Biomedicine: 2023; 43(2): 696-700 March-April 2023 DOI: https://doi.org/10.51248/.v43i02.2250 Biomedicine- Vol. 43 No. 2: 2023 Research Article Risk factors of pneumonia among elderly with robust Poisson regression - A study on mimic III data Kalesh M. Karun 1 , Amitha Puranik 2 , Lintu M. K. 3 Deepthy M. S. 4 1 ICMR- National Institute of Traditional Medicine, Dept. of Health Research (Govt. of India), Nehru Nagar, Belagavi, 590010, Karnataka, India 2 Centre for Health Informatics, Computing and Statistics, Lancaster Medical School, Lancaster University, LA1 4YW, United Kingdom 3 Department of Data Science, Prasanna School of Public Health, Manipal Academy of Higher Education, Manipal, 576104, Karnataka, India 4 Department of Biostatistics, Jawaharlal Institute of Postgraduate Medical Education and Research, Puducherry, 605006, India (Received: November 2022 Revised: March 2023 Accepted: April 2023) Corresponding author: Kalesh M. Karun. Email: karunkmk@gmail.com ABSTRACT Introduction and Aim: Pneumonia is a common and serious illness among the elderly. Early identification of the risk factors for pneumonia is essential for improving the survival outcomes among elderly. The present study aimed to identify an optimal regression approach to determine the risk factors for pneumonia among elderly patients. Materials and Methods: The present study utilized data from the Medical Information Mart for Intensive Care (MIMIC III) to evaluate the use of alternative generalized linear models to identify the risk factors for pneumonia. The regression model with the smallest AIC, BIC and SE was considered as the appropriate regression model for the data. Robust Poisson model was considered the best fit for the current data as it had the lowest AIC, BIC and standard error compared to other regression models. Results: Variables such as BMI, renal failure, hypertension, diabetes and asthma were identified as the significant risk factors for pneumonia. The risk of pneumonia was found to be significantly higher in the underweight category of BMI [RRadj=1.70; 95% CI=1.38, 2.08]; diabetic patients [RRadj =1.29; 95% CI=1.03, 1.61); asthmatic patients [RRadj =1.35; 95% CI=1.15, 1.58] and patients with renal failure [RRadj =1.16; 95% CI= 1.05, 1.29]. Conclusion: Among various binary regression models, Poisson regression with robust variance (sandwich Poisson regression) provided unbiased estimates of the relationship. In the present study, variables such as BMI, renal failure, diabetics, hypertension and asthma were identified as the significant risk factors for pneumonia in the elderly using robust Poisson regression. Keywords: Regression models; robust Poisson regression; negative binomial; elderly; pneumonia; risk factors; MIMIC III. INTRODUCTION neumonia is an inflammation and infection of the lung tissue caused by infectious agents such as bacteria, viruses, or fungi (1). Community-acquired pneumonia (CAP) has a global incidence of 1.5 to 14 cases per 1000 person-years (2). This condition is highly common among persons at the extremes of age due to their weakness and vulnerability. A US-based retrospective study reported that CAP had higher hospitalization rates and total expenditures than stroke, myocardial infarction, and osteoporotic fractures in the elderly population (3). Elderly patients with pneumonia who require hospital care are more apparent and are more likely to develop complications that require longer hospital stays (4). Studies have reported that different factors such as gender, age, diabetes, hypertension, body mass index, dementia, history of heart failure, history of stroke, chronic respiratory disease, and chronic liver disease are associated with hospitalization for pneumonia among the elderly (5-8). The selection of an appropriate regression technique to model the data is very important in any field of research for the proper identification of the risk factors for events of interest. Several types of generalized linear regression models are available for modelling the binary event in cohort studies such as Poisson, log binomial, robust (sandwich) Poisson and negative binomial regression models (9). The present study aimed to identify an optimal regression approach to determine the risk factors for pneumonia among elderly patients in the intensive care unit. MATERIALS AND METHODS Data source The current study utilized data from MIMIC III to evaluate the use of alternative generalized linear models for the analysis of binary outcome i.e. pneumonia status and compared the results of these selected models to identify the most suitable model for this data (10). The MIMIC III database consists of de- identified clinical data of 49785 patients, among which a total of 5214 patients aged greater than 65 years and BMI between 15 and 50 years were included P 696