A Predictive Enrichment Model for Auricular Acupoint Stimulation in Post-Stroke Shoulder-Hand Syndrome: Identifying Baseline Effect Modifiers

Background Auricular acupoint stimulation (AAS) improves upper limb function in poststroke shoulder–hand syndrome (SHS), but treatment response heterogeneity limits personalized application. We aimed to identify baseline effect modifiers that predict differential response to AAS versus comprehensive rehabilitation alone. Methods This retrospective analysis included 80 SHS patients (AAS = 40, control = 40), with FMA-UE assessed at baseline and 2 and 4 weeks. Linear mixed-effects models (LMM) estimated individual recovery slopes (BLUPs), regressed on treatment, baseline variables, and interactions to identify effect modifiers; 200 bootstrap resamples provided internal validation. Predictive enrichment here denotes an exploratory framework using baseline FMA-UE to characterize differential AAS recovery, not a validated decision tool. Results LMM confirmed superior trajectory in AAS (Group × Time β = 1.294, 95% CI: 0.844–1.744, p < .001). Baseline FMA-UE demonstrated a significant interaction with treatment (interaction β = 0.276, p = .032), indicating that the treatment effect varies with baseline motor function. The Johnson–Neyman technique yielded a mathematical intersection at a centered FMA-UE value of 20.90, which lies outside the clinically meaningful range of the scale. Patients with FMA-UE ≤ 5 derived significant benefit (ΔFMA + 5.0 points, p < .001). In the smaller FMA-UE > 5 subgroup ( n = 14), the median change was 0.0 points (range −5 to 5), which did not reach statistical significance—likely reflecting limited statistical power in this stratum rather than a true absence of effect, as the continuous interaction model consistently predicts benefit across the entire FMA range. Conclusion AAS provides consistent benefit across the spectrum of baseline upper limb impairment, with baseline FMA-UE significantly modifying treatment response.