Designing Clinical Trials of Interventions for Mobility Disability: Results from the Lifestyle Interventions and Independence for Elders Pilot (LIFE-P) Trial

Abstract

Background

Clinical trials to assess interventions for mobility disability are critically needed; however, data for efficiently designing such trials are lacking.

Methods

Results are described from a pilot clinical trial in which 424 volunteers aged 70–89 years were randomly assigned to one of two interventions -- physical activity or a healthy aging education program -- and followed for a planned minimum of 12 months. We evaluated the longitudinal distributions of four standardized outcomes to contrast how they may serve as primary outcomes of future clinical trials: ability to walk 400 meters, ability to walk 4 meters in ≤10 seconds, a physical performance battery, and a questionnaire focused on physical function.

Results

Changes in all four outcomes were inter-related over time. The ability to walk 400 meters as a dichotomous outcome provided the smallest sample size projections (i.e. appeared to be the most efficient outcome). It loaded most heavily on the underlying latent variable in structural equation modeling with a weight of 80%. A four-year trial based on the outcome of 400 meter walk is projected to require N = 962 to 2,234 to detect an intervention effect of 30% to 20% with 90% power.

Conclusions

Future clinical trials of interventions designed to influence mobility disability may have greater efficiency if they adopt the ability to complete a 400 meter walk as their primary outcome.

INTRODUCTION

As the life expectancy of older Americans increases, prevention of age-associated physical function decline and disabilities has emerged as a major clinical and public health priority.¹ A critical factor in an older person’s ability to function independently is mobility, the ability to move without assistance.² Older people who lose mobility are less likely to remain in the community, have higher rates of morbidity and mortality, have more hospitalizations, and experience a poorer quality of life.³

Clinical trials are necessary to establish interventions for improving mobility. For major investments to be made, it is important that these trials are designed efficiently, which includes informed decisions on which outcomes have the best measurement characteristics, what effect sizes should be targeted, and how many participants are required. We describe how data collected by the Lifestyle Interventions and Independence for Elders Pilot (LIFE-P) trial have been used to address these design issues.

METHODS

The study design, eligibility criteria, and recruitment procedures of LIFE-P were described previously.⁴ Participants were aged 70–89 years and able to complete a 400 meter walk in 15 minutes. Major exclusion criteria included presence of severe heart failure, uncontrolled angina, and other severe illnesses that might interfere with physical activity. All participants completed a 1-week behavioral run-in prior to random assignment, with equal probability, to either a physical activity intervention or a health education control. Written informed consent was obtained; the NIH and Institutional Review Boards for all participating institutions approved the protocol and consent forms. Between May, 2004 and February, 2005, 424 participants were enrolled. At baseline, 6- and 12-months, comprehensive standard assessments were conducted by trained research staff who were masked to intervention assignment. The ability to walk 400 meters (W400) was assessed using a standard walking course. Participants were permitted to stop during the walk, but not allowed to sit or receive help from others (cane use was permitted during follow-up assessments) and were required to complete the course in 15 minutes. Central adjudication was used to classify individuals who did not complete the walk.⁴ The ability to walk 4 meters in <10 seconds (W4), corresponding to a gait speed ≥0.4 m/sec, was assessed by asking the participants to walk at their usual pace.⁶ The Short Physical Performance Battery (SPPB) is a brief performance-based test that includes the above 4 meter walk, repeated chair stands, and a balance test.^7,8 Its three components are each scored 0 to 4, with 4 indicating the highest level of performance, and were summed to yield an overall score. To be eligible for LIFE-P, participants must have had SPPB ≤ 9 at baseline. The Self-Reported Disability Scale (SRDS) has been used in clinical trials of physical activity and observational studies^9–11 and is based on a 22-item questionnaire that assesses perceived difficulties in activities of daily living during the past month.¹² Respondents provide ordered numeric responses of no difficulty (1), a little difficulty (2), some difficulty (3), a lot of difficulty (4), or unable to do (5), which were averaged across the 22 items for an overall score. The Quality of Well-Being Scale (QWBS), which has a range of possible scores from 0 (worse) to 1 (best), was used to assess health-related quality of life.⁵

Statistical methods

For the dichotomous measures (W400, W4), we described rates and transition probabilities over time using log-linear models.¹³ For the measures analyzed as continuous data (SPPB, SRDS), we examined mean changes over 6 month epochs. To project the required sample sizes for continuous measures, we simulated data that had the same variances and longitudinal correlations observed in LIFE-P, but longer (i.e. 4 years) follow-up. For the control group, the mean changes over 6 month intervals time were set to equal those observed in LIFE-P, with those from 6–12 months repeated to extend throughout 4 years. To introduce 20–30% intervention effects, we increased any beneficial mean changes and decreased any non-beneficial mean changes by these amounts to simulate data for the intervention group. In a parallel manner, we projected sample sizes for categorical measures by simulating data that had transition rates based on LIFE-P data. For the control group, the 6-month transition rates were set to equal those observed in LIFE-P, with those from 6–12 months repeated to extend throughout 4 years. To introduce 20–30% intervention effects, we increased the rates of beneficial transitions and decreased the rates of non-beneficial transitions by these amounts to simulate data for the intervention group. Missing data were randomly introduced at accumulating rates of 4% per 6 months (corresponding to what LIFE-P observed) in both groups. We used changes from baseline over follow-up (in repeated measures models) as the outcome for continuous data. For dichotomous measures, we examined two potential outcomes: times until the first failure during follow-up and the occurrence of two successive failures. Data were simulated (100,000 sequences) and analyzed with general linear models or survival analyses, depending on the outcome, to project power.

We fitted a structural equation to examine how the four outcomes tracked over time against an underlying construct or “commonality” and fitted this model using Gibbs sampling^14,15 (see appendix). Standardized weights estimated from this model express how strongly changes in each outcome were related to changes in the underlying commonality.

We examined whether changes in outcomes were related to changes in health related quality of life, a factor expected to be associated with mobility disability. To do this, we examined the correlations that 1-year changes in QWBS had with 1-year changes in SPPB and SDRS, after covariate-adjustment for intervention assignment. We also examined mean changes in QWBS for participants who did and did not complete the 400 meter and 4 meter walks during follow-up.

RESULTS

Table 1 provides a description of the 424 LIFE participants at enrollment. The mean (standard deviation) SPPB and SRDS scores were 7.52 (1.41) and 1.36 (0.36), respectively. All participants successfully completed the 400 meter walk as part of eligibility criteria, with average walk speeds of 0.85 m/sec (0.18 m/sec); all but 9 (2.1%) completed the 4 meter walk in ≤10 seconds, with average walk speeds of 0.74 m/sec (0.16 m/sec). Mean QWBS score was 0.64 (0.10). Data collection rates at 6 months were 94.8% (400 meter walk), 93.9% (4 meter walk), 94.6% (SPPB), and 96.2% (SRDS). At 12 months these were 92.2% (400 meter walk), 93.9% (4 meter walk), 93.9% (SPPB), and 94.6% (SRDS). One participant in each intervention condition died during each 6-month epoch (i.e. 4 total deaths).

Table 1.

Characteristics of the LIFE cohort at enrollment

Characteristic	Total (N = 424)
Age, yrs
70–74	147 (34.7)
75–79	162 (38.2)
80–89	115 (27.1)
Sex
Female	292 (68.9)
Male	132 (31.1)
Short Performance Physical Battery
≤6	97 (22.9%)
7	80 (18.8%)
8	117 (27.6%)
9	130 (30.7%)
Mean (SD)	7.52 (1.42)
Self-Report Disability Score
1–1.9	392 (92.7)
≥ 2	31 (7.3%)
Mean (SD)	1.36 (0.36)
400 Meter Walk
Success	424 (100%)
Failure	0 (0%)
4 Meter Walk in ≤10 Seconds
Success	415 (97.9%)
Failure	9 (2.1%)
Intervention Assignment
Successful aging	211 (49.8%)
Physical activity	213 (50.2%)

Table 2 describes estimated rates (standard errors) at which participants transitioned between success and failure for completing the 400 meter walk and 4 meter walk from baseline to 6 months and from 6 months to 12 months. For example, at 6 months, 6.9% (1.9%) of the successful aging participants were unable to complete the 400 meter walk. Of these who were later evaluated, 33.3% (13.4%) reverted to success at 12 months. Of those who were successful at 6 months, 7.9% (2.1%) were unsuccessful at 12 months.

Table 2.

Transition rates and standard errors for 400 meter walk and 4 meter walk outcomes: estimates from log-linear models applied to participants enrolled in the successful aging and physical activity interventions^*

Successful Aging
Outcome and Transition	Transition Rates (SE)
	Between 0 and 6 Months	Between 6 and 12 Months
400 meter walk
Success to failure	6.9% (1.9%)	7.9 (2.1%)
Failure to success	--	33.3% (13.4%)
4 meter walk
Success to failure	4.7% (1.5%)	2.3% (1.2%)
Failure to success	--	66.7% (16.0%)
Physical Activity Intervention
Outcome and Transition	Transition Rates (SE)
	Between 0 and 6 Months	Between 6 and 12 Months
400 meter walk
Success to failure	7.8% (1.9%)	3.6% (1.5%)
Failure to success	--	50.0% (17.4%)
4 meter walk
Success to failure	2.6% (1.1%)	3.3% (1.3%)
Failure to success	--	40.0% (22.0%)

^*During the first year of follow-up, 20 of the 213 participants assigned to the physical activity intervention and 24 of the 211 assigned to the successful aging intervention were unable to complete the 400 meter walk at the 6- and/or 12-month assessment. Of participants who were successful at completing a 4 meter walk at baseline, 11 of the 206 who were assigned to the physical activity intervention and 13 of the 202 assigned to successful aging were unable to complete the 4 meter walk at the 6- and/or 12-month examination.

Table 3 describes mean (standard error) 6-month changes in the SPPB and SRDS scores among successful aging participants. Scores on the SPPB improved by an average of 0.589 (0.139) units during the first 6 months and worsened slightly by 0.168 (0.115) units from 6 to 12 months. Changes in SRDS scores were less marked and small relative to their standard error. Also in Tables 2 and 3 are estimates for participants randomized to the physical activity intervention.

Table 3.

Estimated mean changes, standard errors, and covariance structure for the short physical performance battery (SPPB) and self-reported disability scale (SRDS) for participants assigned to the two intervention programs

Successful Aging
Measure	Mean (SE) Change		Variance and Correlation Estimates
	0–6 Months N = 203	6–12 Months N = 195	0–6 Mo Change σ12	6–12 Mo Change σ22	Correlation of Changes ρ12
SPPB	0.589 (0.139)	−0.168 (0.115)	3.885	2.509	−0.27
SRDS	0.021 (0.022)	0.017 (0.022)	0.102	0.097	−0.42
Physical Activity Intervention
Measure	Mean (SE) Change		Variance and Correlation Estimates
	0–6 Months N = 204	6–12 Months N = 177	0–6 Mo Change σ12	6–12 Mo Change σ22	Correlation of Changes ρ12
SPPB	1.241 (0.126)	−0.258 (0.115)	3.154	2.576	−0.34
SRDS	−0.027 (0.024)	0.024 (0.024)	0.119	0.112	−0.28

Table 4 lists the numbers of participants required to detect intervention effects of 20%, 25%, and 30%, respectively, for each outcomes. We project that 1414 participants would be required for 90% statistical power to detect a 25% effect size for the outcome of time until the first failure of the 400 meter walk; 2839 participants would be required for 90% power to detect this effect if the outcome is time until two successive failures. The projected sample sizes for the other outcomes were uniformly larger.

Table 4.

Total number of participants enrolled in trial (assigned in equal numbers to the physical activity and successful aging interventions) required to provide 80% and 90% power to detect intervention effects of 20%, 25%, and 30% over 4 years of follow-up.^*

Outcome	80% Power			90% Power
	Intervention Effect			Intervention Effect
	20%	25%	30%	20%	25%	30%
400 meter walk
Time until first failure	1669	1056	719	2234	1414	962
Two successive failures	3453	2120	1403	4624	2839	1678
4 meter walk
Time until first failure	5178	3134	2085	6934	4196	2792
Two successive failures	16881	9650	6597	22603	12921	8833
SPPB
Mean difference over follow-up	4673	2991	2077	6257	4105	2781
SRDS
Mean difference over follow-up	14010	8967	6233	18760	12006	8338

^*Projections are based on simulations incorporating means and transition rates from Tables 2 and 3 for the successful aging group. As an examples, for a 20% intervention effect the 400 meter transition rates for the intervention group were calculated as 6.9% × 0.8 = 5.5%, 7.9% × 0.8 = 6.3%, and 33.3% × 1.2 = 40.0% and the mean changes in the SPPB for the intervention group were calculated as 0.589 × 1.2 = 0.707 and −0.168 × 0.8 = −0.134.

The same (percent) effects sizes are used for each of the outcome measures in Table 4. It is likely, however, that these measures vary in their fidelity to mobility disability. Some may more closely represent its changes than others, so similar effect sizes may correspond to different changes in mobility disability. Table 5 provides results from an analysis aimed at describing how changes in the outcomes portray underlying changes. The fitted weights indicate the relative strength of the relationship between changes in each outcome and changes in the underlying commonality (reported in standard deviation units). Weights from each outcome are bounded away from zero (changes in measures are interrelated) and relationships are in the expected directions. The 400 meter walk expressed changes in the underlying construct most directly: an intervention producing a one standard deviation change in this outcome was estimated to yield a 0.80 standard deviation change in the commonality (e.g. “mobility disability”). The SPPB, which is negatively correlated with the other outcomes due to how its scoring was ordered, was the second most direct outcome, with a 0.32 standard deviation change. The dichotomous 4 meter walk outcome was projected to receive a 0.22 unit change. The SRDS was relatively inefficient, receiving only a 0.13 unit change in the underlying construct.

Table 5.

Estimated proportion of change in one standard deviation of the commonality estimated from a structural equation model for the distinct outcomes of mobility disability.

Outcome	Variable (See Appendix)	Proportion Conveyed to Underlying Construct*	95% Credible Interval
400 meter walk	w1	−0.80	[−1.11, −0.56]
4 meter walk	m1	−0.22	[−0.32, −0.18]
SPPB	s1	0.32	[0.29, 0.34]
SRDS	d1	−0.13	[−0.15, −0.11]

^*Signs denote direction of the relationship. Results of 400 meter walk and 4 meter walk were coded as 1 = failure and 0 = success. Thus, higher scores for the walks and SRDS and lower scores for SPPB were associated with worse performance.

QWBS scores showed little overall change from baseline to 1-year among participants assigned to both the physical activity and successful aging interventions. Mean (standard deviation) changes were 0.01 (0.10) and 0.01 (0.09), respectively. After covariate adjustment for intervention assignment, the correlations between 12-month changes in QWBS and 12-month changes in SPPB and SRDS were r = 0.13 (p = 0.01) and r = −0.31 (p < 0.001), respectively. Mean QWBS scores (standard error) worsened over 12 months among participants who failed the 400 meter walk either at the 6 and/or 12 month visit, but improved among who did not: −0.020 (0.012) versus 0.011 (0.005), p = 0.02. Failure of the 4 meter walk either at the 6 and/or 12 month visit was not associated with 1-year changes in QWBS scores (p = 0.67).

DISCUSSION

Trials conducted in older individuals offer many challenges, including the potential of special barriers to recruitment and adherence, greater safety concerns, higher rates of missing data, competing risks, and informative drop-out.^16–18 LIFE-P was intended to provide critical data and experience for the development of a full scale randomized trial to assess whether a physical activity intervention may prevent decline in physical functioning in older persons.⁴ Among its goals was to obtain information to inform the selection of the primary outcome for the full scale trial. Based on the results of the current manuscript, the LIFE-P investigators adopted the 400 meter walk as the primary outcome in the full scale trial.

Mobility Disability Outcomes

We evaluated four distinct mobility-related outcomes. Three were objectively measured (400 meter walk, 4 meter walk, and SPPB), while the fourth (SRDS) was based on self-report. The 400 meter walk and 4 meter walk assess mobility directly, with inability to complete the latter task representing the more severe form of mobility disability.¹⁹ While not considered pure measures of mobility, the SPPB includes three elements (gait, transfers, and balance) that serve as the building blocks for mobility and the SRDS includes several mobility items, including walking across a small room, walking one block, walking a quarter of a mile, walking several blocks, and climbing a flight of stairs. The SPPB and SRDS have served as primary outcome measures in previous intervention trials.^20,21 The physical activity intervention in LIFE-P led to relative improvements in SPPB scores that were statistically significant.²² Increasing evidence supports the reliability and validity of the 400 meter walk as a measure of major mobility disability.^21,23,24 In trials for which it is impossible to mask participants to intervention assignment, objective measures may be preferred to those based on subjective self-report.¹⁷

LIFE-P data suggest that outcomes may vary in their ability to express underlying changes in mobility disability. Outcomes that most clearly express these changes are preferable, and targeted effects should be of sufficient magnitude to produce meaningful changes in the underlying construct. While the structural equation model we fitted indicated that the 400 meter walk is most strongly inter-correlated with the other measures under consideration, the model certainly falls short of capturing the full complexity of interrelationships and our underlying construct may not correspond directly to mobility disability.

Sample Size Requirements

Sample sizes targets are based on assumptions regarding the magnitude of the intervention effect and the distribution of outcomes. A pilot trial does not have sufficient size to estimate accurately an intervention effect; this is the purpose of a full scale trial. Although results presented in tables 2 and 3 by intervention assignment are encouraging that the physical activity intervention may be beneficial, more definitive assessment is required. LIFE-P provided transition rates for categorical outcomes and means and covariances of changes for continuous outcomes, which we have used in our calculations.

If an intervention produces effects of 20–30% in each of the four measures, our projections indicate that the 400 meter walk is the most efficient primary outcome. Fewer than half as many participants would be required as for the 4 meter walk or SPPB, and considerably fewer than for the SRDS. In general, categorizing continuous outcomes reduces power; however, a categorical outcome may be a more efficient expression of an underlying continuous commonality than continuous outcomes that are less directly related.

Among participants who failed the 400 meter walk at 6 months, those in the physical activity group were more likely to complete the test at 12 months successfully. Nonetheless, our results indicate that nearly 50% more participants would be required for a primary outcome based on two successful failures to complete the 400 meter walk. Thus, although the outcome of persistent disability may be important clinically, it would likely require much larger clinical trials.

Associations with Health Related Quality of Life

We found that 1-year changes in SPPB, SRDS, and 400 meter walk were associated with relatively better changes in a commonly used index of health related quality of life. No such association was found for changes in 4 meter walk; our power to detect such an association, however, may have been low. To our knowledge, these four mobility-related outcomes have not been previously evaluated in a single study. Nonetheless, each has been previously linked to poorer quality of life and subsequent adverse outcomes, including loss of independence, institutionalization, and mortality.^25–32

Limitations

The one year experience from LIFE-P may not allow us to accurately project what would occur over the course of a longer trial; it is possible, for example, that the covariance, transitions, and trajectories in later years will differ from those in the pilot phase. Participants in LIFE-P, while recruited across four centers in a manner to enhance its diversity, may not broadly represent physically impaired older persons. As for any trial, the generalizability of our results may be limited by inclusion criteria and the characteristics of participants. In particular, LIFE-P participants were required to complete a 400 m walk at baseline, hence our results may not extend to persons who have more limited mobility. Our simulations incorporated the same rates of missing data for all four outcomes. While high rates of data collection were observed for each outcome during the LIFE pilot trial, the SRDS was successfully collected on 1–2% more participants than the other outcomes at both time points. We have not factored in either cost or participant burden considerations into our recommendation on outcomes; some advantages may accrue according to these considerations for questionnaire-based outcomes.

We have not explicitly addressed differences among clinical sites in the distributions of outcomes. These depend, in part, on how sites are selected and monitored. If not accounted for in analyses, variance may be inflated and generalizability may be limited.³³ The LIFE-P analysis plan anticipated this eventuality by including site as a covariate when evaluating the impact of interventions on outcomes.^4,22 In our results, these differences contribute to the overall variability we describe. If future trials anticipate different levels of intra-site correlations, some adjustment to our projected sample size may be required. There is also the potential that the effect sizes of interventions may vary among sites; if information on the likely variance of these differences is available, further adjustments may be made.

The LIFE-P intervention featured walking as the primary means to enhance physical activity, thus the inter-correlations among outcomes may reflect those that are most sensitive to walking.

Summary

Our analyses of data from LIFE-P suggest that its planned full-scale trial may have greater efficiency if, among the outcomes it considered, it adopts the ability to complete a 400 meter walk as its primary outcome. It is likely, however, that no universally optimum outcome exists for trials of mobility disability. Primary outcomes should not be selected solely on the basis of their inter-relationships with alternative outcomes and statistical efficiency, but also on the basis of their associations with other health markers, participant burden and safety, and their ability to influence clinical practice.^34,35 Thus other outcomes may prove to be more useful for trials in different populations and contexts and the relative advantages of competing outcomes will be better understood as experience accrues. To this end, it may be profitable to include more than one measure of mobility disability in future study designs.

Appendix

In our structural equation, we labeled the underlying construct, X. X expresses the greatest “commonality,” i.e. underlying inter-correlation of changes in outcomes. Loosely, we consider and interpret it as an expression of the degree of mobility disability. We assumed that each individual “i” ages along a trajectory, which over a relatively short range of one year could be described by a linear model:

$${\text{X}}_{\text{ik}}({\text{t}}_{\text{k}})={\text{a}}_{\text{i}}+{\text{b}}_{\text{i}}{\text{t}}_{\text{k}}$$

in which a_i and b_i are the intercept and rate of decline for participant i, k denotes occasion, and t_k denotes time. We assumed that a_i and b_i were random effects that followed Gaussian distributions. If mobility disability, X, were observable, it would be the focus of clinical trials. Because X is not observable, we assumed that changes in the 400 meter walk (“unable” coded as W400 = 1 and “able” coded as W400 = 0), 4 meter walk (“unable” coded as W4 = 1 and “able” coded as W4 = 0), SPPB (labeled “S”) and SRDS (labeled “D”) each were driven in some part by changes in X. Critically, we viewed interventions as designed to have a beneficial effect on how X changes over time. How interventions do this is expressed indirectly through their effects on other outcomes. Thus,

$$log(\text{odds}[\text{W}400({\text{t}}_{\text{k}})]=1)={\text{w}}_0+{\text{w}}_1({\text{X}}_{\text{ik}}({\text{t}}_{\text{k}})),$$

$$log(\text{odds}[\text{W}4({\text{t}}_{\text{k}})]=1)={\text{m}}_0+{\text{m}}_1({\text{X}}_{\text{ik}}({\text{t}}_{\text{k}})),$$

$$\text{E}[{\text{S}}_{\text{ik}}({\text{t}}_{\text{k}})]={\text{s}}_0+{\text{s}}_1({\text{X}}_{\text{ik}}({\text{t}}_{\text{k}})),\text{ and}$$

$$\text{E}[{\text{D}}_{\text{ik}}({\text{t}}_{\text{k}})]={\text{d}}_0+{\text{d}}_1({\text{X}}_{\text{ik}}({\text{t}}_{\text{k}})),$$

where “E” refers to the expected value and intervention effects are modeled as influencing the mean of X_ik(t_k). The slopes (w_1, m₁, s₁, and d₁) in these relationships are reported in Table 4 in standard deviation units to express, using a common yardstick, the degree to which changes in individual outcomes are associated with changes in X.

Many statistical algorithms may be applied to fit this model. We used the Bayesian algorithm of Gibbs sampling,¹⁴,¹⁵ which provided flexibility for addressing missing data. This approach required us to define prior distributions for parameters, which we chose to be non-informative. Potential values for parameters were iteratively sampled and accumulated to produce a posterior distribution for each. We report the medians of these posterior distributions as point estimates and 95% equal-tail credible intervals, which are roughly analogous of 95% confidence intervals in frequentist approaches. A “burn-in” phase (during which the first 50,000 samples were discarded) was used to remove the influence of initial starting values. Analyses were re-run from a range of starting points; the congruence of these results and graphical inspection of the sequential samples allowed us to conclude that the estimation process was stationary and that 20,000 samples were sufficient to provide stable estimates.