Artificial neural network machine learning prediction of the smoking behavior and health risks perception of Indonesian health professionals

Nuryunarsih, Desy; Okatiranti, Okatiranti; Herawati, Lucky

doi:2023.38.1.e2023003

Environ Anal Health Toxicol > Volume 38:2023 > Article

Brief Report

Published online March 6, 2023

Environmental Analysis Health and Toxicology 2023;38(1):e2023003-0. doi: https://doi.org/10.5620/eaht.2023003

Artificial neural network machine learning prediction of the smoking behavior and health risks perception of Indonesian health professionals

Desy Nuryunarsih¹

, Okatiranti Okatiranti²

, Lucky Herawati³

¹School of Health and Life Sciences, Glasgow Caledonian University, United Kingdom
²Department of Nursing, the University of Nottingham United Kingdom, Nottingham, United Kingdom
³Department of Environmental Health, Yogyakarta Health Polytechnic, Yogyakarta, Indonesia

Corresponding Author: Desy Nuryunarsih ,Email: nuryunars@gmail.com

Received: November 29, 2022; Accepted: February 23, 2023.

ABSTRACT

Health professionals (HPs) can play an important role in influencing the smoking behavior of their patients and the implementation of smoke-free workplace policies. In some countries physicians and dentists may not have a no-smoking policy in place. Breathing in other people’s tobacco smoke (second-hand smokers) increase the risk of smoking related diseases. Environmental Tobacco smoke ETS causes a similar range of diseases to active smoking, including various cancers, heart disease, stroke, and respiratory diseases. Little is known about the smoking-related attitudes and clinical practices of HPs in Indonesia. Evidence suggests that high smoking rates remain among male HPs; however, the risk perceptions and attitudes to smoking among Indonesian HPs have not been investigated using prediction model artificial neural networks. For this reason, we developed and validated an artificial neural network (ANN) to identify HPs with smoking behavior. The study population consisted of 240 HPs, including 108 (45%) physicians, and 132 (55%) dentists, with more female (n=159) than male participants (n=81) for both professions. Participants were randomly divided into two sets, the training (192) and test (48) sets. The input variables included gender, profession (doctor or dentist), knowledge regarding smoking-related diseases and awareness of smoking provided to their patients, smoke-free policy in place at their workplace, and smoking status. ANN was constructed with data from the training and selection sets and validated in the test set. The performance of ANN was simultaneously evaluated by discrimination and calibration. After the training, we completed the process using the test dataset with a multilayer perceptron network, determined by 36 input variables. Our results suggested that our final ANN concurrently had good precision (89%), accuracy (81%), sensitivity (85%), and area under the curve (AUC; 70%). ANN can be used as a promising tool for the prediction of smoking status based on health risk perceptions of HPs in Indonesia.

Keywords: Neural network, smoking, health professionals, health risks