The impact of obesity on hospitalized patients with ulcerative colitis

Data source

We used the NIS database for the year 2016. The NIS is an all-payer, annually collected abstraction of approximately 7 million hospital discharges from 46 states in addition to the District of Columbia, representing more than 97% of the US population. It is available through the Healthcare Cost and Utilization Project (HCUP) for research purposes. Since 2012, there have been changes in the sampling strategy of the NIS data to improve its precision and decrease sample error. These strategies include sampling 20% from all participating hospitals instead of sampling hospitals, using statewide data from the HCUP to feed the NIS database, and removing all identifiers of hospitals or states, as well as data not uniformly available in all states [20].

Patient selection

Patients who had a primary diagnosis of UC were identified using the tenth revision of the International Statistical Classification of Diseases and Related Health Problems (ICD-10) code (K51) in the first 2 discharge diagnoses. This code has been used in multiple previous studies [21,22].

We excluded all subjects less than 18 years of age, and anyone with a diagnosis of underweight (ICD-10 R636) or a BMI less than 20 kg/m² (ICD-10 Z681). We then defined patients with obesity based on ICD-10 codes of obesity or a BMI >30 kg/m² (Supplementary Table 1) and compared them to other patients in the study population.

The weighted Charlson Comorbidity Score [23-25], which unlike other indexes does not include obesity as one of its components, was calculated for all individuals in the study population, based on the ICD-10 code’s definition of the included comorbidities, and was used to adjust between obese and non-obese patients. Demography (age, race, sex), median household income for the patient’s ZIP Code based on the current year, the expected primary payer for the patient’s hospital stay, patient location (based on a 6-category urban–rural classification scheme for the United States counties developed by the National Center for Health Statistics), whether the patient was admitted to the hospital electively, and the type of facility that the patients were transferred from (if any) were considered as potential confounders.

Missing data

The randomness of the missing values in our data was evaluated using the missing data pattern plot to recognize if there was any systematic pattern. We used multiple imputations by chained equations [26], a method recommended by the HCUP for computing missing values in the NIS database [27]. Five imputed datasets were created; we ran the analyses repeatedly on each one and then combined the results.

Outcomes

Our primary outcomes of interest were: 1) length of hospital stay; 2) total charges of hospital admission; and 3) in-hospital mortality.

We also examined other secondary outcomes including: 1) rate of IBD-related surgeries; 2) venous thromboembolism; 3) infections, including pneumonia, urinary tract infection, and any bacterial or viral infection; and 4) cardiac events during the patients’ hospitalization.

Statistical analysis

R version 3.5.1 (R foundation for statistical computing, Vienna, Austria) was used for the analysis. Descriptive statistics were used to describe the baseline hospital and patient characteristics. We used the chi-square test to compare categorical variables and 2-group analysis of variance to compare continuous variables, while the Kruskal and Fisher tests were used for variables not fitting a normal distribution. We then estimated propensity scores by applying logistic regression models to all 5 imputed datasets to estimate the log odds of the probability of being obese based on the possible confounders listed above. The common support of propensity score distribution between the obese and non-obese groups was evaluated visually using Kernel density and box-and-whiskers plots. We then calculated the inverse probability of treatment weighting (IPTW) using propensity scores. We evaluated the covariate balance between the obese and non-obese groups to confirm adequate balance with acceptable absolute standardized effect sizes below 0.1 standard deviations [28]. All analyses were performed using survey-adjusted methods, accounting for NIS-specific hospital weighting and the IPTW. As a confirmatory measure, we re-estimated propensity scores using generalized boosting modeling (GBM) and compared the final results to the previous ones.

Descriptive analysis

We identified a total of 12,215 patients (61,075 weighted sample size) with a primary discharge diagnosis of UC. Among those, 1204 patients (6020 weighted sample size) fulfilled the criteria for obesity based on their BMI or ICD-10 coding of obesity. These patients tended to be older (54 vs. 51 years), were more likely to be female (62% vs. 54%), to be admitted electively to the hospital (28% vs. 20%), and to have Medicare (37% vs. 32%). Overall, patients with UC had a lower Weighted Charlson Comorbidity Score; however, obese patients tended to have a higher score (1.17 for patients with obesity vs. 0.77 for non-obese), and more specifically, a higher rate of congestive heart failure, chronic obstructive pulmonary disease, diabetes mellitus, mild liver disease and renal disease (Supplementary Table 2). Other characteristics in both groups are summarized in Table 1.

Table 1 Baseline characteristics of patients with UC comparing obese and non-obese patients

Missing data

The total amount of missing data was 6.8%. Most of the missing values referred to race (4.3%), followed by median household income for the patient’s ZIP Code (1.7%), indicator of a transfer into the hospital (0.5%), elective admissions (0.3%), patient location (0.3%), expected primary payer (0.1%), and sex (0.1%). The missing data pattern plot showed that data appeared to be missing at random (Supplementary Fig. 1).

Propensity score and IPTW

After propensity scores had been created using a regression model on all 5 imputed datasets, the common support was reviewed using Kernel density plots and box-and-whiskers plots and was judged by the authors as sufficient to use (Supplementary Figs. 2-5). IPTW was calculated for each individual based on their propensity score. The covariate balance was achieved by using the IPTW, based on absolute standardized effect sizes below 0.1 standard deviations in all the covariates (Supplementary Tables 3,4, Supplementary Fig. 6).

Outcomes

The length of hospital stay was greater in obese patients, with a mean difference (MD) of 0.57 days (95% confidence interval [CI] 0.22-0.93; P=0.002) compared to non-obese patients. Similarly, the MD of total charges was $6,341.71 higher (95%CI 2499.7-10183.71; P=0.001) obese patients compared to non-obese. There was no significant difference in hospital mortality, with an odds ratio (OR) of 0.28 and a wide CI that included the null value (95%CI 0.04-2.05; P=0.212; Fig. 1).

Figure 1 Outcomes plot of hospitalized patients with ulcerative colitis comparing obese and non-obese patients in relation to length of stay, total charges and hospital mortality

CI, confidence interval; MD, mean difference; OR, odds ratio

Our results from estimating propensity scores using GBM were similar, with length of stay MD 0.56 days (95%CI 0.14-0.97; P=0.008), total charges MD $5657.59 (95%CI 1439.66-9875.52; P=0.009), and no significant difference in hospital mortality with OR 0.56 (95%CI 0.08-4.07; P=0.566). Additionally, these results were consistent in subgroup analyses that stratified patients based on the presence of a simultaneous diagnosis of Crohn’s disease or other types of colitis, being overweight, or being younger than 65 years (Supplementary Figs. 7,8).

For the secondary outcomes, there was no statistical differences between the 2 groups in the rate of IBD-related surgery, venous thromboembolic events or the rate of infections, with ORs of 1.12 (95%CI 0.88-1.43; P=0.354), 1.21 (95%CI 0.94-1.55; P=0.13), and 1.24 (95%CI 0.96-1.6; P=0.1), respectively. On the other hand, obese patients appeared to have a higher risk of having myocardial infarction, with an OR of 1.26 (95%CI 1.03-1.55; P=0.025).