Combined electronic medical records and gene polymorphism characteristics to establish an anti-tuberculosis drug-induced hepatic injury (ATDH) prediction model and evaluate the prediction value.

BACKGROUND

Anti-tuberculosis drug-induced hepatic injury (ATDH) lacks specific diagnostic markers. The characteristics of gene polymorphisms have been preliminarily used for the risk classification of ATDH, and the activation of Pregnane X receptor/aminole-vulinic synthase-1/forkhead box O1 (PXR/ALAS1/FOXO1) axis is closely related to ATDH. Therefore, we consider combining general clinical features of the electronic medical record, laboratory indications, and genetic features of key genes in this axis for predictive model construction to help early clinical diagnosis and treatment.

METHODS

The general characteristics derived from the Hospital Information System (HIS) medical record system, the biochemical tests and hematology tests were detected by Roche automatic biochemical immunoassay analyzer cobas8000 and Sysmex automatic hemocytometer XE2100. The single nucleotide polymorphisms (SNPs) genotyping work was conducted with a custom-designed 48-plex SNP scan TM Kit. A total of 746 cases were included which were divided into training set and validation set according to the ratio of 3:2 randomly. Taking the occurrence of confirmed ATDH as the outcome variable, lasso regression and logistic regression were used to identify the predictors preliminarily. alanine aminotransferase, aspartate aminotransferase, monocyte, uric acid, albumin, fever, the polymorphisms of rs4435111 (FOXO1) and rs3814055 (PXR) were chosen from all variables to combine the predictive model. The goodness of fit, predictive efficacy, discrimination, and consistency, and clinical decision curve analysis was used to assess the clinical applicability of the models.

RESULTS

The best model had a discriminant efficacy C-index of 0.8164, a sensitivity of 34.25%, specificity of 97.99%, a positive predictive value of 78.13% and negative predictive value of 87.69%, the two-tailed value of Spiegelhalter Z test of consistency test S:P =0.896, maximum absolute difference Emax =0.147, and average absolute difference Eave =0.017. In the validation set, performance was close. The clinical decision curve showed the clinical applicability of the prediction model when the prediction risk threshold was between 0.1 and 0.8.

CONCLUSIONS

The ATDH prediction model was constructed using a machine learning approach, combining general characteristics of the study population, laboratory indications, and SNP features of and genes with good fit and certain predictive value, and has potential and value for clinical application.