Progression of frailty and prevalence of osteoporosis in a community cohort of older women—a 10-year longitudinal study

In this study, we investigate 75-year-old community-dwelling women (the OPRA cohort) [16]. The cohort was randomly selected from population registries and women invited on their 75th birthday. No exclusion criteria were applied. A total of 1044 women participated in the baseline investigation between 1995 and 1999, representing a participation rate of 67%. Reasons for non-attendance have previously been detailed [17]. Follow-up investigations were performed at 5 years (age 80, n = 715 attended) and at 10 years (age 85, n = 382 attended). Similarly, reasons for non-attendance have been detailed [18] (Fig. 1).

The participants were extensively investigated at each visit. A questionnaire provided information on lifestyle (education, work, physical activity, smoking, and alcohol), health (medications, surgery, injuries, other diseases, food/nutrition, and hormonal function) falls, and fractures. The questionnaire was revised at follow-up to include supplemental information and events over the intervening 5 years. Self-estimated risk of falling was assessed using a Likert scale with 5 as the highest risk.

Physical assessment included balance (modified Romberg method), gait speed, and number of steps (30-m walk, 2 × 15 m with one turn) and thigh muscle strength (Biodex Medical Systems®, v4.5.0, Biodex Corporation, Shirley, N.Y., USA) as previously described [19]. Biochemical markers (CRP and creatinine) were assayed as described [18, 20]. BMD was measured using dual-energy x-ray absorptiometry (DXA) (GE Lunar, Madison, WI) [19] and the same machine used throughout. In this study, femoral neck BMD is used with osteoporosis being defined as a T-score below − 2.5. Precision of DXA was assessed by duplicate measurements on healthy individuals (precision error was 0.009–0.010 g/cm2 at FN). No drifts in phantom measured results were observed [21].

Mortality as date of death was acquired in October 2012, from the Swedish National Population Register (individuals still alive were a maximum 91.5 years of age).

Participants provided written informed consent, and the regional ethical review board at Lund University approved the study, which was performed according to the principles of the Helsinki Declaration.

Being an osteoporosis cohort, we were unable to define frailty according to the most commonly used frailty phenotype [5]. Instead, using the principles of Searle et al. [22], we followed a stepwise process to construct an index with available data that allowed us to capture frailty across all assessment points.

As described in detail below, the final frailty index (FI) used in these analyses consists of 13 variables at all visits (Table 1). Covering a number of physiological domains, e.g., mobility, strength, co-ordination, and poly-medication, the index represents, for each OPRA participant, the number of “deficits in health.” The FI was calculated by dividing the number of deficits present by the total number of deficits examined, giving a score from 0.0–1.0, where a higher score indicates a higher frailty status. Where an individual had missing information for a particular variable, the total deficits were reduced by one.

To score variables (deficit present/non-present), we used either clinically relevant cut-points or identified the cut-off values by plotting the variable against an interim FI [22]. Categorical values were converted to binary values 1 (=deficit present) and 0 (=deficit absent); those with more than two categories were scored between 0 and 1 (e.g., high = 1.0; medium = 0.5; low = 0.0). To estimate cut-points of continuous data for dichotomization, curve estimation regression was performed, plotting potential frailty index variables against an intermediate frailty index. The resulting categories were then tested for differences in survival using Cox proportional hazard regression [22].

Descriptive statistics are reported as mean and standard deviation (SD), median, and IQR or frequency and percentage.

The frailty index, which shows a typically positively skewed (gamma) distribution [22], was used both as a continuous variable and stratified into quartiles (Q1 = lowest level of frailty; Q4 = highest level of frailty). Statistical comparisons were calculated overall or for Q1 vs Q4 as appropriate.

Annual linear progression of frailty over 10 years was calculated as the average, based on mean values of the whole cohort. For mortality, hazard ratios (HR) and 95% confidence intervals (95% CI) were estimated using Cox proportional hazard regression with the healthiest quartile (Q1) as the reference category. Time to death was 5 years, 10 years, or until end of study (i.e., October 2012). HRs are presented unadjusted.

For osteoporosis, differences in T-score between the frailty categories were estimated using the non-parametric Kruskal Wallis test.

Over 10 years of follow-up, mean frailty increased giving an approximate annual linear frailty score progression of 6–7% (Table 2, Fig. 1). At baseline, the proportion scoring least frail (FI 0.0–0.1), i.e., most robust, constituted 48% of the cohort. At age 80 and 85, that proportion dropped to 25 and 14%, respectively. Among those rated least frail at age 75, although they progressed in frailty, the majority only reached intermediate levels (FI 0.2–0.6). As many as 11% had no change in frailty status and remained robust during the 10 years. Figure 2 illustrates the progression towards increased frailty among the participants.

Those who died during the first 5-year period had the highest average frailty scores at baseline (n = 105; mean FI 0.30, median 0.29); approximately similar to the mean FI at age 85. The same trend was observed tracing those who attended the 5-year visit and comparing their frailty score at the 10-year follow-up (Fig. 1).

Mortality was substantially higher in the highest quartile of frailty compared to the lowest based on their baseline frailty score; after 10 years, 49.1% had died compared to 17.2% (p = 8.4 × 10⁻¹⁵). At 10 years, mortality was also higher in Q3 and 70% of those dead contained in Q3–Q4 (Table 3). The corresponding mortality risk over the first 5 years was highest in the frailest women (Q4 vs Q1; HR_unadj 3.26 [1.86–5.73]; p < 0.001) and continued to be elevated at 10 years (HR_unadj 3.58 [2.55–5.03]; p < 0.001) (Fig. 3). At age 85, only the least frail (i.e., most robust) had 2–3 times lower mortality, compared to the other quartiles.

Study participation may serve as an indicator of societal participation. Women who were alive but did not attend 5-year follow-up at age 80 were more frail at baseline than those who attended again (mean FI 0.23, median 0.19 vs FI 0.13, median 0.09). Further demonstrating the applicability of this frailty index in a long-term perspective, initial baseline frailty score was lowest in those who attended 10-year follow-up (FI 0.11, median 0.08), and increased stepwise in those who were alive but did not attend (FI 0.15, median 0.10) and those who had died (FI 0.20, median 0.18) (Fig. 1).

Aging is associated with osteoporosis and since this cohort was specifically designed for this purpose, we tested the association between frailty and osteoporosis. The proportion with osteoporosis increased with age as expected in the population overall; at baseline, 28.1% were osteoporotic rising to 49.0% after 10 years. At age 75, femoral neck BMD was 0.773 g/cm² (SD 0.131) in the least frail compared to 0.759 (SD 0.150) in the frailest, and not statistically different. After adjustment for BMI, BMD was significantly associated with frailty (overall; p = 0.0006 and Q1 vs Q4 p = 0.0003). The pattern was similar at age 80, while femoral neck BMD at age 85 was similar across frailty quartiles (Table 3), adjustment for BMI did not result in a statistically significant difference (data not shown).

Firstly, one potential limitation is that our frailty index was derived and applied in the same population and external validation of the index has not been performed. While validation would be valuable, this is however, part of the problem in the emerging field of frailty and mirrors the lack of consensus and inconsistency across studies in terms of collected information. Further to this, making direct comparison with other studies is difficult; however, in our index, the cut-off for frailty coincides with the lower limit of Q4 and while a consensus threshold is lacking; this is close to the empirical cut-off point of > 0.25 for a frailty index based on accumulated deficits as described by Rockwood [6].

Secondly, our index has fewer variables than the suggestion of 30–40; however, in its development, we demonstrate a very high correlation and an almost identical ability to predict mortality between a 40-item index and the 13-item index used in this study. This most likely reflects the high inter-relationship between the included variables, whereby one variable can capture and substitute multiple variables. It can also be argued that this high redundancy between variables is an advantage as it indicates the possibility to use simpler constructs and facilitate use. Thirdly, due to constraints from the original study design and subsequent lack of information in certain domains, data on social and mental factors are unfortunately not included. Fourth, being a longitudinal study of older persons, there is an inherent problem of loss to follow-up and a potential bias of healthy participants. Indeed, we also show that those continuing in the study are less frail, and with regard to mortality, this is not problematic, but a loss of power may occur for other outcomes, although the descriptive information is still robust. Fifth, caution should be exercised in terms of generalizing the findings to other populations such as younger women or other ethnicities.

Strengths of this study include that the participants are 75-year-old community-dwelling rather than institutionalized women, representing a pivotal period for continued healthy aging or deteriorating health. The fact that all women were at the same age at inclusion is advantageous as it minimizes the influence of chronological age on accumulated health deficits. Another strength is the provision of longitudinal data for up to 15 years allowing us to quantitatively assess change in frailty status with advancing age. Additionally, we demonstrate that it is possible to develop a meaningful frailty index from available data and with the same discriminatory ability as a more comprehensive, larger item index. This is important since research on frailty in relation to osteoporosis is still in its infancy but potentially beneficial for future research. Taken together, this study contributes with data on frailty in averagely healthy older women including tracking over time and its association to bone health.

In conclusion, the relevance of this study lies in demonstrating the pattern of frailty longitudinally in older community-dwelling women and its association to mortality up to 15 years. Frailty was associated with a threefold increased risk of death in both a short and longer perspective with a higher frailty score being equivalent to being chronologically five to 10 years older. Conversely, only one in ten older women escaped progression of frailty. In addition, higher frailty is associated with osteoporosis, despite the fact that the frailest individuals may have a higher BMI.