Sensitivity and specificity calculations
Sensitivity and specificity were first calculated for the custom alert
(Table A.5 in the supplementary appendix). Of all custom alerts, 161
(31%) had a risk score of < 10 points. A score of <
10 indicates low risk for QTc-prolongation, and a custom alert will not
fire. This cutoff led to a sensitivity of 81%, a specificity of 35%, a
PPV of 32% and an NPV of 84% to predict QTc prolongation.
When factoring in the additional points from manual chart review for
patients with QTc prolongation on previous ECG readings, sensitivity was
improved by 6%, but the specificity remained the same. The PPV and NPV
increased to 33% and 88%, respectively. Nevertheless, this component
is not readily available from the chart and the risk score will be
unable to automatically include it in the final calculation.
Five out of 100 patients had a follow up ECG that was prolonged per our
definition; yet their baseline ECG was more prolonged. This was
reflected in 24 risk scores which varied between 8.5 and 18.75 points
(median 14.5). We eliminated these scores and recalculated sensitivity
and specificity results, however, there was minimal to no change to the
values.
In addition to the custom alert sensitivity and specificity
calculations, we performed a calculation on the existing DDI warnings
(Table A.6 in the supplementary appendix). We found a sensitivity of
54%, a specificity of 65%, a PPV of 42% and an NPV of 75%. The
alerts that generated were the alerts that actually triggered for these
patients. Alerts not generated refer to those which were filtered at
that time, but were archived and identified by the investigators.