Mortality and complications after percutaneous endoscopic gastrostomy: a retrospective multicentre study

We conducted this study in Region Norrbotten, a region in the northernmost part of Sweden with approximately 250 000 inhabitants. Primary care is provided through 28 primary care centres spread across the region, while secondary care is provided at five hospitals located in Kiruna, Gällivare, Sunderbyn, Kalix and Piteå. PEG is only conducted in two of these hospitals, Gällivare and Sunderby hospital. The procedure is performed by approximately 30 different surgeons and surgical residents in Region Norrbotten. The procedure is mostly performed in an endoscopic ward in sedation not requiring an anaesthesiologist present, however with the possibility of doing it in general anaesthesia in an operation theatre when needed. Patients routinely receive a single dose of prophylactic antibiotics given through the gastrostomy and the end of the procedure. Standard diameter of the PEG tube used in Region Norrbotten is 20 French.

Our primary outcome was all-cause mortality at 30 days, with secondary outcomes including mortality at 90 days, as well as minor- and major complications occurring within 30 days from the procedure. Minor complications included accidental dislocation of gastrostomy tube, local infection, leakage around stoma, abdominal pain requiring surgical consultation or tube blockage. Major complications included aspirational pneumonia, organ damage, major bleeding requiring blood transfusion, intraabdominal leakage requiring surgery or peritonitis requiring surgical consultation. All outcomes were measured as dichotomous variables (yes/no). Our main exposure variable was the indication for the PEG procedure, classified into three categories: neurological (such as stroke, Parkinson disease, motor neuron disease etcetera), malignancy (any tumour obstruction proximal to the ventricle causing need for enteral nutrition or tumour obstruction distal to the ventricle causing need for gastric decompression) and other causes of dysphagia (trauma, infections, unknown dysphagia etcetera). Other variables that were screened for having an impact on the outcomes included age, sex, body mass index (BMI), method (introducer method vs pull method), which hospital performed the procedure, comorbidity (diabetes mellitus, heart failure, ischemic heart disease, chronic obstructive pulmonary disease, anticoagulation therapy) and laboratory tests (haemoglobin (Hb), CRP, thrombocytes, albumin level and creatinine level) measured closest prior to the procedure but up to a maximum of 30 days in advance.

We analysed patient characteristics and missing data, followed by unadjusted and adjusted analyses. Patient characteristics and missing data are presented as absolute numbers and proportions. In the unadjusted analyses all scale variables were transformed to the ordinal scale for better visualization. Missing data were handled by using multiple imputation, using all confounders and outcomes as predictors while only imputing confounders. Two variables that was planned for inclusion were excluded; the first being albumin, which was excluded from all further analyses due to 58% missing data, the second being treatment with a higher than prophylactic dose of low-molecular-weight heparin (anticoagulation therapy), where only four patients (1%) in the study group were treated with a higher dose and hence it was also excluded from further analyses. Finally, it should be noted that there were 378 unique patients, but 11 of these received a PEG twice during the study period meaning that 389 PEG operations were conducted. We concluded that each PEG operation was valuable to include, and since the number of patients receiving a PEG twice was low, we chose to use the term “patients” throughout the manuscript.

We used multivariable logistic regression analyses to adjust for confounding factors. Multicollinearity was addressed by constructing a matrix for bivariate correlations as well as addressing variance inflation factor statistics. The assumption of linearity was tested using the Box-Tidwell test. Age, thrombocytes, and creatinine did not fulfil the assumption for all outcomes and hence was transformed to the ordinal scale in the final model for this outcome; otherwise, the models were built identically for all outcomes. The following variables were included in the final multivariable models: indication, year of procedure, hospital, operation method, age, sex, BMI, diabetes, heart failure, ischemic heart disease, chronic obstructive pulmonary disease, CRP, Hb, thrombocytes and creatinine levels.

All confounders were chosen based on clinical experience and probability of availability in the medical records. The robustness of the multivariable analyses was checked by performing analyses only on patients undergoing their first PEG operation which showed similar results, and as complete-case analyses on unimputed data which showed similar but attenuated results. All variable and the modelling strategy were decided a priori. We performed all statistical analyses in IBM SPSS Statistics 26. Ethical approval was obtained from the Swedish Ethical Review Authority (DNR 2020-01,917).

We found a total of 393 patients. Four patients were excluded because of age under 18 years, which left a total of 389 patients, of whom all were included in the study. In all, 176 (45%) were women, the median age was 72 years, and the median BMI was 22. Ischemic heart disease was the most common comorbidity, occurring in 80 (21%) patients. In all, 261 (73%) patients received their gastrostomy using the pull method, and the most common indication for the procedure was neurological disease, which affected 281 (72%) patients. Most of the variables had no missing data, and among variables with missing data, the level was generally low. Baseline data and patient characteristics are shown in Table 1.

Mortality was 15% (n = 59) at 30 days and 28% (n = 108) at 90 days. The most common cause of death was malignancy, followed by aspiration pneumonia (Table 2). The unadjusted analyses revealed a much higher mortality rate at both 30 and 90 days for patients where the indication for PEG was malignancy (25% at 30 days and 51% at 90 days) compared to neurological deficit (13% at 30 days and 21% at 90 days) (Table 3). A visualization of this is provided as a Kaplan–Meier curve (Fig. 1). In the adjusted analyses malignancy as indication was significantly associated with higher mortality at 90 days (Odds Ratio [OR] 4.41, 95% Confidence interval [CI] 2.20–8.88) (Table 4). Every year increase in age significantly increased mortality at 30 days (OR 1.03 95% CI 1.01–1.06). Each point increase in CRP significantly increased the mortality rate at 30 days (OR 1.01, 95% CI 1.00–1.01) and at 90 days (OR 1.01, 95% CI 1.00–1.01). Mortality at 90 days was significantly higher for women (OR 1.95, 95% CI 1.11–3.44) while it significantly decreased with increasing BMI (OR 0.91 per point, 95% CI 0.85–0.98). Mortality at 90 days was also significantly higher if the patient had diabetes (OR 2.11, 95% CI 1.07–4.19) or heart failure (OR 2.29 95% CI 1.02–5.14). There was also a significant difference in mortality between the two hospitals at both 30 days (OR 0.18 95% CI 0.08–0.41) and 90 days (OR 0.39 CI 95% 0.19–0.78). Tables 3 and 4 shows a full account of the unadjusted and adjusted analyses.

Minor complications were seen in 41 (11%) patients, and major complications were seen in 57 (15%) patients. The most common minor complications were dislocation of the tube (49%) and local infection (30%), while the most common major complications were aspiration pneumonia (74%) and organ damage (9%). If aspiration pneumonias are excluded, the total number of major complications is 15 (4%) (Table 5). In the adjusted analyses diabetes significantly increased the risk for minor complications (OR 2.61, 95% CI 1.04–6.55), while there was a significantly higher risk for major complications among patients aged 75 years or older, compared to those younger than 65 years (OR 2.23, 95% CI 1.02–4.85). Finally, we also found a significant difference in the risk of major complications between the two hospitals (OR 0.42, 95% CI 0.19–0.91) (Table 4).