Excess mortality after hip fracture in elderly persons from Europe and the USA: the CHANCES project

Authors


Abstract

Background

Hip fractures are associated with diminished quality of life and survival especially amongst the elderly.

Objective

All-cause mortality after hip fracture was investigated to assess its magnitude.

Methods

A total of 122 808 participants from eight cohorts in Europe and the USA were followed up for a mean of 12.6 years, accumulating 4273 incident hip fractures and 27 999 deaths. Incident hip fractures were assessed through telephone interviews/questionnaires or national inpatient/fracture registries, and causes of death were verified with death certificates. Cox proportional hazards models and the time-dependent variable methodology were used to assess the association between hip fracture and mortality and its magnitude at different time intervals after the injury in each cohort. We obtained the effect estimates through a random-effects meta-analysis.

Results

Hip fracture was positively associated with increased all-cause mortality; the hazard ratio (HR) in the fully adjusted model was 2.12, 95% confidence interval (CI) 1.76–2.57, after adjusting for potential confounders. This association was stronger amongst men [HR: 2.39, 95% CI: 1.72–3.31] than amongst women [HR: 1.92, 95% CI: 1.54–2.39], although this difference was not significant. Mortality was higher during the first year after the hip fracture [HR: 2.78, 95% CI: 2.12–3.64], but it remained elevated without major fluctuations after longer time since hip fracture [HR (95% CI): 1.89 (1.50–2.37) after 1–4 years; 2.15 (1.81–2.55) after 4–8 years; 1.79 (1.57–2.05) after 8 or more years].

Conclusion

In this large population-based sample of older persons across eight cohorts, hip fracture was associated with excess short- and long-term all-cause mortality in both sexes.

Introduction

As the population ages, bone fractures are becoming an increasingly important health problem amongst the elderly with substantial burden for the individual and society. Hip fractures are the most relevant fractures in terms of severity, functional dependence, social and economic cost and fatality [1-3].

Despite a well-known increase in mortality shortly after hip fracture [3-5], the evidence on the long-term mortality following a hip fracture is not consistent [6-11]. Some studies have demonstrated a persistent increase in all-cause mortality in the long term after the injury [6-9], whereas others report from low to no elevated long-term mortality after hip fracture [10, 11]. The higher mortality rates were mostly observed in elderly, ill or impaired populations [6, 7]. A recent meta-analysis exploring the magnitude and duration of excess mortality risk after hip fracture found the highest risk in the first 3 months after the fracture (fivefold to eightfold increase), and mortality remained elevated, compared to age-matched controls, even after 10 years. The excess risk increased with age and, at any given age, was higher for men than for women [12].

The aim of this study was to investigate both short- and long-term mortality after hip fracture in a large cohort of community dwellers, aged 60 years and older, from Europe and the USA who were followed up prospectively.

Materials and methods

The CHANCES project

The Consortium on Health and Ageing: Network of Cohorts in Europe and United States (CHANCES) project is a large collaboration, funded by the European Commission within the Seventh Framework Programme, combining 14 major cohorts/studies from Europe and the USA, to provide evidence on ageing-related health characteristics and determinants of healthy ageing. The study protocol of each individual cohort/study has been approved by local ethics committees, and all participants have given written informed consent before enrolment. All procedures have been carried out in accordance with the Declaration of Helsinki. Variables were harmonized across the cohorts following predetermined standardized procedures. The study design and population characteristics of the cohorts included in the CHANCES project have been described in detail elsewhere [13].

Eight cohorts with available information on hip fractures during follow-up as well as mortality were included in the present analysis: EPIC-Elderly Greece and EPIC-Elderly Umeå, Sweden [14]; the ESTHER (Epidemiological Study on the Chances of Prevention, Early Recognition and Optimised Treatment of Chronic Diseases in the Older Population) Study from Germany [15]; the Tromsø Study from Norway [16]; the Swedish Mammography Cohort (SMC) and the Cohort Of Swedish Men (COSM) studies [17]; the Nurses' Health Study (NHS) from the USA [18]; and the Health, Alcohol and Psychosocial factors in Eastern Europe (HAPIEE) study with data from the Czech Republic [19]. Further details about the participating cohorts are available in the Appendix S1.

Information on incident hip fractures

Information on incident hip fractures was collected through telephone interviews or questionnaires to elicit self-reported data in EPIC-Elderly Greece, ESTHER and NHS and through national inpatient registries or fracture registries in EPIC-Elderly Umeå, the Tromsø Study, COSM, SMC and the Czech HAPIEE cohort [16, 20]. To verify self-reported hip fractures, validation studies were conducted for EPIC-Elderly Greece and ESTHER in the context of the CHANCES project. The rate of verification ranged from 52% to 86%. A validation study was also conducted as part of the NHS in which all self-reported hip fractures were confirmed by medical records [21], whilst COSM, SMC and the Tromsø Study had shown high validity of incident hip fracture diagnosis using the national inpatient register [20, 22]. Hip fractures identified as International Classification of Diseases and Related Health Problems 10th Revision (ICD-10) codes S72.0–S72.2 were included in the analyses.

Information on all-cause mortality

Vital status of the participants was assessed either by contacting relatives or household members, or through record linkage with nationwide or local death registries. All causes of death were verified through death certificates, whereas ICD coding was used across the cohorts.

Statistical analysis

Individual cohorts

To describe the socio-economic, lifestyle, medical and anthropometric characteristics of the participants, the distribution of the corresponding variables, separately for men and women in every cohort, is presented. Cox regression was applied for the cohort-specific analyses to calculate hazard ratios (HRs) and 95% confidence intervals (95% CIs) for mortality following the occurrence of a hip fracture event. The survival time was calculated from the date of enrolment in the study until the date of death (for those who died during follow-up) or the date of last follow-up (for those who were alive at that time). Once the exposure of interest in this study was the hip fracture event, which occurred during follow-up, we treated hip fracture as a time-dependent variable to capture the association between hip fracture and mortality. The same methodology was used to assess the aforementioned association at different time intervals from the beginning of the hip fracture event.

Models were applied with three levels of adjustment with an increasing number of confounders. Specifically, model 1 was only adjusted for age (in years; continuous variable) and sex. Model 2 was additionally adjusted for the continuous variables; body mass index (BMI) (in kg m−2), height (in m), daily energy intake (in kcal day−1) and alcohol intake (in g day−1) and the categorical variables; vigorous physical activity (yes/no), educational level (none/less than primary/vocational or technical secondary/secondary, not vocational and not technical/college or university), living alone (yes; for single, widowed, separated or divorced/no; for married or living together), employment status (full-time or part-time employment and not of pensionable age/self-employment/housewife and not of pensionable age/pensionable age and still working/pensionable age and not working/stopped working before retirement age due to poor health/unemployed and not of pensionable age) and smoking status (never/former/current smoker). Finally, model 3 was additionally adjusted for hypertension (yes/no) and chronic diseases (cardiovascular disease, diabetes or cancer; yes/no).

After following a consistent harmonization procedure [13], there were minor differences in the definitions of variables used; the following variables were not available in all cohorts and were not included in the analysis for specific studies: alcohol intake (many missing values in the Tromsø Study), energy intake (not available in the Tromsø Study and ESTHER), education (all participants in NHS educated to the same level), prevalent cancers (excluded in COSM at baseline), living alone (not available in SMC), vigorous physical activity (not available in EPIC-Sweden) and prevalent hip fractures (not available in EPIC-Greece and EPIC-Sweden). Moreover, educational level was used in three levels in the Tromsø Study (primary or less/high school or lyceum/college or university).

Participants aged ≥60 years at enrolment without a prevalent hip fracture event were included in the present analysis. Model 3 was chosen as the main (fully adjusted) model. We excluded from our analyses those participants without any follow-up or with missing values in variables included in Model 3. We performed three further subanalyses restricted to (i) men, (ii) women and (iii) subjects aged ≥70 years at enrolment.

We also tried to assess interaction on an additive scale between hip fractures and other risk factors using the relative excess risk due to interaction (RERI) index [23]:

display math

λ11=hazard rate when hip fracture and the other risk factor are present; λ10=hazard rate when hip fracture is present and the other risk factor is absent; λ01=hazard rate when hip fracture is absent and the other risk factor is present; λ00=hazard rate when hip fracture and the other risk factor are absent.

In other words:

display math

HR11=hazard ratio when hip fracture and the other risk factor are present; HR10=hazard ratio when hip fracture is present and the other risk factor is absent; HR01=hazard ratio when hip fracture is absent and the other risk factor is present.

All cohort-specific analyses were carried out using Stata statistical software versions 10–13 (Stata Corp LP, College Station, TX, USA).

Meta-analysis

All meta-analyses of cohort-specific estimates were performed using the DerSimonian–Laird method with random effects [24]. We estimated the HRs and 95% CIs for mortality following hip fracture events, after combining all results from each cohort separately. The heterogeneity between cohorts was measured using the I2 statistic and tested for statistical significance with the chi-squared test from Cochran's Q statistic [25]. Moreover, we tested effect modification with a meta-analysis of all available estimates from different strata by calculating the chi-squared test for heterogeneity.

For the meta-analysis of interaction on the additive scale, we calculated the RERI (index of interest) in all cohorts and then performed a meta-analysis of these indices.

For all meta-analyses, we used stata, version 11 (Stata Corp LP). All tests were two-sided, and P-values less than 0.05 were considered statistically significant.

Results

The study population consisted of 122 808 participants from eight cohorts (seven from Europe and one from the USA); during a mean follow-up time of 12.6 years (range 7.9–13.7 years), there were 4273 incident hip fractures. Participants were mostly recruited during the 1990s, and a total of 27 999 participants died during follow-up (Table 1). The percentage of hip fracture varied from 1.2% to 10.3%. Once all participants were at least 60 years old, small age differences were observed amongst the cohorts. The percentage of participants with one or more missing values for any of the variables included in the analysis varied from 8% to 44% across the cohorts; however, the distribution of the variables in each cohort before and after exclusions was essentially the same (see Appendix S1). The baseline characteristics of the study participants are presented by sex and cohort in Tables 2A and B.

Table 1. Description of the participating cohorts
Cohort nameCountry n a Females, n (%)Hip fractures, n (%)Deaths, n (%)Mean age (years) at enrolment (SD)Baseline periodMean follow-up period (years) (SD)
  1. a

    Number of participants (without missing values for any confounding variable included in model 3).

EPIC-GreeceGreece90375488 (61)209 (2)1907 (21)67.3 (4.5)1994–199910.3 (3.3)
EPIC-SwedenSweden31081641 (53)64 (2)460 (15)60.3 (1.1)1992–199613.3 (3.0)
Nurses' Health StudyUSA68 46868 468 (100)1260 (2)10 126 (15)61.0 (0.6)1986–201013.0 (6.1)
The Tromsø studyNorway53732930 (55)378 (7)2817 (52)69.6 (6.9)1994–199512.0 (4.9)
ESTHERGermany49572541 (51)62 (1)956 (19)65.9 (4.1)2000–200210.8 (2.4)
COSMSweden15 7440 (0)936 (6)7143 (45)69.0 (5.2)199812.7 (4.1)
SMCSweden12 92312 923 (100)1327 (10)4191 (32)69.0 (5.6)199813.7 (3.5)
HAPIEECzech Republic31981649 (52)37 (1)399 (13)64.7 (2.9)2002–20057.9 (1.6)
Table 2. (A) Characteristics of male participants at baseline by participating cohort (based on the number of observations in the fully adjusted model 3). (B) Characteristics of female participants at baseline by participating cohort (based on the number of observations in the fully adjusted model 3)
(A)EPIC-GreeceEPIC-SwedenNurses' Health StudyTromsø studyESTHERCOSMSMCHAPIEE-Czech
Men, n (%)3549 (39)1467 (47)0 (0)2443 (45)2416 (49)15 744 (100)0 (0)1549 (48)
Body mass index (kg m−2), mean (SD)28.0 (4.0)25.9 (3.6)25.8 (3.5)27.8 (3.8)25.7 (3.2)24.6 (10.6)
Height (m), mean (SD)1.66 (0.06)1.75 (0.06)1.74 (0.07)1.73 (0.06)1.76 (0.06)1.74 (0.06)
Energy intake (kcal day−1), mean (SD)2049 (613)1916 (636)2466 (692)2051 (964)
Alcohol intake (g day−1), mean (SD)15.2 (23.3)4.3 (5.0)10.2 (11.6)11.1 (12.4)23.3 (27.0)
Education (primary or less), n (%)3209 (90)789 (54)1319 (54)1744 (72)6960 (44)108 (7)
Living alone, n (%)231 (7)270 (18)642 (26)310 (13)2622 (17)202 (13)
Currently working, n (%)980 (28)884 (60)540 (22)205 (8)2208 (14)441 (29)
Never smokers, n (%)1131 (32)756 (52)370 (15)759 (31)5948 (38)519 (34)
Vigorous physical activity, n (%)686 (19)813 (33)1159 (48)14 866 (94)1077 (70)
Hypertension, n (%)1432 (40)423 (29)569 (23)1594 (66)5018 (32)894 (58)
Prevalent cancer, n (%)102 (3)42 (3)201 (8)165 (7)86 (6)
Prevalent diabetes, n (%)551 (16)64 (4)108 (4)346 (14)1344 (9)302 (19)
Prevalent CVD, n (%)384 (11)76 (5)503 (21)356 (15)2812 (18)255 (16)
(B)EPIC-GreeceEPIC-SwedenNurses' Health StudyTromsø studyESTHERCOSMSMCHAPIEE-Czech
  1. CVD, cardiovascular disease.

Women; n (%)5488 (61)1641 (53)68 468 (100)2930 (55)2541 (51)0 (0)12 923 (100)1649 (52)
Body mass index (kg m−2), mean (SD)30.1 (4.8)25.9 (4.4)26.8 (5.4)26.5 (4.7)27.5 (4.3)25.2 (3.9)25.3 (10.7)
Height (m), mean (SD)1.53 (0.06)1.62 (0.06)1.64 (0.06)1.59 (0.06)1.62 (0.06)1.64 (0.06)1.61 (0.06)
Energy intake (kcal day−1), mean (SD)1648 (500)1393 (423)1758 (529)1713 (511)2010 (967)
Alcohol intake (g day−1), mean (SD)2.5 (5.4)1.4 (2.2)5.8 (10.1)3.5 (5.7)4.4 (6.4)4.6 (8.6)
Education (primary or less), n (%)5064 (92)906 (55)2137 (73)2002 (79)6518 (50)367 (22)
Living alone, n (%)1599 (29)389 (24)9274 (14)1490 (51)917 (36)595 (36)
Currently working, n (%)669 (12)1086 (66)44 538 (65)428 (15)241 (9)1638 (13)243 (15)
Never smokers, n (%)5151 (94)1135 (69)29 894 (44)1553 (53)1820 (72)8222 (64)1037 (63)
Vigorous physical activity, n (%)1179 (21)20 999 (31)458 (16)857 (34)12 146 (94)1131 (69)
Hypertension, n (%)2710 (49)558 (34)27 815 (41)758 (26)1574 (62)3513 (27)914 (55)
Prevalent cancer, n (%)218 (4)134 (8)7971 (12)231 (8)207 (8)751 (6)133 (8)
Prevalent diabetes, n (%)766 (14)31 (2)4742 (7)160 (5)263 (10)663 (5)240 (15)
Prevalent CVD, n (%)209 (4)16 (1)2484 (4)299 (10)145 (6)977 (8)121 (7)

Table 3 and Fig. 1 show that the occurrence of hip fracture was positively associated with all-cause mortality (in model 3: HR 2.12, 95% CI: 1.76–2.57) after adjusting for all available potential confounders. After excluding HAPIEE from the analysis due to the exceptionally high HR, overall associations decreased but remained statistically significant (in model 3: HR 1.98, 95% CI: 1.65–2.38). The association between hip fracture and mortality slightly decreased after adjusting for increasing number of confounders (i.e. from model 1 to model 3). Model 3 showed that this relationship was somewhat stronger amongst men (HR 2.39, 95% CI: 1.72–3.31) than women (HR 1.92 (95% CI: 1.54–2.39), and was weaker but still significant amongst participants aged ≥70 years old (HR 1.84, 95% CI: 1.46–2.33), as the underlying risk of these (more elderly) participants is higher. When a sensitivity analysis was applied restricting the analysis to cohorts that included both sexes, the differences remained largely unchanged and statistically significant (in model 3: HR 2.37 and 1.94 for men and women, respectively). By contrast, when the association amongst participants aged ≥70 years was compared with the association in the primary analysis of participants ≥60 years, after excluding EPIC-Sweden, NHS and HAPIEE which do not contribute to the HR of subjects aged ≥70 years old (because they have very few or no participants in this age group at baseline), the difference was small [in model 3: HR 1.91 and 1.84 for all participants (≥60 years old) and those ≥70 years old, respectively]. Although the heterogeneity of the associations was high in all these comparisons (in general: 70% ≤ I2 ≤ 90%), the relationship between hip fracture and mortality was positive in all countries, but differed in magnitude (Fig. 1).

Table 3. Hazard ratio (HR) for mortality (95% confidence interval) after hip fracture amongst participants in three models
 Number of cohortsHR from model 1aHR from model 2bHR from model 3cI2 for model 3 (P-value)
  1. a

    Model 1: adjusted for age (in years; continuous) and sex (male/female).

  2. b

    Model 2: adjusted for the same variables as in model 1 and additionally for the continuous variables body mass index (in kg m−2), height (in m), daily energy intake (in kcal day−1) and alcohol intake (in g day−1), and the categorical variables vigorous physical activity (yes/no), educational level (none/less than primary/vocational or technical secondary/secondary, not vocational and not technical/college or university), living alone (yes/no), employment status (full-time or part-time employment and not of pensionable age/self-employment/housewife and not of pensionable age/pensionable age and still working/pensionable age and not working/stopped working before retirement age due to poor health/unemployed and not of pensionable age) and smoking status (never/former/current smoker).

  3. c

    Model 3: adjusted for the same variables as in model 2 and additionally hypertension (yes/no) and chronic diseases (cardiovascular disease, diabetes or cancer; yes/no).

Total population82.39 (1.95–2.92)2.21 (1.82–2.68)2.12 (1.76–2.57)90% (<0.001)
Men62.87 (1.90–4.35)2.54 (1.78–3.62)2.39 (1.72–3.31)78% (<0.001)
Women72.07 (1.67–2.56)1.97 (1.59–2.44)1.92 (1.54–2.39)84% (<0.001)
Elderly (≥70 years at baseline)51.91 (1.49–2.45)1.88 (1.49–2.38)1.84 (1.46–2.33)90% (<0.001)
Figure 1.

Forest plot showing hazard ratios for mortality after hip fracture in model 3 (i.e. the fully adjusted model).

Although the proportionality assumption was not violated in any of the cohorts, we also estimated the time-dependent effect of hip fracture on mortality (Table 4). We found that the short-term effect of hip fractures was higher than the mid- and long-term effects. Specifically, the HR in the first year after hip fracture was 2.78 (95% CI: 2.12–3.64), whereas in the longer term hip fractures were associated with an almost twofold increase in mortality (1–4 years after hip fracture: HR 1.89, 95% CI: 1.50–2.37; 4–8 years after hip fracture: HR 2.15, 95% CI: 1.81–2.55; and ≥8 years after hip fracture: HR 1.79, 95% CI: 1.57–2.05). In this analysis, we considered the effects of all cohorts for all time periods, except HAPIEE, which does not contribute to the overall HR for ≥8 years. However, the conclusions were unchanged when we excluded the HAPIEE cohort from this entire analysis (data not shown).

Table 4. Hazard ratio (HR) and 95% confidence interval (CI) for mortality after hip fracture in model 3 (i.e. the fully adjusted model) by time since fracture occurrence
Time since hip fractureNumber of cohortsHR for model 3a95% CII2 for model 3 (P-value)
  1. a

    Model 3 adjusted for sex (male/female), the continuous variables age (in years), body mass index (in kg m−2), height (in m), daily energy intake (in kcal day−1) and alcohol intake (in g/day) and the categorical variables vigorous physical activity (yes/no), educational level (none/less than primary/vocational or technical secondary/secondary, not vocational and not technical/college or university), living alone (yes/no), employment status (full-time or part-time employment and not of pensionable age/self-employment/housewife and not of pensionable age/pensionable age and still working/pensionable age and not working/stopped working before retirement age due to poor health/unemployed and not of pensionable age) and smoking status (never/former/current smoker), and hypertension (yes/no) and chronic diseases (cardiovascular disease, diabetes or cancer; yes/no).

≥0 to <1 year82.782.12–3.6481% (<0.001)
≥1 to <4 years81.891.50–2.3781% (<0.001)
≥4 to <8 years82.151.81–2.5557% (0.021)
≥8 years71.791.57–2.050% (0.918)

Finally, we found that the associations between the combination of hip fracture and prevalent chronic disease and mortality were super-additive (RERI > 0), as evidenced by a 42% (95% CI: 10–75%) excess risk of mortality due to the joint presence of hip fracture and chronic disease (Fig. 2). When we investigated any possible excess risk due to the interaction between hip fracture and obesity and living alone, we found no significant deviation from additivity.

Figure 2.

Forest plot showing relative excess risk due to interaction (RERI) between hip fractures and chronic diseases for mortality in model 3 (i.e. the fully adjusted model).

Discussion

In this large sample of individuals, aged 60 years and older from Europe and the USA, there was evidence that hip fracture is associated with excess short- and long-term all-cause mortality in both sexes. Participants who had experienced a hip fracture during follow-up had the highest risk of dying during the first year after the fracture, and an almost twofold increase in mortality persisted even 8 years or more after the injury. Small differences were observed according to sex, with the magnitude of the increase in all-cause mortality somewhat larger amongst men. Associations were significant even after controlling for chronic comorbidities and lifestyle factors. Furthermore, prevalence of chronic diseases at baseline was found to have a super-additive effect with hip fractures on mortality (as tested using the RERI index), implying that individuals with chronic diseases need particularly careful management following a hip fracture.

Our results with respect to short-term excess all-cause mortality confirm those of other studies and the most recent meta-analysis (almost threefold increase in the present study compared to threefold to fivefold increase during the first 6 months in the recent meta-analysis) [5, 9, 12]. To the best of our knowledge, excess short-term mortality following hip fracture, especially during the first 3–6 months, was observed in all previously published studies. Factors that contribute the most to this finding are linked to postoperative complications after surgery such as cardiac and pulmonary complications, infections (e.g. pneumonia and septicaemia) and increased risk of thromboembolism [26, 27]. Other factors, such as multiple comorbid conditions, have also been implicated [12, 25].

A difference in excess all-cause mortality after hip fracture amongst men and women, and specifically a higher excess mortality amongst men, although minimal in this study, has been a consistent finding in previous studies [5, 12, 27-29]. It seems that although hip fracture incidence in men is substantially lower compared to women, mortality after hip fracture is higher in men [29]. Efforts to explore further the causes of this gender difference have shown, in most instances, that such differences remained even after controlling for chronic comorbidities and medications [28].

Long-term mortality after hip fracture was significantly elevated, not only for the first 8 years, but also after that period. The excess long-term risk of death after hip fracture has been found in the majority but not all relevant studies; however, the mechanisms underlying this excess risk remain unclear [6-11]. One explanation has been the co-existence of chronic disease, but excess mortality remained in the studies that collected and had the ability to adjust for such data [4, 6, 7, 12]. On the other hand, hip fracture is associated with increased functional decline and disability in the elderly [30]. Recently, hip fracture occurrence has also been associated with an exaggerated persistent inflammatory response, whilst, in parallel, chronic inflammation might play a role in the functional decline and the onset or acceleration of frailty [31-33]. These mechanisms could provide a possible explanation of the observed decline in health and the increased long-term mortality after hip fracture. In addition, the detrimental effect of long-standing pain and diminished quality of life, especially when followed by loss of independence, should not be underestimated [34].

The strengths of our study include the large, population-based sample of more than 100 000 elderly participants from Europe and the USA, the prospective design, the use of harmonized variables across the cohorts and the implementation of a common statistical analysis with individual data. The meta-analysis of harmonized individual data possibly reduced the potential heterogeneity, which generally occurs when performing a meta-analysis of published data. Moreover, by meta-analysing results from different cohorts without knowing a priori the associations that would be estimated, we have overcome the problem of publication bias [35] that may be present in other meta-analyses of previous publications [12].

A limitation of this study is the different periods of enrolment of the participants in the cohorts as both life expectancy and some aspects of hip fracture treatment have changed during these years. Although the majority of participants entered the studies during the 1990s, subjects were also recruited during the late 1980s to the NHS and during the 2000s to the NHS, ESTHER and HAPIEE-Czech. Nevertheless, although heterogeneity was observed between cohorts (perhaps partially explained by the different periods of recruitment of the participants in the cohorts along with the fact that participants had different characteristics across cohorts; see Tables 1 and 2 A and B), the association between hip fracture and subsequent mortality showed the same positive direction in all cohorts. Moreover, heterogeneity decreased according to the period after hip fracture, possibly due to the decreased number of events (deaths) over time. Furthermore, we could not determine the cause of hip fracture; more specifically, we were not able to differentiate between high-energy (e.g. traffic accidents) and low-energy trauma (e.g. falls from standing height), although the majority of hip fractures in older subjects (≥60 years of age) are low-energy fractures. The different methods of hip fracture and mortality ascertainment used across the participating cohorts could potentially have resulted in differing degrees of under- and over-reporting of hip fracture cases and deaths that could further influence the association under study. Also, although extensive harmonization was undertaken in the context of the CHANCES project, different methods of data collection were used, and not all covariates were assessed in all cohorts. Residual confounding may also exist because of the inability to control for other parameters such as medication (e.g. bisphosphonates), supplement use and access to health care across the cohorts. Additionally, covariates such as BMI, alcohol intake, physical activity and comorbidities were assessed at baseline and not updated during follow-up. It is unlikely, however, that such changes in the covariates would have had a major impact on the results. Information on nursing home status at the time of hip fracture was not available, and thus, we could not differentiate between nursing home residents and community-dwelling participants in our analyses. Nursing home residents have been shown to experience higher mortality in comparison with community dwellers both amongst individuals with hip fracture, especially in the immediate postinjury period, and amongst those without hip fracture [36]. The magnitude of missing data could have affected our findings. However, the extent would be small as there was no significant difference in the characteristics of the available participants and of those included in the analysis (see Appendix S1). Finally, the findings of this study cannot be extrapolated to populations other than white men and women aged ≥60 years with similar sociodemographic characteristics to those of the study participants.

Conclusions

In conclusion, our study confirms that elderly individuals who have suffered a hip fracture are at increased risk of dying, compared to those who have not, in the short term after the fracture but also years later. Appropriate measures need to be implemented for primary and secondary prevention of hip fracture in order to ensure better quality of life and survival in the elderly.

Funding sources

The research presented herein was funded by the European Union Seventh Framework Programme (FP7/2007-2013) under grant agreement no. HEALTH–F3-2010-242244. The project is coordinated by the Hellenic Health Foundation, Greece. The national cohorts are supported by (i) EPIC-Elderly Greece: the Hellenic Health Foundation; (ii) EPIC-Elderly Umea, Sweden: the Swedish Cancer Society and the Swedish Research Council; (iii) ESTHER, Germany: the Baden-Württemberg State Ministry of Science, Research and Arts (Stuttgart, Germany), the Federal Ministry of Education and Research (Berlin, Germany) and the Federal Ministry of Family Affairs, Senior Citizens, Women and Youth (Berlin); (iv) the Tromsø Study: UiT–The Arctic University of Norway, the National Screening Services, the Research Council of Norway, Northern Norway Regional Health Authority, the Norwegian Council on Cardiovascular Diseases, the Norwegian Foundation for Health and Rehabilitation, the Norwegian Diabetes Association, the Cancer Registry of Norway, the Odd Berg Group Research Fund and Troms County Council; (v) COSM and SMC, Karolinska Institutet, Sweden: the Swedish Research Council Karolinska Institutet's Strategic Foundation and Uppsala University, and the Swedish Cancer Society; and (vi) NHS: the National Cancer Institute (grant P01CA87969).

Conflict of interest statement

No conflicts of interest to declare.

Ancillary