Physical Function and Frailty for Predicting Adverse Outcomes in Older Primary Care Patients

Abstract

Objective:

To explore the predictive ability of the short physical performance battery (SPPB), Late Life Function and Disability Instrument-Function component (LLFDI-function) and frailty phenotype, for falls, hospitalizations, emergency department (ED) visits and low self-rated health (SRH) over 1 and 2 years in older adults.

Design:

Secondary analysis of data from a longitudinal study, the Boston Rehabilitative Impairment Study of the Elderly.

Setting:

Primary care.

Participants:

391 adults ≥65 years old at risk for disability who completed ≥1 follow-up call.

Interventions:

None.

Main Outcome Measures:

We computed separate logistic regression models using the SPPB, LLFDI-function and frailty phenotype as independent variables, and falls, hospitalizations, ED visits and SRH over 1 and 2 years as dependent variables. Receiver operating characteristic (ROC) curves were constructed and the areas under the curves (AUCs) calculated.

Results:

Participants had a mean age of 76.5 (7.1) years. The SPPB, LLFDI-function and frailty phenotype all predicted hospitalizations and low SRH over a 1- and 2-year timeframe (OR min-max 1.35-1.51 and 1.67-3.07, respectively). Over 2 years, the SPPB predicted ED visits (OR 1.28) and the LLFDI-function predicted falls (OR 1.31). The LLFDI-function predicted low SRH better than the frailty phenotype over 1 year. There were no differences between the measures for any of the other outcomes.

Conclusion:

The SPPB, LLFDI-function and frailty phenotype had similar accuracy for predicting falls, hospitalizations, ED visits and low SRH over 1 and 2 years among older primary care patients at risk for disability. As a result, when considering the optimal screening tool for older adults, the choice between a measure of function and frailty may ultimately depend on clinical preference and context.

The world’s population is aging with people ≥65 years making up over 8 percent of the population, a number that is expected to grow 2-fold by 2050.¹ This demographic change will affect healthcare costs given that older adults account for a large percentage of utilization of costly services including emergency department (ED) visits² and hospitalizations with prolonged lengths of stay.³ Given that older adults are likely to have multiple health conditions, determining the risk of adverse outcomes such as falls and hospitalizations based on chronic diseases and their severity is difficult.⁴ To circumvent this problem, there is growing interest in other clinically feasible methods for identifying those who are at most risk for decline.⁵ Identifying such individuals is imperative for guiding targeted interventions and to do so, clinicians need access to screening tools that are quick to perform and easily interpretable. Currently, indices of both frailty and physical function are commonly advocated for screening older adults,^6-11 however they are conceptually different and comparisons of their predictive validity are scarce.

Frailty is a clinical syndrome characterized by increased vulnerability to stressors, and adverse health outcomes.¹² Measures of frailty predict falls,⁶ hospitalizations,⁶ and mortality^6-8 in community-dwelling older adults. A challenge in applying frailty measures is that the criteria used to determine frailty can vary and measures may be comprised of multiple constructs (e.g., physical function, cognitive function, psychosocial factors).¹³ In a review of frailty measures, 67 different tools were found that consisted of 2 to 92 items.¹⁴ The most highly cited instruments included a measure of physical function but the inclusion of other domains such as comorbidity and cognition varied amongst the instruments indicating that they stemmed from different underlying theories of frailty.¹⁴ Therefore, the clinical utility of frailty measures as screening tools is dependent on the underlying constructs that determine their design.

In contrast, physical function is a tangible construct recognized by leading disablement paradigms¹⁵ and coined as the “sixth vital sign” for its ability to phenotype older adults.¹⁶ In the context of direct clinical care, a short test of physical function may be easier to administer and interpret and can be used to guide treatment towards specific impairments. A number of studies demonstrate the predictive ability of a simple functional measure, the Short Physical Performance Battery (SPPB),⁹ for adverse outcomes such as falls¹⁰ and mortality.^9,11 Similarly, a patient-reported measure of function, the Late Life Function and Disability Instrument- function component (LLFDI-function) predicts low self-rated health, hospitalizations and falls.¹⁷

Only two studies have directly compared the predictive ability of physical function and frailty. Conflicting results were observed,^18,19 likely due to variation in the number of items in the frailty measures used and the differing patient populations. In this study, our aim was to explore the predictive validity of a commonly used measure of frailty, Fried’s frailty phenotype,⁶ and well-validated measures of physical function in older primary care patients. Our specific aims were to: 1) Determine the extent to which total scores on the SPPB, LLFDI-function, and Fried’s frailty phenotype predict falls, hospitalizations, ED visits, and self-rated health (SRH) over 1 and 2 years and 2) Explore the predictive ability of the domains of each measure for falls, hospitalizations, ED visits and SRH. We hypothesized that the three measures would have similar predictive value, based on visual inspection of standardized odds ratios and comparison of the areas under the receiver operating characteristic curves. This would be clinically relevant given the simplicity of defining and measuring physical function and would suggest that either screening tool could be used depending on clinician preference and context.

METHODS

This was a secondary analysis of the Boston Rehabilitative Impairment Study of the Elderly (RISE), a longitudinal prospective cohort study of 430 English-speaking primary care patients ≥65 years who were considered at risk for developing disability, defined as difficulty or task modification with walking ½ mile or climbing one flight of stairs.⁵ Participants were recruited from 9 primary care practices between December 2009 and January 2012. Potential participants were excluded if they had a terminal disease, had had major surgery or a myocardial infarction in the last 6 months, had planned major surgery, planned to move out of the Boston area in the next 2 years, had a mini-mental state exam score <18, had a baseline SPPB score <4 or had any major medical problems that would interfere with safe and successful testing.⁵ Participants attended two visits at each time point (baseline, 1 year and 2 years) during which they completed a variety of questionnaires and physical tests. Participants received a phone call every three months over two years to track falls, hospitalizations and ED visits.⁵ Ethics approval to access the data was obtained from the Partners Human Research Committee/Institutional Review Board (#2018P000349/PHS).

Independent Variables Measured at Baseline

Short Physical Performance Battery (SPPB)

The SPPB is a test of lower extremity physical function that is comprised of standing balance tasks (narrow stance, semi-tandem stance and tandem stance), a timed 5-repetition chair stand and a 4-metre walk at usual walking pace.⁹ The balance tests are hierarchical with each subsequent test only performed if the participant achieves 10 seconds holding the previous position. The chair stand test is performed as quickly as possible with the arms crossed over the chest and only completed if the participant is able to stand up from the chair without using their arms for support. It takes 5-7 minutes to complete.²⁰ Participants are assigned a score from 0-4 for each of the three domains (the three balance tasks combined have a maximum score of 4 points based on ability to perform the stance and the time it is held) and the scores are summed for a total score of 0-2 points, with higher scores indicating better performance.⁹ Compared to scores of 10-12, declining scores are associated with an increased risk for all-cause mortality.²¹ Scores derived from the SPPB are reliable and predict a number of adverse outcomes among community-dwelling older adults.^9-11,22

Late Life Function and Disability Instrument-function component (LLFDI-function)

The LLFDI is comprised of two components- function and disability- that can be used independently to measure respondents’ limitations in each area. The function component of the LLFDI (LLFDI-function) measures limitations in the ability to perform discrete activities without help.²³ It is a 32-item questionnaire that is comprised of three dimensions- upper extremity (7 items, for example, unscrewing the lid off a jar), basic lower extremity (14 items, for example, stepping down from a curb) and advanced lower extremity functions (11 items, for example, getting up from the floor).²³ Respondents are asked to select the level of difficulty they experience in doing each task on a typical day without assistance. Raw scores are scaled to 0-100 with higher scores indicating better function.²³ This questionnaire has good convergent, divergent and known-groups validity, test-retest reliability and sensitivity to change.²⁴

Frailty Phenotype

Fried’s frailty phenotype is an established method for classifying frailty that is comprised of five criteria: unintentional weight loss, exhaustion, weakness, slow walking speed and low physical activity.⁶ Participants are given a total score from 0-5 and classified as non-frail (score of 0), pre-frail (score of 1-2), or frail (score of 3-5)⁶. For this analysis, the operationalization of the frailty criteria had to be modified from the original, as a result of the data that were available from the Boston RISE. Body mass index (BMI) was used as a surrogate for weight loss,²⁵ and a BMI ≤21kg/m² was considered positive for unintentional weight loss. Participants completed the Avlund Mobility-Tiredness Scale, a 6-item scale requiring participants to report whether they become tired performing mobility-related tasks.²⁶ Participants scoring in the lowest quintile of the sample were considered positive for exhaustion. Weakness was determined by the single leg press one repetition maximum and participants in the lowest quintile, stratified by gender and BMI quartiles, were considered positive. Walking speed was determined by the 4-meter walk test,²⁷ a measure of usual gait speed. Participants in the lowest quintile, stratified by gender and gender-specific height cut-offs,⁶ were considered positive for slow walking speed. Physical activity was determined by the Physical Activity Scale for the Elderly (PASE), a 28-item questionnaire that asks respondents to report their physical activity during the previous week²⁸ and participants scoring in the lowest quintile, stratified by sex, were considered positive for low physical activity. Fried’s frailty phenotype has been modified extensively in the literature, with up to 87% of studies describing the modification of at least one of the criteria.²⁹ While BMI, 4- metre walk and PASE have been used previously for the weight loss, slowness, and low physical activity criteria, respectively,²⁹ to our knowledge the Avlund Mobility-Tiredness Scale and leg press have not been used before.

Dependent Variables

As part of Boston RISE, participants were asked every three months about falls (“Have you fallen to the ground?”), hospitalizations (“Did you stay overnight in a hospital because of any other reason than a fall?”) and ED visits (“Did you go to an emergency room because of a health problem?”) in the previous 3 months. Whether or not each participant experienced these outcomes from baseline to 1-year follow-up (via the 3, 6, 9 and 12-month phone calls) and from baseline to 2-year follow-up (via the 3, 6, 9, 12, 15, 18, 21 and 24-month phone calls) were the dependent variables. Participants also rated their general health as ‘poor’, ‘fair’, ‘good’, ‘very good’ or ‘excellent’ at the 1- and 2-year follow-up visits. In a meta-analysis of 8 studies, SRH of ‘fair’ and ‘poor’ was associated with increased all-cause mortality.³⁰ Therefore, SRH of ‘fair’ or ‘poor’ was considered positive for the adverse outcome of low SRH.

Statistical Analysis

The Boston RISE dataset consists of 430 participants, thirty-nine of whom were excluded from the present analyses due to the absence of any follow up data. Among the remaining 391 participants, eight participants attended both testing sessions at baseline but did not complete the more difficult physical measures from the second testing session, including the leg press. They were given a score of 1 for the frailty criterion of weakness indicating they were unable to complete the physical measures due to poor physical condition. Thirty-four (8.7%) participants had missing frailty scores due to missing at least one frailty domain. These scores were considered to be missing completely at random³¹ since there were no significant differences in participant characteristics between those with and without scores.

Participants’ characteristics and all predictor and outcome variables were summarized using means and standard deviations or frequencies and percentages as appropriate. We computed separate multivariable logistic regression analyses to examine the association between the independent variables (SPPB, LLFDI-function and frailty phenotype) and the outcomes (falls, hospitalizations, ED visits, and low SRH over 1 and 2 years). The independent variables were standardized to allow comparison of odds ratios (ORs) and 95% confidence intervals (CIs) between variables. All models were adjusted for age and sex. Goodness-of-fit was assessed using the Hosmer-Lemeshow test. A receiver operating characteristic (ROC) curve was constructed and the area under the curve (AUC) was calculated for all models. An AUC of ≥ 0.70 was deemed acceptable and statistical significance was set at p < 0.05. The AUCs for each outcome were compared between predictors using chi-square tests. Finally, similar analyses were conducted to explore which domains of the SPPB (balance, chair rise, gait), LLFDI-function (upper extremity, basic lower extremity, advanced lower extremity) and frailty phenotype (low BMI, tiredness, weakness, low physical activity, slow gait) predicted the adverse outcomes. Multiple imputation using the fully conditional specification method (also called multivariate imputation by chained equations) was used for imputing the missing components required to determine the frailty phenotype (BMI, PASE and leg press). We set the maximum number of iterations as 5 and generated 5 imputation datasets. The results from complete case analysis are reported here and results from the multiple imputation are in the supplementary material. Analyses were performed using Stata, version 14.2 (College Station, TX).

RESULTS

Table 1 shows the participants’ characteristics. Of the 391 participants who had at least one phone call over the 2-year follow-up, 130 (33.3%) were men and the mean age was 76.5 (7.1) years. The mean baseline SPPB score was 8.8 (2.2) points and the mean baseline LLFDI-function score was 55.6 (8.0) points. At baseline, 192 participants (51.8%) were classified as pre-frail and 31 (8.4%) were classified as frail. Participants who were classified as frail had worse scores on both function measures (mean score (SD) 6.2 (1.9) for SPPB and 48.4 (6.8) for LLFDI-function) compared to those classified as pre-frail (SPPB 8.3 (2.2) and LLFDI-function 53.7 (7.0)) and non-frail (SPPB 10.0 (1.5) and LLFDI-function 60.2 (7.2)).

Table 1.

Participant Characteristics (N=391)

Characteristic	n(%)*
Age, y, mean (SD)	76.5 (7.1)
Male	130 (33.3)
Marital status
Previously married (separated/divorced/widowed)	201 (51.4)
Married/living as married	136 (34.8)
Never married	54 (13.8)
Education completed
≤Grade 11	46 (11.8)
High school/general equivalency diploma	113 (28.9)
College/Graduate/Professional school	232 (59.3)
SPPB at baseline, 0-12, mean (SD)	8.8 (2.2)
LLFDI-function at baseline, 0-100, mean (SD)	55.6 (8.0)
Frailty phenotype at baseline†
Non-frail	148 (39.9)
Pre-frail	192 (51.8)
Frail	31 (8.4)
Falls over 1 and 2 years	165 (42.2) and 222 (56.8)
Hospitalizations over 1 and 2 years	98 (25.1) and 155 (39.6)
ED visits over 1 and 2 years	134 (34.3) and 213 (54.5)
Low SRH at 1 year‡ and 2 years§	60 (17.4) and 56 (20.3)

^*unless stated otherwise

^†n=371

^‡n=344

^§n=276.

Table 2 summarizes the results from the logistic regression analyses for adverse outcomes over 2-year follow-up. All measures predicted hospitalizations and low SRH (OR min-max 1.35-1.51 and 1.67-2.11, respectively), LLFDI-function score predicted falls (OR = 1.31, 95% CI 1.06, 1.62) and SPPB score predicted ED visits (OR = 1.28, 95% CI 1.03, 1.58). The results were similar for the outcomes over 1 year except that none of the measures predicted falls or ED visits (supplementary table S1). ROC curves were constructed for all models. The AUCs are presented in table 3 for 2-year outcomes and in supplementary table S2 for 1-year outcomes. The AUCs ranged from 0.51 to 0.75 and the confidence intervals overlapped in all cases. The LLFDI-function was a better predictor of low SRH over 1 year than frailty phenotype (p=0.025) but there were no other differences between measures in the prediction of any of the other outcomes. The AUC was greater than the cut-off of 0.70 for the prediction of low SRH at 1 year follow-up for both the SPPB (0.72, 95%CI 0.65-0.78) and LLFDI-function (0.75, 95%CI 0.69-0.82).

Table 2.

Physical Function, Patient-reported Function and Frailty as Predictors of Adverse Outcomes over 2 Years (N=391)

	≥1 fall	≥1 hospitalization	≥1 ED visit	Low SRH*
Predictor		OR (95% CI)
SPPB	1.14 (0.92, 1.41)	1.35 (1.09, 1.67)	1.28 (1.03, 1.58)	1.84 (1.33, 2.53)
LLFDI-function	1.31 (1.06, 1.62)	1.51 (1.21, 1.90)	1.18 (0.96, 1.46)	2.11 (1.47, 3.02)
Frailty phenotype†	1.12 (0.90, 1.39)	1.42 (1.14, 1.77)	1.18 (0.95, 1.46)	1.67 (1.24, 2.23)

SPPB and LLFDI-function reversed so higher scores mean worse function; all predictors converted to z-scores; all models adjusted for age and sex

^*n=276 for SPPB and LLFDI-function and n=253 for frailty phenotype

^†n=356 for falls, hospitalizations and ED visits and n=253 for SRH.

Table 3.

Areas Under the ROC Curves for Prediction of Adverse Outcomes over 2 Years (N=391)

	≥1 fall	≥1 hospitalization	≥1 ED visit	Low SRH*
Predictor		AUC (95% CI)
SPPB	0.55 (0.49, 0.60)	0.59 (0.54, 0.65)	0.59 (0.53, 0.64)	0.66 (0.59, 0.74)
LLFDI-function	0.58 (0.52, 0.63)	0.60 (0.54, 0.65)	0.56 (0.50, 0.62)	0.67 (0.59, 0.74)
Frailty phenotype†	0.53 (0.47, 0.59)	0.60 (0.54, 0.66)	0.54 (0.48, 0.60)	0.64 (0.56, 0.73)

^*n=276 for SPPB and LLFDI-function and n=253 for frailty phenotype

^†n=356 for falls, hospitalizations and ED visits and n=253 for SRH.

Additional logistic regression models computed for each domain of the SPPB, LLFDI-function and frailty phenotype allowed an exploration of the components that may be driving the associations between each measure and the adverse outcomes. Over 2 years, the SPPB balance score and the combined lower extremity function components of the LLFDI-function were the only predictors of all four adverse outcomes (ORs 1.23-1.57 and 1.02-1.06, respectively) and each domain was predictive of at least one adverse outcome (table 4). The results over 1 year were similar (supplementary table S3) except that none of the domains were predictive of falls, the balance domain was not predictive of ED visits, and strength and physical activity were not predictive of any of the outcomes. The results from the multiple imputation analyses were consistent with these findings (supplementary tables S4-S9).

Table 4.

Domains of the SPPB, LLFDI-function and Frailty Phenotype that Predict Adverse Outcomes over 2 Years (N=391)

	≥1 fall*	≥1 hospitalization*	≥1 ED visit*	Low SRH†
		OR (95% CI)
SPPB
Balance score	1.25 (1.03, 1.52)	1.34 (1.11, 1.62)	1.23 (1.01, 1.48)	1.57 (1.19, 2.06)
Walk score	1.18 (0.88, 1.59)	1.47 (1.10, 1.97)	1.15 (0.86, 1.53)	1.86 (1.25, 2.76)
Chair stand score	0.96 (0.81, 1.14)	1.08 (0.91, 1.28)	1.16 (0.98, 1.38)	1.29 (1.00, 1.65)§
LLFDI-function
Upper extremity	1.02 (1.00, 1.04)	1.01 (0.99, 1.02)	1.01 (0.99, 1.02)	1.03 (1.00, 1.05)§
Basic LE	1.02 (1.00, 1.04)	1.02 (1.00, 1.04)	1.01 (0.99, 1.02)	1.04 (1.01, 1.07)
Advanced LE	1.01 (1.00, 1.03)‡	1.03 (1.02, 1.05)	1.02 (1.00, 1.03)	1.06 (1.04, 1.09)
Frailty Phenotype
BMI	0.83 (0.33, 2.09)	0.24 (0.07, 0.86)	0.57 (0.23, 1.44)	0.94 (0.26, 3.47)
Tiredness	1.24 (0.78, 1.96)	1.63 (1.04, 2.58)	1.07 (0.68, 1.69)	2.46 (1.32, 4.59)
Strength	1.30 (0.78, 2.19)	1.92 (1.15, 3.20)	1.55 (0.92, 2.60)	1.27 (0.60, 2.69)
Physical activity	1.37 (0.81, 2.30)	1.22 (0.74, 2.03)	0.91 (0.55, 1.51)	2.12 (1.03, 4.35)§
Gait speed	1.03 (0.63, 1.70)	1.88 (1.14, 3.08)	1.42 (0.86, 2.35)	2.63 (1.34, 5.18)

For SPPB and LLFDI-function domains, higher scores are worse; frailty phenotype domains were dichotomized using frailty cut-offs for this sample; LE = lower extremity

^*n=390 for BMI, n=359 for strength, n=389 for physical activity

^†n=276 except n=275 for BMI, n=256 for strength, n=274 for physical activity

^‡not significant

^§overall model not significant (p=0.06-0.17).

DISCUSSION

In this primary care cohort of older adults, a measure of physical function (SPPB), and a measure of patient-reported function (LLFDI-function) performed at least as well as a measure of frailty (Fried’s frailty phenotype) in the prediction of falls, hospitalizations, ED visits and low SRH over 1 and 2 years. These findings are clinically relevant for those working in primary care, where screening is relevant and necessary, given that it is the first point of contact with the healthcare system. Our results suggest that, if screening for adverse health outcomes, a simple measure of either performance-based or patient-reported physical function may work as well as the frailty phenotype.

Few studies have compared measures of frailty to physical function in terms of predictive validity.^18,19 In contrast to our findings, in a sample of 192 community-dwelling older adults, the Tilburg Frailty Indicator better predicted falls over 1 year compared to two single physical function measures, the timed up and go and the one leg stance.¹⁹ Because the SPPB is comprised of walking, balance tasks, and a chair stand, it is possible that the cumulative effect of each of these components contributed to the predictive ability of the SPPB in the current analysis and that single task tests are less robust for this purpose. In addition, the measure of frailty used in our analysis, while one of the most commonly used,¹⁴ is not as comprehensive as the Tilburg Frailty Indicator which includes 15 items covering multiple domains. Although it is not surprising that predictive validity increases with increased information, use of a longer frailty measure such as this requires access to more information and more time for completion, limiting its utility as a screening tool.

In line with our findings, in a previous study, the LLFDI-function was found to predict postoperative complications better than frailty phenotype.¹⁸ The patient-reported LLFDI-function emerged as the only predictor for falls over 2-year follow-up (OR 1.31, 95%CI 1.06–1.62). It also predicted hospitalizations and low SRH over both 1- and 2-year follow-up and was a better predictor of low SRH over 1 year than the frailty phenotype (p = 0.025). The LLFDI-function does not require any equipment, is a relatively short questionnaire with only 32 items, and can be administered over the phone or self-administered by participants.³² This makes it an appealing measure, particularly when it performs as well as other well-established measures in the prediction of adverse outcomes.

An exploration of the domains of each measure that were significant predictors of the adverse outcomes provides some insight into the driving factors for the relationships between each measure and adverse outcome. The emergent predictors were similar for the 1- and 2-year outcomes and many of the component predictors are conceptually intuitive. For example, the SPPB balance score and LLFDI-function lower extremity domains were predictive of falls, low strength was predictive of hospitalizations and patient-reported tiredness was predictive of patient-reported SRH.

Study Limitations

This study had several limitations. The first limitation relates to missing data. A frailty score was not calculated for 34 participants due to missing data. However, the results from multiple imputation analysis were consistent with those from the complete case analyses. Missing outcomes data may have also affected the results as participants were classified has having had an adverse outcome or not regardless of the proportion of follow-up calls they completed. This may have resulted in false negatives (e.g., classifying a faller as a non-faller because they missed the call to report it). The second limitation relates to the operationalization of the frailty phenotype, particularly for the criteria of weight loss, exhaustion and weakness. We used BMI as a surrogate for weight loss and although a low BMI (≤21kg/m²) was considered positive for the criterion of weight loss in the frailty phenotype (therefore, contributing to an increased frailty score), on its own, BMI ≤21kg/m² was protective for hospitalizations over 2 years and ED visits over 1 year. Rather than BMI, body composition is likely a better marker of frailty with frail people having lower muscle mass and bone mass, and higher body fat and waist circumference than non-frail people.^33,34 Using a mobility-tiredness scale rather than a measure of overall exhaustion likely resulted in a greater number of participants being considered positive for this criterion, and leg press may not have the same relationship with overall strength that grip strength does.³⁵ Therefore, the predictive validity of this measure may not be generalizable to other methods of operationalization. In addition, the cut-off points for the frailty criteria of slow gait speed, weakness, inactivity and tiredness were determined by the lowest quintile of the study sample. Had we used literature-derived cut-offs, it is possible that more participants would have been classified as frail.³⁶ Further, this analysis was limited to the frailty phenotype and our results may not extend to other popular measures of frailty such as the frailty index.³⁷ And finally, the Boston RISE study was not a population-based study and was limited to primarily English-speaking primary-care patients from a single health care system in the Northeastern United States and may not be generalizable to other ethnically diverse populations.

CONCLUSIONS

In summary, in this study, the SPPB, LLFDI-function and frailty phenotype had similar accuracy for predicting falls, hospitalizations, ED visits and low SRH over 1 and 2 years among older primary care patients at risk for disability. As a result, when considering the optimal screening tool for older adults, the choice between a measure of function and frailty may ultimately depend on clinical preference, construct of interest, and on contextual factors. For example, in primary care and rehabilitation, a measure of physical function may be the preferred method as it is quick to assess, is easily interpreted and guides treatment toward specific impairments. Alternatively, when exploring risk of adverse events in a large dataset when many variables are readily accessible, a measure of frailty may provide a more comprehensive assessment of risk. Our results should be confirmed through a prospective longitudinal study that includes multiple measures of frailty, including more comprehensive measures such as the Frailty Index, and an objective measurement of adverse outcomes including time to event data. Future research could also further investigate the domains of each measure driving the predictions, potentially resulting in an even more simplified battery of tests to identify those at risk.

Supplementary Material

Supplementary Tables S1-S9

Supplementary Table S1. Physical Function, Patient-reported Function and Frailty as Predictors of Adverse Outcomes over 1 Year (N=391)

Supplementary Table S2. Areas Under the Receiver Operating Characteristic Curves for Prediction of Adverse Outcomes over 1 Year (N=391)

Supplementary Table S3. Domains of the SPPB, LLFDI-function and Frailty Phenotype that Predict Adverse Outcomes over 1 Year (N=391)

Supplementary Table S4. Results from Multiple Imputation for Physical Function, Patient-reported Function and Frailty as Predictors of Adverse Outcomes over 1 Year (N=391)

Supplementary Table S5. Results from Multiple Imputation for Areas Under the Receiver Operating Characteristic Curves for Prediction of Adverse Outcomes over 1 Year (N=391)

Supplementary Table S6. Results from Multiple Imputation for Domains of the SPPB, LLFDI-function and Frailty Phenotype that Predict Adverse Outcomes over 1 Year (N=391)

Supplementary Table S7. Results from Multiple Imputation for Physical Function, Patient reported Function and Frailty as Predictors of Adverse Outcomes over 2 Years (N=391)

Supplementary Table S8. Results from Multiple Imputation for Areas Under the Receiver Operating Characteristic Curves for Prediction of Adverse Outcomes over 2 Years (N=391)

Supplementary Table S9. Results from Multiple Imputation for Domains of the SPPB, LLFDI-function and Frailty Phenotype that Predict Adverse Outcomes over 2 Years (N=391)

Highlights.

1. Function and frailty both predict adverse outcomes among older adults.

2. The choice of screening measure may depend on clinical context.

3. Results need to be confirmed prospectively using multiple frailty measures.

LIST OF ABBREVIATIONS

AUC

area under the curve

BMI

body mass index

CI

confidence interval

ED

emergency department

LLFDI-function

late life function and disability instrument- function component

OR

odds ratio

PASE

physical activity scale for the elderly

RISE

rehabilitative impairment study of the elderly

ROC

receiver operating characteristic

SPPB

short physical performance battery

SRH

self-rated health