Predicting Hypoglycemia in ICU Patients - A Machine Learning Approach

Q: What outcome does the study predict for ICU patients?

The primary outcome is whether a patient experiences any hypoglycemic episode during the ICU stay.

Q: How is hypoglycemia defined in this study?

Hypoglycemia is defined as blood glucose less than 3.9 mmol/L (70 mg/dL), used as a standard literature cutoff.

Q: Which machine-learning model performed best and how was it evaluated?

CatBoost performed best, with the highest AUROC (0.99) and strong classification metrics including accuracy, precision, sensitivity, specificity, and recall.

Aug 2, 2026

Uploaded by Noah

This study develops and validates a robust machine-learning model using electronic health records to forecast hypoglycemia risk in ICU patients in Jordan. The cohort includes 13,567 patients admitted 26,248 times over ten years, with a final analysis set of 1,896 patients after exclusions. Hypoglycemia is defined as blood glucose <3.9 mmol/L during ICU stay. Eight trained models are evaluated with AUROC and classification metrics, with CatBoost achieving the strongest overall performance. Results support earlier risk anticipation to reduce incidents and improve patient outcomes.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages22

Reading Time55 min

AccessFree to download

Introduction

Glycemic control and hypoglycemia risk in ICU

Evidence gaps and study aim

Methodology

Data source and cohort selection

Data preprocessing and outcome definition

Predictor variables

Predicting Hypoglycemia in ICU Patients: A Machine Learning Approach Abstract Background: The current study sets out to develop and validate a robust machinelearning model utilizing electronic health records (EHR) to forecast the risk of hypoglycemia among ICU patients in Jordan. Research Design and Methods: The present study drew upon a substantial cohort of 13,567 patients admitted 26,248 times to the intensive care unit (ICU) over ten years from July 2012 to July 2022. The primary outcome of interest was the occurrence of any hypoglycemic episode during the patient's ICU stay. Developing and testing predictor models was conducted using Python machine learning libraries. Results: A total of 1,896 were eligible to participate in the study, 206 experienced at least one hypoglycemic episode. Eight machine-learning models were trained to predict hypoglycemia. All models showed predicting power with a range of (74.53-99.69) for AUROC. Except for Naive Bayes, the six remaining models performed distinctly better than the basic logistic regression usually used for prediction in epidemiological studies. CatBoost model was consistently the best performer with the highest AUROC (0.99), accuracy and precision, sensitivity and specificity, and recall. Conclusions: We used machine learning to anticipate the likelihood of hypoglycemia, which can significantly decrease hypoglycemia incidents and enhance patient outcomes. Keywords: Hypoglycemia, Machine Learning, ICU, Real World data, Predictive models, 1. Introduction Maintaining proper glycemic control has been and will continue to be integral to patient care [1] . Being one of the most life-threatening conditions in the world [2], hypoglycemia which is routinely defined as having a blood glucose level of less than 70 mg/dL (3.9 mmol/L) [3,4], could result in a wide range of complications, including stroke [5], neurological impairment [6], seizure [2], coma, and most importantly, death [7] . The implications imposed by hypoglycemia are not only limited to threatening patient health but also constitute a heavy economic burden on the patients, their families, and society as a whole. [8], where these costs are either direct costs allocated for managing such a severe condition or indirect costs attributed to decreased productivity [9]. Therefore, hypoglycemia might considerably impair the quality of life. [8] . Unfortunately, hypoglycemia frequently occurs among critically ill patients, with an incidence rate reaching 45%, and is associated with increased morbidity and mortality [5] . The potential negative consequences of hypoglycemia are compounded by the fact that hypoglycemia could be entirely asymptomatic in some cases [10], and even when some symptoms exist, they are not necessarily identical among patients [11] . Besides, the usual manifestations associated with hypoglycemia are not easily detected among critically ill patients, in particular, primarily because of their already impaired physiological response [12] or the use of sedation [13] . Consequently, all these factors may impede any early deduction of many hypoglycemic episodes among severely ill patients. Considering the urgent demand for an evidence-based approach to minimize inpatient hypoglycemia [14], a noticeable shift from conventional linear and logistic predicting modeling methods, which often yield modest predictive results, toward advanced machine learning techniques is evident [14,15] . These techniques rely on the availability of rich clinical datasets stored in hospitals’ electronic record systems to produce robust predictive models and determine the best-performing model. [14,15] . With the availability of rich health records including a wide variety of variables, and the use of more advanced machine learning algorithms, recent methods have been developed to predict hypoglycemia in hospital settings accurately [15] . Although, in the past few years, machine learning has been progressivel

FAQs

What outcome does the study predict for ICU patients?

The primary outcome is whether a patient experiences any hypoglycemic episode during the ICU stay.

How is hypoglycemia defined in this study?

Hypoglycemia is defined as blood glucose less than 3.9 mmol/L (70 mg/dL), used as a standard literature cutoff.

Which machine-learning model performed best and how was it evaluated?

CatBoost performed best, with the highest AUROC (0.99) and strong classification metrics including accuracy, precision, sensitivity, specificity, and recall.

Aug 2, 2026

Uploaded by Noah