AI-Based Analysis of Post-Surgical Patient Data for Risk Stratification After Total Knee Arthroplasty

Abstract

Post-surgical care generates large volumes of heterogeneous data, but most health information systems rely on sparse measurements collected at scheduled intervals. As a result, early deviations in recovery trajectories are often difficult to detect. Here, a data analytics framework is described for analyzing longitudinal post-surgical data, using total knee arthroplasty (TKA) as a representative use case. Systematically gathered electronic health record data, patient-reported outcomes, and activity assessments were utilized to delineate recovery trajectories and ascertain deviations correlated with unfavorable outcomes. Interpretable machine learning models were trained and evaluated on prospectively collected data from 1,000 patients to estimate the probability of outcome events based on temporal patterns observed across multiple data streams. The model exhibiting the highest efficacy attained an area under the receiver operating characteristic curve quantified at 0.896, demonstrating consistent performance across various validation folds. An examination of the model’s features revealed that longitudinal alterations had a more significant impact on predictive efficacy compared to discrete measurements, thereby highlighting the importance of temporal modeling. The results of this research indicate that longitudinal datasets can be utilized to enhance scalable risk stratification in intricate healthcare data infrastructures. The suggested framework operates as a data-processing intermediary designed to facilitate human evaluation instead of automating the decision-making process.

Ähnliche Arbeiten