Is Postoperative Infection a Predictor of 1-year Mortality? A Propensity Score Analysis

Author(s):
William O’Brien; Kalpana Gupta; Kamal Itani

Background:

Postoperative infection is associated with subsequent higher cost, morbidity and mortality. The purpose of this study is to determine whether patients with a postoperative infection face increased risk of 1-year mortality, after adjusting for confounding using a propensity score approach to balance observed patient baseline characteristics between control and exposure groups.

Hypothesis:

Patients with a 30-day postoperative infection have a higher risk of mortality during the year after surgery.

Methods:

We obtained chart-reviewed data on patient and surgery characteristics, 30-day infections (the exposure), and 1-year mortality (the outcome), from the VHA Surgical Quality Improvement Program (VASQIP) during 2008-2015. Surgeries were included if the patient had no prior or subsequent surgery within 1 year, and if the patient had neither 30-day preoperative infection nor 30-day postoperative death. 30-day infection was defined as occurrence of surgical site infection, bloodstream infection, urinary tract infection, or pneumonia. A propensity score model was estimated, and stabilized inverse probability of treatment weights were used in a Cox proportional hazards model to estimate risk of death as a function of exposure to postoperative infection.

Results:

There were 538,610 surgeries included in the study, in which 16,178 (3.0%) had a 30-day infection. Of these, 6,570 (40.6%) had an SSI, 4,442 (27.5%) had a UTI, 2,579 (15.9%) had PNA, and 1,137 (7.0%) had a BSI. The remaining 1,450 (9.0%) of patients with infection had a combination of 2 or more infection types (e.g. SSI plus UTI). 1-year mortality occurred in 1,289 patients (0.24%), and median postoperative day of death was 47 (IQR 37-66). Baseline characteristics of patients with and without infection were well balanced in the weighted sample, with surgical specialty being the only covariate with a standardized mean difference greater than 0.2. Mean age in control patients was 59.0 (SD 13.8), and in exposed patients was 59.2 (13.6). 91.2% of controls were male, compared with 88.9% of exposed patients. In regression analysis, the hazard ratio of mortality in exposed patients was 4.41 (95% CI 3.68 – 5.28).

Conclusions:

The occurrence of postoperative infection leads to a 4.4 times higher probability of death at any given time during the 1-year postoperative interval, compared with those patients having no postoperative infection. This is one of the largest studies to date describing long-term harms of surgical infection, using recently developed statistical methods capable of inferring causality in a non-randomized cohort.