Development and validation of a machine learning model to predict prognosis in HIV-negative cryptococcal meningitis patients: A multicentre retrospective study

Abstract Background: An increasing number of HIV-negative cryptococcal meningitis (CM) patients have been reported with fatality approaching 30%.At present, HIV-negative CM patients are stratified according to clinical guidelines and clinical experience for individualized treatment, but the effect seems to be not ideal in clinical practice. Therefore, an accurate model that predict the prognosis for HIV-negative CM patients is needed to provide reference for precision treatment. Methods: This retrospective study involved 490 HIV-negative CM patients diagnosed between January 1, 1998, and March 31, 2022, by neurologists from 3 tertiary Chinese centres. Prognosis was evaluated at 10 weeks after the initiation of antifungal therapy. We used least absolute shrinkage and selection operator (LASSO) for feature filtering and developed a machine learning (ML) model to predict the prognosis in HIV-negative CM patients. Fifty-six patients from 2 other hospitals were analysed for external validation. An artificial intelligence (AI)-based detection model was also developed to automate the rapid counting of microscopic cryptococcal counts. Results:The final prediction model for HIV-negative CM patients comprised 8 variables: CSF cryptococcal count, CSF white blood cell (WBC), altered mental status, hearing impairment, CSF chloride levels, CSF opening pressure (OP), aspartate aminotransferase levels at admission and decreased rate of CSF cryptococcal count within 2 weeks after admission. The areas under the curve (AUCs) in the internal and external validation sets were 0.87 (95% CI 0.794-0.944) and 0.86 (95% CI 0.744-0.975), respectively. An AI model was trained to detect and count cryptococci, and the mean average precision (mAP) was 0.993. Additionally, an online and freely available platform for predicting prognosis and detecting and counting cryptococci in HIV-negative CM patients was established. Conclusions:A ML model for predicting prognosis in HIV-negative CM patients was built and validated, and the model might provide a reference for personalized treatment of HIV-negative CM patients. The change in the CSF cryptococcal count in the early phase of HIV-negative CM treatment can reflect the prognosis of the disease. In addition, utilizing AI to detect and count CSF cryptococci in HIV-negative CM patients can eliminate the interference of human factors in detecting cryptococci in CSF samples and reduce the workload of the examiner.