Diagnostic Accuracy and Prediction of COVID-19 Outcome Using Artificial Intelligence Based on Radiological Data, Clinical and Laboratory Parameter at Dr. Sardjito General Hospital, Yogyakarta
Harik Firman Thahadian 1 Ika Trisnawati1 Eko Budiono1 Bambang Sigit Riyanto 1 Heni Retnowulan1 Nur Rahmi Ananda1 Sarah Ulfa 1 Tani Prima Auladina1 Imam Manggalya Adhikara 1 Sumardi
1 Department of Internal Medicine of Faculty of Medicine, Public Health, and Nursing, Universitas Gadjah Mada Yogyakarta
Introduction: The application of the “color heat-map” method through identifying and analyzing chest X-ray images transferred into AI (artificial intelligence) to generate scores. The aim of this research to was to evaluate the diagnostic accuracy of artificial intelligence scores of Chest X-Ray for predicting the clinical outcome of COVID-19 patients and establishing a scoring system using predictor variables based on AI scoring data based on chest X-rays, clinical parameters, and laboratories of COVID-19 patients.
Methods: A retrospective study collected data from hospitalized COVID-19 patients in Dr. Sardjito General Hospital, Yogyakarta, between 2020 and 2022. The data collected is clinical, laboratory parameters, patient outcomes, and values from AI Chest X-Ray readings. Artificial intelligence was used to detect radiographic abnormalities using CAD4COVID-Xray software (Thirona, Nijmegen, Netherlands). Receiver operator curve (ROC) to evaluate the predictive value of the AI probability score and AI Affected Lung Area score. Multiple logistic regression analysis selected some variables to develop the scoring model.
Results: Four hundred forty-nine (449) patients were included in the study: 237 males (52,8%), median age 56 years (IQR = 45-65). ROC analysis shows that the AI probability score (AUC = 0.875, CI 95% 0.801-0.948) and AI ALA score (AUC = 0.836, CI 95% 0.766-0.906) have sufficient discrimination ability to determine the degree of disease severity of COVID-19 confirmed subjects. Multiple logistic regression analysis from clinical, laboratory, and clinical outcomes showed that this scoring system uses seven variables (5 clinical and two laboratory variables) and has a good prognostic ability to predict the severity of COVID-19 patients. Based on the stratification of scoring results, we found that the scoring value of low-risk patients (1-2 points) had a mortality proportion of 7.8.%, moderate risk ((3-5) points) had a mortality proportion of 38.7%, and high-risk ((6-9) points) had a mortality proportion of 76.9%.
Discussion: Using an AI-based score derived from radiographic, clinical, and laboratory parameters may be beneficial to estimate prognosis in confirmed COVID-19 patients.
