We conducted a retrospective cohort study utilizing patient-level insurance claims data from Optum’s de-identified Clinformatics^® Data Mart Database (2000-2018) (CDM) and Medicare (2007-2017). CDM includes anonymized data on demographics, diagnoses, procedures, and medications dispensed. Data on providers includes clinical specialty and subspecialty, and whether healthcare services are provided by an individual provider or a medical facility. Medicare is a government-run health insurance plan for older (age >65 years) and disabled Americans. CDM data includes patients with managed care Medicare plans. We also used a 20% random sample of national fee for service Medicare data from 2007 to 2017. Medicare benefits are available in three separate plans. Part A covers inpatient care, part B covers outpatient care and part D includes pharmacy benefits. There is no overlap between CDM and Medicare data.

We included patients enrolled in the health plan on January 1, 2016 and who had a prevalent diagnosis of IBD prior to and remained enrolled on January 1, 2017. Prevalent diagnosis of IBD was based on our algorithm previously validated in the commercial healthcare population, which uses a combination of diagnoses and prescriptions for IBD therapies with positive predictive values ranging from 78% - 94% depending on diagnoses and prescriptions combinations.⁶ We excluded patients with missing race data, coverage gaps of more than 30 days in 2016, or with one or more diagnosis of psoriasis, psoriatic arthritis, rheumatoid arthritis, multiple sclerosis or ankylosing spondylitis prior to January 1, 2017. This was done to minimize missing data, lower the risk of misclassification, and increase certainty that prescriptions were for treatment of IBD. Medicare beneficiaries were required to have Parts A, B and D coverage on January 1, 2017. Patients exclusively categorized as a race or ethnicity other than Asian, Black, Hispanic or White, including Native American or Alaskan Native, were excluded.

The primary exposure was a composite variable consisting of race and ethnicity. This was chosen because CDM and Medicare similarly classify race and ethnicity as a single composite variable consisting of Asian, Black, Hispanic, White and Other categories. CDM race classification is from a proprietary algorithm based on geography and name. Medicare race classification uses enrollment information, Social Security, and a proprietary race imputation algorithm. The “Other” category was not included in the study due to heterogeneity. Given race and ethnicity are categorized as a single variable, “race” will hereafter refer to both “race and ethnicity” for clarity. Moreover, for the purpose of this research, race and ethnicity are recognized as social constructs rather than biological clades.

We measured healthcare utilization during 2017 using medications dispensed, outpatient encounters, emergency department encounters, hospitalizations, surgeries, lower endoscopy and cross-sectional imaging (Table 1).

Patients were categorized as CD, UC or IBD not otherwise specified (IBDnos) as in our prior research.¹ To be categorized as CD or UC, the majority of the IBD diagnoses and the most recent diagnosis before January 1, 2017 had to be CD or UC, respectively. All other patients were categorized as IBDnos. We collected data on covariates that were hypothesized a priori to be associated with race and measures of healthcare utilization. These were measured in 2016 and included age, sex or gender as recoded in the data, geographic region of residence (Midwest, Northeast, South, West U.S.), number of Elixhauser score comorbidities, number of non-IBD medications, number of non-gastroenterology office visits, estimated household income (in CDM only) and dual Medicare and Medicaid eligibility (in Medicare only).⁷ Socioeconomic risk adjustment used household income and dual Medicare and Medicaid eligibility as previously described.⁸ We also captured whether patients had a documented outpatient gastroenterology visit in 2016.

All analyses were performed for separately for children (age <20 years), adults aged 20-64 years, and adults aged 65 or older.

Descriptive statistics were used to compare characteristics and healthcare utilization by race. Statistical comparisons used chi-squared test for categorical variables and analysis of variance (ANOVA) for continuous variables. To formally test the association between race and healthcare utilization, we used generalized estimating equations to compute odds ratios (OR) and 95% confidence intervals (CI) clustered on the last gastroenterology provider seen in 2013 to 2016. If no gastroenterology outpatient encounter existed during this period, a unique pseudo-provider was assigned to the patient to maintain consistent clustering in the regression model. Race was included as a categorical variable with non-Hispanic White as the reference group because that group had the most individuals, which increased comparison stability. All models also included number of days of insurance coverage in 2017 to account for variable duration of follow-up.

We constructed statistical models in a stepwise fashion to assess for potential mediating covariates for both data sets. We started with models adjusted only for age, sex and IBD type as covariates. Second, we added geographic region of residence. Third, we added number of Elixhauser comorbidities, number of non-IBD medications, and non-gastroenterology office visits. The fourth iteration included household income categorized as <$50,000 per year, $50,000-$99,999, >$100,000 and missing within CDM, or dual Medicare and Medicaid eligibility for fee for service Medicare beneficiaries as a surrogate for socioeconomic status.⁹ The final a priori model added the number of gastroenterology visits in 2016. Additionally, we created a post hoc model to adjust for oral steroid, immunomodulator and advanced IBD therapy (e.g. biologics or JAK inhibitor) use in 2016 to account for disease severity. We qualitatively compared the strength of association in sequential models to identify covariates that were confounding or mediating the association of race with outcomes. Models were not implemented for healthcare utilization categories when any race category had fewer than 5 beneficiaries with utilization of the resource. Within Medicare, the IBDnos group was only included in analysis of use of advanced therapies, imaging and bowel resection surgery as small numbers prevented model convergence for other healthcare utilization outcomes when IBDnos was included. To aggregate data for those age 65 or older, we used fixed effects meta-analysis methods using the most adjusted models.

We report nominal p values for all comparisons and applied a correction for multiple testing. Because we made an a priori assumption to do all analyses stratified by age and have multiple racial groups, we used a false discovery rate (FDR) to determine significance.¹⁰ The FDR was applied separately to each age group and only to the models with the most covariates. We used an FDR less than 10% and a nominal p value less than 0.05 to define statistical significance.

Statistical power was estimated for the smallest possible groups that could be included in comparisons given the requirement for a minimum of 5 people with the healthcare utilization outcome. Power calculations assumed at least a 10:1 ratio of White patients to the smallest racial group. With 50 patients of whom five (10%) or 100 patients of whom five (5%) had the healthcare utilization outcome, we have 80% power to detect a relative risk of 2.4 and 2.5, respectively, before correcting for multiple comparisons.

Unadjusted rates of healthcare utilization are reported in Supplemental Table 1. Hispanic children were more likely to have undergone lower endoscopy (OR 1.61, 95% CI 1.04-2.49, FDR 0.239) and imaging (1.70, 95% CI 1.03-2.82, FDR 0.239) compared to non-Hispanic White children (Figure 1 and Supplemental Table 2). Steroid usage (1.53, 95% CI 0.99-2.37) and having a gastroenterologist visit (1.62, 95% CI 0.97-2.72) were only nominally higher in Hispanic children. In contrast, Asian (0.50, 95% CI 0.27-0.92, FDR 0.239) and Black children (0.63, 95% CI 0.37-1.07, FDR 0.239) were less likely than non-Hispanic White children to have seen a gastroenterologist. Post hoc analysis adjusted for IBD therapies in the year prior showed less use of advanced IBD therapy among Asian children and nominally lower use among Black and Hispanic pediatric patients (Supplemental Table 2). Notably, none of these findings reached the set FDR threshold. In general, the nominally significant associations among children increased in strength with each additional a priori adjustment of the regression models among Hispanic children, but not Black or Asian children (Supplemental Table 2).

Black Americans demonstrated a higher steroid, hospitalization, and emergency department use in minimally adjusted analyses, but these associations were attenuated in models that accounted for socioeconomic risks (Figure 2). Use of 5-aminosalicylates (5-ASA) was lower only in Black adults (0.86, 95% CI 0.78-0.96, FDR 0.031) compared to White subjects. Asian Americans had a less use of advanced therapies (0.69, 95% CI 0.57-0.83, FDR 0.001) but higher use of intravenous iron (1.94, 95% CI 1.28-2.93, FDR 0.013). Hispanic Americans used less advanced therapies (0.74, 95% CI 0.65-0.84, FDR <0.001), less steroids (0.88, 95% CI 0.79-0.98, FDR 0.098), and were less likely to have seen a gastroenterologist (0.84, 95% CI 0.76-0.94, FDR 0.013). Hispanic adults also had a numerically lower likelihood of undergoing lower endoscopy (0.88, 95% CI 0.79-0.99, FDR 0.116) and increased likelihood receive intravenous iron (1.41, 95% CI 1.02-1.96, FDR 0.135), though neither achieved the FDR cutoff of 10%. In contrast to children, the strength of associations among adults under age 65 generally were attenuated by adding additional covariates to the regression models (Supplemental Table 3).

Figure 3 and Supplemental Table 4 summarize the analyses evaluating the association between race and healthcare utilization in adults over 65. Older Asian Americans had decreased use of advanced IBD therapies (0.59, 95% CI 0.38-0.92, FDR 0.112), immunomodulators (0.67, 95% CI 0.48-0.93, FDR 0.112), steroids (0.62, 95% CI 0.45-0.86, FDR 0.032) and gastroenterology office visits (0.85, 95% CI 0.73-0.99, FDR 0.139) compared to White adults over 65. Black adults had lower use of steroids (0.77, 95% CI 0.70-0.85, FDR <0.001) and higher use of emergency department care (1.41, 95% CI 1.19-1.67, FDR 0.001). Healthcare use was also lower in older Hispanics with IBD with decreased use of glucocorticoids (0.75, 95% CI 0.67-0.84, FDR <0.001). Significant associations in advanced IBD therapy usage among Hispanic adults and gastroenterology visits across all racial groups disappeared after accounting for household income or dual Medicare and Medicaid eligibility. In the post hoc analysis, older black adults also demonstrated a significantly lower advanced IBD therapy use after adjustment for IBD treatment in the prior year. All other healthcare use metrics were not significantly affected by this analysis (Supplemental Table 4).