Recommendations for improving national clinical datasets for health equity research

Abstract

Health and healthcare disparities continue despite clinical, research, and policy efforts. Large clinical datasets may not contain data relevant to healthcare disparities and leveraging these for research may be crucial to improve health equity. The Health Disparities Collaborative Research Group was commissioned by the Patient-Centered Outcomes Research Institute to examine the data science needs for quality and complete data and provide recommendations for improving data science around health disparities. The group convened content experts, researchers, clinicians, and patients to produce these recommendations and suggestions for implementation. Our desire was to produce recommendations to improve the usability of healthcare datasets for health equity research. The recommendations are summarized in 3 primary domains: patient voice, accurate variables, and data linkage. The implementation of these recommendations in national datasets has the potential to accelerate health disparities research and promote efforts to reduce health inequities.

INTRODUCTION

Rationale

For 3 decades, numerous national scientific organizations and federal agencies have recommended improvements in data collection with the goal of a more robust understanding of health and healthcare disparities, and their eventual elimination.^1–7 Landmark publications such as the 1985 Malone-Heckler Report,¹ a 2008 call to action by the World Health Organization’s Commission on Social Determinants of Health,⁸ multiple reports by the Institute of Medicine (National Academy of Medicine) on racial and ethnic disparities in health care³ and on LGBTQ health,² and legislation such as the Affordable Care Act⁴ have consistently highlighted a broad need for the role of research on disparities in improving health outcomes. Yet technical challenges to systematic and comprehensive data collection remain the most prominent hurdle to research on social determinants of health (SDH) that can effectively inform the mitigation of disparities.⁹

One recommended approach to improve the quality of research focused on eliminating health and healthcare disparities is by effectively leveraging electronic health record (EHR) and related datasets. EHRs are recommended as a potential source of data to improve the quality of research on eliminating health and healthcare disparities. Published recommendations document both the utility and limitations of EHR data,^10,11 while highlighting specific use cases for EHR use research such as patient clinical trial recruitment and tracking.^12–18 Although these same reports offer recommendations to standardize, modify, add, and revise available values for EHR data fields, the Office of the National Coordinator for Health Information Technology (ONC) guidance for certification focuses on the safe use of EHR in clinical operations and not research.^12,19,20

Increasing attention in recent years has been given to the inclusion of SDH data into the EHR for the purposes of understanding inequities. The National Academy of Medicine set forth 11 recommended domains for SDH in EHRs.⁷ In response, screening tools and protocols have been developed by the Centers for Medicare and Medicaid Services²¹ and the National Association of Community Health Centers Protocol for Responding to and Assessing Patients’ Assets, Risks, and Experiences,²² respectively. The National Academy of Medicine has also developed means of capturing social and behavioral information in the EHR.²³ All of these tools offer opportunities and potential for health equity research. However, a comprehensive set of recommendations for improving EHRs and other commonly used data sources to study health equity is lacking in the health services literature.

The purpose of this article is to report on the development of recommendations to improve the usability of healthcare datasets for health equity research using a consensus-based approach among a Patient-Centered Outcomes Research Network (PCORNet) working group of investigators, program leaders, clinicians, and patients with expertise in health equity and health equity research.

PCORnet and the Common Data Model

In order to produce these recommendations, we selected and convened experts from both the PCORnet Health Disparities Collaborative Research Group and the broader health equity research community.

PCORnet, established by PCORI, is a network of clinical research networks (CRNs) developed to promote research collaborations. PCORnet is well positioned to be a leader in providing datasets that are relevant and ready for health equity research.

PCORnet launched its Common Data Model (CDM) to better standardize data collection for research. The PCORnet CDM provides a framework for data structure and definition that enhances the research-readiness by allowing rapid data queries across distributed networks like the Food and Drug Administration Sentinel Initiative CDM⁵ and others.

With geographic, sociodemographic, provider, and payor diversity, PCORnet has the potential to aggregate demographic and healthcare data on millions of individuals who suffer health and healthcare disparities using the CDM. Since its inception, data quality and completeness have been central and implemented through a 5-step process with diagnostic queries, characterization queries, empirical data characterization reports and mitigation plans, and making changes. The first 2 steps involve running specific queries to test for data accuracy, the reports provide sight across the data, which highlight potential inaccuracies and mitigation plans and changes work to address any issues identified.²⁴ In addition to work at the PCORnet level, each CRN engages in data curation and improvement activities.^25–27 However, there are improvements to be made that would prepare the PCORnet CDM to optimally measure inequities in health and healthcare outcomes.

In support of catalyzing new research and leveraging the value of the CDM, PCORnet developed collaborative research groups (CRGs). These groups brought together experts from across research networks to accelerate research to improve health care in select content areas. The health disparities CRG, comprising 3 research interest groups (RIGs), was created in the program’s first year as part of the Accelerating Data Value Across a National Community CRN and continues its work to initiate and encourage health equity research using the CDM. The 3 RIGs are health disparities in racial and ethnic minority populations, health disparities in sexual orientation and gender identity minority populations, and SDH. After 3 years, the Health Disparities Collaborative Research Group was recognized for special expertise in data science and was charged with developing data science recommendations for the PCORnet Data Committee to improve completeness, comprehensiveness and research readiness for health disparities projects. This article describes the process and outcomes of providing PCORI with data science recommendations to improve health disparities research.

MATERIALS AND METHODS

To develop recommendations, we assessed and categorized practices for collecting and reporting relevant data in the EHR and solicited engagement and input from new and existing stakeholders. Stakeholders were asked about research and data needs for their own health disparities work (questions in Supplementary Appendix) and for the field of health disparities research. While our final recommendations encompassed broad areas of study and may have clinical relevance or relevance in other areas of research, they were focused on research in social determinants of health, racial and ethnic minority health, and sexual and gender minority health.

The Collaborative Research Group Project Team executed a thematic approach, analyzing, and formulating qualitative data into recommendations in 3 phases:

• Phase I: The Project Director, a qualitative researcher, trained the Project Coordinator in research interviewing, and together they completed interviews and focus groups with experts, researchers, clinicians and patients. Members of the RIGs and the OCHIN Research Patient Engagement Panel (a panel of patients trained to serve as advisors and patient investigators for research) participated in focus groups or interviews based on their preference and availability. Each member of the RIGs is an expert in health disparities research and represents a different aspect of this body of work. Based on their skills and knowledge, there were serving on the RIGs and were subsequently recruited to participate in a focus group. Experts outside of the RIGs and in each focus area were recruited specifically for this information gathering phase of the project and interviewed individually. Participants included Principal Investigators for the National Institutes of Health and other nationally funded health equity, translational, and clinical research studies; directors of research centers and patient advocacy networks; health equity award winners; and healthcare providers known nationwide for specialization with specific marginalized populations. Patients from the Patient Engagement Panel all have experience consulting on health equity research and their own experiences in both clinical care and different aspects of research.

• Phase II: The Principal Investigator and Project Director completed a thematic analysis to identify overarching themes and focus areas through both an inductive and deductive process ensuring both accuracy to the data and relevance to the questions to be addressed.

• Phase III: An initial set of draft recommendations were compiled by the Principal Investigator and Project Director. This draft was reviewed and revised in an iterative process by all individuals who provided input and a subset of contributors who ultimately authored the final recommendations.

RESULTS

The analysis of the data collected revealed the following recommendations from our experts: (1) inclusion of patient voice, (2) accurate and relevant variables, and (3) important additional data. Each theme emphasized the need for data quality and relevance to vulnerable populations. The experts specifically decided not to state a prioritized order for the recommendations, citing the global importance of each issue and a lack of clear rationale for prioritizing one over another. In addition to the recommendations, select implementation strategies and opportunities are identified. Table 1 includes the categories, recommendations and suggestions for implementation.

Table 1.

Recommendations and implementation suggestions by area

Area	Rec #	Recommendation	Implementation
A. Inclusion of patient voice	A1	Prioritize health information collected directly from patients using patient-reported outcomes, measures of patient experience, and emerging patient-reported measures.28,29	Include measures from PROMIS—they are public domain, validated, and available in multiple languages.30 Leverage existing patient experience measures such as the Hospital Consumer Assessment of Healthcare Providers and Systems and PROs such as the Adverse Childhood Experiences assessment. Incentivize development of EHR fields/instruments that allow patients to communicate priorities/most significant healthcare needs/barriers.
	A2	Develop text fields and repositories of written and oral patient narratives describing experiences navigating and receiving health care.28,29	Apply the DIPEx methodology for gathering and maintaining a narrative repository.6 Include text fields to encompass the multidimensional experience of care such as transportation, childcare, time away from work, school or other responsibilities, past experiences, social support, and health literacy.
B. Accurate and relevant variable names, categories, and values	B1	Improve the accuracy of basic sociodemographic data by standardizing basic demographic variable names and categories, enhance demographic variable values, and incentivize demographic data completeness and transition to these enhanced variable values.11	Pilot expanded fields for ethnic subgroups (Hispanic, Asian, others). Sponsor lines of research aimed at determining healthcare-relevant categories of race, ethnicity, and other sociodemographic variables.31 Evidence of what is healthcare relevant is required to collect the most relevant and actionable data, Example question: What level of specificity in race and ethnicity (for instance, Hispanic/Latino subgroup) data is meaningful and useful in health and healthcare research? Incentivize the standardized collection of insurance/payment information, with fields for non–fee-for-service programs.
	B2	Integrate a core set of individual-level social risk factors data into data collection.32	Use 11 National Academy of Medicine–recommended domains for SDH in EHRs,7 the Centers for Medicare and Medicaid Services Accountable Health Community Health-Related Social Needs screening tool,21 and PRAPARE.22 Create coding system to standardize and subsequently aggregate data from a variety of sources. Develop explicit language around data security and privacy related to sensitive topics (ie, work, refugee status, immigration, incarceration history).33 Provide funding to evaluate the utility of these measures for health equity research.
	B3	Integrate data on food insecurity and housing stability into data collection.34,35	Include items revealing level of food security and housing stability by relying on existing measures as described above.
	B4	Improve the standardization of data fields/formats for behavioral risk and protective factors.7,13,32,36	Use existing measures such as those from the National Academy of Medicine15 and PROMIS measures focused on behavioral risk and protective factors to capture these data in a standardized fashion.30 Provide additional values to differentiate between smoking and smokeless tobacco use, provide items to increase specificity in multisubstance use (substance and frequency for each) and include diet and exercise in behavioral health questions.
	B5	Develop data fields for barriers unique or significant to rural populations.37	Include distance to hospital, access to transportation, healthy food, internet availability, size of community, social service availability, clinic/provider/caregiver density, immigration status, and pharmacy density in the dataset. Engage rural patients in identification and piloting of relevant fields leveraging existing networks such as the CRNs in PCORnet or the patient-powered research networks (previously part of PCORnet, which leveraged specific patient communities to inform and contribute to research). Patient experience measures (discussed above) to include questions about perceived stigma related to demographic or health status and privacy concerns accompanying small community living.
C. Important additional data	C1	Create data fields for integration of the datasets such as those listed below. Provide technical expertise and best practice workflows for linkage to these existing datasets and incentivize the development of these linkages.17,29	1. High priority data: a. Geocoded data providing community-level variables (eg, American Community Survey data) b. National Death Index, state-level vital statistics c. Medicaid and Medicare claims data 2. Secondary: Criminal justice system (including juvenile justice), Department of Human Services data (foster care, Adult Protective Services), state cancer registries.
	C2	Include emergency department and hospital data.17	Provide incentives for patient-level data that includes ambulatory, emergency department, inpatient information and long-term care.

CRN: clinical research network; DIPEx: Database of Individual Patients’ Experiences’; EHR: electronic health record; PCORnet: Patient-Centered Outcomes Research Network; PRAPARE: Protocol for Responding to and Assessing Patients’ Assets, Risk, and Experiences; PRO: patient-reported outcome; PROMIS: Patient-Reported Outcomes Information System; SDH: social determinants of health.

Inclusion of patient voice highlights for the accurate representation of patient values, experiences, perspectives, and priorities related to healthcare needs and barriers. This includes data collected from patients with validated and standardized measures including patient-reported outcomes, as well as patient narratives and various types of patient experience data. The use of patient-reported outcomes in EHRs for both clinical care and research is becoming more common and availability of tools is increasing with the PROMIS (Patient-Reported Outcomes Information System).³⁰

The Database of Individual Patients’ Experiences’ (DIPEx)⁶ is an example of systematic qualitative data collection that amplifies patient voice and provides high-quality data for multiple uses. Qualitative narratives outside of the EHR and the clinical setting are gathered and banked for multiple uses in research, like any repository or research data embedded or attached to a patient’s medical record. Qualitative data collection is resource heavy and carries a high patient burden; a database of qualitative data for sharing creates an invaluable efficiency for both patients and researchers while also providing increased opportunities for research. While all patient voices may be challenging to gather, patient experience may prove invaluable as research data and is one step toward greater incorporation of patient voice in health disparities research.^11,28,29 The inclusion of the patient’s perspective in datasets also requires an awareness of patient burden and how the data collection process may impact them; this must subsequently inform methodology and may inform research questions.

Accurate and relevant variables includes both the standardization of variables and the determination of their definitions, categories, and purpose. Variables without widespread consensus, operational definition, and standardization include race, ethnicity, sexual identity, gender identity, income, insurance (including coverage with any capitated or alternative payment programs), and health behaviors.^11,34 The National Academy of Medicine identifies 11 domains for collecting data on SDH providing a first step in standardization of these items. The domain are the following: race or ethnic group, education, financial resource strain, stress, depression, physical activity, tobacco use, alcohol use, social connection or isolation, intimate partner violence, and neighborhood income.⁷

High-risk variables, such as nativity and immigration status, which may be crucial to health status, have significant challenges and risks to collection. Many other core sociodemographic elements are currently collected in a format that is not useful for clinical or research purposes. In addition, response categories may be either overly broad or overly restrictive. Therefore, they do not meaningfully represent the population of interest.

In addition to these technical concerns, certain information is often absent or inconsistently present in research datasets.³¹ Individual social risk factors such as food insecurity and housing should be included because of their clear relationship to health and healthcare outcomes.³⁴ In addition, capturing and standardizing data on patients’ prioritization of individual social risk factors, can inform health care how to best address such need.^11,28,34 Collection of behavioral factors is inconsistent and lacks standardization. The multidimensional aspects of care for patients in rural areas, including access and transportation challenges, unique experiences associated with rural living, and the effects of lower healthcare density, are simply missing from many data sources.³⁷

Adding variables to the CDM is a key theme for improving research readiness for this information. Important data linkages suggests that linking clinical data to datasets held by federal or state agencies (ie, hospital and emergency room records, Medicaid claims data, birth and death data, Department of Health and Human Services data), maintained registries, or collected by other research initiatives would significantly increase the breadth and depth of equity research.¹⁷ Linkages to existing datasets in a systematic way would increase the research readiness of the CDM substantially by providing a more comprehensive picture of patient health.²⁹ Because ambulatory data are often not in the same dataset as emergency department and hospital data, priority should be given to merging these records whenever possible. Including hospital data in more networks and specifically emergency department data would provide a more complete picture of health and health care for patients included in the CDM.

Each area described previously includes multiple recommendations. In addition, a key component of recommendation development was consideration of high-impact implementation strategies. Table 1 lists the recommendations in each area (in no particular order) and provides examples of fundamental implementation tasks in each recommendation prioritized based on potential impact and feasibility.

DISCUSSION

These recommendations are primarily to inform PCORnet, who may implement some of them in updates to the CDM. Such changes would improve the data available for health equity research across multiple institutions and nationwide. Increasing numbers of researchers are seeking to use PCORnet data in their work and such changes could have significant positive impact on their research capacity and quality. These recommendations may also be implemented by others as well, such as national and state registry holders, the ONC, or EHR manufacturers. An example may be ONC adopting recommended strategies for improving demographic or behavioral health data or requiring the use of a standardized measure for collecting data on individual-level social factors. Individual EHR manufacturers may choose to apply the strategies recommended here to standardized and improve data for their customers. State and federal registries may adopt some of these strategies to standardize data within their registry or across registries.

Health Level 7 International is a not-for-profit organization committed to the development and support of standards for interoperability and electronic data sharing. Projects in their index may serve as examples or even models for implementing these recommendations.³⁸ Nursing and allied health data such as those captured in SNOMED CT (Systematized Nomenclature of Medicine Clinical Terms) may be an additional rich resource for potential linkage or to serve as models for data collection and variables. Several suggested strategies include incentivizing work. Federal bodies such as the National Institutes of Health, Centers for Disease Control and Prevention, and Agency for Healthcare Research and Quality, among others, could provide funding specific to piloting some of the strategies. They have the opportunity to set a standard by requiring use of specific variables or measures for projects they fund. There are multiple avenues for using and applying these recommendations and strategies to promote better data for use in health equity research.

Limitations

Our recommendations have limitations. This was a qualitative data collection, thematic analysis, and consensus-based recommendation development with a select group of experts heavily involved in this work. While this is not a representative sample of experts, which does limit the rigor of our findings, we have not found these ideas elsewhere in the literature and, given the expertise of the group, felt that it was imperative to articulate these to the broader community.

Implementation of these recommendations must be done in conjunction with clinicians and healthcare centers. Gathering information from patients can require special training and significant effort for healthcare providers. Potential risk to patients must be considered including traumatization or retraumatization, time and energy costs, the process of information gathering must be made therapeutic and meaningful for patients, not taxing and invasive. An additional risk in collecting data from patients around social needs is the inability to have timely and relevant interventions to address such needs.³⁹ The recommendations require significant coordination across networks, and without EHR standardization guidelines across clinics and health systems, that will likely pose a challenge. PCORnet is the prime resource to support this coordination so while it may not be easy, there is infrastructure and support for coordinated change across engaged networks.

CONCLUSION

There are significant limitations to the currently available data for use in health equity research. The Health Disparities Collaborative Research Group was tasked with developing data science recommendations to nationally improve the data for health equity research. While these recommendations may prove useful to research in all areas, the focus here was specifically health equity research. Recommendations include inclusion of patient voice, accurate and relevant variables, and important additional data. These recommendations offer early steps in improving data science to increase capacity, quality, and relevance of health disparities research. Implementation of these ideas will be challenging, but with national networks such as PCORnet, it is possible and essential for greater health equity in our society.

FUNDING

This work was supported by contract Patient-Centered Outcomes Research Institute Award CDRN-1306-04716 (to JP). The opinions presented in this work are solely the responsibility of the author(s) and do not necessarily represent the views of the Patient-Centered Outcomes Research Institute, its Board of Governors, or Methodology Committee.

AUTHOR CONTRIBUTIONS

All authors contributed to this manuscript in accordance with the International Committee of Medical Journal Editors guidelines for authorship including substantial contributions to the analysis and interpretation of data, drafting guidelinesproviding critical revisions; approving final content; and agreeing. to be accountable. Additional author-specific contributions are as follows: RGB co-led conception and design, analysis, interpretation of data, drafted and led revision, submitted final approval for publication, and is primarily accountable for all aspects of the work. JP provided data science expertise; . EC provided expertise in SDH and patient voice ; MRL provided expertise in vulnerable populations and healthcare disparities; MJD, ARQ, and BC provided expertise in healthcare disparities; WP made substantial contributions to conception of the work and interpretation of data including supporting drafting guidelines. JH co-led conception and design of the work and analysis and interpretation of data, drafted and revised of the work, submitted final approval of the version to be published, and is primarily accountable for all aspects of the work.

SUPPLEMENTARY MATERIAL

Supplementary material is available at Journal of the American Medical Informatics Association online.