Item-level analysis and network analysis of the functional independence measure cognitive items at discharge after acute stroke.

PURPOSE: This study characterized item-level profiles and inter-item relationships of the five Functional Independence Measure (FIM) cognitive items at discharge after acute stroke, using clinically anchored coding and network analysis. METHODS: Discharge FIM cognition scores were retrospectively analyzed in 974 consecutive patients. Items were dichotomized into supervision-or-better (scores 5-7) versus requiring assistance beyond supervision (scores 1-4). Attainment rates, item-rest correlations, and conditional response probability curves were described. Original 7-point scores were used for supplementary item-level and dimensionality analyses and ordinal partial-correlation network estimation. Edge-weight accuracy and centrality stability were assessed by bootstrapping, with Pearson-based sensitivity analysis. RESULTS: Supervision-or-better attainment ranged from 67.0% to 77.7%, and item-rest correlations ranged from 0.79 to 0.91. Analyses supported a dominant common dimension. The ordinal network showed two major conditional associations: comprehension-expression (r = 0.71) and memory-problem solving (r = 0.55). Comprehension showed the highest strength centrality, with good stability (CS coefficient = 0.677). Stronger edges were estimated more precisely, with a comparable Pearson-based pattern. CONCLUSION: Discharge FIM cognition profiles demonstrated clinically interpretable item-level heterogeneity and a conditional association structure dominated by two edges. These findings provide a discharge-oriented descriptive framework warranting replication and longitudinal validation against external outcomes.