Dr. Korea University Anam Hospital Seoul, Seoul-t'ukpyolsi, Republic of Korea
Objectives: To develop multivariate time-series artificial intelligence (AI) based clustering analysis algorithm that comprehensively considers recovery patterns in patients with fragility hip fracture.
Design: Total of 211 patients who underwent surgery for acute unilateral fragility hip fracture at 3 hospitals were enrolled. Excluding patients missing 2 or more functional assessments, 132 patients were selected for analysis. Patients performed functional evaluations, including gait ability and balance (KOVAL, FAC, FIM, MRMI, BBS), activities of daily living (K-MBI, K-IADL), QOL (EQ5D), frailty (K-FRAIL), cognition (K-MMSE) and mood (K-GDS). Evaluations were performed on admission for rehabilitation, at discharge, and at the 3, 6 and 12-month follow-up visits after surgery. The distance between the data was measured by dynamic time warping method, and the patient were classified using multivariate time series clustering analysis. Using data during hospitalization period, the recovery pattern of the subsequent period was predicted by the XGBoost machine learning model.
Results: Five clusters (HIGH, MID1-3, LOW) with different recovery patterns were derived. Table 1 presents the demographic and clinical characteristics of each cluster, and Figure 1 presents the recovery trajectory for each functional outcome in each cluster. The HIGH and MID1 clusters were younger, had better premorbid ambulatory function, and had better functional recovery. The MID1 showed a lower MMSE than HIGH and very low function in the acute phase. Compared to the MID3, the MID2 cluster had lower premorbid ambulatory ability, but a similar level of function at 1 year postoperatively with rapid recovery to premorbid levels after surgery. However, the MID2 cluster had lower cognitive function and IADL improvement. The LOW cluster showed poor premorbid function and recovery.
Conclusions: By analyzing the comprehensive patterns of 11 functional outcomes, 5 clusters were derived according to the recovery patterns in patients with fragility hip fracture.
