PneumoMLPred: Innovating pneumonia management with guideline adherence review and explainable machine learning-based mortality prediction - A single-centre retrospective study

Authors

• Kamel A. Samara, Jefferson Abington Hospital, Abington, Pennsylvania, United States of America. Electronic address: kamel.samara@jefferson.edu.
• Hiba Jawdat Barqawi, Department of Clinical Sciences, College of Medicine, University of Sharjah, Sharjah, United Arab Emirates; Research Institute of Medical and Health Sciences, University of Sharjah, Sharjah, United Arab Emirates.
• Saryia Adra, Harlem Hospital Center, New York, United States of America.
• Batool Aldaher, The University of Tennessee at Chattanooga, Tennessee, United States of America.
• Drishti D. Kampani, Al Jalila Children's Hospital, Dubai Health, Dubai, United Arab Emirates.
• Enad Haddad, Jefferson Abington Hospital, Abington, Pennsylvania, United States of America.
• Hebah Soudan, Bon Secour Mercy Health St. Vincent Medical Center, Toledo, Ohio, United States of America.
• Islam Ibrahim Anwar, Al Qassimi Women and Children Hospital, Sharjah, United Arab Emirates.
• Mohamad Besher Adi, Ascension Saint John Hospital, Moross Rd, Detroit, Michigan, United States of America.
• Omar Hamadi, Advocate Health - MidwestFollow
• Yumna Almour, Al Qassimi Women and Children Hospital, Sharjah, United Arab Emirates.
• Mohamed Nassef, Al Qassimi Hospital, Sharjah, United Arab Emirates.
• Zaher Al Aghbari, Department of Computer Science, University of Sharjah, Sharjah, United Arab Emirates.
• Eman Abu Gharbieh, Department of Clinical Sciences, College of Medicine, University of Sharjah, Sharjah, United Arab Emirates; Research Institute of Medical and Health Sciences, University of Sharjah, Sharjah, United Arab Emirates; School of Pharmacy, The University of Jordan, Amman, Jordan.

Recommended Citation

Samara KA, Barqawi HJ, Adra S, et al. PneumoMLPred: Innovating pneumonia management with guideline adherence review and explainable machine learning-based mortality prediction - A single-centre retrospective study. Respir Med. Published online May 19, 2026. doi:10.1016/j.rmed.2026.108896

Affiliations

Advocate Illinois Masonic Medical Center

Abstract

Introduction: Community-acquired pneumonia (CAP) is one of the leading causes of death globally. Changing antibiotic resistance profiles, paucity of reliable CAP mortality prediction models, and ageing populations will increase its burden. This study aims to evaluate adherence to CAP management guidelines as well as develop an ML-based mortality prediction model.

Methods: This retrospective, chart review included all CAP cases between 2016 and 2021. Variables collected included patients' demographics, comorbidities, vitals, laboratory, imaging, and culture findings, as well as initial and targeted antibiotic regimens. Data was pre-processed and analysed in python-3. Adherence to the American thoracic society/infectious diseases society of America 2019 guidelines (ATS/IDSA) was evaluated. The minimum required antibiotic dosages used to determine guideline adherence were defined according to the 2019 ATS/IDSA guidelines for an average adult with normal renal function. Random forest was used for feature selection, while several base classifiers and ensembles were evaluated. SHAP was used to generate interpretable factors for the best-performing model.

Results: 783 cases were included, all of which were Pneumonia Severity Index (PSI) class IV/V; overall mortality rate was 15.75%. Etiological diagnosis was rarely established and was not used to guide antibiotic usage. Prescribed doses were frequently below the recommended levels. Less than one-third of cases were adhering to the 2019 ATS/IDSA management guidelines. PneumoMLPred, an explainable support vector classifier-based pneumonia prognostication model, had accuracy = 88.5%, specificity = 96.3%, and AUROC = 0.87. SHAP revealed that PSI, CRP, and kidney diseases were the top three drivers behind in-hospital mortality.

Conclusions: Adherence to the 2019 ATS/IDSA guidelines was very limited. PneumoMLPred has great potential for in-hospital CAP mortality prediction.

Type

Article

PubMed ID

42162923