Data extraction and risk of bias assessment

Three reviewers (K.I., T.I., and S.I.) independently screened titles and abstracts in the first round, followed by full-text review in the second round. Reference management and screening were performed using EndNote, version 21 (Clarivate, Philadelphia, PA, USA). Extracted data included study characteristics, patient demographics, ICU setting, details of the biotic intervention (type, dose, and duration), and outcome data for all predefined endpoints. When relevant information was unclear or unavailable in the published report, the registered study protocol was consulted. Discrepancies among reviewers were resolved through discussion and, when necessary, adjudicated by a fourth reviewer (J.H.). Risk of bias in individual studies was independently assessed using the Cochrane Risk of Bias 2 (RoB 2) tool [16], and disagreements were resolved by consensus.