Abstract
Background and objective In the context of Japan's rapidly aging society, there is an increasing demand for effective rehabilitation strategies to support individuals with diminished physical function. This study aimed to identify the optimal combination of systemic condition and physical function indicators that most strongly influence functional independence measure (FIM) gains, a key outcome index in rehabilitation. Methods This retrospective study included 31 inpatients (mean age: 71.5 ± 13.6 years) with cerebrovascular or musculoskeletal diseases in a convalescent rehabilitation ward. Systemic condition was assessed using 14 blood biomarkers (e.g., mean corpuscular hemoglobin concentration (MCHC), albumin (Alb), and total protein (TP)), while physical function was evaluated using grip strength, knee extensor strength-to-weight ratio, the six-minute walk test, and the Berg Balance Scale. Random forest models with five-fold cross-validation were used to identify predictors of FIM gain. Feature importance was assessed using the mean decrease in Gini (MDG) index, and a composite score was calculated to rank variable combinations. Results The mean FIM gain during hospitalization was 25.0 ± 17.1 points. The strongest individual predictors among systemic condition indicators were MCHC (MDG = 1.3607), Alb level (1.0777), and TP (0.8868). For physical function, the most important variables were balance ability (2.1481), paralytic-side muscle strength (2.0245), and walking ability (1.9111). Composite score analysis showed that the combination of "balance ability + MCHC" had the highest predictive value (score = 0.6253), followed by "non-paralytic grip strength + MCHC" (0.5754) and "balance ability + Alb" (0.559). Conclusions Our findings suggest that combining indicators of physical function, especially balance ability and grip strength, with systemic markers such as MCHC and Alb can enhance the prediction of functional recovery. The use of such combined indices may facilitate more tailored and efficient rehabilitation interventions. Notably, MCHC, a lesser-studied indicator in rehabilitation, showed high predictive value, highlighting the potential role of anemia-related biomarkers in optimizing functional outcomes. Although the sample size was small, this exploratory analysis provides hypothesis-generating insights that may inform future large-scale rehabilitation studies.