Accuracy of Electronic Health Record Food Insecurity, Housing Instability, and Financial Strain Screening in Adult Primary Care

Abstract

This study assesses the accuracy of electronic health record–based screening questionnaires about social risk factors using external single-domain questionnaires as a comparator.

Public health, care delivery organizations, federal agencies, and researchers have advocated for better collection of patient social risk factor information. Social factors can determine referrals to community partners, increase awareness of patients’ needs, measure population health, or improve risk prediction models.

Popular electronic health record (EHR) systems include screening questionnaires assessing various social risk factors.¹ While these multidomain screening questionnaires are commonly used in practice, their accuracy has not been established.² We assessed the accuracy of EHR-based multidomain questionnaires using single-domain questionnaires on food insecurity, housing instability, and financial strain as external standards.

Methods

Adults receiving care at 1 of 11 primary care clinics in Indianapolis, Indiana, and Gainesville, Florida, between January and September 2022 completed EHR-based and single-domain questionnaires concurrently during visits or via phone or email after their visit. The self-administered questionnaires were offered in English and Spanish.

Patients provided demographic information (see Table 1 for information on race and ethnicity) and completed questionnaires that are part of the Epic EHR as well as the US Department of Agriculture’s Six-Item Short Form of the Food Security Survey,³ the Housing Instability Index,⁴ and the Consumer Financial Protection Bureau’s Financial Well-being Scale (financial strain),⁵ in that order. Per the instructions of the questionnaire developers, we created binary indicators for positive screens (eTable in Supplement 1). We compared differences in the percentage of patients with a positive screen for each social factor using a z test. Using the single-domain questionnaire results as the external standard, we calculated sensitivity, specificity, positive predictive values, and the area under the curve (AUC) in Stata version 17 (StataCorp). Statistical significance was defined as a 2-tailed P < .05. This study was approved by the Indiana University institutional review board, and written informed consent was obtained.

Table 1. Characteristics of Adult Primary Care Patients Completing Screening Questions on Food Insecurity, Housing Instability, and Financial Strain.

Characteristics	No. (%) (N = 826)
Sexa
Female	528 (63.9)
Male	296 (35.8)
Race and ethnicityb
Black non-Hispanic	324 (39.2)
Hispanic/Latinx	56 (6.8)
White non-Hispanic	370 (44.8)
Multiple	40 (4.8)
Other or unknownc	36 (4.4)
Age, mean (SD), y	49.7 (17.7) [n = 825]
Language
English	818 (99.0)
Spanish	8 (1.0)
Education
Less than high school	54 (6.5)
High school graduate or equivalent	201 (24.3)
Some postsecondary education or higher	552 (66.8)

^a Sex as recorded in the electronic health record.

^b Participants self-reported race and ethnicity using Office of Management and Budget categories.

^c Other includes American Indian or Alaska Native and Native Hawaiian or Other Pacific Islander.

Results

Of 898 patients with encounters, 74.0% completed surveys during clinic visits. Of the 4716 contacted by phone or email after a visit, 4.2% responded. The sample of 826 patients (Table 1) was predominantly female (64%) and had a mean age of 49.7 (SD, 17.7) years. The sample was racially and ethnically diverse, with 39% Black non-Hispanic, 7% Hispanic/Latinx, and 45% White non-Hispanic individuals. For food insecurity, the prevalences from the EHR-based screening questionnaire and the single-domain questionnaire were not significantly different (38.0% [95% CI, 34.7%-41.4%] vs 36.0% [95% CI, 32.7%-39.3%]; P = .38). For housing instability, the prevalence from the EHR-based questionnaire was significantly lower than from the single-domain questionnaire (27.1% [95% CI, 24.0%-30.1%] vs 38.4% [95% CI, 35.1%-41.8%]), as was the prevalence of financial strain (10.4% [95% CI, 8.3%-12.5%] vs 33.2% [95% CI, 29.9%-36.4%]) (P < .001 for both).

The EHR screening questionnaire for food insecurity had an AUC of 0.938 (95% CI, 0.921-0.955) and both high sensitivity (94.5%; 95% CI, 91.2%-96.8%) and high specificity (93.1%; 95% CI, 90.6%-95.2%) (Table 2). The performance of both housing instability (AUC, 0.760; 95% CI, 0.731-0.790) and financial strain (AUC, 0.626; 95% CI, 0.599-0.653) was lower. For housing instability, the EHR screening questionnaire demonstrated high specificity (93.0%; 95% CI, 90.4%-95.1%) but low sensitivity (59.0%; 95% CI, 53.4%-64.5%). Likewise, for financial strain, specificity was high (98.1%; 95% CI, 96.6%-99.1%) and sensitivity was low (27.1%; 95% CI, 21.8%-32.8%). Each factor had a positive predictive value greater than 80%, meaning that a positive EHR-based screen would reflect the presence of the respective social factor in more than 8 of 10 patients.

Table 2. Diagnostic Performance of EHR-Based Social Risk Factor Screening Questions Among Adult Primary Care Patients.

Social risk factor	EHR-based prevalence, % (95% CI)	External questionnaire prevalence, % (95% CI)	Sensitivity, % (95% CI)	Specificity, % (95% CI)	Positive predictive value, % (95% CI)	Area under the curve (95% CI)
Food insecurity	38.0 (34.7-41.4)	36.0 (32.7-39.3)	94.5 (91.2-96.8)	93.1 (90.6-95.2)	88.4 (84.3-91.7)	0.938 (0.921-0.955)
Housing instability	27.1 (24.0-30.1)a	38.4 (35.1-41.8)	59.0 (53.4-64.5)	93.0 (90.4-95.1)	84.2 (78.7-88.7)	0.760 (0.731-0.790)
Financial strain	10.4 (8.3-12.5)a	33.2 (29.9-36.4)	27.1 (21.8-32.8)	98.1 (96.6-99.1)	87.8 (78.7-94.0)	0.626 (0.599-0.653)

Abbreviation: EHR, electronic health record.

^a P < .001 per z test for proportions comparing EHR-based and external questionnaire prevalences.

Discussion

Commonly used EHR-based social screening questionnaires underidentified patients with housing instability and financial strain compared with single-domain screening tools. This suggests that the primary goal of screening may not be achieved for these 2 risks. In contrast, EHR-based food insecurity screening was accurate.

Limitations include a sample from 2 health systems in 2 states and that the results may not be generalizable to other multidomain screening questionnaires, social factor domains, languages, or care settings. However, many other screening questionnaires use the same items as the EHR questionnaire in this study. While the screening and external single-domain questionnaires share much of the same wording, they may measure different constructs.

Assessing performance is the first step in understanding the utility and implications of screening results. Further research is needed on which and how such screening tools should be used.

Section Editors: Jody W. Zylke, MD, Deputy Editor; Kristin Walter, MD, Senior Editor.

Supplement 1.

eTable. Questionnaire Items

eReferences

Supplement 2.

Data Sharing Statement