1 INTRODUCTION

Tens of millions of United States (US) residents have gained increased access to insurance coverage due to the expansion of Medicaid eligibility that resulted from the Patient Protection and Affordable Care Act (ACA).1, 2 Prior to ACA, Medicaid eligibility was state-based but typically included the elderly (age 65+), persons with disabilities, pregnant women, and low-income families. Post-ACA, nonelderly adults with incomes ≤138% of the federal poverty line became eligible for Medicaid in states that participate in expansion.2 Medicaid expansion has improved the affordability of and access to healthcare, with subsequent lower rates of noninsurance and improvement in outcomes for various conditions.3-5

Within oncology, Medicaid expansion has been associated with lower rates of noninsurance among nonelderly patients with new cancer diagnoses.1, 6, 7 Researchers have demonstrated increased rates of screening, early-stage diagnoses, and/or timely treatment for various screening-amenable malignancies including cervical cancer in expansion states.1, 3, 7, 8 Medicaid expansion has also been associated with greater use of cancer surgery among low-income US residents9 as well as diminished racial and socioeconomic disparities in oncologic care access.8, 10 Given noninsurance predicts poorer outcomes and mortality in cancer patients,11-18 Medicaid expansion also has the potential to improve survival in patients with cancer but data is limited. One study of patients with lung cancer found that Medicaid was not associated with improved overall survival compared with noninsurance, suggesting the need for further intervention at the policy level.19 A recent study found that in patients with lung, colorectal, and breast cancer had improved survival with Medicaid expansion.20

Cervical cancer is a highly screenable cancer in which Medicaid expansion may have a significant impact. There are approximately 126 929 people alive in the US who were diagnosed with cervical cancer from 2001 to 2016. In 2017, 12 831 new cases were reported and 4207 people died of cervical cancer in the US.21 Studies have shown that patients who are uninsured, of lower socioeconomic status, or members of racial/ethnic minority groups have lower rates of cervical cancer screening,22, 23 which is associated with late-stage disease at diagnosis.24, 25 Studies conducted prior to the ACA have also demonstrated that noninsurance is associated with increased rates of late-stage disease at presentation among patients with cervical cancer,26 and also partially mediate the increased risk of cervical cancer-specific mortality in non-Hispanic Black patients, compared with non-Hispanic White patients.27 Finally, increased time to treatment is associated with worse cervical cancer outcomes.28 Currently, there exists limited data on the effect of Medicaid expansion on cervical cancer stage at diagnosis, treatment, and mortality in Medicaid expansion states compared with nonexpansion states.7, 29, 30 Accordingly, the impact of Medicaid expansion on cervical cancer outcomes merits further study.

Given studies suggesting the deleterious effects of noninsurance on patients with cervical cancer,16, 23-25, 27 we hypothesized that Medicaid expansion would be associated with improved access to care and screening, and thus earlier stage at diagnosis, greater rates of timely treatment, and improved overall survival in cervical cancer. Using a nationally representative database, we assessed whether Medicaid expansion was associated with changes in cervical cancer outcomes. We examined the interaction between residence in a Medicaid expansion state and diagnosis in the postexpansion period with regards to insurance coverage, stage at diagnosis, timely treatment, and survival.

2 METHODS 2.1 Data source and patients

Data were obtained from the National Cancer Database (NCDB), a nationwide hospital-based cancer registry, which captures approximately 70% of new cancer diagnoses in the US from 30% of all US hospitals and contains data on patient, tumor, treatment, and hospital characteristics as well as survival.31, 32 Our study population included patients (aged 40–64 years) from 2011 to 2015 with newly diagnosed, invasive cervical cancer. Patients <40 years were excluded as NCDB does not report expansion status for this age range. Patients >65 years were excluded as routine screening for cervical cancer is not recommended for these individuals. Those on Medicare were also excluded. Patients diagnosed in 2016 were excluded as survival data were not available for this group. Patients with noninvasive in situ cancers (0.2%) or missing sociodemographic, clinical, geographic, and treatment variables (16%) were excluded. For the analysis of FIGO (International Federation of Gynecology and Obstetrics) stage at diagnosis as the outcome, those with missing/unknown stage were excluded (29%). For the analysis of time to treatment, patients who died <30 days after diagnosis were excluded (0.6%). For analyses with stage at diagnosis as a covariate, an indicator variable was included in the model for missing/unknown stage so that all patients could be included in the analyses. NCDB reports Medicaid expansion status for the state of patient's residence at time of diagnosis as “nonexpansion states,” “January 2014 expansion states,” “early expansion states (2010–2013),” “late expansion states (after January 2014).” We restricted our analyses to include patients from 19 states that expanded Medicaid in January 2014 for the Medicaid expansion states, excluding patients residing in states that expanded Medicaid before (six states) or after (seven states) January 1, 2014, since these patients may dilute the effect of Medicaid expansion that occurred on January 1, 2014. This study was approved by the institutional review board.

2.2 Study design

We performed a difference-in-differences (DID) cross-sectional analysis from 2011 to 2015 to compare insurance status, stage at diagnosis, time to treatment, and survival among cervical cancer patients residing in Medicaid expansion and nonexpansion states before (2011–2013) and after (2014–2015) Medicaid expansion. January 1, 2014 was used as the timepoint for Medicaid expansion as it corresponded with the time when 19 states expanded Medicaid.

2.3 Independent variables

The primary independent variables of the study were residence in a Medicaid expansion state, diagnosis in the postexpansion period (2014–2015), and the interaction between the two.

2.4 Outcomes

The primary outcomes of interest were: (1) insurance status (uninsured, Medicaid), (2) FIGO stage at diagnosis (curable [stage I–III], metastatic [stage IV]), (3) time to initial cancer-directed treatment (surgery, radiation, or systemic therapy within 30 and 90 days from diagnosis), and (4) overall survival.

2.5 Control variables

All models were adjusted for age, Charlson/Deyo comorbidity score,33 and urban/rural location.

2.6 Statistical analyses

Chi-square and Kruskal-Wallis tests were used to compare categorical and continuous variables, respectively. Multivariable linear regression was used to calculate the adjusted DID estimates for insurance status, stage at diagnosis, and timely treatment as a function of residing in a Medicaid expansion state, diagnosis in the postexpansion period, and an interaction between the two; variables included in the model were: age, Charlson/Deyo comorbidity score,33 and urban/rural location. Linear models were used based on previous DID studies1 and given they provide easily interpretable percentage point estimates of absolute changes.34, 35 Cox proportional hazard models were used to conduct analogous multivariable analyses of survival since diagnosis. The Kaplan-Meier method was used to generate survival curves and log-rank tests were used to compare the survival curves. A two-sided p < .05 was considered as statistically significant. All statistical analyses were performed using Stata/SE 15.1 in May 2020. Four sensitivity analyses were performed (Appendix Methods 2).

3 RESULTS 3.1 Patient characteristics

A total of 15 265 cervical cancer patients diagnosed between 2011 and 2015 were included in the study. We first compared baseline characteristics of patients in Medicaid expansion (n = 6351) and nonexpansion (n = 8914) states (Table 1). There were no significant differences in age, histology, FIGO stage, comorbidities, diagnosis year, chemotherapy, or radiation treatment. Compared with those living in nonexpansion states, patients residing in expansion states were less likely to be Black (13% vs. 19%) and uninsured (8% vs. 17%) and were more likely to have Medicaid (31% vs. 23%), a higher median household income ($50 354–63 332 vs. $40227–50 353), receive care at an academic cancer center (53% vs. 42%), have a shorter distance to the hospital facility (11 vs. 14 miles), and have had a hysterectomy (46% vs. 43%).

TABLE 1. Baseline patient and treatment characteristics among expansion versus nonexpansion states Variables Expansion states n = 6351 Nonexpansion states n = 8914 p-value Sociodemographic Age (year), median (IQR) 50 (45–56) 50 (45–56) .16 Race, % <.01 White 4998 (78.7) 6848 (76.8) Black 841 (13.2) 1668 (18.7) Asian 342 (5.4) 235 (2.6) Other 170 (2.7) 163 (1.8) Insurance (%) <.01 Noninsured 500 (7.9) 1550 (17.4) Medicaid 1954 (30.8) 2035 (22.8) Other (private, other government) 3897 (61.4) 5329 (59.8) Median household incomea (%) <.01 < $40 227 1461 (23.0) 2795 (31.4) $40 227–50 353 1451 (22.9) 2439 (27.4) $50 354–63 332 1431 (22.5) 1921 (21.6) ≥ $63 333 2008 (31.6) 1759 (19.7) Clinical Histology (%) <.01 Squamous cell carcinoma 4217 (66.4) 6060 (68.0) Adenocarcinoma 1824 (28.7) 2368 (26.6) Other 310 (4.9) 486 (5.5) FIGO stage at diagnosis (%) .09 I 2319 (36.5) 3255 (36.5) II 900 (14.2) 1383 (15.5) III 813 (12.8) 1173 (13.2) IV 463 (7.3) 606 (6.8) Missing/Unknown 1856 (29.2) 2497 (28.0) Charlson/Deyo score (%) .22 0 5382 (84.7) 7497 (84.1) 1 774 (12.2) 1120 (12.6) 2 145 (2.3) 199 (2.2) ≥ 3 50 (0.8) 98 (1.1) Geographic Facility location (%) <.01 Atlantic 2312 (36.4) 3944 (44.3) Central 2971 (46.8) 2643 (29.7) Southwest 66 (1.0) 2100 (23.6) Pacific 1002 (15.8) 227 (2.6) Rural/Urban (%) <.01 Urban 6260 (98.6) 8707 (97.7) Rural 91 (1.4) 207 (2.3) Hospital facility type (%) <.01 Community cancer program 447 (7.0) 433 (4.9) Comprehensive community cancer program 1842 (29.0) 3279 (36.8) Academic/research program 3367 (53.0) 3724 (41.8) Integrated network cancer program 695 (10.9) 1478 (16.6) Distance to hospital (miles), median (IQR) 10.5 (4.7–27.2) 14.0 (6.6–33.4) <.01 Treatment Chemotherapy (%) .76 Yes 3892 (61.3) 5484 (61.5) None 2459 (38.7) 3430 (38.5) Radiation (%) <.01 None 2236 (35.2) 3168 (35.5) EBRT alone 1589 (25.0) 2118 (23.8) Brachytherapy alone 799 (12.6) 1282 (14.4) EBRT + brachytherapy 1727 (27.2) 2346 (26.3) Surgery (%) <.01 None 2938 (46.3) 4351 (48.8) Local excision 524 (8.3) 695 (7.8) Hysterectomy 2889 (45.5) 3868 (43.4) Year of diagnosis (%) .52 2011 1202 (18.9) 1709 (19.2) 2012 1281 (20.2) 1716 (19.3) 2013 1218 (19.2) 1741 (19.5) 2014 1314 (20.7) 1910 (21.4) 2015 1336 (21.0) 1838 (20.6) Abbreviations: EBRT, external beam radiation therapy; FIGO, International Federation of Gynecology and Obstetrics; IQR, interquartile range. 3.2 Insurance status

Adjusted trends in insurance status at the time of diagnosis are presented in Figure 1(A) (uninsured) and Figure 1(B) (Medicaid), stratified by Medicaid expansion status. Parallel trends were observed for insurance status in the pre-ACA period (2011–2013; p > .05). The unadjusted difference and adjusted DID for insurance status are shown in Table 2. From the pre- to postexpansion period, both expansion (difference = −6.2%, 95%CI = −7.5, −4.9) and nonexpansion (difference = −3.2%, 95%CI −4.8, −1.6) states had a significant decrease in the proportion of uninsured patients. These decreases were significantly different between expansion and nonexpansion states (adjusted DID = −3.0%, 95%CI, −5.2, −0.8). There was a significant increase in the proportion of patients covered by Medicaid in expansion states (difference = 7.5%, 95%CI = 5.2, 9.8) and a significant decrease in the proportion covered by Medicaid in nonexpansion states (difference = −3.3%, 95%CI = −5.1, −1.6) from the pre- to postexpansion period. These changes were significantly different (adjusted DID = 11.0%, 95%CI = 8.2, 13.8). Changes in insurance coverage by year in expansion and nonexpansion states are shown in Appendix Table 1.

Adjusted Trends in Health Insurance Status (A and B), Cancer Stage at Initial Diagnosis (C and D), and Timely Treatment (E and F) for Cervical Cancer in Medicaid Expansion versus Nonexpansion States: (A) Uninsured, (B) Medicaid, (C) Curable Stage (Stages I–III) Cancer, (D) Metastatic Stage (Stage IV) Cancer, (E) Time to Treatment within 30 days of Diagnosis, (F) Time to Treatment within 90 days of Diagnosis. Participants include patients aged 40–64 years old diagnosed with cervical cancer between January 1, 2012 to December 31, 2015 from the National Cancer Database. Error bars show 95% confidence intervals of estimated margins. The vertical red line represents January 1, 2014, the date of Medicaid expansion

TABLE 2. Changes in insurance status, cancer stage at initial diagnosis, and timely treatment in medicaid expansion versus nonexpansion states Expansion states Nonexpansion states Adjusted DID (95% CI) and DID p-value Before After Unadjusted diff (95% CI) Before After Unadjusted diff (95% CI) Insurance status Uninsured (%) 10.5 4.3 −6.2 (−7.5 to −4.9) 18.7 15.6 −3.2 (−4.8 to −1.6)

−3.0 (−5.2 to −0.8)

p < .01

Medicaid (%) 27.6 35.1 7.5 (5.2 to 9.8) 24.2 20.9 −3.3 (−5.1 to −1.6)

11.0 (8.2 to 13.8)

p < .01

Stage at diagnosis Stage I–III (%) 90.1 89.2 −0.9 (−2.7 to 0.9) 91.0 90.0 −1.0 (−2.5 to 0.4)

0.0 (−2.3 to 2.3)

p = .99

Stage IV (%) 9.9 10.8 0.9 (−0.9 to 2.7) 9.0 10.0 1.0 (−0.4 to 2.5)

0.0 (−2.3 to 2.3)

p = .99

Time from diagnosis to treatment ≤30 days (%) 56.6 55.0 −1.5 (−4.0 to 1.0) 58.7 55.4 −3.3 (−5.3 to −1.2)

1.6 (−1.6 to 4.8)

p = .33

≤90 days (%) 95.9 95.7 −0.2 (−1.2 to 0.8) 96.0 95.7 −0.3 (−1.2 to 0.5)

0.1 (−1.2 to 1.4)

p = .94

Abbreviations: CI, confidence interval; DID, difference-in-difference. 3.3 Stage at diagnosis

Adjusted trends in FIGO stage at diagnosis are presented in Figure 1(C) (curable, stage I–III) and Figure 1(D) (metastatic, stage IV), stratified by expansion status. Unadjusted difference and adjusted DID for stage at diagnosis are shown in Table 2. Parallel trends were observed for stage at diagnosis in the pre-ACA period (2011–2013; p > .05). Patients diagnosed with curable stage disease nonsignificantly decreased in both expansion and nonexpansion states, and there was no statistically significant difference between the decreases seen in the two groups (adjusted DID = 0.0%, 95%CI = −2.3 to 2.3). The proportion of patients diagnosed with metastatic disease nonsignificantly increased from the pre- to postexpansion period in expansion and nonexpansion states, and the increases were not significantly different (adjusted DID = 0.0%, 95%CI = −2.3, 2.3).

3.4 Time to treatment

Adjusted trends in timely treatment for cervical cancer are presented in Figure 1(E) (treatment ≤30 days of diagnosis) and Figure 1(F) (treatment ≤90 days of diagnosis), stratified by expansion status. Table 2 summarizes the unadjusted difference and adjusted DID for time to treatment. Parallel trends were observed for treatment within 30 and 90 days in the pre-ACA period (2011–2013; p > .05). There was a slight decrease in the proportion of patients treated ≤30 days in both expansion (difference = −1.5%, 95%CI = −4.0, 1.0) and nonexpansion (difference = −3.3%, 95%CI = −5.3, −1.2) states; however, these decreases were not statistically different between the groups (adjusted DID = 1.6%, 95%CI = −1.6, 4.8). Treatment within 90 days remained similar from the pre- to postexpansion periods in both expansion and nonexpansion states, and these changes were not statistically different between the groups (adjusted DID = 0.1%, 95%CI = −1.2, 1.4%).

3.5 Overall survival

Kaplan-Meier survival curves (Figure 2; Appendix Figure 1) of expansion and nonexpansion states in the pre- and postexpansion periods show statistically significant differences between the four survival curves (p < .01) but no significant differences between the pre- and postexpansion period curves in both expansion (p = .72) and nonexpansion (p = .53) states. The adjusted Cox regression model (Table 3) demonstrated no significant change for overall survival in the expansion and nonexpansion states over time. The DID ratio comparing the hazard ratios of death in nonexpansion to expansion states (DID-HR = 0.95, 95%CI = 0.83, 1.09) indicates no difference in survival between expansion and nonexpansion states. A DID ratio > 1 indicates a greater improvement in expansion versus nonexpansion states, or less worsening in expansion versus nonexpansion states. Results of the four sensitivity analyses are described in Appendix Results 1.

Kaplan–Meier Survival Curves for Medicaid Expansion versus Nonexpansion States in the Pre- and Post-Expansion Period. Participants include patients aged 40–64 years old diagnosed with cervical cancer between January 1, 2012 to December 31, 2015 from the National Cancer Database. There exist statistically significant differences in overall survival between the four curves (log-rank test p = .02) but no significant differences between the pre- and postexpansion period curves in expansion (p = .72) and nonexpansion (p = .53) states

TABLE 3. Cox regression of survival in medicaid expansion versus nonexpansion states Adjusted Post- to pre-expansion HR (95% CI); p-value Difference-in-difference ratioa (95% CI); p-value Nonexpansion states 0.96 (0.88–1.05); p = .41 Reference Expansion states 1.01 (0.91–1.13); p = .82 0.95 (0.83–1.09); p = .48 Abbreviations: CI, confidence interval; HR, hazard ratio. 4 DISCUSSION

Using a large, nationally representative sample of patients with newly diagnosed cervical cancer, we found that the ACA was associated with significant expansion of Medicaid insurance and decrease in the rate of uninsured patients. However, no significant differences were observed in the stage of cervical cancer at diagnosis, timely treatment, or survival associated with Medicaid expansion.

Our study represents one of few reports investigating the effects of Medicaid expansion on cervical cancer stage at diagnosis, treatment, and mortality. Prior studies have reported mixed results with some demonstrating positive effects of Medicaid expansion on stage at diagnosis and timely treatment.7, 29, 30 Barnes, et al. looked at the impact of early Medicaid expansion in 2010–2011 on outcomes of cervical cancer patients undergoing radiation treatment and found that the six Medicaid expansion states had lower rates of late stage diagnosis (adjusted DID = −5.9%, p < .01) compared with nonexpansion states, though without any difference in survival.29 Kim, et al. analyzed stat