Proton pump inhibitors and 180-day mortality in the elderly after Clostridium difficile treatment

To reduce the potential for systematic error and to mitigate bias, we followed protocols for the optimal conduct of retrospective studies. Before data were abstracted, we a priori defined the pertinent predictor and outcome variables to be extracted from the medical record in a standardized manner. Trained abstractors used a standardized collection form to query the EMR to obtain longitudinal data pertaining medication use prior to and during treatment for CDI. Demographic data, including age at CDI diagnosis, sex, race and ethnicity, were collected. We also calculated a Charlson comorbidity index (CCI) score to characterize the individual’s medical comorbidities. Details of the initial CDI treatment, including if the case of CDI was hospital acquired, and the patient received dual antibiotic coverage were also collected. We defined the initial CDI as severe by using both the Infectious Disease Society of America (IDSA) 2010 clinical practice guidelines (i.e. presence of leukocytosis greater then 15,000, serum creatinine 1.5× greater then baseline, hypotension, shock, megacolon, ileus or need for surgical intervention), or if the treating physician felt the case was severe and escalated therapy. Abstractors were blinded to outcome status and mortality data was obtained after all patients in the cohort had their data extraction complete. These methods were used to reduce information bias during clinical record review.

We used the EMR to determine the medications each participant was exposed to. We specifically queried the EMR for medication exposures initiated during treatment for CDI. We had particular interest in antibiotics (both those used in the treatment of CDI and those used to treat other bacterial infections) and acid reducing medications. Data on both prevalent (daily medication use prior to CDI) and incident acid-reducing medication use (initiated during treatment for CDI) including both PPIs and H2 blocker medications were collected.

To determine mortality outcomes, patients in our cohort were queried in the Center for Disease Control National Death Index (NDI). Six-month all-cause mortality was determined for each individual CDI case. Only after all study information was collected for all participants did two independent physician reviewers (JH, EB) make case and control assignments. We performed this same procedure for both 90-day and 1-year mortality.

We used Chi-square tests to compare categorical variables and the student’s t-test for continuous variables between CDI patients with 180-day mortality and survivors. Bivariate risk ratios were determined by logistic regression. After examining the proportional hazards assumption for Cox proportional hazards model, it was determined that the variables did not violate the model’s assumptions and the Cox model was used to identify associations between predictor variables and the outcome of 180-day mortality. Clinically important treatment factors included antibiotic treatment during initial CDI treatment, acid-reducing PPI medication exposure, and probiotic treatment. The other main variables of interest included patient demographic characteristics (including age, sex, race), infection characteristics that included hospital acquired CDI and IDSA defined severe CDI, location of residence (community or nursing home), and medical comorbidities using the CCI score. The final adjusted model included clinically important variables we defined a priori and variables associated with the outcome at a significance level of p < 0.10 from our bivariate analysis. Kaplan–Meier survival analysis was used to determine time to all cause mortality. The log rank test was used to compare the survival curves of time to mortality. Significance was set at a p value < 0.05 for all analysis. We used multiple imputation to address missing data, after exploring the data and determining missing data was missing at random. We used Stata version 13.1 for all analyses (Stata Corp LP, College Station, TX).

Exposure to PPIs during the treatment course of CDI may help reduce intestinal inflammation. In vitro, neutrophils stimulated by Helicobacter pylori extracts that were treated with omeprazole or lansoprazole showed decreased expression of endothelial attachment factors, an effect not seen when the same neutrophils were treated with famotidine or ranitidine [22]. This data suggests PPIs may have an anti-inflammatory effect by preventing migration of neutrophils into tissue. Another in vitro study showed that coloncyte cultures grown in the presence of omeprazole showed some notable gene expression changes including down-regulation of a protein involved in leukocyte tissue transmigration and down-regulation of a C. difficile toxin target [23]. These changes in in vitro gene expression can be both hypothesized to promote or inhibit infection and inflammation. Changes in host and pathogen gene expression after exposure to PPIs may alter the natural disease course, however human evidence is lacking.

Another hypothesized mechanism through which acid-reducing medication could effect the course of CDI is through changes in colonic flora, which has been observed with PPI treatment [9, 24]. The effect of PPI in this setting has not been well studied and how the altered colonic flora after PPI exposure might influence the course CDI is not known.

In this article, we report a decrease in all-cause mortality in patients treated for CDI who were also treated with PPI, we did not assess the effect of PPI on mortality specifically related to CDI. As such, it may be that the effect PPI has on mortality in CDI patients may also be independent of any direct effect of the disease itself. PPI remains an effective treatment for certain conditions, such as patients hospitalized with upper GI bleeding or those critically ill in the ICU [5]. In these patients the usual barriers that protect the gastric mucosa, such as mucous or bicarbonate secretion, have broken down and PPIs limit proton secretion from undamaged mucosa, leading to increased gastric pH that limits direct damage to mucosa from acidic environment and also reduces pepsin activation which enzymatically degrade damaged mucosa or fibrin clots that may have formed [25]. It may be that patients with CDI are also at risk for stress ulcers or other conditions treated effectively by PPI and this is the reason for the observed mortality benefit.

In patients hospitalized for other medical reasons, data exist to show that PPIs are effective prophylactically to prevent upper GI bleeding for those at high risk, such as critically ill patients [5]. However, acid-reducing medication does not appear to generally benefit all hospitalized patients. PPI use has been associated with increases in all cause mortality after hospitalization for myocardial infarction [26] and when used in long-term in hemodialysis patients [27]. It is clear that PPI does not generally lead to decreased mortality and only benefits patients with certain conditions, of which CDI may be one.

The risk of developing CDI has been well linked to PPIs use. It has been hypothesized that the association is due to the nature of patients treated with PPI who are generally more elderly and have more medical comorbidities and this are at a higher risk for CDI in general. In 2014, Novack et al. attempted to address this concern by performing a case control study in which controls were patients suspected of having CDI, but with negative testing, as opposed to other matching methodologies. They reported that use of PPI use was associated with a negative test as opposed to increased risk [15]. Other studies that compare CDI patients to those that were tested but were negative also fail to find an association between PPI use and CDI [28, 29]. In a prospective cohort of 894 cases of CDI inpatients showed there was no association between concurrent PPI use and risk of recurrent CDI [30]. The same group evaluated use of PPIs in ICU patients and the risk of developing CDI and also found no association [31].

The association between PPI use and severity of CDI and mortality after CDI treatment not been thoroughly investigated. There are some reports that cases of CDI associated with PPI use are more severe and lead to higher complication rates and mortality [17‐19]. This could be explained by the average patient treated with PPIs being generally older and having more medical comorbidities. One study that attempted to address this specific concern followed 285 patients treated for CDI at Mayo Clinic found that after adjustment for age and medical comorbidities, CDI patients with PPI exposure were not more likely to suffer complications or be associated with treatment failure [20]. Although these authors did not note a protective effect, all cause mortality in these patients was not explored. In a meta analysis evaluating risk factors for poor outcomes in patients treated for CDI, PPI was associated with increased mortality in a single study out of the 40 studies included in the analysis [14]. This systematic review found that most studies evaluating mortality failed to find an association. Here we report a positive association with PPI use and reduced mortality even after multivariate adjustment.

H2 blockers and PPIs are both used to suppress gastric acid production and are typically used for similar indications. Prior studies have found that H2 blockers and PPI similarly affect CDI risk and recurrence [13, 32]. In studies evaluating severity of CDI in the setting of gastric acid suppression these agents have been typically analyzed together as representing a similar exposure [14, 17] and the specific contribution of H2 blockers has not been reported. In our analysis, treatment with H2 blockers did not confer a similar mortality benefit to PPI. These medications act by different mechanisms, and it may be that the mortality benefit seen in PPI use is not related specifically to gastric acid suppression. Alternatively, the efficacy of the gastric acid blockade is typically thought to be greater in PPIs then in H2 blockers [33], and hence, it may be that the mortality benefit results from a more robust acid blockade in PPIs.

Other factors contributing to mortality in our data set included increasing age, higher CCI score, hospital acquired CDI and severe or complicated infection as determined by 2010 IDSA guidelines. The finding of increased mortality related to advancing age and severe infection was seen among most studies in a 2014 meta analysis looking at outcomes of CDI and makes intuitive sense [14]. The association with higher CCI score with 30 day mortality was also reported in two of the studies included in that analysis [14]. In our analysis, it was significantly associated with 180-day mortality, suggesting that these patients are particularly high risk and require close attention when CDI develops during the hospital course. The use antibiotics during CDI treatment was seen to increase risk of complication [14], however, this did not appear to effect mortality in our study.

Our study found a decrease in mortality associated with PPI use that was consistent and seen at 90-days, 6 months and 1 year, suggesting a lasting benefit associated with these medications. All CDI cases in the study period were included, which should limit referral bias. Our study included both patients treated within the hospital and at home and looked beyond inpatient morbidity and mortality. Taken together our study suggests PPIs may play a beneficial role in reducing death when taken during CDI treatment.

However, all-cause mortality was measured, not mortality related specifically incident or recurrent CDI. We also did not collect data on other outcomes associated with CDI treatment such as hospital length of stay, recurrences, or treatment complications. Although the mortality benefit appears robust, how the use of PPI affects the clinical course of CDI remains to be studied in depth. Additionally our investigation was a retrospective cohort study design, which may introduce selection and information biases. We attempted to limit this bias by applying standard protocols for the optimal conduct of retrospective study.