Piecing Together the Risk of Sudden Cardiac Death on Dialysis

According to ESRD Death Notification forms, 40% of hemodialysis (HD) patient deaths with a known cause and 30% of all HD patient deaths in 2014–2016 were primarily due to cardiac arrest or arrhythmia.¹ A decade earlier, corresponding shares were 36% and 27%.² Amid a secular trend toward improving survival among patients on HD in the United States, proportionate mortality due to sudden cardiac death (SCD) has thus increased modestly. Very likely, the road to better survival on HD passes through the valley of SCD.

Prolongation of the QT-interval is associated with higher risks of torsades de pointes and SCD, and, importantly, prolongation may be drug-induced. Well known drug groups that prolong the QT-interval include macrolide antibiotics, class 3 antiarrhythmics, and antipsychotics. Another such drug is citalopram, a selective serotonin reuptake inhibitor (SSRI). In 2011, the US Food and Drug Administration warned that daily doses exceeding 40 mg should not be used³; in 2012, the US Food and Drug Administration warned that daily doses exceeding 20 mg should not be used in patients older than 60 years.⁴ Escitalopram, the S-enantiomer of citalopram, also prolongs the QT-interval, although to a lesser degree than citalopram.⁴

Against this backdrop, Assimon et al.⁵ have adeptly compared the risk of SCD in patients on HD who were dispensed either citalopram or escitalopram versus other SSRIs (fluoxetine, fluvoxamine, paroxetine, and sertraline). Using Medicare claims data, the investigators identified over 65,000 patients who initiated SSRI exposure between 2007 and 2014. Slightly less than half were dispensed either citalopram or escitalopram, and with these drugs, the risk-adjusted hazard ratio of SCD was 1.18 (95% confidence interval, 1.05 to 1.31), relative to other SSRIs. This translates to an SSRI user population-attributable fraction of 7.8%. Provocatively, among 8% of patients with a diagnosed conduction disorder, the hazard ratio of SCD was 1.47, and among 48% of patients with use of at least one (other) drug that prolongs the QT-interval, the hazard ratio was 1.29.

This is a very well executed pharmacoepidemiologic study. The carefulness of the study design, including the nuanced SSRI discontinuation definition, the thoroughness of risk adjustment, and the sensitivity analyses, including analysis of a negative control outcome (non-SCD), all strengthen the investigators’ conclusion. However, this is an observational study of claims data. Inescapable limitations exist; several merit debate.

First, the 180-day washout period has advantages and disadvantages. The period obviously permits identification of new exposure to an SSRI and baseline comorbidity during a fixed interval. However, it appears to come at a cost. Nearly 60% of patients who were dispensed an SSRI did not meet inclusion criteria: treatment with in-center HD and continuous enrollment in Medicare Parts A, B, and D during the washout period. Excluded patients must have comprised (1) patients on incident dialysis who were dispensed an SSRI; (2) patients on dialysis who were enrolled in Medicare Advantage (their prescription drug claims are apparent in the US Renal Data System database, but their inpatient and outpatient care claims are not); and (3) patients on dialysis who were newly enrolled in Medicare Parts A, B, and/or D, thus lacking 6 months of continuous enrollment before initial SSRI exposure. The reported hazard ratio of 1.18 may be internally valid, but whether it would change upon adding over 100,000 patients is unknown. Second, patients could have used SSRIs before the washout period. All exposure to SSRIs before this period necessarily resulted in zero SCDs, whether with citalopram, escitalopram, or other SSRIs. If prewashout patient-years of SSRI exposure were included in the study, the reported hazard ratio of 1.18 might be attenuated. Third, sudden discontinuation of SSRI exposure was the norm. In fact, median duration of follow-up was 37 days; this is likely meaningful because a 30-day supply of an SSRI and a 7-day grace period (which defines discontinuation) sums to 37 days. In other words, a large proportion of the cohort received 30 days of supply and never refilled. This raises an important question: what proportion of the cohort never consumed the dispensed SSRI? If the proportion were high, then some part of the reported hazard ratio would reflect an old foe: unmeasured confounding. In every subgroup, the unadjusted hazard ratio of SCD with either citalopram or escitalopram was larger than the risk-adjusted hazard ratio, thus indicating some channeling of higher-risk patients toward both citalopram and escitalopram. Collectively, these limitations suggest that the SSRI user population-attributable fraction of SCD due to citalopram and escitalopram may be lower than 7.8%.

Nevertheless, these results should be interpreted in the context of the background treatment, thrice-weekly HD. Fluid overload, as measured by bioimpedance spectroscopy, is present in nearly half of European patients on HD.⁶ The cyclical nature of volume loading between treatments and unloading (i.e., ultrafiltration) during treatment is correlated with several negative patterns, including steadily increasing right ventricular pressure during interdialytic gaps,⁷ high risk of bradycardia during the final 12 hours of each interdialytic gap,⁸ and high risk of atrial fibrillation during HD.⁷ Furthermore, many patients use dialyzable β-blockers (e.g., metoprolol), thereby allowing increased risk of arrhythmia after HD concludes.⁹ In general, the hours before, during, and after each HD treatment constitute recurrent, high-risk periods of arrhythmia. The addition of one or more medications that prolong the QT-interval may be enough to tip the scale from subclinical arrhythmia to SCD. Whether those medication pose the same risk in patients with better volume control (i.e., less time-integrated fluid load and lower ultrafiltration rate) remains to be studied.

According to Medicare Part D data, 30% of patients on dialysis are dispensed an antidepressant.¹⁰ If every patient who were prescribed an SSRI were dispensed paroxetine, which appears to confer the lowest risk of QT-interval prolongation, then only a small proportion of SCDs would likely be prevented, considering the data that Assimon et al. have reported. This observation, however, does not diminish the importance of the study. Addressing every component of SCD risk in patients on HD, from the dialysis prescription itself to the array of medications that can disturb heart rhythm, is likely necessary to improve patient survival. There is no single solution.

Disclosures

The author E.D.W. is also employed by NxStage Medical (Lawrence, Massachusetts), but reports no conflicts of interest with the content.