Multiple logistic regression
Our preliminary risk score had an area under the Receiver Operating Characteristic (ROC)-curve of 0.848 to predict a QTc ≥ 450 ms for males, ≥ 470 ms for females, or an increase from baseline ECG by ≥ 30 ms. The results of the multiple logistic regression are displayed in table 3. Significant p-values are shown in bold.
Panel A includes the full dataset of all components in our preliminary score, whereas panel B includes the components after exclusion of selected insignificant or unimportant factors.
In the full dataset, significant risk factors included age ≥ 65 years, female sex, ischemic cardiomyopathy and/or hypertension, arrhythmia, potassium ≤ 3.5 mmol/L, CRP > 5 mg/L, drugs from CredibleMeds list 1 and 2. After excluding BMI ≥ 30 kg/m2, liver failure, diabetes, neurological disease, CRP > 5 mg/L, eGFR ≤ 30 ml/min, and drugs from CredibleMeds list 3, panel B logistic regression resulted in similar significant factors, except for arrhythmia which was shown to be no longer significant. However, we decided to keep it in our risk score because we believed it would largely determine the risk for QTc prolongation. Although the CRP component was shown to be significant in our full dataset, we excluded it from panel B after discussions with cardiology, with the notion that most patients at our institution have elevated CRP values. We retained calcium and thyroid disturbances in the updated dataset since many patients with QTc-prolongation had low calcium levels and thyroid disorders.
Table 3. Results of the multiple logistic regression