Prognostic factors and predictive modeling of in-hospital mortality among MDRO-positive inpatients: a retrospective cohort and GBD MICROBE study.

Existing studies on prognostic prediction for multidrug-resistant organisms (MDROs) are limited by single-model designs, incomplete consideration of ward environmental observations, and insufficient linkage to national antimicrobial-resistance burden data. This single-center pilot study aimed to identify interpretable prognostic factors for in-hospital mortality among MDRO-positive inpatients, internally evaluate a prognostic model in the non-surgical cohort, describe bed-unit environmental bacterial isolates, and compare local MDRO-related observations with national resistance-burden data. This retrospective cohort study enrolled 1057 MDRO-positive inpatients from 2019 to 2023. The non-surgical cohort was split into 7:3 training and validation sets. Seven machine-learning and statistical models were assessed via fivefold cross-validation. Strict statistical criteria (99.5% CIs, P < 0.005) were adopted for multivariable Cox regression analysis, with covariates selected based on clinical evidence and conservative screening. Environmental monitoring covered 84 bed units across three departments, and GBD 2023 data were used for national burden comparison. Advanced age and higher comorbidity/diagnosis burden were the most consistent internally validated prognostic factors associated with in-hospital mortality in the overall and non-surgical cohorts. Cumulative antibiotic use and urethral catheterization variables lost significance after correction. The XGBoost Cox model achieved the best internal discrimination (training C-index = 0.814, validation C-index = 0.807), whereas conventional Cox regression offered better clinical interpretability. Environmental monitoring identified variation in colony counts and environmental bacterial-isolate detection across bed-unit sites, and local MDRO-related observations were compared descriptively with national high-burden pathogens. Age and comorbidity burden were the most consistent internally validated prognostic factors for in-hospital mortality among MDRO-positive inpatients. Because the cohort was defined by laboratory-recorded MDRO detection and clinical infection could not be independently confirmed for every patient, colonization and infection may have been mixed. The environmental and GBD analyses provide exploratory and surveillance context, respectively; multicenter external validation is required before clinical implementation.