Data source
Data from this study was obtained from the NIS, which is a portion of the Healthcare Cost and Utilization Project and one of the largest all-payer inpatient health care databases. This database was a publically available database and is available to everyone. The information from NIS can be used to make national estimates of health care utilization, charges, quality, as healthcare resource use and clinical characteristics are included. Begin from 2012, the NIS went through a redesign aimed at representing over 95% of the US population. Detailed information is available according to the official website https://www.hcup-us.ahrq.gov, and on account of no patient-identifiable information, approval from ethical institutions is not required.
Study population
All patients aged ≥ 18 years with a primary diagnosis of T(A)AD from 2008 to 2017 were included. First of all, we assessed primary diagnosis variable to identify TAD (Thoracic Aortic Dissection) inpatients by utilizing the Clinical Modification diagnosis code (ICD-9-CM) 441.01, 441.03 and ICD-10-CM diagnosis code I71.01, I71.03. Then we assessed every procedure variable to distinguish T(A)ADs using criteria developed by Sachs et al. [18], with procedure codes for cardioplegia, valve repair, or operations on vessels of the heart, cardiopulmonary bypass or hypothermia which were exclusively performed on T(A)AD inpatients to a great extent. List of ICD-9/10 procedure codes for locating T(A)AD patients are listed in Additional file 1: Table S1. Considering that the WHO classifies obesity into class I (body mass index (BMI) of 30–34.99 kg/m2), class II (BMI of 35–39.99 kg/m2) and class III (morbid obesity; BMI of ≥ 40 kg/m2), we further grouped patients into non-obese, obese (BMI of 30–39.99 kg/m2) and morbid obese (BMI of ≥ 40 kg/m2). Then we identified BMI categories or obesity status of patients by non-obesity, obesity (V85.30-39, 278.00) and morbid obesity (V85.40-45, 278.01) based on ICD-9-CM codes and by obesity (Z68.30-39, E66.8, E66.9, E66.09), morbid obesity (Z68.40-45, E66.01) based on ICD-10-CM codes, according to previous document researches [19]. The process of the cohort selection was shown in Fig. 1 (p. 28).
Covariates and outcome measures
Patient-related variables from NIS were recorded including sex, age, race, elective admission or not, insurance type, income status, smoking, dyslipidaemia, coronary artery disease, prior stroke, long-term anticoagulants and antithrombotics and Elixhauser Comorbidity Index (ECI) which contains common comorbidities and was calculated to assess the severity of comorbidities for each admission. Hospital characteristics like location/teaching status, bed size, and region are also included. All the covariates are given in the first column of Additional file 2: Table S2.
The primary outcomes were the prevalence of morbid obesity in T(A)AD hospitalizations and the association between morbid obesity and in-hospital mortality, total cost. Total cost was derived from total charges in the NIS database by the use of cost-to-charge ratio and Consumer Price Index (CPI) [19]. The CPI was collected from the information published by the U.S. Department of Labor. Secondary outcomes included the association between related clinical factors and obesity/morbid obesity in T(A)AD inpatients involving patient-related and hospital-level variables.
Statistical analysis
Weighting and stratification methods are applied to obtain total national estimates, as the NIS database is based on a complex sampling design. To summarize the baseline characteristics of patients, we performed the analysis of variance and Kruskal–Wallis tests to compare distributions of continuous variables and used the chi-square test to compare differences in categorical variables.
The Cochran-Armitage trend test was conducted to analysis the temporal trend in prevalence of obesity and morbid obesity in T(A)AD patients and Cochran–Mantel–Haenszel test was used to analysis the trend of in-hospital mortality among three different weight categories. In order to evaluate the association between different obesity classes and in-hospital mortality, multivariable-adjusted logistic analyses are conducted by controlling potential confounders. Because of the right skewed distribution, we performed logarithmic transformation for total cost before using multivariable linear models. Furthermore, patient-level subgroup analyses were also explored through multivariable-adjusted logistic analyses.
Besides, we utilized univariate logistic analysis (model 1) and a multivariable-adjusted logistic analysis (model 2) to assess clinical factors associated with obesity and morbid obesity in T(A)ADs. The variables included in model 2 are clinically relevant variables based on the literature search and significant variables on univariate regression model 1 including age, sex, race, dyslipidaemia, long-term anticoagulants and antithrombotics, smoking, alcohol abuse, deficiency anemia, chronic blood loss anemia, congestive heart failure, depression, diabetes, drug abuse, hypertension, hypothyroidism, lymphoma, fluid and electrolyte disorders, renal failure. We also utilized the other two multivariable-adjusted logistic analyses (Model 3 and Model 4) as the sensitivity analyses. Age, sex, race, year of discharge, income level, insurance type, elective admission or not, hospital status and clinical factors involving smoking, prior stroke, dyslipidaemia, coronary artery disease, long-term anticoagulants and antithrombotics and 28 individual ECI comorbidities are adjusted for model 3. There's only one difference between the two models (Model 3 and Model 4) is that Model 4 uses the comorbidities score (ECI score) as a variable to estimate the outcome, which is one of the most recommended models at present (HCUP officially provides response variables).
We filled the missing data by the dominant category for categorical variables and the median for continuous variables referring to the previous studies [20] and the missing rate of all variables baseline characteristics have been reported in the Additional file 3: Table S3. R software (version 4.0.3) and the SAS version 9.4 (SAS Institute Incorporation, Cary, North Carolina, USA) are used for statistical analyses. Statistical significance was defined as a P < 0.05 on two-tailed testing.