Strengths and limitations
We identified several strengths in our study. First, we performed a
thorough review of data and risk factors associated with QTc
prolongation, which allowed us to build and evaluate the risk score for
all components and their association with QTc prolongation in our
patient population. We used the corrected QTcF formula and identified
patients in sinus rhythm to better assess QTc prolongation on a follow
up ECG reading. Furthermore, the preliminary risk score that was built
in the EHR was capable of accurately extracting the data from the
patients’ charts. The ECG timeframe we set was appropriate to detect
medication-related QTc prolongation, eliminating other possible causes.
In addition, we compared our custom alert and DDI warning for the same
patients and within the same timeframe. This added to the accuracy of
our analysis which was based on rigorous data. Performing a multiple
logistic regression also helped us identify the risk factors that were
mostly associated with QTc prolongation, thus allowing us to re-arrange
the score points and increase our sensitivity and specificity results
through an optimized risk score. Moreover, using the missing-indicator
method in our regression analysis to deal with missing values
contributes to a more realistic view and broadens applicability.
Limitations in our study include the inability of the score to
automatically extract a QTc interval from the EHR to be included in our
risk score or to be used as a reference for clinical decision support.
In the future, implementing electronic waveform readings in the EHR
would allow for a more accurate assessment of alerts by incorporating
this component in the risk score. Additionally, the patients we included
already had a baseline ECG reading, which may suggest that these
patients were already at high risk for QTc prolongation and were being
followed closely by cardiology. Smoking is one of the risk factors for
QTc prolongation, however, our score could not reliably extract smoking
habits from the chart. As mentioned earlier, the DDI warning is based on
the FDB database, whereas the custom alert is based on CredibleMeds
list. Most of the DDI warnings were filtered at that time, and the
triggering medications were those with the greatest risk for QTc
prolongation. This explains the high specificity of the DDI warning
compared with the custom alert which included all drugs listed in
CredibleMeds lists 1, 2 and 3, regardless of severity. Furthermore, our
study was conducted in an oncology setting where patient characteristics
and risk factors differ and are affected by various therapies and
comorbidities, which makes extrapolating this data to other patient
populations challenging.