Introduction

For some time now, there has been a broad agreement that social determinants of health (SDoH) highly influence how individuals interact (or don’t) with the healthcare system which, in turn, impacts health outcomes. Researchers believe as much as 80% of the contributors to healthy outcomes are SDoH-related, such as nonclinical factors like behavior, socioeconomics, financial and health literacy, race, and ethnicity.¹

Researchers, clinicians, care management teams, and life sciences firms have been exploring SDoH for some time now; but results have been decidedly mixed.2 That begs the questions, under what circumstances does SDoH actually help improve therapeutic effectiveness or increase adherence? Or help address the diversity of clinical trial recruiting? Or help predict and improve interventions? It’s clear that both commercial and research uses stand to gain, so what’s the secret to generating results with SDoH?

In this paper, we make a case that both the fidelity of SDoH data and a modernized systemic approach to compliance are the two critical factors. More specifically, SDoH data works best when it’s individual level, standardized, and independently verified—that’s why SDoH based on census tracts and EMR extracts often fail to meet expectations.3 Furthermore, compliance must be systemic and “always on,” as inconsistently applied manual governance creates uncertainty with respect to privacy and ethical use. Let’s explore these concepts more fully, then we’ll turn our attention to steps for improving the effectiveness of your SDoH initiatives.

“All the children are above average” in Lake Wobegon, the fictional town featured in Garrison Keillor’s long-running radio show, “A Prairie Home Companion.” Keillor was cleverly poking fun not only at the bias we all feel when it comes to our communities and children, but also perhaps at our reliance on information that really doesn’t tell us much about the reality of a population or a place. Of course, not all children can be above average; averages tell us little about any individual.

Not All Data Is Created Equal

Just as all children in Lake Wobegon can’t be above average, no individual’s SDoH is average. To improve outcomes for a person, it’s vital to know that person’s actual economic stability or health literacy. Using averages for these purposes is, at best, unproductive and, at worst, misleading. Just as precision medicine is meant to be personalized to the individual, so too must SDoH data be individualized.

One traditional source for SDoH has been census tract data. This data is readily available and has many beneficial planning uses inside and outside of healthcare. However, there are over 73,000 census tracts in the U.S., meaning they each depict averages for about 4,500 people. If averages were all that mattered, a person with one foot in boiling water and the other in a bucket of ice would be perfectly comfortable. Using averages can be misleading! Similarly, average economic, housing, or food vulnerability tells us nothing about the circumstances of an actual individual. Census tract fidelity and granularity are simply inadequate for the uses that matter most—understanding how real-world evidence (RWE) and health economics and outcomes research (HEOR) can be tuned to improve individual outcomes. Individualizing SDoH will not only improve the patient outcome but potentially influence future standards of care. It can influence how adherence strategies can be adjusted based on the lived life experience of the patient or how clinical trials might identify and reach a truly representative population for target protocols.

Patient-reported data, sometimes captured in EMRs, has been the other usual source of SDoH information. There are two big challenges that limit the usefulness of this data. First, the incentives for the provider to capture this information and the willingness of the patient to answer truthfully limit utility. That is why EMR SDoH fill rates are lacking, and the attributes captured are too few to address modern analytic needs. Second, standard SDoH coding is not yet available; and two different providers might refer to the same SDoH factor differently. Moreover, this information is often captured in clinical notes, which requires scarce AI resources to obtain appropriate permissions and to scan and structure the data in a consistent way across providers. Most providers simply don’t have those resources available.

What’s needed is individual level, standardized, and independently verified data about the lived life experience of the patient outside the clinical setting.

With Great Data Comes Great Responsibility (hat tip Spiderman)

The other big constraining factor for researchers using “traditional” SDoH data sources is the lack of systemic data use monitoring, particularly with respect to privacy and ethical use.

HIPAA compliance, particularly for uses of deidentified data, becomes more challenging when SDoH attributes about patients are added.

Mathematically, the more attributes known about a de-identified patient, the more likely that patient can be re-identified, creating exposure for civil and criminal penalties—plus the inevitable brand reputation impact. Unfortunately, the typical convention for protecting de-identified data is the occasional engagement of a HIPAA expert to render an opinion of compliance. Not only is the certification process slow, expensive, and inflexible, it’s unusual for that certification to reflect the actual current reality of widespread “anonymized” information available to their organization. The lurking risk of re-identification is often underappreciated.

Another key dynamic is ethical use. The risk of adverse decisions for disadvantaged populations also increases when additional SDoH is available. Unfortunately, some engrained healthcare practices reinforce disparate outcomes which, stated another way, are negative drivers to quality measures. There is a critical need to demonstrate that the incorporation of SDoH can improve patient outcomes, not disadvantage them. Unfortunately, many organizations are long on datause principles and short of mechanisms to ensure they are actually followed.

What’s needed is systemic, “always on” monitoring to detect potential privacy or ethical use missteps before they actually occur.

A Better Way to Understand Patients: the Multidimensional Patient Journey

Healthcare progress has always been based on evidence, usually clinical readings, diagnoses, procedures delivered, and medications prescribed. By adding independently verified data about the patient’s nonclinical experience, we’ll better understand previously hidden drivers to patient outcomes, particularly those occurring outside clinical settings. We describe the linking of patient clinical and nonclinical data over time as creating a multidimensional patient journey. Adding social, physical, and behavioral determinants of health and race/ethnicity to a patient’s clinical journey provides critical added dimensions that affect adherence, access, outcomes, and economics. In time, this could form the basis for creating new evidence-based standards of care. Researchers, providers, care management teams, life sciences firms, and of course the patient all stand to benefit.

A Better Way to Understand Patients: the Multidimensional Patient Journey (continued)

At Change Healthcare, we’ve mapped the major social, physical, and behavioral determinants of health and race/ethnicity to most of the commercially insured U.S. population’s claims data. This patient longitudinal data is updated daily with new care activity. From this data, we know that a major factor affecting the healthcare experience is economic stability (see chart that follows). You can very clearly see that higher economic instability is directly correlated with higher healthcare utilization (and often acuity too). Measuring utilization trends just at the therapeutic level, without understanding the economic stability of patients/members, simply provides an “average” that masks the underlying diversity of healthcare experiences. Of course, meeting the patient where they are, clinically and nonclinically, is the key to achieving quality outcomes for all.

A major advantage of this dataset is it links medical and pharmacy claims with major social, physical, and behavioral determinants of health at the de-identified patient level, enabling enormous flexibility to examine unique patient journey populations in new ways. Let’s illustrate its advantages in comparison to census tract data.

We’d all say that Garrett Park, Maryland, Durham, North Carolina, and Marin County, California, are affluent communities with average incomes of $186,000, $127,000, and $250,000 respectively. Yet underneath the average is a wide range in individual patient income. In fact, 22.6%, 23.2%, and 12.2% of individuals in those communities have incomes under the poverty level.⁴ Basing care protocols on an area’s average economic means is a recipe to dramatically misrepresent the real circumstances of actual individuals in that geography.

The preceding paragraph reflects income; but our research has found averages to deceive across a wide variety of nonclinical factors, including health literacy, race/ethnicity, vulnerabilities in food, economic stability, housing, transportation, and much more. Simply said, if you want to improve a person’s health outcome, you need to understand them individually and not draw conclusions just by where they live.

Multidimensional Journeys Require a Modernized Approach to Monitoring Data Use

Healthcare was one of the first industries to implement a national privacy law, HIPAA in 1997. This regulation has served healthcare well, helping to fuel research, safety, and performance benchmarking among other beneficial uses. That said, the traditional method of demonstrating compliance with the HIPAA Privacy Rule by occasionally “certifying” is no longer adequate. Not only is the process slow, expensive, and inflexible, just occasionally certifying is no longer sufficient given the wide availability of “anonymized” information about people on the internet, wearables, and on mobile phones. The world has clearly changed since the ‘90s. Accordingly, monitoring HIPAA Privacy Rule adherence needs to change from manual to automatic, from infrequent to “always on,” and from general assumptions to the actual protections needed for a given task. Change Healthcare’s Data Science as a Service (DSaaS) was designed, and is deployed, for this exact purpose.

DSaaS is a secure-hosted cloud designed and provisioned for individual clients focused on “always on” compliance monitoring for privacy, ethical use, anticompetitive activity, and other key requirements you determine. Once in place, DSaaS persistently and consistently monitors all analytic activity, including all queries, models, and output. Nothing is displayed or leaves DSaaS unless it’s first checked against compliance rules. This happens automatically and persistently.

DSaaS also provides instant scale. DSaaS data is prepopulated with permissioned and de-identified medical/pharmacy claims data that has already been integrated with SDoH for most of the commercially insured population. Furthermore, we can integrate any data you provide to us. That means cohorts for clinical trials, RWE, HEOR, rising risk models, etc. can be assembled in minutes, not weeks or months, with adherence to your data governance built into the process. You start faster, analyze more robustly, iterate more frequently, all with confidence instilled via an “always on” compliance monitoring system.

Monitoring HIPAA Privacy Rule adherence needs to change from manual to automatic, from infrequent to “always on,” and from general assumptions to the actual protections needed for a given task. Change Healthcare’s Data Science as a Service (DSaaS) was designed, and is deployed, for this exact purpose.

Contenders and Pretenders: 7 Key Differentiators

Health technology and data can seem like a Middle Eastern bazaar, a cacophony of similar sounding concepts. Many offer SDoH data; others offer “HIPAA-certified environments,” as if HIPAA were a single thing. How does one sort through it all? Here’s a checklist that covers the key attributes of an effective SDoH platform:

1. Patients, not averages: SDoH should be about individuals, not averages, for a geography. Improving quality outcomes begins by meeting people where they are in their health journey.
2. Greater than sum of its parts: SDoH data is at its most powerful when linked to clinical data and vice versa. Connecting what happens in a clinical setting to life outside it is critical to understanding the whole patient.
3. Standardized and portable: SDoH should be standardized so that SDoH classifications are immediately understood by every healthcare constituent. No Rosetta Stone required!
4. Trust but verify: SDoH should be independently validated across multiple sources.
5. Permissioned and secured in the cloud: Deidentified claims data should be appropriately permissioned and provisioned in a secure hosted cloud. This accelerates cycle time and amplifies depth of analysis.
6. “Always on” compliance: Data governance monitoring should be systemic and “always on,” not infrequently or manually checked. Compliance monitoring needs to be modernized to a world awash with anonymized information about people. Furthermore, HIPAA Privacy Rule monitoring should be “always on”, not episodic. Don’t be the organization that underappreciates your risk of re-identification.
7. Good (data) governance: Systems should monitor data representativeness (to avoid skew) and outputs (to avoid disparate treatments/outcomes), as well as monitor potential anticompetitive activity, using the FTC Safe Harbor as a guide.

Conclusion

Healthcare is at the precipice of a new era, where clinical insights are enhanced by a deeper understanding of how social, physical, and behavioral determinants of health impact individual lives. We describe this as a multidimensional patient journey. For the first time, decision-making can confidently access and effectively use nonclinical factors that influence up to 80% of health outcomes. Guard rails monitor use for accordance with privacy regulations, ethical use considerations, and avoidance of anticompetitive activity. Frictionless access to appropriately permissioned SDoH and clinical data (yet always adhering to modern data governance principles) creates the opportunity for dramatic increases in cycle time and insight. These new superpowers are key to actually improving clinical trial diversity and uncovering previously hidden opportunities in RWE, HEOR, market access, and adherence. They unlock new opportunities for providers to improve quality outcomes for all. They help care management teams better identify and act on rising risk. And our seven-step checklist shows the way to effectively compare alternatives.

While other industries have been transformed using data and AI, healthcare has proven to be more difficult because of the complexity and diversity of how we live our lives and because of reasonable concerns over privacy and the ethical use of healthcare data. But we now see how DSaaS – combining clinical, true SDoH and AI – can change our view of both individual patients and population segments leading to research breakthroughs and, ultimately, better health outcomes. A new era beckons.

_{^{1. Hood, C. M., K. P. Gennuso, G. R. Swain, and B. B. Catlin. 2016. County Health Rankings: Relationships Between Determinant Factors and Health Outcomes. American Journal of Preventive Medicine 50(2):129-135. https://doi.org/10.1016/j.amepre.2015.08.024}}

_{^{2. Halamka, J. 2021. Can Social Determinants of Health Predict Your Patient’s Future? Dispatch from the Digital Health Frontier. http://geekdoctor.blogspot.com/2021/08/can-social-determinants-of-health.html}}

_{^{3. N Krieger, J Chen, P Waterman, M Soobader, S Subramanian, and R Carson. 2003. Choosing area based socioeconomic measures to monitor social inequalities in low birth weight and childhood lead poisoning: The Public Health Disparities Geocoding Project (US). Journal of Epidemiol Community Health 57(3): 186–199. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1732402/}}

_{^{4. Data from Change Healthcare DSaaS platform – closed claims for patients with claims activity July 2020 – June 2021.}}