Accuracy of a Single Financial Security Question to Screen for Social Needs Free

Children's Community Pediatrics has 45 community pediatric primary care offices that span 6 urban counties and 10 rural counties and shares an electronic health record with 1 academic pediatric primary care practice. Practice size ranges from 2 to 40 clinicians and 1200 to 11 000 patients.

Routine electronic health record (EpicCare Verona, Wisconsin)-based screening for social needs began in August 2021 for all children <11 years of age who were presenting for a well-child visit. Screening could be completed in English by a parent/guardian on the patient portal, by tablet in the waiting room, or verbally with staff. Parents/guardians who preferred a language other than English were asked questions verbally with the use of an interpreter. Responses were immediately available in the medical record. We used data from children <11 years of age assigned the questionnaire between January 1 and December 31, 2022.

The questions used in the social needs screener are the standardized questions within EpicCare and they are always administered in the same order, with the financial needs question occurring first. The questions included in the EpicCare module were validated items, derived from a 2014 Institute of Medicine (now National Academy of Medicine) report, “Capturing Social and Behavioral Domains and Measures in Electronic Health Records.”¹⁰  The overall financial stress question, “How hard is it for you to pay for the very basics like food, housing, medical care, and heating?” has 5 response options (not hard at all, not very hard, somewhat hard, hard, very hard).¹¹  The item on financial stress used was recommended in the Institute of Medicine report, for inclusion in electronic health record screening modules, and subsequently incorporated into EpicCare after this recommendation.¹⁰  Screening also included questions about food,¹²  housing,¹³  and transportation.¹⁴  We created binary categorizations of food insecurity, housing instability, and transportation needs (Supplemental Information).

Demographic variables were extracted from the electronic health record including the child’s age, sex, race, ethnicity, and insurance provider. Race and ethnicity were collected from 2 demographic fields within the child’s electronic health record, self-identified by the parent/guardian. Race and ethnicity are collected through clinical and administrative personnel at the time of the patient’s first encounter and updated at parent/guardian’s request. Race and ethnicity, both social constructs, are included because of the potential for experiences of racism to affect both the risk for experiencing health-related social needs and acceptability of disclosing social needs to the health care system.^15,16  Race and ethnicity are reported separately (eg, “white” includes both “non-Hispanic white” and “Hispanic white”). Race and ethnicity categories are reported if they have a frequency of 2% or more in the overall sample; categories with frequencies <2% are combined into “another race or multiple races.” Insurance providers were categorized as commercial, Medicaid, other governmental, or self-pay/missing.

We report the frequency of demographic variables and reported social needs in the entire sample and by age group (0–1, 2–5, and 6–10 years old). We also report these variables stratified by financial stress. We report the area under the receiver operating curve (AUROC) for the 5 response options of the financial stress question, as well as sensitivity, specificity, positive predictive value, and negative predictive value of the financial stress question (not hard at all versus any other response; not hard at all and not very hard versus any other response) to detect other reported social needs. These analyses are performed for the entire sample and stratified by age group. We chose to use age groups for subgroup analyses because the costs of raising children vary as they age,¹⁷  and thus the relationships with the outcomes of interest may also vary by age. Also, some community and governmental resources, such as the Special Supplemental Nutrition Program for Women, Infants, and Children, are only available to families with children in certain age groups (<5 years of age).¹⁸  The age groups correspond roughly to infancy, preschool age, and elementary school age in our sample.

The analysis was performed under the Univeristy of Pittsburgh Medical Center's Quality Improvement Committee Project 3373. The Quality Improvement Committee approves projects that do not meet the definition of human subjects research and therefore are reviewed by this committee instead of an institutional review board. Survey responses were collected during usual clinical care; participant consent was not required for our retrospective analyses of these deidentified clinical data.

Of 137 261 eligible children, 127 753 (93.1%) had social needs data collected at their first eligible visit and 130 414 (95.0%) had data collected within at least 1 visit during 2022. Overall and by age group, demographic data and reported social needs are included in Table 1. In the entire sample, the reported racial categories were 79.8% white, 11.8% Black, and 8.4% another racial identity (including Alaska Native, American Indian, Chinese, Filipino, Guam/Chamorro, Hawaiian, Indian [Asian], Japanese, Korean, other Asian, other Pacific Islander, Samoan, and Vietnamese). In the overall sample, 48.7% of the children were female, 2.0% identified as Hispanic, and the primary insurance category was commercial (61.0%). As shown in Table 2, within each age group, reporting financial stress was associated with higher rates of other social needs (eg, in the entire sample, 1.4% of those without financial stress report food insecurity compared with 29.2% of those with financial stress [P ≤ .001]). Children of parents reporting financial stress were more likely to identify as Hispanic ethnicity (P ≤ .001), nonwhite race (P ≤ .001), and be insured through Medicaid (P ≤ .001). Likewise, children with missing social needs data were more likely to identify as Hispanic ethnicity (P ≤ .001), nonwhite race (P ≤ .001), and be insured through Medicaid (P ≤ .001) (Table 3).

Reports of social needs were common (Table 1). In the overall sample, 9.8% of families reported food insecurity, 9.3% reported housing instability, 4.2% reported transportation needs, and 16.8% reported any of these needs. Overall, 30.8% of families answered the financial stress question with a response other than not hard at all and 10.6% of families answered the financial stress question with responses other than not hard at all or not very hard.

The AUROC, sensitivity, specificity, positive predictive value, and negative predictive value of the financial stress question’s ability to detect other social needs are presented in Table 4 and the AUROC are illustrated in Fig 1. The AUROCs ranged from 0.73 to 0.89. When using not hard at all versus all other responses to the financial stress question, sensitivity ranged from 0.712 to 0.788, specificity ranged from 0.671 to 0.903, positive predictive values ranged from 0.942 to 0.986, and negative predictive values ranged from 0.092 to 0.414. In this sample, using the financial stress question as the first step of a 2-step screening process with not hard at all versus all other responses would have stopped screening for 69% of respondents. This would have missed 9.7% of the families who reported food insecurity, 22.6% who reported housing instability, and 33.0% who reported transportation needs.

When using not hard at all or not very hard versus all other responses to the financial stress question, sensitivity improved with a range from 0.911 to 0.969, specificity worsened with a range from 0.447 to 0.623, positive predictive values ranged from 0.902 to 0.974, and negative predictive values ranged from 0.180 to 0.751. In this sample, using the financial stress question as the first step of a 2-step screening process with not hard at all or not very hard would have stopped screening for 90% of respondents. This would have missed 37.7% of the families who reported food insecurity, 50.1% who reported housing instability, and 55.3% who reported transportation needs. The age-stratified AUROC found similar results as the overall AUROC (Supplemental Figs 2–4).

In a large sample of outpatient pediatric practices, we found that a single question about financial stress had inadequate sensitivity to be used as an initial screening question for food insecurity, housing instability, or transportation needs. Ideally, 2-step screening would employ a highly sensitive test, followed by a highly specific test,⁸  to minimize respondent burden without missing families in need. We found that the financial stress question used in our electronic health record, and standardized in a widely used electronic health record, does not function well as a first-step item.

The American Academy of Pediatrics was the first national medical society to endorse health-related social needs screening.¹  Other national health care organizations have similarly recommended that clinicians screen for and assist patients with unmet social needs to improve patient care and reduce health disparities.^19–21  Additionally, national organizations, such as the Centers for Medicare and Medicaid and the Joint Commission, are in the process of setting new quality measures requiring health-related social needs screening.^22–24  Despite the growing interest and investment in integrating social needs screening in clinical care settings,²⁵  barriers (eg, time, family burden) exist, and it is still unclear how to most effectively implement social needs screening in busy pediatric health care settings.²⁶  The results of this study are consistent with and expand previous work related to health-related social needs screening in pediatric care settings. Although this is the first study to evaluate using a question regarding financial stress as a prescreen, 1 study evaluated using the 2-item Hunger Vital Sign as prescreen for other social needs.²⁷  Similar to our results, the authors found that utilizing the Hunger Vital Sign as a prescreen would potentially miss a large portion of families (>20%) who endorsed other health-related social needs. Further research is needed to identify alternative strategies to optimize the balance efficiency and effectiveness of social needs screening in busy clinical settings in a family-centered approach.

The findings of this study have important implications and suggest directions for future study. Perhaps the most important implication is that using a single-item financial strain prescreen would miss many families experiencing health-related social needs. This does not mean, however, that a 2-step workflow is infeasible. Instead, future studies should investigate whether other items might have acceptable sensitivity to be used in a 2-step work flow. This work could be important for reducing the burden of screening in clinics and on families. Another interesting research question we were not able to evaluate in this study (because of the way the data are presented and stored within the electronic health record) is whether the order of questions or the mode of question administration (through the patient portal, on tablets, or verbally by clinicians or staff) affects the sensitivity of the financial strain item. Previous work on this topic has suggested that patients and caregivers often have preferences about mode of administration and may be less likely to disclose unmet social needs if asked verbally,^28–30  but it is unclear whether different modes lead to a meaningfully different relationship between responses.^2,31–33  We think this would be an important area for future investigation, because it is ideal to support modes that are concordant with patient and caregiver preferences, and elicit the most accurate information. It is important to note that, although the children with missing social needs data were relatively small, they were more likely to identify as Hispanic ethnicity, nonwhite race, and be insured through Medicaid. Insuring that social needs screening is acceptable to all patients should be a priority.

The findings of this study should be interpreted in light of several limitations. This study was conducted in a single health system in a single state and single language of administration. However, despite this, the needs identified in this sample are similar to nationally reported rates.³⁴  Second, the data used in this analysis were from 2022, toward the end of the coronavirus disease 2019 pandemic, a time when access to support for social needs was changing rapidly. Despite this volatility, we see no reason why these changes in the circumstances that create needs or the services available to address needs would affect the interrelationship between needs, which was the focus of this study. Thus, we would expect that the associations seen here would be generalizable to other contexts, although that could be confirmed empirically. Third, on the basis of how the data are stored in the electronic health record, we are unable to assess how the social needs questions were presented to parents/guardians or what parents/guardians perceived the intent of these questions to be. Parents/guardians may have been less likely to report social needs if they had concerns about who would have access to the results and whether reporting social needs could be used in a punitive way (eg, notifying child protective services).^35,36 

These limitations were balanced by several strengths. This study used a large sample of parents and guardians who presented to a set of outpatient pediatric clinics that span a variety of built environments (eg, urban, rural, suburban) and economic environments (eg, area deprivation index scores in the catchment area range from the fourth to 100th national percentile).³⁷  Further, completion of screening was high (95% of children presenting for a well-child visit during the year), suggesting that the results are representative of the patients seen in these clinics.

Developing efficient and effective evidence-based, health-related social needs screening strategies could improve implementation in clinical practice. Although 2-step screening for health-related social needs may offer benefits, we found that this specific 2-step screening strategy had inadequate performance. At this time, we recommend clinicians use multidomain screening tools to ensure they are able to detect social needs within their clinical population, ideally selected by the institution’s ability to address any identified needs.

AUROC

area under the receiver operating curve