Abstract
Background:
Women with gynecologic cancer face socioeconomic disparities in care that impact survival outcomes. The Affordable Care Act offered states the option to expand Medicaid enrollment eligibility criteria as a means of improving timely and affordable access to care for the most vulnerable. Variable uptake of expansion by states created a natural experiment, allowing for quasi-experimental methods, which offer more unbiased estimates of treatment effects from retrospective data than traditional regression adjustment.
Objective:
We sought to use a quasi-experimental, difference-in-difference framework to create unbiased estimates of impact of Medicaid expansion on women with gynecologic cancer.
Study Design:
We performed a quasi-experimental retrospective cohort study from National Cancer Database files for women with invasive cancers of the uterus, ovary and fallopian tube, cervix, vagina, and vulva diagnosed 2008–2016. Using a marker for state Medicaid expansion status, we created difference-in-difference models to assess the impact of Medicaid expansion on outcomes of access to and timeliness of care. We excluded women under 40 due to suppression of the state Medicaid expansions status in the data, and women 65 and over due to universal Medicare coverage availability. Our primary outcome was rate of uninsurance at diagnosis. Secondary outcomes included Medicaid coverage, early-stage diagnosis, treatment at academic facility, and any treatment or surgery within 30 days of diagnosis. Models were run within multiple subgroups, and on a propensity-matched cohort to assess robustness of treatment estimates. The assumption of parallel trends was assessed with event study time-plots.
Results:
Our sample included 335,063 women. Among this cohort, 121,449 were from non-expansion states, and 213,614 from expansion states, with 79,886 post-treatment cases diagnosed after expansion took full effect in expansion states. Groups had minor differences in demographics, and we found occasional pre-period event study coefficients diverging from the mean, but outcome trends were generally similar between expansion and non-expansion states in the pre-period, satisfying the necessary assumption for difference-in-difference analysis. In a basic difference-in-difference model, January 2014 Medicaid expansion was associated with significant increases in insurance at diagnosis, treatment at academic facility, and treatment within 30 days of diagnosis. In an adjusted model including all states and accounting for variable expansion implementation time, there was a significant treatment effect of Medicaid expansion for reduction in uninsurance at diagnosis (−2.00%; 95%CI −2.3- −1.7; p<0.001), and increases in early stage diagnosis (0.80%; 95%CI 0.2–1.4; p=0.02), treatment at academic facility (0.83%; 95%CI 0.1–1.5; p=0.02), treatment within 30 days (1.62%; 95%CI 1.0–2.3; p<0.001), and surgery within 30 days (1.54%; 95%CI 0.8–2.3; p<0.001). Particularly large gains were estimated for women living in low-income zip codes, Hispanic women, and women with cervical cancer. Estimates from subgroup and propensity-matched cohorts were generally consistent for all outcomes besides early stage diagnosis and treatment within 30 days.
Conclusion:
Medicaid expansion was significantly associated with gains in access and timeliness of treatment for non-elderly women with gynecologic cancer. Implementation of Medicaid expansion could greatly benefit women in non-expansion states. Gynecologists and gynecologic oncologists should advocate for Medicaid expansion as a means of improving outcomes and reducing socioeconomic and racial disparities.
Keywords: quasi-experimental methods, affordable care act, health disparities, ovarian cancer, uterine cancer, cervical cancer, vulvar cancer, vaginal cancer, epidemiology
Introduction
Approximately 110,000 new cases of gynecologic cancer are diagnosed annually in the United States (US).1 Women without insurance coverage have lower rates of cancer screening, present with more advanced disease, and are less likely to receive evidence-based treatment, leading to racial and socioeconomic disparities in survival.2–4 The Affordable Care Act (ACA) was signed in 2010, with the aim of improving access to and affordability of care, primarily through insurance coverage expansions targeting non-elderly, low-income populations.5 ACA policies included extension of dependent coverage to age 26, establishment of insurance exchanges with income-based subsidies, prohibition of coverage discrimination due to pre-existing medical conditions, limits on out-of-pocket spending, and expansion of Medicaid eligibility. Medicaid expansion broadened required eligibility criteria from minimums that included no coverage guarantee for non-pregnant, childless adults, to covering all children and adults up to 133% of the federal poverty level.6,7 Unlike other ACA policies, secondary to a Supreme Court decision, Medicaid expansion was not implemented universally, but left as an option for individual states with variable uptake since ACA passage.
The aim of this study was to use a quasi-experimental, difference-in-difference framework to create unbiased estimates of the impact of Medicaid expansion, independent of other ACA policies, on the primary outcome of uninsurance rate at time of diagnosis, and other measures of access to and timeliness of care for women with gynecologic cancer.
Materials and Methods
We performed a retrospective cohort study of the National Cancer Database (NCDB), a quality improvement initiative of the American College of Surgeons and the American Cancer Society. NCDB contains hospital-based patient data from select cancer programs, including upwards of 70% of new cancer diagnoses in the United States.8 We selected women with International Classification of Diseases for Oncology9 codes for cancers of the uterus (C54.0-C54.9), ovary and fallopian tube (C56.9-C57.0), cervix (C53.0-C53.9), and vulva/vagina (C51.0-C52.9) diagnosed from 2008 to 2016. Women over 64 were excluded due to universal availability of Medicare coverage, and women under age 40 were excluded, as state Medicaid expansion status was not reported by NCDB for these patients (40.8% and 7.1% of observations, respectively). From this cohort, we further excluded women with unknown insurance status (1.9%), those without tissue confirmation of diagnosis (0.5%), and those with in situ disease (2.8%).
Variable uptake of Medicaid expansion across states created a natural experiment, facilitating a quasi-experimental difference-in-difference framework by which to assess the impact of the policy change.10 This methodology facilitates more unbiased estimates of treatment effect than can be calculated by traditional regression analyses of retrospective data, and can nearly approximate a randomized trial. Mathematically equivalent difference-in-difference estimates can be calculated via either (1) algebraic differences between group rates or (2) regression on an interaction term between post-period and treatment group status. These estimates are robust to differences between treatment and control groups, so long as the assumption of parallel trends is satisfied, whereby treatment and control groups have similar trends in the outcome prior to the policy change, and can be assumed to have changed in the same way over time in absence of the policy change.11
Medicaid expansion under the ACA occurred in three waves, and is reported by the NCDB by expansion status of the patient’s state of residence: (1) non-expansion, (2) early expansion, (3) January 2014 expansion, and (4) late expansion (see Figure 1). Nineteen non-expansion states chose not to expand Medicaid during the time period under study. Five early expansion states and Washington DC, sought waivers from the federal government to expand Medicaid enrollment eligibility early, with launch dates ranging from January 2010 to March 2011.12 Nineteen states instituted expansion on the standard date of January 2014. Finally, seven late expansion states chose to expand Medicaid after the standard program launch, with start dates ranging from April 2014 to July 2016.13
Figure 1. Map and timeline diagram depicting Medicaid expansion status and timing by state.
Color-coded map depicting state Medicaid expansion status (top), and with timeline depicting exact timing of expansion implementation for expansion states and corresponding definitions of full or partial treatment by state Medicaid expansions status and year as used for complex difference-in-difference models. Map created via MapChart.net.
We compared descriptive statistics between expansion and non-expansion states, before and after the policy change. We assessed for differences in age, non-white race (vs. white), Hispanic ethnicity (vs. non-Hispanic), cancer site (uterus, ovary or fallopian tube, cervix, vulva or vagina), Charlson comorbidity score (0, 1, ≥2), and residence in low-income and low-education zip codes (median annual income ≤$38,000, and ≥21% without high school degree respectively, corresponding to the bottom quartiles from 2008–2012 American Community surveys). Significance was assessed with Pearson’s χ2. Women with missing data (<0.3% of sample) were excluded from adjusted analyses.
Our primary outcome was uninsurance at diagnosis. Secondary outcomes of access to and timeliness of care included Medicaid coverage, treatment at academic facility (versus Community, Comprehensive Community, or Integrated Network facilities), early stage diagnosis (defined as Stage I or Stage II disease), and any treatment (surgery, radiation, systemic, or other therapy) and surgery (surgical diagnostic and/or staging procedure, excluding cases for which primary surgery was not performed) within 30 days of diagnosis. Treatment at academic facility and treatment within 30 days have previously been demonstrated to impact outcomes from gynecologic cancers in various settings,14–17 and treatment within 30 days was consistent with reporting by related studies of the ACA using the NCDB.18,19 Women with missing data for secondary outcomes were excluded from respective analyses (8.2% stage of diagnosis, 5.3% days to treatment, 14.8% days to surgery).
We first created a basic difference-in-difference model to demonstrate this methodology in its simplest form. We used linear regression to estimate the impact of January 2014 Medicaid expansion relative to non-expansion states (excluding early and late expansion states due to variable timing in implementation) in the pre-period (2008–2013) and post-period (2014–2016). These estimates are mathematically equivalent to the differences-in-differences by subtraction (see Appendix A).
We then created a complex model including the full sample, designed to accommodate the variable timing of Medicaid expansion across the different types of expansion states by including separate difference-in-difference indicators for partial treatment (observations from 2010 in early expansion states and 2014–2016 in late expansion states, during which some, but not all states in the respective group had expanded Medicaid), and full treatment (observations from 2011–2016 in early expansion states, and 2014–2016 in January 2014 expansion states). This model was further adjusted for age, race, ethnicity, comorbidity, cancer site, and low-income and low-education zip code of residence (see Appendix A).
To assess the robustness of our results, we first checked the sensitivity of our treatment effect estimate to inclusion of different groups of expansion states, and limiting within certain demographic subgroups. Next, in an attempt to account for different and changing demographics in expansion versus non-expansion states, we performed propensity matching on likelihood of living in an expansion state, matched within year of diagnosis. We then ran the above models on the matched cohort (see Appendix A).20
The assumption of parallel trends was first assessed with visual inspection of time-plots for similar trends in outcomes in the pre-expansion period. This was then formally tested with event studies regression on the outcomes. This technique uses interaction terms for expansion status with time relative to expansion (in order to accommodate differing timing of early and late expansion). These year-by-year coefficients are plotted and any violation of the assumption of parallel trends would be indicated by notable trends in the pre-expansion period interaction coefficients (see Appendix A).
We considered p<0.05 to be statistically significant. STROBE guidelines for observational research were followed. Study data was de-identified and granted exempt status by the University of Pennsylvania IRB. Analyses were conducted with STATA v15.1 (Statacorp, College Station, TX, USA). Figures were created with Microsoft PowerPoint/Excel v14.7.7 (Microsoft Corporation, Redmond, WA, USA).
Results
A total of 335,063 women met inclusion criteria. Among this cohort, 121,449 women (36.3%) came from states with no Medicaid expansion during the study. Of the 213,614 women from expansion states, 60,923 (28.5%) came from 6 early expansion states, 106,372 (49.8%) came from the 19 states expanding Medicaid in January 2014, and 46,319 (21.7%) came from the 7 late expansions states (see Figure 1).
Demographic characteristics were compared between pre- and post-expansion, and expansion and non-expansion states (see Table 1). As expected with non-randomized implementation, states that selected to expand Medicaid were different from those that chose not to expand. The gynecologic cancer populations in non-expansion states were older, more non-white, had a higher proportion of cervical cancer, and were more likely to reside in low-income and low-education zip codes. The pre- and post-periods were also statistically significantly different for most characteristics within both expansion and non-expansion states. However, with the exception of increases in the Hispanic (5.1% to 11%) and low-education (14% to 18%) rates, and decrease in the low-income (15% to 11%) rate for expansion states in the post-period, absolute differences in characteristic rates tended to be small, with statistical significance of driven by the large sample size. Propensity-score matching was performed, with balance demonstrated among post-matching demographic characteristics (see Table B.1).
Table 1.
Demographic characteristics of women ages 40–64 with gynecologic cancer before and after Medicaid expansion in expansion versus non-expansion states
| Characteristic | Medicaid Expansion States | No Medicaid Expansion States | P Value† | |||||
|---|---|---|---|---|---|---|---|---|
| Pre-expansion* n = 111,178 | Partial Expansion* n = 22,550 | Full expansion* n = 79,886 | P value | 2008–2013 n = 77,451 | 2014–2016 n = 43,998 | P value | ||
| Age | <0.001 | <0.001 | <0.001 | |||||
| 40–49 | 25,147 (23%) | 4,729 (21%) | 16,770 (21%) | 18,893 (24%) | 9,980 (23%) | |||
| 50–59 | 53,723 (48%) | 10,843 (48%) | 38,696 (48%) | 36,348 (47%) | 20,703 (47%) | |||
| 60–64 | 32,308 (29%) | 6,978 (31%) | 24,420 (31%) | 22,210 (29%) | 13,315 (30%) | |||
| Race/Ethnicity | ||||||||
| Non-White | 17,370 (16%) | 3,407 (15%) | 15,355 (19%) | <0.001 | 13,704 (18%) | 8,716 (20%) | <0.001 | <0.001 |
| Hispanic | 5,679 (5.1%) | 1,213 (5.4%) | 8,481 (11%) | <0.001 | 5,908 (7.6%) | 3,815 (8.7%) | <0.001 | <0.001 |
| Primary Cancer | <0.001 | <0.001 | <0.001 | |||||
| Uterus | 65,197 (59%) | 13,359 (59%) | 47,705 (60%) | 42,430 (55%) | 25,427 (58%) | |||
| Ovary | 25,351 (23%) | 4,999 (22%) | 18,159 (23%) | 18,025 (23%) | 9,366 (21%) | |||
| Cervix | 15,197 (14%) | 3,028 (13%) | 10,607 (13%) | 12,606 (16%) | 6,871 (16%) | |||
| Vulva/Vagina | 5,433 (4.9%) | 1,164 (5.2%) | 3,415 (4.3%) | 4,390 (5.7%) | 2,334 (5.3%) | |||
| Comorbidities‡ | <0.001 | <0.001 | <0.001 | |||||
| 0 | 87,194 (78%) | 17,730 (79%) | 63,868 (80%) | 60,051 (78%) | 33,692 (77%) | |||
| 1 | 19,099 (17%) | 3,701 (16%) | 12,494 (16%) | 14,087 (18%) | 7,872 (18%) | |||
| ≥2 | 4,885 (4.4%) | 1,119 (5.0%) | 3,524 (4.4%) | 3,313 (4.3%) | 2,434 (5.5%) | |||
| Patient Zip Code§ | ||||||||
| Low Income | 16,671 (15%) | 3,492 (16%) | 9,044 (11%) | <0.001 | 18,742 (24%) | 10,609 (24%) | 0.63 | <0.001 |
| Low Education | 15,832 (14%) | 3,010 (13%) | 14,713 (18%) | <0.001 | 17,214 (22%) | 9,991 (23%) | 0.07 | <0.001 |
Pre-expansion: Early expansion states 2008–2009, January 2014 expansion states 2008–2013, late expansion states 2008–2013. Partial expansion: earlyexpansion states 2010, late expansion states 2014–2016. Full expansion: early expansion states 2011–2016, January 2014 expansion states 2014–2016.
Comparison of patients from Medicaid expansion versus non-expansion states, including all years 2008–2016
Comorbidity by Charlson comorbidity score
Zip code characteristics defined by 2012 bottom quartiles: low income = median income ≤$38,000; low education = ≥21% without high school degree
The difference-in-difference framework is robust to these incongruities between groups, so long as the assumption of parallel trends is satisfied. This was first tested visually with time-plots for outcomes under study (see Figure 2, Figure 3). We identified no gross violations of our primary assumption visually. To test this statistically, event study regressions were run on each outcome on the original and propensity-matched cohorts (see Figure 4, Figure B.1, Figure B.2). In the pre-expansion years (t<0) we did observe some significant differences between treated observations and the omitted category. However, these are only marginally significant, and are not substantial in magnitude relative to the large level shift we observed at the time of expansion. Moreover, there were no consistent patterns to the differences in coefficients in the pre-period, indicating that the differences seen in the post-period were not likely driven by the continuation of pre-treatment differences in trends, and that the assumption of parallel trends is generally satisfied.
Figure 2. Time-plot of (A) uninsurance and (B) Medicaid rates for women ages 40–64 with gynecologic cancer, by state Medicaid expansion status.
Time-plots for outcomes of access to care for visual assessment of parallel trends in the pre-expansion time period (2008–2009 for early expansion; 2008–2013 for January 2014 and late expansion states). No gross violations of the parallel trends assumption were identified.
Figure 3. Time-plot of (A) early stage diagnosis, (B) treatment at academic facility, and (C) treatment within 30 days rates for women ages 40–64 with gynecologic cancer, by state Medicaid expansion status.
Time-plots for outcomes of secondary outcomes of timeliness of care for visual assessment of parallel trends in the pre-expansion time period (2008–2009 for early expansion; 2008–2013 for January 2014 and late expansion states). No gross violations of the parallel trends assumption were identified.
Figure 4. Event study time-plot of association of Medicaid expansion with uninsurance at diagnosis for women ages 40–64 with gynecologic cancer for (A) original cohort, (B) propensity-matched cohort.
Displayed according to time t relative to expansion year in order to accommodate variable implementation time. All coefficients estimated relative to group-level differences at t-2. Pre-treatment coefficients are centered by the overall pre-treatment mean so that the level differences in the post-period are presented relative to the relevant pre-period used in the main regression analysis.
We first present the basic difference-in-difference model comparing non-expansion to January 2014 expansion states (see Table 2). We found significant treatment effect of January 2014 Medicaid expansion on reduction in uninsurance at diagnosis (−2.12%; 95%CI −2.58- −1.67; p<0.001), and increases in Medicaid coverage (7.19%; 95%CI 6.64– 7.73; p<0.001), treatment at academic facility (2.22%; 95%CI 1.38–3.07; p<0.001), and treatment within 30 days of diagnosis (1.48% 95%CI 0.65– 2.31; p<0.001). The treatment effect for increase in Medicaid coverage was larger than the reduction in uninsurance, and balanced by a shift away from private coverage in expansion versus non-expansion states (−4.98%; 95%CI −5.8- −4.2; p<0.001).
Table 2.
Basic difference-in-difference model estimates for impact of January 2014 Medicaid expansion versus non-expansion on women ages 40–64 with gynecologic cancers
| January 2014 Expansion | No Medicaid Expansion | Difference-in-Difference* | ||||
|---|---|---|---|---|---|---|
| Outcome | 2008–2013 n = 69,085 | 2014–2016 n = 36,433 | 2008–2013 n = 77,451 | 2014–2016 n = 43,998 | Treatment Effect Estimate (95%CI) | P value |
| Insurance | ||||||
| Uninsured | 4,187 (6.06%) | 862 (2.31%) | 8,559 (11.1%) | 4,147 (9.43%) | −2.12% (−2.58 to −1.67) | <0.001 |
| Medicaid | 7,787 (11.3%) | 6,467 (17.3%) | 7,644 (9.9%) | 3,851 (8.75%) | 7.19% (6.64 to 7.73) | <0.001 |
| Medicare/other | 6,511 (9.42%) | 3,881 (10.4%) | 8,819 (11.4%) | 5,481 (12.5%) | −0.09% (−0.62 to 0.45) | 0.75 |
| Private | 50,600 (73.2%) | 26,077 (69.9%) | 52,429 (67.7%) | 30,519 (69.4%) | −4.98% (−5.76 to −4.19) | <0.001 |
| Early Stage (I/II) | 43,398 (68.8%) | 23,275 (67.9%) | 47,974 (67.3%) | 26,808 (66.0%) | 0.40% (−0.41 to 1.27) | 0.31 |
| Academic Facility | 35,265 (51.1%) | 20,021 (53.7%) | 29,204 (37.7%) | 16,777 (38.1%) | 2.22% (1.38 to 3.07) | <0.001 |
| ≤30 Days to Treatment | 43,537 (66.5%) | 22,628 (63.4) | 52,446 (70.7%) | 27,698 (66.1%) | 1.48% (0.65 to 2.31) | <0.001 |
| ≤30 Days to Surgery | 38,999 (65.8%) | 19,792 (61.5%) | 46,492 (70.2%) | 23,909 (64.7%) | 1.20% (0.32 to 2.07) | 0.007 |
Treatment effect estimates can be calculated either via linear regression model comparing non-expansion to January 2014 expansion states (outcome = α + β1(treatment) + β2(post) + βDiff-in-Diff(treatment*post)), or by subtraction of group rates (βDiff-in-Diff =((Posttreatment–Pretreatment)–(Postcontrol–Precontrol)). These are mathematically equivalent. 95%CI and p value derived from linear regression model.
We next present complex models accounting for the variable timing of Medicaid expansion implementation (see Table 3). From the full sample adjusted model, we find significant treatment effects for Medicaid expansion on all outcomes, including a reduction in uninsurance at diagnosis (−2.00%; 95%CI −2.3- −1.7; p<0.001), and increases in Medicaid (4.64%; 95%CI 4.2–5.1; p<0.001), early stage diagnosis (0.80%; 95%CI 0.2–1.4; p=0.02), treatment at academic facility (0.83%; 95%CI 0.1–1.5; p=0.02), treatment within 30 days (1.62%; 95%CI 1.0–2.3; p<0.001), and surgery within 30 days (1.54%; 95%CI 0.8–2.3; p<0.001).
Table 3.
Treatment effect difference-in-difference estimates from adjusted models on original and propensity-matched cohorts for impact of Medicaid expansion on women ages 40–64 with gynecologic cancer
| Original n = 334,177 | Propensity-Matched* n = 275,011 | |||
|---|---|---|---|---|
| Estimate (95%CI) | P value | Estimate (95%CI) | P value | |
| Insurance | ||||
| Uninsured | −2.00% (−2.3– −1.7) | <0.001 | −2.00% (−2.4– −1.6) | <0.001 |
| Medicaid | 4.64% (4.2– 5.1) | <0.001 | 4.69% (4.3– 5.2) | <0.001 |
| Early stage (I/II) | 0.80% (0.2– 1.4) | 0.02 | −0.57% (−1.3– 0.2) | 0.14 |
| Treatment at Academic facility | 0.83% (0.1–1.5) | 0.02 | 2.38% (1.6– 3.2) | <0.001 |
| ≤30 Days to Treatment | 1.62% (1.0– 2.3) | <0.001 | 0.66% (0.0– 1.3) | 0.053 |
| ≤30 Days to Surgery | 1.54% (0.8– 2.25) | <0.001 | 3.18% (2.4– 3.9) | <0.001 |
Adjusted model controls: year, expansion status, age, Charlson Comorbidity score, cancer site, low education zip code*, low income zip code*, non-white race, Hispanic ethnicity
Propensity-matching done to 20 nearest neighbors, with replacement, with maximum caliper 0.012, matching performed separately by year of diagnosis
The full treatment effect was robust to adjustment, and inclusion of early and late expansion states, with estimates ranging from −1.97% to −2.21% (see Table B.2). We also ran an identical model on the propensity-matched cohort and identified similar improvements in uninsurance at diagnosis (−2.00%; 95%CI −2.4- −1.6; p<0.001), Medicaid coverage (4.69%; 4.3–5.2; p<0.001), treatment at academic facility (2.38%; 1.6–3.2; p<0.001), and surgery within 30 days (3.18%; 2.5–3.9%; p<0.001), but no significant gains in early stage diagnosis or treatment within 30 days.
Finally, we ran the model within subgroups of the original sample corresponding to certain disadvantaged populations (Table 4) and individual cancer sites (Table 5). The difference-in-difference estimates were largely consistent across subgroups, with the exception of early stage diagnosis, which only met significance for increases within non-white women among subgroups assessed. The estimates of treatment impact on uninsurance rate were relatively larger for women from low-income zip codes (−4.45%; 95%CI −5.5- −3.4; p<0.001), Hispanic women (−3.60%; 95%CI −1.7- −5.5; p<0.001), and for women with cervical cancer (−3.66%; 95%CI −4.8- −2.5; p<0.001). The largest gains in rate of treatment within 30 days were seen for Hispanic women (3.28%; 95%CI 0.8–5.7; p=0.008), and for women with cervical cancer (3.19%; 95%CI 1.3–5.1; p=0.001) and vulvar/vaginal cancer (5.95%; 95%CI 2.7–9.2; p<0.001).
Table 4.
Treatment effect difference-in-difference estimates from adjusted models for impact of Medicaid expansion on women ages 40–64 with gynecologic cancer, within disadvantaged subgroups
| Low Income Zip Code* n = 58,558 | Non-White race n = 58,410 | Hispanic ethnicity n = 25,029 | ||||
|---|---|---|---|---|---|---|
| Estimate (95%CI) | P value | Estimate (95%CI) | P value | Estimate (95%CI) | P value | |
| Insurance | ||||||
| Uninsured | −4.45% (−5.5– −3.4) | <0.001 | −0.99% (−1.9– −0.1) | 0.04 | −3.60% (−5.5– −1.7) | <0.001 |
| Medicaid | 8.21% (6.8– 9.6) | <0.001 | 6.76% (5.5– 8.0) | <0.001 | 3.67% (1.6– 5.7) | <0.001 |
| Early stage (I/II) | 1.23% (−0.5– 2.9) | 0.15 | 2.41% (0.8– 4.0) | 0.003 | 0.65% (−1.7– 3.0) | 0.58 |
| Treatment at Academic facility | −1.44% (−3.2– 0.3) | 0.11 | 2.35% (0.7– 4.0) | 0.005 | −3.4% (−5.8– −0.9) | 0.008 |
| ≤30 Days to Treatment | 2.33% (0.6– 4.1) | 0.008 | 1.89% (0.3– 3.5) | 0.02 | 3.28% (0.8– 5.7) | 0.008 |
| ≤30 Days to Surgery | 1.83% (−0.1– 3.7) | 0.06 | 2.18% (0.4–3.9) | 0.02 | 2.66% (0.0– 5.3) | 0.05 |
Adjusted model controls: year, expansion status, age, Charlson Comorbidity score, cancer site, low education zip code*, low income zip code*, non-white race, Hispanic ethnicity
Zip code characteristics defined by 2012 bottom quartiles: low income = median income <$38,000; low education = >=21% without high school degree
Table 5.
Treatment effect difference-in-difference estimates from adjusted models for impact of Medicaid expansion on women ages 40–64 with gynecologic cancer, by primary cancer site
| Uterus n = 193,611 | Ovary/FT* n = 75,702 | Cervix n = 48,172 | Vulva/Vagina n = 16,592 | |||||
|---|---|---|---|---|---|---|---|---|
| Estimate (95%CI) | P value | Estimate (95%CI) | P value | Estimate (95%CI) | P value | Estimate (95%CI) | P value | |
| Insurance | ||||||||
| Uninsured | −1.68% (−2.1– −1.3) | <0.001 | −1.65% (−2.4– −0.9) | <0.001 | −3.66% (−4.8– −2.5) | <0.001 | −3.17% (−4.9– −1.5) | <0.001 |
| Medicaid | 4.17% (3.7– 4.7) | <0.001 | 4.02% (3.2– 4.9) | <0.001 | 7.40% (5.8– 9.0) | <0.001 | 4.47% (2.3– 6.7) | <0.001 |
| Stage I/II Diagnosis | 0.74% (0.0– 1.5) | 0.06 | 0.55% (−0.9– 2.0) | 0.47 | 1.34% (−0.5– 3.2) | 0.14 | 0.31% (−2.7– 3.3) | 0.84 |
| Treatment at Academic facility | 0.99% (0.1– 1.9) | 0.03 | 1.61% (0.2– 3.1) | 0.03 | −0.38% (−2.2– 1.4) | 0.68 | −−1.48% (−4.6– 1.7) | 0.36 |
| ≤30 Days to Treatment | 1.44% (0.5– 2.4) | 0.002 | 0.44% (−0.4– 1.3) | 0.30 | 3.19% (1.3– 5.1) | 0.001 | 5.95% (2.7– 9.2) | <0.001 |
| ≤30 Days to Surgery | 1.61% (0.7– 2.6) | 0.001 | 0.36% (−0.8– 1.5) | 0.54 | 4.74% (2.7– 3.6) | <0.001 | 3.73% (0.2– 7.3) | 0.04 |
FT = Fallopian tube
Adjusted models controls: year, expansion status, age, Charlson Comorbidity score, low education zip code*, low-income zip code*, non-white race, Hispanic ethnicity
Zip code characteristics defined by 2012 bottom quartiles: low income = median income <$38,000; low education = >=21% without high school degree
Comment
Principal Findings
We used quasi-experimental, difference-in-difference modeling to create unbiased estimates of the impact of Medicaid expansion on measures of access to and timeliness of care for non-elderly women with gynecologic cancers. We find statistically significant benefits of Medicaid expansion, including approximately a 2% reduction in uninsurance at diagnosis, our primary outcome, as well as consistent gains in Medicaid coverage, treatment at academic facility, and surgery within 30 days of diagnosis. We also identified gains in early stage diagnosis and treatment within 30 days in some models. We generally observed larger treatment effects for women from low-income zip codes, Hispanic women, and women with cervical cancer.
Results in Context
Several recent studies have assessed the impact of the ACA on women with gynecologic cancer. Smith and Fader used NCDB data and difference-in-difference techniques to show increases in insurance at diagnosis, early stage diagnosis, and fertility sparing treatment for women ages 21–26 from expanded dependent coverage, relative to those ages 27–35.21 The same group compared women with ovarian cancer ages 21–64 to those 65 and over and found increases in early stage diagnosis and treatment within 30 days from ACA changes.18 Our work differs in several important ways. First, our study is aimed to assess the impact of Medicaid expansion, independent of the other policies of the ACA that were universally implemented. Second, rather than comparing populations of different age ranges, we compare the same age range in expansion versus non-expansion states.
Another group used Surveillance, Epidemiology, and End Results data from 14 states to compare expansion to non-expansion states in 2011–2014, and found absolute reductions of 3–4% in uninsurance at diagnosis for gynecologic cancer patients,22 and similar impacts on patients with non-gynecologic cancers.23 Our NCDB study demonstrates similar results, with differences likely related to the different included cohorts, as we include all 50 states (versus 14), but we were unable to include women under age 40 due to data suppression by NCDB.
Prior work in non-gynecologic cancers has shown increases in timely treatment, particularly for African Americans with advanced/metastatic cancer,24 and small absolute increases in insurance at diagnosis and early stage diagnosis comparable to our findings from a similarly structured difference-in-difference study of the NCDB assessing patients with breast, colon, and lung cancers.19
Finally, we find interesting shifts in private insurance coverage. In our data, women were covered by private insurance at increased rates in non-expansion states, and decreased rates in expansion states. We cannot rule out an element of “crowd-out,” whereby low-income individuals are switching from private coverage to Medicaid in expansion states. However, prior work assessing for this effect from Medicaid expansion has shown that the observed divergence is mostly related to the fact that income-based subsidies for insurance on the state exchanges close some portion of the coverage gap in non-expansion states (those at incomes from 100 to 133% of the FPL are eligible for subsidies for private insurance), whereas in expansion states, these individuals are eligible for Medicaid, but not subsidies.25
Clinical Implications
Our results are notable for showing real clinical impacts of high-level policy changes. Although small in absolute differences, these small shifts would have a sizable impact in terms of number of patients. Assuming approximately 110,00 women will be diagnosed with gynecologic cancer in the US in 2020, a 1% shift in an outcome would correspond to a positive impact for upwards of 1,000 women annually.1 Improvements in access to and timeliness of care have been shown to have positive impact on outcomes in cancer patients in general,26–29 and gynecologic cancer patients specifically.17,30–32
Moving forward, non-expansion states still have the standing option to expand their Medicaid programs. Costs of expansion are covered 100% by the Federal government initially, decreasing to 90% over a phase-in period. Furthermore, state and federal policies such as Medicaid work requirements, and proposed shifts to block grant funding threaten to contract Medicaid coverage nationally. We provide evidence to support policies that protect or expand insurance coverage for our patients.
Research Implications
We observed small overall average declines in rates of diagnosis at early stage and treatment within 30 days of diagnosis. Medicaid expansion helped ameliorate this negative trend, and implementation of the program in non-expansion states may similarly benefit our patients, but we do not have a clear explanation for this troubling trend and it warrants investigation in the future.
Additionally, the impact of Medicaid expansion on women with gynecologic cancers can be further studied. Insurance coverage could impact receipt of evidence-based treatments by stage, within individual cancers. Furthermore, as data mature, the impact of Medicaid expansion on survival could be studied.
Strengths and Limitations
The strengths of our study include the large sample, representing about 70% of new gynecologic cancer diagnoses over nine years, encompassing all 50 states.
Additionally, quasi-experimental methods facilitated unbiased estimates of the treatment effect of the policy change, controlling for differences between groups, and secular changes over time, that would nearly approximate estimates from a study of a randomly applied policy change. We tested multiple specifications and a propensity-matched cohort to verify robustness of results.
Our primary limitations are related to the nature of the NCDB. Insurance status is defined as coverage at the time of diagnosis, but is subject to change over time (i.e. an uninsured patient may obtain coverage in response to cancer diagnosis). The NCDB is not population-based, and we cannot assess true population rates, nor account for substitution effects over time between reporting and non-reporting facilities. Concurrent variation in implementation of state insurance exchanges, or other similarly timed insurance related policy choices that affect trends in the post-expansion period could have some role in observed treatment effects, but these policies were largely consistent state to state. NCDB does not identify state of residence, which would allow more accurate modeling of the treatment effect. Women under 40 were excluded due to suppression of the treatment indicator by NCDB (12% of non-elderly women diagnosed with gynecologic cancer). Data from 2017 and beyond was not available at the time of analysis; therefore we were unable to assess the full treatment effect in late expansion states, where implementation occurred as late as July 2016.
Conclusions
We have demonstrated relevant clinical impacts on gynecologic cancer care for women within a short period after implementation of Medicaid expansion. Non-elderly women with gynecologic cancers, and likely those with other primary cancers, or non-cancer obstetric or gynecologic conditions,33–35 would benefit from Medicaid expansion by way of improved access. Gynecologists and gynecologic oncologists should continue to advocate for coverage expansion as a means of supporting our patients.
Supplementary Material
1. Condensation:
Using quasi-experimental methods, we demonstrate that Medicaid expansion was associated with gains in access to and timeliness of care for women ages 40–64 with gynecologic cancer.
2. AJOG at a Glance:
A. Why was the study conducted?
Within the Affordable Care Act, Medicaid expansion aimed to improve access to and timeliness of care for the most vulnerable Americans. Timely care can improve outcomes for women with gynecologic cancers. Medicaid expansion remains unimplemented by nineteen states and is an active topic of political debate.
B. What are the key findings?
We use quasi-experimental methods to demonstrate robust and unbiased evidence for the impact of Medicaid expansion on our primary outcome, reducing rate of uninsurance at diagnosis, as well as on secondary outcomes, increasing rates of early stage diagnosis, treatment at academic facility, and treatment within 30 days of diagnosis.
C: What does this study add to what is already known?
Medicaid expansion was associated with gains in access to and timeliness of care for women with gynecologic cancer, adding to a growing literature describing benefits for vulnerable populations achieved by insurance coverage expansion.
ACKNOWLEDGEMNTS
We have no acknowledgements to report.
4. Dr. Moss is supported by a BIRCWH (Building Interdisciplinary Research Careers in Women’s Health) career development award from the NIH. Dr. Ko is supported by a grant from Tesaro, unrelated to this work. Other authors have no funding to report.
5. This study is not a clinical trial.
6. This work has not been previously presented at a meeting.
7. No authors are employed by the Federal Government or Armed Forces.
Footnotes
The authors report no conflicts of interest.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
REFERENCES
- 1.Siegel RL, Miller KD, Jemal A. Cancer statistics, 2020. CA Cancer J Clin. 2020;70(1):7–30. doi: 10.3322/caac.21590 [DOI] [PubMed] [Google Scholar]
- 2.Temkin SM, Rimel BJ, Bruegl AS, Gunderson CC, Beavis AL, Doll KM. A contemporary framework of health equity applied to gynecologic cancer care: A Society of Gynecologic Oncology evidenced-based review. Gynecol Oncol. 2018;149(1):70–77. doi: 10.1016/j.ygyno.2017.11.013 [DOI] [PubMed] [Google Scholar]
- 3.Chen H-Y, Kessler CL, Mori N, Chauhan SP. Cervical cancer screening in the United States, 1993–2010: characteristics of women who are never screened. J Womens Health (Larchmt). 2012;21(11):1132–1138. doi: 10.1089/jwh.2011.3418 [DOI] [PubMed] [Google Scholar]
- 4.Jemal A, Lin CC, Davidoff AJ, Han X. Changes in insurance coverage and stage at diagnosis among nonelderly patients with cancer after the affordable care act. J Clin Oncol. 2017;35(35):3906–3915. doi: 10.1200/JCO.2017.73.7817 [DOI] [PubMed] [Google Scholar]
- 5.Obama BH. United States Health Care Reform: Progress to Date and Next Steps. JAMA. 2016;316(5):525–532. doi: 10.1001/jama.2016.9797 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Moy B, Polite BN, Halpern MT, et al. American Society of Clinical Oncology policy statement: opportunities in the patient protection and affordable care act to reduce cancer care disparities. J Clin Oncol. 2011;29(28):3816–3824. doi: 10.1200/JCO.2011.35.8903 [DOI] [PubMed] [Google Scholar]
- 7.Mitchell A. The ACA Medicaid Expansion; 2014. www.crs.gov. Accessed March 10, 2020.
- 8.Boffa DJ, Rosen JE, Mallin K, et al. Using the national cancer database for outcomes research: a review. JAMA Oncol. 2017;3(12):1722–1728. doi: 10.1001/jamaoncol.2016.6905 [DOI] [PubMed] [Google Scholar]
- 9.International Classification of Diseases for Oncology: Third Edition. Geneva, Switzerland; 2013. [Google Scholar]
- 10.Moss HA, Melamed A, Wright JD. Measuring cause-and-effect relationships without randomized clinical trials: Quasi-experimental methods for gynecologic oncology research. Gynecol Oncol. 2019;152(3):533–539. doi: 10.1016/j.ygyno.2018.11.006 [DOI] [PubMed] [Google Scholar]
- 11.Wing C, Simon K, Bello-Gomez RA. Designing Difference in Difference Studies: Best Practices for Public Health Policy Research Keywords. Annu Rev Public Heal. 2018;39:453–469. doi: 10.1146/annurev-publhealth [DOI] [PubMed] [Google Scholar]
- 12.States Getting a Jump Start on Health Reform’s Medicaid Expansion.; 2012. https://www.kff.org/health-reform/issue-brief/states-getting-a-jump-start-on-health/. Accessed February 2, 2020.
- 13.Status of State Medicaid Expansion Decisions: Interactive Map. Kaiser Family Foundation. https://www.kff.org/medicaid/issue-brief/status-of-state-medicaid-expansion-decisions-interactive-map/. Published 2020. Accessed February 2, 2020. [Google Scholar]
- 14.Cliby WA, Powell MA, Al-Hammadi N, et al. Ovarian cancer in the United States: Contemporary patterns of care associated with improved survival. Gynecol Oncol. 2015;136(1):11–17. doi: 10.1016/j.ygyno.2014.10.023 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Rauh-Hain JA, Starbuck KD, Meyer LA, et al. Patterns of care, predictors and outcomes of chemotherapy for uterine carcinosarcoma: A National Cancer Database analysis. Gynecol Oncol. 2015;139(1):84–89. doi: 10.1016/j.ygyno.2015.08.014 [DOI] [PubMed] [Google Scholar]
- 16.Seagle B, Butler S, Strohl A, … WN-N- G, 2017. undefined. Chemotherapy delay after primary debulking surgery for ovarian cancer. Elsevier https://www.sciencedirect.com/science/article/pii/S0090825816315785. Accessed February 10, 2020. [DOI] [PubMed] [Google Scholar]
- 17.Tewari KS, Java JJ, Eskander RN, Monk BJ, Burger RA. Early initiation of chemotherapy following complete resection of advanced ovarian cancer associated with improved survival: NRG oncology/gynecologic oncology group study. Ann Oncol. 2016;27(1):114–121. doi: 10.1093/annonc/mdv500 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Smith AJB, Fader AN. Association of the Affordable Care Act With Ovarian Cancer Care. Obstet Gynecol. 2020;135(2):257–265. doi: 10.1097/AOG.0000000000003630 [DOI] [PubMed] [Google Scholar]
- 19.Takvorian SU, Oganisian A, Mamtani R, et al. Association of Medicaid Expansion Under the Affordable Care Act With Insurance Status, Cancer Stage, and Timely Treatment Among Patients With Breast, Colon, and Lung Cancer. JAMA Netw open. 2020;3(2):e1921653. doi: 10.1001/jamanetworkopen.2019.21653 [DOI] [PubMed] [Google Scholar]
- 20.Austin PC. Optimal caliper widths for propensity-score matching when estimating differences in means and differences in proportions in observational studies. Pharm Stat. 2011;10(2):150–161. doi: 10.1002/pst.433 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Smith AJB, Fader AN. Effects of the affordable care act on young women with gynecologic cancers. Obstet Gynecol. 2018;131(6):966–976. doi: 10.1097/AOG.0000000000002592 [DOI] [PubMed] [Google Scholar]
- 22.Moss HA, Havrilesky LJ, Chino J. Insurance coverage among women diagnosed with a gynecologic malignancy before and after implementation of the Affordable Care Act. Gynecol Oncol. 2017;146(3):457–464. doi: 10.1016/j.ygyno.2017.06.012 [DOI] [PubMed] [Google Scholar]
- 23.Moss HA, Havrilesky LJ, Zafar SY, Suneja G, Chino J. Trends in Insurance Status Among Patients Diagnosed With Cancer Before and After Implementation of the Affordable Care Act. J Oncol Pract. 2018;14(2):e92–e102. doi: 10.1200/jop.2017.027102 [DOI] [PubMed] [Google Scholar]
- 24.Adamson BJS, Cohen AB, Estevez M, et al. Affordable Care Act (ACA) Medicaid expansion impact on racial disparities in time to cancer treatment. J Clin Oncol. 2019;37(18_suppl):LBA1–LBA1. doi: 10.1200/jco.2019.37.18_suppl.lba1 [DOI] [Google Scholar]
- 25.Frean M, Gruber J, Sommers BD. Premium subsidies, the mandate, and Medicaid expansion: Coverage effects of the Affordable Care Act. J Health Econ. 2017;53:72–86. doi: 10.1016/j.jhealeco.2017.02.004 [DOI] [PubMed] [Google Scholar]
- 26.Ward E, Halpern M, Schrag N, et al. Association of Insurance with Cancer Care Utilization and Outcomes. CA Cancer J Clin. 2008;58(1):9–31. doi: 10.3322/ca.2007.0011 [DOI] [PubMed] [Google Scholar]
- 27.Ellis L, Canchola AJ, Spiegel D, Ladabaum U, Haile R, Gomez SL. Trends in cancer survival by health insurance status in California from 1997 to 2014. JAMA Oncol. 2018;4(3):317–323. doi: 10.1001/jamaoncol.2017.3846 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Neal RD, Tharmanathan P, France B, et al. Is increased time to diagnosis and treatment in symptomatic cancer associated with poorer outcomes? Systematic review. Br J Cancer. 2015;112(1):S92–S107. doi: 10.1038/bjc.2015.48 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Chamoun K, De Lima MJG, Caimi PF, et al. Insurance status and survival of multiple myeloma (MM) patients. J Clin Oncol. 2019;37(18_suppl):LBA107-LBA107. doi: 10.1200/jco.2019.37.18_suppl.lba107 [DOI] [Google Scholar]
- 30.Elit LM, O’Leary EM, Pond GR, Seow H-Y. Impact of Wait Times on Survival for Women With Uterine Cancer. Obstet Gynecol Surv. 2014;69(3):143–144. doi: 10.1097/OGX.0000000000000038 [DOI] [PubMed] [Google Scholar]
- 31.Shalowitz DI, Epstein AJ, Buckingham L, Ko EM, Giuntoli RL. Survival implications of time to surgical treatment of endometrial cancers. Am J Obstet Gynecol. 2017;216(3):268.e1–268.e18. doi: 10.1016/j.ajog.2016.11.1050 [DOI] [PubMed] [Google Scholar]
- 32.AlHilli MM, Elson P, Rybicki L, Khorana AA, Rose PG. Time to surgery and its impact on survival in patients with endometrial cancer: A National cancer database study. Gynecol Oncol. 2019;153(3):511–516. doi: 10.1016/j.ygyno.2019.03.244 [DOI] [PubMed] [Google Scholar]
- 33.Eliason EL. Adoption of Medicaid Expansion Is Associated with Lower Maternal Mortality. Women’s Heal Issues. February 2020. doi: 10.1016/j.whi.2020.01.005 [DOI] [PubMed] [Google Scholar]
- 34.Gordon SH, Sommers BD, Wilson IB, Trivedi AN. Effects Of Medicaid Expansion On Postpartum Coverage And Outpatient Utilization. Health Aff (Millwood). 2020;39(1):77–84. doi: 10.1377/hlthaff.2019.00547 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Antonisse L, Garfield R, Rudowitz R, Guth M. The Effects of Medicaid Expansion under the ACA: Updated Findings from a Literature Review; 2019. https://www.kff.org/medicaid/issue-brief/the-effects-of-medicaid-expansion-under-the-aca-updated-findings-from-a-literature-review-august-2019/. Accessed March 10, 2020.
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.