All aspects of this study were approved by the Institutional Review Board (IRB Protocol #: 844350) at the University of Pennsylvania (Philadelphia, PA). This retrospective cohort study included patients with ESRD on thrice-weekly HD at Penn Presbyterian Medical Center (PPMC, Philadelphia, PA) who underwent endoscopic procedures from 2016 to 2022. Data were collected from the electronic medical record between June and August 2023 and managed using REDCap electronic data capture tools [15, 16].

Any patient with ESRD on thrice-weekly HD who received either an upper endoscopy (EGD), colonoscopy, or combined EGD and colonoscopy at PPMC between 2016 and 2022 was included in the dataset.

Any patient who was not on a regular HD schedule or transitioned to other forms of dialysis, who required emergent intensive care unit (ICU) bedside procedures, or who were missing data on last HD session were excluded. Additionally, procedures with incomplete or missing anesthesia procedure notes and other endoscopic procedures, such as endoscopic ultrasounds and endoscopic retrograde cholangiopancreatographies, were excluded.

The primary exposure variable was timing of endoscopic procedures relative to last HD session. Procedures were categorized into 0 vs. 1 or 2 days post-HD for the primary analysis and 0 vs. 1 day post-HD for a secondary analysis as described below.

Outcome measures included intraoperative cardiovascular complications (arrhythmia and/or cardiac arrests), intraoperative fluid administration, hemodynamic instability (any pressure administration, need for second pressor administration, phenylephrine administration), and mortality (in-hospital and/or 30-day). Arrhythmias were defined as any new or clinically significant arrhythmias documented in the anesthesia procedure note. Fluids were defined as any intravenously administered crystalloid solutions and administration was quantified in milliliters (mL). Any pressor administration was defined as administration of any of the following pressors: phenylephrine, vasopressin, epinephrine, and ephedrine.

Extracted data included patient age, biological sex, race, BMI, American Society of Anesthesiologists (ASA) classification, procedure date, last HD date, encounter type (outpatient, inpatient), procedure type (esophagogastroduodenoscopy (EGD), colonoscopy, combined EGD/colonoscopy), procedure indication (screening/surveillance, gastrointestinal (GI) bleed/anemia, other GI symptoms, abnormal imaging), procedure interventions (biopsy, polypectomy, hemostasis), procedure duration, and anesthesia type (MAC, general). The procedure indication for abnormal imaging included further evaluation of the following findings: esophageal thickening, esophageal narrowing, esophageal varices, duodenal narrowing, duodenitis, gastric lesion, colitis, and colonic lesions concerning for malignancy. All cases of MAC involved propofol with other agents as deemed appropriate by the anesthesiologist.

Statistical analysis was conducted in collaboration with the University of Pennsylvania Biomedical Data Science Core. Analyses were performed using RStudio [17]. We summarized the cohort characteristics as frequencies with percentages for categorical variables and medians with interquartile ranges (IQR) or means with standard deviations (SD) for continuous variables. We utilized Chi-squared analyses to compare frequencies and ANOVA analyses to compare means.

Pre-procedural covariates included biological sex, race, BMI, ASA classification, encounter type, procedure type, and procedure indication. Intra-procedural covariates included procedure duration and procedure interventions. The race variable was collapsed to Black vs. non-Black due to too few patients noted in the race categories of other and unknown.

To balance covariates, we first generated propensity scores for the probability that a patient would undergo endoscopy on the same day as HD compared to 1–2 days post-HD. Two propensity scores were generated using logistic regression. The first propensity score was based on pre-procedural covariates alone (Model 1). The second propensity score included both pre- and intra-procedural covariates (Model 2). Propensity scores were rounded to the hundredth decimal point and extreme values less than 0.01 or greater than 0.99 were capped at these thresholds to avoid over-weighting outliers (three records met this criteria). The primary analysis utilized the GEE model incorporated with inverse probability treatment weighting (IPTW) [18] logistic regression to compute odds ratios (OR) with 95% confidence intervals (CI); the balance of covariate distributions were analyzed with standardized mean differences (SMDs) [19] with a SMD ≤ 0.1 representing good balance. For covariates with SMD values greater than 0.11 (i.e. sex, race, ASA class, encounter type, procedure indication (screening/surveillance and GI bleed/anemia), procedure intervention (biopsy), and procedure type), we tested for confounding of the association of endoscopy day with the outcomes of interest by adding each variable separately to the IPTW logistic regression model.

A secondary analysis compared 0 vs. 1 day post-HD, an additional set of propensity scores was generated and analyzed following the same protocol.

We conducted a sensitivity analysis limited to patients with ASA classification 3 and inpatient encounter type. The propensity scores were re-derived for these subgroups and the process of testing for residual confounding was repeated. To assess for the potential impact of unmeasured confounders, we computed the E-value [20, 21].