Public Data as Public Health Infrastructure

Abstract

Public health efforts addressing chronic disease and poverty require information that describes local conditions, preserves historical comparisons, and can be interpreted by the communities expected to act on it. Delaware’s My Healthy Community (MHC) platform brought health and social data together within a shared interface. Its 2025 decommissioning and transition to distributed sources offer a case for examining what community access to public data requires. We argue that access depends not on publication alone but on sustained stewardship and interpretive capacity. Adapting Penchansky and Thomas’s five dimensions of access (availability, accessibility, accommodation, affordability, and acceptability), we use Delaware examples, including a dated, topic-by-topic inventory, to show how access can erode even when data remain technically public. Substitute measures may not answer the original question, historical releases may become difficult to retrieve, and distributed sources require documentation and assistance. Free data also carry costs in staff time and expertise, and unexplained changes can weaken trust. Artificial intelligence makes analysis easier to produce without making it easier to verify, increasing the importance of these supports. We propose maintaining historical releases, documenting changes, and funding the partnerships and interpretive support required for responsible community use.

Introduction

Efforts to reduce chronic disease depend on understanding the conditions in which people live, including whether they can afford housing, obtain nutritious food, and receive care when it is needed. These circumstances influence both the development of disease and the resources available to manage it, though their effects may accumulate across years and remain poorly characterized in clinical records.¹ Patients who frequently require emergency medical care for diabetes, for example, may do so as a direct consequence of difficulty obtaining medications, unstable living circumstances, limited access to affordable nutrition, or limited outpatient care. Responding to those needs requires information that extends beyond a singular healthcare encounter, and observable patterns in the prevalence of disease can highlight disparities and opportunities for intervention.

Delaware’s My Healthy Community (MHC) platform brought health and social data together to describe these relationships locally, reflecting a collaboration between public health informatics and epidemiology.² In late 2025, the Delaware Department of Health and Social Services (DHSS) decommissioned MHC, intending to transition source data to the Delaware Open Data Portal while the next phase was planned.^3,4 A statewide resource supports a common understanding of the incidence, prevalence, contextual factors, and growth or contraction of health issues requiring attention. When that source becomes difficult to retrieve and future updates are uncertain, disruption of this data commons draws resources away from deciding how to respond, and limits participation to those with the capacity and skills to find the data elsewhere.

Using MHC’s decommissioning as a case, we propose that community access to public health data depends not on publication alone but on sustained stewardship and interpretive capacity. Our concern is community accessibility, meaning whether residents and community-based organizations can obtain, interpret, and continue using relevant data. We first describe MHC and adapt the five dimensions of access to public health data. We then examine each dimension in turn, drawing on Delaware examples from chronic disease and poverty measurement, State Health Improvement Plan (SHIP) implementation, and community data partnerships. We close by considering artificial intelligence (AI) and outlining what data transitions must preserve under each dimension.

My Healthy Community and the Five Dimensions of Access

A Common Basis for Understanding and Action

MHC integrated health outcomes, health care utilization, environmental conditions, and social circumstances within a common data interface. Users ranging from seasoned scientists to everyday residents could examine topics across Delaware geographies, including units smaller than the county.² By bringing these measures together, MHC provided a basis for discussing how conditions overlapped within communities, while keeping definitions and source documentation close to the displayed findings.

This arrangement also supported collaboration. Organizations working on housing, food access, or chronic disease could begin with a common measure and identify where their questions differed, without each independently reconstructing the source, denominator, or geographic boundaries.

What the Transition Changed

Our May 2026 comparison of MHC’s former content with the Open Data Portal identified gaps involving substance use, hospital discharge data, integrated Social Vulnerability Index displays, and Equity Counts presentations.⁵ The comparison identified seventeen categories missing outright. We distinguished source datasets from dashboards and narrative reports, because losing an interface differs from losing the underlying records. The comparison did not cover every standalone agency publication or reconstruct every former holding. Its contribution is to identify specific discontinuities in access rather than to estimate the total amount of Delaware data that remain publicly available.

Adapting the ‘Five As’ to Public Data

In 1981, Penchansky and Thomas described access to health care as the fit between people and the system serving them, distinguishing availability, accessibility, accommodation, affordability, and acceptability.⁶ We adapt these five dimensions to public health data as a conceptual framework to organize and analyze the impact of MHC discontinuation on access to a common data set.

(1) Availability concerns whether an appropriate indicator exists for the population, geography, and period of interest. (2) Accessibility describes whether people can locate and retrieve it through the digital resources available to them, with reproducible methods, and whether the data persist over time. (3) Accommodation concerns how publication is organized around users’ needs, including understandable documentation, ease of interpretation, usable formats, language, and technical assistance. (4) Affordability includes the staff time, expertise, and software required to prepare and interpret the data even when the download itself is free. (5) Acceptability concerns whether communities regard the source, categories, and proposed uses as credible and appropriate, including whether their knowledge informs interpretation.

These dimensions specify what a transition must preserve to meet the needs of a diverse user base, and each can erode even when data remain technically public. A file may remain available but through migration becomes difficult to locate. A dashboard may be easy to navigate for a single condition but lack coherence with other variables and datasets. An organization may retrieve both and still lack the staff time to reconcile discrepancies. Community accessibility therefore depends on the relationship between the resource and its intended users, including the support necessary to use it responsibly.

1. Availability and Appropriate Measures

When a familiar source disappears, a similar measure is often substituted. Availability depends not on whether some indicator exists, but on whether the indicator answers the original question for the intended population, geography, and period. As an example, the loss of block-group resolution examined in the next section affects both dimensions as it changes the geography at which a measure is available as well as the effort required to retrieve it.

Substitute Measures are Not Equivalent

Poverty can be described through income, assets, material hardship, or eligibility for assistance, each capturing different circumstances. When individual socioeconomic information is incomplete, investigators frequently link health records to area-based measures, which vary substantially in variables, geography, and intended application.⁷ The Index of Concentration at the Extremes, for example, characterizes concentrations of privilege and deprivation, including racialized economic segregation,⁸ while the CDC/ATSDR Social Vulnerability Index (SVI) was designed to support preparedness and response to hazards and emergencies.⁹ These measures cannot be assumed to be interchangeable. Rollings and colleagues found that Area Deprivation Index (ADI) and SVI rankings had only moderate correlation, with fewer than half of census tracts differing by no more than one decile.¹⁰ Substituting an available index may therefore change which communities are classified as most disadvantaged, and none of these area-level measures establishes the income, resources, or experience of every resident.

Outcome substitution creates a related problem. Replacing observed diabetes-related emergency visits with modeled diabetes prevalence changes the question from health care utilization to estimated disease burden. CDC PLACES provides useful small-area estimates, but its models draw on survey and census information, including a county poverty measure,¹¹ and so associations between modeled outcomes and socioeconomic indices may partly reflect shared model inputs.

Geography can also interrupt comparisons. Postal ZIP codes describe mail delivery, while Census ZIP Code Tabulation Areas provide approximations for statistical analysis,¹² and community service areas may cross several of these boundaries. Changes in boundaries, population composition, or denominators can interrupt comparisons even when the measures themselves remain unchanged, and small populations produce estimates that are unstable or cannot be released. Federal policy changes may further complicate this estimation process. The Department of Commerce’s June 2026 disclosure-avoidance policy establishes coarsening (grouping or rounding information) as the preferred approach for covered statistical products, and the Census Bureau acknowledges potential reductions in detail for small areas and population groups.¹³ Because the policy applies prospectively and implementation varies by program, Delaware users should examine each release for changes in categories or resolution rather than assume that a new estimate preserves the detail of its predecessor.

Some Needed Measures Were Never Available

Availability limits predated MHC’s transition. Many chronic diseases develop through exposures operating across years, and a profile of a neighborhood at a single point in time cannot reconstruct earlier exposures or an individual’s residential history.¹ Repeated observations of a neighborhood also differ from following the same residents over time, particularly where displacement or migration, often driven by economic forces, changes the population being described.

In 2022–2023, United Way of Delaware’s Equity Counts Partnership Initiative provided data consultation to four community-based organizations, helping them use local data to assess community needs, guide programming, and support grant applications.¹⁴ This work revealed concrete gaps in the available data. ZIP-code-level adverse childhood experience data available at that time came from the 2015 Delaware School Survey, with gaps where reliable estimates were unavailable, and organizations sought measures of eviction, foreclosure, and social cohesion that existing sources could not adequately provide. Restoring an interface alone would not resolve the need for timely, sufficiently detailed data.

2. Accessibility and Persistence Over Time

Preserving the Ability to Examine Change

Historical releases permit examination of gradual change, while retained definitions help distinguish population trends from changes in measurement. The 2025 KIDS COUNT in Delaware FOCUS book, Strength in Numbers, shows how incremental changes become visible across thirty years of measurement and partnership, and documents difficulty comparing racial and ethnic categories across changing collection practices.¹⁵

The FAIR principles identify findability, accessibility, interoperability, and reusability as foundations for scientific data stewardship, with metadata and provenance supporting both human and automated use.¹⁶ The National Institutes of Health (NIH) updated 2026 data management and sharing guidance to explicitly encourage practices consistent with FAIR,¹⁷ although the NIH requirement does not govern every state public health dataset. For community data systems, these principles support retaining the exact releases, definitions, denominators, and transformations used in earlier work. A current file cannot reproduce an earlier analysis if the accompanying definitions or geographic assignments have disappeared.

Discontinuities Require Active Stewardship

Preservation of data is not a passive process, as data discontinuities can arise in myriad ways, including ended collections, interrupted updates, altered definitions, and withdrawal of an interface. These events require different responses, but each can prevent users from continuing an analysis that was previously feasible. The relevant question is whether the original measure, its interpretation, and a comparable subsequent observation can still be recovered.

MHC's distinguishing feature was the resolution at which it published (Figure 1). The platform presented health indicators at geographic levels reaching the census block group, an area of as few as 600 residents, and it was one of few systems nationally able to publish health data at that resolution.² Small-area publication is ordinarily limited by disclosure risk, because a single case within a small population can identify a person. MHC addressed that risk in software rather than by withholding whole geographic levels. Any frequency based on fewer than eleven individuals was suppressed at display time and the data automatically aggregated over a longer period, a larger geographic unit, or the attribute itself until the threshold was met, a (k, p)-anonymity scheme with k = 11 and p = 0.¹⁸ That allowed publication at the census block group, a geography of as few as 600 people, wherever the threshold was met, and its developers identified the algorithm as among the platform’s most important innovations.^2,18 The method was certified under the expert determination provision of the HIPAA Privacy Rule, 45 CFR 164.514(b)(1), which permits release when a qualified expert determines that the risk of identifying an individual is very small.² A privacy methodology of this kind is not a limit on small-area release but its precondition, and it is a capability separate from the data itself.

Figure 1. My Healthy Community Data Architecture.

Consequently, this resolution has not carried over. Green River’s 2024 case study reported nearly 1,900 indicators from more than 30 datasets by July 2021.¹⁹ Our May 2026 inventory documented 53 source datasets.⁵ As of May 30, 2026, the only current sub-county health content is communicable disease surveillance at the census tract. Table 1 provides a comparison by topic area of holdings previously reported in MHC which currently can no longer be retrieved or reproduced at the equivalent geographic specificity from the Open Data Portal.

Table 1. My Healthy Community Content and Delaware Open Data Portal Status, by Topic Area, May 2026.

Topic area	Published Geography	Status on the Open Data Portal, May 2026
Substance Use	ZIP	Not found
Chronic Disease and Hospital Utilization	Census Tract	Not found
Mental Health	Census Tract	Not found
Sexual Violence	County	Not found
Social Vulnerability Index	Census Tract	Not found
Equity Counts Data Center	Census Tract	Not found
Youth Health Behavior	State	Not found
Prescription Monitoring	ZIP	Not found
Emergency Department and Syndromic Surveillance	ZIP	Not found
Community Characteristics	Census Block Group	Not found below the county
Communicable Disease	Census Block Group	Census tract, headline indicators only
Maternal and Child Health	Census Tract	County only, 2009-2016
Vital Statistics and Mortality	Census Tract	County only, 2009–2016 (cancer mortality: state only, 2008–2012)
Cancer	Census Tract	State only, 2008-2012
Air Quality	Census Tract	Facility or state emissions only

Footnote: Portal status reflects a search of the portal's Health category and relevant Environment entries conducted May 30, 2026, with the vital statistics, cancer, and communicable-disease holdings re-verified September 4, 2026. “Not found” means that no matching dataset was identified in that search. MHC geographies are the finest published geographies documented in the May 2026 inventory. Tract-level chronic disease measures include ECDC small-area estimates; hospital discharge measures were published at ZIP level. Sub-county portal values appear on three communicable-disease datasets only, where geography is recorded as a text label without federal geographic identifiers, and so joining them to census-tract data requires a name-based crosswalk.

Currently no portal dataset reaches the census block group, and as no publicly documented suppression methodology accompanies the portal, the resolution that distinguished the MHC platform is unlikely to be replicated.

Discontinuities in national dataset availability illustrate how fragility to digital data permanence can occur at scale. Jacobs and colleagues identified in October 2025 unexplained pauses in 38 of 82 CDC databases previously updated at least monthly, and recommended visible update status, reasons for interruptions, and anticipated resumption.²⁰ The Federal Data Terminations Tracker and Data Rescue Project similarly distinguish loss of collections, data elements, and tools.^21,22 Continuity does not maintain itself but requires active stewardship.

Updating SHIP Objectives

The 2025–2028 SHIP was informed by the 2022–2023 State Health Assessment (SHA), which was undertaken by the University of Delaware Partnership for Healthy Communities in partnership with the Delaware Department of Health and Social Services Division of Public Health.^23,24 This work drew on MHC for several baseline measures. Its cancer-mortality objective, for example, specifies an age-adjusted baseline of 164.8 deaths per 100,000 for 2016–2018 and a target of 148.3 by 2028, citing MHC as the source.²⁴ Updating such an objective requires comparable definitions, time periods, and adjustment methods, together with a clear account of how the original baseline was produced.

Loss of the MHC interface and incomplete transfer to the Open Data Portal have consequently made it difficult or impossible to update some objectives through their originally specified sources. The July 2026 Community Data Resource Guide developed by the SHIP team and KIDS COUNT in Delaware identifies alternative publications, portals, selected archives, and contacts for requesting data.²⁵ This work can support recovery, although each alternative must be reconciled with the objective’s original measure.

3. Accommodation of Users’ Needs

Integration and Documentation

MHC’s integrated design reduced the work of combining sources and understanding their definitions, and distributed sources shift that work to users. The Open Data Portal provides a general repository across Delaware agencies, while other platforms organize data around particular populations and purposes. Healthy Communities Delaware uses IP3 ASSESS to support assessment through vital-conditions frameworks, geographic comparisons, and shareable visualizations, but does not cover the scope or share the sources of MHC.²⁶

KIDS COUNT illustrates why documentation matters. Its national indicators draw on federal sources for comparisons across states, while independently operated state organizations contribute locally sourced measures, so similarly named national and state indicators may contain different data.²⁷ A common website does not ensure uniform content, geographic detail, or historical coverage. Nescott, Barlow, and Perez-Rivera’s description of Delaware poverty alongside employment, housing, food, and other basic needs demonstrates the interpretive work state partners contribute in determining which measures describe local circumstances and how their limitations should be communicated.²⁸

Technical Assistance and Research Literacy

United Way of Delaware’s Equity Counts Partnership Initiative illustrates the assistance community use requires. Supported by Highmark Blue Cross Blue Shield Delaware, the initiative worked with four organizations (Children and Families First, Cornerstone West, Network Connect, and Wilmington Alliance) from October 2022 through July 2023 on questions ranging from neighborhood population and wealth change to adverse childhood experiences and workforce development.¹³ A data specialist met weekly or biweekly with designated representatives to refine questions, locate information, and prepare findings.

Support extended beyond finding data in MHC. The initiative also helped organizations assess alternative sources, interpret uncertainty, and distinguish population conditions from outcomes attributable to their programs. Participant feedback also informed the addition of state, county, and census-tract views to the Equity Counts Data Center in July 2023. Feedback was favorable, though the initiative’s small size and descriptive evaluation limit conclusions about lasting organizational independence or improved health outcomes.

Goodman and Sanders Thompson’s Public Health Research Methods for Partnerships and Practice provides a methodological resource for community engagement in building capacity to understand data, research, and public health, and their related Community Research Fellows Training program combined research education with community participation.^29,30 A similar Delaware implementation could pair research literacy with recurring technical assistance and shared documentation. The relevant capacity includes recognizing when a question cannot be answered by the available data and explaining that limit to partners and funders.

SHIP provides a structure for sustaining this support across organizations. The University of Delaware’s Partnership for Healthy Communities has supported sustained, reciprocal partnerships aligned with SHIP, and the current plan is implemented through a Partnership Coalition spanning CBOs, nonprofits, government, health systems, education, transportation, and housing.^31,32 The coalition includes 162 individual members representing 83 distinct organizations, as of August 2026.

4. Affordability and the Cost of Free Data

Open data support independent analysis and allow users to construct questions beyond those anticipated by a dashboard. However, publication leaves substantial work involving discovery, preparation, linkage, and interpretation, described by Kucharczyk as the “last-mile” public data problem.³³ Martin and colleagues’ study of community-based organizations (CBOs) identified constraints in funding, staffing, skills, and data systems, alongside population data that did not adequately describe local communities.³⁴

These costs are concrete. In addition to the data specialist’s time, Equity Counts participants committed approximately two to four hours per month to meetings and a similar amount outside meetings, and progress had to be documented to survive changes in participating staff.¹⁴

For CBOs, these costs affect decisions about initiating and sustaining services. Population data help establish need in grant applications, identify service areas, and describe the conditions against which programs operate. When an update is delayed or no longer accessible, organizations may be unable to refresh those accounts even as funders expect current evidence. Missing data do not demonstrate that need has diminished. Nevertheless, smaller communities may become harder to represent in funding decisions precisely because their populations cannot support stable, publicly reportable estimates.

5. Acceptability, Credibility, and Community Knowledge

Shared Definitions and Community Knowledge

A shared dataset cannot resolve every disagreement about priorities or causation, but explicit definitions and jointly interpreted findings can reduce repeated disputes about what the numbers represent, allowing more attention to the actions partners can take and the changes they will assess together.

Community knowledge also shapes what the numbers mean. Delaware’s Community Health Advocacy Mobilization Group (CHAMG), working with the Division of Public Health and Delaware State University, examined chronic health conditions and community assets in 52 census tracts with high social vulnerability, selected using tract-level SVI published in MHC. Its 2025 report combined these profiles with resident conversations to develop recommendations concerning care, food access, transportation, and other conditions affecting health.³⁵ As described in Table 1, tract-level SVI is not available on the portal, and so the next cycle of this work cannot reproduce its tract selection from the portal alone.⁵ The SHA’s community conversations in all three counties provided complementary detail, with New Castle County participants describing transportation constraints, difficulty finding information about services, and forms that made assistance difficult to obtain.²³ These accounts help explain how an apparently available resource can remain inaccessible and identify questions for quantitative assessment, without establishing causal effects or implying that participants’ experiences represent every resident.

Trust and Misinformation

Public confidence depends on whether people can understand changing evidence and evaluate the information presented to them. In 2020, Lee and colleagues described conflicting COVID-19 messages and the need to work through trusted churches, barbershops, and salons in Delaware.³⁶ Troisi placed the growth of vaccine misinformation and disinformation within a longer history of hesitancy, identifying declining trust during and after COVID-19 as a continuing public health concern.³⁷ Credible information travels through established relationships.

Unexplained data interruptions complicate this work. When an agency cannot show why a number changed, where a prior release went, or when an update is expected, community partners have less capacity to distinguish a legitimate revision from an unsupported claim, which may weaken confidence in both the data and their interpretation. Transparent corrections, retained source material, and accessible explanations give partners a basis for addressing misinformation while acknowledging uncertainty and legitimate criticism.

Accountability depends on the same transparency. As an example, SHIP coalition members need to know whether an apparent lack of progress on an objective reflects worsening health, an unchanged estimate, or an unavailable update. Recording that distinction preserves accountability without treating missing information as evidence of either success or failure.

Responsible AI and Verification Capacity

The increasing accessibility of AI tools makes community data capacity more consequential. An organization may now produce a map or summary without staff who can determine whether the system substituted a national estimate, combined incompatible geographies, or interpreted an area-level association as an individual risk. AI lowers the cost of producing analyses but not the cost of verifying them. This increases the need for informed review, particularly when findings support service decisions or funding requests.

The National Institute of Standards and Technology’s AI Risk Management Framework provides a basis for governing and evaluating these uses.³⁸ As for human work, AI-assisted work should retain exact source releases, executable calculations, and documented transformations, with review by people who understand epidemiology and the affected community. Training should include checking denominators, recognizing suppression and uncertainty, and identifying unsupported causal claims. Unavailable measures should be identified explicitly, any substitution justified and disclosed, and privacy protections maintained when public sources are combined.

Conclusion and Recommendations for Data Transitions

Public data function as public health infrastructure whose value depends on maintenance, responsible ownership, and the people equipped to use it. MHC’s decommissioning shows that access can erode along each dimension even when data remain technically public. Replacement systems should therefore be assessed against community tasks, such as updating a neighborhood profile or following a specific objective, using the five dimensions of access to identify what must be preserved and what additional support is needed.

1. Availability. When a measure is discontinued or replaced, identify the loss explicitly and document how any substitute differs in definition, geography, and population.

2. Accessibility. Accompany changes to collections and interfaces with advance notice, a named responsible source, visible update status, and a plan for maintaining historical releases with their definitions and denominators.

3. Accommodation. Keep documentation close to the data and sustain technical assistance and research-literacy training for community partners.

4. Affordability. Provide recurring funding that covers both data maintenance and the staff time required for CBOs to interpret and use the data.

5. Acceptability. Explain revisions and interruptions transparently and include community partners in interpreting findings.

For communities addressing chronic disease and poverty, accessible data and documented measures allow earlier questions to be revisited as conditions change, while sustained partnerships provide the interpretation needed to guide subsequent action.

This article includes recommendations about the importance of persistence and reproducibility/verification of data. The authors have used Zenodo as document repositories for two data pieces they feel are important to include: Knapp’s Indicator inventory and Open Data Portal Comparison, and Kauffman’s Equity Counts Partnership Project report (https://doi.org/10.5281/zenodo.22841341).

Funding Statement

MBK founded Green River, the firm that designed and built the My Healthy Community platform under contract to DHSS. He transferred ownership of Green River to its employees in November 2024 and retains a financial interest in the firm through seller financing from that transition. He is the principal of Sedoha, LLC, which conducted the inventory comparison (Knapp, 2026) without compensation. Sedoha, LLC has no current contract relating to the platform or its successor.

This work was supported by an Institutional Development Award (IDeA) from the National Institute of General Medical Sciences of the National Institutes of Health under grant number U54-GM104941 (PI: Hicks) and the State of Delaware.