Risk Factors for Mobility Decline in Community-Dwelling Older Adults: A Systematic Literature Review

Mobility is essential to maintaining independence for older adults. This systematic review aimed to summarize evidence about self-reported risk factors for self-reported mobility decline; and to provide an overview of published prognostic models for self-reported mobility decline among community-dwelling older adults. Databases were searched from inception to June 2, 2020. Studies were screened by two independent reviewers who extracted data and assessed study quality. Sixty-one studies (45,187 participants) were included, providing information on 107 risk factors. High-quality evidence and moderate/large effect sizes for the association with mobility decline were found for older age beyond 75 years, the presence of widespread pain, and mobility modi ﬁ cations. Moderate – high quality evidence and small effect sizes were found for a further 21 factors. Three model development studies demonstrated acceptable model performance, limited by high risk of bias. These ﬁ ndings should be considered in intervention development, and in developing a prediction instrument for practical application.

Mobility is defined as "the ability to move oneself within community environments that expand from one's home, to the neighborhood, and to regions beyond" (Webber, Porter, & Menec, 2010).Maintaining mobility is fundamental to aging well for older adults (Guralnik et al., 1993).Limited mobility is linked to functional decline, mortality, and increased health care costs (Guralnik et al., 1993).Changes in mobility are important as conceptually they precede disability within models of disablement and are potential targets for rehabilitation (Verbrugge & Jette, 1994).
Mobility may be assessed using objective or self-reported measures.While it is acknowledged that different aspects of mobility may be measured with each, multiple studies have found a moderate correlation between the two approaches (Alexander et al., 2000;Syddall, Westbury, Cooper, & Sayer, 2015).Selfreported mobility is important as it is the most commonly used measure clinically (Chung, Demiris, & Thompson, 2015).Objective physical performance measures do not reflect the individual's perceptions about their own mobility, and are most often conducted in a supervised, controlled environment, thus unreflective of a real-world situation.Potential risk factors for mobility decline may also be measured objectively or by self-report.Self-reported measures of these factors are also important, as they have low response burden, can capture both current and historical information, and can assess psychological and social factors.
Two previous reviews of risk factors for mobility decline combined objective and self-reported measures of risk factors and mobility, but these were narrative rather than systematic reviews (Rantakokko, Mänty, & Rantanen, 2013;Yeom, Fleury, & Keller, 2008).Consolidated evidence of self-reported risk factors contributing to self-reported mobility decline is needed to inform clinical decision making and to inform the development of a prognostic tool to help clinicians and researchers identify older adults at risk of mobility decline.The primary objective of this systematic review was to synthesize available evidence for self-reported factors which predict decline in self-reported mobility after 12 months to 5 years of follow-up among community-dwelling older adults.The secondary objective was to identify, describe, and synthesize the predictive accuracy of prognostic models developed to predict risk of selfreported mobility decline in community-dwelling older adults.

Methods
This systematic review is registered at the International Prospective Register of Systematic Reviews database (PROSPERO 2019 CRD42019135420.Available from: https://www.crd.york.ac.uk/ prospero/display_record.php?ID=CRD42019135420).The review is reported according to the Preferred Reporting Items for Systematic Review and Meta-Analyses guidelines (Moher, Liberati, Tetzlaff, & Altman, 2009), and the CHecklist for critical Appraisal and data extraction for systematic Reviews of prediction Modelling Studies (CHARMS) (Moons et al., 2014).

Search Strategy
The search strategy was developed conjointly with an information scientist (SK).We searched the MEDLINE (via OVID), EMBASE (via OVID), PsycINFO (via OVID), and CINAHL (via EBSCOhost) databases to identify studies that assessed the predictive value of selfreported risk factors on self-reported mobility decline.Search terms included controlled vocabulary (e.g., MeSH) terms and free-text terms.No date or language limits were applied to the search.The search strategy for MEDLINE is presented in Supplementary Table S1 (available online), and this was translated to the relevant controlled vocabulary headings and syntax for each database.The search was conducted on June 26, 2019, with an update run on June 2, 2020.Supplementary searches of the reference and citation lists of included studies and relevant systematic reviews were undertaken to identify additional eligible studies.

Study Selection
Inclusion criteria.
• Studies involving community-dwelling older adults with a minimum mean age of 60 years at recruitment.• Where a study collected both self-reported measures and objective measures, the study was eligible if data reported the relationship between self-reported measures and selfreported outcomes.• Follow-up period of between 12 months and 5 years.
• Published in English, Spanish, or Portuguese (languages spoken by the research team).
• Studies including only patients with a specific medical condition (e.g., Parkinson's disease, dementia, stroke).• Studies conducted in residential aged care settings.
• Studies published in abstract form only.
Search results were imported into Covidence (Veritas Health Innovation, Melbourne, Australia, www.covidence.org) for screening.Titles and abstracts of all identified citations were independently screened by two reviewers (PN and MS).Following title and abstract screening, the full text of all potentially eligible articles was retrieved and screened independently by the same two reviewers.Inclusion disagreements were resolved through discussion, with an adjudicator (PN and MS) available.

Data Extraction
A data extraction form in Microsoft Excel (Microsoft Corporation, Redmond, WA) was developed and piloted independently by two reviewers before independent data extraction from all articles (PN and MS).Disagreements were resolved by discussion.
Studies were categorized as etiological, predictor finding, or model development using the following definitions (Moons, Royston, Vergouwe, Grobbee, & Altman, 2009): • Etiological studies: A study examining causal association between predefined risk factor/s and mobility decline.• Predictor finding studies: A study examining the contribution of multiple risk factors to predict mobility decline, without quantification of the predictive ability of the model.• Model development studies: A study that aimed to develop a multivariable prediction model, assigning weighting to each identified risk factor and developing a final prediction model.These studies may or may not include internal validation studies.
Data extracted from all studies included: country of origin, year of study conduct, recruited and analyzed sample sizes, study design, type of study, length of follow-up, and participant characteristics (population source and setting, eligibility criteria, recruitment method, and sociodemographic characteristics).Method of assessment was recorded for each risk factor and mobility outcome.
For etiological studies, we extracted data for the specific risk factor/s of interest as stated in the study.For predictor finding studies, we extracted data on all included risk factors.We extracted unadjusted and adjusted effect estimates (odds ratio [OR], relative risk [RR], or hazard ratio [HR]) and corresponding SE or confidence intervals for each risk factor.For model development studies, we extracted modeling method, handling of risk factors and missing data, method for selection of risk factors, model performance measures, and model evaluation.We contacted authors when insufficient information was reported.

Assessment of Risk of Bias in Included Studies
Two reviewers (PN and MS) independently assessed the risk of bias in included studies using the Quality In Prognosis Studies tool for etiological and predictor finding studies and the Prediction model Risk of Bias Assessment Tool for model development studies (Hayden, van der Windt, Cartwright, Cote, & Bombardier, 2013;Wolff et al., 2019).
The Quality In Prognosis Studies tool assesses risk of bias across six domains: study participation, study attrition, risk factor measurement, outcome measurement, study confounding, and statistical analysis and reporting (Hayden et al., 2013).Each domain was rated as low, moderate, or high risk of bias (Hayden et al., 2013).The Prediction model Risk of Bias Assessment tool includes 20 signaling questions grouped within four domains (participant selection, predictors, outcome, and analysis) to evaluate the risk of bias of prediction models (Wolff et al., 2019).Studies were classified as low risk of bias if all domains were rated as low risk of bias.Any disagreements in scoring were resolved through discussion.

Data Synthesis and Analysis
Descriptive characteristics of all included studies were summarized and assessed for suitability for meta-analysis by considering the clinical heterogeneity of included studies based on population, definition of risk factors and/or mobility decline, and methodology (Riley et al., 2019).Adjusted effect estimates (minimum adjustment for age and gender/sex) were reported in the narrative analysis.Odds ratios of 1.4-2.5 were considered as small effect, 2.5-4.25 were classified as moderate, and 4.25 or greater as large (Chinn, 2000).Odds ratios were considered to indicate no association if the 95% confidence interval crossed 1.0.Information from risk factor and prognostic model studies were evaluated separately because their objectives and model-building methods are different (van den Berg et al., 2013).Performance of prognostic models were summarized and presented narratively.

Narrative Analysis
We applied the following criteria proposed by Hayden, Tougas, Riley, Iles, and Pincus (2014) to categorize the consistency of evidence for each risk factor.Consistent evidence defined as two or more studies from two or more cohorts with ≥75% of studies showing the same direction of effect (i.e., four studies from three cohorts, of which three studies show the same direction of effect).Limited evidence defined as: single study or multiple studies from the same cohort.Inconsistent evidence defined as: two or more studies from two or more cohorts with <75% of studies showing the same direction of effect (i.e., four studies from three cohorts, of which two studies show the same direction of effect).

Assessment of Quality of Evidence
The quality of the body of evidence was synthesized using the Grading of Recommendations Assessment, Development and Evaluation approach modified for use in reviews of risk factor studies (Huguet et al., 2013).Overall quality of evidence was rated as high, moderate, low, or very low considering study limitations, inconsistency of results, indirectness, imprecision, publication bias, moderate/large effect size, and the presence of a dose effect.

Study Selection
The process of study selection is shown in Figure 1.In total, 26,079 citations (18,201 with duplicates removed) were identified.Of these, 187 articles progressed to full-text review, and 61 studies were included in the review.Among included studies, 47 were classified as etiological studies, 11 as predictor finding studies, and three were model development studies.
The results of the Prediction model Risk of Bias Assessment tool appraisal for the model development studies are presented in Supplementary Table S4 (available online).All three studies (Chaves et al., 2000;Papachristou et al., 2017;Reynolds & Silverstein, 2003) were judged as having high risk of bias for the analysis domain.This was due to the absence of model performance measures, selection of predictor variables based on univariable analyses, and lack of reporting results of multivariable analyses.All three studies were judged to have both high risk of bias overall and high concern regarding applicability to the review question.
a When more than one study examining the same factor was derived from the same cohort, the highest number of participants was added to the participant total.b Two or more studies from two or more cohorts; >75% of studies showing the same direction of effect.c Mobility modifications defined as any of resting during the task, using an aid, using handrails, reduced frequency, or reduced speed of performing the task.d Single study or multiple studies from single cohort.e Two or more studies from two or more cohorts; <75% of studies showing the same direction of effect, or single study with insufficient data to assess association.
variation in methods of measurement for risk factors and/or mobility outcomes, we were unable to formally pool the results using meta-analysis (Riley et al., 2019).
Of the 107 risk factors examined in these studies, there was consistent evidence (less than or equal to two studies from less than or equal to two cohorts) that 24 factors were associated with mobility decline.Of these 24, three were associated with a moderate/large effect and 21 associated with a small effect.There was also consistent evidence of no association with mobility decline for three factors.The results and discussion will focus on these 27 factors.
Among the remaining 80 risk factors, there was limited evidence (single study or multiple studies from a single cohort) of association with mobility decline for 30, limited evidence of no effect on mobility decline for 30, and inconsistent or unclear effects on mobility decline for 20 factors (Table 1 and Supplementary Table S5 [available online]).
Factors with consistent moderate/large effect on mobility decline.Three factors were consistently associated with mobility decline with moderate-large effect sizes.There was high quality evidence for older age, with large effect sizes reported for participants aged 75 years or older compared to those younger than 75 years.There was also high quality evidence and a dose-response effect observed for mobility modifications (resting during the task, using an aid, using handrails, reduced frequency or reduced speed of performing the task).Moderate quality evidence was found for the presence of widespread pain as a risk factor for mobility decline.
Factors with consistent small effect on mobility decline.For the remaining 21 risk factors consistently associated with mobility decline, the strength of the association represented by ORs was considered small (Table 1).There was high-quality evidence for five demographic and socioeconomic risk factors.Female gender was associated with increased risk of mobility decline, as was low annual income; low number of financial assets; low diversity in social relations; and low social engagement.
Five risk factors were associated with physical-and fatiguerelated issues.Low physical activity was associated with mobility decline with high quality evidence.There was moderate quality evidence for not walking every day, having at least one fall in the past year, higher fatigue after exertion, and greater tiredness in daily activities as risk factors for mobility decline.
Evidence for the consistent small effect of 10 health factors on mobility decline ranged from high to low quality.High quality evidence was found for increasing number of health conditions.Specific health conditions presenting as risk factors for mobility decline included depression (high quality evidence), eye conditions (high quality evidence), heart conditions (high quality evidence), history of stroke (moderate quality evidence), and sensory difficulties (low quality evidence).Other factors that showed a consistent small association with increased risk of mobility decline were hip pain (moderate quality evidence), knee pain (moderate quality evidence), poor self-rated health (high quality evidence), and reporting a hospital stay in the past year (Table 1 and Supplementary Table S5 [available online]).
Factors with consistent lack of effect on mobility decline.
Moderate quality evidence was found for three factors which consistently demonstrated no association with increased risk of mobility decline: weight gain in the past 12 months, history of or current cancer diagnosis, and reporting a higher number of physical environment barriers (e.g., long distances, busy traffic, rough or hilly terrain, high crime area; Table 1 and Supplementary Table S5 [available online]).

Model Development Studies
The three model development studies are summarized in Table 2 and Supplementary Table S2 (available online).Follow-up periods varied from recruitment: 18 months (Chaves et al., 2000); 36 months (Papachristou et al., 2017); and 60 months (Reynolds & Silverstein, 2003).Overall, the three models showed a moderate discriminative ability (area under curve [AUC] = 0.68-0.81),but calibration measures were poorly reported or missing.Only one model was internally validated (Chaves et al., 2000) and none undertook external model validation.

Discussion
This systematic review provides evidence on self-reported factors which predict decline in self-reported mobility between 12 months and 5 years in older adults.Among the 107 risk factors examined in the included studies, we found that 24 had consistent moderate or high quality evidence indicating their association with mobility decline.Effect sizes were generally small, with moderate/large effect sizes observed for three factors: older age (≥75 years), widespread pain, and mobility modifications.Three existing model development studies were identified, all demonstrating moderate model discriminatory performance, and limited by high risk of bias because of analytical shortcomings.
To our knowledge, this is the first systematic review to comprehensively summarize available evidence about self-reported risk factors for self-reported mobility decline among older adults.Among the 107 risk factors we reviewed, only one quarter had sufficient evidence from which to draw firm conclusions regarding association with mobility decline.This does not mean that the remaining factors do not play a role in mobility decline in later life, rather that evidence was insufficient to draw such a conclusion.Significant heterogeneity across studies in definitions and assessments of risk factors and mobility decline precluded quantitative synthesis of the data, and consequently our findings must be interpreted with caution.
A 2008 narrative review concluded that the association between older age and increased mobility limitations was statistically significant over 70 years of age (Yeom et al., 2008).Agerelated changes in mobility may result from physiological changes, such as decreased muscle strength and power, reduced bone mass, and decreased response to balance perturbations (Byrne, Faure, Keene, & Lamb, 2016).However, age-related changes in mobility may also be associated with, or accelerated by, lack of physical activity, leading to a negative cycle of decline (Brach et al., 2003).
The negative effects of widespread pain on physical functions, including mobility, are well established (Butera, Roff, Buford, & Cruz-Almeida, 2019).However, the pathway from widespread pain to mobility limitation is not clear.Nagi's disablement model suggests that widespread pain may lead to decreased physical activity, and in turn loss of muscle strength and the development of mobility limitations (Verbrugge & Jette, 1994).However, Leveille et al. concluded that pain is a unique domain as a cause of mobility difficulty, independent of the usual pathway via physical impairments (Leveille et al., 2007).
Mobility modifications, such as resting, using an aid, using handrails, reduced frequency, or speed of performing a task are important risk factors among older adults as they have been identified as "preclinical mobility decline" (Fried et al., 2000;Rantakokko et al., 2013).Making such modifications may compensate for the impact of underlying health changes, such as pain, reduced muscle strength, or fatigue on mobility (Fried et al., 2000).An individual's transition to mobility difficulty easily occurs when any internal or external changes overwhelm the compensations being employed.Our findings reiterate the importance of assessing the use of mobility modifications to identify people at high risk of mobility decline so the underlying causes can be addressed.This review identifies several opportunities for targeting modifiable factors associated with mobility decline, which should be considered in management of older adults at risk of mobility decline, and intervention development.We found high quality evidence of the association between two social factors (low diversity in social relations and low social engagement) and increased risk of mobility decline.This highlights the importance of assessing social outcomes, which are often neglected, and the potential of social prescribing interventions.We also found high quality evidence for low physical activity as a risk factor for mobility decline.Many physical activity programs have been used to improve mobility in older adults, with positive effects (Rantakokko et al., 2013).Assessing and addressing fears of falling in older adults is also important, as indicated by our finding of moderate quality evidence for fear of falling as a risk factor for mobility decline.
Our findings highlight the complexity of managing older adults at risk of mobility decline.Mobility decline in older age is multifactorial, and many older adults may report multiple risk factors concurrently.Combinations of multiple risk factors may have greater impact than the sum of their individual effects, and future research should explore this.Future interventions targeting maintenance of mobility should be multidimensional, combine interventions for multiple risk factors, and control for confounding factors.Our findings also highlight the limited evidence for existing prognostic models for mobility decline.These results may guide future longitudinal studies to develop a prediction instrument for practical application.

Limitations of the Review
Despite the authors' effort to construct a sensitive search strategy, there is a chance that relevant studies were overlooked.Of the included studies, 70% (43/61) came from two countries (United States and Finland), which may impact the generalizability of findings to other populations.Only 3% of studies (2/61) included participants from low-income countries (World Bank, 2020).None of the included studies were conducted in Africa, and only two were conducted in Asia (Hong Kong and Japan).Differing lifestyles and cultures may mean that risk factors are not the same in these settings, and findings should not be generalized.Studies in more diverse international settings are warranted.Grouping together everyone with mobility decline may dilute associations linked with specific subtypes of mobility decline, such as gradual versus rapid decline.Analysis of risk factors according to definitions of mobility decline may give more insight into causal pathways.Almost all included studies (59/61, 97%) used narrow measures of mobility, focusing on ability to walk and/or climb stairs.Use of broader measures of mobility, such as life-space assessment may enable identification of additional risk factors, and/or which risk factors are associated with different aspects of mobility.Finally, by excluding studies with follow-up periods of <12 months or >5 years, there is a risk that studies showing short-or long-term results were missed.

Conclusion
In conclusion, our systematic review identified 24 wide-ranging self-reported risk factors consistently associated with self-reported mobility decline in older people.Older age beyond 75 years, the presence of widespread pain, and mobility modifications were the risk factors with the highest effect.

Figure 1 -
Figure 1 -Preferred Reporting Items for Systematic Reviews and Meta-Analyses flowchart.

•
Longitudinal cohort studies evaluating the ability of self-reported factors to predict self-reported mobility decline.Randomized controlled trials were eligible if they provided appropriate data and treatment-effect modification analyses were reported.•Prognostic model studies, with or without validation.Models had to contain at least two self-reported risk factors, and include age and sex as a minimum of adjustment factors.
• A self-reported measure of mobility, defined as difficulty with walking and/or difficulty in climbing stairs and/or life-space assessment, assessed at baseline and follow-up.•A clear definition of mobility decline.
(Huguet et al., 2013)e to Table 1 Overview of Risk Factors for Mobility Decline, Presented by Direction of Association, Using the Adapted GRADE(Huguet et al., 2013)

Table 2
Synthesis of Risk Factors for Mobility Decline Examined in Included Model Development StudiesNote.Apparent performance is the performance observed in the development data.AUC = area under the curve; IADL = instrumental activities of daily living; NR = not reported.Percentage of individuals whose predicted and observed probabilities are the same.b Age, sex, hypertension, diabetes, cancer, lung disease, heart condition, psychiatric problems, arthritis, stroke, married, ethnicity (Hispanic and Black), family network, asset complexity, negative affect, cognition, home modification, weight, impairment, smoke, service use, health insurance, onset of hypertension, onset of diabetes, onset of cancer, onset of lung disease, onset of heart condition, onset of psychiatric problems, a onset of arthritis, onset of stroke, deterioration in affect, deterioration in cognition, new home modifications, weight loss, onset of Nagi impairments, onset of IADL impairments, smoking less, new service use, vigorous exercise, and preventive test.