We conducted a mixed methods cohort study at Monroe Carell Jr. Children’s Hospital at Vanderbilt, a single, freestanding, tertiary care children’s hospital from February 2023 to February 2024. English- and Spanish-speaking caregivers of children 18 years or younger admitted to a pediatric hospital medicine team with a positive inpatient FI screen on the 2-question Hunger Vital Sign screener (answer of “Sometimes“ or “Often” to either question) were eligible for study inclusion. Only the first encounter was included for patients with multiple admissions during the study period. We aimed to include a cohort that was representative of the patients with positive FI screens admitted to our hospital, thereby supporting our goal to understand the feasibility of our automated, pragmatic approach.

This study took place at an academic tertiary care pediatric hospital located in the Southeast region of the United States. The hospital’s patient population included children from urban, suburban, and rural areas across 9 states. In 2023, over half (54%) of admitted patients were insured by Medicaid. English (87.6%), Spanish (9.1%), and Arabic (1.4%) were the 3 most common preferred languages. The prevalence of FI in the state in 2023 was 11.7% [29].

Inpatient FI screening was implemented as part of routine clinical care at the hospital in 2021. Screening was iteratively expanded and adapted using Quality Improvement methodology to reach the goal of sustainably screening 80% of eligible families. The full details of the hospital’s screening process and associated outcomes are detailed in a previous publication [30]. At the time of this study, families admitted to one unit of the hospital were routinely screened for FI with the validated 2-question Hunger Vital sign [31,32]. Screening occurred at the time of nursing admission intake; participant responses were entered into the electronic health record (EHR). Caregiver contact information was also reviewed as part of the nursing admission process. All families with documented FI were provided information at discharge in their preferred language that included a food resource handout containing general information on federal support programs (eg, Supplemental Nutrition Assistance Program [SNAP]) and links to ZIP code–based searches for emergency community food resources. The discharging nurse was responsible for providing and reviewing this information at the time of hospital discharge [30]. The Text Connect intervention was added to this existing workflow for the duration of the study. All other aspects of the workflow were unchanged.

To evaluate the feasibility of the Text Connect process and acceptability of the intervention, we used (1) primary measures of system-reported reach and engagement among the cohort of all caregivers included in the study (henceforth “included cohort”); and (2) secondary measures of caregiver-reported reach and engagement among a subset of caregivers who responded to text messages (henceforth “text message respondents”). We also assessed usability and acceptability through semistructured interviews with a subset of English- and Spanish-speaking caregivers.

We defined the primary, system-reported measure of the intervention’s reach as the proportion of the included cohort who received the Text Connect message based on Twilio message delivery data. Twilio reports the status of each message (ie, sent, delivered, undeliverable) and generates error notifications for any undeliverable message (eg, number is not in service, landline that cannot receive text messages, etc). Among text message respondents, a secondary, caregiver-reported measure of intervention reach was defined by caregiver report of receiving the Text Connect messages.

We defined the primary, system-reported measure of intervention engagement as the proportion of included caregivers that either clicked the FindHelp link or responded to at least one text message. This combined measure was selected to be inclusive of caregivers who engaged with the Text Connect system by responding via text message but who may have been unable or chose not to use the embedded FindHelp link. Among text message respondents, we defined secondary measures of caregiver-reported engagement as (1) the proportion of caregivers who reported reading the text message; and (2) the proportion of caregivers who used the information to search for food resources.

We assessed resource connection as a preliminary measure of effectiveness. Among text message respondents who reported using the included information to look for a resource, we defined resource connection as the proportion of caregivers that reported getting a new food resource since their child’s hospitalization.

Interview participants completed the validated, 9-question Simplified System Usability Scale (SUS) to quantify intervention usability. The SUS is scaled from 0 to 100, with scores >75 indicating good usability [41]. Open-ended follow-up questions were asked to characterize identified usability issues. Acceptability was assessed among interview participants based on caregiver agreement with the statement, “The text message is an acceptable way to get food resource information.”

We used descriptive statistics to summarize primary and secondary measures as well as clinical and sociodemographic characteristics of the included cohort, those receiving Text Connect based on the system-reported measure, and those responding to follow-up messages or clicking the FindHelp link. We compared demographic variables between text message respondents and nonresponders using the Wilcoxon rank sum test for continuous variables and the Pearson chi-square test for categorical variables. Usability scores were scaled to 100 for comparison against established benchmarks and summarized for the interviewed cohort. For qualitative analyses, we used rapid thematic analysis to (1) explore caregiver preferences for receiving food resource information through text message compared to typical methods of paper handouts and meeting with a person in the hospital (ie, a resource navigator); and (2) identify usability issues and barriers to engagement with the intervention. Primary themes were informed by quantitative data and inductively assigned by PI (CQF) and reviewed by LSM and DW. Themes were compared between English- and Spanish-speaking caregivers. We used triangulation between quantitative outcomes and themes that emerged between English and Spanish caregivers to ensure the trustworthiness of qualitative analyses. Quantitative analyses were conducted using R statistical software.

This study posed no greater than minimal risk to participants and was deemed exempt from review by the Vanderbilt institutional review board (#222183) based on analysis of deidentified data. Informed consent was waived due to the low-risk nature of the study and approval under “Exempt” status by the institutional review board. Privacy and confidentiality of participants were maintained through data storage on password-protected servers and analysis of deidentified data. All interview participants received a US $30 gift card as compensation for their time.

During the study period, 1670 patients on pediatric hospital medicine teams were screened for FI. Of those screened, 194 (194/1670, 12%) had a positive screen for FI, and 187 (187/194, 96%) of these preferred English or Spanish and were included in the study cohort (Figure 3). Most participants reported non-White race (107/187, 60%), were publicly insured (165/187, 90%), preferred English (133/187, 71%), and lived in a nonrural ZIP code (137/187, 75%) (Table 1).

Three types of system errors were identified and addressed during the study: (1) failure to import data for certain demographic and clinical variables into the report (10 participants); (2) failure to automatically import data into the REDCap system due to a communication breakdown between the EHR and REDCap following an EHR system update (9 participants); and (3) failure to trigger the initial text due to a REDCap connection error (15 participants). After these system errors were identified during regular quality checks and resolved, the text message series was manually triggered for all affected participants. The affected 24 participants with a phone number in REDCap were subsequently able to participate in other study activities.

Among the included cohort (n=187), 175 (94%) were reached by the intervention. Twelve (6%) participants were not reached due to (1) phone number did not import into REDCap (n=10, 83%); or (2) an undeliverable message (n=2, 17%) (Figure 3). The reached cohort had similar distributions of demographic and clinical characteristics as the full cohort (Table 1). System-collected measures indicated that 102 (102/175, 58%) of the reached participants engaged with the text message exchange in some way, with 65 (65/175, 37%) engaging by clicking the FindHelp link and 94 (94/175, 53%) responding to at least one message. 54 (54/175, 31%) participants did both. (Figure 3).

Ninety-two (92/175, 53%) of the reached caregivers responded to at least one text message. There was a higher proportion of Hispanic caregivers (40% [35/92] vs 23% [21/95], P=.03) and caregivers who preferred Spanish (38% [35/92] vs 20% [19/95], P=.01) among text message respondents compared to nonrespondents. Eighty-eight (88/92, 96%) of text message respondents reported receiving the text message with food resources, 83 (83/92, 90%) reported reading the text message, and 42 (42/92, 46%) reported using the information to look for a new resource (Figure 3).

Questions were asked in a sequential, nested fashion such that only caregivers who reported using the included information were asked about connection to a new resource. Among the 42 text message respondents asked about connection to a food resource, 7 of the 11 participants (64%) with a response to the relevant question reported getting a new food resource.

We interviewed 20 caregivers who preferred English and 11 who preferred Spanish. Most caregivers were female (28/31, 90%), the patient’s mother (28/31, 90%), and non-White (18/31, 60%); 13/31 (42%) reported Hispanic ethnicity. The median caregiver age was 29.5 years (23.3‐38.0) (Table 2).

Among interview participants, the median SUS score was 86.1 (66.7‐91.7), indicating excellent usability. In looking at individual items assessed within the SUS, 81% (25/31) felt confident using the text message to look for food resources, and 94% (29/31) would use the information in the text message again. In total, 97% of participants agreed or strongly agreed that SMS text messaging is an acceptable way to receive food resource information. The lowest usability item was whether the text message was “easy to use” (24/30, 80% of respondents agreed or strongly agreed). Issues with usability and acceptability of the intervention included (1) technical issues such as the text or website being in the wrong language or website being difficult to navigate; (2) preference to discuss available resources with a person; and (3) distrust in the information in the text message (eg, wary of scams).

Nearly all caregivers reported a preference for receiving information through text message compared to a paper handout. Reasons included that (1) handouts are easily lost among many papers received at discharge; (2) text messages can be searched directly on a personal phone and are accessible at any time; and (3) text messages felt more personalized. Preferences were mixed when asked about resource connections provided by a resource navigator. Perceived benefits of a navigator included feeling like participants would receive more personalized information and real-time assistance with questions; perceived benefits of a text message included being more private and more convenient when juggling caregiving responsibilities. Illustrative quotations for these themes are included in Tables 3 and 4. Each row in the table represents responses from an individual caregiver.

Our team developed and piloted Text Connect, a novel, automated mHealth intervention to provide ZIP code–specific food resource information to English- and Spanish-speaking families with FI identified during a hospital admission. The Text Connect intervention reached 94% of families with FI, with over half of recipients engaging with the text messages despite no prior communication from the hospital that the text would be sent to them. Caregiver-reported usability and acceptability of the intervention were high, and 64% of text message respondents reported getting a new food resource due to information contained in the text message. Interview participants preferred Text Connect over paper handouts due to the message accessibility and tailoring.

These results are especially relevant in the context of recent attention to social and medical care integration, where hospital systems across the country are actively implementing social risk screening in response to federal mandates from the Centers for Medicare and Medicaid Services and The Joint Commission [42,43]. Though addressing disclosed social risk information is a key component of many national recommendations, evidence to guide best practices for responding to positive FI screens is lacking [44]. As a result, existing approaches are heterogeneous, many have not been evaluated, and many are person-time intensive or rely on a static, standardized paper handout that is not tailored to patient needs or location [19]. Online referral platforms are increasingly popular and may overcome these limitations, but they also can require substantial investment from the health system and community partners for purchase, implementation, referral, and outcome reporting.

Our study demonstrates that an automated, EHR-generated text message with location-specific resource information offers a promising alternative to a paper resource handout as an initial, low-intensity strategy for providing tailored information about local food assistance programs. Over half of the reached cohort responded to text messages and one-third used the ZIP code–specific FindHelp community food resource link. This aligns with prior work showing that not all families with FI identified through screening desire assistance [45-47] and that patients are more likely to respond to text messages than click on embedded links [48]. Despite no preceding in-hospital interaction about FI resources, our text message response rate was similar to a pragmatic, automated, postdischarge SMS text messaging program that first contacted participants during admission, in which 55% of included patients responded to at least the first text message [49]. Applying caregiver interview data to the TAM-RLS framework, we hypothesize that this preference for text message use is due to (1) a need to engage through mechanisms that require only basic cellular service; and (2) greater trust in the confidentiality of responding to a text message as compared to clicking on an unfamiliar link. These findings suggest that future effective interventions that similarly provide tailored resources links to patients are likely to benefit from avoiding dependence on cellular data or internet and allowing two-way messaging interactions between patients and their care teams.

Though this study was not designed or powered to assess the effectiveness of the Text Connect intervention (eg, connection to a new food resource or subsequent health outcomes), text message respondents and interview participants reported getting a new food resource due to the information in the text message, suggesting that SMS text messaging can help a proportion of caregivers access food resources. By providing additional points of contact with families postdischarge that convey continued investment in their family’s well-being, the Text Connect intervention may also positively impact family and population health outcomes by fostering deeper trust in and connection between health systems and families they serve [50]. Caregiver interviews suggest that augmenting Text Connect by providing an option for caregivers to receive additional assistance through a follow-up phone call with a social worker or resource navigator may increase families’ food resource connections. This individualized assistance may also encourage connection to unique programs that exist in a caregiver’s community, such as Food is Medicine initiatives or other community partnerships. Improving access to nutritious foods for all children is critical to improving physical, mental, and developmental health outcomes. Thus, additional studies should rigorously compare the effectiveness of different methods of responding to positive FI screens, particularly attending to the effectiveness of different approaches across patient sub-populations.

Our findings should be interpreted in light of several key limitations. First, although automated EHR data collection and intervention related to food insecurity is feasible, low cost, and scalable, systematic monitoring is required to ensure system components are functioning as intended. We identified several Text Connect system issues in real time that resulted in occasional delays, pauses, and required manual import of records. However, the end user was unaware of these technical issues, so we expect minimal impact on their perception of acceptability or usability. The use of pragmatic data collection measures among the full cohort via Twilio was also a strength in that it lowered risk of response bias and facilitated more complete data collection. Second, though we saw a 50% text message response rate, we found a notable drop-off in completed responses as participants progressed through text message questions, limiting sample size and ability to draw conclusions about intervention effectiveness in facilitating connection to food resources. Additionally, though the screener used has shown 97% sensitivity for identifying families with FI [51], we likely missed a small fraction of caregivers with FI.

Finally, this study was conducted among patients on one unit and one service at a tertiary care, free-standing children’s hospital, and our results may not generalize to all caregivers with FI or other health systems. Future studies should test the feasibility of this intervention in other settings.

Based on results of this study, the Text Connect intervention was approved by our health system as a method for providing food resource information to any family with FI identified during admission to the Children’s Hospital. We adapted message content in response to input from caregivers and are currently testing Text Connect’s effectiveness in a pragmatic randomized controlled trial (ClinicalTrials.gov; NCT06919445). Data from the next trial will inform cocreation of a stepped-care approach to resource connections that incorporates resource navigator follow-up for caregivers who are not engaged or assisted via text message alone.

An automated text message is a feasible, usable, and acceptable way to provide families with FI identified during hospital admission with relevant food resource information. The Text Connect intervention is a system-level approach that was built with input from community members and can engage caregivers with varying levels of technology access. This approach is widely generalizable and could be adapted and implemented in other health systems that use modern EHR systems and have the resources to generate automated text messages to provide tailored content for FI and other health-related social needs (eg, housing insecurity).