A colleague recently alerted me to a news story entitled “Antidepressants Linked to Doubling of C difficile Risk” (Lowry, 2013). Clostridium difficile (C. difficile) infection (CDI) is a common and clinically significant cause of diarrhea associated with antibiotic drugs. This news story described the findings from a recently published study (Rogers et al., 2013), which was conducted in response to an unexpected finding from a previous study (Dalton, Lye-MacCannell, Henderson, MacCannell, & Louie, 2009). Because of the potential public health significance of this finding, I think my following critical review of the methodology and findings from this study would be instructive for nurses.
Antidepressant Drug Use Analyses and CDI
In their first analysis (Study 1), Rogers et al. (2013) used a data set from a longitudinal investigation of a nationally representative sample (N = 16,781) of older Americans, linking data from biennial interviews to diagnosis data extracted from administrative files of the Centers for Medicare & Medicaid Services (CMS). These files contain diagnosis code data recorded from physician and emergency department visits, stays in hospital and skilled nursing facilities, home health visits, and other outpatient visits. CDI diagnosis and various depressive disorder diagnosis codes were extracted from CMS files. The investigators used the number of infection-related visits (based on CMS diagnosis codes) as a crude proxy for exposure to antibiotic agents. Actual antibiotic agent, PPI, H2RA, antidepressant agent, or other drug use data were not directly ascertained (i.e., from the biennial interviews or by using prescription claims data).
In their second analysis (Study 2), Rogers et al. (2013) conducted a case-control investigation of hospitalized adults (N = 4,047) who had been tested for CDI, comparing 468 cases (CDI positive) and 3,579 controls (CDI negative). Data regarding patient demographics, selected medical comorbidities, and medications prior to CDI testing were extracted from the hospital electronic data system. In a third analysis (Study 3), Rogers et al. (2013) conducted a case-crossover study, where hospitalizations in which hospital-acquired CDI occurred (406 CDI patients) were compared with subsequent hospitalizations for the same patient in which CDI did not occur (i.e., the 406 patients had 949 subsequent hospitalizations for reasons other than CDI). Data regarding medications for each hospitalization were extracted from the electronic system.
In Study 1, the rate (per 100,000 person-years) of CDI among patients with depression was 282.9, 95% confidence interval (CI) [226.3, 339.5] compared to nondepressed patients, whose CDI rate was 197.1, 95% CI [168.0, 226.1]. If CI ranges overlap, the numerical difference between CDI rates (282.9 versus 197.1) would not be considered statistically significant. Because the two CI ranges did not overlap, the higher CDI rate among patients with depression was considered statistically significant (although slight).
Calculating an odds ratio (OR) is a statistical approach for quantifying how the presence of a variable (compared to its absence) relates to the chances (risk) of having a particular outcome. When an OR is significantly >1, there is a greater risk of an outcome. When an OR is significantly <1, there is a reduced risk. When an OR is close to 1, there is no significantly altered risk of an outcome. After adjusting for certain baseline variables (age, gender, race, body mass index, smoking, end-stage renal disease, various comorbid bowel diseases, total number of medical-related visits, and number of infection-related visits), Rogers et al. (2013) found the OR was significantly >1 for depression. Hence, depressed patients (compared to non-depressed) had a significantly greater risk of CDI. They also found significantly greater OR risks of CDI in patients who were widowed, divorced, or separated, or lived alone. Psychiatrist visits were not associated with a significantly altered OR risk of CDI. Curiously, Rogers et al. (2013) did not adjust their depression analysis for baseline marital status, living alone, or psychiatrist visits. They used a crude measure of antibiotic drug exposure (rather than prescription claims data) and had no measures of PPI or H2RA exposure, psychotropic drug exposure, or other nonpsychotropic drug exposure. Without these data, they cannot adjust their depression analyses in Study 1 for concurrent psychotropic drug use (e.g., antidepressant agents) or nonpsychotropic drug use (e.g., antibiotic or PPI/H2RA drugs).
In their case-control analysis (Study 2), Rogers et al. (2013) found no difference between cases and controls regarding age, gender, race, or type of admission (elective, urgent, or emergent). Similar percentages of cases and controls were taking antibiotic agents, immunosuppressant drugs, PPIs, H2RAs, statin (cholesterol) drugs, antianxiety drugs, and antipsychotic drugs. They then determined the OR risks for CDI for 12 different antidepressant agents. After adjusting for certain baseline variables (age, gender, race, antibiotics, PPIs, H2RAs, statins, and comorbid bowel disease), patients taking mirtazapine (Remeron®) or fluoxetine (Prozac®) (compared to non-users) had a significantly increased OR risk of CDI. The use (compared to non-use) of other antidepressant drugs (escitalopram [Lexapro®], citalopram [Celexa®], sertraline [Zoloft®], paroxetine [Paxil®], nortriptyline [Pamelor®], amitriptyline [Elavil®], trazodone [Oleptro®], duloxetine [Cymbalta®], venlafaxine [Effexor®], and bupropion [Wellbutrin®]) was not associated with significantly altered OR risks for CDI. Although not statistically significant, the OR for duloxetine was >1 (1.15), whereas the OR for venlafaxine was <1 (0.71). It is notable that drugs from the same class (serotonin-norepinephrine reuptake inhibitor [SNRI]) had OR risks in opposite directions, and fluoxetine alone (among five serotonin reuptake inhibitor [SRI] class drugs) was associated with a significant risk of CDI. These observations call into question the validity and plausibility of the study findings.
These antidepressant drug use analyses did not adjust for baseline depression, marital status, living alone, or psychotropic drug use, which may confound or bias the findings. Antidepressant agents are used for nondepression psychiatric as well as nonpsychiatric reasons, but Rogers et al. (2013) do not have data to take these reasons into account in any analysis. They considered whether interactions between antidepressant agents could alter the risk of CDI, but only reported findings pertaining to mirtazapine and trazodone. For patients taking both drugs, the OR risk for CDI was substantially greater than for patients receiving neither drug. The OR risk for this combination was higher than for any of the 12 antidepressant agents. Rogers et al. (2013) do not describe or report on other antidepressant agent combinations or antidepressant agents combined with other psychotropic drugs. Trazodone is used primarily for sleep, either alone or together with an SRI or SNRI. Combining mirtazapine and trazodone is highly unusual, especially for relatively older patients. Why these drugs were combined and why their combination might be associated with an especially increased risk of CDI is not discussed by Rogers et al. (2013). The finding is likely to be spurious.
For their case-crossover analysis (Study 3), Rogers et al. (2013) calculated OR risks for CDI for the 12 antidepressant agents and the mirtazapine-trazodone combination. After adjusting for a limited number of baseline variables (i.e., antibiotic agents, PPIs, H2RAs, statin drugs, immunosuppressant agents, and blood transfusions), Rogers et al. (2013) found a substantially increased OR risk for mirtazapine-trazodone combined (compared to non-use of both drugs). Many of the ORs they calculated in Study 3 are <1, although not all are statistically significant. The reduced ORs for citalopram and trazodone, however, are each statistically significant, but they do not discuss or comment on these findings. Trazodone was used in 171 cases (more than any other drug), but Rogers et al. (2013) do not critically discuss their finding that trazodone alone is associated with a reduced risk of CDI, whereas adding trazodone to mirtazapine substantially increases the risk.
In the case-crossover analysis, the OR for the SNRI drugs duloxetine (0.44) and venlafaxine (5.05) are on opposite sides of 1 (neither is statistically significant). In the previous case-control analysis, the ORs of these drugs also are on opposite sides of 1, but their respective ORs have reversed direction (1.15 for duloxetine and 0.71 for venlafaxine in the previous analysis). Similarly, the OR for paroxetine reversed direction from the case-control study (0.86) to the case-crossover study (4.71), but neither is statistically significant. Methodological differences are noted in the case-control and case-crossover analyses, but the discrepant findings do not provide valid evidence that antidepressant agents increase the risk of CDI.
In their study, Rogers et al. (2013) highlight the findings of an animal study by Varghese et al. (2006) in which “behavioral depression” causes colitis, but they neglect to discuss the fact that antidepressant agents attenuated this effect in this particular study. Rogers et al. (2013) focus much of their discussion on functional and inflammatory bowel disease rather than CDI. They never discuss the generally favorable use of antidepressant agents in patients with functional/inflammatory bowel disease (Ford, Talley, Schoenfeld, Quigley, & Moayyedi, 2009; Goodhand et al., 2012; Halpert et al., 2005).
There is considerable literature on the broad antimicrobial effects of antidepressant agents and other psychotropic drugs (Munoz-Bellido, Munoz-Criado, & Garcia-Rodriguez, 2000; Wainwright, Amaral, & Kristiansen, 2012). Surprisingly, Rogers et al. (2013) did not reference this literature in discussing their findings. Indeed, sertraline has demonstrated antimicrobial activity against C. difficile (Munoz-Bellido et al., 2000). If antidepressant agents have “broad-spectrum” antibiotic properties, could this render antidepressant-treated patients susceptible to CDI? Rogers et al. (2013) describe CDI as “not a common outcome” (p. 9). If antidepressant agents increase the risk of CDI, you might expect higher rates, as 12% of Americans take antidepressant agents and they are the third most commonly prescribed drugs (Pratt, Brody, & Gu, 2011).
Rogers et al. (2013) also do not discuss the complex bidirectional relationships among depression, immune system function, and infectious disease (Anders, Tanaka, & Kinney, 2013; Tanaka, Anders, & Kinney, 2012). An infection-defense hypothesis has been articulated, which proposes that depression offers advantages in fighting existing infections and preventing new infections, and that infections, immune processes, and mood influence one another (Anders et al., 2013). In addition, there is evidence that various antidepressant therapies increase immunological defenses (Tanaka et al., 2012), which is certainly relevant to the work of Rogers et al. (2013).