Adherence to early mobilization and acute functional recovery after lower-limb orthopedic surgery: A longitudinal cohort study

Study design

A longitudinal cohort study was conducted at the department of orthopedic trauma of a tertiary referral center between June 2025 and September 2025 (study duration: 4 months). The minimum sample size was calculated using the single-proportion formula: n = Z²_1-α/2 × p(1−p)/d², where a = 0.05, p = 0.76 (76% adherence prevalence per Rosas Hernández et al.,^[14]), and d = 0.05; this yielded a minimum of 281 patients.

Study population

Adult patients (≥18 years) scheduled for primary total or hemi-hip arthroplasty, total knee arthroplasty, or acute internal fixation of a lower-extremity fracture were eligible. Patients were required to be fluent in the local language and possess the cognitive capacity to comprehend the protocol and provide informed consent. Exclusion criteria included life-threatening emergency surgery or polytrauma requiring immediate critical-care management, immediate severe surgical failure or catastrophic complications precluding rehabilitation, severe psychiatric disorders, the American Society of Anesthesiologists (ASA) physical status IV–V, absolute contraindication to early mobilization, or severe cognitive impairment.

Data collection protocol

Each participant was followed from the pre-operative period through hospital discharge (mean stay 6.5 days), assessing the immediate short-term functional trajectory. Data were collected at three points by trained research personnel. Baseline assessment (T1, within 24 h of surgery) recorded sociodemographic information, medical and surgical history, ASA classification, and baseline functional status through structured interview and medical record review. Post-operative day 5 (T2) and discharge day (T3) involved standardized re-assessment of functional status and adherence. Surgical and post-operative data were extracted from the institution’s electronic medical records.

Variables and measurement instruments

The primary outcome was a change in post-operative functional status, quantified using the WOMAC. The WOMAC is a validated 24-item instrument assessing pain, stiffness, and physical function; raw scores were normalized to a 0–100 scale, with higher scores indicating worse symptoms. The primary outcome variable was the absolute improvement in the normalized total WOMAC score from T1 to T3; thus, a positive change reflects clinical improvement, and a positive regression coefficient indicates a greater symptom reduction.

The primary independent variable was adherence to the prescribed mobilization protocol, measured using the self-report EARS (6 items; score range: 0–24; higher scores indicate greater adherence). Patients were categorized into three groups using distribution-derived cutoffs of the discharge EARS score (low ≤15; medium 16–18; high ≥19), yielding approximately equal-sized strata (n = 102, 89, 90, respectively). These cutoffs were selected to approximate equal thirds of the distribution, given the absence of validated clinical EARS thresholds for the perioperative setting.

Confounding variables comprised demographics (age, sex, body mass index [BMI], education), clinical characteristics (comorbidities, ASA class, surgery type, anesthesia), and pre-operative laboratory parameters (albumin, total protein, neutrophil and lymphocyte percentages, platelet count, creatinine).

Statistical analysis

Statistical analyses were performed using R (v4.3.3). Continuous data are reported as mean standard deviation (SD), and categorical data as frequency (%). For the primary outcome (WOMAC change), we applied analysis of covariance (ANCOVA) with baseline WOMAC score as a covariate, the established approach for adjusting change scores in observational research with paired baseline–follow-up measurements.^[15,16] For predictors of the discharge EARS score, we used multivariable linear regression with patient characteristics as predictors; Day-5 EARS was not used as a baseline covariate because doing so would introduce autoregressive adjustment that absorbs the variance attributable to the predictors of interest.

Variables for both models were selected via univariable screening (p < 0.20) combined with clinical relevance, followed by backward stepwise elimination. To address potential multicollinearity, generalized variance inflation factors (GVIFs) were computed for all predictors; both serum albumin and total protein were retained in the final models because all GVIF values were below 1.32, well below the conventional threshold of 2.24 (the single-df equivalent of VIF = 5).

A pre-specified interaction term (adherence × surgery type) tested whether the effect of adherence differed between fracture fixation and arthroplasty subgroups. For this binary subgroup analysis, hip and knee replacements were grouped as arthroplasty (n = 53), and lower-limb fracture fixation was analyzed separately (n = 228). To assess robustness against sparse-data bias for the hypertension subgroup (n = 22; 7.8%), we computed bootstrap 95% CIs from 1000 resamples. To quantify potential bias due to unmeasured confounding, we calculated E-values as described by VanderWeele and Ding.^[17] We further conducted a sensitivity analysis excluding seven patients with biologically implausible BMI values (<14 or >40 kg/m²; 2.5%). A sensitivity power analysis (G*Power 3.1) indicated that n = 281 with k = 8 predictors at a = 0.05 provided ≥80% power to detect a Cohen’s f² ≥ 0.057 (small-to-medium standardized effect). Following Hoenig and Heisey,^[18] we do not interpret observed power as independent validation of sample adequacy. Coefficients are reported as b with 95% confidence intervals (CIs) and p-values; statistical significance was set at two-sided p < 0.05. Supplementary analyses and supporting materials are provided in the online supplementary material [Supplementary material].

Cohort characteristics

A total of 281 patients were enrolled [Table 1]. The mean age was 55.7 years (SD 16.3); 158 (56.2%) were female. Most of the patients (94.3%) were classified as ASA I, and internal fixation for lower-extremity fracture was the predominant procedure (228, 81.1%). The mean operative duration was 39 (SD 6) min, reflecting an institutional case-mix of simple closed lower-limb fractures amenable to intramedullary nailing or percutaneous plating.

Adherence and functional outcomes

Mean adherence (EARS) at discharge was 16.78 (SD 3.26). Distribution-based cutoffs categorized 102 patients (36.3%) as low adherence, 89 (31.7%) as medium, and 90 (32.0%) as High. The mean pre-operative baseline WOMAC score (normalized 0–100) was 22.2 (SD 4.3); the mean discharge score was 0.2 (SD 0.6), with 93.2% of patients scoring zero at discharge. This pronounced floor effect is addressed in the discussion. Across all participants, mean WOMAC improvement was 22.1 points (SD 4.3) on the normalized 0–100 scale, with the largest change in the pain subscale (25.4 points, SD 6.0). The longitudinal trajectory of total WOMAC scores by discharge adherence group is shown in Figure 1. Exploratory WOMAC trajectories stratified by adherence-change status from post-operative day 5 to discharge are shown in Figure 2.

Close

Figure 1: Longitudinal trajectory of total WOMAC scores by discharge adherence group from preoperative assessment to post-operative day 5 and discharge. Points indicate mean values and error bars indicate 95% confidence intervals. Lower WOMAC scores indicate better function. WOMAC: Western Ontario and McMaster Universities osteoarthritis index.

Export to PPT

Close

Figure 2: WOMAC trajectory stratified by adherence-change status between post-operative day 5 and discharge. Improved indicates an increase in adherence category, Maintained indicates no category change, and Declined indicates a decrease. Points indicate mean values and error bars indicate 95% confidence intervals. Lower WOMAC scores indicate better function. WOMAC: Western Ontario and McMaster Universities osteoarthritis index.

Export to PPT

Independent predictors of adherence

Table 2 presents the multivariable linear regression model for the discharge EARS score. After adjustment, hypertension was the strongest positive predictor of adherence (β = 3.71, 95% CI 2.35–5.06; p < 0.0001). To evaluate stability against sparse-data bias given the small hypertension subgroup (n = 22), we computed a bootstrap 95% CI (1000 resamples), which remained away from the null (β = 3.71, bootstrap 95% CI 2.27–5.57). Higher educational attainment was independently associated with better adherence for selected categories (Junior High School vs No Formal Education: b = 4.3, 95% CI 1.4–7.2; p = 0.022). Higher pre-operative neutrophil percentage was associated with higher adherence (β = 0.10, 95% CI 0.05–0.14; p < 0.0001).

Association of adherence with functional trajectory

Table 3 presents the ANCOVA model for WOMAC improvement, with baseline WOMAC severity as a covariate. After adjustment, the Medium adherence group showed a small additional improvement compared with Low adherence (β = 0.19, 95% CI 0.02–0.36; p = 0.029); the High adherence group showed a similar magnitude with marginal statistical evidence (β = 0.20, 95% CI −0.02–0.42; p = 0.073). The near-equivalent estimates between medium and high groups are compatible with a threshold-like pattern rather than a strict dose–response relationship. However, interpretation is constrained by the severe floor effect at discharge.

The ANCOVA model adjusts the change score (T1 − T3) for the baseline WOMAC score (T1), per Vickers and Altman^[15] and Senn.^[16] Model R² = 0.988 reflects substantial floor effect in discharge WOMAC scores (93.2% scoring zero); absolute effect sizes are mathematically constrained by this distributional feature, addressed in the discussion.

Effect modification by surgical context

A pre-specified interaction analysis (adherence × surgery type) was statistically significant (p = 0.014; ANOVA). Stratified estimates suggested that the adherence association was driven mainly by the fracture fixation subgroup (n = 228; medium vs. low b = +0.31; high vs. low b = +0.33), whereas the arthroplasty subgroup (n = 53, including hip and knee replacement) showed minimal additional variance attributable to adherence after baseline adjustment (medium b = −0.08; high b = +0.03). The corresponding adjusted point estimates stratified by surgical context are shown in Figure 3. A bootstrap analysis restricted to arthroplasty patients showed a high versus low difference of b = 3.68, 95% CI 1.35–6.45 on the unadjusted scale, indicating directional consistency but attenuation after baseline adjustment.

Close

Figure 3: Adjusted point estimates for the association between discharge adherence group and WOMAC improvement, stratified by surgical context. Estimates are from ANCOVA models adjusted for baseline WOMAC score and covariates. Low adherence was the reference group. Positive β values indicate greater WOMAC improvement. The adherence-by-surgical-context interaction p-value was 0.014. WOMAC: Western Ontario and McMaster Universities osteoarthritis index.

Export to PPT

Sensitivity analyses

Excluding seven patients with implausible BMI values (n = 274) yielded virtually unchanged estimates for the primary association (medium vs. low: b = 2.29; high vs. low: b = 2.93 on unadjusted models). The E-value for the primary effect (medium adherence threshold and WOMAC improvement) was 1.25 (Cohen’s d = 0.044). Coupled with the floor effect at discharge, this E-value indicates that the estimate is sensitive to unmeasured confounding (notably post-operative pain control and physiotherapy attendance, which were not collected) and should be interpreted as hypothesis-generating.

Limitations

This study has several limitations. First, the observational design precludes causal inference; despite ANCOVA adjustment, residual confounding from unmeasured variables, particularly post-operative pain intensity, physiotherapy attendance frequency, analgesia regimen, and provider-level mobilization practices, may persist. The E-value of 1.25 indicates that a moderate-strength unmeasured confounder could plausibly explain the observed association, and this caveat should temper interpretation. Second, adherence was measured through self-report, which is subject to recall and social desirability biases; future studies should incorporate accelerometry or wearable sensor data. Third, a substantial floor effect was observed in the discharge WOMAC: 93.2% of patients scored zero at discharge, and 67.7% scored zero by Day 5, indicating that the WOMAC instrument lacks discriminative sensitivity for acute postoperative functional trajectories in fracture cohorts. Patients on protected weight-bearing protocols or using ambulatory aids may report “no difficulty” on items such as walking on flat surfaces because these activities are not attempted, rather than because function is restored. The high R² values observed in the ANCOVA model are attributable to this floor effect rather than to genuine predictive power, and absolute effect sizes are mathematically constrained. Fourth, the brief mean operative duration (39 min) reflects an institutional case mix predominantly comprising simple closed lower-limb fractures; under Vietnam’s tertiary referral system, complex polytrauma is typically routed to specialized trauma centers, limiting generalizability to facilities that manage the full orthopedic trauma spectrum. Fifth, the cohort is predominantly ASA I (94.3%), with low BMI and minimal comorbidity, which further limits external validity to standard Western or Asian arthroplasty populations, who frequently present with ASA II–III status. Finally, regression estimates for sparse subgroups (ASA II/III, hypertension n = 22) are inherently unstable; the bootstrap analysis for hypertension and the directionally consistent arthroplasty subset analysis partially address this concern, but caution is warranted.

Strengths

Strengths include the prospective design with 3 time points, use of validated instruments for both functional outcomes (WOMAC) and adherence behavior (EARS), pre-specified interaction analysis, multiple sensitivity analyses (BMI exclusion, bootstrap, E-value), and transparent acknowledgment of measurement limitations.

Recommendations:

1. Routine measurement of adherence using a validated instrument (e.g., EARS) may be considered at Day 5 and discharged as a structured post-operative mobility-adherence measure.

2. Pre-operative risk stratification should integrate educational level, serum albumin, and total protein to identify patients at risk for low adherence and constrained recovery.

3. Targeted pre-operative interventions for at-risk patients, including structured patient education and nutritional optimization, warrant evaluation in prospective trials.

4. Future research in acute post-operative orthopedic settings should consider alternative patient-reported outcome measures (e.g., Lower Extremity Functional Scale, Timed Up and Go) better suited to detecting variation in early functional recovery, given the demonstrated WOMAC floor effect.