A predictive model of response to metoprolol in children and adolescents with postural tachycardia syndrome

Background: The present work was designed to explore whether electrocardiogram (ECG) index-based models could predict the effectiveness of metoprolol therapy in pediatric patients with postural tachycardia syndrome (POTS).

Methods: This study consisted of a training set and an external validation set. Children and adolescents with POTS who were given metoprolol treatment were enrolled, and after follow-up, they were grouped into non-responders and responders depending on the efficacy of metoprolol. The difference in pre-treatment baseline ECG indicators was analyzed between the two groups in the training set. Binary logistic regression analysis was further conducted on the association between significantly different baseline variables and therapeutic efficacy. Nomogram models were established to predict therapeutic response to metoprolol. The receiver-operating characteristic curve (ROC), calibration, and internal validation were used to evaluate the prediction model. The predictive ability of the model was validated in the external validation set.

Results: Of the 95 enrolled patients, 65 responded to metoprolol treatment, and 30 failed to respond. In the responders, the maximum value of the P wave after correction (Pcmax), P wave dispersion (Pd), Pd after correction (Pcd), QT interval dispersion (QTd), QTd after correction (QTcd), maximum T-peak-to-T-end interval (Tpemax), and T-peak-to-T-end interval dispersion (Tped) were prolonged (all P < 0.01), and the P wave amplitude was increased (P < 0.05) compared with those of the non-responders. In contrast, the minimum value of the P wave duration after correction (Pcmin), the minimum value of the QT interval after correction (QTcmin), and the minimum T-peak-to-T-end interval (Tpemin) in the responders were shorter (P < 0.01, < 0.01 and < 0.01, respectively) than those in the non-responders. The above indicators were screened based on the clinical significance and multicollinearity analysis to construct a binary logistic regression. As a result, pre-treatment Pcmax, QTcmin, and Tped were identified as significantly associated factors that could be combined to provide an accurate prediction of the therapeutic response to metoprolol among the study subjects, yielding good discrimination [area under curve (AUC) = 0.970, 95% confidence interval (CI) 0.942-0.998] with a predictive sensitivity of 93.8%, specificity of 90.0%, good calibration, and corrected C-index of 0.961. In addition, the calibration curve and standard curve had a good fit. The accuracy of internal validation with bootstrap repeated sampling was 0.902. In contrast, the kappa value was 0.769, indicating satisfactory agreement between the predictive model and the results from the actual observations. In the external validation set, the AUC for the prediction model was 0.895, and the sensitivity and specificity were 90.9% and 95.0%, respectively.

Conclusions: A high-precision predictive model was successfully developed and externally validated. It had an excellent predictive value of the therapeutic effect of metoprolol on POTS among children and adolescents.