Association of acupuncture timing and treatment intensity with short-term functional recovery in stroke rehabilitation inpatients: a retrospective cohort and prediction study

Background In inpatient stroke rehabilitation settings where acupuncture is part of routine care, several practical questions arise: when treatment should begin, how intensively it should be delivered, and whether data on treatment patterns can help guide individualized discharge planning. We examined treatment-pattern associations in a retrospective acupuncture-treated cohort; lesion-side analyses were treated as secondary exploratory heterogeneity analyses. Methods We analyzed 1,048 stroke rehabilitation inpatients who received acupuncture during hospitalization. Median onset-to-first acupuncture was 26 days (IQR 20–35), median total exposure was 20 sessions (IQR 15–26), and median weekly frequency was 5.0 sessions/week (IQR 4.5–5.5). Primary outcomes were admission-to-discharge changes in Barthel Index (BI), Fugl-Meyer Assessment upper extremity (FMA-UE), lower extremity (FMA-LE), and balance; composite recovery was the mean standardized z-score across these four primary change domains. Secondary/exploratory outcomes were Montreal Cognitive Assessment (MoCA) and FMA-ROM/Pain. Home discharge was analyzed as a pragmatic discharge-planning endpoint. Analyses included regression models, subgroup interaction tests, total-session, extended-adjustment, and missingness-oriented sensitivity analyses, inverse-probability weighting, and 3-fold internally validated machine-learning benchmarks. Results Patients improved in BI (13.74 +/− 6.57), FMA-UE (10.18 +/− 5.86), FMA-LE (4.90 +/− 3.08), and FMA-balance (2.85 +/− 2.22). Because these instruments have different score ranges, standardized effect sizes and recovery ratios were used to support cross-domain interpretation. Earlier acupuncture initiation was the most consistent favorable exposure signal; for BI gain, the timing coefficient was beta = −1.55 (95% CI −1.96 to −1.14; p < 0.001). Mean total sessions were higher among patients discharged home than among non-home patients (21.86 +/− 8.19 vs. 19.60 +/− 7.62; p < 0.0001), and total-session models were associated with BI gain (beta = 2.83, 95% CI 2.46–3.20; p < 0.001) and home discharge (OR 1.39, 95% CI 1.19–1.62; p < 0.001). Left-versus-right interactions were not significant; stroke-type interaction was evident only for timing with FMA-LE gain ( p = 0.012). In prediction benchmarks, LARE had AUROC 0.804 (95% bootstrap CI 0.778–0.829) for home discharge but was not statistically distinguishable from CatBoost or RandomForest; BI-gain regression was moderate. Conclusion Within this acupuncture-treated inpatient cohort, earlier initiation, weekly frequency, and total exposure were associated with short-term ADL and motor recovery. These findings are consistent with timing- and dose–response hypotheses from the broader rehabilitation and acupuncture literature, but prospective studies with non-acupuncture comparators and acute neurologic severity measures are needed before causal efficacy claims can be made.