Examination of Elective Bariatric Surgery Rates Before and After US Affordable Care Act Medicaid Expansion

Key Points Question Was Affordable Care Act Medicaid expansion associated with increased uptake of elective bariatric surgery? Findings In this cohort study using a difference-in-differences analysis of 637 557 bariatric surgeries from 2010 to 2017 from 11 states that expanded Medicaid and 6 states that did not, Medicaid expansion was associated with a 31.6% annual increase in the rate of bariatric surgery during 2014-2017 among Medicaid-covered and uninsured non-Hispanic White adults aged 26 to 64 years. No increase was observed among non-Hispanic Black and Hispanic adults. Meaning This study suggests that additional policy changes and clinical programs may be necessary to address barriers disproportionately faced by racial and ethnic minority populations to ensure more equitable access to evidence-based treatment of obesity.


Identification of bariatric surgery
Our focus was on identifying primary elective bariatric surgery in the discharge data which include ICD-9-CM (2012 to 2015Q3) and ICD-10-CM / ICD-10-PCS (2015Q4-2017) codes. We did not study surgery revisions or repairs. As prior studies have differed in the choice of codes used we used two approaches. First, to be comprehensive we identified all codes used in any of the prior studies identified. Second, as use of some codes may have varied over time and across providers, we also identified counts based on the dominant single code used for laparoscopic Roux-en-Y Gastric Bypass (RYGB) gastric bypass and sleeve gastrectomy (SG) procedures across the study period (2012-2017) (these codes are shaded in the tables below). As we document below, these two codes accounted for an overwhelming share of the total volume in both the ICD-9 and ICD-10 periods. The list of prior studies we reviewed are listed below.
We identified elective primary bariatric surgeries as all discharges that meet the following conditions: 1. Use of RYGB, SG and other gastric procedure identified below. 2. Diagnosis of obesity identified as a principal or secondary diagnosis codes. 3. No diagnosis of abdominal neoplasm, ulcer or other conditions (identified below) identified in the principal or secondary diagnosis codes. 4. Patient admission was not through the emergency department. Such discharges were identified using the "Source" code in the discharge record.  • more than 30% of the discharges in the hospital had the race reported as "other"; • more than 50% of the discharges had no information on the race of the patient; • all of the discharges in the hospital had race coded as white, other, or missing; • 100% of the discharges had race coded as white and the hospital had more than 50 beds We followed the same procedure and excluded all bariatric surgeries from hospitals that met one of the above criteria (see eTable 3 below). As our data did not include number of hospital beds, we modified the fourth criterion and excluded hospitals with 100 percent of discharges coded as non-Hispanic White.
We used the combined categorization of race and ethnicity into the 5 groups identified in Table 1. This categorization was developed by AHRQ and the combined categorization field was included in the raw data we obtained. Notes: 1) These counts do not include the cases excluded for the criteria in eTable 3.
2) We find no noticeable change in surgery volume associated with shift from ICD9 to ICD10.The volume in the last ICD-9 quarter (2015Q3) was 23,101 and in the first ICD-10 quarter (2015Q4) was 24,273. This magnitude of change may also be due to (a) secular longitudinal increase in procedure volume, and (b) seasonal differences.
3) The first two procedure codes (0D164ZA, 0DB64Z3) account for 97.5 percent of the total volume during this period.

Estimation data and models:
The analytic data is longitudinal state-cohort level data for each of the three outcomes of interest: number of bariatric surgeries, count of census population and the rate of bariatric surgery (per 10,000 census population). Each regression was based on 264 observations, consisting of 17 states x (up to) 8 years x 2 age groups (26-64 and 65-74). As noted in Table 1, we had 8 years of data for all states, except WI (6 years) and AR and NY (7 years).
denotes the year indicator relative to the base year described above. We included fixed effects indicators for each state. Inclusion of state fixed effects adjusts for the unobserved time-invariant heterogeneity across areas by utilizing only the within-state changes in surgery counts; specifically, the estimates are robust to state-level differences in time-invariant characteristics, such as the sociodemographic composition, local healthcare environment (including provider availability) and rural location. We obtained standard errors clustered at the state level (Hansen 2020).
Our preferred specification was a log-linear model wherein we used the logarithm of the outcome measure ( ). This enables interpretation of the interaction estimates as percentage change in outcome in the relative year ( ) relative to that in the base year, associated with Medicaid expansion. Included in the relative years is the comparison of the three pre-expansion years (2010 to 2013). This estimate is an important indicator of the validity of the non-expansion states as a suitable comparator ("parallel trends test"); 2 since these two years precede Medicaid expansion, a significant non-zero estimate weakens the case for validity of the comparison group of state as it indicates the presence of systematic difference in the longitudinal trend in outcome measure prior to expansion. This three-way difference-in-differences regression model is estimated separately for each outcome for a variety of subgroups (insurance coverage, race/ethnicity, age and sex).
Alongside the above regression model, in each case we also estimated a modified version of the relative year wherein all the years following expansion are combined, to give the average change in surgery volume during the 2014-2017 period (relative to the reference year); this is denoted as 2 .

(Model 2)
As sensitivity analysis we also estimated the Poisson specification of the above linear models. 1) The estimates in each row are from a separate regression using observations for the respective age group. Each regression was based on 264 observations, consisting of 17 states x (up to) 8 years x 2 age groups (26-64 and 65-74). As noted in Table 1, we had 8 years of data for all states, except WI (7 years) and AR and NY (7 years).   1) The estimates in each row are from a separate regression using observations for the respective sex group. Each regression was based on 264 observations, consisting of 17 states x (up to) 8 years x 2 age groups (26-64 and 65-74). As noted in Table 1, we had 8 years of data for all states, except WI (7 years) and AR and NY (7 years). 2) Estimates reported here from the log-linear regression model specification (Model 1).
3) The estimated change in each of the three outcome measures associated with Medicaid expansion is obtained as 100*[exp(coefficient) -1] and denotes the percentage change in outcome in the expansion states among those aged 26-64 relative to those aged 65-75 within each state and those aged 26-64 in non-expansion states. The estimates of percent change in each outcome measure in the base year as the reference year. Base year is the year preceding the expansion year, which is 2013 for all states except Pennsylvania (for which 2014 was the base year). The percent change estimates reflect the change associated with Medicaid expansion. See Online Supplement for the model specification details. 4) Confidence intervals were obtained based on standard errors clustered at the state level. Estimates in bold indicate significance at p<0.05 level. 5) Note that the first two columns (baseline levels) give the absolute levels of the three outcome measures: total volume of bariatric surgery, count of census population by payer and rate of bariatric surgery (# surgeries per 10,000 census population by payer). The remaining columns are all in percentage terms, indicating the percentage change in the respective outcome measure associated with Medicaid expansion. eTable 9. Bariatric Surgery by Type As noted previously (eTable 2a, 2b), two procedures -laparoscopic Roux-en-Y Gastric Bypass (RYGB) gastric bypass and sleeve gastrectomy (SG) proceduresaccount for majority of bariatric surgeries. Over time the share of SG has been rising. Following tables quantify these trends separately by expansion and non-expansion states.