A. Problems

The first and possibly biggest hurdle to utilizing income data is that it is neither generated nor collected as part of regular medical practice. And even when income-related information is collected in medical settings, it is not clear how it is or should be recorded, classified, and accessed. Further, the measures currently used for assessing such data are wanting, lacking the granularity necessary for patient care and research.

Income-related data can be collected in two ways. First, the International Classification of Diseases (ICD)-10, which comprises standard medical diagnosis categories, also captures socio-behavioral information through its z-code category. However, the z-code standards relating to income are relatively vague. The Z59 code comprises all “problems related to housing and economic circumstances.”³ The subcodes include, inter alia, homelessness and inadequate housing (Z59.0 and Z59.1 respectively). There is only one subcode for income — Z59.6, “low income” (though extreme poverty, Z59.5, and insufficient welfare support, Z59.7, are closely related). An unspecified “low income” category is hard to use for precision research. The value of such research lies in its ability to produce algorithms that combine hundreds, even thousands of data points about an individual to make predictions about them. Such predictions become harder the less precise measures are.

Relative to the z-Code standards, the Systematized Nomenclature of Medicine — Clinical Terms (SNOMED-CT) is clearer. Its hierarchical approach, designed for EHR use, presents somewhat more granular categories. For example, it distinguishes between “variable income,” and “stable income,” among high, low, moderate, and no household income, between income that is insufficient to buy necessities, to “meet needs,” to buy anything more than necessities, and income that is sufficient to meet needs. It also sets apart “low” income and “no” income. And yet, the distinctions are vague. What is, for example, “low income”? How is one to distinguish between “needs” and “necessities”?

Most important, perhaps, is the impressionistic nature of these measurements. However granular our approach, concepts such as these are subjective, especially for clinical staff that lack training in them. More importantly, in the era of telemedicine, can a doctor in New York City make these judgments about someone living outside Rochester?

B. Approaches to Collection

Rather than creating new income data categories specific to the clinical context, it makes more sense to use existing poverty measurement approaches to collect income information. This would allow linking clinical research to existing literature and understandings of poverty. Such analysis would require obtaining actual income numbers, and combining those numbers with the demographic data that poverty studies have historically used to ensure proper analysis. Much like the Meaningful Use” program, which sets out “stages” of data collection and employment for Medicare providers, each stage harder than the next, poverty data can be collected in increasingly sophisticated steps.⁴

C. Privacy Objections

Another objection comes from privacy considerations. Just as commentators and litigants raise concerns about employers' gaining access to personal health information through wellness programs,²² some may raise concerns about payers gaining access to private salary information. Entities may then use the information to discriminate against individuals based on their income, shuttling them into lower value plans, or refusing to reimburse them, confident that lower income individuals will lack the resources to challenge denials. Finally, many might have similar concerns about providers seeing the information — at worse, low income patients might receive inferior treatment, at best, they might receive less courtesy when providers or their staff see that they are low income.^{Reference Konnoth23}

Using data segmentation techniques, neither the payer, nor even, if the patient so chooses, the provider, should be able to see the content of the transmitted data.²⁴ As discussed below, even if not seen, the information might be valuable for patient care through diagnostic support systems for providers. Further, data segmentation techniques can also be used to ensure that only some individuals within the practice and not others can see the data.

Further, questions about privacy should be separated from questions about discrimination. Providers and payers already discriminate based on a range of patient characteristics, such as race, sexuality, gender identity, diagnosis, and disability,^{Reference MacReady25} some of which may be learned through access to health information that might be considered more sensitive than income data. Discrimination is therefore a larger problem that should be addressed through education, attitudinal change, and enforcement.

Finally, studies show that a seeming desire to keep information private actually reflects embarrassment about discussing certain information in face-to-face interactions.^{Reference Newman26} Automatically and electronically collecting patient information on the sensitive topic of income might better comport with preferences. It is plausible that many patients recognize the importance of information such as income to their medical care, but do not want the awkwardness of discussing the information with the doctor themselves.

A. Individual Interventions

First, data can be collected to determine how best to intervene to help patients — at least those patients below a certain level of income. Financial stability can have a range of implications on outcomes. For example, studies show that low income individuals have problems adhering to medication due to costs, and a lack of transportation.^{Reference McCabe and Kurlander28} Low income patients are also disproportionately affected by surgical follow-up costs, including those involving travel, childcare, and lost wages.^{Reference Scott29} Studies have therefore shown less post discharge follow up and hospital utilization among this group.^{Reference Delia30} Other studies show that low income individuals are more subject to cardiac events; the causes merit further investigation.^{Reference Jones31}

Collecting and analyzing data would allow a more precise calibration between poverty and health. The risks of particular health events might plausibly be linked with different levels of income. More importantly, by analyzing — in the long run, hopefully, in real time — the effects of various kinds of interventions, providers and others might more precisely apply solutions in order to address poverty's harms.

First, providers themselves might adopt clinical solutions, that is, medical solutions to physiological problems caused by social conditions such as poverty.^{Reference Konnoth32} One way to do this would be to incorporate income-based approaches in Clinical Decision Support systems (CDS). CDS systems offer providers ranges of recommended options based on the patient's customized profile (which might include their income among other characteristics). CDS is used, for example, to warn clinicians of potential complications arising from drug interactions based on other medications that the patient is taking. To be sure, providers might ignore these recommendations, and other commenters have considered the liability issues (or lack thereof) this would raise.^{Reference Fox and Thompson33}

CDS, or, if available, the poverty information itself can offer a provider guidance in determining, for example, the kinds of follow up questions they should ask. Whether the patient has access to regular meals might determine what kind of medication is prescribed. Whether the patient has access to electricity, or running water will determine what kind of care a provider can reasonably expect a patient could administer for herself at home. Whether a patient has access to straightforward transportation or childcare options might also determine what kind of follow up the provider can reasonably expect—and again, might affect the care she provides today. In a fee for service regime, it might affect the kinds of procedures recommended. To be sure, the physician might get a sense of the patient's income from the kinds of recommendations the CDS makes, but that again is far more limited information than if the physician collected the data herself, and may not exceed the kinds of hints a physician already gathers by discovering that a patient receives Medicaid.

Biological or clinical solutions do not address the root of the problem — poverty. Rather, they seek to address the harm ex post. However, the information might also be used to offer structural solutions, what I refer to elsewhere as a social approach, to solving the problems of poverty itself.^{Reference Konnoth34} These social interventions involve interventions outside the clinic, such as ensuring access to resources that help ameliorate or eliminate poverty. These interventions can take place in contexts that are closely related to the clinic. For example, the information can be used with payers to get telemedicine waivers if transport is hard. It might be used to waive co-payments; this, in the short term, might lose a payer money, but in the long run, might ensure better care (and lower costs) by preventing catastrophic events.

Income data can allow providers to connect individuals with other services such as counselors and social service providers to ensure that they remain adherent to expensive or hard-to-administer medication. The “cascades of care” approach that the HIV context has long used and community health care models offer blueprints — as soon as someone is diagnosed with HIV, for example, they are connected to social service providers and others to ensure adherence.^{Reference Mugavero35} Indeed, continued adherence might require access to other kinds of medical care — in the HIV context, for example, depression counseling has been shown to be critical.^{Reference Pence36} In the poverty context, we might see referrals to counselors, insurance co-pay programs, housing programs, and others, to ensure that patients maintain access to medication.

The data can also enable individuals to be connected to services far beyond the clinic. Certain Accountable Care Organizations (ACOs) integrate social services into the care model. These ACOs link with social services organizations and engage in interventions such as providing transport for clinical visits, and helping to address environmental causes of disease.³⁷ Similarly, medical-legal partnerships that have caught the imagination of many law schools and community organizations, help address legal trouble that patients may face, ranging from landlord/tenant disputes to insurance denials.^{Reference Matthew38} Indeed, many of these interventions can be achieved automatically with the right technology. At Boston Medical Center, for example, a medical-legal partnership used electronic health records to automatically generate electricity shut off protection letters to electric companies.^{Reference Gottlieb39}

B. System-wide Interventions

Apart from determining how interventions should be carried out with respect to specific patients, collecting income-related data is helpful for identifying systemwide changes to improve the health of individuals that do not involve direct clinical or non-clinical interventions in the lives of individuals.

Clinical or clinical-adjacent transformations may include changing how we provide care for patients, by changing staffing. We might seek to hire more individuals with social work expertise rather than those with clinical training for example. Another important area of consideration involves payment models. Recently, there has been a renewed push for moving away from fee for service models to performance based models. Objections to such changes include the fact that some patient panels have greater social hardships, including poverty, than others, and therefore have worse outcomes, all else (including provider performance) being equal.^{Reference Roberts40} In such cases, it is unfair to penalize providers for taking care of more vulnerable patients. While payment adjustments exist today for such providers,^{Reference Markovitz41} collecting and analyzing data will help determine exactly how much to weight poverty in determining value based payments.

More importantly, perhaps, collecting and analyzing this data would give us a sense of how healthcare relates to other kinds of social services. It would give us a sense of how to allocate social resources. Commentators have questioned the value of an emphasis on medical research that benefits only the well off, but fails to increase health outcomes more widely. They question how much we allocate to clinical care and research relative to say, social services, income assistance, and other programs.^{Reference Baum42} Their claims arise from the fact that clinical and non-clinical methods are both plausible approaches to solving health problems. The clinical approach takes for granted the social determinant, such as poverty, and works on the back end, seeking to stem or mitigate the harmful effects of the social condition. To do so, it might suggest preventative clinical measures, such as regular vitamin tablet consumption.^{Reference Bazzano43} More often, the clinical approach works ex post, in the form of different clinical treatment approaches as outlined above. Non-clinical approaches work ex ante. They seek to nip the problem in the bud by ameliorating the social harm in the first place. Our research might therefore tell us what interventions would provide the most value for money — instead of cancer research, we may determine that outcomes would be improved by increasing TANF payments. Ultimately, such research can form the basis of determining how to allot social resources.