Construction and internal validation of a predictive model for risk of gastrointestinal bleeding in children with abdominal Henoch-Schönlein purpura: A single-center retrospective case-control study

Abstract

Early identification of gastrointestinal (GI) bleeding in children with abdominal Henoch-Schönlein purpura (HSP) is essential for their subsequent treatment, and a risk prediction model for GI bleeding in abdominal HSP was constructed in this study to assist physicians in their decision-making. In a single-center retrospective study, the children collected were divided into two parts, a training set and a validation set, according to the time of admission. In the training set, univariate analysis was performed to compare demographic data and laboratory tests between the two groups of children with GI and non-GI bleeding, and the independent risk factors were derived using binary logistic equations to develop a scoring model for predicting GI bleeding in children by odds ratio (OR) values and receiver operating characteristic curves. The scoring model was then internally validated in validation set. The results showed that there were 11 indicators were statistically different between the two groups in the training set, including white blood cells, neutrophil-to-lymphocyte ratio, platelets, eosinophils (EO), high sensitivity C-reactive protein (hsCRP), activated partial thromboplastin time (APTT), sodium, potassium (K), albumin (ALB), Total bilirubin, and Immunoglobulin E (IgE) in the univariate analysis. Among them, the independent risk factors for GI bleeding included the six indicators of EO ≤ 0.045×10^9/L, hsCRP ≥ 14.5 mg/L, APTT ≤ 28.1 s, K ≥ 4.18 mmol/L, ALB ≤ 40.6 g/L, and IgE ≥ 136 ng/mL. According to the OR values, where EO ≤ 0.045 ×10^9/L, hsCRP ≥ 14.5 mg/L, APTT ≤ 28.1 s, ALB ≤ 40.6 g/L each scored 3 points, K ≥ 4.18 mmol/L, IgE ≥ 136 ng/mL each scored 2 points, and the total score was 0-16 points. The sensitivity and specificity of predicting GI bleeding were 88.7% and 64.2%, respectively, when the child scored ≥ 7 points. In the validation set, the sensitivity, specificity and accuracy of the model in predicting GI bleeding were 77.4%, 74.5% and 75.2%, respectively. In conclusion, the construction of a scoring model to predict the risk of GI bleeding from abdominal HSP would greatly assist pediatricians in predicting and identifying children at high risk for GI bleeding at an early stage.

Keywords: Henoch-Schönlein purpura; IgA vasculitis; binary logistic regression; gastrointestinal bleeding; gastrointestinal involvement; predictive scoring model.

Figure 1

Flow chart of study subject inclusion. GI, gastrointestinal; HSP, Henoch-Schönlein purpura.

Figure 2

ROC curve of scoring model and single indicator. (A) ROC curve of the scoring model. The AUC (95% CI) for predicting GI bleeding was 0.832 (0.779, 0.884), P < 0.001. (B) ROC curve of the single indicator. A comparison of the AUC of the ROC curves for different indicators showed that the AUC of the scoring model was higher than that of the single indicator. The AUC (95% CI) for predicted probability was 0.833 (0.781, 0.885), P < 0.001, and the difference in AUC values between the scoring model and the predicted probability was not significant. ROC, receiver operating characteristic; AUC, area under the curve; CI, confidence interval.