Two-stage integrated evidence synthesis: Systematic review and meta-analysis of hyperbaric oxygen therapy for frostbite plus exploratory development of frostbite amputation risk prediction model using MIMIC-IV and eICU-CRD databases

Abstract

## Study Objective This two-stage integrated study aims to address the unresolved clinical question: which patients with severe frostbite may benefit from adjunctive hyperbaric oxygen therapy (HBOT). Stage 1 will synthesize existing clinical evidence of HBOT for acute frostbite via systematic review and meta-analysis. Stage 2 will develop an exploratory multivariable prediction model to quantify individual amputation risk among critically ill frostbite patients, to inform eligibility screening for future interventional trials of adjunctive frostbite treatments. ## Study Design **Stage 1: Systematic review and meta-analysis (PRISMA 2020 compliant)** We will search multiple databases without language restriction for clinical studies comparing HBOT versus standard care for patients with acute frostbite. The primary outcome is the rate of major amputation. Secondary outcomes include wound healing time, infectious complications, and in-hospital mortality. Pooled effect sizes will be calculated using Freeman-Tukey double arcsine transformation for proportion meta-analysis. We will assess risk of bias using ROBINS-I for non-randomized studies, conduct pre-specified subgroup and sensitivity analyses, evaluate publication bias with funnel plots and Egger’s regression test, and rate overall evidence certainty using the GRADE framework. **Stage 2: Exploratory prognostic prediction model (TRIPOD compliant)** A retrospective multicenter critical care cohort will be constructed from MIMIC-IV and eICU-CRD. Adult patients with cold injury (ICD-10 T33-T35, T68-T69, T14.3) admitted to ICU will be included. The primary outcome is in-hospital amputation. Predictors are selected a priori based on clinical pathophysiology and prior literature, including age, vital signs on admission, serum creatine kinase, lactate, creatinine, white blood cell count, frostbite injury extent, comorbidities, and administered treatments. Firth penalized logistic regression will be applied to handle the rare binary outcome. Internal validation will be performed with 1000 bootstrap resampling. Model performance will be assessed via AUC, Brier score, calibration curve, net reclassification improvement (NRI), integrated discrimination improvement (IDI), and decision curve analysis (DCA). We will transparently report feasibility of external validation in eICU-CRD; if external validation is infeasible due to limited event numbers, this limitation will be explicitly stated. ## Expected Output A single integrated manuscript combining the two analytical stages. This registration covers both the systematic review protocol and prediction model analytical plan. If later separated into two independent publications, this OSF registration will be cited in both articles. ## Limitations Pre-specified 1. The prediction model is exploratory, limited by the small number of frostbite amputation events in ICU databases. 2. The meta-analysis is anticipated to include mostly non-randomized retrospective studies with high heterogeneity and low-to-very-low certainty evidence. 3. We predefine variables by clinical knowledge instead of data-driven variable selection (LASSO) due to limited sample size, to reduce overfitting risk.