Predictors of 30-Day Readmissions for Adrenal Insufficiency

A Retrospective National Database Study

Asim Kichloo; Zain El-amir; Hafeez Shaka; Farah Wani; Sofia Junaid Syed


Clin Endocrinol. 2021;95(2):269-276. 

In This Article

Materials and Methods

Design and Data Source

This was a retrospective cohort study involving adult hospitalizations for AI in the United States in 2018. Data were sourced from the Nationwide Readmissions Database (NRD) for 2018. The NRD is the largest publicly available all-payer inpatient healthcare readmission database drawn from the Agency for Healthcare Research and Quality (AHRQ) Healthcare Cost and Utilization Project (HCUP) State Inpatient Databases (SID).[6] The NRD 2018 contains discharge data from 28 geographically dispersed states accounting for 59.7% of the total US resident population and 58.7% of all US hospitalizations. It contains both patient-level information and hospital-level information. Up to 40 discharge diagnoses and 25 procedures are collected for each patient using the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM/PCS). Hospitals are stratified according to ownership control, number of beds, teaching status, urban/rural location and geographic region. The NRD allows for weighted analysis to obtain 100% of the US admissions within a given year.

Study Population

The study involved hospitalizations from the NRD with AI as the reason for index admission. Hospitalizations for AI included patients with principal diagnostic codes for primary adrenal insufficiency (E27.1), Addisonian crisis (E27.2), drug-induced adrenocortical insufficiency (E27.3), and other and unspecified adrenocortical insufficiency (E27.4). Hospitalizations with age less than 18, and December and elective hospitalizations were also excluded. December admissions were excluded when searching for index admissions, as these hospitalizations would lack data for at least 30 days following discharge to determine whether there was a readmission per the study design. Using unique hospitalization identifiers, index hospitalizations were identified and one subsequent hospitalization within 30 days was tagged as a readmission. Elective and traumatic admissions were excluded from readmissions. Comorbidity burden was assessed using Sundararajan's adaptation of the modified Deyo's Charlson Comorbidity Index.[7] A score of >3 has about a 25% 10-year mortality, while a score of 2 or 1 has a 10% and 4% 10-year mortality, respectively. This cut-off point was chosen as a means of assessment of increased risk of mortality.

Outcome Measures

The primary outcome was rate and reasons for 30-day readmission in patients with AI. Secondary outcomes included comparison of mortality, length of hospital stay (LOS), total hospital charges (THC), cost of hospitalization (COH) and predictors of 30-day all-cause readmissions of AI. The COH was obtained from the cost to charge ratio data provided by HCUP.[8] Mean length of stay in days, total hospital charges and hospital costs in USD are continuous variables; however, mortality is a dichotomous variable.

Statistical Analysis

We analysed the data using Stata® Version 16 software (StataCorp, Texas, USA). All analyses were conducted using the weighted samples for national estimates in adjunct with HCUP regulations for using the NRD database. Comorbidities were calculated as proportions of the cohort, and chi-squared test was used to compare these characteristics between the index and readmissions. Univariate regression was used to compare readmission mortality, LOS, THC and COH. A univariate Cox regression analysis was performed to identify variables with hazard ratios for 30-day readmission <0.20 to obtain confounders for readmission. To obtain independent predictors for readmission, a univariate screening using Cox regression analysis was performed to identify variables with hazard ratios for 30-day readmission with p values <0.20. The identified variables were used to generate a multivariate Cox regression model used to identify independent predictors for readmissions with p values <.05 set as threshold for statistical significance.

Ethical Considerations

The NRD database lacks patient identifiers. In keeping with other HCUP databases, the NRD database does not require Cook County Health Institutional Review Board approval for analysis.

Data Availability Statement

The NRD is a large publicly available all-payer inpatient care database in the United States, containing data on more than 18 million hospital stays. Its large sample size provides sufficient data for analysis across hospital types and the study of readmissions for relatively uncommon disorders and procedures.