Outcomes of Citalopram Dosage Risk Mitigation in a Veteran Population

As shown in Figure 1, the cumulative proportion of patients receiving citalopram prescriptions for dosages ≤40 mg/day increased rapidly after the safety communications were issued. Substantial numbers of subjects were censored after a 30-day gap in citalopram resupply that may have resulted in part from unrecognized citalopram dosage reductions or discontinuation. Of those who filled citalopram prescriptions for ≤40 mg/day during the year of follow-up (N=18,407), 3.9% (cumulative proportion; N=519) subsequently filled prescriptions for >40 mg/day.

Selected characteristics of subjects who continued to have active citalopram prescriptions for >40 mg/day after the FDA and VA safety communications were issued (N=35,848), and who did (N=18,407) or did not (N=17,441) have their dosage reduced thereafter, are described in Table 1. Their last prescribed dosages before the safety communications were issued averaged 64 mg/day (SD=8.6) and 63 mg/day (SD=7.9), respectively. The two groups had a similar number of VA health care encounters and diagnoses, as well as medications related to mental health or cardiac disorders. All 109 baseline variables, including standardized differences, are summarized in the data supplement in Table S1 and, after matching on propensity scores, in Table S2. The majority of standardized differences were negligible (<10%) before matching on propensity scores, and only one slightly exceeded 10% in the matched groups.

The unadjusted incidence of all-cause hospitalizations or deaths was more than 2.5 times higher after citalopram dosages were reduced to ≤40 mg/day (Table 2). The unadjusted hazard ratio for the primary all-cause outcome was 4.4 (95% CI=4.0–4.8; p<0.001). The mortality rate continued to be 1.9 per 100 person-years after dosages were reduced, and the adjusted hazard ratio was 1.0 (95% CI=0.8–1.3; p=0.97). Therefore, the higher risk observed for the primary composite outcome was driven by a higher risk of hospitalizations (unadjusted hazard ratio=5.3, 95% CI=4.6–6.0; p<0.001), with deaths analyzed as a competing risk. The only remarkable difference in principal discharge diagnoses was in the mental health disorder category, which accounted for 20.1% (N=401) of the 1,997 all-cause hospitalizations that occurred after the dosage was reduced to ≤40 mg/day, compared with 14.0% (N=132) of the 944 all-cause hospitalizations that occurred when higher dosages were continued. PTSD accounted for nearly half of the difference (6.6% [N=132] and 3.8% [N=36], respectively). There was little difference in the other 20 major diagnostic categories of all-cause hospitalizations, including circulatory system disorders (19.6% compared with 18.8%).

The unadjusted incidence rates for the secondary composite outcome of hospitalizations for depression, diagnoses of self-injury, or death were also higher after citalopram dosages were reduced (Table 2). The unadjusted hazard ratio was 9.8 (95% CI=6.2–15.5; p<0.001) when deaths were analyzed as a competing risk rather than as an outcome event. The composite unadjusted incidence of hospitalizations for cardiac arrhythmias or all-cause deaths (Table 2) was not lower after dosages of citalopram were reduced, and neither was the unadjusted hazard ratio (hazard ratio=1.3, 95% CI=1.0–1.6; p=0.02). The unadjusted hazard ratio of these more specific arrhythmia-related hospitalizations was 7.1 (95% CI=2.9–17.1; p<0.001) when deaths were analyzed as a competing risk rather than as an outcome. The increased risk of this composite outcome after citalopram dosages were reduced came from more hospitalizations with a principal discharge diagnosis of syncope or collapse. Excluding these less specific diagnoses, the unadjusted incidence of hospitalizations for cardiac arrhythmias or all-cause mortality was approximately 2.1 per 100 person-years, regardless of whether high citalopram dosages were reduced or continued.

The multivariable regression outcome model and the propensity matching are presented in Table 3. Neither had a substantial effect on the hazard ratio estimates for the primary or secondary composite outcomes. Allowing a 90-day gap, rather than a 30-day gap, for receipt of another citalopram supply allowed us to include many more subjects and more follow-up time in the analysis, but the adjusted time frame did not substantially alter the estimated adjusted hazard ratios (Table 3).

Reductions of prescribed dosages of citalopram to a new safety limit of 40 mg/day were associated with a substantial increase in all-cause and depression-related hospitalizations. Rapidly reducing dosages from >60 mg/day to ≤40 mg/day may have precipitated worsening symptoms of depression, PTSD, and other mental health disorders. On the other hand, there was no noticeable reduction in hospitalizations for cardiac arrhythmias or all-cause mortality. Although citalopram can prolong the QT interval in a dose-dependent manner, the mortality risk may have been too low to detect even in this large cohort of patients who had survived high dosages of citalopram (26, 27). Previous research has found no increased risk of ventricular arrhythmia or death when citalopram dosages >40 mg/day were compared with lower dosages (28). In the present study, deaths that were directly or possibly indirectly related to worsening mental health (e.g., poor adherence to cardiovascular medications and self-care) may have offset deaths from cardiac arrhythmias incurred by a continuance of higher prescription dosages. The safety warning may have prompted electrocardiograms that led to dosage reductions that prevented some hospitalizations for cardiac arrhythmias and deaths. However, the net effect of the large number of citalopram dosage reductions that occurred shortly after the safety communications were issued appeared to manifest as an increase in hospitalizations for mental health disorders. The outcomes may differ if high citalopram dosages are more selectively reduced for patients at greatest risk of clinically significant QT interval prolongation, although the FDA dosage limit was applicable to all patients treated with citalopram.

Citalopram prescriptions for ≤40 mg/day increased impressively shortly after the safety warnings were issued by the FDA and disseminated to all VA health care systems by the VA Center for Medication Safety. There were relatively few dosage reductions during the weeks before the new dosage limit was issued, and these dosage reductions were excluded from the outcome analyses. Baseline clinical characteristics were similar in the groups that did or did not have their prescribed citalopram dosages reduced. Therefore, most of the observed citalopram dosage reductions were presumably prompted by the new safety limit. If worsening mental health prompted outpatient visits when citalopram dosages were reduced, it seems more likely that these dosage reductions would be prompted by the dosage limit rather than the patients’ worsening mental health, although clinicians and patients may have thought the high dosages were no longer effective. If so, worsening depression prior to dosage reductions may have led to some hospitalizations and deaths following dosage reduction.

Outcome events could be misclassified into the group that continued on higher dosages if the citalopram dosages were reduced or discontinued but not reflected in the medical records. To limit the effects of this type of misclassification, follow-up was censored as soon as there was a 30-day gap in the dispensed citalopram supply. The results were essentially the same when a 90-day gap was allowed before follow-up was censored. Furthermore, this type of misclassification would place a bias against finding a higher incidence of hospitalizations or death following citalopram dosage reductions.

We used multivariable Cox regression models of the composite outcome and the propensity to reduce the dosage of citalopram to try to reduce bias from differences in measured variables. However, we could not exclude the possibility of residual confounding based on differences in unmeasured confounders (24). Bias arising from differences in the unmeasured severity of depression or PTSD may not have been controlled via its relationship to the prescribed dosage of citalopram, use of other prescription medications, use of VA health care, and so forth. The risk of QT interval prolongation may not have been controlled by incorporating into the regression models other cardiac dysrhythmias and conditions, other medications that can alter the QT interval, and the amount of previous exposure to high dosages of citalopram. If some of the variables in the propensity model were related to the likelihood of citalopram dosage reductions but not to the study outcome, their inclusion could have increased bias in the estimated hazard ratios (25). Nevertheless, we did not find substantial differences in the large number of measured confounders, and accordingly neither regression method we used to try to reduce bias substantially altered the estimated hazard ratios. Any residual bias due to differences in unmeasured confounders would need to be much larger to affect the estimated hazard ratios. Thus, residual confounding bias most likely does not entirely explain the observed association between the citalopram dosage reductions and the adverse patient outcomes.

Our analysis is also limited by lack of data for some non-VA hospitalizations. We cannot exclude the possibility that the observed higher relative risk of VA hospitalizations was because of less utilization of non-VA hospital care after citalopram dosages were reduced, compared with when high VA dosages of citalopram continued to be prescribed. We did not pursue Medicare data to try to explore this issue because the majority of the at-risk population was younger than age 65, and many subjects had private health care insurance.

Presumably, any diagnostic misclassification in the VA data would not be related to whether clinicians decided to prescribe lower dosages of citalopram, and it certainly would not affect the primary composite outcome of all-cause hospitalizations or death. Counting hospitalizations that were not directly or indirectly associated with the dosage reductions as outcomes would be expected to reduce the estimated hazard ratio, assuming unrelated hospitalizations were equally likely when citalopram dosages were or were not reduced as they appeared to be when major discharge diagnosis categories were examined.

A national VA medication safety communication helped ensure that VA health care providers were aware of the new FDA dosage limit for citalopram. Indeed, the majority of veterans who continued to receive citalopram prescriptions had lower dosages prescribed soon after the warning was disseminated to all VA health care systems. Many VA health care providers may have felt obligated to prescribe dosages that did not exceed 40 mg/day.

In conclusion, reduction of prescribed citalopram dosages to a new safety limit was associated with a higher rate of hospitalization in a large patient population who had been treated with substantially higher dosages. We hope this research will encourage more empirical studies of patient outcomes after risk mitigation initiatives and thereby help to improve medication safety warnings and clinical risk management.

Andrea Cutting, M.S., from the Center for Chronic Disease Outcomes Research at the Minneapolis VA Health Care System, extracted the data from the VA Corporate Data Warehouse, and under the direction of the principal investigator, processed the data into an analyzable data file.