Eur Rev Med Pharmacol Sci. 2022 Oct;26(20):7344-7348. doi: 10.26355/eurrev_202210_30003.
ABSTRACT
OBJECTIVE: An interaction between hereditary and environmental variables is thought to be the cause of keratoconus, a progressive ectatic corneal condition. The identification of risk factors is necessary since they are currently the subject of intense debate and are crucial to the management and prevention of the disease. The objective of this study is to gain a better understanding of the risk factors associated with the onset and progression of keratoconus. It would be valuable for both eye care professionals and patients in Saudi Arabia.
PATIENTS AND METHODS: Patients seeking eye care at Qassim University eye clinic were included in this study. Participants were divided into: cases (with keratoconus) and control (without keratoconus but with other ocular problems). Keratoconus diagnoses of the participants were made by the attending optometrists or ophthalmologists. Multivariate logistic analyses were performed to identify the risk factors for keratoconus. Moreover, by performing logistic regression and CART analysis, supervised learning algorithms were developed to predict the likelihood of keratoconus based on the risk factors.
RESULTS: There were 75 keratoconus patients and 75 control. The CART model to predict the chances of keratoconus occurrence has an accuracy of 73%. Our prediction model can be a baseline model for future risk factor analysis studies that will be done in the Middle Eastern region. The models can be better trained by refining the risk factor quality and also by increasing the keratoconus population in the study. Including clinical parameters in the prediction models would result in complex as well as models with better prediction accuracy.
CONCLUSIONS: Clinical ocular parameters including the corneal topographic variables have to be obtained to better correlate the risk factors with specific changes or the subtypes of the keratoconus. Complex diseases like keratoconus require machine learning models apart from statistical analysis for association and causation. Machine learning models would not only predict the disease but also provide insight into how the risk factors interact with each other.
PMID:36314304 | DOI:10.26355/eurrev_202210_30003
