Assessing the Relationship Between Electronic Medical Record (EMR) Generated QTc Alerts and Cancer Patient Mortality

BACKGROUND: EMR generated drug associated QTc alerts are generated frequently in cancer patient populations and in the last few years their annual frequency has outgrown the number of unique patient visits at our institution. Anecdotally, they are largely ignored by providers, contributing to provider cognitive burnout and alert fatigue. While they may be considered nuisance alerts, the EMR collects rich data on the circumstances underlying the alert, the patients impacted, and their outcomes. By querying the EMR, we aimed to risk stratify patients by cancer site and demographic factors and provide meaning to these alerts allowing providers to incorporate the information we have provided in clinical decision making regarding the care of cancer patients. Our project is the first large-scale analysis of EMR generated QTc prolongation alerts at a tertiary referral center in the United States. OBJECTIVE: Acute mortality of cancer patients varies significantly by cancer site and demographic factors following EMR generated drug associated QTc interval prolongation alerts. METHODS: UT-Southwestern's EMR was queried to identify all patients between 10/04/2005 (the date of the 1st recorded alert) and 08/13/2019 over 18-years of age with a diagnosis of cancer and EMR generated drug associated QTc interval prolongation alerts yielding a sample of 19,223 patients. We collected the alert triggering medications, patient demographics, cancer site, and mortality data to identify the time between a patient's 1st alert and recorded death date. Rates of death by age, ethnicity, gender, race, number of alerts, and primary cancer site were identified in the following intervals: within 10 days, 11-180 days, and 181-365 days. Kaplan-Meier Overall Survival Analysis with Cox regression and multinomial multivariable analysis controlling for age, race, ethnicity, and gender, and median survival data analytic methods were used to identify if there were statistically significant differences in mortality at the pre-specified time points based on the above listed variables. Head and neck cancer patients were used as our reference group for comparison when analyzing mortality by primary cancer site. This group was chosen as their rates of within 10-day mortality closely aligned with those predicted by the null hypothesis that there would be no difference in mortality rates within 10-days across primary cancer sites used in our preliminary chi-squared goodness of fit model. Additionally, for patients with EKG's recorded within 10 days of their 1st alert, QTc intervals utilizing Bazett's correction algorithm were collected and the recorded interval of patients with deaths within 10 days was compared to those who were alive after 10 days, with participants separated by gender. RESULTS: Analysis of mortality from a patient's 1st QTc alert demonstrated statistically significant (p<0.05) higher risk of ≤10-day mortality in: patients with increasing number of QTc alerts, particularly patients with 6-10, HR=1.67 or >11 HR=1.88 alerts. Additionally, increased ≤10-day mortality was demonstrated in cancer patients with male gender, (female patients demonstrated a HR=0.83, compared to a male reference group), African American, HR=1.26, or other/unknown race, HR=1.29, and age >70, hazard ratio (HR)=2.32. Chi-squared goodness of fit testing identified head and neck cancer patients had a ratio of expected (by chance) to observed mortality of 0.98 and given the close concordance with our null model, were used as our reference group in multivariable analysis. Compared to a head and neck cancer patient baseline, significantly increased ≤10-day mortality was seen in: GI, HR=2.16; lung, HR=1.94; blood, HR=1.46; soft tissue, HR=2.26; and female genital, HR=1.55 cancers. Male genital, HR=0.39; breast, HR=0.57; and endocrine, HR=0.48 cancers had significantly decreased ≤10-day mortality. Of patients with EKG's, male patients who died ≤10-days of their 1st alert had significantly longer QTc intervals than males who survived to day 11 (469 vs 450 milliseconds, p<0.0001). In patients with any cancer and a QTC alert, 0.01 (male genital)-2.50% (GI) died ≤10 days of their 1st alert. The majority of deaths recorded <1 year after a patient's 1st alert occurred between 11-180 days during which 2.93 (male genital)-23.8% (GI) of the total sample with that primary cancer site diagnosis died. 63.2 (GI)-95.9% (endocrine) of cancer patients were alive >1 year after their 1st alert. CONCLUSION: Our research supports the anecdotal suggestion that very few patients die within 10 days of their initial QTc alert, suggesting that in many cases they function as distractions, especially in male genital, breast, and endocrine cancer patients and females or individuals <50 years of age. However, they may also identify patients at imminent risk of death, particularly those with lung, soft tissue, GI, blood, and female genital cancers, or males, African Americans, and individuals >70 years of age. Further, our analysis shows that QTc alerts may be a negative prognostic factor as the patients with more alerts (>5) have greater ≤10-day mortality rates. Additionally, of the patients who die within 365 days of their first alert the vast majority across cancer sites die between 11-180 days.