Evaluation of Cancer Care After Medicaid Expansion Under the Affordable Care Act

This cohort study evaluates different approaches for estimating the population of individuals eligible for Medicaid expansion under the Affordable Care Act in the context of researching changes in cancer care after Medicaid expansion.


Introduction
By 2017, more than 17 million previously uninsured people in the US had gained insurance through Medicaid expansion (ME), which was instituted as part of the Affordable Care Act of 2010. 1 Given prior associations between insurance status and cancer outcomes, there has been growing interest in the impact of ME on the 1.7 million people diagnosed with cancer annually in the US. 2 However, individuals more likely to benefit from ME would have been those previously uninsured who gained insurance by meeting eligibility criteria for ME, an estimated 76% of uninsured patients. 3 Focusing on expansion-eligible patients for research can be challenging because most cancer databases lack data on key eligibility criteria for expansion (income Յ138% of the federal poverty level and the absence of other insurance coverage opportunities).
Attempts to understand the impact of ME on cancer prevalence and care have approached study of the expansion-eligible population differently. Some have studied the entire cancer population, 4 whereas others have studied cohorts of patients likely to be ME eligible (eg, those with low incomes or in racial/ethnic minority groups). 5 Different approaches have resulted in varying proportions of expansion-eligible patients within study populations. Using the prevalence of stage I breast cancer diagnosis as an example of an outcome potentially impacted by access to insurance (ie, more insurance equals more screening and, consequently, earlier cancer diagnosis and care), we evaluated 3 previously described approaches and 2 novel approaches to focus on expansion-eligible patients. Five approaches to focusing on expansion-eligible patients were evaluated (Table); 3 of these have been previously reported. 4,5 The remaining 2 approaches have not previously been described, 1 restricting the cohort to Medicaid and uninsured patients before and after expansion, the other using propensity score matching to match Medicaid patients after expansion with uninsured patients before expansion (see the Table). The primary outcome was the proportion of newly diagnosed female patients who had clinical stage I breast cancer. Difference-in-difference (DiD) modeling comparing expansion and nonexpansion states was used to evaluate the outcome. 4,5 The parallel trends assumption was tested both graphically and through a linear regression, including an interaction between time and expansion status during the preexpansion period ( Figure). The interaction effect was not significant for all methods except for the "all patients" approach. Statistical significance was set at P < .05, and all tests were 2-tailed. Analyses were performed using SAS version 9.4 (SAS Institute Inc) and STATA version 16 (StataCorp LLC).

Methods
Author affiliations and article information are listed at the end of this article.
Open Access. This is an open access article distributed under the terms of the CC-BY License.

Results
In this cohort study of different approaches to focus on the population of patients with breast cancer who could have been expansion eligible, the study populations varied in size, from 231 938 (all patients) to 9036 (propensity score matched). The proportions of uninsured and Medicaid patients varied by cohort, ranging from 3.6% and 9.2% for all patients to 50% and 50% for matched patients (Table).
The percentage of the study population (women age 40 to 64 years) who had newly diagnosed stage I breast cancer was tracked at 3 time points: well before expansion (

Discussion
This retrospective cohort study suggests that different approaches to focusing on expansion-eligible patients are associated with different impressions of the influence of ME on patients with breast cancer. This phenomenon may not be limited to breast cancer; similar variability was seen for other screening-detectable cancers, including prostate, lung, and colon cancer (data available upon request). In the present study, as the proportion of uninsured and Medicaid patients increased within the approaches, the association of ME with a shift toward stage I diagnosis appeared to strengthen.
Propensity score matching, which simultaneously considered a range of predictors for expansioneligible status, was associated with the largest difference after expansion. This approach combined multiple factors associated with insurance status (unlike stratification approaches using solitary predictors such as race/ethnicity or income), potentially yielding the highest concentration of expansion-eligible patients (albeit with less sensitivity). It is important to note that each approach might have its own limitations; for example, the all-patients approach violated the parallel trends assumption (see Table for details).
It should also be noted that other populations could benefit from ME (eg, patients who were on high-deductible plans who were then eligible to switch to Medicaid), but those who were more likely to benefit from ME were individuals who were previously uninsured and gained insurance through ME.
Although there is no perfect way to focus on expansion-eligible patients to model the impact of ME on cancer care, investigators and their audiences should recognize the potential for variations among approaches. Efforts should be made to identify the approach that best aligns with the intended research question, even though it will likely reflect a trade-off between sensitivity and specificity for the expansion-eligible population.