PLOS ONEResearch HighlightsOpen Access

Interpretable Deep Learning Framework for Predicting Postoperative Infections in ICU Patients

09 Apr 20264 min read0 viewsJournal Feed

GIST

by Xiaoyu Wu, Isaac Luria, Meisheng Xiao, Patrick Tighe, Fei Zou, Baiming Zou 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.

Clinical Editorial

Interpretable modeling objective and approach

Objective: Predict postoperative infections in ICU patients and identify contributing factors to support postoperative management.
Study design: Analytic framework applied to electronic health records, leveraging a deep neural network (DNN) enhanced with interpretability mechanisms.
Methodology: The framework integrates Permutation Feature Importance Test (PermFIT) to statistically assess each feature’s impact on infection risk, enabling identification of influential factors before feeding inputs to the DNN.

Data source and population context

Data source: MIMIC-III, a large ICU EHR database.
Population scope: ICU surgical patients undergoing procedures with potential risk for postoperative infections.

Modeling strategy and features

Interpretability emphasis: PermFIT is used to quantify the contribution of individual features to risk, addressing a key barrier to adopting ML in clinical settings.
Feature selection: Only features deemed important by PermFIT are used as inputs to the DNN, with the aim of improving predictive performance and model transparency.
Predictive objective: The model aims to deliver accurate postoperative infection risk predictions while revealing which factors most strongly influence risk.

Key findings and implications

Signal identification: The PermFIT-enabled DNN effectively highlights significant factors associated with postoperative infections.
Predictive performance: By restricting inputs to identified important features, the model achieves enhanced predictive accuracy for infection risk.
Clinical relevance: The framework demonstrates potential utility in guiding postoperative care decisions for ICU surgical patients and advancing understanding of infection mechanisms.

Context and limitations

Reported scope: The study emphasizes the interpretability advantages and predictive gains within the PermFIT-DNN construct, grounded in the MIMIC-III ICU surgical patient cohort.