Association Between the Maryland Medicaid Behavioral Health Home Program and Cancer Screening in People With Serious Mental Illness

We used Maryland Medicaid administrative claims data from October 1, 2012, to September 30, 2016, which included the year prior to implementation of BHHs (baseline) and the three years after BHH implementation. The unit of analysis was person-year. This study was waived by the institutional review board at Johns Hopkins University.

In Maryland, 53 (37%) of 145 psychiatric rehabilitation programs (PRPs) implemented BHHs during the study period (9). To qualify for PRP services, individuals must have significant functional impairment as a result of mental illness. Our sample consisted of 12,176 Medicaid enrollees with greater than five uniquely dated PRP services during the study period. We censored individuals when they were no longer enrolled in Medicaid.

To be consistent with U.S. Preventive Services Task Force (USPSTF) guidelines, we measured cervical cancer screening among women ages 21 to 64 (N=6,876), breast cancer screening among women ages 50 to 64 (N=1,914), and colorectal cancer screening among men and women ages 50 to 64 (N=3,670) (10). We excluded individuals with a history of cervical cancer (N=52), endometrial cancer (N=13), breast cancer (N=256), colorectal cancer (N=39), or symptoms within 3 months of receipt of a cancer screening test (N=201) that may indicate testing for nonscreening purposes. We performed sensitivity analyses that included all of these individuals.

Outcomes were receipt of a Pap smear for cervical cancer screening; mammography for breast cancer screening; and colonoscopy, sigmoidoscopy, or fecal occult blood test (FOBT) for colorectal cancer screening (10). (Procedure and diagnostic codes are provided in supplement A in the online supplement to this report.) Claims from acute settings (emergency department, inpatient hospital) were excluded because they were unlikely to represent routine screening events.

BHH enrollment occurred on a rolling basis throughout the study period. All Medicaid beneficiaries receiving services at a PRP with a BHH program were eligible and were required to provide consent to participate in the BHH. Staff of the BHHs reported attempting to enroll all eligible clients but also reported staffing shortages that may have impeded enrollment (9). We assumed that once an individual was enrolled in a BHH, he or she remained enrolled for the remainder of the study, consistent with intent-to-treat analysis principles. Approximately 27% (N=3,298) of the population were enrolled in a BHH at some point during the study, 25% (N=3,040) received psychosocial rehabilitation services at a program without a BHH and therefore never enrolled in a BHH, and 48% (N=5,838) never enrolled in a BHH even though the program where they received psychosocial rehabilitation services had a BHH.

We accounted for rolling enrollment into BHHs by using time-invariant and time-varying confounders. Time-invariant confounders included baseline age, sex, race, psychiatric diagnosis, size of PRP, region of residence, and enrollment in a Medicaid managed care organization. Observed time-varying confounders, measured yearly, included eligibility for Medicaid via disability, substance use disorder diagnosis, Charlson comorbidity index, number of PRP services received, somatic and psychiatric hospitalizations, number of primary care visits, and prior receipt of cancer screening.

Traditional regression adjustment for time-varying confounders can introduce bias into results; therefore, we used a marginal structural modeling approach (11). First, we constructed the inverse-probability-of-treatment weighting. We estimated the inverse probability of BHH enrollment for each person-year, adjusting for time-invariant and time-varying confounders. The inverse-probability-of-treatment weighting for each individual-year was calculated as the product of the inverse probabilities in the years leading up to that year, and therefore weighting controlled for an individual’s history of treatment and confounders. Second, we calculated censoring weights, which account for confounding factors that may influence censoring from the sample due to Medicaid disenrollment. We calculated the inverse probability of censoring for each person-year, adjusting for confounders. The censoring weight for a given person-year was the product of the inverse probabilities up to that year. The final weight for any given person-year observation was the product of the treatment weight and censoring weight. (Full details of the weighting process are available in supplement B of the online supplement.)

Third, we fit weighted logistic regression models to estimate the average effect of BHH participation on annual cancer screening. To improve interpretability of results, we calculated predicted annual screening rates, interpreted as the expected screening rate if all participants were enrolled in a BHH (versus not enrolled). As a sensitivity analysis, we tested for interaction between BHH enrollment and psychiatric disorder, substance use disorder, PRP or primary care utilization at baseline, and sex.

Unweighted baseline demographic characteristics differed by BHH enrollment (see online supplement C). Compared with non-BHH participants (N=8,878, 73%), BHH participants (N=3,298, 27%) were more likely to be white (N=1,560, 47%, versus N=3,134, 36%; p<0.001); to have schizophrenia (N=2,098, 64%, versus N=3,232, 36%; p<0.001); and to qualify for Medicaid by disability (N=2,843, 86%, versus N=5,167, 58%; p<0.001); used more PRP services (9.8±0.07 versus 5.1±0.04 services, p<0.001); and were less likely to have a substance use disorder (N=716, 22%, versus N=2,450, 28%; p<0.001). BHH participants had fewer baseline primary care visits than did non-BHH participants (5.1±0.07 versus 5.4±0.05 visits, p<0.001), but the percentage of individuals without any primary care visits (27%) did not differ by BHH enrollment (enrolled, N=887; not enrolled, N=2,353).

There were similar demographic differences between the BHH participants and the subsets of individuals who received psychosocial services at a program with a BHH but who were not enrolled in the BHH and individuals who received services at a program that did not implement a BHH (see online supplement C). Programs with a BHH were larger than those without a BHH (see online supplement C). For example, 28% (N=15) of programs with a BHH served more than 500 clients compared with 5% (N=5) of programs without a BHH (p<0.001). Weighting improved the balance between baseline characteristics of BHH and non-BHH participants to a level that indicates that these factors caused minimal residual confounding (see online supplement C).

During the postintervention period, 53% (N=3,576) of eligible women (N=6,811) received a Pap smear. BHH enrollment was associated with an increase in the likelihood of receiving cervical cancer screening compared with no BHH enrollment (odds ratio [OR]=1.20, 95% confidence interval [CI]=1.07–1.35; p=0.002) (Table 1). The predicted annual screening rate was 31% for BHH enrollment and 27% for nonenrollment.

During the postintervention period, 44% (N=737) of eligible women (N=1,658) had a mammogram. BHH enrollment was associated with an increase in the likelihood of receiving breast cancer screening compared with no BHH enrollment (OR=1.30, 95% CI=1.06–1.59; p=0.01). The predicted annual screening rate was 28% for BHH enrollment and 23% for nonenrollment.

During the postintervention period, 30% (N=1,021) of eligible individuals (N=3,430) received colorectal screening, of whom 27% (N=911) had a colonoscopy, 0.4% (N=13) had a sigmoidoscopy, and 7% (N=252) had a FOBT. BHH enrollment was not associated with the likelihood of colorectal cancer screening. Similar results were observed when screening modality was stratified by colonoscopy, sigmoidoscopy, or FOBT.

Our findings were unchanged after we included participants who may have received testing for reasons other than routine screening (see online supplement D). Over longer intervals, the differences between screening rates among BHH-enrolled and non–BHH-enrolled individuals for cervical and colorectal cancer (3-year interval) and for breast cancer (2-year interval) were similar to the results after one person-year, both in magnitude and statistical significance.

When we examined whether subsets of the population benefited more than others from BHH enrollment, we found that the positive association between BHH enrollment and cervical cancer screening was stronger among low utilizers of PRP services relative to high utilizers (OR=1.08; 95% CI=1.04–1.12). BHH enrollment was also associated with higher rates of breast cancer screening among individuals without schizophrenia relative to those with schizophrenia (OR=1.10, 95% CI=1.04–4.58) and with higher rates of colorectal cancer screening among individuals with a substance use disorder compared with individuals without such a disorder (OR=1.05, 95% CI=1.01–1.08). We found no other differences between BHH-enrolled and nonenrolled individuals by patient or PRP characteristic (see online supplement D).

The predicted annual screening rates for both cervical cancer and breast cancer were lower for both BHH participants (31% and 28%, respectively) and nonparticipants (27% and 23%, respectively) than estimated annual rates in the general population (42% and 43%, respectively) (5, 6). Our study screening rates were similar to those reported in a California Medicaid population with serious mental illness (5, 6).

The higher rates of breast and cervical cancer screening associated with BHHs are consistent with clinical trial findings (8). Although the relative increase in screening was modest, it suggests that BHHs may have a clinically meaningful impact. The core care coordination, population health management, and health home activities implemented by BHH programs in Maryland and across the country (12, 13) are designed to improve delivery of preventive services such as cancer screening. In our study sample, the majority of individuals were connected to primary care: the proportion of BHH participants and nonparticipants, respectively, with no primary care visit was 26.9% and 26.5% during the baseline period and 11.1% and 11.0% at the end of the follow-up. Thus increases in screening were more likely to be due to BHH care coordination activities than to new linkages to primary care providers. Maryland implemented its BHH program within PRPs, which had engaged in coordination of mental health and social services prior to the BHH program (9). Existing capacity for care coordination may have facilitated the ability of these programs to improve cancer screening rates. Prior research has identified differences in the degree of integration of behavioral and general medical care at Maryland BHH sites (14); future research should consider the effects of these differences.

The low overall rates of cancer screening in our population suggest that some screening barriers remain unaddressed. Subsets of BHH-enrolled participants were more likely to get specific types of cancer screening, but no overarching trend was observed. Potential reasons include impairment in memory or executive function—which may influence understanding of risks and benefits of screening (15)—or the high prevalence of poverty, disability, and housing instability (9, 15)—which has been associated with lower cancer screening rates (16). People with serious mental illness may prioritize preventive services differently or experience greater levels of stigma (15), which may influence care. Colorectal cancer screening may have additional barriers, such as consumers’ perception of its invasive nature and requirements for bowel preparation (16). Furthermore, potential bias, low prioritization of preventive screening, and discomfort in discussing cancer screening with individuals with mental illness may also contribute to barriers at the provider level (15, 16).

These results should be considered in light of several limitations. We analyzed data at the person-year level to account for potential time-dependent confounding. Our primary analyses, therefore, examined annual screening rates rather than the multiyear intervals recommended by clinical guidelines (10). Second, our marginal structural modeling approach assumes there were no unobserved confounders. No data existed on which clients were approached for enrollment, so we could not delineate between eligible participants who were not invited to participate in the BHH and those who were invited to participate but declined to enroll. Third, Maryland’s Medicaid BHH program was implemented in the subset of people who qualify for PRP services, which may limit our study’s generalizability. Finally, while we accounted for PRP size and utilization, we were unable to determine programmatic differences or cancer screening priorities at the level of the BHH. The state of Maryland has identified improving cancer screening as an explicit BHH goal (12), but the degree to which such screening was prioritized across different BHHs is unknown.

Our study provides exploratory evidence that enrollment in BHHs is associated with higher rates of cervical and breast cancer screening for individuals with serious mental illness in real-world settings. BHH enrollment had no effect on colorectal cancer screening. Observed cancer screening rates improved after BHH implementation, but they remained suboptimal, which suggests that other interventions are needed to improve cancer screening rates for people with serious mental illness.

The authors thank Michael Rosenblum for helpful comments in reviewing the manuscript.