Piecing together the puzzle of disparities in AYAs

As we sit poised to witness the evolution of the delivery of health care in the United States, we are hard pressed not to consider the impact of the health care delivery model on patient outcome. In order to assess what impact, if any, we have made on patient outcome, it is our duty to evaluate each piece of this model both before it is changed and then after it has evolved. This is especially important when we consider vulnerable populations as in many instances, these are the patients that reform is aiming to protect, whose health promotion and disease treatment we are aiming to improve.^{1, 2} The National Cancer Institute has deemed adolescents and young adults (15–39 years of age) with cancer to be a vulnerable population as through time, these AYAs have not seen the same improvements in survival outcome as have children or adults 40 years and older³. Young adults represent the largest group of uninsured in the US both pre- and post-ACA implementation,⁴ however this age group has seen the steepest increase in coverage in the immediate post-ACA period.⁵

The multiple facets of the ACA are voluminous; those most applicable to the AYA population include regulations aimed at (1) improving the general health of the population by requiring new health plans to offer at least the minimum health benefits; (2) limiting gaps in coverage by outlawing pre-exisiting condition exclusions along with annual or lifetime limits and allowing young adults to remain on parents’ plans until 26 years; (3) making health insurance more affordable by creating a marketplace exchange; (4) minimizing out-of-pocket costs with the establishment of a temporary high-risk pool along with Medicaid eligibility expansion; (5) containing cost; and (6) increasing access for cancer patients including mandated coverage for clinical trials and concurrent hospice/therapeutic care for children.⁶

Via these mechanisms, a higher proportion of AYAs with cancer is likely covered; however, with the changing patterns of benefits and coverage of plans themselves, it is unclear whether other elements in the delivery of health care in AYA oncology are changing as well. As we design our before/after evaluations of the United States health care delivery model, studies such as Rosenberg’s study the current issue of Cancer are crucial.

Dr. Rosenberg and colleagues report a population-level analysis of the impact of insurance status on patient outcomes in AYAs with cancer. Aiming to distinguish associations between insurance status and both advanced-stage cancer and cancer-specific mortality, they interrogate SEER data to look at common AYA malignancies in 15–39 year-old patients diagnosed within the three years prior to the implementation of the early parts of the Affordable Care Act (ACA). The diagnoses included were all consistently staged using American Joint Committee on Cancer (AJCC) criteria: thyroid cancer, breast cancer, Hodgkin and non-Hodgkin lymphomas, female genitourinary cancers (including cervical cancer), male genitourinary cancers (including testicular germ cell tumors), melanoma, colon cancer bone/soft-tissue sarcomas (excluding Kaposi sarcoma), upper gastrointestinal cancers, lung cancer, hepatic tumors, renal tumors and non-pelvic germ cell tumors; central nervous system tumors and leukemias, both common malignancies in the AYA population, were unfortunately excluded as they are not staged with AJCC and thus could not be analyzed similarly. They identified nearly 58,000 eligible patients with available data - 54,765 patients (20–39 years) were included in their analysis of stage and 48,816 patients (25–39 years) were included in their analysis of survival. The authors draw a line to evaluate patients less than 25 years of age versus those older (20–24 years versus 25–39 years), a choice which is supported by their use of likelihood ratio testing; this would have been more beneficial as a policy-level analysis if the age cutoff instead had been 26 years, as the ACA expanded coverage in 2010 to allow young adults to remain on their parent policies until they turned 26 years.

SEER data is eternally fraught with limitations but remains the best available broad source of cancer-specific population-level data. Without granular treatment data, it is always difficult to draw unfaltering conclusions; however with stage, age, race/ethnicity and gender in the multivariable models, the authors have done as well as administrative data allows in adjusting for clinical variability.

Some details of this study limit the external applicability of the findings to all AYAs. In particular, the 15–19 year group was not included in the analyses. Although the authors did not address this decision specifically, many investigators looking at AYAs do not consider the adolescents when regarding health system factors such as payor, as federally mandated Title V coverage often takes care of the majority of this age group⁷. This decision ranges from one that is statistical in nature to one that is rooted in policy, whether governmental or institutional. Other studies address adolescents without including young adults at all⁸ or include only adolescents and the young AYAs (<25 years; yAYA)⁹ without including the older AYAs (>25 years; oAYA); they may focus on pediatric facilities or not have inclusive data for all ages. Despite the NCI’s definition of AYA, countries, institutions, investigators and cooperative groups have approached evaluations of this vulnerable population with age range poised as a moving target. Age cutoffs can fall anywhere from 19 to 39 years and everywhere in between. While this may not affect clinical care within a particular program, it is becoming problematic in the interpretation of the AYA literature. It is not accurate for us to say, as investigators, that one factor or another is key in the AYA Gap as a whole when we address only one age group in certain studies. Including the entire National Cancer Institute-designated age range of 15 to 39 years in a study would provide harmonization of the AYA investigations in order to better assess both clinical and policy implications of our findings. Providing negative studies for certain portions of the age group would be important to document in order to tease apart each element of this model, each piece of this puzzle.

The authors found that lacking private health insurance (having Medicaid or no insurance) was associated with a higher stage of cancer at diagnosis as compared to Stage I disease. The magnitude of this effect varied from modest for Stage II disease (increased likelihood of 20–70%) to more dramatic for stage III and IV disease (increased likelihood of 200 to 320% in oAYA). The effect in yAYA was more modest in the uninsured than in Medicaid or in oAYA; this reflects the nature of the expected payor variables in administrative data such as SEER. The ‘self-pay’ variable can reflect anything from a healthy young adult who risks going without coverage, to patients from abroad bringing cash in hand for services, while the Medicaid variable may indicate continuous or discontinuous Medicaid coverage at the time of diagnosis or retroactive coverage for a patient who had no coverage at the time of presentation. Regardless, the disease-specific analyses performed within the oAYA validate these effects with only a few outliers (melanoma has a very high likelihood of being diagnosed in advanced stages). Without sufficient numbers to perform the same analysis in the yAYA who would most benefit from the age-related ACA policy change¹⁰, we trust that the same would hold true there.

These blurred lines of the administrative data insurance variable often lead investigators such as Rosenberg and colleagues and others to consider Medicaid lumped with no insurance as a variable that indicates a lack of private insurance¹¹ or in contrast lumped with private insurance as a variable that indicates lack of any insurance.¹² While both cases may be made convincingly as to why Medicaid should be considered with private insurance (it indicates coverage) or un-insurance (it may be granted retroactively), Medicaid status acts differently in different situations. Ideal is likely the investigation of insurance status as three or more separate entities;^13–16 while this may lead to different findings in different studies it would allow evidence-driven decision making as to how to analyze these groups. This leads us to the conclusion that payor is not the definitive element in the model that is driving the outcome but rather it is one part of a puzzle. Payor is one element in the structure and process of health care delivery.

The role played by insurance coverage in presentation in advanced stages of cancer is supported by the findings in Pole’s study¹⁷ comparing Canadian AYAs (20–29 years) to US AYAs in the two decades prior to the ACA. That yAYAs saw improvement through time in Canada under a system of universal coverage, but not in the US under the pre-ACA framework of health insurance suggests that health system factors contribute a piece to the puzzle, that one piece resides along the spectrum between universal and fragmented coverage. Aizer¹² lends further support with SEER-based findings similar to Rosenberg’s, using two years’ of data (20–40 years); private insurance (as compared to Medicaid and no insurance lumped together) was associated with less metastatic disease, more definitive therapy and a lower mortality. Further support is provided by Robbins¹¹ who reports a higher risk of presenting with distant disease in AYAs (15–39 years) with Medicaid or no insurance as compared to private insurance in the 7 years prior to ACA implementation; this spanned not only diagnoses amenable to AJCC staging but also leukemias and central nervous system tumors. Smith¹⁸ presents a comparable association between not having private insurance and a higher likelihood of presenting with advanced Hodgkin Lymphoma in California AYAs (15–40 years). While addressing these pieces of the puzzle is integral to moving forward, it is also crucial to create a body of evidence that can support advocacy for policy-level changes on behalf of all AYAs. These arguments would be cleaner if a uniform age group was explored with consistent inclusion of adolescents.

Furthermore, the different landscape of treatment sites between studies points towards the need to measure the treatment site variability. While SEER, California^{12, 18} and Canadian registries¹⁷ include all potential treatment sites (as well as treatment protocol data in Canada), National Cancer Database (NCDB) studies¹¹ include only Commission on Cancer accredited hospitals. The NCDB thus may exclude young adults whose coverage limits them from accessing certain hospitals and commits them to community physician care.

The absolute difference in median survival in the present study is modest. However, once the analysis is adjusted for age and race and stratified by stage, the hazard ratios provide a solid effect size with the largest effects seen in oAYAs with lymphomas, low-grade male and female genitourinary cancers, low grade colon cancers and all low grade cancers in yAYAs. Insurance was associated with both all-cause mortality in other studies¹² and cancer-related death as here. Based on these findings and the other studies described above, one might consider associating these outcomes with access to preventive health care and recommended screening. However, it is important to note that while the majority of these studies focused on adult-type cancers with available screening, those cancers that cannot be screened (lymphomas) also had similar findings thus again the elements that contribute to this are multifactorial. This is supported by a higher risk of mortality in uninsured AYAs (15–39 years) with leukemia¹⁹ in the decade prior to ACA implementation as compared to those with any coverage; and clearly we cannot screen preventively for leukemia. However one conceives of it, the hypothesis that payor is associated with both advanced stage and outcome is proven in both cancers that can be screened and cancers that cannot.

Despite the inclusion of some sociocultural variables in some studies^{11, 12} the above are prime examples of situations in which a number of other biologic, individual, sociocultural and health system factors are unmeasured that impact disparities in survival.²⁰ The Institute of Medicine may posit that the unmeasured items that affect outcome would group into a domain representing the quality of the cancer care delivered²¹, although still we have no systematic means of measuring quality in this context. Nevertheless, even if collecting population-level data with the ability to represent a complete model is nearly impossible, we should consider each piece of the puzzle and how it fits together.

When Donabedian broke down the delivery of health care²² into structure, process and outcome, he dissected what clinicians perceive as a routine and fluid process, into minute elements. Each of these elements in turn has one or more downstream effects on the quality of health care delivered to an individual patient or to a patient group. The domain that he termed ‘structure’ denoted the organization of the delivery system, or the infrastructure, the scaffolding; this is where everything from a payor structure and financing to information systems and facilities would fit. The process domain denoted the actual provision and receipt of care; this would include the actual physical actions and interactions as well as how they are actually carried out. Factors unique to the individual and/or environment certainly have an impact on outcome that cannot be ignored; however these modify the relationship among these factors rather than define the model. Studies such as Rosenberg’s and similar studies addressing the payor element are integral to piecing together each part of the model. But by hanging our hat on one piece of the model we ignore the interconnectedness that Donabedian²² laid out for health care delivery and in turn have been laid out for AYA cancer.²⁰ Each element cannot be mutually exclusive, but will contribute to the overall outcome.

As a generation, we will be faced with not only designing studies to evaluate the impact of the ACA for our vulnerable populations, our cancer patients, our AYAs, but also in interpreting the findings of these studies. We will be faced with the temptation to envision the entirety of the puzzle with one piece or even a few. Going back to the scaffolding of health services research, we can employ a methodical approach to investigating each piece of the puzzle and the multiple interconnections; this will serve us well as we propose a solution moving forward. Some of these have been done and some have not. AYA clinical trial enrollment has been associated with both insurance status²³ and outcome.²⁴ Treatment site is being examined.^{25, 26} Socioeconomic status is being explored.^{18, 19} In the structure domain, what other facets impact the delivery of health care that could be affecting access to both preventive screening and disease treatment and what facets of the treatment sites themselves could be contributing to either the access or quality delivered? In the process domain, what facets could be contributing to the provision of care either within the healthcare team or facility? While a quick answer is helpful to advocate in the public arena, this means that we have to take a deep breath. Before we say that providing private insurance for young adults via the ACA is solving the challenges in AYA outcomes, the investigations will need to be performed, using as uniform a model as possible, to dig the other pieces of the puzzle out from the bottom of the box.