Abstract
Background
Falls are a leading cause of morbidity and loss of independence in older adults. Existing fall prediction models vary widely in design and performance, and the added value of incorporating nonbiological factors remains unclear. This study aimed to evaluate the predictive performance of the biological, sociodemographic, behavioral, and environmental domains proposed by the World Health Organization (WHO) for discriminating fall risk in community-dwelling older adults.
Methods
We conducted a cross-sectional analysis using baseline data from the prospective INITIATE cohort of community-dwelling adults aged ≥ 65 years. LASSO regression with 10-fold cross-validation was used to select biological predictors for any fall and injurious fall outcomes, followed by multivariable logistic regression. Sequential models reflecting sociodemographic, behavioral, and environmental domains were constructed to assess incremental discriminative performance. Model performance was assessed using the area under the ROC curve (AUC), with DeLong’s test used for comparisons. Bootstrap validation (1,000 iterations) was used to assess internal validity.
Results
A total of 433 participants were included (median age 76 years; 64% female). For any falls (n = 194, 44.8%), the reduced biological model (mobility limitation, balance, visual acuity, fear of falling) achieved an AUC of 0.64 (95% CI 0.58–0.69). For injurious falls (n = 40, 28.4%), the reduced biological model (Timed Up and Go [TUG] time, grip strength, executive function, global cognition, and pain interference) achieved an AUC of 0.73 (95% CI 0.64–0.82). Adding sociodemographic, behavioral, and environmental variables produced minimal, nonsignificant improvements for both outcomes (any falls: AUC 0.65, p = 0.46; injurious falls: AUC 0.78, p = 0.30).
Conclusions
Parsimonious models based primarily on biological measures can provide clinically meaningful discrimination while remaining feasible for community and outpatient use. The distinct risk profiles for any falls and injurious falls highlight the need for outcome-specific screening approaches. Prospective evaluation and external validation are needed prior to clinical implementation.
Supplementary Information
The online version contains supplementary material available at 10.1186/s12877-026-07417-7.
Keywords: Falls, Injurious falls, Fall risk assessment, Older adults, Community-dwelling, Prediction models, LASSO regression, Biological predictors, Model performance
Background
Each year, approximately 23–30% of community-dwelling older adults experience a fall [1], and approximately one-third of those fallers sustain injuries requiring medical attention [2]. Globally, falls are the second leading cause of unintentional injury-related deaths and the leading cause of death among adults aged 70 years and older [3]. In Canada alone, falls account for more than 65% of injury-related hospitalizations among adults aged 65–74 years and over 80% among those aged 75 years and older, resulting in an estimated $10.3 billion annually in direct and indirect healthcare costs [4]. These individual and societal consequences underscore the need for early identification of high-risk individuals to guide effective prevention strategies and alleviate the burden on healthcare systems.
Accurately predicting who will fall remains a challenge because fall risk is inherently multifactorial and shaped by biological, sociodemographic, behavioral, and environmental factors [5]. Despite this multifactorial etiology, most fall risk prediction models have focused on a relatively narrow subset of predictors, largely emphasizing mobility- and performance-based biological measures, and may overlook the broader determinants that influence fall occurrence in real-world settings [6, 7].
This narrow focus is also evident in clinical practice and international guidelines, where fall risk assessment commonly prioritizes gait, balance, and functional performance [8–11]. This focus is justified, as impairments in these capacities are strong and well-established risk factors for falls [6, 7]. However, fall risk profiles are not uniform across the older adult population. Among community-dwelling older adults, fall risk can be shaped not only by biological impairment but also by living circumstances, everyday behaviors, and exposure to potentially hazardous environments, which are often not observed during clinic-based assessments [12]. Thus, while functional assessments remain essential, their predominant biological focus likely provides only a partial representation of fall risk.
Importantly, contemporary frameworks such as those described in the WHO fall-prevention reports emphasize that fall risk extends beyond biological factors alone and includes behavioral, social, and environmental domains that interact to influence fall risk in older adults [5, 13]. While these broader constructs are increasingly recognized conceptually, relatively few studies have examined whether incorporating them into fall-risk prediction models provides meaningful additive value beyond biological predictors alone [6]. Accordingly, the aim of this paper is to examine whether fall risk among community-dwelling older adults is influenced by factors beyond biological ones. Specifically, we aimed to (i) determine the extent to which biological predictors are associated with the retrospective 12-month incidence of any falls and injurious falls and (ii) examine whether the inclusion of sociodemographic, behavioral, and environmental factors improves discriminative performance compared with a biological-only model. We hypothesized that models incorporating predictors across multiple domains would demonstrate greater discriminative performance than biological-only models.
Methods
This study was reported in accordance with the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) Statement [14]. The completed checklist is provided in Additional file 1.
Study design and population
This cross-sectional analysis used baseline data from the Initial Test for Fall Risk Assessment in the Elderly (INITIATE) cohort, a prospective cohort study of community-dwelling older adults designed to investigate risk factors for falls. Community-dwelling older adults aged ≥ 65 years were recruited between March 2020 and September 2023. Recruitment paused during the COVID-19 pandemic and subsequently resumed in March 2022. All baseline data collection was completed by October 2024. In the INITIATE study, participants undergo a baseline assessment and are prospectively followed for 12 months to record fall events. The present analysis used baseline measurements and retrospectively reported falls occurring in the preceding 12 months. These data were collected at baseline as part of the INITIATE prospective cohort using standardized questionnaires and clinical assessments administered by trained research staff, as previously described [15].
Participants were eligible if they were community-dwelling, English-speaking, able to walk 10 m without physical assistance, and capable of providing informed consent. The exclusion criteria included severe visual or hearing impairments, insufficient English proficiency, or cognitive impairment precluding written consent.
Participants completed a baseline assessment, during which retrospective fall data from the preceding 12 months were collected. Predictor and outcome data were collected at baseline via telephone or online surveys and a two-hour, in-person assessment of clinical and performance-based measures. The outcome assessors were not blinded to the participants’ retrospective fall status; however, they were trained in standardized assessment administration and followed predefined protocols. Ethical approval was obtained from the Hamilton Integrated Research Ethics Board (#7380), and all participants provided written informed consent prior to enrollment in the study.
Outcomes
Two dichotomous fall outcomes were defined based on self-reported falls in the 12 months preceding baseline: [1] any falls, indicating the occurrence of one or more falls, and [2] injurious falls, defined as a fall requiring medical attention from a health professional within 48 h [16]. As per the Prevention of Falls Network Europe recommendations [17], a fall was defined as “an unexpected event in which the individual comes to rest on the ground, floor, or lower level.” All eligible falls were retained, including recurrent falls and falls occurring during sports or recreational activities, as each fall has the potential to contribute to and perpetuate a cycle of increased fall risk [18, 19]. The outcomes were coded as 0 = no and 1 = yes.
Predictors
Candidate predictors were selected a priori to reflect the multifactorial nature of fall risk and to minimize selection bias [20]. The selection of predictors was informed by existing literature [6, 21, 22], clinical applicability, feasibility of administration, and relevance to geriatric and rehabilitation practice. The predictors were grouped according to the WHO’s four-domain fall-risk model [5]:
Biological: frailty, multimorbidity, mobility limitation, Timed Up and Go (TUG) time, balance, grip strength, vision, hearing, sensory loss, global cognition, executive function, number of fall risk–increasing drugs (FRIDs), pain interference, fear of falling, and depression.
Sociodemographic: education, living alone, social support.
Behavioral: participation in strengthening exercise and nutritional risk.
Environmental: neighborhood walkability.
In total, 23 predictors, corresponding to 23 degrees of freedom, were available for model building. Operational definitions, measurement details, and questionnaire sources are provided in Additional file 2: Table S1.
Sample size
The sample size was determined by the primary objective of the parent study, the INITIATE cohort. No a priori sample size calculation was performed for the present analysis; however, concerns regarding events per variable (EPV) were mitigated by the use of LASSO penalization. All participants who completed the baseline assessment and provided retrospective data on falls over the preceding 12 months were eligible for inclusion in the analytic sample [15].
Statistical analysis
Sample characteristics were summarized by sex (female, male) using means and standard deviations (SDs) for normally distributed variables, medians and interquartile ranges (IQRs) for skewed variables, and frequencies and relative percentages for categorical variables (Table 1). Missing data were handled via complete case analysis, whereby only participants with complete predictor and outcome data were included in the models.
Table 1.
Design of tiered prediction models for falls and injurious falls
| Model | Independent Variables |
|---|---|
| Tier 1 | Age, sex, biological predictors retained by LASSO |
| Tier 2 | Tier 1 + education, living alone, social support |
| Tier 3 | Tier 2 + participation in strengthening exercise, nutrition risk |
| Tier 4 | Tier 3 + neighborhood walkability |
To identify the most informative biological predictors, the least absolute shrinkage and selection operator (LASSO) was applied with 10-fold cross-validation [23]. LASSO was chosen to address the high dimensionality of the predictor set, reduce multicollinearity, and mitigate overfitting by shrinking small coefficients toward zero [24, 25]. Predictors with nonzero coefficients were retained to form the biological base model for each outcome. Subsequent logistic regression models were then constructed to reflect the domains described by the WHO risk factor framework and evaluate the added discriminative contribution of nonbiological domains. LASSO retained biological predictors were entered into Tier 1, with age (continuous) [26] and sex (female/male) [27] forced into all models due to their known confounding effects. Sociodemographic variables were added in Tier 2, behavioral variables in Tier 3, and an environmental variable in Tier 4. Interaction terms were not examined due to the limited number of outcome events. The full list of variables considered at each tier is presented in Table 1.
For the final adjusted models, assumptions of logistic regression were assessed. The linearity of continuous predictors with the log-odds of the outcome was assessed visually. Multicollinearity was assessed via the calculation of variance inflation factors (VIFs); a VIF > 5 was regarded as indicating serious multicollinearity [28]. Influential observations were identified using leverage and standardized residual plots. The results are presented as adjusted odds ratios (ORs) with 95% confidence intervals (CIs). Statistical significance was set at a two-sided p < 0.05.
Model discrimination was evaluated using the area under the receiver operating characteristic (ROC) curve (AUC) [29], with higher values indicating better discrimination. AUC values were interpreted descriptively, with values around 0.6–0.7 reflecting poorer discrimination, 0.7–0.8 fair to acceptable discrimination, and higher values indicating stronger performance [30, 31]. AUC differences between models were tested using Stata’s ROCCOMP command, which applies DeLong’s method for comparing correlated ROC curves [32]. Internal validation was performed using bootstrap resampling with 1,000 iterations, to assess the stability of model performance estimates and reduce optimism in classification results [33]. Model calibration was assessed using the Brier score, with lower values indicating better calibration [34]. All analyses were performed using Stata/SE 18.5 (StataCorp LLC, College Station, TX).
Results
Participant characteristics
Among the 514 participants who consented to participate (median age = 76 years, IQR = 10; 63.8% female), the outcome data were complete for all participants. Eighty-one participants were excluded due to missing predictor data, yielding a final analytic sample of 433 participants (median age = 76 years, IQR = 10; 64.0% female). Missing predictor data were low overall, with no single variable exceeding 5% missingness (Additional file 3: Figure S1). The sociodemographic, lifestyle, and health-related characteristics of the analytic sample are shown in Table 2. The baseline distribution of candidate predictor variables included in the modelling analyses is presented in Additional file 4: Table S2. Nearly half of the participants (44.8%, n = 194) reported at least one fall in the preceding year. Among fallers, 28.4% (n = 40) experienced an injurious fall in line with our study’s definition.
Table 2.
Sociodemographic and health-related characteristics of the participants
| Variables | Female (n = 277) |
Male (n = 156) |
Total (N = 433) |
|---|---|---|---|
| Age, yr, median (IQR) | 75 (10) | 76 (9.5) | 76 (10) |
| White, n (%) | 262 (94.6) | 126 (80.8) | 388 (89.6) |
| Living alone, n (%) | 120 (43.5) | 31 (19.9) | 151 (35.0) |
| Education, n (%) | |||
| < Secondary school | 16 (5.8) | 9 (5.7) | 25 (5.8) |
| Secondary school graduation but no post- secondary education | 26 (9.4) | 15 (9.6) | 41 (9.5) |
| Some post-secondary education | 51 (18.4) | 19 (12.2) | 70 (16.2) |
| Post-secondary degree/diploma | 184 (66.4) | 113 (72.4) | 297 (68.6) |
| Marital status, n (%) | |||
| Single | 23 (8.5) | 4 (2.8) | 27 (6.5) |
| Married/living with a partner in common-law relationship | 133 (49.1) | 116 (80.6) | 249 (60.0) |
| Widowed | 76 (28.0) | 17 (11.8) | 93 (22.4) |
| Divorced or separated | 39 (14.4) | 7 (4.9) | 46 (11.1) |
| BMI, kg/m2, mean (SD) | 28.0 (6.2) | 28.1 (4.3) | 28.0 (5.6) |
| Systolic blood pressure, mmHg, mean (SD) | 136.5 (20.2) | 133.9 (17.5) | 135.6 (19.3) |
| Diastolic blood pressure, mmHg, mean (SD) | 74.9 (10.1) | 75.0 (11.0) | 74.9 (10.4) |
| Smoking status, n (%) | |||
| Current | 9 (3.3) | 3 (1.9) | 12 (2.8) |
| Previous | 120 (43.3) | 79 (50.6) | 199 (46.0) |
| Never | 148 (53.4) | 74 (47.4) | 222 (51.3) |
| Current alcohol consumption, n (%) | 202 (72.9) | 126 (80.8) | 328 (75.8) |
| Self-reported health status, n (%) | |||
| Excellent/very good/good | 250 (92.3) | 128 (88.9) | 378 (91.1) |
| Fair/poor | 21 (7.7) | 16 (11.1) | 37 (8.9) |
| Self-reported mental health status, n (%) | |||
| Excellent/very good/good | 254 (93.7) | 132 (92.3) | 386 (93.2) |
| Fair/poor | 17 (6.3) | 11 (7.7) | 28 (6.8) |
Values are presented as the median (IQR), mean (SD), or n (%)
Abbreviations: n Number, IQR Interquartile range, SD Standard deviation, BMI Body mass index, BP Blood pressure
Selection and discriminative performance of the biological prediction models
LASSO regression was performed using all the predefined biological variables for each fall outcome. From the initial pool of fifteen candidate variables, along with the forced covariates of age and sex, the selection process identified four predictors for the any fall outcome: mobility limitation, balance, visual acuity, and fear of falling. For the injurious fall outcome, five predictors were selected: TUG time, grip strength, executive function, global cognition, and pain interference. None of the predictors overlapped across the models.
The reduced biological model showed modest discrimination for any falls (AUC 0.64, 95% CI 0.58–0.69), comparable to the full biological model (AUC 0.65, 95% CI 0.60–0.70). For injurious falls, both the reduced (AUC 0.73, 95% CI 0.64–0.82) and full models (AUC 0.76, 95% CI 0.68–0.84) demonstrated acceptable discrimination. The differences in the AUCs between the reduced and full models were not statistically significant (any falls: p = 0.22; injurious falls: p = 0.19).
Multivariate analysis of LASSO-selected biological predictors
The retained variables were entered into multivariable logistic regression analyses. Assumptions of logistic regression were assessed for all final models, and no major violations were identified. In the adjusted logistic regression model for any falls, poorer visual acuity (OR 6.64, 95% CI 1.40–31.56) and greater fear of falling (OR 1.04, 95% CI 1.01–1.08) were significantly associated with increased fall risk. For injurious falls, greater pain interference was associated with lower odds of sustaining an injurious fall (OR 0.74, 95% CI 0.57–0.96). Table 3 presents the adjusted odds ratios following penalized LASSO selection for both outcomes. No other biological variables demonstrated statistically significant associations with either fall outcome after adjustment for covariates.
Table 3.
Adjusted associations between LASSO-selected biological predictors and fall outcomes
| Predictor | Any falls model (n = 433) OR (95% CI) |
Injurious falls model (n = 141) OR (95% CI) |
|---|---|---|
| Constant | 1.51 (0.07–34.23) | 0.11 (0.00–262.92) |
| Age (years) | 0.99 (0.95–1.02) | 1.03 (0.95–1.12) |
| Female sex (ref = male) | 1.23 (0.81–1.86) | 0.75 (0.23–2.38) |
| Preclinical mobility limitation | 1.07 (0.87–1.31) | – |
| Balance | 0.98 (0.93–1.03) | – |
| Better-eye acuity-logMAR | 6.64 (1.40–31.56) | – |
| Fear of falling | 1.04 (1.01–1.08) | – |
| TUG (s) | – | 1.12 (0.99–1.28) |
| Grip strength (kg) | – | 0.95 (0.89–1.01) |
| Executive function | – | 1.01 (0.99–1.02) |
| Global cognition | – | 0.97 (0.84–1.13) |
| Pain interference | – | 0.74 (0.57–0.96) |
Complete case analysis. Significant predictors are indicated in bold
Abbreviations: OR Odds ratio, CI Confidence interval, TUG Timed Up and Go, logMAR Logarithm of the minimum angle of resolution
Discriminative performance of the extended models
To evaluate whether additional WHO domains improved discrimination, the biological base models (Model 1) were sequentially extended to include sociodemographic (Model 2), behavioral (Model 3), and environmental (Model 4) predictors for each outcome.
For any falls, model performance improved only minimally, from an AUC of 0.64 in Model 1 to 0.65 (95% CI 0.60–0.71) in the fully extended Model 4. For injurious falls, discrimination increased from 0.73 to 0.78 (95% CI 0.69–0.86). However, none of the incremental improvements in the AUCs were statistically significant (any falls: p = 0.46; injurious falls: p = 0.30). The AUC values for all the models are presented in Fig. 1 and Additional file 5: Table S3. Given the lack of significant improvement with the inclusion of additional domains, Model 1 for each outcome was retained as the final model. The full prediction models are presented in Additional file 6: Table S4.
Fig. 1.
Parsimony plot showing the discriminative performance of sequentially expanded models for predicting any falls and injurious falls. Black bars indicate the 95% confidence intervals. Model 1 = biological factors (age and sex forced); Model 2 = Model 1 + sociodemographic factors; Model 3 = Model 2 + behavioral factors; Model 4 = Model 3 + environmental factors. Abbreviations: AUC, area under the curve
Model validation
Internal validation was performed using the bootstrap method with 1,000 resamples. For any falls, Model 1 yielded a Brier score of 0.23 (95% CI 0.22–0.24), identical to Model 4. For injurious falls, Models 1 and 4 achieved Brier scores of 0.17 (95% CI 0.14–0.21) and 0.16 (95% CI 0.13–0.20), respectively. All models showed acceptable overall calibration and model fit across outcomes. Full bootstrap validation results are presented in Additional file 7: Table S5.
Discussion
This study sought to identify key biological predictors of fall risk in community-dwelling older adults and to examine whether expanding fall risk prediction models beyond biological factors to include sociodemographic, behavioral, and environmental domains improved discriminative performance. Contrary to our hypothesis, incorporating predictors from additional WHO risk factor domains did not meaningfully improve model discrimination compared with biological-only models for either any fall or injurious fall outcomes. Predictor profiles differed by outcome, indicating outcome-specific patterns in fall-risk prediction. Additionally, several predictors identified in this study, notably visual acuity and fear of falling, are not consistently incorporated in fall risk prediction models.
Although the biological models demonstrated only modest discrimination, their performance was consistent with that of previously published models from community-based samples. A recent systematic review of any fall prediction models reported AUCs ranging from 0.49 to 0.87, with a narrower range of 0.62 to 0.69 among models evaluated in independent samples [35]. Notably, many developed models lacked internal validation [35], which may contribute to optimistic performance estimates at the upper end of this range. In this context, the discrimination observed in the present study aligns with that typically reported for validated, community-based models and reflects the inherent challenge of predicting heterogeneous any fall outcomes.
In the present study, the injurious fall models demonstrated modestly higher discrimination than the any fall models; however, the addition of sociodemographic, behavioral, and environmental predictors resulted in only minimal, nonsignificant improvements in performance for both outcomes, despite their theoretical relevance and emphasis in global guidelines [5, 9]. This pattern is consistent with prior studies reporting modestly greater discrimination for models predicting injurious falls, with AUCs of 0.75 for women and 0.77 for men [36]. This suggests that injurious falls may represent a more homogeneous and clinically distinct outcome and may therefore be more amenable to prediction than any fall outcome. Nonetheless, the overall modest discriminative performance observed here aligns with broader evidence indicating that fall risk prediction models rarely achieve high discrimination, even when incorporating multiple domains [6, 35]. However, cross-study comparisons should be interpreted cautiously, as differences in study design, population, and the measurement or operationalization of risk factors may contribute to variability in discriminative performance.
The limited incremental contribution of nonbiological domains in this study should not be interpreted as evidence that these factors lack predictive relevance for falls. Rather, their contribution likely reflects how these constructs were operationalized, the limited availability of well-validated measures for several nonbiological domains, and the characteristics of the cohort. Few predictors represented each nonbiological domain, and some factors in the INITIATE data set were difficult to operationalize for model inclusion. For instance, available environmental measures, such as home hazard assessments, are structured as checklist-based indicators rather than continuous or composite scores [37]. In addition, the relatively healthy, well-educated, and homogeneous nature of the sample may have further limited the variability in sociodemographic and behavioral factors. Prior work suggests that individuals with higher educational attainment often have greater health literacy and access to supportive resources [38, 39], which may mitigate the influence of contextual risk factors observed in more socioeconomically diverse populations. In this context, the comparable performance of the developed models supports the value of parsimonious modeling approaches. From a clinical perspective, focusing on a smaller set of well-measured biological predictors may enhance feasibility, interpretability, and implementation of screening tools without compromising predictive accuracy.
Beyond overall model performance, predictor profiles differed meaningfully by fall outcome, highlighting the importance of outcome definition in fall-risk modeling. For any falls, LASSO-selected predictors reflected a profile dominated by sensory function and psychological vulnerability (i.e., poorer visual acuity and greater fear of falling) whereas injurious falls were characterized by a more impairment-driven biological profile, including markers of reduced physical capacity, cognitive performance, and pain-related burden. Although both outcomes were biologically driven, the lack of overlap between predictor sets supports growing evidence that any falls and injurious falls represent related but distinct outcomes [40, 41]. This distinction suggests that screening and prevention strategies may need to be tailored according to the specific fall outcome of interest rather than relying on a single, unified risk profile.
Several predictors identified in this study diverged from those emphasized in commonly used screening tools. Performance-based measures that primarily assess physiological and functional aspects of mobility and postural stability, such as gait speed and balance, and are commonly prioritized in clinical practice and international guidelines [9, 42], were not independently associated with fall outcomes. Instead, visual acuity and fear of falling emerged as key predictors of any falls. While consistent with prior evidence linking sensory impairment and fear-related adaptations to fall risk [43, 44], these results suggest that mobility-focused screening alone may overlook important contributors to fall risk, particularly among active, community-dwelling older adults. This highlights the value of considering sensory and psychological dimensions alongside performance-based measures when evaluating the risk of any falls.
Consistent with this broader interpretation, the association between visual acuity and any falls was characterized by a large odds ratio and a wide confidence interval, indicating substantial uncertainty around the effect estimate. This imprecision likely reflects a combination of limited representation at poorer levels of visual acuity, the scaling properties of the logMAR metric (where a one-unit change represents a clinically large decrement in vision) and unmeasured variability in the causes and functional consequences of vision impairment that are not captured by visual acuity alone. Accordingly, while the magnitude of the observed effect should be interpreted cautiously, the direction of the association aligns with prior evidence linking poorer vision to increased fall risk [43].
Similarly, careful interpretation is warranted for the inverse association observed between pain interference and injurious falls. Although this finding diverges from studies reporting higher risk of injurious falls among older adults with pain [45, 46], it may reflect behavioral adaptation, whereby individuals experiencing greater pain modify or limit their participation in activities or environments associated with greater injury risk. Such adaptations, whether due to discomfort, reduced confidence, or functional limitation, may reduce exposure to hazardous situations and help explain why greater pain interference was associated with fewer injurious falls in this cohort. This pattern suggests that biological indicators may influence fall risk through accompanying behavioral and environmental pathways, reinforcing the value of multidomain assessment when evaluating fall risk.
Strengths and limitations
This study’s key strength is the comparison of fall risk prediction across the four WHO risk-factor domains, which, to the authors’ knowledge, has not been previously applied. Additional strengths include the use of penalized regression to reduce overfitting, a large community-based sample, standardized fall definitions, and the inclusion of a broad set of candidate predictors not commonly incorporated in prior models [47]. Modeling both any falls and injurious falls, the latter of which is less frequently examined [35, 48], provides complementary insight into outcome-specific patterns of fall risk.
Several limitations should be considered. The cross-sectional design and retrospective fall reporting preclude causal inference and may introduce recall bias, while reliance on self-reported measures may be subject to measurement error and reporting bias, particularly among participants with cognitive impairment or depressive symptoms. Additionally, fall status was coded dichotomously (faller and non-faller), which did not distinguish between single and recurrent fallers or account for heterogeneity in fall circumstances. This may obscure differences in fall risk severity and profiles by grouping falls that occur during higher-risk activities, such as sports, with those more reflective of underlying vulnerability. The representation of non-biological domains was constrained by data availability and challenges in operationalizing certain constructs. The relatively healthy, community-dwelling sample and voluntary participation in the cohort also limit generalizability to frailer or institutionalized populations and introduce selection bias. Findings should also be extrapolated cautiously to low- and middle-income countries, where differing socioeconomic and health system contexts may influence the transportability and stability of model estimates.
Future research should prioritize prospective collection of fall outcomes, external validation in diverse populations, and the incorporation of a broader range of nonbiological constructs. Alternative measurement approaches, such as device-based and ecological momentary approaches, may enhance the ability to capture fall-related determinants and improve the prediction of any falls.
Conclusion
Falls remain a major health concern for older adults, highlighting the need for effective and feasible screening approaches. This study demonstrates that parsimonious, biologically-based models can meaningfully discriminate fall risk and perform comparably to more complex multidomain models in community settings. Given the differences in risk profiles by fall outcome, clinicians should recognize that a “one-size-fits-all” approach is unlikely to be adequate. Prevention strategies should be tailored to the fall type to enhance clinical relevance and impact.
Supplementary Information
Additional file 1. TRIPOD checklist: Prediction model development
Additional file 2. Table S1. Candidate predictors and operational definitions. [49–74]
Additional file 3. Figure S1. Missing data across candidate predictors (N = 514)
Additional file 4. Table S2. Baseline distribution of candidate predictors in the analytic sample
Additional file 5. Table S3. Area under the curve (AUC) values for sequentially expanded fall risk prediction models
Additional file 6. Table S4. Regression coefficients for the final prediction models
Additional file 7. Table S5. Internal validation of predictive models for any falls and injurious falls using bootstrap resampling (1000 iterations)
Authors’ contributions
FE: conceptualized the study, conducted the statistical analyses, interpreted the data, and drafted the initial manuscript.SS: helped to conceptualize the study, acquired the data, interpreted the data, revised the manuscript, and co-supervised the first author.CD: helped to conceptualize the study, acquired the data, interpreted the data, revised the manuscript, and co-supervised the first author.MB: conceptualized the study, acquired the data, interpreted the data, revised the manuscript, and supervised the first author.
Funding
This work was supported by Canadian Institutes of Health Research grant number PJT-165844.
Data availability
The data sets used and/or analyzed during the current study are available from the corresponding author upon reasonable request.
Declarations
Ethics approval and consent to participate
This study was conducted in accordance with the Declaration of Helsinki. Ethics approval was obtained from the Hamilton Integrated Research Ethics Board (HiREB; #7380), and all participants provided informed consent prior to participation.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
- 1.Salari N, Darvishi N, Ahmadipanah M, Shohaimi S, Mohammadi M. Global prevalence of falls in the older adults: a comprehensive systematic review and meta-analysis. J Orthop Surg Res. 2022;17(1):334. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Stevens JA, Mack KA, Paulozzi LJ, Ballesteros MF. Self-reported falls and fall-related injuries among persons aged > or = 65 years–United States, 2006. J Saf Res. 2008;39(3):345–9. [DOI] [PubMed] [Google Scholar]
- 3.James SL, Lucchesi LR, Bisignano C, Castle CD, Dingels ZV, Fox JT, et al. The global burden of falls: global, regional and national estimates of morbidity and mortality from the Global Burden of Disease Study 2017. Inj Prev. 2020;26(Supp 1):i3–11. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Ontario PH. National Fall Prevention Month 2022 2022 [cited 2025 13 September]. Available from: https://www.publichealthontario.ca/en/About/News/2022/Fall-Prevention-Month-2022.
- 5.Ageing, WHO UL. WHO global report on falls prevention in older age. Geneva: WHO; 2008. [Google Scholar]
- 6.Gade GV, Jørgensen MG, Ryg J, Riis J, Thomsen K, Masud T, et al. Predicting falls in community-dwelling older adults: a systematic review of prognostic models. BMJ Open. 2021;11(5):e044170. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Saunders S, D’Amore C, Hao Q, El-Moneim NA, Richardson J, Kuspinar A, et al. Risk Factors for Falls in Community-Dwelling Older Adults: An Umbrella Review. J Am Med Dir Assoc. 2025;26(9):105765. [DOI] [PubMed] [Google Scholar]
- 8.National Institute for Health and Care Excellence (NICE). Falls: assessment and prevention in older people and in people 50 and over at higher risk [Internet]. London: NICE; 2025. Available from: https://www.nice.org.uk/guidance/ng249. Cited 1 Apr 2026. [PubMed]
- 9.Montero-Odasso M, van der Velde N, Martin FC, Petrovic M, Tan MP, Ryg J, et al. World guidelines for falls prevention and management for older adults: a global initiative. Age Ageing. 2022;51(9):afac205. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Society AG, Society BG, Surgeons AAO, Persons PoFPoO. Guideline for the prevention of falls in older persons. J Am Geriatr Soc. 2001;49(5):664–72. [PubMed] [Google Scholar]
- 11.Scott V, Votova K, Scanlan A, Close J. Multifactorial and functional mobility assessment tools for fall risk among older adults in community, home-support, long-term and acute care settings. Age Ageing. 2007;36(2):130–9. [DOI] [PubMed] [Google Scholar]
- 12.Beauchamp MK, Hao Q, Kuspinar A, Amuthavalli Thiyagarajan J, Mikton C, Diaz T, et al. A unified framework for the measurement of mobility in older persons. Age Ageing. 2023;52(Suppl 4):iv82–5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.World Health Organization. Step safely: strategies for preventing and managing falls across the life-course [Internet]. Geneva: World Health Organization; 2021. Available from: https://www.who.int/publications/i/item/9789240021914. Cited 1 Apr 2026.
- 14.Collins GS, Reitsma JB, Altman DG, Moons KGM. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD Statement. BMC Med. 2015;13(1):1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Saunders S, Kuspinar A, Sibley KM, D’Amore C, Noble T, Griffith LE, et al. The INITIATE (Initial Test for Fall Risk Assessment in the Elderly) prospective cohort study: baseline results. BMC Geriatr. 2025;26:182. 10.1186/s12877-025-06724-9. [DOI] [PMC free article] [PubMed]
- 16.Schwenk M, Lauenroth A, Stock C, Moreno RR, Oster P, McHugh G, et al. Definitions and methods of measuring and reporting on injurious falls in randomised controlled fall prevention trials: a systematic review. BMC Med Res Methodol. 2012;12:50. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Lamb SE, Jørstad-Stein EC, Hauer K, Becker C. Development of a common outcome data set for fall injury prevention trials: the Prevention of Falls Network Europe consensus. J Am Geriatr Soc. 2005;53(9):1618–22. [DOI] [PubMed] [Google Scholar]
- 18.Chandrasekaran S, Hibino H, Gorniak SL, Layne CS, Johnston CA. Fear of Falling: Significant Barrier in Fall Prevention Approaches. Am J Lifestyle Med. 2021;15(6):598–601. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Saunders S, Mayhew A, Kirkwood R, Nguyen K, Kuspinar A, Vesnaver E, et al. Factors Influencing Mobility During the COVID-19 Pandemic in Community-Dwelling Older Adults. Arch Phys Med Rehabil. 2023;104(1):34–42. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Moons KG, Kengne AP, Woodward M, Royston P, Vergouwe Y, Altman DG, et al. Risk prediction models: I. Development, internal validation, and assessing the incremental value of a new (bio)marker. Heart. 2012;98(9):683–90. [DOI] [PubMed] [Google Scholar]
- 21.Lusardi MM, Fritz S, Middleton A, Allison L, Wingood M, Phillips E, et al. Determining Risk of Falls in Community Dwelling Older Adults: A Systematic Review and Meta-analysis Using Posttest Probability. J Geriatr Phys Ther. 2017;40(1):1–36. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Tromp AM, Pluijm SM, Smit JH, Deeg DJ, Bouter LM, Lips P. Fall-risk screening test: a prospective study on predictors for falls in community-dwelling elderly. J Clin Epidemiol. 2001;54(8):837–44. [DOI] [PubMed] [Google Scholar]
- 23.Pavlou M, Ambler G, Seaman SR, Guttmann O, Elliott P, King M, et al. How to develop a more accurate risk prediction model when there are few events. BMJ. 2015;351:h3868. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Ranstam J, Cook JA. LASSO regression. Br J Surg. 2018;105(10):1348. [Google Scholar]
- 25.McNeish DM. Using Lasso for Predictor Selection and to Assuage Overfitting: A Method Long Overlooked in Behavioral Sciences. Multivar Behav Res. 2015;50(5):471–84. [DOI] [PubMed] [Google Scholar]
- 26.Gale CR, Cooper C, Aihie Sayer A. Prevalence and risk factors for falls in older men and women: The English Longitudinal Study of Ageing. Age Ageing. 2016;45(6):789–94. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Welmer A-K, Rizzuto D, Calderón-Larrañaga A, Johnell K. Sex Differences in the Association Between Pain and Injurious Falls in Older Adults: A Population-Based Longitudinal Study. Am J Epidemiol. 2017;186(9):1049–56. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Vatcheva KP, Lee M, McCormick JB, Rahbar MH. Multicollinearity in regression analyses conducted in epidemiologic studies. Epidemiol (Sunnyvale). 2016;6(2). 10.4172/2161-1165.1000227. [DOI] [PMC free article] [PubMed]
- 29.Delacour H, Servonnet A, Perrot A, Vigezzi JF, Ramirez JM. [ROC (receiver operating characteristics) curve: principles and application in biology]. Ann Biol Clin (Paris). 2005;63(2):145–54. [PubMed] [Google Scholar]
- 30.Çorbacıoğlu ŞK, Aksel G. Receiver operating characteristic curve analysis in diagnostic accuracy studies: A guide to interpreting the area under the curve value. Turk J Emerg Med. 2023;23(4):195–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.de Hond AAH, Steyerberg EW, van Calster B. Interpreting area under the receiver operating characteristic curve. Lancet Digit Health. 2022;4(12):e853–5. [DOI] [PubMed] [Google Scholar]
- 32.Zou L, Choi YH, Guizzetti L, Shu D, Zou J, Zou G. Extending the DeLong algorithm for comparing areas under correlated receiver operating characteristic curves with missing data. Stat Med. 2024;43(21):4148–62. [DOI] [PubMed] [Google Scholar]
- 33.Austin PC, Steyerberg EW. Events per variable (EPV) and the relative performance of different strategies for estimating the out-of-sample validity of logistic regression models. Stat Methods Med Res. 2017;26(2):796–808. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Steyerberg EW, Vickers AJ, Cook NR, Gerds T, Gonen M, Obuchowski N, et al. Assessing the performance of prediction models: a framework for traditional and novel measures. Epidemiology. 2010;21(1):128–38. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Dormosh N, van de Loo B, Heymans MW, Schut MC, Medlock S, van Schoor NM, et al. A systematic review of fall prediction models for community-dwelling older adults: comparison between models based on research cohorts and models based on routinely collected data. Age Ageing. 2024;53(7):afae131. [DOI] [PMC free article] [PubMed]
- 36.Ek S, Rizzuto D, Calderón-Larrañaga A, Franzén E, Xu W, Welmer AK. Predicting First-Time Injurious Falls in Older Men and Women Living in the Community: Development of the First Injurious Fall Screening Tool. J Am Med Dir Assoc. 2019;20(9):1163–e83. [DOI] [PubMed] [Google Scholar]
- 37.Centers for Disease Control and Prevention (CDC). Check for safety: a home fall prevention checklist for older adults [Internet]. Atlanta (GA): CDC; 2015. Available from: https://www.cdc.gov/steadi/pdf/steadi-brochure-checkforsafety-508.pdf. Cited 1 Apr 2026.
- 38.Bayati T, Dehghan A, Bonyadi F, Bazrafkan L. Investigating the effect of education on health literacy and its relation to health-promoting behaviors in health center. J Educ Health Promot. 2018;7:127. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Cho YI, Lee S-YD, Arozullah AM, Crittenden KS. Effects of health literacy on health status and health service utilization amongst the elderly. Soc Sci Med. 2008;66(8):1809–16. [DOI] [PubMed] [Google Scholar]
- 40.Ek S, Rizzuto D, Fratiglioni L, Johnell K, Xu W, Welmer AK. Risk Profiles for Injurious Falls in People Over 60: A Population-Based Cohort Study. J Gerontol Biol Sci Med Sci. 2018;73(2):233–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Poh FJX, Shorey S. A Literature Review of Factors Influencing Injurious Falls. Clin Nurs Res. 2018;29(3):141–8. [DOI] [PubMed] [Google Scholar]
- 42.Summary of the Updated American Geriatrics Society/British Geriatrics. Society clinical practice guideline for prevention of falls in older persons. J Am Geriatr Soc. 2011;59(1):148–57. [DOI] [PubMed] [Google Scholar]
- 43.Wieczorek M, Isler M, Landau K, Becker MD, Dawson-Hughes B, Kressig RW, et al. Association Between Visual Acuity and Prospective Fall Risk in Generally Healthy and Active Older Adults: The 3-Year DO-HEALTH Study. J Am Med Dir Assoc. 2024;25(5):789–e952. [DOI] [PubMed] [Google Scholar]
- 44.Asai T, Oshima K, Fukumoto Y, Yonezawa Y, Matsuo A, Misu S. The association between fear of falling and occurrence of falls: a one-year cohort study. BMC Geriatr. 2022;22(1):393. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 45.Cai Y, Leveille SG, Shi L, Chen P, You T. Chronic Pain and Risk of Injurious Falls in Community-Dwelling Older Adults. Journals Gerontology: Ser A. 2020;76(9):e179–86. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 46.Alrawaili SM, Alkhathami KM, Elsehrawy MG, Obaidat SM, Alhwoaimel NA, Alenazi AM. Multisite Pain and Intensity were Associated with History Fall among Older Adults: A Cross-Sectional Study. J Multidiscip Healthc. 2024;17:1241–50. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 47.Gadkaree SK, Sun DQ, Huang J, Varadhan R, Agrawal Y. Comparison of simple vs. performance-based fall prediction models: data from the National Health and Aging Trends Study. Gerontol Geriatr Med. 2015;1:1–10. 10.1177/2333721415584850. [DOI] [PMC free article] [PubMed]
- 48.Ward RE, Leveille SG, Beauchamp MK, Travison T, Alexander N, Jette AM, et al. Functional performance as a predictor of injurious falls in older adults. J Am Geriatr Soc. 2015;63(2):315–20. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 49.Rockwood K, Song X, MacKnight C, Bergman H, Hogan DB, McDowell I, et al. A global clinical measure of fitness and frailty in elderly people. CMAJ. 2005;173(5):489–95. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 50.Han SJ, Jung HW, Lee JH, Lim J, Moon SD, Yoon SW, et al. Clinical Frailty Scale, K-FRAIL questionnaire, and clinical outcomes in an acute hospitalist unit in Korea. Korean J Intern Med. 2021;36(5):1233–41. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 51.Jung HW, Baek JY, Kwon YH, Jang IY, Kim DY, Kwon HS, et al. At-Point Clinical Frailty Scale as a Universal Risk Tool for Older Inpatients in Acute Hospital: A Cohort Study. Front Med (Lausanne). 2022;9:929555. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 52.Nguyen H, Manolova G, Daskalopoulou C, Vitoratou S, Prince M, Prina AM. Prevalence of multimorbidity in community settings: A systematic review and meta-analysis of observational studies. J Comorbidity. 2019;9:2235042X19870934. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 53.Mänty M, Heinonen A, Leinonen R, Törmäkangas T, Sakari-Rantala R, Hirvensalo M, et al. Construct and predictive validity of a self-reported measure of preclinical mobility limitation. Arch Phys Med Rehabil. 2007;88(9):1108–13. [DOI] [PubMed] [Google Scholar]
- 54.Herman T, Giladi N, Hausdorff JM. Properties of the ‘timed up and go’ test: more than meets the eye. Gerontology. 2011;57(3):203–10. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 55.Marques A, Almeida S, Carvalho J, Cruz J, Oliveira A, Jácome C. Reliability, Validity, and Ability to Identify Fall Status of the Balance Evaluation Systems Test, Mini-Balance Evaluation Systems Test, and Brief-Balance Evaluation Systems Test in Older People Living in the Community. Arch Phys Med Rehabil. 2016;97(12):2166–73. e1. [DOI] [PubMed] [Google Scholar]
- 56.Bohannon RW. Grip Strength: An Indispensable Biomarker For Older Adults. Clin Interv Aging. 2019;14:1681–91. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 57.Lord SR, Smith ST, Menant JC. Vision and falls in older people: risk factors and intervention strategies. Clin Geriatr Med. 2010;26(4):569–81. [DOI] [PubMed] [Google Scholar]
- 58.Tomioka K, Ikeda H, Hanaie K, Morikawa M, Iwamoto J, Okamoto N, et al. The Hearing Handicap Inventory for Elderly-Screening (HHIE-S) versus a single question: reliability, validity, and relations with quality of life measures in the elderly community, Japan. Qual Life Res. 2013;22(5):1151–9. [DOI] [PubMed] [Google Scholar]
- 59.Yang F. Identification of optimal foot tactile sensation threshold for detecting fall risk among community-dwelling older adults. Phys Ther. 2021;101(8):pzab119. 10.1093/ptj/pzab119. [DOI] [PubMed]
- 60.Nasreddine ZS, Phillips NA, Bédirian V, Charbonneau S, Whitehead V, Collin I, et al. The Montreal Cognitive Assessment, MoCA: a brief screening tool for mild cognitive impairment. J Am Geriatr Soc. 2005;53(4):695–9. [DOI] [PubMed] [Google Scholar]
- 61.Wagner S, Helmreich I, Dahmen N, Lieb K, Tadic A. Reliability of three alternate forms of the trail making tests a and B. Arch Clin Neuropsychol. 2011;26(4):314–21. [DOI] [PubMed] [Google Scholar]
- 62.de Vries M, Seppala LJ, Daams JG, van de Glind EMM, Masud T, van der Velde N. Fall-Risk-Increasing Drugs: A Systematic Review and Meta-Analysis: I. Cardiovascular Drugs. J Am Med Dir Assoc. 2018;19(4):371.e1-.e9. [DOI] [PubMed] [Google Scholar]
- 63.Seppala LJ, Wermelink A, de Vries M, Ploegmakers KJ, van de Glind EMM, Daams JG, et al. Fall-Risk-Increasing Drugs: A Systematic Review and Meta-Analysis: II. Psychotropics. J Am Med Dir Assoc. 2018;19(4):e37111–17. [DOI] [PubMed] [Google Scholar]
- 64.Seppala LJ, van de Glind EMM, Daams JG, Ploegmakers KJ, de Vries M, Wermelink A, et al. Fall-Risk-Increasing Drugs: A Systematic Review and Meta-analysis: III. Others. J Am Med Dir Assoc. 2018;19(4):372.e1-.e8. [DOI] [PubMed] [Google Scholar]
- 65.Cleeland CS, Ryan KM. Pain assessment: global use of the Brief Pain Inventory. Ann Acad Med Singap. 1994;23(2):129–38. [PubMed] [Google Scholar]
- 66.Yardley L, Beyer N, Hauer K, Kempen G, Piot-Ziegler C, Todd C. Development and initial validation of the Falls Efficacy Scale-International (FES-I). Age Ageing. 2005;34(6):614–9. [DOI] [PubMed] [Google Scholar]
- 67.Radloff LS, The CES-D, Scale. A Self-Report Depression Scale for Research in the General Population. Appl Psychol Meas. 1977;1(3):385–401. [Google Scholar]
- 68.Song J, Wu X, Zhang Y, Song P, Zhao Y. Association between changes in depressive symptoms and falls: The China health and retirement longitudinal study (CHARLS). J Affect Disord. 2023;341:393–400. [DOI] [PubMed] [Google Scholar]
- 69.Sherbourne CD, Stewart AL. The MOS social support survey. Soc Sci Med. 1991;32(6):705–14. [DOI] [PubMed] [Google Scholar]
- 70.Washburn RA, Smith KW, Jette AM, Janney CA. The Physical Activity Scale for the Elderly (PASE): development and evaluation. J Clin Epidemiol. 1993;46(2):153–62. [DOI] [PubMed] [Google Scholar]
- 71.D’Amore C, Lajambe L, Bush N, Hiltz S, Laforest J, Viel I, et al. Mapping the extent of the literature and psychometric properties for the Physical Activity Scale for the Elderly (PASE) in community-dwelling older adults: a scoping review. BMC Geriatr. 2024;24(1):761. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 72.Keller HH, Goy R, Kane SL. Validity and reliability of SCREEN II (Seniors in the community: risk evaluation for eating and nutrition, Version II). Eur J Clin Nutr. 2005;59(10):1149–57. [DOI] [PubMed] [Google Scholar]
- 73.Saelens BE, Sallis JF, Black JB, Chen D. Neighborhood-based differences in physical activity: an environment scale evaluation. Am J Public Health. 2003;93(9):1552–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 74.Carlson JA, Frank LD, Ulmer J, Conway TL, Saelens BE, Cain KL, et al. Work and Home Neighborhood Design and Physical Activity. Am J Health Promotion. 2018;32(8):1723–9. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Additional file 1. TRIPOD checklist: Prediction model development
Additional file 2. Table S1. Candidate predictors and operational definitions. [49–74]
Additional file 3. Figure S1. Missing data across candidate predictors (N = 514)
Additional file 4. Table S2. Baseline distribution of candidate predictors in the analytic sample
Additional file 5. Table S3. Area under the curve (AUC) values for sequentially expanded fall risk prediction models
Additional file 6. Table S4. Regression coefficients for the final prediction models
Additional file 7. Table S5. Internal validation of predictive models for any falls and injurious falls using bootstrap resampling (1000 iterations)
Data Availability Statement
The data sets used and/or analyzed during the current study are available from the corresponding author upon reasonable request.