An Expert-Derived Data-Driven Approach to Identifying Medicaid Children at Risk of Out-of-Home Placement Authors: Rose Y. Hardy, David P. Ciccone, Emelie Bailey, Samantha Bozada, Erin Donnelly, Krystel Tossone, Deena J. Chisolm Published: 2026 ABSTRACT Background:Out-of-home placement (OOHP) through child welfare, residential substance treatment, or extended inpatient care has life-altering impacts on child health and generates an outsized proportion of pediatric Medicaid costs. Early risk identification may prevent OOHP but is challenged by the difficulty of navigating state and federal data sharing regulations and multiple disconnected data systems. Identifying ways to streamline and simplify this process are critical to OOHP prevention.Objectives:To describe the development and implementation of an expert-derived data-driven algorithm to identify OOHP risk among Medicaid-enrolled children residing in 2 Ohio counties between 2022 and 2024 (n=27,000), discuss practical considerations for data sharing and linkage across multiple agencies, and identify lessons learned.Findings:A cross-sector team of government and academic partners developed a 3-category OOHP risk stratification algorithm incorporating data from Medicaid claims, child welfare information systems, area-level social determinants of health, and patient-reported health risk assessments as part of Ohio’s Integrated Care for Kids Model. Substantial administrative and legal processes were required to link data sources. Lessons learned include: (1) understanding legal, political, and security structures associated with use of administrative data is critical; (2) strong working relationships with data partners can ensure success; (3) as administrative data are dynamic and may be retroactively updated, appropriate lookback periods are necessary for accuracy; and (4) while executing data use agreements can be challenging, the relationships built can have lasting benefits.Conclusions:Data-driven risk stratification models have the potential to reduce cost, time, and redundancies in identifying families who would benefit from OOHP prevention supports. Balancing data-sharing challenges with the value gained from linking data is vital. Medicaid is the largest insurer of children in the US. In 2022, Medicaid covered 39% of all US children and children accounted for ∼$97.9 billion in Medicaid expenditures.1,2 As a federal and state partnership, Medicaid is one of the largest single line items in many state budgets. Compared with children covered by private insurance, children covered by Medicaid are more likely to have special health care needs, health-related social needs, and behavioral health concerns.3 Providing high-quality, need-appropriate, cost-efficient care to this medically and socially complex population is a goal of every state Medicaid program; however, systemically fragmented individual-level data make early risk identification and mitigation a challenge. One example is the intersection of Medicaid and Child Welfare. The child welfare system works to ensure all children are in safe and supportive family environments.4 In 2020, child welfare agencies spent $31.4 billion in federal, state, local, or other funds.5 An average of over 600,000 children are served each year.6 Although placement outside the home may be necessary, particularly to ensure child safety, it can have negative effects on involved children, including worse mental health,7 lower educational attainment,8 chronic diseases like cardiovascular disease9 and chronic obstructive pulmonary disease,10 and mortality.9 The reasons for removal from the home are multifaceted. While abuse accounts for about 16% of cases,6 other reasons for removal include parental incarceration, family substance abuse, housing challenges, and severe physical or behavioral health issues of the child.4,6 Many of these challenges can be identified using existing data sources, but for supports to reach families in time, the systems caring for these children need to use such data to identify needs and risks earlier. This task is challenged by administratively disparate organizations with limited data-sharing capacities due to data-sharing regulations and independent information systems. Medicaid has a vested interest in early identification of children at risk of out-of-home placement (OOHP) whether through child welfare, as discussed above, or through mechanisms including residential substance use treatment, juvenile justice involvement, or extended inpatient care. In 2020, ∼$1 billion in federal Medicaid dollars went to child welfare agencies alone, costs that have been increasing over time.11 Although Medicaid data can serve as the backbone of models informing early intervention and primary prevention to maximize access to needed services, gaps remain in existing risk identification models. Multisector administrative databases have been identified as valuable tools to address child welfare by creating more comprehensive information to inform intervention.12 This paper presents an example of using multiple linked data sources to recognize risk, discusses the practical considerations required to link such data, describes patient-level and system-level benefits to implementation of such a model, and identifies lessons learned during this process. INTEGRATED CARE FOR KIDS (INCK) MODEL The Centers for Medicare and Medicaid Services (CMS) InCK Model is “a child-centered local service delivery and state payment model that aims to reduce expenditures and improve the quality of care for children under 21 covered by Medicaid through prevention, early identification, and treatment of behavioral and physical health needs.”13 Children with identified needs, specifically higher levels of OOHP risk, are engaged in integrated care coordination and case management to address those needs (Fig. 1). In Ohio, the InCK model includes Medicaid-enrolled children, aged 0–20, in 2 counties. These counties are primarily rural, with one formally designated as Appalachian. These counties were selected based on data demonstrating high rates of children’s services engagement and a high need for expanded access to behavioral health services. The risk tiering approach described here was developed within Ohio’s implementation of the InCK Model. Nationwide Children’s Hospital (NCH) is the lead organization for this work and has implemented the model in partnership with the Ohio State University Government Resource Center (GRC) and the Ohio Department of Medicaid (ODM). In response to a call for papers on the use of data linkages to promote health equity, our multidisciplinary team convened our internal and external data team, our operational team, and our community partners to discuss the challenges we faced, and in some cases overcame, in applying data to identify and address risks before an OOHP exists. Risk stratification components of the InCK model. We structured this series of conversations using an adaptation of the “Five Whys” method for root cause analysis, in this case, the “Five Hows.” This method is used to develop more details of a solution to a problem under consideration. Each “How” drives deeper understanding of the underlying process, problems, and solutions. Our driving question, question #1, is: How can we reduce unnecessary OOHP stays in Medicaid enrollees in our target region? Each subsequent question expands on how we approached this challenge. Question #2 – How can we identify those at-risk in the community? Answer – Use Existing Evidence and Data Sources. While the most direct and arguably most valid approach to assessing risk in a community would be to survey everyone, interviewing families of over 40,000 children one or more times each year was not operationally feasible due to a combination of resource limitations and low-quality contact information for enrollees. Our project sought to reduce the burden of direct data collection by leveraging administrative data from the state Medicaid program and other state agencies serving children enrolled in Medicaid and by enriching these data with publicly available geocoded data representing social drivers of health (SDH). CMS specified the framework for the escalating model-wide OOHP risk categories and the Ohio InCK team customized this framework applying published evidence and input from clinicians, data scientists, epidemiologists, and social service providers to develop an algorithm that considered behavioral health need, physical health concerns, food insecurity, housing instability, functional symptoms and impairments, and child welfare involvement using available Ohio data. Other variables, including parental health, family functioning, and school engagement, were considered but necessary data partnerships or linkages could not be implemented (see question #3). Per the model, OOHP risk was categorized into 3 progressive, mutually exclusive categories: Service Integration Level (SIL) 1 – low risk, SIL 2 – risk of medical/behavioral health need, or SIL 3 – imminent risk of OOHP. A full listing of the data sources and measures can be found in Table 1. In 2025, there were ∼27,000 Medicaid beneficiaries residing in these 2 counties, with ∼3325 stratified to SIL 2, ∼722 to SIL 3, and the rest to SIL 1. TABLE 1Identified Need Domains in OOHP AlgorithmIdentified need domainsData sourceData measureHealth and quality of life risk indicators Behavioral healthMedicaid claimsDiagnosis code value sets flag beneficiary needs associated with behavioral health and severe behavioral health needs. This includes a comprehensive behavioral health value set and a serious mental illness (SMI Dx) value set Physical healthMedicaid claimsPediatric Medical Complexity Algorithm (PCMA), an open-source algorithm that identifies clinically complex chronic conditions. Children with more than one body system involved, one or more progressive conditions, or one or more malignant conditions are classified as having a physical health need Functional symptomsMedicaid claimsAmong children, severe behavioral health diagnoses, serious mental illness (SMI Dx), neonatal abstinence syndrome (NOWS Dx), and substance use-related diagnoses (SUD Dx), excluding tobacco use disorder. A parental substance use flag links adults with substance use-related diagnoses (Parent BH Dx) who share a familial connection or shared household with the child Functional impairmentsPhysical, emotional, social, and school functional impairments: Pediatric Quality of Life (PedsQL) Inventory Physical and social impairments: Short Form 12 (SF-12) Early Childhood Development functional impairments: Survey of Wellbeing of Young Children (SWBYC) Self-care and daily activities functional impairments: Ohio Medicaid Health Risk AssessmentPhysical, emotional, social, and school functional impairments. Administered by NCH service integration coordinators (ie, registered nurses and social workers) at the time of enrollment and on a bi-yearly (SIL 3) or yearly (SIL 3) schedule. Early childhood development is administered by NCH service integration coordinators (ie, registered nurses and social workers) at the time of enrollment and on a bi-yearly (SIL 3) or yearly (SIL 3) schedule. Administered by NCH service integration coordinators (ie, registered nurses and social workers) at the time of enrollment and on a bi-yearly (SIL 3) or yearly (SIL 3) schedule. The Ohio Medicaid Health Risk Assessment screener question, “If for any reason you need help with day-to-day activities such as bathing, etc., do you get the help you need?” and Ohio InCK assigns a self-care need if the respondent answers “I could use a little more help” or “I need a lot more help”Social risk indicators Housing instabilityMeasure 1: Medicaid enrollee address files Measure 2: Ohio Medicaid Health Risk AssessmentMeasure 1: Ohio InCK uses geocoded Ohio Medicaid street address files to assign a housing instability flag to any eligible individual who has experienced 2 or more address changes in the past 12 mo, has a nonresidential address including being unhoused, or a Medicaid eligibility linked to a nonpermanent living arrangement (eg, shelter, domestic violence shelter, etc.). Measure 2: The Ohio Medicaid Health Risk Assessment screener question “Describe your current living situation” and Ohio InCK assigns a housing instability need if the answer is “I have a place to live today, but I am worried about losing it in the future” or “I do not have a steady place to live (I am temporarily staying with others, in a hotel, in a shelter, living outside on the street, on a beach, in a car, abandoned building, bus or train station, or in a park)” at least 1 time in the last 18 mo Food/nutrition insecurityMeasure 1: Medicaid enrollee address data and USDA ERS Food Access Research Atlas Measure 2: Ohio Medicaid Health Risk AssessmentMeasure 1: Ohio InCK links Medicaid enrollee address data to the USDA ERS Food Access Research Atlas at the tract level to identify area-level food deserts associated with primary residence. Measure 2: Ohio Medicaid Health Risk Assessment screener question, “At any time in the past year, have you run out of food before you got money to buy more?” and Ohio InCK assigns a food insecurity need to an answer of “Yes” at least 1 time in the last 18 moMaternal and child health service needs indicators Special education and early intervention are eligibleMedicaid claimsChildren who meet early intervention (EI Dx) clinical eligibility up to age 6 are flagged as having an EI need Women, Infants, and Children Supplemental Nutrition (WIC) Program EligibleMedicaid claimsChildren who meet enrollment criteria for Women, Infants, and Children Supplemental Nutritional Program (WIC Dx) up to age 21 are flagged as having a maternal child health support need Child Welfare InvolvementMeasure 1: Medicaid claims Measure 2: SACWIS dataMeasure 1: positive OOHP identified through Medicaid aid category codes using an 18 mo lookback for any prior involvement with child welfare. Measure 2: Children who are currently in an OOHP or have previously experienced OOHP in the past 18 moEI Dx indicates conditions eligible for enrollment in Special Education or Early Intervention; NOWS Dx, neonatal opioid withdrawal syndrome; OOHP, out-of-home placement; Parent BH Dx, parental substance use disorder; PMCA, pediatric medical complexity algorithm; SACWIS, statewide automated child welfare information system; SMI Dx, severe mental illness diagnosis; SUD Dx, substance use related diagnosis; WIC Dx, conditions eligible for enrollment in WIC (Women, Infants, and Children Supplemental Nutritional Program). Identified Need Domains in OOHP Algorithm EI Dx indicates conditions eligible for enrollment in Special Education or Early Intervention; NOWS Dx, neonatal opioid withdrawal syndrome; OOHP, out-of-home placement; Parent BH Dx, parental substance use disorder; PMCA, pediatric medical complexity algorithm; SACWIS, statewide automated child welfare information system; SMI Dx, severe mental illness diagnosis; SUD Dx, substance use related diagnosis; WIC Dx, conditions eligible for enrollment in WIC (Women, Infants, and Children Supplemental Nutritional Program). We designed a hybrid approach to risk stratification, which begins with a data-driven approach followed by interviews with those identified as potentially at risk. The data-driven component requires a dataset built through linking 4 data sources for the deterministic rule-based risk stratification algorithm: (1) Medicaid claims, address, and enrollment/eligibility records with a rolling 12-month to 18-month lookback period, (2) child welfare involvement data for eligible individuals [Statewide Automated Child Welfare Information System (SACWIS)], (3) enrollee responses to the Ohio Medicaid Health Risk Assessment (HRA), a health-related social needs screening tool administered by managed care organizations and the InCK team at NCH and reported to ODM, and (4) geocoded, area-linked data from the US Department of Agriculture Economic Research Service Food Access Research Atlas to measure area-level food access.14 To increase assurance that we could effectively identify behavioral health needs with Medicaid claims, we conducted an IRB-approved validation study in which we assessed behavioral health in SIL 1 beneficiaries using the Short Form 1215,16 or the Diagnostic Interview Schedule for Children.17 We identified behavioral health needs in 40% of SIL 1 beneficiaries. With the inclusion of these assessments, the algorithm correctly assigned all 30 beneficiaries to SIL 1. Feeling comfortable with the algorithm, we went about implementation, which required combining data from multiple state agencies and publicly available sources, leading to our next question. Question 3: How do we gain access to the necessary data? Answer: Build partnerships, trust, and data use agreements. The vision for our hybrid screening model was predicated on access to data from administratively separate agencies, each of whom operated under state and federal regulations that defined how their data could be shared and used. Operationalizing our screening model required a multipartner data use agreement (DUA) formalizing rules of implementation for proper handling and use of data. Partners included the lead institution, each state agency contributing data, and an “honest broker” data intermediary responsible for data linkage and ensuring that only DUA-covered data were transmitted to the lead organization team. The time associated with scoping data flow, convening agreement parties, and negotiating terms was substantial. From the inception of initial data requests to the first DUA execution spanned a period of ∼2 years. Additional DUA revisions were required as new agencies were added to the data partnership. Despite persistent and ongoing discussions required to execute the DUA, the time involved in convening meetings and coordinating activities supporting data linkage resulted in working relationships that proved invaluable during project implementation once data sharing terms were established. These working relationships were initially developed through standing biweekly work group meetings where technical and administrative questions could be formulated and resolved collaboratively. Ultimately, we were not able to reach legal agreements with every state agency whose data we sought (eg, Department of Education and Workforce, Department of Behavioral Health) and even within our successful agreements there were data restrictions that were not overcome (eg, no access to parent data for included children). However, the available data allowed the algorithm to serve as a strong first indicator of risk tier, with other risk factors of interest assessed from the parent/guardian during intake interviews with those identified with preliminary risk. Our DUA enabled access to multiple datasets sourced from structurally separated state agencies and public systems with different data formats and structures, but these data could not be used without a rigorous approach to data cleaning and linkage, leading to our next question. Question #4: How do we make this rich data work together towards our goal of reducing OOHP? Answer – Lots of data science! Data are linked by an “honest broker” with a history of government contracting for data linkage and analysis. To develop the InCK datasets, we begin with Medicaid eligibility data to identify beneficiaries who reside in one of the 2 counties, have at least 1 month of Medicaid eligibility in the prior year, and are under 21. We clean the data using a comprehensive process and complete simple deterministic linkage on person identifiers (SSN, first name, last name, DOB, and sex) where exact matches on SSN and first name, last name, DOB, and sex are treated equally. Medicaid IDs that share an SSN and/or first name, last name, DOB, and sex after data cleaning are linked together and assigned a unique person ID that is retained over the course of the model. Once the population is identified and deduplicated, we prepare the other data sources for linkage. Data preparation includes importing files, checking for inconsistencies, and reformatting data fields. Data on unique beneficiaries are then linked to Medicaid claims, Medicaid address, HRA, child welfare, and NCH screening data. Medicaid eligibility and address data from ODM are updated on a weekly and monthly basis. The Ohio Medicaid Managed Care Organizations (MCOs) are required to ensure the completion of an HRA within 90 days of enrollment for new members and on an annual basis for all members. The HRA data is reported on a quarterly basis. We generate domain-specific flags and use the algorithm to assign a SIL to each unique person ID. There are a few unique features of this process that require special treatment. The data cleaning and deduplication of Medicaid IDs is a particular challenge in this work. While the Medicaid ID is intended to serve as a unique identifier, there are scenarios in which a person can have multiple Medicaid IDs. In Ohio, Medicaid members receive a new Medicaid ID when they enter or change custody to protect the youth’s data and privacy. It is also possible that additional Medicaid IDs are generated in error due to procedural reasons. While the state actively corrects erroneous duplicate IDs, in some cases, service utilization may be spread across multiple IDs for a single beneficiary. Our deduplication data process links multiple Medicaid IDs together to accurately capture all health care utilization, diagnoses, and contact information for each beneficiary. It includes extensive data cleaning on person-identifying fields (name, SSN, and DOB) and uses SAS functions SOUNDEX, COMPGED, and SUBSTR to compare likeness between records. The data cleaning and deduplication steps are detailed in the Appendix, Supplemental Digital Content 1, http://links.lww.com/MLR/D142. Beneficiaries with multiple Medicaid IDs in Quarter 2 of 2024 are reflected in Table 2. Beneficiaries with 2 Medicaid IDs represented 3.8% of the total population, and there is a small percentage of beneficiaries with 3 or more Medicaid IDs. If these beneficiaries were not deduplicated, they would have artificially increased the number of total beneficiaries identified from 25,999 to 27,131. Given that the beneficiaries with OOHP are more likely to have more than one Medicaid ID and therefore a higher SIL, 58% of beneficiaries with 2 Medicaid IDs, there is a practical benefit to reducing multiple Medicaid IDs to one record to ensure that outreach is not sent to the same home repeatedly for higher SILs. TABLE 2Beneficiaries With Multiple Medicaid IDs (Quarter 2 of 2024)CategoryBeneficiaries with OOHP identified (n=623) (%)Total beneficiaries (n=25,999) (%)Beneficiaries with 1 Medicaid ID55 (8.8)25,011 (96.2)Beneficiaries with 2 Medicaid IDs543 (87.2)925 (3.6)Beneficiaries with 3+Medicaid IDs25 (4.0)63 (0.22)Note 1: OOHP were identified using an 18-month lookback from the end of Q2 2024.Note 2: Percentages represent the percentage of the column total for both columns.OOHP: Out-of-Home placement. Beneficiaries With Multiple Medicaid IDs (Quarter 2 of 2024) Note 1: OOHP were identified using an 18-month lookback from the end of Q2 2024. Note 2: Percentages represent the percentage of the column total for both columns. OOHP: Out-of-Home placement. Linking data to other sources beyond Medicaid claims is an essential part of correctly identifying SIL. To be in SIL 3 one must have an OOHP identified either through SACWIS or Medicaid aid category data (derived in part from SACWIS, but on a delay). In Quarter 2 of 2024, 5% of beneficiaries were linked to SACWIS, 95% to claims data, and 9% to HRA data (Fig. 2). HRA data completeness averaged 3%–9% in 2023 and 12%–16% in 2024 for the InCK population. As part of the Ohio InCK model, we have worked to increase completion rates by having pediatric primary care practices and NCH staff conduct HRAs in addition to the health plans. Other files contribute to the criteria for different SILs like having a behavioral health diagnosis or having a housing instability or food insecurity flag (See Fig. 3 for more details). Once a SIL is assigned, care coordinators complete the HRA and other screeners for children enrolled in services. Successfully linking data together increases the opportunity to correctly assign beneficiaries and improve data quality, but then what? Overlap of contributing InCK data sources. Claims indicate Medicaid administrative claims; HRA, health risk assessment; SACWIS, Statewide Automated Child Welfare System. Service integration level algorithm stratification algorithm. PMCA: Pediatric Medical Complexity Algorithm EI Dx: Conditions eligible for enrollment in Special Education or Early Intervention WIC Dx: Conditions eligible for enrollment in WIC (Women, Infants, and Children Supplemental Nutritional Program) NOWS Dx: Neonatal opioid withdrawal syndrome FAS Dx: Fetal alcohol syndrome diagnosis SMI Dx: Severe mental illness diagnosis SUD Dx: Substance use related diagnosis Parent BH Dx: Parental substance use disorder SACWIS: Statewide Automated Child Welfare Information System. This information is securely sent to NCH, which is responsible for contacting eligible families about their eligibility for care coordination activities and begins to answer our fifth question. Question #5 How does this work get us closer to a reduction in OOHP? Answer: With a lot more hard work. Involvement with child protective agencies introduces a unique complexity to engaging with families and beneficiaries about opportunities like those provided by InCK. Because a child’s parent is not always the current legal guardian, linking SACWIS and Medicaid data benefits care coordinators because they have enhanced access to information regarding custodial arrangements, allowing for greater timeliness, specificity, and accuracy related to a child’s OOHP. In addition, more accurate contact information increases the probability of successfully reaching the child’s legal guardian and receiving faster confirmation of the legal guardian’s interest in care coordination services. For example, this can lead to a smoother referral to a mental health professional and more streamlined access to child mental health services. Furthermore, the ability to track children and families across systems allows us to provide value to our partners who may not have easy access to this information. From the patient’s perspective, families involved in multiple systems are often asked to recount potentially traumatic or triggering experiences. By linking multiple data systems, care coordinators have access to a more complete picture of a child and the family’s health and social circumstances. When contact with the family is necessary, care coordinators are more knowledgeable about a family’s situation and can employ trauma-informed strategies and motivational interviewing before broaching difficult topics and needs assessments. As a result, families are able to open up more quickly to coordinators they view as trusted, knowledgeable, and empathetic professionals. Lessons Learned There are many lessons learned during this process. First, working with administrative data can be challenging. Navigating secondary data successfully requires strong working relationships where partners are invested in the project and willing to supply data dictionaries or use cases for various data fields. Further, data changes over time, even retroactively at times. Appropriate lookback periods are necessary to be as accurate as possible when employing these data in real-world interventions. With appropriate planning, ongoing quality assurance, and partnership investment, data linkage across various administrative data sources can be successful. In addition, these data sharing and linkages require care and attention to data sharing uses and roles to request data and execute data use agreements (DUA). While these can be complicated and lengthy processes, this relationship-building can have benefits beyond the DUA approval process. Those agency relationships have been invaluable during the implementation phase of the project. Furthermore, having a single neutral entity to coordinate this complex data process has been key. Their data linkage expertise and insights during the development and implementation of the algorithm helped make this project successful. DISCUSSION This work demonstrates how information from Medicaid, SACWIS, community-level SDH data, and patient/family-reported assessments can be linked to assess OOHP risk among Medicaid beneficiaries and inform interventions. Such work can have myriad implications for patient and family-centered outcomes research. First, linking existing data from disparate organizations can better identify families in need while reducing strain on scarce resources. Second, linking multiple data sources can reduce the burden on families who are often asked to repeat their stories to multiple organizations and complete similar need assessments. A fully screener-based assessment and stratification methodology, relying on patient and family report, is resource-intensive and receiving complete, accurate, and timely data for every family is challenging. In our care coordination outreach experience, we found that many enrollees did not have valid contact information or were unable to be reached, suggesting that completing this type of direct screening to the entire enrollee population would have limited reach. Furthermore, better ways to classify OOHP risk can ensure that supports and early interventions get to those in need at the right time and disrupt a child’s life as little as possible. While there has been work to predict OOHP and understand the outcomes associated with OOHP, the literature in the US has been challenged by small samples and limited data.18–21 The ability to link multiple data sources from multiple perspectives can help overcome some of these challenges. For example, we find that self-reported and claims-based measures of housing instability and food insecurity vary. However, they measure different dimensions of the concepts. With the combination, we can identify nuances and different mechanisms to address a family’s needs. Residing in a food desert may not mean a specific family experiences food insecurity, but if a substantial number of families live in such an area, a community-level intervention may be more valuable than one designed to help individual families. Our ability to multidimensionally measure food insecurity and housing instability allows us to better tailor interventions to the communities we serve. Our algorithm, which separates into low risk, risk of medical/behavioral health need, and imminent risk of OOHP, allows us to consider different interventions for different families in need. While families in higher levels of OOHP risk are eligible for care coordination services, we also implement interventions within the community that may have a greater impact on those with more medical and behavioral health needs. Locally, interventions include evidence-based home visiting models, early childhood mental health services, school-based health clinics, and mobile crisis response services. Defining these types of risk can mean more effective interventions can be implemented targeting specific mechanisms addressing these needs. Richer data can provide evidence-based information on how best to support families in need and ensure that children reside in safe and supportive homes. Finally, often fewer resources and supports are available in rural communities and thus finding innovative ways to address these needs and use resources effectively is imperative to ensuring that children flourish, regardless of the community in which they reside. Despite these benefits, we recognize the ethical considerations around using linked data to identify OOHP risk. This project had a waiver of consent, meaning there was no direct consenting process. However, we did implement several guardrails to mitigate potential privacy breaches and stigmatization, including no long-term labels in the records and no repopulation of algorithm information into beneficiaries’ permanent records. In addition, we worked to ensure that only the needed information at the “right” time is available to staff, highlighting the importance of a data intermediary. Next steps include understanding how families feel about the use of their data in these models. Limitations There are important limitations to consider. First, this algorithm is used to identify OOHP risk among children with Medicaid coverage in 2, primarily rural, counties in the Midwest. As such, it may lack generalizability to other communities and other populations. Nonetheless, we believe this can serve as a blueprint for engaging data owners across multiple systems. While situations vary, data linkage processes and considerations are likely similar across contexts. Second, administrative data, like Medicaid claims and the data from Ohio’s child welfare information system, have challenges and limitations that are important to navigate when implementing health interventions. Administrative data are not designed for data collection or health intervention work and are also subject to changes in the public policy environment. They do not always precisely align with research questions, can be messy and in need of data cleaning and organization, and may need linkage to pull data fragments together. The effects of these limitations can be mitigated with a thorough assessment of the variables, processes for data collection, as well as administrative data expertise and appropriate linkage strategies.22,23 For example, the Families First Coronavirus Response Act (FFCRA), which was signed into law in March 2020, prohibited states from disenrolling members from Medicaid, even if they were found to be ineligible. This was to ensure members did not lose vital health care coverage during the pandemic. Ohio’s Medicaid program resumed routine eligibility operations on February 1, 2023. This pandemic-specific process occurred during the implementation phase of the InCK model. We anticipate that the FFCRA implementation may have impacted the administrative data used by the Ohio InCK project, including less frequent disenrollment and address changes. In this case, changes in the administrative data might have affected the accuracy of our claims-based housing instability flag. Given that address data are updated monthly, we have reasonable confidence in its accuracy but acknowledge it as a limitation during the COVID-19 public health emergency. The evidence that bias exists in big datasets, like administrative data, is well established.24,25 In the expert-derived data-driven algorithm, we use linked data from multiple data sources, an open-source algorithm to identify pediatric medical complexity,26–28 and a confirmatory process to ensure children most in need are identified. However, biases may still exist as identifiers may change over time, data collection may differ across counties, and administrative claims data means we are only getting information on children touching the health care system. CONCLUSIONS Ohio InCK illustrates how data linkage across existing administrative datasets from multiple agencies can be used to improve the health and well-being of children. Finding ways to link other important data (eg, food/nutrition, early intervention, and housing) has the potential to enrich our understanding of children and their families and provide better support when needed. As these data linkages require significant time and effort to protect privacy and collaborate with multiple partners, balancing the information gained with the effort put into these data-sharing partnerships is an important consideration. Greater intention should be placed on how information adds to project goals to ensure that we value and protect the data we have access to while advancing patient-centered outcomes research. Supplementary Material