Risk score development and data collection
Based on sensitivity and specificity results as well as the ability to identify risk factors in the EHR, we built our risk score using the components of the RISQ-PATH model5 by Vandael and colleagues, with a few modifications to the lookback timeframe and methods of extracting the data. Data extracted from the EHR and added to the risk score included age, sex, body mass index (BMI, kg/m2), cardiovascular history, liver failure, neurological disorders, thyroid disturbances, potassium (mmol/L), calcium (mmol/L), C-reactive protein (CRP; mg/L), estimated glomerular filtration rate (eGFR; ml/min), and medications associated with a risk of QTc-prolongation and/or TdP (lists 1, 2 and 3 of CredibleMeds). CredibleMeds10 is an American organization that provides lists of drugs associated with a risk of QTc-prolongation and TdP (list 1: drugs with a known risk for TdP; list 2: drugs with a possible risk for TdP; list 3: drugs with a conditional risk for TdP; list 4: drugs to be avoided by patients with congenital long QT syndrome, including all drugs in lists 1, 2 and 3, in addition to drugs with special risk).
Table 1 lists the components included in our preliminary risk score. Each risk factor is allocated a certain number of points contributing to the total score, with a risk cutoff of 10 points. However, due to limitations in the EHR, the score was unable to automatically extract a previous ECG reading from the patient’s chart. This information was collected manually through chart review and was not automatically included in the risk score built in the EHR. The risk score and point allocation were discussed with the cardiology department at MD Anderson Cancer Center, who were in agreement with our methods and risk factors. Details on how we extracted these components from the EHR are listed in table A.1 (Supplementary appendix).
During the validation phase, each time a new QTc-prolonging medication was signed or verified, the modified QTc-prolongation risk score was calculated in the EHR and was saved in an archived file every 12 hours, accessed solely by the investigators. The custom alert associated with the score calculation was not visible during the validation phase. The current DDI warnings that triggered at the time of risk score calculation were saved in a separate file and retrieved by the investigators. A chart review was conducted on patients who met the inclusion criteria to validate the risk score and compare it to the current DDI warnings. QTc intervals were collected manually using baseline and follow-up 12-lead ECG readings, and were corrected using the Fridericia formula (QTcF). We defined a prolonged QTc interval as ≥ 450 ms for males, ≥ 470 ms for females, or an increase from baseline by ≥ 30 ms.
Table 1. Preliminary QTc-prolongation risk score at MD Anderson Cancer Center