Plastic Surgery

MySurgeryRisk maintains high multicenter accuracy for ICU admission, MV, AKI, and mortality

In a 14-institution cohort of 508,097 major inpatient operations, MySurgeryRisk models achieved AUROCs of 0.93 for ICU admission, 0.94 for mechanical ventilation, 0.92 for AKI, and 0.95 for in-hospital mortality. Models used routinely collected EHR variables with primary procedure code and clinician-specific factors as the top predictors. Development (2012–2020) and validation (2020–2023) splits suggest preserved performance across time and sites. Before deploying for clinical decisions, confirm local calibration and assess integration with existing risk workflows. 1

OPTION-12/OPTION-5 scores show modest, setting-dependent changes in observed shared decision-making

Systematic review of 174 studies (~20,000 consultations) found baseline OPTION-12 mean 25.1 and OPTION-5 mean 31.8, with higher postintervention scores (OPTION-12 38.4; OPTION-5 47.7). Multivariable analysis identified clinical setting as the independent predictor for OPTION-12 and consultation duration for OPTION-5. Overall improvement since 2015 was modest and heterogeneity was high across studies. Interpret OPTION score changes cautiously because the clinical threshold for meaningful change remains unclear. 2

Global medical AI evidence is dominated by preclinical studies with few randomized trials

Scoping review of 218 systematic reviews identified 4,667 primary studies; 88.2% were preclinical and only 2.4% were randomized controlled trials. Most RCTs were single-center (67.3%) and many did not report adherence to reporting guidelines. Although 82.3% of RCTs reported favorable outcomes, common methodological concerns included allocation concealment and blinding. For clinical adoption, prioritize AI tools supported by well-reported, multicenter randomized evidence with transparent methods. 3