P.177 Artificial intelligence-based outcome prediction for moderate to severe traumatic brain injury: a systematic review and methodological appraisal

Abstract

Background: Artificial intelligence (AI) holds promise to predict outcomes for patients sustaining moderate to severe traumatic brain injury (msTBI). This systematic review sought to identify studies utilizing AI-based methods to predict mortality and functional outcomes after msTBI, where prognostic uncertainty is highest. Methods: The APPRAISE-AI quantitative evidence appraisal tool was used to evaluate methodological quality of included studies by determining overall scores and domain-specific scores. We constructed a multivariable linear regression model using study sample size, country of data collection, publication year and journal impact factor to quantify associations with overall APPRAISE-AI scores. Results: We identified 38 studies comprising 591,234 patients with msTBI. Median APPRAISE-AI score was 45.5 (/100 points), corresponding to moderate study quality. There were 13 low-quality studies (34%) and only 5 high-quality studies (13%). Weakest domains were methodological conduct, robustness of results and reproducibility. Multivariable linear regression highlighted that higher journal impact factor, larger sample size, more recent publication year and use of data that were collected in a high-income country were associated with higher APPRAISE-AI overall scores. Conclusions: We identified several study weaknesses of existing AI-based prediction models for msTBI; this work highlights methodological domains that require quality improvement to ultimately ensure safety and effiicacy of clinical AI models.

Ähnliche Arbeiten