Developing a Risk Prediction Model for the Identification of Necrotizing Fasciitis (NF) in Patients
Author(s)
Nandana Acharjee, MBBS1, Riddhi Kumar Markan, M.SC. Economics2, Anuj Gupta, MSc1, Rajesh Ganguly, MBA3, Vikash Kumar Verma, MBA, PharmD3, Abhinav Nayyar, MBBS, MBA3, Ina Kukreja, MBA, PT3, Abhimanyu Roy, MBA3, Arunima Sachdev, MA3, Rahul Goyal, BS Tech4, Louis Brooks Jr, MA5, Marissa Seligman, BS Pharma2.
1Optum, Noida, India, 2Optum, Boston, MA, USA, 3Optum, Gurgaon, India, 4Optum, Phoenix, AZ, USA, 5Optum, Bloomsbury, NJ, USA.
1Optum, Noida, India, 2Optum, Boston, MA, USA, 3Optum, Gurgaon, India, 4Optum, Phoenix, AZ, USA, 5Optum, Bloomsbury, NJ, USA.
Presentation Documents
OBJECTIVES: Necrotizing fasciitis (NF) is a rapidly progressing, high-mortality surgical emergency often misdiagnosed due to its atypical presentation. Early and accurate diagnosis remains challenging, relying heavily on clinical judgment and ancillary tests. Our study develops and evaluates a machine learning (ML) model for early detection and management of NF.
METHODS: This study analyzed 17,901 patients from Optum® de-identified Market Clarity database, covering January 1, 2021, to December 31, 2023. The index event was NF diagnosis (ICD-10 code M72.6). We ensured continuous eligibility for three-year baseline period and examined for comorbidities and clinical activities during this period. The study cohort included NF-diagnosed patients, while the control cohort had patients without NF, matched by age, gender, and region. The data was divided into 80:20 ratio for training and testing. Supervised machine learning techniques (Logistic Regression, XGBoost, Random Forest) were used to predict occurrence of NF. Model performance was evaluated using AUC, accuracy, and F1 score.
RESULTS: The accuracy scores for Logistic Regression, XGBoost, and Random Forest were 77.52%, 77.46%, and 76%, respectively. The F1 scores were 78%, 77%, and 76%, respectively, and the AUC scores were 83%, 82%, and 73%, respectively. Logistic Regression showed significant associations between various risk predictors and NF likelihood. Odds ratios (OR) indicated higher probabilities of NF in patients with Peripheral vascular disease (OR: 2.57), Diabetes (OR: 1.71), Skin ulcer (OR: 2.32), and Open wound (OR: 5.37). Further, we will analyze association of NF with other comorbidities.
CONCLUSIONS: The ML models identified key risk factors for NF, improving early detection and aiding healthcare professionals in making informed decisions. This can lead to improved patient outcomes and reduction in morbidity and mortality. Further validation is required to enhance models’ clinical applicability.
METHODS: This study analyzed 17,901 patients from Optum® de-identified Market Clarity database, covering January 1, 2021, to December 31, 2023. The index event was NF diagnosis (ICD-10 code M72.6). We ensured continuous eligibility for three-year baseline period and examined for comorbidities and clinical activities during this period. The study cohort included NF-diagnosed patients, while the control cohort had patients without NF, matched by age, gender, and region. The data was divided into 80:20 ratio for training and testing. Supervised machine learning techniques (Logistic Regression, XGBoost, Random Forest) were used to predict occurrence of NF. Model performance was evaluated using AUC, accuracy, and F1 score.
RESULTS: The accuracy scores for Logistic Regression, XGBoost, and Random Forest were 77.52%, 77.46%, and 76%, respectively. The F1 scores were 78%, 77%, and 76%, respectively, and the AUC scores were 83%, 82%, and 73%, respectively. Logistic Regression showed significant associations between various risk predictors and NF likelihood. Odds ratios (OR) indicated higher probabilities of NF in patients with Peripheral vascular disease (OR: 2.57), Diabetes (OR: 1.71), Skin ulcer (OR: 2.32), and Open wound (OR: 5.37). Further, we will analyze association of NF with other comorbidities.
CONCLUSIONS: The ML models identified key risk factors for NF, improving early detection and aiding healthcare professionals in making informed decisions. This can lead to improved patient outcomes and reduction in morbidity and mortality. Further validation is required to enhance models’ clinical applicability.
Conference/Value in Health Info
2025-05, ISPOR 2025, Montréal, Quebec, CA
Value in Health, Volume 28, Issue S1
Code
MSR66
Topic
Methodological & Statistical Research
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics
Disease
SDC: Rare & Orphan Diseases, SDC: Sensory System Disorders (Ear, Eye, Dental, Skin)