval of the hazard ratio. © 2021 Finegood ED et al. JAMA Health Forum. eTable 2. BW Model Among Siblings and Twins in Groups/Pairs Where All Family Members Have <= $382,500 in Net Worth at M1 HR 95% CI p Age 1.11 1.09 1.12 < 0.001 Female 0.74 0.58 0.95 0.01 Non white 0.67 0.37 1.23 0.20 Parent education 0.97 0.92 1.02 0.27 Heart disease 2.04 1.58 2.63 < 0.001 Cancer 1.56 1.14 2.14 0.005 Smoking 1.87 1.45 2.41 < 0.001 Alcohol use 1.04 0.81 1.34 0.71 Net worth (between family) 0.84 0.78 0.90 < 0.001 Net worth (within family) 0.89 0.82 0.96 0.004 Note: N=2,110, which includes 321 deaths. HR=hazard ratio, 95% CI= 95% confidence interval of the hazard ratio. © 2021 Finegood ED et al. JAMA Health Forum. eTable 3. Multi-Level Cox Regression Analysis in Full Analysis Sample HR (95% CI) p value Age 1.10 (1.09 1.11) < 0.001 Female 0.82 (0.72 0.94) 0.005 Non white 1.13 (0.87 1.46) 0.35 Parent education 0.97 (0.95 0.99) 0.03 Heart disease 1.87 (1.61 2.17) < 0.001 Cancer 1.44 (1.22 1.70) < 0.001 Smoking 1.75 (1.53 2.00) < 0.001 Alcohol use 1.04 (0.91 1.19) 0.47 Net worth 0.95 (0.94 0.97) < 0.001 © 2021 Finegood ED et al. JAMA Health Forum. eTable 4. Multi-Level Cox Regression Analysis in Sample of Siblings and Twins HR (95% CI) p value Age 1.10 (1.09 1.11) < 0.001 Female 0.78 (0.64 0.96) 0.01 Non white 0.85 (0.48 1.50) 0.58 Parent education 0.97 (0.93 1.01) 0.25 Heart disease 1.85 (1.46 2.35) < 0.001 Cancer 1.53 (1.18 1.99) 0.001 Smoking 2.03 (1.63 2.52) < 0.001 Alcohol use 0.92 (0.74 1.14) 0.46 Net worth (between family) 0.96 (0.93 0.99) 0.01 Net worth (within family) 0.95 (0.92 0.98) 0.002 © 2021 Finegood ED et al. JAMA Health Forum. eTable 5. Multi-Level Cox Regression Analysis Across Sibling Subsamples Siblings (n = 1,214) DZ twins (n = 740) MZ twins (n = 536) HR (95% CI) p value HR (95% CI) p value HR (95% CI) p value Age 1.10 (1.08 1.12) < 0.001 1.12 (1.09 1.14) < 0.001 1.11 (1.08 1.15) < 0.001 Female 0.64 (0.47 0.87) 0.005 0.88 (0.60 1.29) 0.51 1.03 (0.58 1.83) 0.89 Non white 1.50 (0.66 3.42) 0.32 0.25 (0.05 1.09) 0.06 1.16 (0.31 4.29) 0.81 Parent edu. 1.00 (0.94 1.06) 0.87 0.96 (0.90 1.03) 0.38 0.91 (0.82 1.02) 0.11 Heart disease 2.07 (1.51 2.86) < 0.001 1.72 (1.09 2.69) 0.01 2.39 (1.10 5.21) 0.02 Cancer 1.92 (1.37 2.70) < 0.001 1.01 (0.57 1.77) 0.96 1.55 (0.58 4.14) 0.38 Smoking 2.21 (1.63 2.98) < 0.001 2.08 (1.39 3.11) < 0.001 2.11 (1.17 3.80) 0.01 Alcohol use 0.73 (0.52 1.01) 0.06 1.18 (0.81 1.72) 0.36 1.04 (0.61 1.76) 0.86 Net worth (between) 0.98 (0.93 1.02) 0.41 0.90 (0.84 0.97) 0.005 0.95 (0.89 1.03) 0.26 Net worth (within) 0.94 (0.90 0.98) 0.005 0.94 (0.86 1.02) 0.19 0.95 (0.90 1.01) 0.16 © 2021 Finegood ED et al. JAMA Health Forum. eReferences 1. von Elm E, Altman DG, Egger M, et al. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement: Guidelines for Reporting Observational Studies. PLoS Med. 2007;4(10):e296. doi:10.1371/journal.pmed.0040296 2. Brim OG, Ryff CD, Kessler RC, eds. How Healthy Are We? A National Study of Well Being at Midlife. Chicago, IL: University of Chicago Press; 2004. 3. StataCorp. Stata Statistical Software: Release 16. StataCorp LLC; 2019. 4. Allison PD. Fixed Effects Regression Models. Sage Publications, Inc.; 2009. 5. Sjölander A, Lichtenstein P, Larsson H, Pawitan Y. Between within models for survival analysis. Stat Med. 2013;32(18):3067 3076. doi:10.1002/sim.5767 6. Carlin JB, Gurrin LC, Sterne JAC, Morley R, Dwyer T. Regression models for twin studies: A critical review. Int J Epidemiol. 2005;34(5):1089 1099. doi:https://doi.org/10.1093/ije/dyi153 7. Turkheimer E, Harden KP. Behavior Genetic Research Methods: Testing Quasi Causal Hypotheses Using Multivariate Twin Data. In: Reis HT, Judd CM, eds. Handbook of Research Methods in Social and Personality Psychology. 2nd ed. Cambridge University Press; 2014:159 187. 8. McGue M, Osler M, Christensen K. Causal inference and observational research: The utility of twins. Perspect Psychol Sci. 2010;5(5):546 556. doi:10.1177/1745691610383511 9. Muthén LK, Muthén BO. Mplus User’s Guide. Eighth Ed. Los Angeles, CA: Muthén & Muthén. 1998 2017. © 2021 Finegood ED et al. JAMA Health Forum.

Journal

JAMA Health Forum – American Medical Association

Published: Jul 23, 2021

There are no references for this article.