Evaluating Redundancy and Biases in EHR Social Determinants of Health Data Screening

This study analyzes SDOH screening data from 1.8 million patients to demonstrate methods for identifying redundant questions and demographic biases, offering strategies to streamline clinical workflows and ensure more equitable data collection.

The Gist

Imagine your healthcare system as a giant, busy kitchen. For a long time, chefs (doctors) only cooked based on the ingredients they could see right in front of them, like fresh vegetables or meat. But recently, they realized that a meal's success also depends on things happening outside the kitchen: Is the family hungry? Can they afford the groceries? Do they have a car to get to the store? These outside factors are called Social Determinants of Health (SDOH).

To help, the kitchen started handing out a long questionnaire to every customer before they ordered, asking about their life outside the hospital. But the authors of this paper asked a simple question: "Is this questionnaire actually working, or is it just making a mess?"

They looked at data from 1.8 million patients (a massive crowd!) to see if the questions were doing their job. Here is what they found, explained through some everyday analogies:

1. The "Double-Check" Problem (Redundancy)

Imagine you are filling out a form, and Question 1 asks, "Do you have enough money for food?" and Question 2 asks, "Are you worried about your grocery bill?" and Question 3 asks, "Have you skipped meals recently?"

The researchers found that for many patients, these questions were all asking the exact same thing in slightly different words. It's like a security guard asking you for your ID, then asking for your driver's license, and then asking for your passport, even though they all prove the same thing: who you are.

The Result: The team found that questions about food and money were often "echoing" each other. By realizing this, the hospital can cut out the extra questions, saving time for both the staff and the patients. It's like removing the extra steps from a recipe to make dinner faster without changing the taste.

2. The "Unfair Invitation" Problem (Bias)

Next, they looked at who was actually being asked these questions. Imagine a party where the host only invites people wearing red shirts to the dance floor, while ignoring everyone in blue shirts. That would be unfair, right?

The study found that the hospital's system was doing something similar:

• Who got asked: Female patients and White patients were more likely to be asked these important questions than other groups. It was like the host only dancing with certain people.
• Who said "No": When people were asked, some declined to answer. The study found that American Indian/Alaska Native and Hispanic/Latino patients were less likely to say "no" compared to others. This suggests that the way the questions were asked, or the trust levels in the system, might be different for different groups.

The Big Takeaway

The authors aren't saying "stop asking about people's lives." They are saying, "Let's ask smarter."

They showed hospitals how to:

1. Trim the Fat: Cut out the duplicate questions so the process is faster and less annoying (like editing a long, repetitive email down to just the important points).
2. Level the Playing Field: Make sure everyone gets asked the questions fairly, regardless of who they are, so the hospital gets a true picture of its community's needs.

In short: This paper is a guide for hospitals to stop asking the same question three times and to make sure they aren't accidentally ignoring the people who need help the most. It's about making healthcare data collection fair, efficient, and actually useful.

Technical Summary

1. Problem Statement

Healthcare organizations are increasingly integrating Social Determinants of Health (SDOH) screening into clinical workflows to address factors influencing patient outcomes. However, two critical issues hinder the efficacy and equity of these initiatives:

• Redundancy: SDOH screening tools often contain overlapping questions that provide duplicate risk information, leading to unnecessary clinical burden and workflow inefficiency.
• Bias: There is a lack of understanding regarding whether the administration of these screens and patient responses are systematically biased based on demographic factors (e.g., race, gender, ethnicity). This can result in skewed data and unequal access to support services.

2. Methodology

The study utilized a large-scale retrospective analysis of Electronic Health Record (EHR) data from 1.8 million patients within the UNC Health system. The authors employed a two-pronged analytical approach:

• Redundancy Analysis:
  • The team analyzed response patterns across pairs of SDOH questions.
  • They calculated response agreement metrics to determine the degree of correlation between different questions. High agreement indicates that multiple questions are measuring the same underlying risk factor, suggesting potential redundancy.
• Bias Evaluation:
  • Demographic Disparities in Screening: Logistic regression models were used to evaluate whether the likelihood of a patient being asked specific SDOH questions varied significantly by demographic groups (gender, race, ethnicity).
  • Demographic Disparities in Refusal: Similar logistic regression models were applied to analyze the likelihood of patients declining to answer questions across different demographic groups.

3. Key Contributions

• Methodological Framework: The paper establishes a reproducible, data-driven methodology for healthcare organizations to audit their own SDOH screening protocols. It moves beyond theoretical concerns to provide statistical tools for quantifying redundancy and bias.
• Workflow Optimization Strategy: By identifying specific question pairs with high redundancy, the study offers a concrete path to streamline clinical workflows, reducing the time clinicians and patients spend on screening without sacrificing data quality.
• Equity Audit Mechanism: The study provides a framework for detecting systemic biases in both the administration of screens (who gets asked) and the response rates (who refuses), enabling organizations to implement more equitable screening protocols.

4. Key Results

• Redundancy Findings: Significant redundancy was identified in the risk information gained from the screening tool. Specifically, questions related to food insecurity and financial insecurity showed high response agreement, suggesting that multiple questions in these domains may be capturing the same risk profile.
• Screening Bias: The analysis revealed demographic disparities in who was asked questions. Female and White patients were statistically more likely to be administered certain SDOH questions compared to other demographic groups.
• Response Bias: Disparities were also found in refusal rates. American Indian or Alaska Native and Hispanic or Latino patients were significantly less likely to decline answering questions compared to other groups, indicating complex interactions between trust, cultural factors, and screening administration.

5. Significance

This research is significant for the future of value-based care and health equity:

• Operational Efficiency: By eliminating redundant questions, healthcare systems can reduce "screening fatigue" for both providers and patients, potentially increasing overall compliance and data completeness.
• Data Integrity and Equity: Identifying demographic biases allows organizations to standardize screening protocols, ensuring that SDOH data collection is consistent across all patient populations. This is crucial for accurately targeting social interventions and avoiding the exacerbation of existing health disparities.
• Scalability: The methods presented are applicable to any large healthcare system seeking to refine their SDOH data collection strategies, promoting a shift from ad-hoc screening to systematic, evidence-based protocols.