Table 3.
Intervention effect on Lower Extremity Functional Scale (LEFS) scores*
| Adjustments | Intervention, no. participants | Usual care, no. participants | Difference in means | 95% CI | P |
|---|---|---|---|---|---|
| Primary analysis (intent‐to‐treat, no imputation) of LEFS scores at 12 months | |||||
| Model 1 | 87 | 86 | 4.47 | (0.20, 8.74) | 0.040 |
| Sensitivity analyses (intent‐to‐treat, no imputation) | |||||
| Model 2† | 87 | 86 | 4.44 | (0.18, 8.70) | 0.041 |
| Model 3‡ | 4.27 | (0.10, 8.44) | 0.045 | ||
| Model 4 | 3.95 | (–0.26, 8.17) | 0.066 | ||
| Analysis using MICE to account for missing data (intent‐to‐treat) | |||||
| Model 1 | 89 | 91 | 4.31 | (–0.18, 8.80) | 0.060 |
| Model 2† | 4.29 | (–0.19, 8.77) | 0.060 | ||
| Model 3‡ | 3.93 | (–0.45, 8.32) | 0.079 | ||
| Model 4 | 3.80 | (–0.49, 8.10) | 0.082 | ||
| Per‐protocol analyses | |||||
| Model 1 | 69 | 86 | 6.12 | (1.60, 10.64) | 0.008 |
| Model 2† | 6.12 | (1.60, 10.65) | 0.008 | ||
| Model 3‡ | 6.34 | (1.87, 10.82) | 0.005 | ||
| Model 4 | 5.54 | (1.10, 9.97) | 0.014 | ||
| Primary analysis (intent‐to‐treat, no imputation) of LEFS scores at 3 months | |||||
| Model 1 | 87 | 86 | 8.07 | (3.75, 12.40) | <0.001 |
| Primary analysis (intent‐to‐treat, no imputation) of LEFS scores at 6 months | |||||
| Model 1 | 87 | 86 | 5.41 | (1.06, 9.77) | 0.015 |
Modeling strategies for analysis of LEFS scores were as follows: model 1, linear mixed regression adjusted for stratification variables, accounting for clustering within patient; model 2, linear mixed regression adjusted for stratification variables, accounting for clustering within patient and surgeon; model 3, linear mixed regression adjusted for stratification variables and baseline imbalance variables (level of education, working status, and anxiety on the Hospital Anxiety and Depression Scale [HADS] anxiety subscale), accounting for clustering within patient and surgeon; model 4, linear mixed regression adjusted for stratification variables and whether the patient had received additional physical therapy during the trial, accounting for clustering within patient and surgeon. 95% CI = 95% confidence interval; MICE = multivariate imputation via chained equations.
The variance of the random effect associated with surgeon level was significant; this level was kept for the following sensitivity analyses.
Variables that were imbalanced at baseline: level of education, working status, preoperative HADS anxiety.