Dynamic evaluation system for improving hospital outsourcing service performance: a G1-Critic and LSTM+Dropout approach

BACKGROUND: Static scorecards dominate hospital outsourcing evaluation, but fixed weights cannot adapt to shifting operational priorities. We developed a dynamic evaluation system that intentionally re-weights indicators each month to steer contractors toward emerging risk areas. METHODS: A two-campus, 1350-bed medical group was followed for 24 months. During 2023, baseline performance of six domains (service quality, efficiency & timeliness, cost control, compliance, safety, staffing & satisfaction) was captured with equal weights. In 2024, a G1–CRITIC hybrid supplied the initial weight vector, after which an Long Short-Term Memory (LSTM)+Dropout network forecast next-month scores. Weights were automatically up-regulated for indicators predicted to deteriorate and down-regulated for those expected to improve, then normalised to 1. Impact was quantified with interrupted time-series analysis (ITSA). RESULT: The LSTM+Dropout model yielded test RMSE ≤ 1.60 (scale 0–100) across all indicators. After 12 months of dynamic adjustment, ITSA showed immediate and significant improvements in service quality (+23.2 points, p = 0.001), cost control (+24.6 points, p = 0.004) and the standardised mean score (+8.5 points, p = 0.001). No adverse effects were observed in other domains. CONCLUSIONS: Converting a static scorecard into a self-adjusting steering lever significantly accelerated contractor quality gains. The system requires only routine administrative data and open-source software, offering a readily replicable tool for proactive, data-driven governance of outsourced hospital services.