Plasma D-dimer Can Effectively Predict the Prospective Occurrence of Ascites in Advanced Schistosomiasis Japonica Patients

Machine learning algorithms to predict the 1 year unfavourable prognosis for advanced schistosomiasis

Short-term prognosis of advanced schistosomiasis has not been well studied. We aimed to construct prognostic models using machine learning algorithms and to identify the most important predictors by utilising routinely available data under the government medical assistance programme. An established database of advanced schistosomiasis in Hunan, China was utilised for analysis. A total of 9541 patients for the period from January 2008 to December 2018 were enrolled in this study. Candidate predictors were selected from demographics, clinical features, medical examinations and test results. We applied five machine learning algorithms to construct 1 year prognostic models: logistic regression (LR), decision tree (DT), random forest (RF), artificial neural network (ANN) and extreme gradient boosting (XGBoost). An area under the receiver operating characteristic curve (AUC) was used to evaluate the model performance. The important predictors of the optimal model for unfavourable prognosis within 1 year were identified and ranked. There were 1249 (13.1%) cases having unfavourable prognoses within 1 year of discharge. The mean age of all participants was 61.94 years, of whom 70.9% were male. In general, XGBoost showed the best predictive performance with the highest AUC (0.846; 95% confidence interval (CI): 0.821, 0.871), compared with LR (0.798; 95% CI: 0.770, 0.827), DT (0.766; 95% CI: 0.733, 0.800), RF (0.823; 95% CI: 0.796, 0.851), and ANN (0.806; 95% CI: 0.778, 0.835). Five most important predictors identified by XGBoost were ascitic fluid volume, haemoglobin (HB), total bilirubin (TB), albumin (ALB), and platelets (PT). We proposed XGBoost as the best algorithm for the evaluation of a 1 year prognosis of advanced schistosomiasis. It is considered to be a simple and useful tool for the short-term prediction of an unfavourable prognosis for advanced schistosomiasis in clinical settings.

Derivation and external validation of a model to predict 2-year mortality risk of patients with advanced schistosomiasis after discharge

To date, no risk prediction tools have been developed to identify high mortality risk of patients with advanced schistosomiasis within 2 years after discharge. We aim to derive and validate a risk prediction model to be applied in clinical practice. The risk prediction model was derived from 1487 patients from Jingzhou and externally validated by 723 patients of Huangshi, two prefecture-level cities in Hubei province, China (from September 2014 to January 2015, with follow-up to January 2017). The baseline variables were collected. The mean age [SD] was 62.89 [10.38] years for the derivation cohort and 62.95 [12.22] years for the external validation cohort. The females accounted for 36.3% and 43.7% of the derivation and validation cohorts, respectively. 8.27% patients (123/1487) in the derivation cohort and 7.75% patients (56/723) in the external validation cohort died within 2 years after discharge. We constructed 4 models based on the 7 selected variables: age, clinical classification, serum direct bilirubin (DBil), aspartate aminotransferase (AST), alkaline phosphatase (ALP), hepatitis B surface antigen (HBsAg), alpha fetoprotein (AFP) at admission. In the external validation cohort, the multivariate model including 7 variables had a C statistic of 0.717 (95% CI, 0.646–0.788) and improved integrated discrimination improvement (IDI) value and net reclassification improvement (NRI) value compared to the other reduced models.

Therefore, a multivariate model was developed to predict the 2-year mortality risk for patients with advanced schistosomiasis after discharge. It could also help guide follow-up, aid prognostic assessment and inform resource allocation.