Effect of Proton Pump Inhibitors on the Risk and Prognosis of Infections in Patients With Cirrhosis and Ascites

Gitte Dam; Hendrik Vilstrup; Per Kragh Andersen; Lars Bossen; Hugh Watson; Peter Jepsen


Liver International. 2019;39(3):514-521. 

In This Article


The Satavaptan Trials

Three multinational randomized controlled trials were conducted in 2006–2008 to examine whether satavaptan reduces ascites formation in patients with cirrhosis.[25] The three trials had different target populations: Patients with diuretic-manageable ascites (N = 462), patients with ascites managed with diuretics and occasional therapeutic paracentesis (N = 496), and patients with diuretic-resistant ascites managed primarily with therapeutic paracentesis (N = 240). The trials were conducted similarly and included 1198 patients in total.

All three trials excluded patients with a functioning transjugular intrahepatic portosystemic shunt and patients with variceal bleeding or spontaneous bacterial peritonitis in the 10 days before randomization. They also excluded those with a serum creatinine >151 μmol/L, serum potassium ≥5.0 mmol/L, serum sodium >143 mmol/L, serum bilirubin >150 μmol/L, INR >3.0, platelets <30 000/mm3, neutrophils <1000/mm3, hepatocellular carcinoma exceeding the Milan criteria, and those who used a potent modifier of the cytochrome P450 3A pathway or any drug that increased the risk of Q-T interval prolongation.

The planned treatment duration was 52 weeks. All patients were followed for one additional week after planned or premature treatment cessation. Infections and other cirrhosis complications were not recorded after that date (which we hereafter refer to as the "complication follow-up date"), but the survival status was recorded for all trial participants by the end of the planned trial duration (hereafter referred to as the "survival follow-up date"). Thus, follow-up for death continued after follow-up for other complications had ended.

Patients were seen every 4 weeks in their hepatology departments. At every visit, all current medications including their indications and dosages were recorded, blood tests were taken, and patients were examined clinically and had paracentesis of ascites. Clinical events were recorded, including infections between visits. The diagnosis of "infection" was made by the clinician caring for the patient, following everyday clinical criteria. The analysis of the trial data showed that satavaptan did not have the desired beneficial effect on ascites formation. Nor did it affect the incidence rate of infections (satavaptan: 79 first-time infections per 100 person-years; placebo: 78 first-time infections per 100 person-years); hence, we combined the satavaptan and placebo arms for the analyses presented here.

Statistical Analysis

Effect of infections on mortality. We firstly examined whether infections had any effect on patient mortality, as a measure of their clinical importance. To do so, we followed patients from inclusion to the survival follow-up date and compared the mortality rate before an infection or more than 90 days after an infection with the mortality rate during the 90 days following an infection. Infection was a time-dependent variable, so that patients who developed an infection contributed follow-up time to both comparison groups, whereas patients who did not develop an infection contributed follow-up time to the no-infection group only. We conducted this analysis for infections overall and for the commonest sites of infection: spontaneous bacterial peritonitis, upper respiratory tract infection, lower respiratory tract infection, urinary tract infection, skin infection and gastroenteritis.

Effect of PPI use at baseline on all-cause mortality. Secondly, we examined the association between PPI use at the time of inclusion and all-cause mortality. We defined PPI by the ATC-code A02BC. Follow-up began at inclusion and ended at death or in censoring on the survival follow-up date. We used the Kaplan-Meier estimator to compare all-cause mortality between users vs nonusers of PPI at inclusion. With Cox regression, we adjusted for the following confounders, also measured at inclusion: patient gender, age, cirrhosis aetiology (alcoholic cirrhosis vs other), cirrhosis severity as measured by the MELD score (incorporating INR, bilirubin, creatinine and sodium),[26] serum albumin, use of lactulose (yes or no), severity of ascites (refractory or diuretic-responsive), history of variceal bleeding before inclusion (yes or no) and history of spontaneous bacterial peritonitis before inclusion (yes or no).

Effect of PPI use on the incidence of infections. Next, we examined whether PPI use increased the incidence of infections. Follow-up began at inclusion and ended on the date of the first infection or at death. Patients who experienced neither event were censored on the complication follow-up date. We restricted the analysis to first-time infections because it was impossible to determine when one infection ended and another began. We used Cox regression to estimate the association between current PPI use and the hazard rate of first-time infection. In this analysis, PPI use was a time-dependent variable, so that a patient contributed follow-up time to the "PPI user" group when he was using PPI, and to the "PPI nonuser group" when he was not. The hazard ratio for PPI use vs nonuse expresses the risk of a first-time infection during the day for a patient who does vs does not use PPI at the beginning of that day. The proportional hazard assumption assumes that this ratio remains constant throughout the follow-up period, and we used scaled Schoenfeld residuals to examine that assumption. We controlled for the same confounders as described above, but in this analysis, MELD scores, serum albumin and use of lactulose were updated during the follow-up.

We used a marginal structural model to control for time-dependent confounding and to contrast the cumulative risk of an infection between PPI treatment strategies. The effects of PPI use in cirrhotic patients are not fully understood, and time-dependent confounding can arise when the decision to treat a patient with PPI depends on factors that are a result of PPI treatment offered at the previous visit.[27,28] Unlike standard Cox models, marginal structural models can control for time-depending confounding, yielding the so-called "average causal effect."[29] They can also model the effect of PPI use on the cumulative risk of infection. We analysed each four-weekly visit separately in order to estimate the effect of PPI use on the risk of having an infection before the next visit, considering death without infection as a competing risk. We subsequently combined the per-visit models to estimate the cumulative risk of an infection for an "always give PPI" strategy vs a "never give PPI" strategy. We controlled for the same confounders as in the Cox regression model and used the bootstrap method to estimate a 95% confidence interval for the relative risk. The marginal structural model is described in detail in the Appendix S1.

Effect of PPI use on 90-day mortality following infection. We examined whether PPI use was associated with a higher 90-day mortality following infection. The analysis included only patients with an infection, and they were followed from onset of infection and 90 days onwards. Surviving patients with <90 days of follow-up were censored on the survival follow-up date. We used a stratified Cox proportional hazard model to examine the effect of PPI use on the mortality hazard ratio during the 90-day follow-up, adjusting for the confounders described above. The Cox model was stratified by the type of infection (spontaneous bacterial peritonitis, upper respiratory tract infection, lower respiratory tract infection, urinary tract infection, skin infection, gastroenteritis or other infection), meaning that the mortality hazard was allowed to differ between the types of infection, but all model variables were constrained to have the same effect on all types. In this analysis, we ignored changes in PPI use or in MELD scores, albumin levels or lactulose use during the 90-day follow-up, because those data were not collected between the complication follow-up date and the survival follow-up date.