Background
The Affordable Care Act implemented optional Medicaid expansion starting in 2014, but the association between Medicaid expansion and gynecologic cancer survival is unknown.
Objective
To evaluate the impact of Medicaid expansion by comparing 2-year survival among gynecologic cancers before and after 2014 in states that did and did not expand Medicaid using a difference-in-difference analysis.
Study Design
We searched the National Cancer Database for women aged 40 to 64 years, diagnosed with a primary gynecologic malignancy (endometrial, ovarian, cervical, vulvar, and vaginal) between 2010 and 2016. We used a quasiexperimental difference-in-difference multivariable Cox regression analysis to compare 2-year survival between states that expanded Medicaid in January 2014 and states that did not expand Medicaid as of 2016. We performed univariable subgroup difference-in-difference Cox regression analyses on the basis of stage, income, race, ethnicity, and geographic location. Adjusted linear difference-in-difference regressions evaluated the proportion of uninsured patients on the basis of expansion status after 2014. We evaluated adjusted Kaplan–Meier curves to examine differences on the basis of study period and expansion status.
Results
Our sample included 169,731 women, including 78,669 (46.3%) in expansion states and 91,062 (53.7%) in nonexpansion states. There was improved 2-year survival on adjusted difference-in-difference Cox regressions for women with ovarian cancer in expansion than in nonexpansion states after 2014 (hazard ratio, 0.88; 95% confidence interval, 0.82–0.94; P <.001) with no differences in endometrial, cervical, vaginal, vulvar, or combined gynecologic cancer sites on the basis of expansion status. On univariable subgroup difference-in-difference Cox analyses, women with ovarian cancer with stage III–IV disease ( P =.008), non-Hispanic ethnicity ( P =.042), those in the South ( P =.016), and women with vulvar cancer in the Northeast ( P =.022), had improved 2-year survival in expansion than in nonexpansion states after 2014. In contrast, women with cervical cancer in the South ( P =.018) had worse 2-year survival in expansion than in nonexpansion states after 2014. All cancer sites had lower proportions of uninsured patients in expansion than in nonexpansion states after 2014.
Conclusion
There was a significant association between Medicaid expansion and improved 2-year survival for women with ovarian cancer in states that expanded Medicaid after 2014. Despite improved insurance coverage, racial, ethnic, and regional survival differences exist between expansion and nonexpansion states.
Why was this study conducted?
Medicaid expansion has been associated with improved insurance coverage, early-stage diagnosis, and timely treatment of gynecologic cancers, but its impact on survival is unknown.
Key findings
Women with ovarian cancer had improved 2-year survival in the Medicaid expansion states than in nonexpansion states in a difference-in-difference analysis. Racial, ethnic, and geographic differences in survival exist between expansion and nonexpansion states.
What does this add to what is known?
Our findings demonstrate an association between Medicaid expansion and gynecologic cancer survival and add to existing evidence in support of improved insurance coverage throughout the United States.
Introduction
The Affordable Care Act (ACA) became law in 2010 with the goal of expanding affordable health insurance to more Americans. In 2014, the ACA allowed states to expand Medicaid to provide healthcare coverage to adults with incomes below 138% of the federal poverty level, which has been unequally adopted by different states within the United States. As a result of the ACA, there has been a substantial increase in access to and utilization of healthcare services for low-income and young adults, especially in primary care and preventative services. ,
Multiple studies have demonstrated increased cancer screening and insurance coverage for patients with cancer since the advent of the ACA, which has translated into more early-stage cancer diagnoses and improved cancer care in a variety of cancer sites. , In the field of gynecologic oncology, the ACA has been associated with improved early-stage cancer diagnoses, , receipt of fertility-sparing treatment, improved time to treatment after diagnosis , , and improved survival in some patients with endometrial cancer. Specifically, Medicaid expansion has demonstrated improved insurance coverage, early-stage diagnosis, and timely treatment of women with gynecologic malignancies , and has also demonstrated an association with improved survival in lung, breast, and colorectal cancers. Despite these encouraging studies, the ultimate impact of the ACA and its Medicaid expansion mandate on survival in women with gynecologic cancer is unknown.
The primary objective of our study was to use the National Cancer Database to compare 2-year survival among gynecologic cancer patients before and after the advent of Medicaid expansion in 2014 in states that did and did not expand Medicaid using a difference-in-difference (DID) analysis; our secondary objective was to identify subgroup differences in survival pre- and post-2014 on the basis of expansion status.
Materials and Methods
We used a retrospective cohort study design with DID analysis of women diagnosed with a primary gynecologic malignancy within the National Cancer Database (NCDB), which captures hospital-reported patient data from programs accredited by the American College of Surgeons and covers approximately 70% of new cancer diagnoses. , Our primary outcome was 2-year all-cause survival after a new diagnosis of gynecologic cancer. We censored patients who survived >24 months to ensure the same median follow-up for all groups. Cancer-specific survival is not available in the NCDB database.
Within the NCDB, we identified patients using the International Classification of Diseases for Oncology codes and included those diagnosed with primary uterine corpus (C54.0–C54.9), ovarian (C56.9), peritoneal (C48.1–48.2), fallopian tube (C57.0), cervix (C53.0–53.9), vulvar (C51.0–51.9), and vaginal (C52.9) cancers. Ovarian, peritoneal, and fallopian tubes were grouped together for analysis. For uterine corpus diagnosis codes, we excluded sarcoma, lipoma, and complex missed stromal neoplasm histologies; for ovarian and fallopian tube diagnosis codes, we excluded sex cord stromal, germ cell, and trophoblastic neoplasm histologies. Although we included outcomes data through 2017, we only included patients diagnosed with a new gynecologic malignancy between 2010 and 2016 after the implementation of the Affordable Care in 2010 and excluded those diagnosed in 2017 because of a lack of mortality information. We also excluded patients with missing data between their diagnosis and last contact or death and with unknown stage. We excluded women <40 years old, as data on Medicaid expansion status was suppressed for this group in the NCDB database and excluded women >64 years old because of universal Medicare coverage starting at age 65 years ( Table 1 ).
| Step | Selection criteria | Included |
|---|---|---|
| 1 | Primary uterine corpus (C54.0–C54.9), ovarian (C56.9), peritoneal (C48.1–48.2), fallopian tube (C57.0), cervix (C53.0–53.9), vulvar (C51.0–51.9), or vaginal (C52.9) cancer | 1,047,376 |
| 2 | Exclude histology codes 859–958 for fallopian and uterine corpus sites | 981,712 |
| 3 | Age 40–64 y | 572,061 |
| 4 | Exclude patients diagnosed in 2017 because of lack of mortality information | 527,389 |
| 5 | Exclude patients with missing months from diagnosis to last contact or death | 527,344 |
| 6 | Keep stage I–IV disease only | 464,626 |
| 7 | Keep January 2014 expansion (KY, NV, CO, OR, NM, WV, AR, RI, AZ, MD, MA, ND, OH, IA, IL, VT, HI, NY, and DE) and nonexpansion (TN, NC, ID, GA, FL, MO, AL, MS, KS, TX, WI, UT, SC, SD, VA, OK, NE, WY, and ME) states only | 284,132 |
| 8 | Exclude patients diagnosed before 2010 | 169,731 |
Next, we selected patients on the basis of their state’s Medicaid expansion status. The Medicaid expansion states in our study included 19 states (KY, NV, CO, OR, NM, WV, AR, RI, AZ, MD, MA, ND, OH, IA, IL, VT, HI, NY, and DE) that expanded their Medicaid program in January 2014. The nonexpansion states included the 19 states that had not expanded Medicaid (TN, NC, ID, GA, FL, MO, AL, MS, KS, TX, WI, UT, SC, SD, VA, OK, NE, WY, and ME) as of 2016. We did not include states that participated in “early” Medicaid expansion before 2014 or “late” Medicaid expansion between January 2014 and December 2016 to minimize bias owing to unequal timing of Medicaid expansion.
DID analyses provide a quasi-experimental method that attempts to account for both measured and unmeasured confounding to examine outcomes over time in response to a change that affects one group but not the other. , In our analysis, the differences in survival outcomes before and after 2014 were compared between the expansion and nonexpansion states, allowing us to examine the change in outcomes related to Medicaid expansion beyond background trends in cancer survival occurring in all states. To ensure that the trends in outcomes between the expansion and nonexpansion groups were the same before Medicaid expansion in January 2014, we tested for the parallel trends assumption by assessing the significance of the interaction term between time period and Medicaid expansion in a regression model adjusted for all examined covariates. There was no significant interaction term between time period and Medicaid expansion in our analysis, showing that our parallel trends assumption was met , ( Supplemental Table 1 ).
We assessed baseline characteristics including the cancer site, cancer stage using the American Joint Committee on Cancer Stage Group, age group, race, ethnicity, Charlson-Deyo score, health insurance (Medicaid, Medicare, uninsured, other government, private, and unknown), geographic location, median household income (categorized as quartiles), educational attainment (categorized as quartiles on the basis of high-school graduation rates) according to zip code, and urbanicity (metro, urban nonmetro, rural, and unknown). Among the January 2014 expansion and nonexpansion states included in our analysis, geographic location was grouped into Northeast (MA, ME, RI, VT, NY), Midwest (IL, OH, WI, IA, KS, MO, ND, NE, SD), South (DE, FL, GA, MD, NC, SC, VA, WV, AL, KY, MS, TN, AR, OK, TX), and West (AZ, CO, ID, NM, NV, UT, WY, HI, OR). We compared the baseline characteristics differences between the Medicaid expansion groups using chi-squared tests for categorical and t-tests for continuous variables. We assessed the 2-year adjusted survival probability by study period and expansion status ( Supplemental Table 2 ).
The pre- and post Cox proportional hazards model compared the change in 2-year survival between pre- and post-2014 study periods for the expansion and nonexpansion states adjusted for cancer type, cancer stage, age, race, ethnicity, Charlson-Deyo score, health insurance, geographic location, median income, education level, and urbanicity. We reported the hazard ratios (HR) and 95% confidence intervals (CIs) and computed the adjusted 2-year survival rate on the basis of the pre and post multivariable Cox regression models. An HR of <1 indicates an improvement in 2-year survival in the post-2014 study period compared with the pre-2014 study period. We then built a multivariable DID Cox regression model adjusted for the above patient covariates to evaluate the significance of the DID interaction term. We reported the HRs and 95% CIs, with a DID HR <1 indicating a greater improvement in 2-year survival in the expansion states than in the nonexpansion states after 2014.
We performed exploratory univariable subgroup Cox regressions to estimate the DID HRs on the basis of covariates of interest including stage (early I or II vs advanced III or IV), income (lowest vs highest quartile), race (Black vs White), ethnicity (Hispanic vs non-Hispanic), and geographic location (Northeast vs Midwest vs South vs West). As described above, we performed both a pre- and post-2014 Cox regression analysis and a DID Cox regression analysis. To assess whether the insurance coverage changed after Medicaid expansion, we also performed DID analyses for the outcome variable of insurance status (insured vs uninsured) adjusted for the time period of diagnosis, patient age, race, ethnicity, and primary cancer site. The models treated the binary outcome variable as a continuous variable, with the mean estimates representing the proportion of uninsured patients and the results presented as adjusted DID in percentage points with the associated 95% CIs.
Finally, Kaplan–Meier survival curves adjusted for patient age, stage, race, and ethnicity were generated and stratified according to the expansion status and year group for combined gynecologic malignancies and each primary cancer site. A significant P value indicates a difference between any of the generated Kaplan–Meier curves.
Comparisons were considered statistically significant using a 2-sided alpha level of 0.05. This study received institutional review board approval at the MD Anderson Cancer Center. Data analysis was done using SAS enterprise guide 7.1. (SAS Institute, Cary, NC). The Strengthening the Reporting of Observational Studies in Epidemiology guidelines were used in the preparation of this manuscript.
Results
We identified 169,731 patients who were diagnosed with a gynecologic malignancy after 2010 and met the inclusion criteria, including 78,669 (46.3%) in expansion states and 91,062 (53.7%) in nonexpansion states ( Table 1 ). Nonexpansion states had significant differences in all the examined covariates in the pre-2014 compared with the post-2014 study period ( Table 2 ). The expansion states had significant differences in all the examined covariates except the primary cancer site ( P =.08) and geographic location ( P =.25) in the post-2014 compared with the pre-2014 study period ( Table 2 ).
| Covariates | January 2014 expansion states (N=78,669) | Nonexpansion states (N=91,062) | ||||
|---|---|---|---|---|---|---|
| Pre-2014 (2010–13) n (%) | Post-2014 (2014–16) n (%) | P value | Pre-2014 (2010–13) n (%) | Post-2014 (2014–16) n (%) | P value | |
| Primary cancer site | .079 | <.001 a | ||||
| Cervical | 6262 (14.2) | 4883 (14.1) | 8553 (17.1) | 6943 (16.9) | ||
| Endometrial | 24,499 (55.6) | 19,566 (56.5) | 25,694 (51.3) | 21,868 (53.3) | ||
| Ovarian or peritoneal | 11,038 (25.1) | 8460 (24.4) | 12,960 (25.9) | 10,002 (24.4) | ||
| Vaginal | 427 (1.0) | 299 (0.9) | 549 (1.1) | 397 (1.0) | ||
| Vulvar | 1816 (4.1) | 1419 (4.1) | 2290 (4.6) | 1806 (4.4) | ||
| Age group (y) | <.001 a | <.001 a | ||||
| 40–44 | 3804 (8.6) | 2812 (8.1) | 4887 (9.8) | 4031 (9.8) | ||
| 45–49 | 5727 (13.0) | 4201 (12.1) | 6859 (13.7) | 5218 (12.7) | ||
| 50–54 | 9100 (20.7) | 6829 (19.7) | 10,179 (20.3) | 7976 (19.4) | ||
| 55–59 | 12,247 (27.8) | 9796 (28.3) | 13,286 (26.5) | 11,374 (27.7) | ||
| 60–64 | 13,164 (29.9) | 10,989 (31.7) | 14,835 (29.6) | 12,417 (30.3) | ||
| Race | <.001 a | <.001 a | ||||
| American Indian/Alaskan | 238 (0.5) | 160 (0.5) | 208 (0.4) | 237 (0.6) | ||
| Asian/Pacific Islander | 1769 (4.0) | 1667 (4.8) | 940 (1.9) | 931 (2.3) | ||
| Black | 3900 (8.9) | 3336 (9.6) | 6632 (13.3) | 5776 (14.1) | ||
| Other | 485 (1.1) | 543 (1.6) | 504 (1.0) | 569 (1.4) | ||
| Unknown | 578 (1.3) | 422 (1.2) | 326 (0.7) | 364 (0.9) | ||
| White | 37,072 (84.2) | 28,499 (82.3) | 41,436 (82.8) | 33,139 (80.8) | ||
| Ethnicity | <.001 a | <.001 a | ||||
| Hispanic | 2359 (5.4) | 2197 (6.3) | 4111 (8.2) | 3689 (9.0) | ||
| Non-Hispanic | 40,084 (91.0) | 31,753 (91.7) | 44,611 (89.1) | 36,494 (89.0) | ||
| Unknown | 1599 (3.6) | 677 (2.0) | 1324 (2.6) | 833 (2.0) | ||
| Charlson-Deyo score | <.001 a | <.001 a | ||||
| 0 | 34,200 (77.7) | 26,988 (77.9) | 38,805 (77.5) | 31,571 (77.0) | ||
| 1 | 7772 (17.6) | 5797 (16.7) | 9105 (18.2) | 7249 (17.7) | ||
| ≥2 | 2070 (4.7) | 1842 (5.3) | 2136 (4.3) | 2196 (5.4) | ||
| Health insurance | <.001 a | <.001 a | ||||
| Unknown | 849 (1.9) | 440 (1.3) | 1423 (2.8) | 957 (2.3) | ||
| Medicaid | 5039 (11.4) | 5839 (16.9) | 4855 (9.7) | 3551 (8.7) | ||
| Medicare | 3785 (8.6) | 3139 (9.1) | 4447 (8.9) | 3998 (9.7) | ||
| Not insured | 2658 (6.0) | 779 (2.2) | 5618 (11.2) | 3686 (9.0) | ||
| Other government | 477 (1.1) | 393 (1.1) | 1155 (2.3) | 990 (2.4) | ||
| Private insurance | 31,234 (70.9) | 24,037 (69.4) | 32,548 (65.0) | 27,834 (67.9) | ||
| Geographic location | .249 | .006 a | ||||
| Midwest | 16,726 (38.0) | 12,959 (37.4) | 8541 (17.1) | 6676 (16.3) | ||
| Northeast | 12,713 (28.9) | 10,070 (29.1) | 1057 (2.1) | 829 (2.0) | ||
| South | 8116 (18.4) | 6356 (18.4) | 38,755 (77.4) | 32,056 (78.2) | ||
| West | 6487 (14.7) | 5242 (15.1) | 1693 (3.4) | 1455 (3.5) | ||
| Median income | <.001 a | <.001 a | ||||
| First quartile | 6014 (13.7) | 4832 (14.0) | 11,834 (23.6) | 9520 (23.2) | ||
| Second quartile | 8072 (18.3) | 6063 (17.5) | 12,575 (25.1) | 9701 (23.7) | ||
| Third quartile | 9739 (22.1) | 7003 (20.2) | 11,087 (22.2) | 8093 (19.7) | ||
| Fourth quartile (richest) | 15,668 (35.6) | 11,694 (33.8) | 10,771 (21.5) | 8846 (21.6) | ||
| Not available | 4549 (10.3) | 5035 (14.5) | 3779 (7.6) | 4856 (11.8) | ||
| Educational attainment | <.001 a | <.001 a | ||||
| First quartile | 6525 (14.8) | 5756 (16.6) | 11,759 (23.5) | 10,215 (24.9) | ||
| Second quartile | 10,230 (23.2) | 7688 (22.2) | 13,364 (26.7) | 10,402 (25.4) | ||
| Third quartile | 12,362 (28.1) | 8452 (24.4) | 12,397 (24.8) | 9040 (22.0) | ||
| Fourth quartile (most educated) | 10,427 (23.7) | 7762 (22.4) | 8792 (17.6) | 6550 (16.0) | ||
| Not available | 4498 (10.2) | 4969 (14.4) | 3734 (7.5) | 4809 (11.7) | ||
| Locale | .002 a | <.001 a | ||||
| Metro | 35,301 (80.2) | 28,125 (81.2) | 38,619 (77.2) | 32,468 (79.2) | ||
| Rural | 627 (1.4) | 474 (1.4) | 1307 (2.6) | 909 (2.2) | ||
| Unknown | 1502 (3.4) | 1138 (3.3) | 1423 (2.8) | 1204 (2.9) | ||
| Urban nonmetro | 6612 (15.0) | 4890 (14.1) | 8697 (17.4) | 6435 (15.7) | ||
| Stage at diagnosis | <.001 a | <.001 a | ||||
| Stage I | 26,772 (60.8) | 20,828 (60.1) | 28,945 (57.8) | 23,570 (57.5) | ||
| Stage II | 3557 (8.1) | 2611 (7.5) | 4396 (8.8) | 3341 (8.1) | ||
| Stage III | 9109 (20.7) | 7249 (20.9) | 11,086 (22.2) | 9128 (22.3) | ||
| Stage IV | 4604 (10.5) | 3939 (11.4) | 5619 (11.2) | 4977 (12.1) | ||
Our pre- and postadjusted Cox regression models demonstrated that within the nonexpansion states, the 2-year survival improved significantly in the post-2014 compared with the pre-2014 study period for combined cancer sites with a 4% decreased hazard of death (HR, 0.96; 95% CI, 0.92–0.99) and for cervical cancer with a 13% decreased hazard of death (HR, 0.87; 95%, CI 0.81–0.93) ( Table 3 ). Within the expansion states, the 2-year survival significantly improved for combined cancer sites with a 10% decreased hazard of death (HR, 0.90; 95% CI, 0.87–0.94), for endometrial cancer with a 9% decreased hazard of death (HR, 0.91; 95% CI, 0.85–0.99), for ovarian cancer with a 12% decreased hazard of death (HR, 0.88; 95% CI, 0.82–0.94), and for cervical cancer with an 11% decreased hazard of death (HR, 0.89; 95% CI, 0.81–0.97) in the post-2014 compared with the pre-2014 study period ( Table 3 ). Vaginal and vulvar cancers had no significant survival differences after 2014 in either expansion or nonexpansion states.
| Cancer site | Nonexpansion states | January 2014 expansion states | Difference-in-difference HR b | P value | ||
|---|---|---|---|---|---|---|
| Post-2014 vs pre-2014 HR (95% CI) a | P value | Post-2014 vs pre-2014 HR (95% CI) a | P value | |||
| Combined cancer sites | 0.96 (0.92–0.99) | .015 c | 0.90 (0.87– 0.94) | <.001 c | 0.95 (0.90–1.00) | .05 |
| Endometrial | 0.95 (0.89–1.02) | .16 | 0.91 (0.85– 0.99) | .011 c | 0.95 (0.86–1.06) | .35 |
| Ovarian | 0.99 (0.93–1.05) | .67 | 0.88 (0.82–0.94) | <.001 c | 0.90 (0.82–0.98) | .016 c |
| Cervical | 0.87 (0.81–0.93) | <.001 c | 0.89 (0.81–0.97) | .007 c | 1.03 (0.92–1.15) | .67 |
| Vaginal | 1.01 (0.77–1.32) | .96 | 1.01 (0.73–1.41) | .93 | 1.08 (0.71–1.63) | .73 |
| Vulvar | 1.13 (0.94–1.35) | .20 | 1.14 (0.92–1.41) | .22 | 0.99 (0.75–1.30) | .92 |
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
