Comparing Patient-level Social Drivers of Health from Health Surveys and Electronic Health Records for Patients with Comorbid Hypertension and Uncontrolled Diabetes

Abstract

Social drivers of health significantly influence diabetes and hypertension outcomes. By taking into account patients’ social and economic circumstances, healthcare systems can enhance both the quality and efficiency of care delivery, leading to improved health outcomes. This study aims to assess the concordance between patient-level social drivers data gathered from a patient-reported, health-related social needs survey and the data documented in electronic health records. A comparative analysis was conducted among 165 adults diagnosed with coexisting hypertension and uncontrolled diabetes from a singular academic health system. Each participant engaged in a standardized assessment of health-related social needs survey, and the corresponding electronic health record-based social drivers of health data were extracted. Concordance at the patient level for social drivers of health was assessed using Cohen’s Kappa and percent agreement. Overall, agreement between the patient-reported social needs survey and electronic health records data was low, indicating only slight alignment across various social drivers of health domains. These findings suggest that relying solely on electronic health records data may underestimate the true prevalence of patient-reported social needs in this high-risk cohort with diabetes and hypertension. To ensure high-quality care delivery, there is a critical need for healthcare systems to develop more effective and sustainable methods for capturing social drivers of health data.

Introduction

Social drivers of health (SDOH) significantly influence healthcare outcomes and expenditures in the United States. SDOH includes non-medical economic and social factors that impact an individual’s health, such as financial strain, transportation issues, mental health, and living conditions.¹ According to the World Health Organization,² SDOH account for approximately 30-55% of health outcomes. However, over 70% of national healthcare expenditures were allocated to direct medical treatments such as hospital care, prescription drugs, and procedures, indicating a significant imbalance in the distribution of the healthcare budget.³ In chronic disease management, social factors can consistently affect patients’ self-management and treatment adherence in the long term due to factors such as access to care and affordability of medications.⁴ In a high-risk cohort with diabetes and hypertension, individuals with high social risks experience worse clinical outcomes and a lower quality of life compared to others.²⁷

To provide holistic care and support patients beyond the clinical encounter, healthcare systems have been integrating SDOH data using various methods, such as standardized patient surveys, electronic health record (EHR) integration, community health assessments, and utilization of public data sources. Among these, methods that capture individual-level SDOH data have gained significant attention due to their predictive power for various health outcomes, such as enhancing predictions for service referrals and hospitalization risk.^{5, 6} Various standardized measurements have been developed to collect individual-level SDOH data effectively. One example is the Accountable Health Communities (AHC) Health-Related Social Needs (HRSN) Screening Tool, which has become helpful for systematically evaluating individual SDOH factors in healthcare settings.^7,8

Integrating SDOH data into EHRs has also been recognized as a valuable method to capture individual-level SDOH data. Integrating standardized SDOH data into EHRs offers an opportunity to enhance personalized care and health equity by identifying social risks that impact clinical outcomes. For instance, when information about housing instability or food insecurity is included in EHRs, health systems can not only predict clinical outcomes such as 30-day hospital readmission risk, but more importantly, they can proactively identify and address these critical social needs through targeted interventions that improve overall patient health.^{5, 6}

However, the practical application of EHR-based SDOH data in clinical care remains limited due to several challenges. One significant challenge is the lack of comparability in SDOH data collection and documentation. The definition and risk stratifications of social domains vary across healthcare systems.⁹ Social needs are defined and assessed differently depending on the healthcare system, leading to inconsistent data and difficulty in comparing results across institutions.^{8, 10} Researchers also face challenges in using EHR-based SDOH data for comparative studies of social risks due to variation in risk stratification, which affect data usability and representativeness.⁸

Completeness of SDOH documentation represents another critical issue. Linfield et al⁸ observed that certain SDOH domains, such as food insecurity, transportation needs, and utility security, are less frequently documented in EHRs than in other domains. On the other hand, multiple barriers affect SDOH documentation in clinical practice. Healthcare providers often face time constraints resulting in incomplete or rushed SDOH assessments.^{9, 11} Patients may also be reluctant to share sensitive information, such as financial strain or housing instability, which results in underreporting SDOH factors in EHRs.⁸

Furthermore, data accuracy affects the reliability of EHR-based SDOH data. When SDOH documentation fails to capture patients’ social needs, institutions may risk misallocating limited resources and miss opportunities to help those needing support most.¹² Furthermore, it is notable that inaccurate SDOH documentation could disproportionately impact racial minorities. Black and Hispanic patients’ EHRs were more likely to contain erroneous SDOH data, which could potentially skew research findings and exacerbate health inequities.¹³

Given these challenges, evaluating the comparability, completeness, and accuracy of EHR-based SDOH data is crucial. One method for evaluation is to compare it with patient-reported data collected through standardized surveys. The AHC HRSN survey provides a structured and comprehensive assessment of social needs, whereas the customization allowed in the EHR introduces variability and potential data gaps.¹⁴ To increase the comparability and generalizability of social risk data extracted from EHRs, it is essential to understand the link between these two data sources. This comparison will allow us to evaluate how well these two sources align, identify inconsistencies or gaps in EHR-based SDOH documentation, and contribute to a more thorough and accurate representation of patients’ social needs. This knowledge can guide the design of more effective social risk-related interventions for patients with comorbid hypertension and uncontrolled diabetes, leading to better patient care and more targeted strategies to address health disparities. Therefore, this study aims to compare SDOH data collected through the AHC HRSN survey with EHR-based SDOH data for a high-risk cohort with uncontrolled diabetes and hypertension. This study assesses how well the two sources align and identifies which SDOH domains show the greatest or least concordance.

Methods

Parent Study Overview. This study analyzed SDOH data collected within the EXpanding Technology-Enabled, Nurse-Delivered Chronic Disease Care (EXTEND) randomized controlled trial (N=220). Participants were recruited from five primary care and endocrinology clinics in Durham, NC, USA.¹⁵ These clinics serve a diverse patient population, with over 54% identifying as Black/African American and 14% as Latinx. The ongoing EXTEND trial was designed to compare two 12-month interventions for managing uncontrolled diabetes and hypertension. Adults aged 18 years or older were eligible if they had (1) uncontrolled diabetes (HbA1c ≥ 8.0%) and (2) hypertension (systolic blood pressure ≥ 140 mmHg or diastolic blood pressure ≥ 90 mmHg). Participants were randomized to either EXTEND Plus (intervention, n=110) or EXTEND (attention control, n=110) arm.¹⁵ The EXTEND Plus (intervention) group received nurse-led telehealth care, including mobile monitoring with four health-tracking devices, personalized self-management support, and pharmacist-guided medication management through scheduled remote encounters. In contrast, the EXTEND (attention control) group received mobile monitoring-aided self-management alone, using the same four health-tracking devices without structured telehealth support or pharmacist involvement.¹⁶ Both groups were followed to assess changes in HbA1c and blood pressure as primary outcomes. Additional details on the protocol and technology build are published elsewhere.^{15, 16} The EXTEND trial was reviewed and approved by the Duke University Health System Institutional Review Board (Pro00107722). Participants provided written informed consent permitting the use of their survey data and linkage with EHR information for subsequent research.

Study Design. We conducted a comparative analysis of SDOH data from two sources: the AHC HRSN Screening Tool and EHR-based SDOH documentation from a health system’s Epic-based electronic health record (EHR).^{7, 8} The AHC HRSN Screening Tool, hereafter referred to as the HRSN survey, was introduced after the initial EXTEND trial enrollment completed. This was in response to patients reporting a number of social needs to clinical and research staff and barriers to intervention engagement, such as limited access to transportation and challenges with access to resources for managing their diabetes, among other needs. Thus, our analysis included the 165 participants who completed the HRSN survey and had at least one SDOH documentation in their EHR within ±12 months of the HRSN survey completion date. We selected this 12-month timeframe to align with the HRSN survey’s standard recall period, which captures social needs experienced within the past year.⁷ This alignment ensures a more appropriate comparison between the two data sources. This study focused on measuring concordance between the two data sources at approximately similar time points rather than analyzing longitudinal changes in SDOH status.

Data Collection and Measures. The HRSN survey was administered during a follow-up visit in the EXTEND trial to assess SDOH across 13 domains. The survey included five core domains (living situation, food, transportation, utility, and safety) and eight supplemental domains (financial strain, employment, family and community support, education, physical activity, substance use, mental health, and disability). In total, 32 questions assessed social needs within these 13 domains, with six demographic items. The HRSN survey used three response formats: binary (Yes/No), multi-option categorical responses, and culminative scoring to accommodate different types of responses. Binary questions, such as those for utility, classified a participant as at risk if they answered “Yes” (e.g., “In the past 12 months, has the electric, gas, oil, or water company threatened to shut off services in your home?”). Multi-option categorical responses, such as those in the food domain, included choices such as “often true,” “sometimes true,” and “never true.” Cumulative scoring was applied in specific domains, including physical activity, where multiple questions were used to determine risk thresholds. The HRSN survey classified any response other than “never true” as indicating risk, treating even minimal presence of need as positive indicators across all domains. Each domain was evaluated independently according to the survey protocol without calculating an overall score.

SDOH data were also extracted from the EHR. Data extraction focused on twelve social risk domains, including tobacco use, alcohol use, financial resource strain, food insecurity, transportation needs, physical activity, stress, social connections, depression, housing stability, utilities, and health literacy. The EHR used a color-coded risk stratification method: green for no or minimal risk, yellow for moderate risk, red for high risk, and grey for unknown risk or missing data. When multiple data points were available, all relevant dates were included. In October 2024, one researcher extracted the initial EHR data, while the second researcher reviewed and validated the extracted data. Data were documented in REDCap (Research Electronic Data Capture),^{17, 18} a secure electronic database for research study management.

Domain Mapping. To compare social needs consistently between the HRSN survey and the EHR, we standardized domain names to emphasize present unmet needs. We identified nine comparable domains between the two sources through the domain mapping process. Seven had a direct one-to-one match (e.g., “Utilities” in HRSN mapped directly to “Utilities” in Epic). To align with the broader HRSN classification, we merged multiple Epic fields into single domains for two categories. Specifically, for the “Substance Use” domain, two Epic fields (Tobacco Use and Alcohol Use) were combined into a single domain. Likewise, for the “Mental Health Needs” domain, two Epic fields (Stress and Depression) were merged. On the other hand, Safety, Employment, Education, and Disabilities from the HRSN survey had no corresponding EHR fields, whereas Health Literacy was documented only in the EHR (Table 1).

Table 1.

Domain Mapping

Domain	HRSN Screening Tool	Epic-based EHR
Utility Needs	Utilities	Utilities
Transportation Needs	Transportation	Transportation Needs
Financial Strain	Financial Strain	Financial Resource Strain
Food Insecurity	Food	Food Insecurity
Substance Usea	Substance Use	Tobacco Use, Alcohol Use
Housing Instability	Living Situation	Housing Stability
Mental Health Needsa	Mental Health	Stress, Depression
Family/Community Needs Physical Inactivity	Family and Community Support Physical Activity	Social Connections Physical Activity
Safety Needs	Safety	NA
Employment Needs Education Needs	Employment Education	NA NA
Disabilities	Disabilities	NA
Health Literacy Needs	NA	Health Literacy

Note: Domain names were adapted from the original HRSN labels or Epic EHR categories to ensure alignment with the two sources and to emphasize unmet needs. All domains were coded consistently (0=no needs, 1=unmet needs present).

^aMultiple EHR fields were merged for this domain. HRSN: Health-Related Social Needs; EHR: electronic health record; NA: not applicable.

Data Integration. Because the HRSN survey treated any reported need as positive while the EHR initially used multiple risk levels, we recoded the EHR data into a binary format (0 = no unmet need and 1 = any level of unmet need) to align with the HRSN survey’s coding. Records coded as grey or unknown were treated as missing. For domains where multiple EHR fields were merged (Substance Use and Mental Health Needs), the presence of any unmet need in any of the subfields was coded as having an unmet need (1). In contrast, the absence of unmet needs in all component fields was coded as no unmet need (0). The entire domain was coded as missing if any subfield records were missing. We then created a binary outcome variable (1=match, 0=mismatch) to compare HRSN survey data with EHR data for each domain. We only analyzed participants with valid entries in both sources. For each participant, we computed the number of days between the HRSN survey and the closest EHR entry. Since the HRSN survey covered 12 months, we excluded any EHR records that lay outside of ±12 months of the survey date. This screening step ensured that only temporally comparable data were included in the concordance analysis. We then selected the EHR record closest to the survey date and matched the survey responses with corresponding EHR data using unique patient identifiers. If no EHR documentation was available within the ±12-month window of the HRSN survey date, the EHR component of that domain was coded as missing.

Statistical Analyses. Descriptive statistics were used to detail the sample characteristics and key analytic variables. All analyses used SAS (version 9.4, SAS Institute INC., Cary, NC). Missing data were examined by categorizing three types of missingness: HRSN survey only missing, EHR only missing, and missing in both sources. For each domain, we calculated the frequency and percentage of missing values in each category. We calculated the time difference in days between each participant’s HRSN survey date and their closest corresponding EHR documentation date. These differences were converted to absolute values, as our focus was on the magnitude rather than the direction of the time gap. Concordance between the HRSN survey and EHR measures for each domain (each source coded as 0 = no needs, 1 = unmet needs present) was determined using 2 x 2 cross-tabulation tables. The concordance analysis was limited to cases when both sources provided valid responses. Two metrics were used to assess concordance. The percentage of matching responses (percent agreement) among all valid responses was calculated to provide a direct measure of how often these sources aligned. Simple percent agreement can be inflated when most observations fall into a single category (both sources indicated absence or both sources indicated presence of unmet needs). Therefore, Cohen’s Kappa statistic, which adjusts for agreement beyond chance, was also calculated for each domain with 95% confidence intervals (CIs). Landis and Koch’s criterion¹⁹ was applied to describe the degree of concordance using the following Kappa value cutoffs: (1) slight, ≤0.20; (2) fair, 0.21-0.40; (3) moderate, 0.41-0.60; (4) substantial, 0.61-0.80; and (6) almost perfect, 0.81-1.00.

Results

Participant Characteristics. We analyzed 165 participants from the EXTEND trial, of whom 84 participants (50.9%) received additional nurse-led support (EXTEND Plus, intervention) and 81 participants (49.1%) received standard care (EXTEND, attention control). The mean age of the participants was 54.4 years (SD=10.1; range: 48-62) with 69.1% aged 30-59. The majority of participants were Black/African American (76.0%) and female (68.3%). With 40.7% having a bachelor’s degree or higher education, most had completed more education than high school (76.5%). Approximately half were married or partnered (47.5%) and employed (58.0%), and more than half of the participants (59.6%) reported an annual income of less than $40,000 (Table 2).

Table 2.

Demographic and Clinical Characteristics of Participants (N=165)

Characteristic	Na	n (%)
Parent Trial Arms	165
Intervention (EXTEND Plus)		84 (50.9%)
Control (EXTEND)		81 (49.1%)
Age, in years	165
Age 30 to 59		114 (69.1%)
Age 60 or older		51 (30.9%)
Gender	164
Female		112 (68.3%)
Male		52 (31.7%)
Race	150
Black/African American		114 (76.0%)
Otherb		36 (24.0%)
Education Level	162
High school or less		38 (23.5%)
Some college/Associate		58 (35.8%)
Bachelor or higher		66 (40.7%)
Married/Partnered	160	76 (47.5%)
Employed	157	91 (58.0%)
Annual Household Income	141
Less than $40k		84 (59.6%)
$40k or greater		57 (40.4%)

^aNumber of participants with available data for each characteristic.

^bOther race includes White, Asian, American Indian/Alaska Native, and More than one race.

Domain-Specific Data Available and Missing Data. Table 3 details the data available for the concordance analysis and missing data for each social need domain. The missing data results indicated the HRSN survey generally had lower missing rates. The EHR missing data, however, varied considerably, with rates exceeding 70% for Physical Activity, Family/Community Support, and Mental Health. To be included in the concordance analysis, the participant must have HRSN and EHR data available for the specified domain to determine percent agreement and calculate a kappa statistic. Thus, the sample size for the subsequent domain-specific concordance analyses included only those participants with data from both sources (HRSN and EHR data available). Timing distributions for the retained records are provided in Figure 1; all retained entries occurred within ±365 days of survey completion, as records outside that range were excluded from the concordance analysis. The sample sizes for the domain-specific concordance analysis ranged from 45 to 117 participants.

Table 3.

Domain-Specific Data Available and Missing

Domain	HRSN & EHR Data Available n (%)	Missing data pattern
HRSN only missing n (%)	EHR only missing n (%)	HRSN & EHR missing n (%)
Utility Needs	90 (54.5%)	5 (3.0%)	67 (40.6%)	3 (1.8%)
Transportation Needs	117 (70.9%)	4 (2.4%)	44 (26.7%)	-
Financial Strain	108 (65.5%)	9 (5.5%)	46 (27.9%)	2 (1.2%)
Food Insecurity	112 (67.9%)	8 (4.9%)	45 (27.3%)	-
Substance Usea	67 (40.6%)	-	98 (59.3%)	-
Housing Instability	80 (48.5%)	-	85 (51.5%)	-
Mental Health Needsa	48 (29.1%)	-	117 (70.9%)	-
Family/Community Needs	45 (27.3%)	-	120 (72.7%)	-
Physical Inactivity	45 (27.3%)	-	120 (72.7%)	-

Note: Row percents provided for N=165. HRSN & EHR Data Available n (%) for each domain indicates the number and percentage of 165 participants who had valid data from both the HRSN survey and the EHR. HRSN only missing denotes participants lacking HRSN data, EHR only missing denotes participants lacking EHR data, and HRSN/EHR missing indicates participants missing data from both sources.

^aMultiple EHR fields were merged for this domain. HRSN: Health-Related Social Needs; EHR: electronic health record.

Figure 1.

Differences in days between HRSN survey completion and EHR completion. The diamond indicates the mean value, the horizontal line within the box represents the median, and the circles denote outliers exceeding 1.5 times the interquartile range. *Multiple EHR fields were merged for this domain.

Concordance Analysis. The level of concordance between the two data sources was analyzed for cases with valid data from both sources. Table 4 presents both percent agreement and Cohen’s Kappa statistics. For utility needs, the percent agreement was 90%, meaning that the raters agreed in 90% of cases, and the corresponding Kappa value of 0.61 indicates substantial agreement between the two sources. Based on the Kappa values, which adjust for chance agreement, Transportation Needs and Financial Strain showed moderate agreement (κ=0.48-0.49), while Food Insecurity and Substance Use showed fair agreement (κ=0.34-0.38). Housing Instability, Mental Health Needs, and Family/Community Needs also demonstrated fair agreement (κ=0.21-0.27), and Physical Inactivity showed slight agreement (κ=0.10). Notably, physical Inactivity had around 80 % percent agreement. However, its Kappa statistic was only 0.1. This means that 8 out of 10 responses matched; however, most of their agreement was likely due to chance, potentially due to the skewed responses and small sample size.

Table 4.

Concordance: Percent Agreement and Kappa Statistics

Domain	N	Match n (%)	Mismatch n (%)	Percent Agreement	Kappa Statistic	Kappa 95% CI	Kappa Agreement Category
Utility Needs	90	81 (90.0%)	9 (10.0%)	90.0%	0.61	(0.38, 0.84)	Substantial
Transportation Needs	117	98 (83.8%)	19 (16.2%)	83.8%	0.49	(0.30, 0.69)	Moderate
Financial Strain	108	80 (74.1%)	28 (25.9%)	74.1%	0.48	(0.32, 0.65)	Moderate
Food Insecurity	112	79 (70.5%)	33 (29.5%)	70.5%	0.38	(0.21, 0.55)	Fair
Substance Use	67	47 (70.1%)	20 (29.9%)	70.1%	0.34	(0.11, 0.57)	Fair
Housing Instability	80	52 (65.0%)	28 (35.0%)	65.0%	0.27	(0.06, 0.49)	Fair
Mental Health Needs	48	32 (66.7%)	16 (33.3%)	66.7%	0.22	(-0.07, 0.51)	Fair
Family/Community Needs	45	25 (55.6%)	20 (44.4%)	55.6%	0.21	(0.00, 0.42)	Fair
Physical Inactivity	45	35 (77.8%)	10 (22.2%)	77.8%	0.10	(-0.17, 0.37)	Slight

Note: n (%) indicates the number and percentage of N defined as the participants who had valid data from both the HRSN survey and EHR data; Match represents both sources indicating presence or absence of needs; Mismatch represents one resource indicating needs while the other did not; Percent agreement represents the number of matching cases divided by the total number of participants with valid data for each domain. CI: Confidence Interval.

Meanwhile, Mental Health Needs, Family/Community Needs, and Physical Inactivity had small sample sizes (N=45 to 48), which should be considered when interpreting these results. Figure 2 provides a visual summary of the concordance analysis across domains. The Kappa categories are color-coded by agreement level (green=substantial; orange=moderate; red=fair to slight), with 95% confidence intervals. This visualization shows that despite high percent agreement in some domains, Kappa values may indicate weaker agreement when accounting for chance agreement.

Figure 2.

Comparison of Percent Agreement and Kappa Statistics with 95% Confidence Intervals. *Multiple EHR fields were merged for this domain. CI: Confidence Intervals.

Discussion

This study compared the assessment of SDOH from a standardized survey to EHR-based documentation in patients with comorbid hypertension and uncontrolled diabetes who were participating in a structured clinical trial. We found notable discrepancies between the two sources, with certain domains (Utility Needs, Transportation Needs, and Financial Strain) having moderate to substantial agreement while others had fair or slight agreement. We also found significant amounts of missing data in certain EHR-based SDOH domains. These findings align with previous research that EHR SDOH data had high degrees of missingness and frequent mismatches when compared to other data sources, such as standardized social risk screening tools or clinical documentation.^{20, 21} These findings highlight the challenges in utilizing real-world EHR-based SDOH data for clinical practice and research.

Currently, there is substantial variation in SDOH documentation within EHR systems.^{8, 21, 22} Documentation methods range from structured approaches using demographic fields, billing codes, encounter codes, or structured SDOH screening questions to unstructured approaches that rely on free text documentation in clinical notes. Moreover, despite the availability of SDOH documentation capabilities, clinical implementation remains inconsistent. Studies indicate that SDOH information in EHR systems remains low and inconsistent.²³ On the other hand, implementing a standardized screening tool for social needs has shown significant clinical impact in healthcare settings. Studies reported that adopting standardized surveys as a screening for social needs and providing community resource referrals were associated with reducing emergency department visits and avoidable ED visits.²⁴

Comparability of SDOH data in Each Source. The main challenge in comparing survey and EHR data came from two key differences. First, the two sources had considerable discordance in how social needs were categorized. In some cases, EHR fields were combined (e.g., Tobacco Use and Alcohol Use under “Substance Use”) to allow comparison. At the same time, other HRSN survey domains (e.g., Safety Needs, Employment Needs, Education Needs, and Disabilities) had no direct EHR counterparts. Second, the two approaches used different methods to assess patient risk levels. The HRSN survey classified patients as needing at even minimal levels of concern, allowing for early detection and proactive intervention. The EHR used a different scoring approach, but its methodology for determining risk levels was not documented. These differences in categorization and risk assessment make streamlining data usage across diverse settings challenging.⁹ Although screening for social drivers of health is becoming more prevalent in hospitals and clinics, achieving consistent adoption and data standardization remains challenging.^{9, 25} Efforts to improve the comparability of SDOH data must tackle these inconsistencies while addressing concerns about over-screening.²⁸ To ensure sustainable social drivers of health screening in clinical practice, especially for high-risk populations with comorbid hypertension and uncontrolled diabetes, healthcare systems must balance the need for thorough and comparable data collection with a careful emphasis on minimizing patient burden.

Completeness of EHR-based SDOH Data. A notable finding was the data missingness in EHR-based SDOH domains. More than 70% of records were missing in the Mental Health Needs, Family/Community Needs, and Physical Inactivity domains. Similarly, over half of the records were missing in the Substance Use and Housing Instability domains. In contrast, the HRSN survey had consistently high completion rates, with less than six percent missing data in each domain. Of note, this was likely due to the highly structured nature of a clinical trial that included payment for survey completion. These domains with high missing rates require careful interpretation, as incomplete data can bias study results.¹³ In our study, the HRSN survey was administered privately. This may have encouraged more open disclosure from participants. On the other hand, lower completion rates in EHR-based data may stem from practical challenges in real-world, clinical settings. Clinicians often face time constraints during patient visits, follow different documentation routines, and may prioritize immediate clinical concerns over social needs documentation.^{8, 11} The incomplete documentation of certain social drivers could also be attributed to departmental differences and institutional priorities, resulting in data gaps. Patients may also hesitate to share sensitive information during clinical visits, particularly when they don’t have an established patient-provider relationship or don’t understand how their information will be used.²⁶ These findings highlight the need to develop feasible strategies that enhance SDOH data completeness within routine clinical practice and diverse healthcare delivery settings.

Accuracy of EHR-based SDOH Data. The concordance analysis between the two sources showed mixed results. Utility Needs, Transportation Needs, and Financial Strain showed promising alignment, with percent agreement above 74% and Kappa values ranging from 0.48 to 0.61 (moderate to substantial agreement). These domains might serve as reliable indicators of patient’s social needs in EHR data. However, there were notable discrepancies in other domains. Physical Inactivity showed a high percent agreement (78%) but very low Kappa (κ=0.38). Likewise, Food Insecurity had 71% percent agreement yet only fair Kappa (κ=0.38). These discrepancies between percent agreement and Kappa statistics may reflect uneven response distribution or high rates of missing data. It is also possible that EHR data might underrepresent certain risks. However, further research with a larger sample size is needed to understand these disagreements. Timing of documentation also affected accuracy. We found a median gap of 90 to 120 days between HRSN survey completion and the nearest EHR entry, with some records appearing to be substantially outdated. This suggests that SDOH information can fluctuate within short periods and may become outdated between the assessments in clinical practice. Since social needs can change over time, infrequent assessments may not capture patient’s current social needs.⁴ More frequent or routine evaluations of social needs are recommended to improve the accuracy of EHR-based documentation. Accurate documentation of social needs has important implications for healthcare delivery. Unaddressed social needs often lead to poorer clinical outcomes and higher healthcare expenditures.²⁴ Systematic SDOH documentation in EHRs could help organizations better direct limited resources, such as social work or care coordination services, to the patients who need them most.

Social Needs vs Risks. Another key insight was that EHR documentation could capture ‘risk factors’ rather than patients’ subjective needs in certain domains. Health records generally focus on risk-oriented and demographic information rather than patients’ self-identified needs.^{25, 26} While risk assessment data can be clinically valuable, this approach has potential limitations. First, although patients express their need for assistance, this information may not always be recorded appropriately in their EHR. This could potentially lead to missed opportunities for timely support. Additionally, risk-based documentation alone might overestimate how many individuals need active intervention, leading to unwanted referrals for patients who don’t desire assistance.

Limitations. The results may not represent system-wide adoption since this study was conducted within primary care and endocrine clinics with a single academic health system. This restricts the generalizability to other settings or populations. Our analysis only included structured EHR data fields that utilized SDOH screening tools integrated into the EHR. This may have restricted the true extent of SDOH documentation in clinical practice since much of the information regarding social needs might have been recorded in free-text clinical notes. Additionally, we recoded EHR data into binary outcomes and combined multiple EHR fields into single domains. However, this may have concealed more nuanced information. We did not apply imputation for missing data, aiming to accurately capture the current state of EHR documentation, which may have reduced our sample size. Lastly, we limited EHR data to within ±12 months of each participant’s HRSN survey to avoid using outdated records, which could affect data non-concordance due to a lack of temporal synchronization. Additionally, since the AHC HRSN SDOH survey was introduced into the EXTEND trial after enrollment, not all participants completed the survey.

Future work. Future work should focus on three key areas. First, since social needs can change over time, more frequent assessments are needed to capture their dynamic nature. Second, future studies should examine how documented social needs translate into actual support. The current study did not track whether high-risk patients received appropriate referrals or interventions. Lastly, healthcare systems should further develop sustainable methods for comprehensively capturing data on social drivers of health.

Conclusion

By comparing standardized HRSN survey data to EHR-based SDOH documentation, this study highlights both the promise and challenges of using EHR data for identifying social needs in a high-risk population with hypertension and uncontrolled diabetes. While certain domains of EHR data appear well-aligned with survey findings, others reflect significant gaps between what is documented in the EHR and the patient’s reported needs. These findings underscore the importance of enhancing comparability, completeness, and accuracy of EHR-based SDOH documentation. Such efforts could improve clinical care, reduce health disparities, and foster more effective interventions for patients.