Contextual Data in Action

Co-authored with Vera Mucaj

I recently posted an article arguing that real-world data analytics are constrained by their use of transactional health data, and that additional insight would be found through the addition of contextual data. Contextual data includes all of the information that surrounds a healthcare transaction, including the social, environmental, and behavioral attributes of the patient, the experience and behavioral attributes of the treating physician, and the physical resources and policies / procedures of the facilities in which the healthcare encounter occurs.

By necessity, I made this argument from the “40,000 foot” level and included a broad overview of the types of contextual data that I believe are most missing from health analytics. Following the publication of that piece, I was asked for some more specific examples of where an analysis would be improved with the addition of contextual data so that readers could more easily envision the practical applications of this improvement. Herein, Vera Mucaj, chief scientific officer at Datavant, and I describe example scenarios of the “before and after” of health analytics to show the additional insight that can be unlocked through contextual data.

Example 1: Rural Patient Outcomes

A researcher is analyzing the effectiveness of an anti-diabetes drug, and sees that patients in rural areas have poorer outcomes compared to patients living in cities. She is confused about why the medication would be less effective in rural areas.

Without Contextual Data, the researcher can establish that:

• Pharmacy claims data show that patients in rural areas are receiving the medications at the same rate as urban patients.
• Medical claims data indicate a higher rate of diabetes-related complications in rural patients.
• Electronic health records (EHR) confirm the prescription and complication rates, and the clinical interventions put in place, but offer no further insights.

With Contextual Data, the analysis changes:

• Physician profile data reveal that rural patients are treated by less experienced physicians who might not recognize the early signs of an impending complication. It also quantifies the limits in specialist care availability due to geographic isolation.
• System profile data suggest that rural clinics have fewer resources, leading to less frequent monitoring and follow-up.
• Health plan profile data show that rural patients are often on more restrictive insurance plans that don’t cover as many follow up appointments or testing.
• Social Determinants of Health (SDOH) data suggest that rural patients have less access to transportation, making it harder to attend follow-up appointments to monitor the effectiveness of their medication, and to catch potential complications early.

Example 2: Hospital Re-Admission Rates

A researcher uses EHR and medical claims data to compare the re-admission rates of heart failure patients seen at a number of hospitals. She notices that Hospital A has significantly higher readmission rates than the others. Based on this data alone, the researcher might conclude that Hospital A is providing inferior care.

Without Contextual Data, the researcher can establish that:

• Medical claims data indicate higher readmission rates for heart failure patients at Hospital A.
• Electronic health records confirm the diagnoses and treatment plans at Hospital A but do not provide reasons for readmissions.

With Contextual Data, she has a clearer picture:

• Physician profile data show that Hospital A employs a higher proportion of early-career physicians who might be more conservative, leading to precautionary readmissions.
• System profile data suggest that Hospital A has fewer outpatient follow-up resources, making it difficult for patients to receive adequate post-discharge care to prevent re-admission.
• Health plan profile data indicate that a larger percentage of patients at Hospital A are covered by insurance plans with limited coverage for home health care services.
• SDOH data highlight that the hospital serves an economically disadvantaged area where patients might lack access to reliable transportation and face difficulties adhering to post-discharge care plans.

Example 3: Maternal Mortality

Maternal mortality has been under-reported in health statistics and new reporting shows rising numbers (ref. 1) — how do we get the right numbers to understand this issue?

Without Contextual Data, the researcher can establish that:

• Patient demographic information present in claims data show higher mortality rates for vulnerable populations, but not as high as reported by other sources.
• EHR data show that mortality is higher for mothers who present to the emergency room for birth, but don’t give insight into why.
• Clinical complications occurring during delivery and associated interventions can be analyzed but do not record any contributing factors or specific reasons beyond the clinical practice.

With Contextual Data, the researcher now sees that:

• Mortality data shows that number of deaths in vulnerable population is significantly higher than is recorded in health records alone.
• Health plan profile data shows population is less insured and has not had regular pre-natal care to identify and treat complications early.
• SDOH shows that patients of a lower socioeconomic status (SES), who are a minority, or part of other vulnerable populations are more commonly giving birth outside of healthcare setting.
• Behavioral and Lifestyle data shows vulnerable populations are more likely to live in food deserts, less likely to purchase OTC prenatal vitamins, and have less access to prenatal education and services and postnatal community support.
• System profile data on staffing levels and available resources / infrastructure, in combination with SDOH data, shows that vulnerable women are more likely to be treated in overcrowded facilities, with longer ER wait times, and fewer resources or specialists.
• Physician profile data indicates that certain areas have limited training in delivery complications due to the experience levels and practice patterns of physicians. This shortage results in limited rapid response systems for emergency births.

Example 4: Predicting Asthma Attacks

A researcher wants to build a predictive model to identify patients most at risk of an asthma attack.

Without Contextual Data, the researcher’s model can incorporate:

• Diagnosis timelines and associated treatment plans for patients with asthma.
• Patient adherence with prescribed medication(s) — e.g., refill rates and patient reported adherence.
• Correlations between clinical procedures, diagnosis events and patient comorbidities for asthma management.
• Association and segmentation of patients who have frequent asthmas attacks by recorded demographic variables (primarily age, gender, and geographic area).

With Contextual Data, the researcher can improve her model with:

• Occupational data revealing potential exposure triggers and asthma-related absences.
• Environmental data that can highlight the impact of air quality (expose to pollutants) and weather conditions (pollen counts).
• SES and housing data that shows where lower socioeconomic areas have higher risk of exposure to indoor allergens, and have less access to health education, and community health programs.
• Behavioral and Lifestyle data that includes information on smoking status, physical activity level, diet, and stress.
• Healthcare system data that provides insights into specialist provider availability.
• Payer data showing affordability and insurance coverage of rescue inhalers and maintenance medication to understand financial barriers associated with asthma management.

References:

1. https://ourworldindata.org/rise-us-maternal-mortality-rates-measurement. See also: https://pubmed.ncbi.nlm.nih.gov/31639369/

…………………………………………………………………………………………………

Veritas Data Research is on a path to building best-in-class contextual data sets, which will form the foundation for better understanding of patient experiences. Datavant seeks to connect the world’s health data through privacy-preserving linkage of deidentified RWD. Visit them at www.veritasdataresearch.com or www.datavant.com to learn more.