Assessing the degree of residual confounding: a cohort study on the association between disability pension and mortality

Assessing the degree of residual confounding: a cohort study on the association between... Abstract Background Previous studies showed that disability pensioners have a higher risk of premature death than others, but residual confounding has been suggested. The aim was to assess the degree of residual confounding of the association between disability pension (DP) and risk of premature death. Methods Prospective cohort study of everyone aged 19–64 years, living in Sweden 31 December 2004 (n = 5 406 469), followed up through 2010. Mortality hazard rates over time were estimated for three groups; incident disability pensioners during 2005 from start of DP (February–December 2005), prevalent disability pensioners (January 2005 or since before), and individuals not on DP in January 2005, after standardizing populations to characteristics of the incident disability pensioners, stratified by previous hospitalization or not. If DP has no immediate effect on mortality, incident disability pensioners and those not on DP should initially have similar hazard rates, thereby, allowing assessment of the degree of residual confounding. Results For those not previously hospitalized, the mortality hazard rate on the first DP day was: 3.07 (95% CI 2.21, 4.36), 2.09 (1.78, 2.48) and 0.78 (0.73, 0.84) per thousand person-years for incident, prevalent, and non DP, respectively. Among previously hospitalized these figures were: 21.67 (17.73, 26.24), 17.00 (15.76, 18.51) and 18.88 (18.14, 19.64) respectively. Hazard ratios were 1.15 (0.94, 1.40) in the strata with and 3.94 (2.78, 5.57) in the strata without, previous hospitalization comparing incident DP with non-DP. Conclusions Substantial residual confounding was found in the association between DP and premature death among those not previously hospitalized. Introduction Previous studies have shown that disability pensioners have a higher risk of premature death than individuals not on disability pension (DP), even after adjusting for variables such as age, sex, and other potential confounders including proxies for the underlying morbidity.1–15 It has been discussed whether the higher risk should be attributed to being on DP or to unadjusted underlying differences between disability pensioners and other individuals. A potentially strong confounder is the individual’s morbidity. One way to try to capture this potential confounder is to adjust for number of days of previous hospitalization. It has been argued that since the association does not change much after adjusting for hospitalization, ‘DP may per se contain a damaging factor in addition to the underlying disease’.6 Others, however, have postulated that there likely is substantial residual confounding, including confounding by indication, and that this issue could only be adequately addressed by randomized trials, which however would generally be difficult and possibly unethical.1,14 Although it might not be possible to adjust for all confounding without randomized trials, it is possible to investigate the degree of confounding at the start of follow-up. If any association is observed already on the first day, it could mainly be attributed to residual confounding. It has been pointed out that if DP is associated with premature death, this should not impact mortality substantially in the first day, week or month.15 Ideally, if all confounding was adjusted for, mortality should be similar at the start of DP. The putative harmful effects of being granted DP should raise mortality rates gradually over time after the DP granting. The aim was to assess the degree of residual confounding in the association between DP and premature death by investigating whether the excess risk of premature death is present already from the start of follow-up. Methods A population-based prospective cohort study utilizing data from four nationwide Swedish registers from three authorities was conducted. Data were linked at the individual level using the ten-digit unique identification number of all residents in Sweden. The entire population aged 19–64 years and living in Sweden on 31 December 2004 (n = 5 406 469) was included. This cohort was followed up from 1 January 2005 until 31 December 2010 using data from (i) the LISA database of Statistics Sweden; information on age, sex, birth country, educational level, type of living area for December 2004 and year of emigration up through 2010; (ii) the Board of Health and Welfare’s Cause of Death Register (date of death); (iii) and information about hospitalization from the National Patient Register (2000–04), and (iv) the National Social Insurance Agency’s MiDAS database, regarding all on DP at 1 January 2005 and dates regarding all incident DP granted February–December 2005. DP is always granted from the first day of a month. Comparing three populations We compared three populations. The first population consisted of incident disability pensioners that for the first time were granted DP sometime during February–December 2005. This population was followed up from the first day of DP. The second population comprised the prevalent disability pensioners, individuals that already were on DP on 1 January 2005, for many granted long before. The third population consisted of individuals that were not on DP on 1 January 2005. The latter two populations were followed up from 1 January 2005. All analyses were stratified by whether individuals had or had not been hospitalized for at least 1 day in the 5-year period 2000–04, excluding pregnancies without complication (ICD-10 code O80).16 We censored observations at the year after emigration or the end of follow-up. If there was residual confounding at the start of the DP, such censoring would be informative and introduce further bias.17 Although methods to correct for the bias introduced by informative censoring are available, these tend to fail in the presence of strong selection bias.18 Still, not censoring those in the non-DP group granted DP after 2005, may also lead to bias, particularly the latter part of the curve for the non-DP group may be overestimated. However, this choice will not be of importance as the focus of this study was the hazard rate on day 1. Regarding date of death, for 2219 individuals, deaths information on the exact day was missing and for 1024 deaths the exact month was missing. For these interval-censored deaths, the day of death were randomly imputed from a uniform distribution. Standardization to create hypothetical populations To make all three populations comparable, we standardized them to the same distribution of background variables measured at 31 December 2004: age, sex (man/woman), birth place (Sweden, other Nordic countries, other countries within EU25, and countries in the rest of the world), educational level (low (elementary school, ≤9 years), medium (upper secondary school, 10–12 years) and high (college/university, >12 years)), married (yes/no), and type of living area (urban, semi-urban and sparsely populated) and days of hospitalization in 2000–04 in five categories (0, 1–2, 3–6, 7–15, ≥16 days). We chose to standardize the non-DP population and the prevalent DP population to create two hypothetical populations with similar distribution of background variables as the incident disability pensioners. The standardization was performed by estimating probability weights for each of the non-DP population and the prevalent DP population. The weights for standardizing the non-DP population to the covariate distribution of the incident DP population were obtained by logistic regression, estimating the odds of being in the incident DP population relative to the non-DP population in an augmented dataset containing the incident DP population and the non-DP population. The predicted odds were then rescaled to have a mean of one and used as probability weights in subsequent analyses. The weights for standardizing the prevalent DP population to the covariate pattern of the incident DP population were obtained analogously. All variables were included as categorical variables in the logistic regression model, except age which was used as a restricted cubic spline function with three knots placed at the observed quartiles.19 An interaction between sex and age was included. Interactions of previous hospitalization with age and sex as well as a three-way interaction were also included in the strata with previous hospitalizations. Based on the Akaike Information Criterion (AIC),20 these models fitted the data better than simpler models without all interactions. No large probability weights were observed in any analysis, with the largest weight over all analyses being 13.3. Hazard rates across time We estimated the hazard rates for death over the follow-up time in each of the three populations and in the two hypothetical populations. We used flexible parametric models for survival analyses without covariates. We estimated smooth curves separately for each population by using the ‘flexsurv’ package in R.21,22 To estimate the hazard in the two hypothetical populations, we used the weights as described earlier. To avoid forcing the hazard rate to be 0 at time 0, we used a model that constrained the shape parameter of the assumed Weibull distribution to be equal to 1. We placed the boundary knots at the time of the first and last event and three internal knots at years 1, 2 and 4, respectively. We present hazard rates with 95% CIs at day 1, at 2, 4 and 6 years, respectively, in the figures and tables. We also present crude incidence rates for the first 30 days with exact 95% CI, to confirm that the flexible parametric model works as intended. In Sweden, the ordinary old-age retirement age is 65 years. People aged 19–64 and not on old-age pension can be granted DP if they have long-standing or permanent work incapacity caused by disease or injury. In 2005 there was no limit to duration of a sick-leave spell before being assessed for DP. Although we estimated hazards across time, our primary interest was to compare the three populations after standardization at the start of follow-up, because this is where the confounding adjustment will be based on the most recent information. After the first day, the information on the confounding variables was incomplete. If no immediate effect is expected, comparing the incident disability pensioners with those not on DP, at the start of follow-up, will enable an assessment of the degree of confounding at that time point. Furthermore, comparing the incident disability pensioners with the prevalent disability pensioners enable us to assess whether the excess risk of premature death among disability pensioners is already present at the start of follow-up or if it develops later. A 95% CI for this estimate was calculated from the two individual CIs using 1 million bootstrap samples. Results Background characteristics and size of the three populations in each of the two strata are presented in tables 1 and 2. Of all individuals, 21% had at least one inpatient day in the 5-year period 2000–04. These proportions were 44% among those with prevalent DP, 46% among those with incident DP, and 19% in the population with no DP in 2005. In the strata without previous hospitalization, the two DP populations were more likely to be women, older, born outside Sweden, less educated, married and to live in a rural area. In individuals with previous hospitalizations, these differences were less pronounced, although the two DP populations were older, less educated and had more days of previous hospitalization. Table 1 Background characteristics for the three populations; incident disability pensioners, prevalent disability pensioners and those not on disability pension (DP) at 31 December 2004, without hospitalization in the 5 years 2000–04 Not on DP Prevalent DP Incident DP n = 3 946 691 n = 299 029 n = 31 735 n % n % n % Sex     Women 1 807 645 45.8 182 490 61.0 20 087 63.3     Men 2 139 046 54.2 116 539 39.0 11 648 36.7 Age categories     19–24 years 521 087 13.2 7298 2.4 952 3.0     25–34 years 901 951 22.9 14 951 5.0 2502 7.9     35–44 years 972 805 24.6 41 635 13.9 6455 20.3     45–54 years 849 428 21.5 80 463 26.9 9174 28.9     55–64 years 701 420 17.8 154 682 51.7 12 652 39.9 Birth place     Sweden 3 369 294 85.4 238 387 79.7 25 433 80.1     Other Nordic 123 678 3.1 19 316 6.5 1556 4.9     Other EU25 92 399 2.3 10 697 3.6 892 2.8     Rest of the world 361 320 9.2 30 629 10.2 3854 12.1 Educational level     Low 628 539 15.9 119 928 40.1 8613 27.1     Medium 1 940 640 49.2 135 650 45.4 16 204 51.1     High 1 377 512 34.9 43 451 14.5 6918 21.8 Married     No 2 295 249 58.2 163 945 54.8 15 843 49.9     Yes 1 651 442 41.8 135 084 45.2 15 892 50.1 Type of living area     Urban 1 514 667 38.4 95 694 32.0 10 283 31.7     Semi-urban 1 387 758 35.2 105 718 35.4 11 098 34.2     Sparsely populated 1 044 266 26.5 97 617 32.6 11 098 34.2 Not on DP Prevalent DP Incident DP n = 3 946 691 n = 299 029 n = 31 735 n % n % n % Sex     Women 1 807 645 45.8 182 490 61.0 20 087 63.3     Men 2 139 046 54.2 116 539 39.0 11 648 36.7 Age categories     19–24 years 521 087 13.2 7298 2.4 952 3.0     25–34 years 901 951 22.9 14 951 5.0 2502 7.9     35–44 years 972 805 24.6 41 635 13.9 6455 20.3     45–54 years 849 428 21.5 80 463 26.9 9174 28.9     55–64 years 701 420 17.8 154 682 51.7 12 652 39.9 Birth place     Sweden 3 369 294 85.4 238 387 79.7 25 433 80.1     Other Nordic 123 678 3.1 19 316 6.5 1556 4.9     Other EU25 92 399 2.3 10 697 3.6 892 2.8     Rest of the world 361 320 9.2 30 629 10.2 3854 12.1 Educational level     Low 628 539 15.9 119 928 40.1 8613 27.1     Medium 1 940 640 49.2 135 650 45.4 16 204 51.1     High 1 377 512 34.9 43 451 14.5 6918 21.8 Married     No 2 295 249 58.2 163 945 54.8 15 843 49.9     Yes 1 651 442 41.8 135 084 45.2 15 892 50.1 Type of living area     Urban 1 514 667 38.4 95 694 32.0 10 283 31.7     Semi-urban 1 387 758 35.2 105 718 35.4 11 098 34.2     Sparsely populated 1 044 266 26.5 97 617 32.6 11 098 34.2 Table 1 Background characteristics for the three populations; incident disability pensioners, prevalent disability pensioners and those not on disability pension (DP) at 31 December 2004, without hospitalization in the 5 years 2000–04 Not on DP Prevalent DP Incident DP n = 3 946 691 n = 299 029 n = 31 735 n % n % n % Sex     Women 1 807 645 45.8 182 490 61.0 20 087 63.3     Men 2 139 046 54.2 116 539 39.0 11 648 36.7 Age categories     19–24 years 521 087 13.2 7298 2.4 952 3.0     25–34 years 901 951 22.9 14 951 5.0 2502 7.9     35–44 years 972 805 24.6 41 635 13.9 6455 20.3     45–54 years 849 428 21.5 80 463 26.9 9174 28.9     55–64 years 701 420 17.8 154 682 51.7 12 652 39.9 Birth place     Sweden 3 369 294 85.4 238 387 79.7 25 433 80.1     Other Nordic 123 678 3.1 19 316 6.5 1556 4.9     Other EU25 92 399 2.3 10 697 3.6 892 2.8     Rest of the world 361 320 9.2 30 629 10.2 3854 12.1 Educational level     Low 628 539 15.9 119 928 40.1 8613 27.1     Medium 1 940 640 49.2 135 650 45.4 16 204 51.1     High 1 377 512 34.9 43 451 14.5 6918 21.8 Married     No 2 295 249 58.2 163 945 54.8 15 843 49.9     Yes 1 651 442 41.8 135 084 45.2 15 892 50.1 Type of living area     Urban 1 514 667 38.4 95 694 32.0 10 283 31.7     Semi-urban 1 387 758 35.2 105 718 35.4 11 098 34.2     Sparsely populated 1 044 266 26.5 97 617 32.6 11 098 34.2 Not on DP Prevalent DP Incident DP n = 3 946 691 n = 299 029 n = 31 735 n % n % n % Sex     Women 1 807 645 45.8 182 490 61.0 20 087 63.3     Men 2 139 046 54.2 116 539 39.0 11 648 36.7 Age categories     19–24 years 521 087 13.2 7298 2.4 952 3.0     25–34 years 901 951 22.9 14 951 5.0 2502 7.9     35–44 years 972 805 24.6 41 635 13.9 6455 20.3     45–54 years 849 428 21.5 80 463 26.9 9174 28.9     55–64 years 701 420 17.8 154 682 51.7 12 652 39.9 Birth place     Sweden 3 369 294 85.4 238 387 79.7 25 433 80.1     Other Nordic 123 678 3.1 19 316 6.5 1556 4.9     Other EU25 92 399 2.3 10 697 3.6 892 2.8     Rest of the world 361 320 9.2 30 629 10.2 3854 12.1 Educational level     Low 628 539 15.9 119 928 40.1 8613 27.1     Medium 1 940 640 49.2 135 650 45.4 16 204 51.1     High 1 377 512 34.9 43 451 14.5 6918 21.8 Married     No 2 295 249 58.2 163 945 54.8 15 843 49.9     Yes 1 651 442 41.8 135 084 45.2 15 892 50.1 Type of living area     Urban 1 514 667 38.4 95 694 32.0 10 283 31.7     Semi-urban 1 387 758 35.2 105 718 35.4 11 098 34.2     Sparsely populated 1 044 266 26.5 97 617 32.6 11 098 34.2 Table 2 Background characteristics for the three populations; incident disability pensioners, prevalent disability pensioners and those not on DP at 31 December 2004, who had at least 1 day with hospitalization in the 5 years 2000–04 Not on DP Prevalent DP Incident DP n=927 225 n=233 524 n=26 771 n % n % n % Sex     Women 537 672 58.0 135 062 57.8 15 477 57.8     Men 389 553 42.0 98 462 42.2 11 294 42.2 Age categories (years)     19–24 96 899 10.5 5077 2.2 919 3.4     25–34 227 976 24.6 13 414 5.7 2525 9.4     35–44 219 926 23.7 32 451 13.9 4875 18.2     45–54 181 373 19.6 58 659 25.1 7068 26.4     55–64 201 051 21.7 123 923 53.1 11 384 42.5 Birth place     Sweden 788 422 85.0 190 594 81.6 21 830 81.5     Other Nordic 30 654 3.3 15 999 6.9 1426 5.3     Other EU25 16 931 1.8 6398 2.7 654 2.4     Rest of the world 91 218 9.8 20 533 8.8 2861 10.7 Educational level     Low 172 367 18.6 90 100 38.6 7704 28.8     Medium 465 761 50.2 109 936 47.1 13 719 51.2     High 289 097 31.2 33 488 14.3 5348 20.0 Married     No 508 394 54.8 135 129 57.9 14 000 52.3     Yes 418 831 45.2 98 395 42.1 12 771 47.7 Type of living area     Urban 335 394 36.2 73 967 31.7 8682 32.4     Semi-urban 329 156 35.5 82 831 35.5 9368 35.0     Sparsely populated 262 675 28.3 76 726 32.9 8721 32.6 Hospitalization (days)     1–2 394 970 42.6 56 480 24.2 6847 25.6     3–6 306 251 33.0 54 339 23.3 6397 23.9     7–15 150 377 16.2 49 253 21.1 5883 22.0     16 or more 75 627 8.2 73 452 31.5 7644 28.6 Not on DP Prevalent DP Incident DP n=927 225 n=233 524 n=26 771 n % n % n % Sex     Women 537 672 58.0 135 062 57.8 15 477 57.8     Men 389 553 42.0 98 462 42.2 11 294 42.2 Age categories (years)     19–24 96 899 10.5 5077 2.2 919 3.4     25–34 227 976 24.6 13 414 5.7 2525 9.4     35–44 219 926 23.7 32 451 13.9 4875 18.2     45–54 181 373 19.6 58 659 25.1 7068 26.4     55–64 201 051 21.7 123 923 53.1 11 384 42.5 Birth place     Sweden 788 422 85.0 190 594 81.6 21 830 81.5     Other Nordic 30 654 3.3 15 999 6.9 1426 5.3     Other EU25 16 931 1.8 6398 2.7 654 2.4     Rest of the world 91 218 9.8 20 533 8.8 2861 10.7 Educational level     Low 172 367 18.6 90 100 38.6 7704 28.8     Medium 465 761 50.2 109 936 47.1 13 719 51.2     High 289 097 31.2 33 488 14.3 5348 20.0 Married     No 508 394 54.8 135 129 57.9 14 000 52.3     Yes 418 831 45.2 98 395 42.1 12 771 47.7 Type of living area     Urban 335 394 36.2 73 967 31.7 8682 32.4     Semi-urban 329 156 35.5 82 831 35.5 9368 35.0     Sparsely populated 262 675 28.3 76 726 32.9 8721 32.6 Hospitalization (days)     1–2 394 970 42.6 56 480 24.2 6847 25.6     3–6 306 251 33.0 54 339 23.3 6397 23.9     7–15 150 377 16.2 49 253 21.1 5883 22.0     16 or more 75 627 8.2 73 452 31.5 7644 28.6 Table 2 Background characteristics for the three populations; incident disability pensioners, prevalent disability pensioners and those not on DP at 31 December 2004, who had at least 1 day with hospitalization in the 5 years 2000–04 Not on DP Prevalent DP Incident DP n=927 225 n=233 524 n=26 771 n % n % n % Sex     Women 537 672 58.0 135 062 57.8 15 477 57.8     Men 389 553 42.0 98 462 42.2 11 294 42.2 Age categories (years)     19–24 96 899 10.5 5077 2.2 919 3.4     25–34 227 976 24.6 13 414 5.7 2525 9.4     35–44 219 926 23.7 32 451 13.9 4875 18.2     45–54 181 373 19.6 58 659 25.1 7068 26.4     55–64 201 051 21.7 123 923 53.1 11 384 42.5 Birth place     Sweden 788 422 85.0 190 594 81.6 21 830 81.5     Other Nordic 30 654 3.3 15 999 6.9 1426 5.3     Other EU25 16 931 1.8 6398 2.7 654 2.4     Rest of the world 91 218 9.8 20 533 8.8 2861 10.7 Educational level     Low 172 367 18.6 90 100 38.6 7704 28.8     Medium 465 761 50.2 109 936 47.1 13 719 51.2     High 289 097 31.2 33 488 14.3 5348 20.0 Married     No 508 394 54.8 135 129 57.9 14 000 52.3     Yes 418 831 45.2 98 395 42.1 12 771 47.7 Type of living area     Urban 335 394 36.2 73 967 31.7 8682 32.4     Semi-urban 329 156 35.5 82 831 35.5 9368 35.0     Sparsely populated 262 675 28.3 76 726 32.9 8721 32.6 Hospitalization (days)     1–2 394 970 42.6 56 480 24.2 6847 25.6     3–6 306 251 33.0 54 339 23.3 6397 23.9     7–15 150 377 16.2 49 253 21.1 5883 22.0     16 or more 75 627 8.2 73 452 31.5 7644 28.6 Not on DP Prevalent DP Incident DP n=927 225 n=233 524 n=26 771 n % n % n % Sex     Women 537 672 58.0 135 062 57.8 15 477 57.8     Men 389 553 42.0 98 462 42.2 11 294 42.2 Age categories (years)     19–24 96 899 10.5 5077 2.2 919 3.4     25–34 227 976 24.6 13 414 5.7 2525 9.4     35–44 219 926 23.7 32 451 13.9 4875 18.2     45–54 181 373 19.6 58 659 25.1 7068 26.4     55–64 201 051 21.7 123 923 53.1 11 384 42.5 Birth place     Sweden 788 422 85.0 190 594 81.6 21 830 81.5     Other Nordic 30 654 3.3 15 999 6.9 1426 5.3     Other EU25 16 931 1.8 6398 2.7 654 2.4     Rest of the world 91 218 9.8 20 533 8.8 2861 10.7 Educational level     Low 172 367 18.6 90 100 38.6 7704 28.8     Medium 465 761 50.2 109 936 47.1 13 719 51.2     High 289 097 31.2 33 488 14.3 5348 20.0 Married     No 508 394 54.8 135 129 57.9 14 000 52.3     Yes 418 831 45.2 98 395 42.1 12 771 47.7 Type of living area     Urban 335 394 36.2 73 967 31.7 8682 32.4     Semi-urban 329 156 35.5 82 831 35.5 9368 35.0     Sparsely populated 262 675 28.3 76 726 32.9 8721 32.6 Hospitalization (days)     1–2 394 970 42.6 56 480 24.2 6847 25.6     3–6 306 251 33.0 54 339 23.3 6397 23.9     7–15 150 377 16.2 49 253 21.1 5883 22.0     16 or more 75 627 8.2 73 452 31.5 7644 28.6 In individuals without previous hospitalization (figure 1 and table 3), the hazard rates of premature death were relatively low in absolute terms, but increased substantially in relative terms over time for all populations. Before standardization the prevalent DP population had the highest risk, followed by the incident DP population, and those not on DP having the lowest risk. After standardization, the observed differences were smaller. The hazard rates for the incident DP population and the prevalent DP population were after standardization essentially at the same level. Table 3 Mortality hazard rates and 95% CIs across time and number of deaths and number of person-years by population for individuals without previous hospitalization, unstandardized and standardized n Deaths Time at risk (person years) Day 1 2 years 4 years 6 years Not DP 3 946 691 33 674 23 290 943 0.52 (0.48, 0.57) 1.27 (1.25, 1.30) 1.71 (1.68, 1.73) 2.03 (1.98, 2.07) Not DP, standardized to incident DP 3 946 691 33 674 23 290 943 0.78 (0.73, 0.84) 2.18 (2.15, 2.21) 2.99 (2.95, 3.02) 3.64 (3.58, 3.70) Prevalent DP 299 029 11 757 1756964 2.72 (2.35, 3.12) 5.96 (5.78, 6.16) 7.73 (7.56, 7.91) 9.21 (8.87, 9.58) Prevalent DP, standardized to incident DP 299 029 11 757 1 756 964 2.09 (1.78, 2.48) 4.58 (4.42, 4.73) 5.94 (5.78, 6.10) 6.95 (6.67, 7.22) Incident DP 31 735 848 174 188 3.07 (2.21, 4.36) 4.63 (4.15, 5.22) 5.35 (4.93, 5.85) 6.30 (5.35, 7.33) n Deaths Time at risk (person years) Day 1 2 years 4 years 6 years Not DP 3 946 691 33 674 23 290 943 0.52 (0.48, 0.57) 1.27 (1.25, 1.30) 1.71 (1.68, 1.73) 2.03 (1.98, 2.07) Not DP, standardized to incident DP 3 946 691 33 674 23 290 943 0.78 (0.73, 0.84) 2.18 (2.15, 2.21) 2.99 (2.95, 3.02) 3.64 (3.58, 3.70) Prevalent DP 299 029 11 757 1756964 2.72 (2.35, 3.12) 5.96 (5.78, 6.16) 7.73 (7.56, 7.91) 9.21 (8.87, 9.58) Prevalent DP, standardized to incident DP 299 029 11 757 1 756 964 2.09 (1.78, 2.48) 4.58 (4.42, 4.73) 5.94 (5.78, 6.10) 6.95 (6.67, 7.22) Incident DP 31 735 848 174 188 3.07 (2.21, 4.36) 4.63 (4.15, 5.22) 5.35 (4.93, 5.85) 6.30 (5.35, 7.33) Hazards are multiplied by 1000 with the timescale in years. Table 3 Mortality hazard rates and 95% CIs across time and number of deaths and number of person-years by population for individuals without previous hospitalization, unstandardized and standardized n Deaths Time at risk (person years) Day 1 2 years 4 years 6 years Not DP 3 946 691 33 674 23 290 943 0.52 (0.48, 0.57) 1.27 (1.25, 1.30) 1.71 (1.68, 1.73) 2.03 (1.98, 2.07) Not DP, standardized to incident DP 3 946 691 33 674 23 290 943 0.78 (0.73, 0.84) 2.18 (2.15, 2.21) 2.99 (2.95, 3.02) 3.64 (3.58, 3.70) Prevalent DP 299 029 11 757 1756964 2.72 (2.35, 3.12) 5.96 (5.78, 6.16) 7.73 (7.56, 7.91) 9.21 (8.87, 9.58) Prevalent DP, standardized to incident DP 299 029 11 757 1 756 964 2.09 (1.78, 2.48) 4.58 (4.42, 4.73) 5.94 (5.78, 6.10) 6.95 (6.67, 7.22) Incident DP 31 735 848 174 188 3.07 (2.21, 4.36) 4.63 (4.15, 5.22) 5.35 (4.93, 5.85) 6.30 (5.35, 7.33) n Deaths Time at risk (person years) Day 1 2 years 4 years 6 years Not DP 3 946 691 33 674 23 290 943 0.52 (0.48, 0.57) 1.27 (1.25, 1.30) 1.71 (1.68, 1.73) 2.03 (1.98, 2.07) Not DP, standardized to incident DP 3 946 691 33 674 23 290 943 0.78 (0.73, 0.84) 2.18 (2.15, 2.21) 2.99 (2.95, 3.02) 3.64 (3.58, 3.70) Prevalent DP 299 029 11 757 1756964 2.72 (2.35, 3.12) 5.96 (5.78, 6.16) 7.73 (7.56, 7.91) 9.21 (8.87, 9.58) Prevalent DP, standardized to incident DP 299 029 11 757 1 756 964 2.09 (1.78, 2.48) 4.58 (4.42, 4.73) 5.94 (5.78, 6.10) 6.95 (6.67, 7.22) Incident DP 31 735 848 174 188 3.07 (2.21, 4.36) 4.63 (4.15, 5.22) 5.35 (4.93, 5.85) 6.30 (5.35, 7.33) Hazards are multiplied by 1000 with the timescale in years. Figure 1 View largeDownload slide Mortality hazard rates and 95% CI across time for the incident disability pensioners, the prevalent disability pensioners, and those not on DP, without previous hospitalization. Hazards are multiplied by 1000 with the timescale in years Figure 1 View largeDownload slide Mortality hazard rates and 95% CI across time for the incident disability pensioners, the prevalent disability pensioners, and those not on DP, without previous hospitalization. Hazards are multiplied by 1000 with the timescale in years Overall, the hazard rates were much higher in the strata with previous hospitalization (figure 2 and table 4). The standardized and unstandardized prevalent DP population and the unstandardized non-DP population showed a similar pattern with relatively constant hazard rates over time. Comparing the incident DP population with the standardized non-DP population, the two hazard curves at the start of follow-up were very similar. The two hazard rates were not significantly different from each other on the first day. Table 4 Mortality hazard rates and 95% CIs across time and number of deaths and number of person-years by population for individuals with previous hospitalization, unstandardized and standardized n Deaths Time at risk (person years) Day 1 2 years 4 years 6 years Not DP 927 225 22 023 5 461 229 5.57 (5.22, 5.97) 3.78 (3.70, 3.86) 3.95 (3.88, 4.03) 4.13 (4.01, 4.24) Not DP, standardized to incident DP 927 225 22 023 5 461 229 18.88 (18.14, 19.64) 8.92 (8.79, 9.04) 8.63 (8.52, 8.74) 8.64 (8.47, 8.81) Prevalent DP 233 524 24 644 1 321 318 20.91 (19.41, 22.34) 18.27 (17.92, 18.63) 18.36 (18.05, 18.68) 18.81 (18.29, 19.31) Prevalent DP, standardized to incident DP 233 524 24 644 1 321 318 17.00 (15.76, 18.51) 14.73 (14.41, 15.08) 14.87 (14.59, 15.15) 15.39 (14.92, 15.86) Incident DP 26 771 2036 142 536 21.67 (17.73, 26.24) 14.42 (13.44, 15.42) 12.36 (11.56, 13.19) 11.91 (10.51, 13.32) n Deaths Time at risk (person years) Day 1 2 years 4 years 6 years Not DP 927 225 22 023 5 461 229 5.57 (5.22, 5.97) 3.78 (3.70, 3.86) 3.95 (3.88, 4.03) 4.13 (4.01, 4.24) Not DP, standardized to incident DP 927 225 22 023 5 461 229 18.88 (18.14, 19.64) 8.92 (8.79, 9.04) 8.63 (8.52, 8.74) 8.64 (8.47, 8.81) Prevalent DP 233 524 24 644 1 321 318 20.91 (19.41, 22.34) 18.27 (17.92, 18.63) 18.36 (18.05, 18.68) 18.81 (18.29, 19.31) Prevalent DP, standardized to incident DP 233 524 24 644 1 321 318 17.00 (15.76, 18.51) 14.73 (14.41, 15.08) 14.87 (14.59, 15.15) 15.39 (14.92, 15.86) Incident DP 26 771 2036 142 536 21.67 (17.73, 26.24) 14.42 (13.44, 15.42) 12.36 (11.56, 13.19) 11.91 (10.51, 13.32) Hazards are multiplied by 1000 with the timescale in years. Table 4 Mortality hazard rates and 95% CIs across time and number of deaths and number of person-years by population for individuals with previous hospitalization, unstandardized and standardized n Deaths Time at risk (person years) Day 1 2 years 4 years 6 years Not DP 927 225 22 023 5 461 229 5.57 (5.22, 5.97) 3.78 (3.70, 3.86) 3.95 (3.88, 4.03) 4.13 (4.01, 4.24) Not DP, standardized to incident DP 927 225 22 023 5 461 229 18.88 (18.14, 19.64) 8.92 (8.79, 9.04) 8.63 (8.52, 8.74) 8.64 (8.47, 8.81) Prevalent DP 233 524 24 644 1 321 318 20.91 (19.41, 22.34) 18.27 (17.92, 18.63) 18.36 (18.05, 18.68) 18.81 (18.29, 19.31) Prevalent DP, standardized to incident DP 233 524 24 644 1 321 318 17.00 (15.76, 18.51) 14.73 (14.41, 15.08) 14.87 (14.59, 15.15) 15.39 (14.92, 15.86) Incident DP 26 771 2036 142 536 21.67 (17.73, 26.24) 14.42 (13.44, 15.42) 12.36 (11.56, 13.19) 11.91 (10.51, 13.32) n Deaths Time at risk (person years) Day 1 2 years 4 years 6 years Not DP 927 225 22 023 5 461 229 5.57 (5.22, 5.97) 3.78 (3.70, 3.86) 3.95 (3.88, 4.03) 4.13 (4.01, 4.24) Not DP, standardized to incident DP 927 225 22 023 5 461 229 18.88 (18.14, 19.64) 8.92 (8.79, 9.04) 8.63 (8.52, 8.74) 8.64 (8.47, 8.81) Prevalent DP 233 524 24 644 1 321 318 20.91 (19.41, 22.34) 18.27 (17.92, 18.63) 18.36 (18.05, 18.68) 18.81 (18.29, 19.31) Prevalent DP, standardized to incident DP 233 524 24 644 1 321 318 17.00 (15.76, 18.51) 14.73 (14.41, 15.08) 14.87 (14.59, 15.15) 15.39 (14.92, 15.86) Incident DP 26 771 2036 142 536 21.67 (17.73, 26.24) 14.42 (13.44, 15.42) 12.36 (11.56, 13.19) 11.91 (10.51, 13.32) Hazards are multiplied by 1000 with the timescale in years. Figure 2 View largeDownload slide Mortality hazard rates and 95% CI across time for the incident disability pensioners, the prevalent disability pensioners, and those not on DP, with previous hospitalization. Hazards are multiplied by 1000 with the timescale in years Figure 2 View largeDownload slide Mortality hazard rates and 95% CI across time for the incident disability pensioners, the prevalent disability pensioners, and those not on DP, with previous hospitalization. Hazards are multiplied by 1000 with the timescale in years Supplementary table S1a and b present crude incidence rates for the first 30 days, giving similar results to the unstandardized hazard rates on day 1, but with wider CIs, consistently with the results from the parametric model. If being granted DP has no immediate effect on mortality, incident disability pensioners and those not on DP should have similar risk initially. This allows us to estimate the degree of residual confounding. Estimates for the degree of residual confounding in terms of hazard ratios are 21.67/18.88 = 1.15 (95% CI 0.94, 1.40) in the strata with previous hospitalization and 3.07/0.78 = 3.94 (2.78, 5.57) in the strata without previous hospitalization. By comparing incident and prevalent disability pensioners after standardization at the start of follow-up we could estimate how much of the excess risk that was present already then. In the strata without previous hospitalization the hazard ratio was 3.07/2.07 = 1.47 (1.01, 2.14) and in the strata with previous hospitalization the hazard ratio was 21.67/17.00 = 1.27 (1.03, 1.58), which meant that the excess risk of premature death was present already when granted DP and was even higher for incident DPs than for prevalent DPs. Discussion In this study, we investigated the association of being on DP with mortality, considering that if any immediate effect is small or nonexistent, residual confounding is likely important for explaining previous findings of DP being associated with premature death. We found that the excess risk of premature death among disability pensioners is present already when being granted DP. The large, what could be assumed as, residual confounding observed at baseline in those without previous hospitalization, which constitute 79% of the total population, may explain findings from previous studies. The study that first pointed out that adjusting for days of hospitalization did not impact the results, had a very long follow-up of 18 years.[6] We expect that while previous hospitalizations measured at baseline might be a good proxy for morbidity at the start of follow-up, they may not be as good a proxy for morbidity 18 years later. Our finding that incident disability pensioners had a higher risk of premature death than prevalent disability pensioners at the start of follow-up is consistent with a previous study that compared prevalent disability pensioners with those not on DP and found no variation across time regarding the three times higher risk of premature death among those on DP, except for a somewhat higher risk during the first year after being granted DP.5 However, this previous study did not attempt to adjust for morbidity. We limited the comparison to the start of follow-up because this is where confounding adjustment for hospitalization will be most accurate, considering that we did not have updated information on hospitalizations after baseline. Disability pensioners were more similar to other disability pensioners and so the degree of confounding for comparing incident and prevalent disability pensioners was expected to be smaller than when comparing disability pensioners with those not on DP. However, this difference between prevalent and incident disability pensioners might also be accounted for by possible cohort effects. Strengths and limitations To the best of our knowledge, this is the first study attempting to assess the degree of residual confounding in the association of being granted DP and the risk of premature death. We presented a method for assessing the degree of residual confounding at the start of follow-up for exposures that were not expected to have an immediate effect on the outcome. Because our study was based on a large population-based cohort covering a whole country, we had sufficient power to compare incident disability pensioners, prevalent disability pensioners, and those not on DP at the start of follow-up. We used standardization to adjust for confounding. This approach to assessing the degree of residual confounding can be useful when trying to improve the confounding adjustment in future studies as a check whether confounding adjustment is inadequate. Standardization adjusts for confounding by balancing the distribution of measured confounders, similarly to randomization, except randomization also balances unmeasured confounders. The reason why we used standardization instead of regression models for the outcome was that it allowed us to use a simple model for the outcome, whose only assumption was that the hazard rate changed smoothly over time. This facilitated interpretation of our findings. In this study we measured morbidity in terms of previous hospitalization. Other morbidity measures could also be used, such as out-patient visits, medication, sick-leave patterns, self-rated health, as days of hospitalization only captures a limited range of morbidity. Another main limitation of our study is that we did not have information on duration and grade of previous and ongoing sick leave. Based on the regulations in Sweden, after 12 months on sick leave, people should be assessed regarding fulfilling the criteria for DP. In 2005, many people on long-term sick leave had not been assessed for DP, due to practices of the National Social Insurance Agency. If they had been assessed for DP earlier, they could already have been on DP for years, and could have been included in the prevalent DP population in this study. In future studies, this group needs to be included. The Organization for Economic Co-operation and Development (OECD) tackled this problem regarding DP data in Sweden by categorizing all sick-leave spells that have been on-going for more than 2 years as being DP cases.23 Moreover, including information on diagnoses might add to the understanding of the processes studied here. In a previous study24 we investigated sick leave and mortality followed from day 32 of a sick-leave spell. There we found that the higher mortality among those with such sick leave was temporary and had vanished after 6 years. The adjusted hazard ratio comparing DP to non-DP standardized to the total population on sick-leave day 32 was 13.58/0.90 = 15.09. The OECD definition lies in between the definition of DP used in this paper and the 32 days of sick leave used in our previous study. This indicates that including sick leave into the definition of a DP as OECD does might increase the problems with residual confounding. However, this needs to be investigated in future studies. These studies should also include additional measures of morbidity, before and after being granted DP, as well as information on previous and ongoing sick leave. Theoretically, there could be detrimental factors under long-term sick leave before day 1 of DP, contributing to a higher risk of premature death of disability pensioners or to a higher risk of premature death among those on long-term sick leave not yet granted DP. Our analysis does not find support for this hypothesis, since in the stratum with previous hospitalization, where we have most of the deaths; no excess risk at the start of DP was found after standardization. However, this should be investigated in more detail in future studies. Further, people with acute, very severe, and fatal diseases usually are not granted DP, rather they are allowed to retain sick-leave benefits, which usually are higher than the DP benefits. We did not differentiate between DP diagnoses or whether people were granted a full- or a part-time DP. Such analyses would be recommended in future studies, especially regarding whether the associations differ between people on part-time and those on full-time DP. These are not limitations that would change the overall result, but it may be possible to locate more exactly in which subgroups residual confounding is more or less pronounced. The method for assessing residual confounding presented in this study would be of use in such future studies to assess if the confounding adjustment is inadequate. This method may also be useful more generally in studies where the exposure is not expected to have an immediate effect on the outcome. Funding This study was financially supported by the Swedish Research Council of Health, Working Life and Welfare (2007-1762). Conflicts of interest: None declared. Key points Previous studies have shown that disability pensioners have higher risks of premature death. Whether this is due to residual confounding or an effect of the disability pension (DP) itself has been discussed. It is shown that residual confounding is likely important in explaining previous findings of an association between DP and premature death. The residual confounding is primarily found in the comparison of individuals without previous hospitalization within a 5-year period. A method for assessing the degree of residual confounding for exposures considered to have little or no immediate effect on the outcome is presented. References 1 Björkenstam C , Alexanderson K , Björkenstam E . Diagnosis-specific disability pension and risk of all-cause and cause-specific mortality – a cohort study of 4.9 million inhabitants in Sweden . BMC Public Health 2014 ; 14 : 1247 . Google Scholar Crossref Search ADS PubMed 2 Gjesdal S , Haug K , Ringdal P , et al. Sickness absence with musculoskeletal or mental diagnoses, transition into disability pension and all-cause mortality: a 9-year prospective cohort study . Scand J Public Health 2009 ; 37 : 387 – 94 . Google Scholar Crossref Search ADS PubMed 3 Gjesdal S , Mæland JG , Svedberg P , et al. Role of diagnoses and socioeconomic status in mortality among disability pensioners in Norway - A population-based cohort study . Scand J Work Environ Health 2008 ; 34 : 479 – 82 . Google Scholar Crossref Search ADS PubMed 4 Gjesdal S , Svedberg P , Hagberg J , et al. Mortality among disability pensioners in Norway and Sweden 1990–96: comparative prospective cohort study . Scand J Public Health 2009 ; 37 : 168 – 75 . Google Scholar Crossref Search ADS PubMed 5 Karlsson NE , Carstensen JM , Gjesdal S , et al. Mortality in relation to disability pension: findings from a 12-year prospective population-based cohort study in Sweden . Scand J Public Health 2007 ; 35 : 341 – 7 . Google Scholar Crossref Search ADS PubMed 6 Wallman T , Wedel H , Johansson S , et al. The prognosis for individuals on disability retirement An 18-year mortality follow-up study of 6887 men and women sampled from the general population . BMC Public Health 2006 ; 6 : 103 . Google Scholar Crossref Search ADS PubMed 7 Åhs AMH , Westerling R . Mortality in relation to employment status during different levels of unemployment . Scand J Public Health 2006 ; 34 : 159 – 67 . Google Scholar Crossref Search ADS PubMed 8 Medhus A , Kristenson H . Mortality among male disability pensioners . Scand J Soc Med 1977 ; 5 : 73 – 5 . Google Scholar Crossref Search ADS PubMed 9 Gubéran E , Usel M . Permanent work incapacity, mortality and survival without work incapacity among occupations and social classes: a cohort study of ageing men in Geneva . Int J Epidemiol 1998 ; 27 : 1026 – 32 . Google Scholar Crossref Search ADS PubMed 10 Kaprio J , Sarna S , Fogelholm M , et al. Total and occupationally active life expectancies in relation to social class and marital status in men classified as healthy at 20 in Finland . J Epidemiol Community Health 1996 ; 50 : 653 – 60 . Google Scholar Crossref Search ADS PubMed 11 Hasle H , Jeune B , Skytthe A . Differential mortality among semiskilled applicants of disability pension . Scand J Soc Med 1988 ; 16 : 273 – 6 . Google Scholar Crossref Search ADS PubMed 12 Jeune B . Survival experience of semi-skilled disability pensioners in Denmark . Scand J Soc Med 1982 ; 10 : 73 – 6 . Google Scholar Crossref Search ADS PubMed 13 Damlund M , Gøth S , Hasle P , et al. The incidence of disability pensions and mortality among semi-skilled construction workers in Copenhagen. A retrospective cohort study with two control groups . Scand J Soc Med 1982 ; 10 : 43 – 7 . Google Scholar Crossref Search ADS PubMed 14 Vingård E , Alexanderson K , Norlund A . Swedish Council on Technology Assessment in Health Care (SBU) . Chapter 9. Consequences of being on sick leave . Scand J Public Health 2004 ; 32(Suppl) : 207 – 15 . Google Scholar Crossref Search ADS 15 Hult C , Stattin M , Janlert U , et al. Comparing mortality rates and recognizing health selection bias: a response to Wallman and Svärdsudd . Soc Sci Med 2010 ; 70 : 1489 – 91 . Google Scholar Crossref Search ADS 16 World Health Organization . International Statistical Classification of Diseases and Related Health Problems . Geneva : World Health Organization , 2004 . 17 Ranganathan P , Pramesh CS . Censoring in survival analysis: potential for bias . Perspect Clin Res 2012 ; 3 : 40 . Google Scholar Crossref Search ADS PubMed 18 Howe CJ , Cole SR , Chmiel JS , et al. Limitation of Inverse Probability-of-Censoring Weights in Estimating Survival in the Presence of Strong Selection Bias . Am J Epidemiol 2011 ; 173 : 569 – 77 . Google Scholar Crossref Search ADS PubMed 19 Harre FE , Lee KL , Pollock BG . Regression Models in Clinical Studies: determining Relationships Between Predictors and Response . J Natl Cancer Inst 1988 ; 80 : 1198 – 202 . Google Scholar Crossref Search ADS PubMed 20 Akaike H . A new look at the statistical model identification . IEEE Trans Autom Control 1974 ; 19 : 716 – 23 . Google Scholar Crossref Search ADS 21 Royston P , Parmar MKB . Flexible parametric proportional-hazards and proportional-odds models for censored survival data, with application to prognostic modelling and estimation of treatment effects . Stat Med 2002 ; 21 : 2175 – 97 . Google Scholar Crossref Search ADS PubMed 22 Jackson C . flexsurv: Flexible parametric survival and multi-state models. Available at: http://cran.r-project.org/web/packages/flexsurv/index.html ( 2014 , accessed 20 March 2015, date last accessed). 23 OECD . Sickness, Disability and Work: Breaking the Barriers: Sweden Will the Recent Reforms Make It? Paris : OECD Publishing , 2010 . 24 Olsson D , Alexanderson K , Bottai M . Sickness absence and the time-varying excess risk of premature death: a Swedish population-based prospective cohort study . J Epidemiol Community Health 2015 ; 69 : 1052 – 7 . Google Scholar Crossref Search ADS PubMed © The Author(s) 2018. Published by Oxford University Press on behalf of the European Public Health Association. All rights reserved. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png The European Journal of Public Health Oxford University Press

Assessing the degree of residual confounding: a cohort study on the association between disability pension and mortality

Loading next page...
 
/lp/ou_press/assessing-the-degree-of-residual-confounding-a-cohort-study-on-the-MGMBKPuHgC
Publisher
Oxford University Press
Copyright
© The Author(s) 2018. Published by Oxford University Press on behalf of the European Public Health Association. All rights reserved.
ISSN
1101-1262
eISSN
1464-360X
D.O.I.
10.1093/eurpub/cky022
Publisher site
See Article on Publisher Site

Abstract

Abstract Background Previous studies showed that disability pensioners have a higher risk of premature death than others, but residual confounding has been suggested. The aim was to assess the degree of residual confounding of the association between disability pension (DP) and risk of premature death. Methods Prospective cohort study of everyone aged 19–64 years, living in Sweden 31 December 2004 (n = 5 406 469), followed up through 2010. Mortality hazard rates over time were estimated for three groups; incident disability pensioners during 2005 from start of DP (February–December 2005), prevalent disability pensioners (January 2005 or since before), and individuals not on DP in January 2005, after standardizing populations to characteristics of the incident disability pensioners, stratified by previous hospitalization or not. If DP has no immediate effect on mortality, incident disability pensioners and those not on DP should initially have similar hazard rates, thereby, allowing assessment of the degree of residual confounding. Results For those not previously hospitalized, the mortality hazard rate on the first DP day was: 3.07 (95% CI 2.21, 4.36), 2.09 (1.78, 2.48) and 0.78 (0.73, 0.84) per thousand person-years for incident, prevalent, and non DP, respectively. Among previously hospitalized these figures were: 21.67 (17.73, 26.24), 17.00 (15.76, 18.51) and 18.88 (18.14, 19.64) respectively. Hazard ratios were 1.15 (0.94, 1.40) in the strata with and 3.94 (2.78, 5.57) in the strata without, previous hospitalization comparing incident DP with non-DP. Conclusions Substantial residual confounding was found in the association between DP and premature death among those not previously hospitalized. Introduction Previous studies have shown that disability pensioners have a higher risk of premature death than individuals not on disability pension (DP), even after adjusting for variables such as age, sex, and other potential confounders including proxies for the underlying morbidity.1–15 It has been discussed whether the higher risk should be attributed to being on DP or to unadjusted underlying differences between disability pensioners and other individuals. A potentially strong confounder is the individual’s morbidity. One way to try to capture this potential confounder is to adjust for number of days of previous hospitalization. It has been argued that since the association does not change much after adjusting for hospitalization, ‘DP may per se contain a damaging factor in addition to the underlying disease’.6 Others, however, have postulated that there likely is substantial residual confounding, including confounding by indication, and that this issue could only be adequately addressed by randomized trials, which however would generally be difficult and possibly unethical.1,14 Although it might not be possible to adjust for all confounding without randomized trials, it is possible to investigate the degree of confounding at the start of follow-up. If any association is observed already on the first day, it could mainly be attributed to residual confounding. It has been pointed out that if DP is associated with premature death, this should not impact mortality substantially in the first day, week or month.15 Ideally, if all confounding was adjusted for, mortality should be similar at the start of DP. The putative harmful effects of being granted DP should raise mortality rates gradually over time after the DP granting. The aim was to assess the degree of residual confounding in the association between DP and premature death by investigating whether the excess risk of premature death is present already from the start of follow-up. Methods A population-based prospective cohort study utilizing data from four nationwide Swedish registers from three authorities was conducted. Data were linked at the individual level using the ten-digit unique identification number of all residents in Sweden. The entire population aged 19–64 years and living in Sweden on 31 December 2004 (n = 5 406 469) was included. This cohort was followed up from 1 January 2005 until 31 December 2010 using data from (i) the LISA database of Statistics Sweden; information on age, sex, birth country, educational level, type of living area for December 2004 and year of emigration up through 2010; (ii) the Board of Health and Welfare’s Cause of Death Register (date of death); (iii) and information about hospitalization from the National Patient Register (2000–04), and (iv) the National Social Insurance Agency’s MiDAS database, regarding all on DP at 1 January 2005 and dates regarding all incident DP granted February–December 2005. DP is always granted from the first day of a month. Comparing three populations We compared three populations. The first population consisted of incident disability pensioners that for the first time were granted DP sometime during February–December 2005. This population was followed up from the first day of DP. The second population comprised the prevalent disability pensioners, individuals that already were on DP on 1 January 2005, for many granted long before. The third population consisted of individuals that were not on DP on 1 January 2005. The latter two populations were followed up from 1 January 2005. All analyses were stratified by whether individuals had or had not been hospitalized for at least 1 day in the 5-year period 2000–04, excluding pregnancies without complication (ICD-10 code O80).16 We censored observations at the year after emigration or the end of follow-up. If there was residual confounding at the start of the DP, such censoring would be informative and introduce further bias.17 Although methods to correct for the bias introduced by informative censoring are available, these tend to fail in the presence of strong selection bias.18 Still, not censoring those in the non-DP group granted DP after 2005, may also lead to bias, particularly the latter part of the curve for the non-DP group may be overestimated. However, this choice will not be of importance as the focus of this study was the hazard rate on day 1. Regarding date of death, for 2219 individuals, deaths information on the exact day was missing and for 1024 deaths the exact month was missing. For these interval-censored deaths, the day of death were randomly imputed from a uniform distribution. Standardization to create hypothetical populations To make all three populations comparable, we standardized them to the same distribution of background variables measured at 31 December 2004: age, sex (man/woman), birth place (Sweden, other Nordic countries, other countries within EU25, and countries in the rest of the world), educational level (low (elementary school, ≤9 years), medium (upper secondary school, 10–12 years) and high (college/university, >12 years)), married (yes/no), and type of living area (urban, semi-urban and sparsely populated) and days of hospitalization in 2000–04 in five categories (0, 1–2, 3–6, 7–15, ≥16 days). We chose to standardize the non-DP population and the prevalent DP population to create two hypothetical populations with similar distribution of background variables as the incident disability pensioners. The standardization was performed by estimating probability weights for each of the non-DP population and the prevalent DP population. The weights for standardizing the non-DP population to the covariate distribution of the incident DP population were obtained by logistic regression, estimating the odds of being in the incident DP population relative to the non-DP population in an augmented dataset containing the incident DP population and the non-DP population. The predicted odds were then rescaled to have a mean of one and used as probability weights in subsequent analyses. The weights for standardizing the prevalent DP population to the covariate pattern of the incident DP population were obtained analogously. All variables were included as categorical variables in the logistic regression model, except age which was used as a restricted cubic spline function with three knots placed at the observed quartiles.19 An interaction between sex and age was included. Interactions of previous hospitalization with age and sex as well as a three-way interaction were also included in the strata with previous hospitalizations. Based on the Akaike Information Criterion (AIC),20 these models fitted the data better than simpler models without all interactions. No large probability weights were observed in any analysis, with the largest weight over all analyses being 13.3. Hazard rates across time We estimated the hazard rates for death over the follow-up time in each of the three populations and in the two hypothetical populations. We used flexible parametric models for survival analyses without covariates. We estimated smooth curves separately for each population by using the ‘flexsurv’ package in R.21,22 To estimate the hazard in the two hypothetical populations, we used the weights as described earlier. To avoid forcing the hazard rate to be 0 at time 0, we used a model that constrained the shape parameter of the assumed Weibull distribution to be equal to 1. We placed the boundary knots at the time of the first and last event and three internal knots at years 1, 2 and 4, respectively. We present hazard rates with 95% CIs at day 1, at 2, 4 and 6 years, respectively, in the figures and tables. We also present crude incidence rates for the first 30 days with exact 95% CI, to confirm that the flexible parametric model works as intended. In Sweden, the ordinary old-age retirement age is 65 years. People aged 19–64 and not on old-age pension can be granted DP if they have long-standing or permanent work incapacity caused by disease or injury. In 2005 there was no limit to duration of a sick-leave spell before being assessed for DP. Although we estimated hazards across time, our primary interest was to compare the three populations after standardization at the start of follow-up, because this is where the confounding adjustment will be based on the most recent information. After the first day, the information on the confounding variables was incomplete. If no immediate effect is expected, comparing the incident disability pensioners with those not on DP, at the start of follow-up, will enable an assessment of the degree of confounding at that time point. Furthermore, comparing the incident disability pensioners with the prevalent disability pensioners enable us to assess whether the excess risk of premature death among disability pensioners is already present at the start of follow-up or if it develops later. A 95% CI for this estimate was calculated from the two individual CIs using 1 million bootstrap samples. Results Background characteristics and size of the three populations in each of the two strata are presented in tables 1 and 2. Of all individuals, 21% had at least one inpatient day in the 5-year period 2000–04. These proportions were 44% among those with prevalent DP, 46% among those with incident DP, and 19% in the population with no DP in 2005. In the strata without previous hospitalization, the two DP populations were more likely to be women, older, born outside Sweden, less educated, married and to live in a rural area. In individuals with previous hospitalizations, these differences were less pronounced, although the two DP populations were older, less educated and had more days of previous hospitalization. Table 1 Background characteristics for the three populations; incident disability pensioners, prevalent disability pensioners and those not on disability pension (DP) at 31 December 2004, without hospitalization in the 5 years 2000–04 Not on DP Prevalent DP Incident DP n = 3 946 691 n = 299 029 n = 31 735 n % n % n % Sex     Women 1 807 645 45.8 182 490 61.0 20 087 63.3     Men 2 139 046 54.2 116 539 39.0 11 648 36.7 Age categories     19–24 years 521 087 13.2 7298 2.4 952 3.0     25–34 years 901 951 22.9 14 951 5.0 2502 7.9     35–44 years 972 805 24.6 41 635 13.9 6455 20.3     45–54 years 849 428 21.5 80 463 26.9 9174 28.9     55–64 years 701 420 17.8 154 682 51.7 12 652 39.9 Birth place     Sweden 3 369 294 85.4 238 387 79.7 25 433 80.1     Other Nordic 123 678 3.1 19 316 6.5 1556 4.9     Other EU25 92 399 2.3 10 697 3.6 892 2.8     Rest of the world 361 320 9.2 30 629 10.2 3854 12.1 Educational level     Low 628 539 15.9 119 928 40.1 8613 27.1     Medium 1 940 640 49.2 135 650 45.4 16 204 51.1     High 1 377 512 34.9 43 451 14.5 6918 21.8 Married     No 2 295 249 58.2 163 945 54.8 15 843 49.9     Yes 1 651 442 41.8 135 084 45.2 15 892 50.1 Type of living area     Urban 1 514 667 38.4 95 694 32.0 10 283 31.7     Semi-urban 1 387 758 35.2 105 718 35.4 11 098 34.2     Sparsely populated 1 044 266 26.5 97 617 32.6 11 098 34.2 Not on DP Prevalent DP Incident DP n = 3 946 691 n = 299 029 n = 31 735 n % n % n % Sex     Women 1 807 645 45.8 182 490 61.0 20 087 63.3     Men 2 139 046 54.2 116 539 39.0 11 648 36.7 Age categories     19–24 years 521 087 13.2 7298 2.4 952 3.0     25–34 years 901 951 22.9 14 951 5.0 2502 7.9     35–44 years 972 805 24.6 41 635 13.9 6455 20.3     45–54 years 849 428 21.5 80 463 26.9 9174 28.9     55–64 years 701 420 17.8 154 682 51.7 12 652 39.9 Birth place     Sweden 3 369 294 85.4 238 387 79.7 25 433 80.1     Other Nordic 123 678 3.1 19 316 6.5 1556 4.9     Other EU25 92 399 2.3 10 697 3.6 892 2.8     Rest of the world 361 320 9.2 30 629 10.2 3854 12.1 Educational level     Low 628 539 15.9 119 928 40.1 8613 27.1     Medium 1 940 640 49.2 135 650 45.4 16 204 51.1     High 1 377 512 34.9 43 451 14.5 6918 21.8 Married     No 2 295 249 58.2 163 945 54.8 15 843 49.9     Yes 1 651 442 41.8 135 084 45.2 15 892 50.1 Type of living area     Urban 1 514 667 38.4 95 694 32.0 10 283 31.7     Semi-urban 1 387 758 35.2 105 718 35.4 11 098 34.2     Sparsely populated 1 044 266 26.5 97 617 32.6 11 098 34.2 Table 1 Background characteristics for the three populations; incident disability pensioners, prevalent disability pensioners and those not on disability pension (DP) at 31 December 2004, without hospitalization in the 5 years 2000–04 Not on DP Prevalent DP Incident DP n = 3 946 691 n = 299 029 n = 31 735 n % n % n % Sex     Women 1 807 645 45.8 182 490 61.0 20 087 63.3     Men 2 139 046 54.2 116 539 39.0 11 648 36.7 Age categories     19–24 years 521 087 13.2 7298 2.4 952 3.0     25–34 years 901 951 22.9 14 951 5.0 2502 7.9     35–44 years 972 805 24.6 41 635 13.9 6455 20.3     45–54 years 849 428 21.5 80 463 26.9 9174 28.9     55–64 years 701 420 17.8 154 682 51.7 12 652 39.9 Birth place     Sweden 3 369 294 85.4 238 387 79.7 25 433 80.1     Other Nordic 123 678 3.1 19 316 6.5 1556 4.9     Other EU25 92 399 2.3 10 697 3.6 892 2.8     Rest of the world 361 320 9.2 30 629 10.2 3854 12.1 Educational level     Low 628 539 15.9 119 928 40.1 8613 27.1     Medium 1 940 640 49.2 135 650 45.4 16 204 51.1     High 1 377 512 34.9 43 451 14.5 6918 21.8 Married     No 2 295 249 58.2 163 945 54.8 15 843 49.9     Yes 1 651 442 41.8 135 084 45.2 15 892 50.1 Type of living area     Urban 1 514 667 38.4 95 694 32.0 10 283 31.7     Semi-urban 1 387 758 35.2 105 718 35.4 11 098 34.2     Sparsely populated 1 044 266 26.5 97 617 32.6 11 098 34.2 Not on DP Prevalent DP Incident DP n = 3 946 691 n = 299 029 n = 31 735 n % n % n % Sex     Women 1 807 645 45.8 182 490 61.0 20 087 63.3     Men 2 139 046 54.2 116 539 39.0 11 648 36.7 Age categories     19–24 years 521 087 13.2 7298 2.4 952 3.0     25–34 years 901 951 22.9 14 951 5.0 2502 7.9     35–44 years 972 805 24.6 41 635 13.9 6455 20.3     45–54 years 849 428 21.5 80 463 26.9 9174 28.9     55–64 years 701 420 17.8 154 682 51.7 12 652 39.9 Birth place     Sweden 3 369 294 85.4 238 387 79.7 25 433 80.1     Other Nordic 123 678 3.1 19 316 6.5 1556 4.9     Other EU25 92 399 2.3 10 697 3.6 892 2.8     Rest of the world 361 320 9.2 30 629 10.2 3854 12.1 Educational level     Low 628 539 15.9 119 928 40.1 8613 27.1     Medium 1 940 640 49.2 135 650 45.4 16 204 51.1     High 1 377 512 34.9 43 451 14.5 6918 21.8 Married     No 2 295 249 58.2 163 945 54.8 15 843 49.9     Yes 1 651 442 41.8 135 084 45.2 15 892 50.1 Type of living area     Urban 1 514 667 38.4 95 694 32.0 10 283 31.7     Semi-urban 1 387 758 35.2 105 718 35.4 11 098 34.2     Sparsely populated 1 044 266 26.5 97 617 32.6 11 098 34.2 Table 2 Background characteristics for the three populations; incident disability pensioners, prevalent disability pensioners and those not on DP at 31 December 2004, who had at least 1 day with hospitalization in the 5 years 2000–04 Not on DP Prevalent DP Incident DP n=927 225 n=233 524 n=26 771 n % n % n % Sex     Women 537 672 58.0 135 062 57.8 15 477 57.8     Men 389 553 42.0 98 462 42.2 11 294 42.2 Age categories (years)     19–24 96 899 10.5 5077 2.2 919 3.4     25–34 227 976 24.6 13 414 5.7 2525 9.4     35–44 219 926 23.7 32 451 13.9 4875 18.2     45–54 181 373 19.6 58 659 25.1 7068 26.4     55–64 201 051 21.7 123 923 53.1 11 384 42.5 Birth place     Sweden 788 422 85.0 190 594 81.6 21 830 81.5     Other Nordic 30 654 3.3 15 999 6.9 1426 5.3     Other EU25 16 931 1.8 6398 2.7 654 2.4     Rest of the world 91 218 9.8 20 533 8.8 2861 10.7 Educational level     Low 172 367 18.6 90 100 38.6 7704 28.8     Medium 465 761 50.2 109 936 47.1 13 719 51.2     High 289 097 31.2 33 488 14.3 5348 20.0 Married     No 508 394 54.8 135 129 57.9 14 000 52.3     Yes 418 831 45.2 98 395 42.1 12 771 47.7 Type of living area     Urban 335 394 36.2 73 967 31.7 8682 32.4     Semi-urban 329 156 35.5 82 831 35.5 9368 35.0     Sparsely populated 262 675 28.3 76 726 32.9 8721 32.6 Hospitalization (days)     1–2 394 970 42.6 56 480 24.2 6847 25.6     3–6 306 251 33.0 54 339 23.3 6397 23.9     7–15 150 377 16.2 49 253 21.1 5883 22.0     16 or more 75 627 8.2 73 452 31.5 7644 28.6 Not on DP Prevalent DP Incident DP n=927 225 n=233 524 n=26 771 n % n % n % Sex     Women 537 672 58.0 135 062 57.8 15 477 57.8     Men 389 553 42.0 98 462 42.2 11 294 42.2 Age categories (years)     19–24 96 899 10.5 5077 2.2 919 3.4     25–34 227 976 24.6 13 414 5.7 2525 9.4     35–44 219 926 23.7 32 451 13.9 4875 18.2     45–54 181 373 19.6 58 659 25.1 7068 26.4     55–64 201 051 21.7 123 923 53.1 11 384 42.5 Birth place     Sweden 788 422 85.0 190 594 81.6 21 830 81.5     Other Nordic 30 654 3.3 15 999 6.9 1426 5.3     Other EU25 16 931 1.8 6398 2.7 654 2.4     Rest of the world 91 218 9.8 20 533 8.8 2861 10.7 Educational level     Low 172 367 18.6 90 100 38.6 7704 28.8     Medium 465 761 50.2 109 936 47.1 13 719 51.2     High 289 097 31.2 33 488 14.3 5348 20.0 Married     No 508 394 54.8 135 129 57.9 14 000 52.3     Yes 418 831 45.2 98 395 42.1 12 771 47.7 Type of living area     Urban 335 394 36.2 73 967 31.7 8682 32.4     Semi-urban 329 156 35.5 82 831 35.5 9368 35.0     Sparsely populated 262 675 28.3 76 726 32.9 8721 32.6 Hospitalization (days)     1–2 394 970 42.6 56 480 24.2 6847 25.6     3–6 306 251 33.0 54 339 23.3 6397 23.9     7–15 150 377 16.2 49 253 21.1 5883 22.0     16 or more 75 627 8.2 73 452 31.5 7644 28.6 Table 2 Background characteristics for the three populations; incident disability pensioners, prevalent disability pensioners and those not on DP at 31 December 2004, who had at least 1 day with hospitalization in the 5 years 2000–04 Not on DP Prevalent DP Incident DP n=927 225 n=233 524 n=26 771 n % n % n % Sex     Women 537 672 58.0 135 062 57.8 15 477 57.8     Men 389 553 42.0 98 462 42.2 11 294 42.2 Age categories (years)     19–24 96 899 10.5 5077 2.2 919 3.4     25–34 227 976 24.6 13 414 5.7 2525 9.4     35–44 219 926 23.7 32 451 13.9 4875 18.2     45–54 181 373 19.6 58 659 25.1 7068 26.4     55–64 201 051 21.7 123 923 53.1 11 384 42.5 Birth place     Sweden 788 422 85.0 190 594 81.6 21 830 81.5     Other Nordic 30 654 3.3 15 999 6.9 1426 5.3     Other EU25 16 931 1.8 6398 2.7 654 2.4     Rest of the world 91 218 9.8 20 533 8.8 2861 10.7 Educational level     Low 172 367 18.6 90 100 38.6 7704 28.8     Medium 465 761 50.2 109 936 47.1 13 719 51.2     High 289 097 31.2 33 488 14.3 5348 20.0 Married     No 508 394 54.8 135 129 57.9 14 000 52.3     Yes 418 831 45.2 98 395 42.1 12 771 47.7 Type of living area     Urban 335 394 36.2 73 967 31.7 8682 32.4     Semi-urban 329 156 35.5 82 831 35.5 9368 35.0     Sparsely populated 262 675 28.3 76 726 32.9 8721 32.6 Hospitalization (days)     1–2 394 970 42.6 56 480 24.2 6847 25.6     3–6 306 251 33.0 54 339 23.3 6397 23.9     7–15 150 377 16.2 49 253 21.1 5883 22.0     16 or more 75 627 8.2 73 452 31.5 7644 28.6 Not on DP Prevalent DP Incident DP n=927 225 n=233 524 n=26 771 n % n % n % Sex     Women 537 672 58.0 135 062 57.8 15 477 57.8     Men 389 553 42.0 98 462 42.2 11 294 42.2 Age categories (years)     19–24 96 899 10.5 5077 2.2 919 3.4     25–34 227 976 24.6 13 414 5.7 2525 9.4     35–44 219 926 23.7 32 451 13.9 4875 18.2     45–54 181 373 19.6 58 659 25.1 7068 26.4     55–64 201 051 21.7 123 923 53.1 11 384 42.5 Birth place     Sweden 788 422 85.0 190 594 81.6 21 830 81.5     Other Nordic 30 654 3.3 15 999 6.9 1426 5.3     Other EU25 16 931 1.8 6398 2.7 654 2.4     Rest of the world 91 218 9.8 20 533 8.8 2861 10.7 Educational level     Low 172 367 18.6 90 100 38.6 7704 28.8     Medium 465 761 50.2 109 936 47.1 13 719 51.2     High 289 097 31.2 33 488 14.3 5348 20.0 Married     No 508 394 54.8 135 129 57.9 14 000 52.3     Yes 418 831 45.2 98 395 42.1 12 771 47.7 Type of living area     Urban 335 394 36.2 73 967 31.7 8682 32.4     Semi-urban 329 156 35.5 82 831 35.5 9368 35.0     Sparsely populated 262 675 28.3 76 726 32.9 8721 32.6 Hospitalization (days)     1–2 394 970 42.6 56 480 24.2 6847 25.6     3–6 306 251 33.0 54 339 23.3 6397 23.9     7–15 150 377 16.2 49 253 21.1 5883 22.0     16 or more 75 627 8.2 73 452 31.5 7644 28.6 In individuals without previous hospitalization (figure 1 and table 3), the hazard rates of premature death were relatively low in absolute terms, but increased substantially in relative terms over time for all populations. Before standardization the prevalent DP population had the highest risk, followed by the incident DP population, and those not on DP having the lowest risk. After standardization, the observed differences were smaller. The hazard rates for the incident DP population and the prevalent DP population were after standardization essentially at the same level. Table 3 Mortality hazard rates and 95% CIs across time and number of deaths and number of person-years by population for individuals without previous hospitalization, unstandardized and standardized n Deaths Time at risk (person years) Day 1 2 years 4 years 6 years Not DP 3 946 691 33 674 23 290 943 0.52 (0.48, 0.57) 1.27 (1.25, 1.30) 1.71 (1.68, 1.73) 2.03 (1.98, 2.07) Not DP, standardized to incident DP 3 946 691 33 674 23 290 943 0.78 (0.73, 0.84) 2.18 (2.15, 2.21) 2.99 (2.95, 3.02) 3.64 (3.58, 3.70) Prevalent DP 299 029 11 757 1756964 2.72 (2.35, 3.12) 5.96 (5.78, 6.16) 7.73 (7.56, 7.91) 9.21 (8.87, 9.58) Prevalent DP, standardized to incident DP 299 029 11 757 1 756 964 2.09 (1.78, 2.48) 4.58 (4.42, 4.73) 5.94 (5.78, 6.10) 6.95 (6.67, 7.22) Incident DP 31 735 848 174 188 3.07 (2.21, 4.36) 4.63 (4.15, 5.22) 5.35 (4.93, 5.85) 6.30 (5.35, 7.33) n Deaths Time at risk (person years) Day 1 2 years 4 years 6 years Not DP 3 946 691 33 674 23 290 943 0.52 (0.48, 0.57) 1.27 (1.25, 1.30) 1.71 (1.68, 1.73) 2.03 (1.98, 2.07) Not DP, standardized to incident DP 3 946 691 33 674 23 290 943 0.78 (0.73, 0.84) 2.18 (2.15, 2.21) 2.99 (2.95, 3.02) 3.64 (3.58, 3.70) Prevalent DP 299 029 11 757 1756964 2.72 (2.35, 3.12) 5.96 (5.78, 6.16) 7.73 (7.56, 7.91) 9.21 (8.87, 9.58) Prevalent DP, standardized to incident DP 299 029 11 757 1 756 964 2.09 (1.78, 2.48) 4.58 (4.42, 4.73) 5.94 (5.78, 6.10) 6.95 (6.67, 7.22) Incident DP 31 735 848 174 188 3.07 (2.21, 4.36) 4.63 (4.15, 5.22) 5.35 (4.93, 5.85) 6.30 (5.35, 7.33) Hazards are multiplied by 1000 with the timescale in years. Table 3 Mortality hazard rates and 95% CIs across time and number of deaths and number of person-years by population for individuals without previous hospitalization, unstandardized and standardized n Deaths Time at risk (person years) Day 1 2 years 4 years 6 years Not DP 3 946 691 33 674 23 290 943 0.52 (0.48, 0.57) 1.27 (1.25, 1.30) 1.71 (1.68, 1.73) 2.03 (1.98, 2.07) Not DP, standardized to incident DP 3 946 691 33 674 23 290 943 0.78 (0.73, 0.84) 2.18 (2.15, 2.21) 2.99 (2.95, 3.02) 3.64 (3.58, 3.70) Prevalent DP 299 029 11 757 1756964 2.72 (2.35, 3.12) 5.96 (5.78, 6.16) 7.73 (7.56, 7.91) 9.21 (8.87, 9.58) Prevalent DP, standardized to incident DP 299 029 11 757 1 756 964 2.09 (1.78, 2.48) 4.58 (4.42, 4.73) 5.94 (5.78, 6.10) 6.95 (6.67, 7.22) Incident DP 31 735 848 174 188 3.07 (2.21, 4.36) 4.63 (4.15, 5.22) 5.35 (4.93, 5.85) 6.30 (5.35, 7.33) n Deaths Time at risk (person years) Day 1 2 years 4 years 6 years Not DP 3 946 691 33 674 23 290 943 0.52 (0.48, 0.57) 1.27 (1.25, 1.30) 1.71 (1.68, 1.73) 2.03 (1.98, 2.07) Not DP, standardized to incident DP 3 946 691 33 674 23 290 943 0.78 (0.73, 0.84) 2.18 (2.15, 2.21) 2.99 (2.95, 3.02) 3.64 (3.58, 3.70) Prevalent DP 299 029 11 757 1756964 2.72 (2.35, 3.12) 5.96 (5.78, 6.16) 7.73 (7.56, 7.91) 9.21 (8.87, 9.58) Prevalent DP, standardized to incident DP 299 029 11 757 1 756 964 2.09 (1.78, 2.48) 4.58 (4.42, 4.73) 5.94 (5.78, 6.10) 6.95 (6.67, 7.22) Incident DP 31 735 848 174 188 3.07 (2.21, 4.36) 4.63 (4.15, 5.22) 5.35 (4.93, 5.85) 6.30 (5.35, 7.33) Hazards are multiplied by 1000 with the timescale in years. Figure 1 View largeDownload slide Mortality hazard rates and 95% CI across time for the incident disability pensioners, the prevalent disability pensioners, and those not on DP, without previous hospitalization. Hazards are multiplied by 1000 with the timescale in years Figure 1 View largeDownload slide Mortality hazard rates and 95% CI across time for the incident disability pensioners, the prevalent disability pensioners, and those not on DP, without previous hospitalization. Hazards are multiplied by 1000 with the timescale in years Overall, the hazard rates were much higher in the strata with previous hospitalization (figure 2 and table 4). The standardized and unstandardized prevalent DP population and the unstandardized non-DP population showed a similar pattern with relatively constant hazard rates over time. Comparing the incident DP population with the standardized non-DP population, the two hazard curves at the start of follow-up were very similar. The two hazard rates were not significantly different from each other on the first day. Table 4 Mortality hazard rates and 95% CIs across time and number of deaths and number of person-years by population for individuals with previous hospitalization, unstandardized and standardized n Deaths Time at risk (person years) Day 1 2 years 4 years 6 years Not DP 927 225 22 023 5 461 229 5.57 (5.22, 5.97) 3.78 (3.70, 3.86) 3.95 (3.88, 4.03) 4.13 (4.01, 4.24) Not DP, standardized to incident DP 927 225 22 023 5 461 229 18.88 (18.14, 19.64) 8.92 (8.79, 9.04) 8.63 (8.52, 8.74) 8.64 (8.47, 8.81) Prevalent DP 233 524 24 644 1 321 318 20.91 (19.41, 22.34) 18.27 (17.92, 18.63) 18.36 (18.05, 18.68) 18.81 (18.29, 19.31) Prevalent DP, standardized to incident DP 233 524 24 644 1 321 318 17.00 (15.76, 18.51) 14.73 (14.41, 15.08) 14.87 (14.59, 15.15) 15.39 (14.92, 15.86) Incident DP 26 771 2036 142 536 21.67 (17.73, 26.24) 14.42 (13.44, 15.42) 12.36 (11.56, 13.19) 11.91 (10.51, 13.32) n Deaths Time at risk (person years) Day 1 2 years 4 years 6 years Not DP 927 225 22 023 5 461 229 5.57 (5.22, 5.97) 3.78 (3.70, 3.86) 3.95 (3.88, 4.03) 4.13 (4.01, 4.24) Not DP, standardized to incident DP 927 225 22 023 5 461 229 18.88 (18.14, 19.64) 8.92 (8.79, 9.04) 8.63 (8.52, 8.74) 8.64 (8.47, 8.81) Prevalent DP 233 524 24 644 1 321 318 20.91 (19.41, 22.34) 18.27 (17.92, 18.63) 18.36 (18.05, 18.68) 18.81 (18.29, 19.31) Prevalent DP, standardized to incident DP 233 524 24 644 1 321 318 17.00 (15.76, 18.51) 14.73 (14.41, 15.08) 14.87 (14.59, 15.15) 15.39 (14.92, 15.86) Incident DP 26 771 2036 142 536 21.67 (17.73, 26.24) 14.42 (13.44, 15.42) 12.36 (11.56, 13.19) 11.91 (10.51, 13.32) Hazards are multiplied by 1000 with the timescale in years. Table 4 Mortality hazard rates and 95% CIs across time and number of deaths and number of person-years by population for individuals with previous hospitalization, unstandardized and standardized n Deaths Time at risk (person years) Day 1 2 years 4 years 6 years Not DP 927 225 22 023 5 461 229 5.57 (5.22, 5.97) 3.78 (3.70, 3.86) 3.95 (3.88, 4.03) 4.13 (4.01, 4.24) Not DP, standardized to incident DP 927 225 22 023 5 461 229 18.88 (18.14, 19.64) 8.92 (8.79, 9.04) 8.63 (8.52, 8.74) 8.64 (8.47, 8.81) Prevalent DP 233 524 24 644 1 321 318 20.91 (19.41, 22.34) 18.27 (17.92, 18.63) 18.36 (18.05, 18.68) 18.81 (18.29, 19.31) Prevalent DP, standardized to incident DP 233 524 24 644 1 321 318 17.00 (15.76, 18.51) 14.73 (14.41, 15.08) 14.87 (14.59, 15.15) 15.39 (14.92, 15.86) Incident DP 26 771 2036 142 536 21.67 (17.73, 26.24) 14.42 (13.44, 15.42) 12.36 (11.56, 13.19) 11.91 (10.51, 13.32) n Deaths Time at risk (person years) Day 1 2 years 4 years 6 years Not DP 927 225 22 023 5 461 229 5.57 (5.22, 5.97) 3.78 (3.70, 3.86) 3.95 (3.88, 4.03) 4.13 (4.01, 4.24) Not DP, standardized to incident DP 927 225 22 023 5 461 229 18.88 (18.14, 19.64) 8.92 (8.79, 9.04) 8.63 (8.52, 8.74) 8.64 (8.47, 8.81) Prevalent DP 233 524 24 644 1 321 318 20.91 (19.41, 22.34) 18.27 (17.92, 18.63) 18.36 (18.05, 18.68) 18.81 (18.29, 19.31) Prevalent DP, standardized to incident DP 233 524 24 644 1 321 318 17.00 (15.76, 18.51) 14.73 (14.41, 15.08) 14.87 (14.59, 15.15) 15.39 (14.92, 15.86) Incident DP 26 771 2036 142 536 21.67 (17.73, 26.24) 14.42 (13.44, 15.42) 12.36 (11.56, 13.19) 11.91 (10.51, 13.32) Hazards are multiplied by 1000 with the timescale in years. Figure 2 View largeDownload slide Mortality hazard rates and 95% CI across time for the incident disability pensioners, the prevalent disability pensioners, and those not on DP, with previous hospitalization. Hazards are multiplied by 1000 with the timescale in years Figure 2 View largeDownload slide Mortality hazard rates and 95% CI across time for the incident disability pensioners, the prevalent disability pensioners, and those not on DP, with previous hospitalization. Hazards are multiplied by 1000 with the timescale in years Supplementary table S1a and b present crude incidence rates for the first 30 days, giving similar results to the unstandardized hazard rates on day 1, but with wider CIs, consistently with the results from the parametric model. If being granted DP has no immediate effect on mortality, incident disability pensioners and those not on DP should have similar risk initially. This allows us to estimate the degree of residual confounding. Estimates for the degree of residual confounding in terms of hazard ratios are 21.67/18.88 = 1.15 (95% CI 0.94, 1.40) in the strata with previous hospitalization and 3.07/0.78 = 3.94 (2.78, 5.57) in the strata without previous hospitalization. By comparing incident and prevalent disability pensioners after standardization at the start of follow-up we could estimate how much of the excess risk that was present already then. In the strata without previous hospitalization the hazard ratio was 3.07/2.07 = 1.47 (1.01, 2.14) and in the strata with previous hospitalization the hazard ratio was 21.67/17.00 = 1.27 (1.03, 1.58), which meant that the excess risk of premature death was present already when granted DP and was even higher for incident DPs than for prevalent DPs. Discussion In this study, we investigated the association of being on DP with mortality, considering that if any immediate effect is small or nonexistent, residual confounding is likely important for explaining previous findings of DP being associated with premature death. We found that the excess risk of premature death among disability pensioners is present already when being granted DP. The large, what could be assumed as, residual confounding observed at baseline in those without previous hospitalization, which constitute 79% of the total population, may explain findings from previous studies. The study that first pointed out that adjusting for days of hospitalization did not impact the results, had a very long follow-up of 18 years.[6] We expect that while previous hospitalizations measured at baseline might be a good proxy for morbidity at the start of follow-up, they may not be as good a proxy for morbidity 18 years later. Our finding that incident disability pensioners had a higher risk of premature death than prevalent disability pensioners at the start of follow-up is consistent with a previous study that compared prevalent disability pensioners with those not on DP and found no variation across time regarding the three times higher risk of premature death among those on DP, except for a somewhat higher risk during the first year after being granted DP.5 However, this previous study did not attempt to adjust for morbidity. We limited the comparison to the start of follow-up because this is where confounding adjustment for hospitalization will be most accurate, considering that we did not have updated information on hospitalizations after baseline. Disability pensioners were more similar to other disability pensioners and so the degree of confounding for comparing incident and prevalent disability pensioners was expected to be smaller than when comparing disability pensioners with those not on DP. However, this difference between prevalent and incident disability pensioners might also be accounted for by possible cohort effects. Strengths and limitations To the best of our knowledge, this is the first study attempting to assess the degree of residual confounding in the association of being granted DP and the risk of premature death. We presented a method for assessing the degree of residual confounding at the start of follow-up for exposures that were not expected to have an immediate effect on the outcome. Because our study was based on a large population-based cohort covering a whole country, we had sufficient power to compare incident disability pensioners, prevalent disability pensioners, and those not on DP at the start of follow-up. We used standardization to adjust for confounding. This approach to assessing the degree of residual confounding can be useful when trying to improve the confounding adjustment in future studies as a check whether confounding adjustment is inadequate. Standardization adjusts for confounding by balancing the distribution of measured confounders, similarly to randomization, except randomization also balances unmeasured confounders. The reason why we used standardization instead of regression models for the outcome was that it allowed us to use a simple model for the outcome, whose only assumption was that the hazard rate changed smoothly over time. This facilitated interpretation of our findings. In this study we measured morbidity in terms of previous hospitalization. Other morbidity measures could also be used, such as out-patient visits, medication, sick-leave patterns, self-rated health, as days of hospitalization only captures a limited range of morbidity. Another main limitation of our study is that we did not have information on duration and grade of previous and ongoing sick leave. Based on the regulations in Sweden, after 12 months on sick leave, people should be assessed regarding fulfilling the criteria for DP. In 2005, many people on long-term sick leave had not been assessed for DP, due to practices of the National Social Insurance Agency. If they had been assessed for DP earlier, they could already have been on DP for years, and could have been included in the prevalent DP population in this study. In future studies, this group needs to be included. The Organization for Economic Co-operation and Development (OECD) tackled this problem regarding DP data in Sweden by categorizing all sick-leave spells that have been on-going for more than 2 years as being DP cases.23 Moreover, including information on diagnoses might add to the understanding of the processes studied here. In a previous study24 we investigated sick leave and mortality followed from day 32 of a sick-leave spell. There we found that the higher mortality among those with such sick leave was temporary and had vanished after 6 years. The adjusted hazard ratio comparing DP to non-DP standardized to the total population on sick-leave day 32 was 13.58/0.90 = 15.09. The OECD definition lies in between the definition of DP used in this paper and the 32 days of sick leave used in our previous study. This indicates that including sick leave into the definition of a DP as OECD does might increase the problems with residual confounding. However, this needs to be investigated in future studies. These studies should also include additional measures of morbidity, before and after being granted DP, as well as information on previous and ongoing sick leave. Theoretically, there could be detrimental factors under long-term sick leave before day 1 of DP, contributing to a higher risk of premature death of disability pensioners or to a higher risk of premature death among those on long-term sick leave not yet granted DP. Our analysis does not find support for this hypothesis, since in the stratum with previous hospitalization, where we have most of the deaths; no excess risk at the start of DP was found after standardization. However, this should be investigated in more detail in future studies. Further, people with acute, very severe, and fatal diseases usually are not granted DP, rather they are allowed to retain sick-leave benefits, which usually are higher than the DP benefits. We did not differentiate between DP diagnoses or whether people were granted a full- or a part-time DP. Such analyses would be recommended in future studies, especially regarding whether the associations differ between people on part-time and those on full-time DP. These are not limitations that would change the overall result, but it may be possible to locate more exactly in which subgroups residual confounding is more or less pronounced. The method for assessing residual confounding presented in this study would be of use in such future studies to assess if the confounding adjustment is inadequate. This method may also be useful more generally in studies where the exposure is not expected to have an immediate effect on the outcome. Funding This study was financially supported by the Swedish Research Council of Health, Working Life and Welfare (2007-1762). Conflicts of interest: None declared. Key points Previous studies have shown that disability pensioners have higher risks of premature death. Whether this is due to residual confounding or an effect of the disability pension (DP) itself has been discussed. It is shown that residual confounding is likely important in explaining previous findings of an association between DP and premature death. The residual confounding is primarily found in the comparison of individuals without previous hospitalization within a 5-year period. A method for assessing the degree of residual confounding for exposures considered to have little or no immediate effect on the outcome is presented. References 1 Björkenstam C , Alexanderson K , Björkenstam E . Diagnosis-specific disability pension and risk of all-cause and cause-specific mortality – a cohort study of 4.9 million inhabitants in Sweden . BMC Public Health 2014 ; 14 : 1247 . Google Scholar Crossref Search ADS PubMed 2 Gjesdal S , Haug K , Ringdal P , et al. Sickness absence with musculoskeletal or mental diagnoses, transition into disability pension and all-cause mortality: a 9-year prospective cohort study . Scand J Public Health 2009 ; 37 : 387 – 94 . Google Scholar Crossref Search ADS PubMed 3 Gjesdal S , Mæland JG , Svedberg P , et al. Role of diagnoses and socioeconomic status in mortality among disability pensioners in Norway - A population-based cohort study . Scand J Work Environ Health 2008 ; 34 : 479 – 82 . Google Scholar Crossref Search ADS PubMed 4 Gjesdal S , Svedberg P , Hagberg J , et al. Mortality among disability pensioners in Norway and Sweden 1990–96: comparative prospective cohort study . Scand J Public Health 2009 ; 37 : 168 – 75 . Google Scholar Crossref Search ADS PubMed 5 Karlsson NE , Carstensen JM , Gjesdal S , et al. Mortality in relation to disability pension: findings from a 12-year prospective population-based cohort study in Sweden . Scand J Public Health 2007 ; 35 : 341 – 7 . Google Scholar Crossref Search ADS PubMed 6 Wallman T , Wedel H , Johansson S , et al. The prognosis for individuals on disability retirement An 18-year mortality follow-up study of 6887 men and women sampled from the general population . BMC Public Health 2006 ; 6 : 103 . Google Scholar Crossref Search ADS PubMed 7 Åhs AMH , Westerling R . Mortality in relation to employment status during different levels of unemployment . Scand J Public Health 2006 ; 34 : 159 – 67 . Google Scholar Crossref Search ADS PubMed 8 Medhus A , Kristenson H . Mortality among male disability pensioners . Scand J Soc Med 1977 ; 5 : 73 – 5 . Google Scholar Crossref Search ADS PubMed 9 Gubéran E , Usel M . Permanent work incapacity, mortality and survival without work incapacity among occupations and social classes: a cohort study of ageing men in Geneva . Int J Epidemiol 1998 ; 27 : 1026 – 32 . Google Scholar Crossref Search ADS PubMed 10 Kaprio J , Sarna S , Fogelholm M , et al. Total and occupationally active life expectancies in relation to social class and marital status in men classified as healthy at 20 in Finland . J Epidemiol Community Health 1996 ; 50 : 653 – 60 . Google Scholar Crossref Search ADS PubMed 11 Hasle H , Jeune B , Skytthe A . Differential mortality among semiskilled applicants of disability pension . Scand J Soc Med 1988 ; 16 : 273 – 6 . Google Scholar Crossref Search ADS PubMed 12 Jeune B . Survival experience of semi-skilled disability pensioners in Denmark . Scand J Soc Med 1982 ; 10 : 73 – 6 . Google Scholar Crossref Search ADS PubMed 13 Damlund M , Gøth S , Hasle P , et al. The incidence of disability pensions and mortality among semi-skilled construction workers in Copenhagen. A retrospective cohort study with two control groups . Scand J Soc Med 1982 ; 10 : 43 – 7 . Google Scholar Crossref Search ADS PubMed 14 Vingård E , Alexanderson K , Norlund A . Swedish Council on Technology Assessment in Health Care (SBU) . Chapter 9. Consequences of being on sick leave . Scand J Public Health 2004 ; 32(Suppl) : 207 – 15 . Google Scholar Crossref Search ADS 15 Hult C , Stattin M , Janlert U , et al. Comparing mortality rates and recognizing health selection bias: a response to Wallman and Svärdsudd . Soc Sci Med 2010 ; 70 : 1489 – 91 . Google Scholar Crossref Search ADS 16 World Health Organization . International Statistical Classification of Diseases and Related Health Problems . Geneva : World Health Organization , 2004 . 17 Ranganathan P , Pramesh CS . Censoring in survival analysis: potential for bias . Perspect Clin Res 2012 ; 3 : 40 . Google Scholar Crossref Search ADS PubMed 18 Howe CJ , Cole SR , Chmiel JS , et al. Limitation of Inverse Probability-of-Censoring Weights in Estimating Survival in the Presence of Strong Selection Bias . Am J Epidemiol 2011 ; 173 : 569 – 77 . Google Scholar Crossref Search ADS PubMed 19 Harre FE , Lee KL , Pollock BG . Regression Models in Clinical Studies: determining Relationships Between Predictors and Response . J Natl Cancer Inst 1988 ; 80 : 1198 – 202 . Google Scholar Crossref Search ADS PubMed 20 Akaike H . A new look at the statistical model identification . IEEE Trans Autom Control 1974 ; 19 : 716 – 23 . Google Scholar Crossref Search ADS 21 Royston P , Parmar MKB . Flexible parametric proportional-hazards and proportional-odds models for censored survival data, with application to prognostic modelling and estimation of treatment effects . Stat Med 2002 ; 21 : 2175 – 97 . Google Scholar Crossref Search ADS PubMed 22 Jackson C . flexsurv: Flexible parametric survival and multi-state models. Available at: http://cran.r-project.org/web/packages/flexsurv/index.html ( 2014 , accessed 20 March 2015, date last accessed). 23 OECD . Sickness, Disability and Work: Breaking the Barriers: Sweden Will the Recent Reforms Make It? Paris : OECD Publishing , 2010 . 24 Olsson D , Alexanderson K , Bottai M . Sickness absence and the time-varying excess risk of premature death: a Swedish population-based prospective cohort study . J Epidemiol Community Health 2015 ; 69 : 1052 – 7 . Google Scholar Crossref Search ADS PubMed © The Author(s) 2018. Published by Oxford University Press on behalf of the European Public Health Association. All rights reserved. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)

Journal

The European Journal of Public HealthOxford University Press

Published: Oct 1, 2018

References

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off