Comorbidities Associated with Worse Outcomes Among Inpatients Admitted for Acute Gastrointestinal Bleeding

We conducted a retrospective, cross-sectional study using electronic health record (EHR) data and considered eligible inpatients hospitalized at the University Hospital Zurich between January 1, 2010, and May 31, 2018 (total of 273,378 inpatient stays). All EHR data used in this study were routinely collected. We followed the STrengthening the Reporting of OBservational studies in Epidemiology (STROBE) guidelines [17].

Ethics approval was obtained from the Cantonal Ethics Commission Zurich, BASEC 2016-00525, which waived the requirement for obtaining informed consent for this retrospective study.

The University Hospital Zurich is a large tertiary care academic medical center in Zurich, Switzerland. The institution has approximately 900 beds and cares for more than 35,000 inpatients per year. Over 8200 employees work at this hospital, covering all specialties except orthopedic surgery and pediatrics.

The reason for hospitalization was the primary diagnosis, flagged as such in the EHR. All diagnoses were ICD-10 coded (ICD: International Classification of Diseases, WHO, Geneva, Switzerland). The ICD-10 coded primary diagnoses were used to identify our population of interest: patients admitted for the treatment of acute GIB. Iterative literature searches were alternated with research group discussions to reach consensus about the ICD-10 codes that referred to these patients. The same process was used to define the ICD-10 codes for the identification of clinically relevant comorbidities among secondary diagnoses. After the study population was identified, detailed chart reviews were conducted to clarify any case of ambiguity: For instance, the diagnoses sometimes indicated that a patient suffered from both UGIB and LGIB. Therefore, the ICD-10 code K92.2 (gastrointestinal hemorrhage) was verified by chart review as UGIB; except for 7 cases, which were excluded due to concurrent diagnoses indicating LGIB (i.e., hemorrhoids, diverticulitis and diverticulosis with bleeding). The ICD-10 code K92.1 (melena) was allocated to UGIB [18]. If potential subjects had a record that indicated refusal to be involved in research, they were excluded from the study. Patients with incomplete information on potential readmission events (patients discharged within 29 days before end of study period) and patients who died during their stay were also excluded (n = 59).

We used ICD-10 coded diagnoses to identify patients with acute GIB and stratified them into two study groups.

The outcome variables investigated were LOS and 30-day all-cause readmissions, the latter defined as any admission within 30 days of the index admission. The predefined exposure of interest variable was presence of 15 comorbid conditions potentially associated with increased LOS or 30-day readmissions (cf. section Co-Variables, below). We strictly differentiated between UGIB and LGIB.

We used the natural logarithm to transform the skewed outcome variable LOS, as we did previously (described in more detail elsewhere [12]). This allowed the application of linear regression models. The estimated coefficients were back-transformed by raising e to the power of the coefficients, and these results can be interpreted as percentage increases or decreases.

Co-variables were considered according to a list of diseases with clinical relevance, using the bleeding risk score termed HAS-BLED Score [19] and the stroke risk categorization schema CHA2DS2-VASCc-Score [20]. The 15 comorbidities were identified among all secondary diagnoses which were comprehensively ICD-10 coded. The physicians in charge of the patients defined and updated the primary diagnosis and all secondary diagnoses over the course of the hospital stay. After the patients discharge, professional medical coders assigned ICD-10 codes to each diagnosis. Four conditions were based on the ICD-10 coded Elixhauser Comorbidity Index definitions by Quan et al. [21]: anemia (blood loss and iron deficiency), cancer (solid tumor with/without metastasis), coagulopathy (purpura and other hemorrhagic diatheses) and chronic pulmonary diseases. The other 11 conditions (i.e., alcohol use disorders, atrial fibrillation or flutter, chronic kidney disease, depression, diabetes mellitus, heart failure, heart valve disease, hypertension, liver disease, thromboembolic event and vascular disease) were defined by means of iterative literature searches alternated with research group discussions to reach consensus. Thromboembolic event includes the diagnoses pulmonary embolism and deep vein thrombosis. Vascular disease was defined by the diagnoses of acute myocardial infarction, coronary heart disease, chronic ischemic heart disease and peripheral arterial vascular disease. The comprehensive lists with their specific ICD-10 codes are summarized in Supplementary Table S1. Multimorbidity was defined as the presence of at least two diagnoses [22]. We controlled for multimorbidity and burden of disease by adjusting for diagnosis count [23].

To control for potential confounding in the context of the therapeutic conflict of reinstating antithrombotic medication for our inpatients admitted with acute GIB, comprehensive drug ordering data were extracted. Antithrombotic agents were identified by the Anatomical Therapeutic Chemical Classification System (ATC, World Health Organization, Geneva, Switzerland) class B01A of which the following categories of drugs were ordered at least once for the inpatients studied: vitamin K antagonists (ATC class B01AA), heparin (ATC class B01AB), platelet aggregation inhibitors excluding heparin (ATC class B01AC) and direct factor Xa inhibitors (ATC class B01AF). All drug prescriptions including “free-text” orders were part of the data analysis: We (i) manually mapped the most frequent orders lacking ATC information to the correct ATC code, (ii) used an automated character sequence comparison algorithm to map the same ATC code to similar orders and (iii) reviewed the automated mappings and revised where necessary. These steps were repeated several times to augment and improve our mapping table.

LOS appeared to increase by 10% per additional diagnosis in UGIB patients (Table 3). Thromboembolic events showed a clinically important trend toward increased LOS which almost reached statistical significance. Antithrombotic agents were independently associated with 46% longer LOS.

Patients were more likely of having a 30-day readmission with increasing year during the study period (Table 3). Cancer was the only comorbidity statistically significantly associated with a 30-day readmission, with an odds ratio almost three times as high as for UGIB patients without cancer. Clinically important trends toward 30-day readmissions were observed for patients with hypertension, diabetes, chronic kidney disease, chronic obstructive pulmonary disease (COPD) and finally anemia the only one nearly reaching statistical significance.

None of the investigated comorbidities were associated with an increased LOS. However, older patients stayed longer in the hospital and patients seemed to stay almost 10% longer per additional diagnosis. Heart valve disease and vascular disease were statistically significantly associated with a shorter LOS (Table 4). Antithrombotic agents were independently associated with a 25% increased LOS.

Atrial fibrillation/flutter and cancer were independently associated with 30-day readmissions with nearly three and nearly five times increased odds ratios, respectively. Potentially important trends toward 30-day readmissions were observed for patients with anemia, alcohol use disorders, vascular disease and also for patients on anticoagulants; however, none of those close to statistical significance (Table 4).