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


We included 1198 patients with cirrhosis and ascites, and 446 of them had an infection during the follow-up. The mortality rate during the 90 days following an infection was 3.45-fold (95% CI 2.64–4.47) higher than it was at other times, attesting to the clinical importance of infections. However, upper respiratory tract infections did not increase mortality at all (mortality rate ratio = 0.33, 95% CI 0.04–1.21), with only two deaths observed during the 90 days following 82 episodes of infection. The mortality rate ratios for the other infection sites ranged from 1.76 (skin infection) to 6.51 (spontaneous bacterial peritonitis).

Proton pump inhibitor users had higher all-cause mortality than nonusers (Figure 1), and this excess mortality could not be explained by confounding. Thus, the confounder-adjusted mortality hazard ratio for users vs nonusers of PPI was 1.28 (95% CI 1.01–1.62).

Figure 1.

Cumulative all-cause mortality for users and nonusers of proton pump inhibitor at the time of inclusion

Of the 1198 included patients, 524 (44%) used PPI at inclusion, and 645 (54%) used them at some point during the study period. On a given day, there was always between 32% and 44% PPI users. PPI users had experienced more cirrhosis complications than nonusers, but their MELD scores and albumin levels were the same (Table 1).

During the follow-up, 446 patients had an infection. Without confounder adjustment, the hazard ratio for infection for PPI use vs nonuse was 1.50 (95% CI 1.24–1.81), and with confounder adjustment, it was 1.43 (95% CI 1.18–1.74; Table 2). The hazard ratio of infections was higher early in the study period, but this time dependency was not statistically significant (P-value for scaled Schoenfeld residual = 0.11, Figure 2). The increased rate of infections was not restricted to one particular infection site, but it was notable that PPI use did not increase the rate of upper respiratory tract infections, which are likely to be caused by virus rather than bacteria (Figure 3). Table S1 presents the available data on specific pathogens.

Figure 2.

Time dependency of the hazard ratio associated with proton pump inhibitor use. The solid line shows the hazard ratio, which decreases from around week 12. This time dependency is not statistically significant (P = 0.11 using scaled Schoenfeld residuals). The dotted line shows the estimate presented in Figure 1 and elsewhere, which is the average hazard ratio over the entire follow-up period

Figure 3.

Adjusted hazard ratios for infections for current users vs nonusers of proton pump inhibitors among patients with cirrhosis and ascites. Hazard ratios are adjusted for gender, age, cirrhosis aetiology (alcohol vs other) and cirrhosis severity

The results from the marginal structural model were similar to those from the Cox model, yielding an average causal effect of 1.57 (95% CI 1.27–1.95). The similarity of the estimates from the Cox model and the marginal structural model indicates that time-dependent confounding was small. One full year's course of PPI treatment increased patients' risk of having one or more infections, but the effect of PPI use on the risk of infections decreased during the study period (Figure 4). After 6 months (week 24), the estimated cumulative risk of infections was 36.4% for PPI users vs 25.1% for nonusers at 6 months (relative risk = 1.45, 95% CI 1.22–1.73). After 1 year, the cumulative risks were 45.2% vs 37.7% (relative risk = 1.20, 95% CI 0.97–1.40; Figure 4).

Figure 4.

Estimated cumulative risk of having one or more infections for patients who use proton pump inhibitor (PPI) continuously for a full year and for patients who do not use PPI at all. The risk estimates are obtained from a marginal structural model, which uses data from the planned 4-weekly visits during the 1-y follow-up in the satavaptan trials. The model assumes that the decision to use or not use PPI is taken every fourth week for every patient. The bottom graph shows the ratio between the two cumulative risk estimates shown in the top graph

Among the 446 patients with an infection, 81 died during the 90 days following the infection's onset. Mortality was not increased for PPI users: the crude hazard ratio from the stratified Cox model was 0.88 (95% CI 0.57–1.36), and with confounder adjustment, it was 0.83 (95% CI 0.53–1.31). By contrast, a high MELD score and a low albumin were associated with increased mortality (Table 3).