Gastrointestinal Bleeding in Hospitalized covid‐19 Patients: a propensity Score Matched Cohort Study

Accepted Article

---

This article is protected by copyright. All rights reserved

Definition of Gastrointestinal Bleeding

Patients were identified as having a potential GI bleed using billing data with one of the following 

ICD-10 codes: Melena (ICD 10 K92.1), GI bleeding (All K92 codes; Z87.19; D50.0; K62.5; K29.51; 

K26.4; K29.81; K25.4; K25.4; K25.6;K29.91; K92.2; K29.71) , hematemesis (K92.0),  hematochezia 

(K92.1; K75.31), bright red blood per rectum (ICD 10 K62.5), or a  hemoglobin drop by 2 g/dL in 24 

hour period.  All charts of potential patients with GI bleeds were then manually reviewed to confirm if 

there was truly a GI bleed. An upper GI bleed was defined as the presence of hematemesis or melena. 

A lower GI bleed was defined as having hematochezia. 

Statistical Analysis

Patients were stratified into 2 groups: GI bleed on admission (GI bleed within 6 hours of admission), 

and GI bleed encountered during hospitalization. Patients with GI bleed on admission were matched 

1:1 to patients without GI bleed (no GI bleed on admission or during hospitalization) using propensity 

score matching. Demographics (age, sex, race, ethnicity), general comorbidities (Charlson 

Comobidity Index), and GI specific comorbidities and known GI bleed risk factors (end-stage renal 

disease, peptic ulcer disease, diverticulitis, inflammatory bowel disease, and prior GI bleed) were 

included in the propensity score model. Patients were matched on the logit of the propensity score 

using optimal matching with calipers of width equal to 0.2[22]. Matching was performed without 

replacement. Patients with GI bleed during hospitalization and patients not yet discharged at the time 

of data collection were excluded from being considered as potential matches for patients with GI 

bleed on admission. 

Patients with GI bleed during hospitalization were matched under similar parameters, except potential 

matches excluded patients with GI bleed on admission and patients not yet discharged at the time of 

data collection. Additionally, length of stay was included as a covariate to ensure that at the time of 

GI bleed, matches were still hospitalized and had the same length of stay as GI bleeders (this was 

done as the longer the hospitalization the higher chance for a GI bleed event to occur). Complete 

separation was observed with the potential confounder of chronic liver disease for both propensity