Participants are from the EPOSA cohort, which includes random samples from five existing population-based cohorts studies (Germany, the Netherlands, Spain, Sweden, and the United Kingdom), as well as a newly recruited sample from Italy (van der Pas et al., 2013). EPOSA was a population-based study of 2,942 adults between the ages of 65–85 years old, resident in six European countries (Germany, Italy, the Netherlands, Spain, Sweden, and the United Kingdom). After having written informed consent, all participants underwent a baseline clinical examination and interview at home or in a healthcare center by trained researchers, between November 2010, and November 2011, and a follow-up interview 12–18 months later. The local research ethics committees approved the study (Germany: Universitat Ulm Ethikkommission [312/08]. Italy: Comitato Etico Provinciale Treviso [XLIV-RSA/AULSS7]. Netherlands: Medisch Ethische Toetsingscommissie Vrije Universiteit Amsterdam [2002/141]. Spain: Comité Ético de Investigación Clínica del Hospital Universitario La Paz Madrid [PI-1080]. Sweden: Till forskningsetikkommittén vid Karolinska Instituted Stockholm [00–132]. United Kingdom: Hertfordshire Research Ethics Committee [10/H0311/59]).

In this study, IC was operationalized using four domains available in the EPOSA dataset: cognition, psychological, locomotion, and vitality. The sensory domain (vision and hearing), part of the WHO construct, was not available. Each domain was scored from 0 to 2, with 0 or 1 indicating a worse status and 2 indicating a better one. According to previous studies (Muneera et al., 2023; Ma et al., 2021), a composite IC score was then calculated, ranging from 0 to 8, with lower scores representing worse status.

1. For the cognition, the Mini-Mental State Examination (MMSE) was used (Folstein et al., 1975), a widely validated tool for cognitive impairment, with good internal consistency (Tombaugh and McIntyre, 1992). The total score ranging from 0 to 30 (with lower score indicating greater cognitive impairment) was classified as 0 for scores ≤18, 1 for scores between 18–24, 2 for scores ≥24. MMSE scores were not corrected for age or educational level, factors that may influence cognitive performance but which were considered in the analysis.

The primary outcome of the study was fallen in the last 12–18 months, self-reported using the question, ‘Did you fall in the past year?’ coded as ‘no’ and ‘yes’.

The baseline characteristics considered included age, sex, country of residence, education level, marital status, income, alcohol and tobacco consumption, chronic diseases, medications being taken, joint replacements and OA.

Educational level was categorized as up to elementary education versus higher levels of education. Marital status was categorized as being single or never married, divorced, widowed, living apart versus married or cohabitating, or in a registered partnership. A monthly income capable of making ends meet was classified as ‘only with great difficulty’/’with some difficulty’, ‘fairly easily’ and ‘easily’. Self-reported information about tobacco consumption was classified as ‘currently’, ‘in the past’ vs. ‘never’, while those on alcohol consumption ‘yes’ versus ‘no’. Self-reported presence of chronic conditions referred to: non-specific lung disease (i.e., asthma, chronic bronchitis or pulmonary emphysema, etc.), cardiovascular disease (i.e., cardiac valve disease, coronary heart diseases, arrhythmia, pacemaker, cardiac arrest, etc.), peripheral artery disease, diabetes mellitus, stroke, cancer and osteoporosis, lasting at least 3 months or which caused the individual to seek a physician’s attention (each dichotomized as present versus absent). Among the medication used over the past 2 weeks were analgesics and/or anti-inflammatory drugs and psycholeptics drugs; these were categorized as either medication use versus non-use. The presence of previous joint replacements was assessed by asking participants if they had ever had joint replacement surgery. If the response was affirmative, the participant was asked about the location and the reason for the joint replacement; the variable was classified as ‘lower limb’, ‘other joints’ versus ‘no’.

The clinical diagnosis of OA was determined on the basis of the participant’s medical history and a physical examination, according the clinical criteria of the American College of Rheumatology (ACR) (Altman, 1991), and the European League Against Rheumatism (Zhang et al., 2009). Clinical hand OA was diagnosed using specific sections of the Australian Canadian Osteoarthritis Hand Index (AUSCAN) (Bellamy et al., 2002). Clinical hip/knee OA, defined as the presence of OA in at least one or both of these joints, was diagnosed using specific sections of the Western Ontario and McMaster Universities Arthritis Index (WOMAC) examining pain and stiffness (Bellamy et al., 1988; Roorda et al., 2004). Pain in the hip/knee on at least one side was also evaluated during the physical examination. As far as clinical OA was concerned, the participants were classified as: (Organization, 2025) no OA (Sánchez-Sánchez et al., 2024), only hand OA (Cheng and Chang, 2017), hip and/or knee OA (Biasin et al., 2023), hip and/or knee OA combined with hand OA.

Only participants with complete data for all the variables were included in the analyses. Weights based on sex and five-year age categories according to the 2010 Standard European Population, calculated for each individual within each country of the EPOSA study (van der Pas et al., 2013) were used only in the descriptive analyses. Categorical variables were reported as proportions, and continuous variables as means and standard deviations (SD), and medians with interquartile ranges (IQRs). Normal distributions of continuous variables were tested using the Kolmogorov–Smirnov test. Significant differences between countries or groups of participants were evaluated using the Kruskal–Wallis test, the Wilcoxon rank-sum test, or the χ2 test, as appropriate.

The predictors of falls were assessed using logistic regression models adjusted for sex, age, and country. Each independent variable was examined to check the appropriate categories if categorical and the linearity in the logit and the scale if continuous. A multivariable model containing all the variables was fitted using a stepwise selection procedure (P to enter = 0.15 and P to remain = 0.10) to select them. All the interactions between the variables in the final model were checked; interaction terms with P ≤ 0.10 were retained in the final model. Odds ratios (ORs) were presented with their 95% confidence intervals (CIs).

Tests were two-tailed, and P < 0.05 was considered statistically significant. Analyses were performed using SAS software (SAS Institute Inc, Cary, NC), version 9.4.

The 2,597 participants with complete, country-stratified (weighted) data had a median age of 73 years (IQR: 70–78, range: 65–85 years). Women comprised 51.3% of the sample, and nearly 30% had a diagnosis of OA. Additionally, 11% had undergone a joint replacement, approximately 25% reported cardiovascular disease and the use of analgesic/anti-inflammatory medications, 10% took psycholeptic medications and more than 50% had an income sufficient to make ends meet. Alcohol consumption was reported by 76% of participants, while around 8% used tobacco. Significant differences were observed between countries (Supplementary Table S3).

Overall, the median IC score was 6 (IQR: 5–7) (Supplementary Table S4). More than half of the participants exhibited some degree of impairment in the locomotion and vitality domains. Median IC scores were lower in Italy, Spain, and the United Kingdom, while Sweden, Germany, and The Netherland had the highest ones. Specifically, Italy showed greater impairment in locomotion, psychological, and cognitive domains; Spain in vitality, psychological, and cognitive domains; and the United Kingdom in vitality, locomotion, and cognitive domains.

Sex and age differences across IC quartiles were significant (Figure 1). In the group with lower IC scores, 81.7% were women, and 45.9% were older than 78 years. In contrast, in the group with higher IC scores, these proportions were notably lower, at 30.1% and 12.4%, respectively.

After 12–18 months, 569 out of 2,127 participants (26.8%) reported experiencing a fall (Table 1). Individuals who fell differed significantly from those who did not in several ways: they were more likely to be female, older, living in Sweden, Spain, or Germany, and single, divorced, widowed, or living alone. Falls were also more common among individuals who did not use tobacco, had OA, had undergone joint replacements in the lower limbs, and used analgesic or anti-inflammatory medications.

Regarding IC and its domains (Table 2), participants who experienced falls exhibited greater cognitive deficits and more pronounced depressive symptoms, indicating a worse psychological domain. They also had lower strength and, even after accounting for BMI, reduced vitality. Locomotion was particularly impaired, as reflected in slower gait and chair test performances, as well as an overall lower PP score. Consequently, their IC scores were low.

According to the univariable logistic regression analysis (Table 3), adjusted for age, sex, and country, falls after 12–18 months were significantly associated with IC. Other significant predictors included tobacco use, cardiovascular disease, analgesic/anti-inflammatory medication use, clinical OA, and joint replacement. These associations remained significant in the multivariable analysis, which also identified one significant interaction between age and sex. In particular, the estimated odds ratio for a five-year increase in age differed between males and females. Among men, a five-year difference in age increases the odds of falling by almost 1.3 times, whereas among women, there is effectively no increase. In fact, women have a higher risk of falling after 12–18 months at younger ages (65–70), and this effect becomes insignificant at around the age of 73 (Figure 2). The odds of experiencing a fall during the 12–18 months follow-up were 1.5 times higher (95% CI: 1.10–2.04) for individuals with a history of hip and/or knee replacements compared to those without, after adjusting for other covariates. Similarly, the odds of falling were 1.39 times higher (95% CI: 1.05–1.82) for individuals with clinical OA of the hand only or in combination with hip and/or knee. The use of analgesic/anti-inflammatory medication was also associated with an increased risk of falls (OR = 1.31, 95% CI: 1.02–1.67). Interestingly, current tobacco consumption appeared to be a protective factor against falls. Finally, close to significance (P = 0.06) was the fall odds of individuals with cardiovascular disease (OR = 1.25, 95% CI: 0.99–1.57). Low IC (<5) was estimated to be 1.57 (95% CI: 1.19–2.08) times more likely to fall after 12–18 months.

This study has some limitations. First, its observational design does not permit causal inference. Second, the IC construct used here was not originally designed to reflect the WHO ICOPE tool and lacks the sensory domain (vision and hearing). As such, it may not fully capture the multidimensional nature of IC as conceptualized by the WHO, and its psychometric properties remain untested. Additionally, the use of a 12–18 months recall period for falls may have introduced recall bias and underreporting. Heterogeneity across participating countries—in terms of recruitment strategies and population characteristics—could also affect generalizability, even though we adjusted for country in the analyses. Moreover, we excluded physical activity measures (e.g., minutes of activity or kilocalories/week) from the models due to high proportions of missing data and the lack of significant associations in preliminary analyses. Although this decision helped preserve statistical power, it may have limited our ability to capture relevant behavioral factors associated with fall risk.

The main strengths of the study include the use of a large, multicenter European dataset, a multidimensional approach to capturing physical and mental capacities, and the inclusion of relevant clinical and sociodemographic covariates. Furthermore, the analysis of interaction terms allowed us to identify meaningful sex- and age-related differences in fall risk.

In conclusion, this study found that a lower intrinsic capacity score, derived from multiple domains, was associated with a greater likelihood of falling among older adults. However, this association was modest and based on a construct that lacks formal validation. As such, the IC score used in this study should not be interpreted as a substitute for established fall risk tools. Further research is needed to develop standardized and validated IC-based instruments, evaluate their predictive value, and determine their utility in clinical settings. Nonetheless, our findings support the conceptual relevance of IC in understanding vulnerability in aging and highlight the importance of multidimensional approaches to fall risk assessment.