DISCUSSION
In this study, we developed a preliminary QTc-prolongation risk score with a sensitivity of 81%, a specificity of 35%, a PPV of 32%, and an NPV of 84%. Compared with the DDI warnings, our full dataset achieved better sensitivity and NPV, but that resulted in less specificity and PPV. Our multiple logistic regressions allowed us to achieve a sensitivity of 68%, specificity of 66%, a PPV of 45% and an NPV of 83% with an optimized risk score and a cutoff of 5 points. Our aim of reducing false positive alerts by 20% was not achieved, however, many of the DDI warnings were filtered, and those that actually triggered and were included in our calculations were mostly severe, with a high risk of QTc prolongation. On the contrary, medications triggering our risk score calculation and custom alert were based on lists 1, 2 and 3 of CredibleMeds, making it easier to achieve a greater number of custom alerts than DDI warnings. In addition, DDI warnings were only identified in patients who already had the risk score calculated and a custom alert generated. Vandael and colleagues included patients from various hospitals and wards including cardiology, internal medicine and others.4 Patients at our institution, however, were all cancer patients with different characteristics and risk factors affected by various therapies and comorbidities. Moreover, the study by Vandael and colleagues did not compare the risk score to any existing warning, unlike our study which compared the risk score to an existing DDI warning, aiming to reduce alert burden and fatigue. The study by Tisdale and colleagues included only patients on cardiac units, and did not compare the risk score to another existing warning. However, it had a pre and a post-implementation phase which revealed that incorporating a validated risk score influenced the prescribing of noncardiac QTc interval-prolonging drugs and reduced the risk of QTc-prolongation.6