Frailty index as a predictor of mortality: a systematic review and meta-analysis

Frailty index as a predictor of mortality: a systematic review and meta-analysis Abstract Background two popular operational definitions of frailty, the frailty phenotype and Frailty index (FI), are based on different theories. Although FI was shown to be superior in predicting mortality to the frailty phenotype, no meta-analysis on mortality risk according to FI has been found in the literature. Methods an electronic systematic literature search was conducted in August 2016 using four databases (Embase, Medline, CINAHL and PsycINFO) for prospective cohort studies published in 2000 or later, examining the mortality risk according to frailty measured by FI. A meta-analysis was performed to synthesise pooled mortality risk estimates. Results of 2,617 studies identified by the systematic review, 18 cohorts from 19 studies were included. Thirteen cohorts showed hazard ratios (HRs) per 0.01 increase in FI, six cohorts showed HRs per 0.1 increase in FI and two cohorts each showed odds ratios (ORs) per 0.01 and 0.1 increase in FI, respectively. All meta-analyses suggested that higher FI was significantly associated with higher mortality risk (pooled HR per 0.01 FI increase = 1.039, 95% CI = 1.033–1.044, P < 0.001; pooled HR per 0.1 FI increase = 1.282, 95% CI = 1.258–1.307, P < 0.001; pooled OR per 0.01 FI increase = 1.054, 95% CI = 1.040–1.068, P < 0.001; pooled OR per 0.1 FI increase = 1.706, 95% CI = 1.547–1.881, P < 0.001). Meta-regression analysis among 13 cohorts with HR per 0.01 increase in FI showed that the studies with shorter follow-up periods and with lower female proportion were associated with higher mortality risks by FI. Conclusions this systematic review and meta-analysis was the first to quantitatively demonstrate that frailty measured by the FI is a significant predictor of mortality. community-dwelling older people, systematic review, death, frailty, frailty Index, mortality Introduction Frailty has been gaining increasing scientific attention over the last few decades. Frailty is generally considered to be a state characterised by reduced physiological reserve and loss of resistance to stressors caused by accumulated age-related deficits [1]. It has been shown that those who are frail are predisposed to various negative health outcomes, such as falls, fractures, hospitalisation, nursing home placement, disability, poor quality of life and dementia [2–8]. Two of the most popular operational definitions of frailty are the frailty phenotype by Fried et al. [9], using data from the Cardiovascular Health Study, and the Frailty index (FI) by Mitnitski et al. [10], using the Canadian Study of Health and Aging (CSHA) [10]. These two approaches are based on different theories [11]. frailty phenotype describes frailty as a biological syndrome with specific phenotypic presentations and defines frailty as having three or more of five physical components: unintentional weight loss, self-reported exhaustion, weakness, slow walking speed and low physical activity [9]. The frailty phenotype is a well-validated and the most frequently used measure in research and clinical practice. On the other hand, this definition has been criticised for being quite narrow in focus, and for not including potentially important components of frailty such as cognitive impairment [1, 12, 13]. By contrast, the concept of the FI is that frailty is a state caused by the accumulation of health deficits during the life course and that the more deficits one has, the more likely one is to be frail [10]. The FI is calculated as a ratio of the number of deficits present to the number of total deficits considered [10]. The deficits can be symptoms, signs, diseases, disabilities, laboratory, radiographic or electrocardiographic abnormalities and social characteristics [14]. While the exact operationalisation of the FI has varied between studies, standard criteria for constructing a FI are used [14]. Frailty is a strong predictor of mortality [1], as has been shown by previous systematic reviews [15–17]. Two of these reviews systematically collected studies that used different frailty definitions, including frailty phenotype and the FI, and demonstrated that frailty consistently increased the risk of death in most studies [15, 16]. These reviews just listed mortality risk estimates per different units of the FI from the original papers, therefore, it is not possible to directly compare these estimates and no meta-analysis was conducted [15, 16]. The third paper conducted a meta-analysis using the data from only studies using frailty phenotype and showed frailty and pre-frailty significantly predicted mortality in a graded manner [17]. Although the FI was shown to be superior in predicting mortality and other health outcome risks to frailty phenotype in a head-to-head comparison [18, 19], to the best of our knowledge, no meta-analysis on mortality risk according to the FI has been found in the literature. This may be partially because the previous studies provided mortality risks according to different units of the FI, such as per 0.01 of the FI, 0.1 of the FI or per additional deficit, or according to frailty groups based on arbitrary cut-points of the FI. Therefore, the objectives of this study are as follows: (i) to conduct a systematic search of the literature for prospective studies examining mortality risk according to frailty defined by the FI; and (ii) to combine the effect sizes to synthesise pooled risk estimates of mortality by standard units of the FI, per 0.01 or 0.1 of the FI’s increment. Method Data source and search strategy An electronic systematic literature search was conducted in August 2016 by a clinician researcher (GK) based on a protocol developed according to the PRISMA statements [20]. Embase, Medline, CINAHL Plus and PsycINFO were searched for studies published in 2000, given that the first FI paper was published in 2001 [10], or later using a combination of Medical Subject Heading (MeSH) and text terms without language restriction. The search terms used were (‘Mortality (MeSH)’ OR ‘Death (MeSH)’ OR ‘Death and Dying’ OR ‘mortality’ OR ‘death*’) AND (‘Rockwood K (as author)’ OR ‘Mitnitski A (as author)’ OR ‘Rockwood’ OR ‘Mitnitski’ OR ‘frailty index’ OR ‘FI’). The names of Professors Rockwood and Mitnitski were used as a search term as they developed the FI and have since published multiple papers using the FI. We also repeated the literature search in July 2017 using ‘accumulated deficit*’, ‘cumulative deficit*’ and ‘deficit accumulation’ along with abovementioned mortality related terms for additional studies. References of the relevant articles and reviews were also reviewed for additional studies. Forward citation tracking was also conducted on Google Scholar website for the three previous review papers [15–17]. Eligibility criteria The following inclusion and exclusion criteria were used. Inclusion criteria: Prospective study design Adult population with mean age of 20 or greater More than half of the cohort in the community (CSHA included approximately 10% of institutionalised people [10]) Baseline frailty defined by the FI constructed according to the published standard methodology [14] Subsequent all-cause mortality risk assessed as hazard ratio (HR) or odds ratio (OR) per 0.01 or 0.1 increase in FI Exclusion criteria: Selected populations, such as ones with a certain disease or medical condition Mortality risk per additional deficit or per worsening of frailty subgroups, such as by tertile or arbitrary cut-points Conference presentations, review articles, editorials, comments or dissertations Study selection The studies identified by the systematic review were assessed using the above inclusion and exclusion criteria by one author (G.K.). Initially the titles and abstracts were reviewed, and full texts were retrieved for articles that were considered to be eligible or to need a further assessment for eligibility. The full texts and reference lists were examined to identify potentially eligible studies. The original authors were contacted for clarification, if needed. If multiple studies showed the same effect measures using the same cohort, or one study provided multiple results with different conditions, such as for different follow-up periods, the results with the larger number of cohorts, the larger number of deficits used to construct the FI, or longer durations were selected. Each cohort only contributed data once per meta-analysis. Data extraction Data extracted from the included studies by the author (G.K.), using a standardised form, were first author, study name if any, publication year, location, population characteristic, sample size, proportion of female participants, mean age, age range, number of deficits used to create the FI and follow-up period. HRs or ORs of all-cause mortality per 0.01 or 0.1 increase in the FI along with 95% confidence interval (CI) were also collected. The effect measures adjusted confounders were preferred over crude ones. See Appendix 1 for methodological quality assessment and statistical analysis. Results Selection processes The systematic search of the literature using four electronic databases (Embase, MEDLINE, CINAHL Plus and PsycINFO) yielded 2,611 studies. Six additional studies were found by other source. Of the 2,617 studies, 651 duplicate studies were excluded. The title and abstract screening further excluded 1,891 studies, leaving 75 studies. Full-text review of these 75 studies excluded 56, due to the following reasons: no HR or OR for mortality provided (n = 25); effect measures per change in frailty groups based on the FI (n = 17); effect measures per each additional deficit (n = 4); non-standard FI used (n = 3); the same cohort used (n = 3); selected population (hospitalised patients) (n = 1);unit of the FI for effect measures not clearly documented (n = 3). Among these excluded studies, the findings of 28 studies providing mortality risks as HR or OR by frailty status based on the FI in general adult populations were summarised in Appendix 2. All the studies consistently showed worse frailty status defined by the FI in various ways, such as per deficit or grouping, was significantly associated with higher mortality risks. Nineteen studies were left (the references are listed in Appendix 3) and assessed for methodological quality using the modified 8-item Newcastle–Ottawa scale. All studies met five or more of the eight items and were considered to have adequate methodological quality (range = 5–7, mean = 6.1). Two studies provided HR per 0.01 increase in the FI using the Survey of Health, Ageing and Retirement in Europe (SHARE) [23, 24]. The study with the larger number (n = 37,546) showed that all of adjusted hazard ratio and upper and lower limits of 95% CI were the same at 1.04 (aHR = 1.04, 95%CI = 1.04–1.04) [23], which was not possible to be included in the meta-analysis. Therefore, the other study (n = 36,306) was used instead (aHR = 1.05, 95%CI = 1.05–1.06) [24]. A study showed 2-year, 4-year and 7-year mortality risks (age- and gender-adjusted HRs = 1.04 (95%CI = 1.03–1.04), 1.03 (95%CI = 1.03–1.04) and 1.03 (95%CI = 1.03–1.03), respectively) [25]. Since the 7-year mortality HR could not be used for the same reason above, the 4-year mortality HR was used for the meta-analysis instead. One study was included after confirmation with the study authors regarding a FI unit used to calculate the effect measures (HR per 0.1 increase in the FI) [26]. Additional data (HR per 0.01 increase in the FI) were also provided by the authors of this study [26] and included in the meta-analysis. Four series of meta-analyses were conducted for HR per 0.01 increase in the FI (n = 12), HR per 0.1 increase in the FI (n = 4), OR per 0.01 increase in the FI (n = 2) and OR per 0.1 increase in the FI (n = 2). A flow chart of the systematic literature review is shown in Figure 1. Figure 1. View largeDownload slide Flow chart of systematic literature review Figure 1. View largeDownload slide Flow chart of systematic literature review Characteristics of selected studies Table 1 presents characteristics and outcomes of the included studies. A total of 18 cohorts were used by 19 studies, which were summarised according to unit of the FI used to calculate effect measures (HR per 0.01 of the FI, HR per 0.1 of the FI, OR per 0.01 of the FI, OR per 0.1 of the FI). Four cohorts from Canada were used by six studies [23, 27–31], three cohorts from the UK were used by two studies [32, 33], four cohorts from the US were used by four studies [14, 18, 34, 35], four cohorts from China were used by three studies [25, 26, 36], two cohorts, both of which consisted of multinational European populations, were used by three studies [24, 37, 38] and lastly one Dutch cohort was used by one study [39]. The sample sizes ranged from 754 [14] to 36,306 [24]. Two female only cohorts were used by three studies [28, 29, 32] and two male only cohorts were used by three studies [35, 37, 38]. The remaining cohorts were mixed with approximately 50–70% women. The number of deficits used to create the FI ranged from 23 [23] to 70 [24, 31]. The follow-up periods varied with the shortest of 2 years [24, 33] and the longest of 19 years [39]. Twelve studies provided HR for mortality risk per 0.01 increase in the FI for 13 cohorts [14, 18, 23, 25–27, 30, 31, 34, 36, 37, 39], four studies provided HR per 0.1 increase in the FI for six cohorts [26, 31, 32, 38], two studies provided OR per 0.01 increase in the FI for two cohorts [28, 33], and two studies provided OR per 0.1 increase in the FI for two cohorts [29, 35]. All included studies provided effect measures adjusted for at least age and gender, or age only in male only or female only cohorts, except for one study [18] providing an unadjusted effect measure. Table 1. Summary of included studies on Frailty Index and mortality Author/Study  Year  Location  Sample size  Female (%)  Age (range)  Number of deficits  Follow-up period  Risk estimate HR/OR (95% CI)  Adjustment  HR per 0.01 of FI   Searle Yale-PEP  2008  USA  754  64.6  –(72–98)  40  9 years  aHR = 1.03 (1.02–1.04)  Age, gender   Kulminski Cardiovascular Health Study  2008  USA  1,073  –  –(>65)  48  4 years  HR = 1.049 (1.040–1.057)  Unadjusted   Rockwood National Population Health Survey  2011  Canada  14,127  54.2  44 (>15)  42  14 years  aHR = 1.04 (1.03–1.04)  Age, gender, education   Yu Beijing Longitudinal Study of Aging (Urban sample)  2012  China  2,136  51.1  70.1 (55–97)  35  8 years  aHR = 1.042 (1.036–1.049)  Age, gender, education   Yu Beijing Longitudinal Study of Aging (Rural sample)  2012  China  1,121  51.0  70.2–70.3 (55–97)  35  8 years  aHR = 1.041 (0.034–1.049)  Age, gender, education   Bennett Chinese Longitudinal Healthy Longevity Survey  2013  China  6,300  53.0  88.9 (80–99)  38  4 years  aHR = 1.03 (1.03–1.04)  Age, gender   Theou SHARE  2013  Europea  36,306  54.6  65.2 (>50)  70  2 years  aHR = 1.05 (1.05–1.06)  Age, gender   Pena Nova Scotia Health Survey  2014  Canada  3,227  50.1  48.1 (>18)  23  10 years  aHR = 1.04 (1.03–1.05)  Age, gender   Blodgett EMAS  2016  Europeb  2,933  0  60.2 (40–79)  39  4.4 years  aHR = 1.07 (1.06–1.09)  Age   Hao Project of Longevity and Aging in Dujiangyan  2016  China  767  68.0  93.7 (90–108)  35  4 years  aHR = 1.03 (1.02–1.04)  Age, gender, education   Hoogendijk Longitudinal Aging Study Amsterdam  2016  Netherlands  2,218  –  –(57–88)  32  19 years  aHR = 1.03 (1.03–1.04)  Age, gender   Miller NHANES  2016  USA  8,911  –  –(20–)  46  8 years  aHR = 1.03 (1.02–1.04)  Age, gender   Mitnitski CSHA  2016  Canada  1,013  61.6  80.8 (>65)  61  6 years  aHR = 1.041 (1.030–1.052)  Age, gender  HR per 0.1 of FI   Kamaruzzaman BWHHS  2010  UK  4,286  100  –(60–79)  44  8.2 years  aHR = 1.3 (1.2–1.4)  Age, socioeconomic status, smoking, alcohol, marital status, living alone, housing tenure   Kamaruzzaman MRC assessment study  2010  UK  11,195  59.9  –(>75)  44  7.9 years  aHR = 1.3 (1.2–1.3)  Age, gender, smoking, alcohol, marital status, living alone, social contact, housing tenure   Theou CSHA  2012  Canada  2,305  62.1  84.6 (70–105)  70  5 years  aHR = 1.25 (1.20–1.30)  Age, gender   Yu Beijing Longitudinal Study of Aging (Urban sample)  2012  China  2,136  51.1  70.1 (55–97)  35  8 years  aHR = 1.28 (1.23–1.32)  Age, gender, education   Yu Beijing Longitudinal Study of Aging (Rural sample)  2012  China  1,121  51.0  70.2–70.3 (55–97)  35  8 years  aHR = 1.27 (1.21–1.32)  Age, gender, education   Rivindrarajah EMAS  2013  Europeb  2,929  0  59.9 (40–79)  39  4.3 years  aHR = 1.49 (1.33–1.67)  Age, centre, smoking, partner status  OR per 0.01 of FI   Li GLOW  2014  Canada  3,985  100  69.4 (>55)  34  3 years  aOR = 1.05 (1.03–1.06)  Age, BMI, smoking, alcohol, education   Theou TILDA  2015  UK  4,961  54.2  61.9 (>50)  66  2 years  aOR = 1.072 (1.040–1.106)  Age, gender  OR per 0.1 of FI   Armstrong HAAS  2015  USA  3,845  0  77.9 (72–91)  48  6 years  aOR = 1.73 (1.57–1.92)  Age, education   Li GLOW  2016  Canada  3,985  100  69.4 (>55)  34  3 years  aOR = 1.33 (0.87–2.03)  Age  Author/Study  Year  Location  Sample size  Female (%)  Age (range)  Number of deficits  Follow-up period  Risk estimate HR/OR (95% CI)  Adjustment  HR per 0.01 of FI   Searle Yale-PEP  2008  USA  754  64.6  –(72–98)  40  9 years  aHR = 1.03 (1.02–1.04)  Age, gender   Kulminski Cardiovascular Health Study  2008  USA  1,073  –  –(>65)  48  4 years  HR = 1.049 (1.040–1.057)  Unadjusted   Rockwood National Population Health Survey  2011  Canada  14,127  54.2  44 (>15)  42  14 years  aHR = 1.04 (1.03–1.04)  Age, gender, education   Yu Beijing Longitudinal Study of Aging (Urban sample)  2012  China  2,136  51.1  70.1 (55–97)  35  8 years  aHR = 1.042 (1.036–1.049)  Age, gender, education   Yu Beijing Longitudinal Study of Aging (Rural sample)  2012  China  1,121  51.0  70.2–70.3 (55–97)  35  8 years  aHR = 1.041 (0.034–1.049)  Age, gender, education   Bennett Chinese Longitudinal Healthy Longevity Survey  2013  China  6,300  53.0  88.9 (80–99)  38  4 years  aHR = 1.03 (1.03–1.04)  Age, gender   Theou SHARE  2013  Europea  36,306  54.6  65.2 (>50)  70  2 years  aHR = 1.05 (1.05–1.06)  Age, gender   Pena Nova Scotia Health Survey  2014  Canada  3,227  50.1  48.1 (>18)  23  10 years  aHR = 1.04 (1.03–1.05)  Age, gender   Blodgett EMAS  2016  Europeb  2,933  0  60.2 (40–79)  39  4.4 years  aHR = 1.07 (1.06–1.09)  Age   Hao Project of Longevity and Aging in Dujiangyan  2016  China  767  68.0  93.7 (90–108)  35  4 years  aHR = 1.03 (1.02–1.04)  Age, gender, education   Hoogendijk Longitudinal Aging Study Amsterdam  2016  Netherlands  2,218  –  –(57–88)  32  19 years  aHR = 1.03 (1.03–1.04)  Age, gender   Miller NHANES  2016  USA  8,911  –  –(20–)  46  8 years  aHR = 1.03 (1.02–1.04)  Age, gender   Mitnitski CSHA  2016  Canada  1,013  61.6  80.8 (>65)  61  6 years  aHR = 1.041 (1.030–1.052)  Age, gender  HR per 0.1 of FI   Kamaruzzaman BWHHS  2010  UK  4,286  100  –(60–79)  44  8.2 years  aHR = 1.3 (1.2–1.4)  Age, socioeconomic status, smoking, alcohol, marital status, living alone, housing tenure   Kamaruzzaman MRC assessment study  2010  UK  11,195  59.9  –(>75)  44  7.9 years  aHR = 1.3 (1.2–1.3)  Age, gender, smoking, alcohol, marital status, living alone, social contact, housing tenure   Theou CSHA  2012  Canada  2,305  62.1  84.6 (70–105)  70  5 years  aHR = 1.25 (1.20–1.30)  Age, gender   Yu Beijing Longitudinal Study of Aging (Urban sample)  2012  China  2,136  51.1  70.1 (55–97)  35  8 years  aHR = 1.28 (1.23–1.32)  Age, gender, education   Yu Beijing Longitudinal Study of Aging (Rural sample)  2012  China  1,121  51.0  70.2–70.3 (55–97)  35  8 years  aHR = 1.27 (1.21–1.32)  Age, gender, education   Rivindrarajah EMAS  2013  Europeb  2,929  0  59.9 (40–79)  39  4.3 years  aHR = 1.49 (1.33–1.67)  Age, centre, smoking, partner status  OR per 0.01 of FI   Li GLOW  2014  Canada  3,985  100  69.4 (>55)  34  3 years  aOR = 1.05 (1.03–1.06)  Age, BMI, smoking, alcohol, education   Theou TILDA  2015  UK  4,961  54.2  61.9 (>50)  66  2 years  aOR = 1.072 (1.040–1.106)  Age, gender  OR per 0.1 of FI   Armstrong HAAS  2015  USA  3,845  0  77.9 (72–91)  48  6 years  aOR = 1.73 (1.57–1.92)  Age, education   Li GLOW  2016  Canada  3,985  100  69.4 (>55)  34  3 years  aOR = 1.33 (0.87–2.03)  Age  a15 European countries: Austria, Belgium, Czech Republic, Denmark, France, Germany, Greece, Ireland, Israel, Italy, Netherlands, Poland, Spain, Sweden, Switzerland. b8 European countries: Belgium, Estonia, Hungary, Italy, Poland, Spain, Sweden, UK. 95% CI = 95% confidence interval. (a)HR: (adjusted) hazard ratio. (a)OR: (adjusted) odds ratio. BMI: body mass index. BWHHS: British Women’s Heart and Health Study. CSHA: Canadian Study of Health and Aging. EMAS: European Male Ageing Study. GLOW: Global Longitudinal Study of Osteoporosis in Women. FI: frailty index. NHANES: National Health and Nutrition Examination Survey. SES: socioeconomic status. SHARE: Survey of Health, Ageing and Retirement in Europe. TILDA: The Irish LongituDinal study on Ageing. Yale-PEP: Yale Precipitating Events Project. Frailty index as a predictor of mortality Meta-analysis of studies using HR HRs of mortality per 0.01 increase in the FI from the 13 cohorts were combined using a random-effects model due to the significant heterogeneity (P < 0.001, I2 = 86%). Frailty was a significant predictor of mortality (13 cohorts: pooled HR = 1.039, 95% CI = 1.033–1.044, P < 0.001). Combining HRs per 0.1 increase in the FI from six cohorts using a fixed-effect model (heterogeneity P = 0.11, I2 = 45%) also showed that frailty significantly predicted mortality (six cohorts: pooled HR = 1.282, 95% CI = 1.258–1.307, P < 0.001) (Figure 2A and B). Figure 2. View largeDownload slide Forest plots of mortality risk according to frailty measured by the Frailty index. A: Risk of dying (hazard ratio) per 0.01 increase in the Frailty index score. CI: confidence interval, IV: inverse variance, NSHS: Nova Scotia Health Survey. B: Risk of dying (hazard ratio) per 0.1 increase in the Frailty index score. BWHHS: British Women’s Heart and Health Study, CI: confidence interval, IV: inverse variance, MRC: MRC assessment study. Figure 2. View largeDownload slide Forest plots of mortality risk according to frailty measured by the Frailty index. A: Risk of dying (hazard ratio) per 0.01 increase in the Frailty index score. CI: confidence interval, IV: inverse variance, NSHS: Nova Scotia Health Survey. B: Risk of dying (hazard ratio) per 0.1 increase in the Frailty index score. BWHHS: British Women’s Heart and Health Study, CI: confidence interval, IV: inverse variance, MRC: MRC assessment study. Meta-analysis of studies using OR Four studies provided OR as a risk measure of mortality. Two studies showed ORs per 0.01 increase in the FI [28, 33] and another two studies showed ORs per 0.1 increase in the FI [29, 35]. fixed-effects models were used (heterogeneity P = 0.23 and 0.24, I2 = 30 and 29%, respectively) and both showed that frailty is a significant predictor of mortality (Appendix 4A and B). See Appendix 1 for meta-regression and subgroup analysis and publication bias assessment. Discussion The current study identified 19 studies that longitudinally examined mortality risk according to frailty measured by the FI in 18 cohorts and provided the effect measured as HR or OR per 0.01 or 0.1 increase in the FI. The meta-analysis quantitatively combined mortality risks based on frailty measured by the FI and consistently showed increased mortality risk according to the FI regardless of different types of the effect sizes and per units of the FI. Although the included studies constructed the FI based on different numbers and types of deficits, in addition to various populations and study settings, it is of note that the effect measures were in relatively narrow ranges and may support the robustness of this accumulation deficit frailty model. Although in general age is a strong predictor of mortality, the mean age of the cohorts was not a significant modulator in the association between the FI and mortality in the meta-regression analysis. Furthermore, subgroup analysis also showed that pooled estimates of studies with a mean age of >65 [14, 18, 24–26, 30, 36] and <65 [23, 27, 37] (mostly middle aged with the mean age ranging from 44 to 60.2) were almost identical (8 cohorts: pooled HR = 1.04, 95% CI = 1.03–1.05, P < 0.001, I2 = 84%, 3 cohorts: pooled HR = 1.05, 95% CI = 1.03–1.07, P < 0.001, I2 = 92%, respectively). This suggests the FI is a good indicator of mortality risk not only among older people but also among younger populations, regardless of age. Two study characteristics were found in the meta-regression analysis to be related to the association between frailty and mortality: follow-up period and female proportion. In general, women live longer but have more disabilities than men, known as the male–female health-survival paradox [40]. Given the FI can be regarded as a measure of biological age [10] and prevalence of frailty is higher among women than men [9], it is to be expected that female gender is associated with lower mortality risk according to frailty in the meta-regression analysis. Regarding the follow-up period, the meta-regression analysis suggests shorter follow-up periods are associated with higher mortality risk according to the FI. Frailty is a dynamic state and known to change over time, mostly worsening rather than improving [41]. The longer follow-up periods imply that as participants get older they usually get frailer. This may be why the reason the association between frailty and mortality became less prominent in studies with longer follow-up periods. The studies using the same cohorts with different lengths of follow-up showed overall comparable results with little difference [14, 23, 24]. In SHARE, 2-year mortality (aHR = 1.05) [24] was slightly higher than 5-year mortality (aHR = 1.04) [23], while 9-year mortality (aHR = 1.03) [14] was slightly lower than 12-year mortality (aHR = 1.04) [23] in the Yale Precipitating Events Project. This study’s findings should be interpreted with caution due to some limitations. First, all processes of the systematic review and meta-analysis were conducted by one investigator. Second, during the study selection, a large number of studies that used the FI to examine mortality risk were excluded because they did not provide HR or OR for mortality (n = 25); the effect measures provided were based on frailty groups defined by different cut-off points (n = 17); or on each additional deficit (n = 4). Although not all, at least some of them could potentially have been included in the meta-analysis. Lastly, the effect measures and upper and lower limits of 95% CI in many of the included papers were rounded to two decimal places, which could potentially lead to a miscalculation of standard error or weighting in the meta-analysis, especially when effect measures were calculated per 0.01 increase in the FI and were therefore relatively smaller. The current study has multiple strengths. The search strategy of the systematic review of the literature was robust and reproducible, using comprehensive search terms in multiple electronic databases. Additional data were also acquired from the original study’s authors [26]. The included studies were also assessed for heterogeneity, methodological quality and publication bias, and a high degree of heterogeneity was further explored by meta-regression analysis and subgroup analysis. The data from included studies were based on a FI constructed according to the standard methodology [14]. and were mostly controlled for important confounders, age and gender, or age in male only or female only cohorts. Other potential confounders would include education, socioeconomic status, smoking and alcohol consumption. In the subgroup analysis, there was no significant difference in mortality risk between studies adjusting for age and gender or age only and studies additionally adjusting for such confounders (8 cohorts: pooled HR = 1.04, 95% CI = 1.03–1.05, P < 0.001, I2 = 89%, four cohorts: pooled HR = 1.04, 95% CI = 1.03–1.04, P < 0.001, I2 = 74%, respectively. P for subgroup difference = 0.53). Lastly this is the first systematic review and meta-analysis focusing on FI and mortality. There are several features of the FI which distinguish it from frailty phenotype. As mentioned above, the FI can evaluate frailty status in a graded manner, rather than just three frailty categorisations by frailty phenotype (robust, pre-frail and frail), and make a more precise risk prediction. Furthermore, those who have a missing value for specific frailty components may be excluded from analyses in frailty phenotype. However the FI can still be calculated by excluding missing deficits from both numerator and denominator, which is because deficits are considered to be interchangeable if a sufficiently large number of deficits are included [42]. Although one may argue that it is not practical in clinical settings to collect information of 30 or more health deficits to calculate the FI, most of the clinical information could be extracted from electronic medical record systems. A recent study created an electronic FI from readily available data in primary care electronic records and demonstrated robust predictive ability for mortality, hospitalisation and nursing home placement [43]. This systematic review and meta-analysis was the first to quantitatively demonstrate the pooled mortality risk estimate according to frailty defined by the FI. Frailty measured by the FI is a strong predictor of death among older people as well as younger and middle-aged populations. A shorter follow-up period and lower female proportion seem to be associated with higher mortality risks according to frailty. Key points The Frailty Index is among the most popular frailty definitions and predicative of mortality. The mortality risk according to the Frailty index has never been quantified with meta-analysis in the literature. 14 studies providing hazard ratios or odds ratios per 0.01 or 0.1 increase in the Frailty index were included. All meta-analyses suggested that frailty measured by the Frailty index is a significant predictor of mortality. Studies conducted in Europe, with shorter follow-up periods and lower female proportion had higher mortality risks. Supplementary Data Supplementary data mentioned in the text are available to subscribers in Age and Ageing online. Funding None. Conflict of interest None. References Note: The very long list of references supporting this article has meant that only the most important are listed here. The full list of references is available in the Supplementary data. 1 Clegg A, Young J, Iliffe S, Rikkert MO, Rockwood K. Frailty in elderly people. Lancet  2013; 381: 752– 62. Google Scholar CrossRef Search ADS PubMed  2 Kojima G. Frailty as a predictor of future falls among community-dwelling older people: a systematic review and meta-analysis. J Am Med Dir Assoc  2015; 16: 1027– 33. Google Scholar CrossRef Search ADS PubMed  3 Kojima G. Frailty as a predictor of fractures among community-dwelling older people: a systematic review and meta-analysis. Bone  2016; 90: 116– 22. Google Scholar CrossRef Search ADS PubMed  4 Kojima G. Frailty as a predictor of hospitalisation among community-dwelling older people: a systematic review and meta-analysis. J Epidemiol Community Health  2016; 70: 722– 9. Google Scholar CrossRef Search ADS PubMed  5 Kojima G. Frailty as a predictor of nursing home placement among community-dwelling older adults: a systematic review and meta-analysis. J Geriatr Phys Ther  2016. 6 Kojima G. Frailty as a predictor of disabilities among community-dwelling older people: a systematic review and meta-analysis. Disabil Rehabil  2016; 39: 1897– 908. Google Scholar CrossRef Search ADS PubMed  7 Kojima G, Iliffe S, Jivraj S, Walters K. Association between frailty and quality of life among community-dwelling older people: a systematic review and meta-analysis. J Epidemiol Community Health  2016; 70: 716– 21. Google Scholar CrossRef Search ADS PubMed  8 Kojima G, Taniguchi Y, Iliffe S, Walters K. Frailty as a predictor of alzheimer disease, vascular dementia, and all dementia among community-dwelling older people: a systematic review and meta-analysis. J Am Med Dir Assoc  2016; 17: 881– 8. Google Scholar CrossRef Search ADS PubMed  9 Fried LP, Tangen CM, Walston J et al.  . Frailty in older adults: evidence for a phenotype. J Gerontol A Biol Sci Med Sci  2001; 56: M146– 56. Google Scholar CrossRef Search ADS PubMed  10 Mitnitski AB, Mogilner AJ, Rockwood K. Accumulation of deficits as a proxy measure of aging. Sci World J  2001; 1: 323– 36. Google Scholar CrossRef Search ADS   11 Walston JD, Bandeen-Roche K. Frailty: a tale of two concepts. BMC Med  2015; 13: 185. Google Scholar CrossRef Search ADS PubMed  12 Avila-Funes JA, Amieva H, Barberger-Gateau P et al.  . Cognitive impairment improves the predictive validity of the phenotype of frailty for adverse health outcomes: the three-city study. J Am Geriatr Soc  2009; 57: 453– 61. Google Scholar CrossRef Search ADS PubMed  13 Rothman MD, Leo-Summers L, Gill TM. Prognostic significance of potential frailty criteria. J Am Geriatr Soc  2008; 56: 2211– 16. Google Scholar CrossRef Search ADS PubMed  15 Kane RL, Shamliyan T, Talley K, Pacala J. The association between geriatric syndromes and survival. J Am Geriatr Soc  2012; 60: 896– 904. Google Scholar CrossRef Search ADS PubMed  16 Shamliyan T, Talley KM, Ramakrishnan R, Kane RL. Association of frailty with survival: a systematic literature review. Ageing Res Rev  2013; 12: 719– 36. Google Scholar CrossRef Search ADS PubMed  17 Chang SF, Lin PL. Frail phenotype and mortality prediction: a systematic review and meta-analysis of prospective cohort studies. Int J Nurs Stud  2015; 52: 1362– 74. Google Scholar CrossRef Search ADS PubMed  19 Rockwood K, Andrew M, Mitnitski A. A comparison of two approaches to measuring frailty in elderly people. J Gerontol A Biol Sci Med Sci  2007; 62: 738– 43. Google Scholar CrossRef Search ADS PubMed  20 Moher D, Liberati A, Tetzlaff J, Altman DG. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. Br Med J  2009; 339: b2535. Google Scholar CrossRef Search ADS   40 Weinstein M, Lane MA, eds. Sociality, Hierarchy, Health: Comparative Biodemography: A Collection of Papers . Washington DC: 2014 by the National Academy of Sciences, 2014. Google Scholar PubMed PubMed  41 Mitnitski A, Song X, Rockwood K. Trajectories of changes over twelve years in the health status of Canadians from late middle age. Exp Gerontol  2012; 47: 893– 9. Google Scholar CrossRef Search ADS PubMed  42 Rockwood K, Mitnitski A, Song X, Steen B, Skoog I. Long-term risks of death and institutionalization of elderly people in relation to deficit accumulation at age 70. J Am Geriatr Soc  2006; 54: 975– 9. Google Scholar CrossRef Search ADS PubMed  43 Clegg A, Bates C, Young J et al.  . Development and validation of an electronic frailty index using routine primary care electronic health record data. Age Ageing  2016; 45: 353– 60. Google Scholar CrossRef Search ADS PubMed  © The Author 2017. Published by Oxford University Press on behalf of the British Geriatrics Society.All rights reserved. For permissions, please email: journals.permissions@oup.com http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Age and Ageing Oxford University Press

Frailty index as a predictor of mortality: a systematic review and meta-analysis

Loading next page...
 
/lp/ou_press/frailty-index-as-a-predictor-of-mortality-a-systematic-review-and-meta-XujOQJeLHF
Publisher
Oxford University Press
Copyright
© The Author 2017. Published by Oxford University Press on behalf of the British Geriatrics Society.All rights reserved. For permissions, please email: journals.permissions@oup.com
ISSN
0002-0729
eISSN
1468-2834
D.O.I.
10.1093/ageing/afx162
Publisher site
See Article on Publisher Site

Abstract

Abstract Background two popular operational definitions of frailty, the frailty phenotype and Frailty index (FI), are based on different theories. Although FI was shown to be superior in predicting mortality to the frailty phenotype, no meta-analysis on mortality risk according to FI has been found in the literature. Methods an electronic systematic literature search was conducted in August 2016 using four databases (Embase, Medline, CINAHL and PsycINFO) for prospective cohort studies published in 2000 or later, examining the mortality risk according to frailty measured by FI. A meta-analysis was performed to synthesise pooled mortality risk estimates. Results of 2,617 studies identified by the systematic review, 18 cohorts from 19 studies were included. Thirteen cohorts showed hazard ratios (HRs) per 0.01 increase in FI, six cohorts showed HRs per 0.1 increase in FI and two cohorts each showed odds ratios (ORs) per 0.01 and 0.1 increase in FI, respectively. All meta-analyses suggested that higher FI was significantly associated with higher mortality risk (pooled HR per 0.01 FI increase = 1.039, 95% CI = 1.033–1.044, P < 0.001; pooled HR per 0.1 FI increase = 1.282, 95% CI = 1.258–1.307, P < 0.001; pooled OR per 0.01 FI increase = 1.054, 95% CI = 1.040–1.068, P < 0.001; pooled OR per 0.1 FI increase = 1.706, 95% CI = 1.547–1.881, P < 0.001). Meta-regression analysis among 13 cohorts with HR per 0.01 increase in FI showed that the studies with shorter follow-up periods and with lower female proportion were associated with higher mortality risks by FI. Conclusions this systematic review and meta-analysis was the first to quantitatively demonstrate that frailty measured by the FI is a significant predictor of mortality. community-dwelling older people, systematic review, death, frailty, frailty Index, mortality Introduction Frailty has been gaining increasing scientific attention over the last few decades. Frailty is generally considered to be a state characterised by reduced physiological reserve and loss of resistance to stressors caused by accumulated age-related deficits [1]. It has been shown that those who are frail are predisposed to various negative health outcomes, such as falls, fractures, hospitalisation, nursing home placement, disability, poor quality of life and dementia [2–8]. Two of the most popular operational definitions of frailty are the frailty phenotype by Fried et al. [9], using data from the Cardiovascular Health Study, and the Frailty index (FI) by Mitnitski et al. [10], using the Canadian Study of Health and Aging (CSHA) [10]. These two approaches are based on different theories [11]. frailty phenotype describes frailty as a biological syndrome with specific phenotypic presentations and defines frailty as having three or more of five physical components: unintentional weight loss, self-reported exhaustion, weakness, slow walking speed and low physical activity [9]. The frailty phenotype is a well-validated and the most frequently used measure in research and clinical practice. On the other hand, this definition has been criticised for being quite narrow in focus, and for not including potentially important components of frailty such as cognitive impairment [1, 12, 13]. By contrast, the concept of the FI is that frailty is a state caused by the accumulation of health deficits during the life course and that the more deficits one has, the more likely one is to be frail [10]. The FI is calculated as a ratio of the number of deficits present to the number of total deficits considered [10]. The deficits can be symptoms, signs, diseases, disabilities, laboratory, radiographic or electrocardiographic abnormalities and social characteristics [14]. While the exact operationalisation of the FI has varied between studies, standard criteria for constructing a FI are used [14]. Frailty is a strong predictor of mortality [1], as has been shown by previous systematic reviews [15–17]. Two of these reviews systematically collected studies that used different frailty definitions, including frailty phenotype and the FI, and demonstrated that frailty consistently increased the risk of death in most studies [15, 16]. These reviews just listed mortality risk estimates per different units of the FI from the original papers, therefore, it is not possible to directly compare these estimates and no meta-analysis was conducted [15, 16]. The third paper conducted a meta-analysis using the data from only studies using frailty phenotype and showed frailty and pre-frailty significantly predicted mortality in a graded manner [17]. Although the FI was shown to be superior in predicting mortality and other health outcome risks to frailty phenotype in a head-to-head comparison [18, 19], to the best of our knowledge, no meta-analysis on mortality risk according to the FI has been found in the literature. This may be partially because the previous studies provided mortality risks according to different units of the FI, such as per 0.01 of the FI, 0.1 of the FI or per additional deficit, or according to frailty groups based on arbitrary cut-points of the FI. Therefore, the objectives of this study are as follows: (i) to conduct a systematic search of the literature for prospective studies examining mortality risk according to frailty defined by the FI; and (ii) to combine the effect sizes to synthesise pooled risk estimates of mortality by standard units of the FI, per 0.01 or 0.1 of the FI’s increment. Method Data source and search strategy An electronic systematic literature search was conducted in August 2016 by a clinician researcher (GK) based on a protocol developed according to the PRISMA statements [20]. Embase, Medline, CINAHL Plus and PsycINFO were searched for studies published in 2000, given that the first FI paper was published in 2001 [10], or later using a combination of Medical Subject Heading (MeSH) and text terms without language restriction. The search terms used were (‘Mortality (MeSH)’ OR ‘Death (MeSH)’ OR ‘Death and Dying’ OR ‘mortality’ OR ‘death*’) AND (‘Rockwood K (as author)’ OR ‘Mitnitski A (as author)’ OR ‘Rockwood’ OR ‘Mitnitski’ OR ‘frailty index’ OR ‘FI’). The names of Professors Rockwood and Mitnitski were used as a search term as they developed the FI and have since published multiple papers using the FI. We also repeated the literature search in July 2017 using ‘accumulated deficit*’, ‘cumulative deficit*’ and ‘deficit accumulation’ along with abovementioned mortality related terms for additional studies. References of the relevant articles and reviews were also reviewed for additional studies. Forward citation tracking was also conducted on Google Scholar website for the three previous review papers [15–17]. Eligibility criteria The following inclusion and exclusion criteria were used. Inclusion criteria: Prospective study design Adult population with mean age of 20 or greater More than half of the cohort in the community (CSHA included approximately 10% of institutionalised people [10]) Baseline frailty defined by the FI constructed according to the published standard methodology [14] Subsequent all-cause mortality risk assessed as hazard ratio (HR) or odds ratio (OR) per 0.01 or 0.1 increase in FI Exclusion criteria: Selected populations, such as ones with a certain disease or medical condition Mortality risk per additional deficit or per worsening of frailty subgroups, such as by tertile or arbitrary cut-points Conference presentations, review articles, editorials, comments or dissertations Study selection The studies identified by the systematic review were assessed using the above inclusion and exclusion criteria by one author (G.K.). Initially the titles and abstracts were reviewed, and full texts were retrieved for articles that were considered to be eligible or to need a further assessment for eligibility. The full texts and reference lists were examined to identify potentially eligible studies. The original authors were contacted for clarification, if needed. If multiple studies showed the same effect measures using the same cohort, or one study provided multiple results with different conditions, such as for different follow-up periods, the results with the larger number of cohorts, the larger number of deficits used to construct the FI, or longer durations were selected. Each cohort only contributed data once per meta-analysis. Data extraction Data extracted from the included studies by the author (G.K.), using a standardised form, were first author, study name if any, publication year, location, population characteristic, sample size, proportion of female participants, mean age, age range, number of deficits used to create the FI and follow-up period. HRs or ORs of all-cause mortality per 0.01 or 0.1 increase in the FI along with 95% confidence interval (CI) were also collected. The effect measures adjusted confounders were preferred over crude ones. See Appendix 1 for methodological quality assessment and statistical analysis. Results Selection processes The systematic search of the literature using four electronic databases (Embase, MEDLINE, CINAHL Plus and PsycINFO) yielded 2,611 studies. Six additional studies were found by other source. Of the 2,617 studies, 651 duplicate studies were excluded. The title and abstract screening further excluded 1,891 studies, leaving 75 studies. Full-text review of these 75 studies excluded 56, due to the following reasons: no HR or OR for mortality provided (n = 25); effect measures per change in frailty groups based on the FI (n = 17); effect measures per each additional deficit (n = 4); non-standard FI used (n = 3); the same cohort used (n = 3); selected population (hospitalised patients) (n = 1);unit of the FI for effect measures not clearly documented (n = 3). Among these excluded studies, the findings of 28 studies providing mortality risks as HR or OR by frailty status based on the FI in general adult populations were summarised in Appendix 2. All the studies consistently showed worse frailty status defined by the FI in various ways, such as per deficit or grouping, was significantly associated with higher mortality risks. Nineteen studies were left (the references are listed in Appendix 3) and assessed for methodological quality using the modified 8-item Newcastle–Ottawa scale. All studies met five or more of the eight items and were considered to have adequate methodological quality (range = 5–7, mean = 6.1). Two studies provided HR per 0.01 increase in the FI using the Survey of Health, Ageing and Retirement in Europe (SHARE) [23, 24]. The study with the larger number (n = 37,546) showed that all of adjusted hazard ratio and upper and lower limits of 95% CI were the same at 1.04 (aHR = 1.04, 95%CI = 1.04–1.04) [23], which was not possible to be included in the meta-analysis. Therefore, the other study (n = 36,306) was used instead (aHR = 1.05, 95%CI = 1.05–1.06) [24]. A study showed 2-year, 4-year and 7-year mortality risks (age- and gender-adjusted HRs = 1.04 (95%CI = 1.03–1.04), 1.03 (95%CI = 1.03–1.04) and 1.03 (95%CI = 1.03–1.03), respectively) [25]. Since the 7-year mortality HR could not be used for the same reason above, the 4-year mortality HR was used for the meta-analysis instead. One study was included after confirmation with the study authors regarding a FI unit used to calculate the effect measures (HR per 0.1 increase in the FI) [26]. Additional data (HR per 0.01 increase in the FI) were also provided by the authors of this study [26] and included in the meta-analysis. Four series of meta-analyses were conducted for HR per 0.01 increase in the FI (n = 12), HR per 0.1 increase in the FI (n = 4), OR per 0.01 increase in the FI (n = 2) and OR per 0.1 increase in the FI (n = 2). A flow chart of the systematic literature review is shown in Figure 1. Figure 1. View largeDownload slide Flow chart of systematic literature review Figure 1. View largeDownload slide Flow chart of systematic literature review Characteristics of selected studies Table 1 presents characteristics and outcomes of the included studies. A total of 18 cohorts were used by 19 studies, which were summarised according to unit of the FI used to calculate effect measures (HR per 0.01 of the FI, HR per 0.1 of the FI, OR per 0.01 of the FI, OR per 0.1 of the FI). Four cohorts from Canada were used by six studies [23, 27–31], three cohorts from the UK were used by two studies [32, 33], four cohorts from the US were used by four studies [14, 18, 34, 35], four cohorts from China were used by three studies [25, 26, 36], two cohorts, both of which consisted of multinational European populations, were used by three studies [24, 37, 38] and lastly one Dutch cohort was used by one study [39]. The sample sizes ranged from 754 [14] to 36,306 [24]. Two female only cohorts were used by three studies [28, 29, 32] and two male only cohorts were used by three studies [35, 37, 38]. The remaining cohorts were mixed with approximately 50–70% women. The number of deficits used to create the FI ranged from 23 [23] to 70 [24, 31]. The follow-up periods varied with the shortest of 2 years [24, 33] and the longest of 19 years [39]. Twelve studies provided HR for mortality risk per 0.01 increase in the FI for 13 cohorts [14, 18, 23, 25–27, 30, 31, 34, 36, 37, 39], four studies provided HR per 0.1 increase in the FI for six cohorts [26, 31, 32, 38], two studies provided OR per 0.01 increase in the FI for two cohorts [28, 33], and two studies provided OR per 0.1 increase in the FI for two cohorts [29, 35]. All included studies provided effect measures adjusted for at least age and gender, or age only in male only or female only cohorts, except for one study [18] providing an unadjusted effect measure. Table 1. Summary of included studies on Frailty Index and mortality Author/Study  Year  Location  Sample size  Female (%)  Age (range)  Number of deficits  Follow-up period  Risk estimate HR/OR (95% CI)  Adjustment  HR per 0.01 of FI   Searle Yale-PEP  2008  USA  754  64.6  –(72–98)  40  9 years  aHR = 1.03 (1.02–1.04)  Age, gender   Kulminski Cardiovascular Health Study  2008  USA  1,073  –  –(>65)  48  4 years  HR = 1.049 (1.040–1.057)  Unadjusted   Rockwood National Population Health Survey  2011  Canada  14,127  54.2  44 (>15)  42  14 years  aHR = 1.04 (1.03–1.04)  Age, gender, education   Yu Beijing Longitudinal Study of Aging (Urban sample)  2012  China  2,136  51.1  70.1 (55–97)  35  8 years  aHR = 1.042 (1.036–1.049)  Age, gender, education   Yu Beijing Longitudinal Study of Aging (Rural sample)  2012  China  1,121  51.0  70.2–70.3 (55–97)  35  8 years  aHR = 1.041 (0.034–1.049)  Age, gender, education   Bennett Chinese Longitudinal Healthy Longevity Survey  2013  China  6,300  53.0  88.9 (80–99)  38  4 years  aHR = 1.03 (1.03–1.04)  Age, gender   Theou SHARE  2013  Europea  36,306  54.6  65.2 (>50)  70  2 years  aHR = 1.05 (1.05–1.06)  Age, gender   Pena Nova Scotia Health Survey  2014  Canada  3,227  50.1  48.1 (>18)  23  10 years  aHR = 1.04 (1.03–1.05)  Age, gender   Blodgett EMAS  2016  Europeb  2,933  0  60.2 (40–79)  39  4.4 years  aHR = 1.07 (1.06–1.09)  Age   Hao Project of Longevity and Aging in Dujiangyan  2016  China  767  68.0  93.7 (90–108)  35  4 years  aHR = 1.03 (1.02–1.04)  Age, gender, education   Hoogendijk Longitudinal Aging Study Amsterdam  2016  Netherlands  2,218  –  –(57–88)  32  19 years  aHR = 1.03 (1.03–1.04)  Age, gender   Miller NHANES  2016  USA  8,911  –  –(20–)  46  8 years  aHR = 1.03 (1.02–1.04)  Age, gender   Mitnitski CSHA  2016  Canada  1,013  61.6  80.8 (>65)  61  6 years  aHR = 1.041 (1.030–1.052)  Age, gender  HR per 0.1 of FI   Kamaruzzaman BWHHS  2010  UK  4,286  100  –(60–79)  44  8.2 years  aHR = 1.3 (1.2–1.4)  Age, socioeconomic status, smoking, alcohol, marital status, living alone, housing tenure   Kamaruzzaman MRC assessment study  2010  UK  11,195  59.9  –(>75)  44  7.9 years  aHR = 1.3 (1.2–1.3)  Age, gender, smoking, alcohol, marital status, living alone, social contact, housing tenure   Theou CSHA  2012  Canada  2,305  62.1  84.6 (70–105)  70  5 years  aHR = 1.25 (1.20–1.30)  Age, gender   Yu Beijing Longitudinal Study of Aging (Urban sample)  2012  China  2,136  51.1  70.1 (55–97)  35  8 years  aHR = 1.28 (1.23–1.32)  Age, gender, education   Yu Beijing Longitudinal Study of Aging (Rural sample)  2012  China  1,121  51.0  70.2–70.3 (55–97)  35  8 years  aHR = 1.27 (1.21–1.32)  Age, gender, education   Rivindrarajah EMAS  2013  Europeb  2,929  0  59.9 (40–79)  39  4.3 years  aHR = 1.49 (1.33–1.67)  Age, centre, smoking, partner status  OR per 0.01 of FI   Li GLOW  2014  Canada  3,985  100  69.4 (>55)  34  3 years  aOR = 1.05 (1.03–1.06)  Age, BMI, smoking, alcohol, education   Theou TILDA  2015  UK  4,961  54.2  61.9 (>50)  66  2 years  aOR = 1.072 (1.040–1.106)  Age, gender  OR per 0.1 of FI   Armstrong HAAS  2015  USA  3,845  0  77.9 (72–91)  48  6 years  aOR = 1.73 (1.57–1.92)  Age, education   Li GLOW  2016  Canada  3,985  100  69.4 (>55)  34  3 years  aOR = 1.33 (0.87–2.03)  Age  Author/Study  Year  Location  Sample size  Female (%)  Age (range)  Number of deficits  Follow-up period  Risk estimate HR/OR (95% CI)  Adjustment  HR per 0.01 of FI   Searle Yale-PEP  2008  USA  754  64.6  –(72–98)  40  9 years  aHR = 1.03 (1.02–1.04)  Age, gender   Kulminski Cardiovascular Health Study  2008  USA  1,073  –  –(>65)  48  4 years  HR = 1.049 (1.040–1.057)  Unadjusted   Rockwood National Population Health Survey  2011  Canada  14,127  54.2  44 (>15)  42  14 years  aHR = 1.04 (1.03–1.04)  Age, gender, education   Yu Beijing Longitudinal Study of Aging (Urban sample)  2012  China  2,136  51.1  70.1 (55–97)  35  8 years  aHR = 1.042 (1.036–1.049)  Age, gender, education   Yu Beijing Longitudinal Study of Aging (Rural sample)  2012  China  1,121  51.0  70.2–70.3 (55–97)  35  8 years  aHR = 1.041 (0.034–1.049)  Age, gender, education   Bennett Chinese Longitudinal Healthy Longevity Survey  2013  China  6,300  53.0  88.9 (80–99)  38  4 years  aHR = 1.03 (1.03–1.04)  Age, gender   Theou SHARE  2013  Europea  36,306  54.6  65.2 (>50)  70  2 years  aHR = 1.05 (1.05–1.06)  Age, gender   Pena Nova Scotia Health Survey  2014  Canada  3,227  50.1  48.1 (>18)  23  10 years  aHR = 1.04 (1.03–1.05)  Age, gender   Blodgett EMAS  2016  Europeb  2,933  0  60.2 (40–79)  39  4.4 years  aHR = 1.07 (1.06–1.09)  Age   Hao Project of Longevity and Aging in Dujiangyan  2016  China  767  68.0  93.7 (90–108)  35  4 years  aHR = 1.03 (1.02–1.04)  Age, gender, education   Hoogendijk Longitudinal Aging Study Amsterdam  2016  Netherlands  2,218  –  –(57–88)  32  19 years  aHR = 1.03 (1.03–1.04)  Age, gender   Miller NHANES  2016  USA  8,911  –  –(20–)  46  8 years  aHR = 1.03 (1.02–1.04)  Age, gender   Mitnitski CSHA  2016  Canada  1,013  61.6  80.8 (>65)  61  6 years  aHR = 1.041 (1.030–1.052)  Age, gender  HR per 0.1 of FI   Kamaruzzaman BWHHS  2010  UK  4,286  100  –(60–79)  44  8.2 years  aHR = 1.3 (1.2–1.4)  Age, socioeconomic status, smoking, alcohol, marital status, living alone, housing tenure   Kamaruzzaman MRC assessment study  2010  UK  11,195  59.9  –(>75)  44  7.9 years  aHR = 1.3 (1.2–1.3)  Age, gender, smoking, alcohol, marital status, living alone, social contact, housing tenure   Theou CSHA  2012  Canada  2,305  62.1  84.6 (70–105)  70  5 years  aHR = 1.25 (1.20–1.30)  Age, gender   Yu Beijing Longitudinal Study of Aging (Urban sample)  2012  China  2,136  51.1  70.1 (55–97)  35  8 years  aHR = 1.28 (1.23–1.32)  Age, gender, education   Yu Beijing Longitudinal Study of Aging (Rural sample)  2012  China  1,121  51.0  70.2–70.3 (55–97)  35  8 years  aHR = 1.27 (1.21–1.32)  Age, gender, education   Rivindrarajah EMAS  2013  Europeb  2,929  0  59.9 (40–79)  39  4.3 years  aHR = 1.49 (1.33–1.67)  Age, centre, smoking, partner status  OR per 0.01 of FI   Li GLOW  2014  Canada  3,985  100  69.4 (>55)  34  3 years  aOR = 1.05 (1.03–1.06)  Age, BMI, smoking, alcohol, education   Theou TILDA  2015  UK  4,961  54.2  61.9 (>50)  66  2 years  aOR = 1.072 (1.040–1.106)  Age, gender  OR per 0.1 of FI   Armstrong HAAS  2015  USA  3,845  0  77.9 (72–91)  48  6 years  aOR = 1.73 (1.57–1.92)  Age, education   Li GLOW  2016  Canada  3,985  100  69.4 (>55)  34  3 years  aOR = 1.33 (0.87–2.03)  Age  a15 European countries: Austria, Belgium, Czech Republic, Denmark, France, Germany, Greece, Ireland, Israel, Italy, Netherlands, Poland, Spain, Sweden, Switzerland. b8 European countries: Belgium, Estonia, Hungary, Italy, Poland, Spain, Sweden, UK. 95% CI = 95% confidence interval. (a)HR: (adjusted) hazard ratio. (a)OR: (adjusted) odds ratio. BMI: body mass index. BWHHS: British Women’s Heart and Health Study. CSHA: Canadian Study of Health and Aging. EMAS: European Male Ageing Study. GLOW: Global Longitudinal Study of Osteoporosis in Women. FI: frailty index. NHANES: National Health and Nutrition Examination Survey. SES: socioeconomic status. SHARE: Survey of Health, Ageing and Retirement in Europe. TILDA: The Irish LongituDinal study on Ageing. Yale-PEP: Yale Precipitating Events Project. Frailty index as a predictor of mortality Meta-analysis of studies using HR HRs of mortality per 0.01 increase in the FI from the 13 cohorts were combined using a random-effects model due to the significant heterogeneity (P < 0.001, I2 = 86%). Frailty was a significant predictor of mortality (13 cohorts: pooled HR = 1.039, 95% CI = 1.033–1.044, P < 0.001). Combining HRs per 0.1 increase in the FI from six cohorts using a fixed-effect model (heterogeneity P = 0.11, I2 = 45%) also showed that frailty significantly predicted mortality (six cohorts: pooled HR = 1.282, 95% CI = 1.258–1.307, P < 0.001) (Figure 2A and B). Figure 2. View largeDownload slide Forest plots of mortality risk according to frailty measured by the Frailty index. A: Risk of dying (hazard ratio) per 0.01 increase in the Frailty index score. CI: confidence interval, IV: inverse variance, NSHS: Nova Scotia Health Survey. B: Risk of dying (hazard ratio) per 0.1 increase in the Frailty index score. BWHHS: British Women’s Heart and Health Study, CI: confidence interval, IV: inverse variance, MRC: MRC assessment study. Figure 2. View largeDownload slide Forest plots of mortality risk according to frailty measured by the Frailty index. A: Risk of dying (hazard ratio) per 0.01 increase in the Frailty index score. CI: confidence interval, IV: inverse variance, NSHS: Nova Scotia Health Survey. B: Risk of dying (hazard ratio) per 0.1 increase in the Frailty index score. BWHHS: British Women’s Heart and Health Study, CI: confidence interval, IV: inverse variance, MRC: MRC assessment study. Meta-analysis of studies using OR Four studies provided OR as a risk measure of mortality. Two studies showed ORs per 0.01 increase in the FI [28, 33] and another two studies showed ORs per 0.1 increase in the FI [29, 35]. fixed-effects models were used (heterogeneity P = 0.23 and 0.24, I2 = 30 and 29%, respectively) and both showed that frailty is a significant predictor of mortality (Appendix 4A and B). See Appendix 1 for meta-regression and subgroup analysis and publication bias assessment. Discussion The current study identified 19 studies that longitudinally examined mortality risk according to frailty measured by the FI in 18 cohorts and provided the effect measured as HR or OR per 0.01 or 0.1 increase in the FI. The meta-analysis quantitatively combined mortality risks based on frailty measured by the FI and consistently showed increased mortality risk according to the FI regardless of different types of the effect sizes and per units of the FI. Although the included studies constructed the FI based on different numbers and types of deficits, in addition to various populations and study settings, it is of note that the effect measures were in relatively narrow ranges and may support the robustness of this accumulation deficit frailty model. Although in general age is a strong predictor of mortality, the mean age of the cohorts was not a significant modulator in the association between the FI and mortality in the meta-regression analysis. Furthermore, subgroup analysis also showed that pooled estimates of studies with a mean age of >65 [14, 18, 24–26, 30, 36] and <65 [23, 27, 37] (mostly middle aged with the mean age ranging from 44 to 60.2) were almost identical (8 cohorts: pooled HR = 1.04, 95% CI = 1.03–1.05, P < 0.001, I2 = 84%, 3 cohorts: pooled HR = 1.05, 95% CI = 1.03–1.07, P < 0.001, I2 = 92%, respectively). This suggests the FI is a good indicator of mortality risk not only among older people but also among younger populations, regardless of age. Two study characteristics were found in the meta-regression analysis to be related to the association between frailty and mortality: follow-up period and female proportion. In general, women live longer but have more disabilities than men, known as the male–female health-survival paradox [40]. Given the FI can be regarded as a measure of biological age [10] and prevalence of frailty is higher among women than men [9], it is to be expected that female gender is associated with lower mortality risk according to frailty in the meta-regression analysis. Regarding the follow-up period, the meta-regression analysis suggests shorter follow-up periods are associated with higher mortality risk according to the FI. Frailty is a dynamic state and known to change over time, mostly worsening rather than improving [41]. The longer follow-up periods imply that as participants get older they usually get frailer. This may be why the reason the association between frailty and mortality became less prominent in studies with longer follow-up periods. The studies using the same cohorts with different lengths of follow-up showed overall comparable results with little difference [14, 23, 24]. In SHARE, 2-year mortality (aHR = 1.05) [24] was slightly higher than 5-year mortality (aHR = 1.04) [23], while 9-year mortality (aHR = 1.03) [14] was slightly lower than 12-year mortality (aHR = 1.04) [23] in the Yale Precipitating Events Project. This study’s findings should be interpreted with caution due to some limitations. First, all processes of the systematic review and meta-analysis were conducted by one investigator. Second, during the study selection, a large number of studies that used the FI to examine mortality risk were excluded because they did not provide HR or OR for mortality (n = 25); the effect measures provided were based on frailty groups defined by different cut-off points (n = 17); or on each additional deficit (n = 4). Although not all, at least some of them could potentially have been included in the meta-analysis. Lastly, the effect measures and upper and lower limits of 95% CI in many of the included papers were rounded to two decimal places, which could potentially lead to a miscalculation of standard error or weighting in the meta-analysis, especially when effect measures were calculated per 0.01 increase in the FI and were therefore relatively smaller. The current study has multiple strengths. The search strategy of the systematic review of the literature was robust and reproducible, using comprehensive search terms in multiple electronic databases. Additional data were also acquired from the original study’s authors [26]. The included studies were also assessed for heterogeneity, methodological quality and publication bias, and a high degree of heterogeneity was further explored by meta-regression analysis and subgroup analysis. The data from included studies were based on a FI constructed according to the standard methodology [14]. and were mostly controlled for important confounders, age and gender, or age in male only or female only cohorts. Other potential confounders would include education, socioeconomic status, smoking and alcohol consumption. In the subgroup analysis, there was no significant difference in mortality risk between studies adjusting for age and gender or age only and studies additionally adjusting for such confounders (8 cohorts: pooled HR = 1.04, 95% CI = 1.03–1.05, P < 0.001, I2 = 89%, four cohorts: pooled HR = 1.04, 95% CI = 1.03–1.04, P < 0.001, I2 = 74%, respectively. P for subgroup difference = 0.53). Lastly this is the first systematic review and meta-analysis focusing on FI and mortality. There are several features of the FI which distinguish it from frailty phenotype. As mentioned above, the FI can evaluate frailty status in a graded manner, rather than just three frailty categorisations by frailty phenotype (robust, pre-frail and frail), and make a more precise risk prediction. Furthermore, those who have a missing value for specific frailty components may be excluded from analyses in frailty phenotype. However the FI can still be calculated by excluding missing deficits from both numerator and denominator, which is because deficits are considered to be interchangeable if a sufficiently large number of deficits are included [42]. Although one may argue that it is not practical in clinical settings to collect information of 30 or more health deficits to calculate the FI, most of the clinical information could be extracted from electronic medical record systems. A recent study created an electronic FI from readily available data in primary care electronic records and demonstrated robust predictive ability for mortality, hospitalisation and nursing home placement [43]. This systematic review and meta-analysis was the first to quantitatively demonstrate the pooled mortality risk estimate according to frailty defined by the FI. Frailty measured by the FI is a strong predictor of death among older people as well as younger and middle-aged populations. A shorter follow-up period and lower female proportion seem to be associated with higher mortality risks according to frailty. Key points The Frailty Index is among the most popular frailty definitions and predicative of mortality. The mortality risk according to the Frailty index has never been quantified with meta-analysis in the literature. 14 studies providing hazard ratios or odds ratios per 0.01 or 0.1 increase in the Frailty index were included. All meta-analyses suggested that frailty measured by the Frailty index is a significant predictor of mortality. Studies conducted in Europe, with shorter follow-up periods and lower female proportion had higher mortality risks. Supplementary Data Supplementary data mentioned in the text are available to subscribers in Age and Ageing online. Funding None. Conflict of interest None. References Note: The very long list of references supporting this article has meant that only the most important are listed here. The full list of references is available in the Supplementary data. 1 Clegg A, Young J, Iliffe S, Rikkert MO, Rockwood K. Frailty in elderly people. Lancet  2013; 381: 752– 62. Google Scholar CrossRef Search ADS PubMed  2 Kojima G. Frailty as a predictor of future falls among community-dwelling older people: a systematic review and meta-analysis. J Am Med Dir Assoc  2015; 16: 1027– 33. Google Scholar CrossRef Search ADS PubMed  3 Kojima G. Frailty as a predictor of fractures among community-dwelling older people: a systematic review and meta-analysis. Bone  2016; 90: 116– 22. Google Scholar CrossRef Search ADS PubMed  4 Kojima G. Frailty as a predictor of hospitalisation among community-dwelling older people: a systematic review and meta-analysis. J Epidemiol Community Health  2016; 70: 722– 9. Google Scholar CrossRef Search ADS PubMed  5 Kojima G. Frailty as a predictor of nursing home placement among community-dwelling older adults: a systematic review and meta-analysis. J Geriatr Phys Ther  2016. 6 Kojima G. Frailty as a predictor of disabilities among community-dwelling older people: a systematic review and meta-analysis. Disabil Rehabil  2016; 39: 1897– 908. Google Scholar CrossRef Search ADS PubMed  7 Kojima G, Iliffe S, Jivraj S, Walters K. Association between frailty and quality of life among community-dwelling older people: a systematic review and meta-analysis. J Epidemiol Community Health  2016; 70: 716– 21. Google Scholar CrossRef Search ADS PubMed  8 Kojima G, Taniguchi Y, Iliffe S, Walters K. Frailty as a predictor of alzheimer disease, vascular dementia, and all dementia among community-dwelling older people: a systematic review and meta-analysis. J Am Med Dir Assoc  2016; 17: 881– 8. Google Scholar CrossRef Search ADS PubMed  9 Fried LP, Tangen CM, Walston J et al.  . Frailty in older adults: evidence for a phenotype. J Gerontol A Biol Sci Med Sci  2001; 56: M146– 56. Google Scholar CrossRef Search ADS PubMed  10 Mitnitski AB, Mogilner AJ, Rockwood K. Accumulation of deficits as a proxy measure of aging. Sci World J  2001; 1: 323– 36. Google Scholar CrossRef Search ADS   11 Walston JD, Bandeen-Roche K. Frailty: a tale of two concepts. BMC Med  2015; 13: 185. Google Scholar CrossRef Search ADS PubMed  12 Avila-Funes JA, Amieva H, Barberger-Gateau P et al.  . Cognitive impairment improves the predictive validity of the phenotype of frailty for adverse health outcomes: the three-city study. J Am Geriatr Soc  2009; 57: 453– 61. Google Scholar CrossRef Search ADS PubMed  13 Rothman MD, Leo-Summers L, Gill TM. Prognostic significance of potential frailty criteria. J Am Geriatr Soc  2008; 56: 2211– 16. Google Scholar CrossRef Search ADS PubMed  15 Kane RL, Shamliyan T, Talley K, Pacala J. The association between geriatric syndromes and survival. J Am Geriatr Soc  2012; 60: 896– 904. Google Scholar CrossRef Search ADS PubMed  16 Shamliyan T, Talley KM, Ramakrishnan R, Kane RL. Association of frailty with survival: a systematic literature review. Ageing Res Rev  2013; 12: 719– 36. Google Scholar CrossRef Search ADS PubMed  17 Chang SF, Lin PL. Frail phenotype and mortality prediction: a systematic review and meta-analysis of prospective cohort studies. Int J Nurs Stud  2015; 52: 1362– 74. Google Scholar CrossRef Search ADS PubMed  19 Rockwood K, Andrew M, Mitnitski A. A comparison of two approaches to measuring frailty in elderly people. J Gerontol A Biol Sci Med Sci  2007; 62: 738– 43. Google Scholar CrossRef Search ADS PubMed  20 Moher D, Liberati A, Tetzlaff J, Altman DG. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. Br Med J  2009; 339: b2535. Google Scholar CrossRef Search ADS   40 Weinstein M, Lane MA, eds. Sociality, Hierarchy, Health: Comparative Biodemography: A Collection of Papers . Washington DC: 2014 by the National Academy of Sciences, 2014. Google Scholar PubMed PubMed  41 Mitnitski A, Song X, Rockwood K. Trajectories of changes over twelve years in the health status of Canadians from late middle age. Exp Gerontol  2012; 47: 893– 9. Google Scholar CrossRef Search ADS PubMed  42 Rockwood K, Mitnitski A, Song X, Steen B, Skoog I. Long-term risks of death and institutionalization of elderly people in relation to deficit accumulation at age 70. J Am Geriatr Soc  2006; 54: 975– 9. Google Scholar CrossRef Search ADS PubMed  43 Clegg A, Bates C, Young J et al.  . Development and validation of an electronic frailty index using routine primary care electronic health record data. Age Ageing  2016; 45: 353– 60. Google Scholar CrossRef Search ADS PubMed  © The Author 2017. Published by Oxford University Press on behalf of the British Geriatrics Society.All rights reserved. For permissions, please email: journals.permissions@oup.com

Journal

Age and AgeingOxford University Press

Published: Mar 1, 2018

There are no references for this article.

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