Research Appraisals
Evidence-based critical appraisals of the latest medical research, systematically evaluated using Oxford CEBM methodology.
Showing 13 appraisals
JMIR medical informatics
Machine Learning for Intraoperative Bleeding Prediction in Patients Undergoing Surgery: Scoping Review
BACKGROUND: Intraoperative bleeding is a critical event that impacts surgical safety and patient outcomes. Machine learning (ML) has demonstrated potential in prediction tasks, yet its methodological rigor and clinical translation face challenges. OBJECTIVE: This scoping review aims to systematically synthesize the current state of development, performance, and validation of ML models for predicting intraoperative bleeding, and to identify key barriers to their clinical implementation. METHODS: Following the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) guidelines, we systematically searched 7 databases (PubMed, Web of Science, Embase, CINAHL, CNKI [China National Knowledge Infrastructure], Wanfang, and VIP [China Science and Technology Journal Database]) from their inception to April 2025. Moreover, 2 reviewers (SY and PZ) independently screened studies, extracted data using the Checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modeling Studies (CHARMS), and assessed the risk of bias using the Prediction Model Risk Of Bias Assessment Tool (PROBAST). A narrative synthesis was used for data analysis. RESULTS: Out of 2651 screened records, 23 studies were included (sample sizes ranging from 48 to 48,543). Tree-based ensemble models (eg, random forests and extreme gradient boosting) were the most frequently used (16/23, 70%), followed by logistic regression (13/23, 57%), and deep learning (11/23, 48%). Model discrimination varied widely (mean area under the curve [AUC] 0.82, SD 0.08, range 0.63-0.93). Integration of multimodal data (electronic health records+imaging) was associated with higher performance. However, model validation was often inadequate; only 6 studies (6/23, 26%) performed external validation, and performance often declined (eg, AUC decreased from 0.85 to 0.63 in 1 study). Reporting exhibited selective bias; AUC was commonly reported (19/23, 83%), whereas key classification metrics, such as calibration (10/23, 43%) and precision (4/23, 17%), were often omitted. PROBAST assessment indicated a high risk of bias in all included studies (23/23, 100%). CONCLUSIONS: While ML models demonstrate technical promise for predicting intraoperative bleeding, our PROBAST assessment revealed a universally high risk of bias across all included studies. This fundamental methodological limitation, coupled with a severe lack of external validation and poor transparency in reporting, severely constrains the current clinical reliability of these models. Future research must prioritize prospective multicenter validation, adherence to Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD) reporting guidelines, and enhanced model interpretability to bridge the gap toward clinical utility.
13 June 2026
Read appraisal →Science robotics
Embodied or virtually represented: Navigating the embodiment debate in human-robot interaction.
The validity of virtually represented robots in HRI experiments depends on when, where, and for whom it matters.
30 May 2026
Read appraisal →JMIR perioperative medicine
AI-Generated Avatar Videos for Postoperative Patient Education Among Health Care Workers: Pilot Randomized Controlled Trial
BACKGROUND: Effective postoperative communication is vital for patient recovery, yet traditional text-based discharge instructions often lead to poor comprehension and adherence, particularly among patients with limited health literacy. Although educational videos improve understanding and retention, their widespread use has been hampered by high production costs. Generative artificial intelligence (AI) offers a scalable solution for creating engaging video content. OBJECTIVE: The primary objective of this pilot study was to assess the feasibility of creating and deploying AI-generated, avatar-led videos for postoperative instruction delivery. Secondary objectives included comparing knowledge retention, engagement, perceived clarity, and user experience between AI-generated video and traditional text-based handout formats among health care workers. METHODS: In this randomized pilot study, 38 health care worker volunteers were recruited as a convenience sample to pilot-test the intervention before patient implementation. Participants were assigned to either a text handout group (n=19, 50%) or an AI-generated video group (n=19, 50%). Both groups received information on 10 common postoperative topics. The primary outcome was objective knowledge, assessed via a 10-item quiz. Secondary outcomes, measured through surveys with 5-point Likert scales, included engagement time, subjective engagement, perceived clarity, usefulness, confidence in understanding, and information retention. Qualitative feedback was also collected. RESULTS: Objective knowledge quiz scores did not differ significantly between groups (mean 8.89, SD 1.20 for the AI-generated video group vs mean 8.21, SD 1.78 for the text handout group; P=.17; Cohen d=0.45). Participants in the AI-generated video group demonstrated significantly higher engagement time (mean 15.11, SD 7.78 minutes vs mean 8.84, SD 4.03 minutes; P=.004; Cohen d=1.04). They also rated instructions as significantly clearer (mean 4.63, SD 0.50 vs mean 4.00, SD 0.82; P=.007; Cohen d=0.93), more engaging (mean 4.05, SD 0.78 vs mean 3.32, SD 1.00; P=.02; Cohen d=0.81), and more effective for retention (mean 4.42, SD 0.84 vs mean 3.37, SD 0.68; P<.001; Cohen d=1.38). Qualitative feedback highlighted the engaging nature of AI-generated videos but noted areas for avatar refinement. CONCLUSIONS: In this pilot study with health care workers, AI-generated avatar videos did not improve objective knowledge scores but significantly enhanced engagement, perceived retention and perceived clarity (Cohen d=0.81-1.38). Future studies in actual patient populations with diverse health literacy levels are needed to determine whether these engagement advantages translate into improved knowledge outcomes.
28 May 2026
Read appraisal →JMIR cancer
Machine Learning-Based Survival Prediction Models for Young Patients With Gastric Cancer: Model Development and Validation Study
BACKGROUND: Despite a global decline in the incidence of gastric cancer (GC), the number of cases diagnosed among younger individuals continues to increase. Several studies have been conducted to develop predictive models of mortality in patients with GC. OBJECTIVE: We developed 3- and 5-year survival prediction models for young patients with GC based on machine learning-based survival modeling approaches. METHODS: Data from 813 young patients (≤50 years) diagnosed with GC between 2013 and 2015 were retrieved from the Gastric Cancer Public Staging Database. Among these 813 patients, data from 569 (70%) and 244 (30%) were allocated to the model training and testing datasets, respectively. Random survival forest, gradient boosting survival analysis, extra survival tree, and the Cox proportional hazards model were applied to predict survival outcomes at the 3- and 5-year time horizons. Model performance was assessed and quantified using the concordance index (C-index) metric. For the machine learning prediction models, the mean C-index values and corresponding 95% CIs were estimated across 100 repeated training iterations. RESULTS: For the random survival forest model, the C-index for predicting 3-year mortality was 95.89% (95% CI 95.80%-95.97%), whereas the C-index for predicting 5-year mortality was 91.82% (95% CI 91.68%-91.96%). In the gradient boosting survival analysis model, the C-index for predicting 3-year mortality was 95.32% (95% CI 95.31%-95.33%), and the C-index for predicting 5-year mortality was 89.98% (95% CI 89.95%-90.01%). For the extra survival tree model, the C-index for predicting 3-year mortality was 95.53% (95% CI 95.46%-95.60%), whereas the C-index for predicting 5-year mortality was 94.60% (95% CI 94.50%-94.70%). In addition, the Cox proportional hazards model showed a C-index of 94.15% for predicting 3-year mortality and 82.26% for predicting 5-year mortality. Tumor stage and tumor size were the primary predictive variables used to train the models for mortality prediction at different time points. Other variables exhibited varying levels of predictive contribution across different time points. CONCLUSIONS: These findings may facilitate the identification of high-risk young patients with GC who may benefit from more aggressive treatment strategies by enabling the prediction of mortality risk at different time points.
28 May 2026
Read appraisal →JMIR research protocols
AI-Supported, Integrative Prediction of Postoperative Delirium: Protocol for the CONFUSED Study
BACKGROUND: Postoperative delirium (POD) is a frequent and serious complication in older surgical patients, characterized by acute cognitive dysfunction and fluctuating levels of consciousness. POD is associated with prolonged hospitalization, long-term cognitive decline, reduced quality of life, and increased mortality. Despite its clinical relevance, the underlying pathophysiological mechanisms remain poorly understood, and reliable biomarkers for early prediction and prevention are lacking. OBJECTIVE: The CONFUSED study aims to identify molecular and clinical predictors of POD by integrating clinical data with proteomic, transcriptomic, and epigenetic analyses. The primary objective is to develop predictive models for POD using multimodal data. Secondary objectives include the identification of delirium-associated genes, proteins, and epigenetic signatures, as well as the exploration of patient subgroups at increased risk for POD. METHODS: CONFUSED is a prospective observational cohort study conducted at a German university hospital. Adult patients undergoing major surgery under general anesthesia will be enrolled until 100 cases of POD have been observed, which is expected to require a total sample size of approximately 200 to 300 patients. Blood samples are collected at 4 predefined time points: before premedication, immediately after surgery, and on postoperative days 2 and 5. Samples undergo comprehensive proteomic profiling, transcriptomic analysis using RNA microarrays, DNA methylation analysis, and genotyping of selected polymorphisms. Clinical data, including demographics, comorbidities, perioperative variables, medications, and delirium assessments using the Confusion Assessment Method (CAM) and CAM for the intensive care unit, are systematically recorded. Statistical analyses include univariate and multivariate methods, as well as machine learning approaches such as random forests and support vector machines, to identify relevant biomarkers and develop predictive models. The study protocol follows STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) and TRIPOD (Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis) guidelines and was approved by the responsible ethics committees. RESULTS: The study was registered in the German Clinical Trials Register (DRKS00033854) on March 18, 2024. Recruitment started in January 2024 and is ongoing at the time of manuscript submission. As of now, 135 patients have been enrolled. Sample collection and laboratory analyses are ongoing. Data analysis began in January 2026, with first results anticipated in July 2026. Final data lock is anticipated after the completion of recruitment. CONCLUSIONS: By integrating multimodal molecular data with clinical parameters and applying advanced machine learning techniques, the CONFUSED study aims to improve the prediction and understanding of POD. The results are expected to support the development of personalized preventive strategies and contribute to improved perioperative care for patients at risk of POD.
26 May 2026
Read appraisal →Renal failure
Machine learning-based risk prediction model for postoperative acute kidney injury in surgical patients with chronic kidney disease (CKD): development, validation, and SHAP-based explainability
This study aimed to develop machine learning models to predict postoperative acute kidney injury (AKI) in surgical patients with pre-existing chronic kidney disease (CKD), a serious complication associated with high morbidity and accelerated disease progression. Using data from 3,851 surgical patients with CKD from the MIMIC-IV database, nine supervised machine learning models were developed to predict AKI, as defined by the KDIGO criteria. Postoperative AKI occurred in 24.2% (932/3,851) of the patients. The XGBoost model demonstrated the best predictive performance, achieving an AUC of 0.844, accuracy of 0.791, sensitivity of 0.79, and F1-score of 0.73, followed by Random Forest and LightGBM. The calibration curve showed a good agreement between the predicted and observed risks. For interpretability, SHAP analysis was used to identify SOFA score, estimated glomerular filtration rate (eGFR), systolic blood pressure, albumin, and phosphorus as the most influential predictors. In conclusion, machine learning models, particularly XGBoost, provide accurate and interpretable predictions of postoperative AKI in this high-risk population. The integration of explainable AI facilitates transparent risk stratification, which may support personalized perioperative care and the development of real-time clinical decision support systems.
21 May 2026
Read appraisal →JMIR formative research
Metaverse-Based Virtual Reality for Remote Anatomy Education: Pilot Randomized Controlled Trial
BACKGROUND: Traditional anatomy teaching relies on cadaveric dissection and 2D resources, which often require in-person attendance and may limit spatial understanding. Virtual reality (VR) provides an immersive, remote alternative that supports 3D visualization from home. Recent evidence suggests that while VR may yield comparable factual knowledge gains to 2D methods, its primary value lies in enhancing learner engagement, motivation, and perceived educational value. OBJECTIVE: This pilot randomized controlled trial compares remote synchronized VR with didactic animated anatomy lectures for the teaching of tracheostomy anatomy. METHODS: Participants were recruited via convenience sampling through the VRiMS (Virtual Reality in Medicine and Surgery) Surgical Society network. All participants first attended a synchronous 20-minute online lecture delivered by a consultant surgeon. They were then individually randomized to one of 2 groups using a computer-generated sequence. Allocation was concealed until the intervention; however, participants and researchers were unblinded. The intervention group completed a 10-minute metaverse-based VR session, delivered by a consultant surgeon via 3D Organon's Medverse platform on the PICO 4 Ultra headset. The control group completed a 10-minute prerecorded 2D animated lecture, accessed on their personal device. Participants then swapped to the other modality. Data were collected via Google Forms at 3 intervals (baseline, postintervention, and postcrossover) to assess confidence, spatial understanding, and knowledge (10-item multiple-choice questions). The analysis of nonparametric data utilized Wilcoxon signed-rank tests for within-group changes and Mann-Whitney U tests for between-group differences. RESULTS: Twenty-four medical students from 11 United Kingdom and Irish medical schools participated. Adherence was 100%, with all participants completing their assigned 10-minute intervention and all assessment points. Ninety-two percent of participants (n=22) reported no prior tracheostomy anatomy teaching. Additionally, 83% (n=20) had no prior remote-synchronized VR anatomy teaching experience. Anatomical confidence improved significantly in the VR group compared with animation (mean change 1.58, SD 1.00 vs mean 0.50, SD 0.80, P=.01). Knowledge scores improved significantly in both groups (VR: mean 1.75, SD 1.54, P=.007; animation: mean 2.83, SD 1.70, P=.003), with no significant postintervention difference between groups (P=.46). VR participants reported significantly superior spatial understanding across all measured domains (all P≤.009). These included depth perception (3.75 vs 2.58, P=.009), appreciation of anatomy from different viewpoints (4.25 vs 2.33, P=.001), mental reconstruction from varying angles (3.83 vs 2.08, P=.002), and spatial depth supporting anatomical understanding (4.08 vs 2.08, P=.001). Following the completion of both modalities, participants rated VR as more engaging (mean 4.54, SD 0.78) and more educationally effective (mean 4.29, SD 0.95) than animation. CONCLUSIONS: Remote VR teaching is feasible and engaging and enhances spatial understanding compared to animation. While knowledge gains were comparable between modalities, VR improved learner confidence and perceived 3D comprehension. Hence, VR may represent a scalable adjunct or alternative to traditional anatomy teaching.
20 May 2026
Read appraisal →PloS one
The carbon footprint of surgical operations: 2023-2025 systematic review update
Surgical care significantly contributes to greenhouse gas emissions, with operating rooms consuming three to six times more energy than other departments relying heavily on single-use materials and anaesthetic gases. We conducted a systematic review of literature from 2023 to 2025 following PRISMA guidelines. The PubMed, Scopus and Healthcare LCA databases were searched, and eligibility criteria were applied. A total of 26 studies met inclusion criteria. Data were extracted on each study's scope and system boundaries, reported emission values, identified carbon hotspots, and any mitigation strategies evaluated. A structured quality assessment was conducted to evaluate the reliability of the included studies and to identify potential sources of bias. The included studies demonstrated wide variability in reported carbon footprints of surgery, ranging from under 5 kgCO₂e for minor procedures to over over 400 kgCO₂e for complex single-stage operations and approximately 1,000 kgCO₂e for whole multistage patient pathways. This systematic review underscores that surgical operations could have a significant carbon footprint, with emissions hotspots concentrated in consumable materials and energy consumption. It also reveals substantial variability and methodological heterogeneity in how surgical carbon footprints are calculated, pointing to the urgent need for standardized life-cycle based frameworks in surgical settings. Establishing common standards will enable more reliable benchmarking of surgical emissions and better comparisons across studies.
19 May 2026
Read appraisal →JMIR medical informatics
From Flow to Feature Using a Proof-of-Concept Spectral-Driven Machine Learning Approach Using Smart Urinary and Drainage Catheter Systems: Algorithm Development and Validation
BACKGROUND: Current urinary and drainage catheter systems collect fluids for visual inspection or manual sampling, offering limited diagnostic value while being labor-intensive and prone to error. Machine learning (ML) has the potential to automate the analysis of these fluids. However, existing methods rely on complex preprocessing steps, which hinder real-time analysis. OBJECTIVE: We aim to develop and evaluate a fully automated, real-time diagnostic approach for smart urinary and drainage catheter systems by leveraging spectral data and ML to differentiate pathological from healthy excreted fluids without the need for manual preprocessing. METHODS: This study proposes a novel, fully automated approach for smart urinary and drainage catheter systems that uses spectra and ML to extract features from excreted fluids, enabling real-time analysis directly. A total of 454 surgical drainage fluid samples (from 181 patients) and 401 urine catheter samples (from 168 patients) were analyzed using smart catheters and drains equipped with compact mini-spectrometer sensors. The collected spectral data were fed into 3 different ML models: a random forest, a partial least squares discriminant analysis regression, and a convolutional neural network (CNN). Each model aimed to extract features and differentiate between pathological and healthy urine and drainage samples based on the various biomarkers available from previously conducted laboratory analyses. RESULTS: All 3 approaches (random forest, partial least squares discriminant analysis regression, and CNN) achieved promising results, demonstrating the potential of the overall approach. In particular, the CNN models trained on the drainage biomarkers hemoglobin and bilirubin achieved the best results. Matthews correlation coefficient scores of 0.83 and 0.81 were obtained for hemoglobin and bilirubin, respectively, when differentiating between pathological and healthy samples using the extracted spectral features. CONCLUSIONS: This work demonstrates the potential of spectral-driven ML for smart urinary and drainage catheter systems. This approach offers a real-time, noninvasive method for analyzing excreted fluids, paving the way for improved diagnostics and personalized patient care. Further research will explore the optimal ML model for this application.
17 May 2026
Read appraisal →Antimicrobial resistance and infection control
Time-related changes in bacterial air contamination of sterile covered items in operating rooms: a systematic review and meta-analysis
BACKGROUND: There is still only limited evidence regarding how bacterial air contamination of sterile items changes over time in the operating room, even when protected by sterile covers. This review aimed to synthesize the available data to inform safe handling and preparation of sterile equipment. METHODS: A systematic search was conducted for controlled studies that assessed time-related bacterial contamination of covered sterile items in operating rooms and that reported the outcome as colony forming units (CFU) measured at two or more time points, generating distinct timelines. Six databases were searched from inception to 1 September 2025: Ovid MEDLINE, the Cochrane Central Register of Controlled Trials, CINAHL, the Cochrane Database of Systematic Reviews, Embase, and Web of Science. A meta-analysis was performed using linear regression with CFU as the dependent variable and time (minutes) as the independent variable. The review followed PRISMA guidelines and was registered in PROSPERO (CRD420251018280). RESULTS: Five timelines from four studies evaluating CFU counts yielded an R2 of 0.085, indicating that time explained 8.5% of CFU variation. Similarly, analysis of eight timelines from four studies assessing mean CFU values produced an R2 of 0.070, attributing 7% of variation to time. Neither analysis demonstrated a statistically significant linear association between time and CFU count (p = 0.176, CI: -1.319 < β < 4.863) or mean CFU values (p = 0.108, CI: 0.191 < β < 0.420) when sterile covers were used. Three additional timelines excluded from the meta-analysis showed mixed results: two indicated a positive association, while one showed no relationship. CONCLUSIONS: Time has only a modest impact on bacterial contamination of sterile items when protective covers are used, with minimal increases in CFU over different waiting periods. Clinically, this confirms that properly covered sterile items can be prepared in advance without compromising microbial safety, thus supporting efficient operating room organization and workflow. These findings reinforce the effectiveness of sterile covers as a key measure to reduce the risk of surgical site infections and ensure the safe handling of sterile equipment. Although covered storage appears safe, minimizing unnecessary waiting time, even with coverage, is good clinical practice.
15 May 2026
Read appraisal →BMC anesthesiology
Automated end-tidal control versus manual regulation in low-flow volatile anesthesia: a systematic review and meta-analysis
BACKGROUND: In low-flow anesthesia, automated monitoring of end-tidal oxygen (etO2) and end-tidal anesthetic agents (etAA) has been favored to reduce clinicians' cognitive workload during anesthesia maintenance. This systematic review, therefore, aimed to determine the comparative efficacy and safety of automated end-tidal controlled anesthesia (etCA) versus manually controlled anesthesia (MCA). METHODS: An electronic search was conducted across four databases: PubMed, Cochrane Central Register of Controlled Trials (CENTRAL), Web of Science, and Scopus. The retrieved references were then analyzed against the studies' screening and eligibility criteria before the full articles were included in the study. The Review Manager (RevMan 5.4.1) was then used to conduct the statistical analysis of the reported outcomes. RESULTS: The electronic search retrieved 965 articles, among which 10 were included in this review. The studies included a pooled sample size of 1269 who underwent surgical procedures under low-flow anesthesia. The analyzed evidence suggests with low certainty that the type of anesthesia control may not have any significant effect on both the anesthesia duration (SMD - 0.00 min; 95% CI [-0.34, 0.34] p = 0.99) and the procedural duration MCA (SMD 0.04 min; 95% CI; 95% CI [-0.26, 0.33] p = 0.80). Moreover, we found with low certainty in the evidence that the amount of sevoflurane consumed may not be significantly different between etCA and MCA (SMD - 0.32 mLs; 95% CI [-0.94, 0.29] p = 0.31). Our narrative synthesis indicated that etCA can potentially improve anesthesia delivery by reducing the required anesthetist interventions and ensuring stable and accurately administered anesthetics. CONCLUSION: This review found very low-quality evidence suggesting that the type of anesthesia control may not significantly affect anesthesia delivery parameters, such as the amount of anesthesia consumed and procedural and anesthesia duration. The analyses, however, had significant heterogeneity, making the estimates highly unreliable. On the other hand, some studies indicate potential advantages of using etCA, such as reducing the number of interventions required during anesthesia; however, the evidence remains limited, preventing a full exploration of these advantages.
15 May 2026
Read appraisal →A&A practice
Hypotension Prediction Index-Guided Hemodynamic Management on Postoperative Organ Outcome: A Systematic Review and Meta-Analysis of Randomized Clinical Trials
Intraoperative hypotension (IOH) is a frequent occurrence during noncardiac surgery. Hypotensive episodes can compromise tissue perfusion, and cumulative exposure has been associated with adverse outcomes such as acute kidney injury (AKI), myocardial injury, and increased mortality. The Hypotension Prediction Index (HPI) is an artificial intelligence tool designed to predict hypotensive events before they occur, allowing clinicians to intervene proactively to prevent or mitigate hypotension. Although randomized trials and pooled analyses have shown its effectiveness in reducing both the incidence and duration of IOH, its impact on patient-centered outcomes remains unclear. From 1946 to August 8, 2025, we prospectively queried MEDLINE (PubMed), Embase, Cochrane Library, and Web of Science. We included randomized controlled trials in adult surgical patients with invasive arterial monitoring, comparing HPI-guided perioperative hemodynamic management to standard care. Primary outcome was postoperative AKI, and secondary outcomes were postoperative myocardial injury, pneumonia, arrhythmia, stroke, postoperative cognitive dysfunction, including delirium, surgical site infection, length of hospital stay, and mortality. Two authors independently evaluated the risk of bias according to the methods described in the Cochrane Handbook for Systematic Reviews of Interventions. A meta-analysis was performed using RevMan 5.4.1 software. Dichotomous data were presented as risk ratio, 95% confidence interval [CI]; continuous data were presented as mean difference, 95% CI. All results were pooled using random-effects models. Of 883 imported references, 14 studies (N = 2030) were included for final analysis. No significant difference in incidence of postoperative AKI was observed between HPI-guided management and standard care, RR = 0.87, 95% CI (0.71-1.07). Postoperative myocardial injury was significantly reduced in HPI-guided management group, RR = 0.61, 95% CI (0.39-0.96), although exclusion-based sensitivity analysis suggests that this finding may not be robust. No other significant differences were observed for hospital length of stay, arrhythmia, stroke, cognitive dysfunction (including delirium), surgical site infection, pneumonia, or mortality. Although HPI-guided management is associated with a reduction in postoperative myocardial injury, its significance does not withstand sensitivity analysis. All other postoperative complications, including AKI incidence, are not reduced in the intervention groups. This study has limitations inherent to the heterogeneity of included patient populations, surgical interventions performed, standard of care practices, and reporting of postoperative outcomes.
15 Apr 2026
Read appraisal →PloS one
An interpretable deep learning framework for predictive modeling of postoperative infections in ICU patients
A significant proportion of intensive care unit (ICU) patients undergo surgical procedures, and some may develop postoperative infections. Accurately predicting postoperative infection risk and identifying key contributing factors is crucial for improving postoperative management and understanding infection mechanisms. However, this task is challenging due to the complex interplay of multiple risk factors. While machine learning models can model these intricate associations to predict postoperative infection risk, their lack of interpretability - failing to uncover each factor's impact-hinders their adoption in clinical settings. To address this difficulty, we introduced an interpretable deep neural network (DNN) model that integrates a permutation feature importance test (PermFIT). PermFIT rigorously evaluates the impact of each feature on postoperative infection risk through a rigorous statistical inference. By using only the identified important features as inputs, the DNN's predictive performance can be further enhanced. We conducted an extensive study using electronic health records (EHRs) from the Medical Information Mart for Intensive Care (MIMIC-III), a large-scale ICU EHR database. Under the PermFIT framework, our DNN model effectively identifies significant factors associated with postoperative infections while delivering the most accurate postoperative infection risk predictions. These findings highlight the clinical utility of our proposed DNN framework in managing postoperative care for ICU surgical patients, ultimately improving their health outcomes.
10 Apr 2026
Read appraisal →