The key to driving political/institutional and wider systemic change needed to address health equity in immune disease is improving evidence-generation and its communication by subject-matter experts of the benefits of AI and BDA to key stakeholders (e.g., payers, clinicians, and regulators). While there are subject-matter experts in AI/BDA, health equity, or immune diseases, there is understandably a limited pool of experts proficient in all these areas. The benefits of AI and BDA therefore need to be communicated more widely, particularly in relation to health equity, to ensure it is widely adopted and these multidisciplinary experts can be fostered.

One approach could be to increase awareness at educational institutions and offer multidisciplinary undergraduate/postgraduate training of immunologists regarding computational biology, programming, and bioinformatics, among other BDA-related topics (Schultze, 2015). Creation of a common, clear, and shared lexicon, built on existing information (Fuller et al., 2017; American Medical Association and Association of American Medical Colleges, 2021), will also be important to minimize inconsistencies and facilitate the synthesis and comparison of data (Palaniappan et al., 2024) so that the benefits of AI and BDA for health equity can be understood. These definitions could be incorporated into data guidelines and standards to inform multidisciplinary consortia involved in decision-making. For example, the Findable, Accessible, Interoperable, Reusable (FAIR) data principles offer domain-independent guidelines for producers and publishers to enhance data reusability through effective data management and stewardship (Wilkinson et al., 2016). More recently, the Gravity Project has provided consensus-based standards on how social determinants of health (SDoH) data are used and shared (Gravity Project, 2025). In addition to establishing a clear lexicon aligned with existing data guidelines and standards, it is important to define trackable key performance indicators (KPIs) to assess the value of health equity initiatives. Multidisciplinary expertise should be engaged to develop a communication strategy for health equity in big data/BDA, effectively conveying these KPIs and highlighting areas with potential or demonstrated ROI.

Evaluating the role of AI, ML, deep learning, and advanced analytics in promoting health equity for immune disorders requires a multidimensional set of KPIs that must address data inclusivity, diagnostic fairness, patient outcomes, and real-world implementation, with a persistent focus on addressing disparities rather than just improving averages. To ensure KPIs are communicated to and used effectively by key stakeholders they could be incorporated into data guidelines, standards, and recommendations [e.g., FUTURE-AI (Lekadir et al., 2025)] and disseminated via a new position paper. Example KPIs for determining the success of big data efforts to achieve health equity are summarized in Table 1. A more exhaustive list is available in a white paper from the National Committee for Quality Assurance (Harrington et al., 2021).

In terms of BDA use more broadly, KPIs need to demonstrate minimized deviation from a gold-standard level of representativeness, minimized outcome disparity, and most importantly usability and scalability within robust healthcare systems. As proposed by Zimmerman and Anderson (2019), multiple metrics for health equity (health inequality, health disparities, and mean health) could be consolidated using an approach based on deviations from the best achievable health, defined by the median experience of the most privileged group (Zimmerman and Anderson, 2019). In addition to the traditional financial ROI metric, (net benefit – cost)/cost), an emerging concept is the creation of a single, blended metric that expands ROI beyond purely financial returns to include concepts such as value and social return on investment (SROI). SROI assigns financial proxy values to non-financial outcomes, making intangible social impacts apparent by demonstrating the broader social value generated from financial investments. Although SROI is recognized by domain experts as a valuable tool for demonstrating social value, awareness across the broader public health field remains limited. There is also a need for standardized methodologies and data reporting practices to ensure its validity and interpretability (Ashton et al., 2024).

Addressing health inequities in immune diseases can be costly and it is challenging to quantify a ROI. However, the unmet need is evident. Immune diseases are lifelong and expensive to treat, with direct costs in the US projected to be around $200–300 billion annually (Wylezinski et al., 2019; Miller, 2023b). Further, the estimated costs of health inequities in 2018, based on medical care expenditures, lost productivity, and costs associated with premature death, were approximately $450 billion for minoritized populations in the US (LaVeist et al., 2023). Considering the contribution of health inequities to disease costs, a recent study has demonstrated that there is great potential for ROI (Yerramilli et al., 2024), particularly in the light of well-documented impacts of SDoH (Braveman and Gottlieb, 2014) and the associated gains in health and productivity (Yerramilli et al., 2024). Despite the potential for strategies targeting SDoH to improve health outcomes and generate cost savings, literature on ROI is scarce (Nikpay et al., 2024). In the last 20 years the US government, non-governmental organizations, and corporations have invested over $179 billion in health equity (Aluko et al., 2023) and despite this—according to one independent analysis—many disparity metrics have shown little or no improvement (Zimmerman and Anderson, 2019). Aluko et al. proposed multiple reasons for the failure of previous health equity investments/approaches, including insufficient governance (see Section 2.6), limited workforce and capacity skill sets, and unsupportive data and technology infrastructure. Some of the proposed solutions from the authors included a refocus on longer-term commitments, and investment in BDA platforms capable of understanding, targeting, and tracking disparities over time (Aluko et al., 2023).

By integrating healthcare data from diverse sources, BDA has the potential to enhance clinical decision-support tools and aid development of personalized or population-based services (Schulte and Bohnet-Joschko, 2022). Indeed, big data has impacted patient care for decades by helping insurance companies incentivize preventive care, ultimately leading to reduced acute care costs and improved care equity (Sabet et al., 2023). A scoping review found that three-quarters of papers reporting economic evaluations of BDA for clinical decision-making corroborated expectations of cost savings, ranging from US$126 per patient to over US$500 million for the entire US healthcare system; however, the interpretation of results was limited by a lack of full and properly performed economic evaluations (Bakker et al., 2020). Even in the absence of robust economic/ROI data, investment in BDA remains attractive and has recently been supported by the US Department of Health and Human Services (HHS), who in 2022 pledged US$90 million to identify and reduce health disparities using new data-driven solutions (Sabet et al., 2023). For investment to continue it will be important for the HHS and other institutions to recurrently evaluate their initiatives, using appropriate KPIs to determine value and ROI. In the absence of large long-term cases, smaller initiatives, such as the EHR-enabled rheumatology registry developed by the American College of Rheumatologists (Gilvaz and Reginato, 2023), may offer the best opportunities to highlight the potential of BDA on health equity and ROI in the near term. Alternatively, researchers may be able to investigate potential for AI-powered insights into health inequalities by registering for established precision medicine initiatives such as the National Institutes of Health-funded “All of US Research Hub” that is built on strong privacy and trust principles including governance, transparency, consent, and data quality (All of Us Research Program, 2025). While funding for this program is declining, opportunities remain for funded partnerships that may help identify solutions to health inequalities (All of Us Research Program, 2025); however, such opportunities are limited and without clear evidence of ROI, competing institutional priorities (e.g., financial sustainability, regulatory compliance, or short-term efficiency gains) may take precedence, to the detriment of health equity.

A survey of US healthcare executives found only 36% have a specific budget dedicated to advancing health equity (Accenture, 2022). A more recent survey found that 43% of life sciences executives and 48% of healthcare executives found it challenging to incorporate health equity into their strategic, financial, and operational processes (Deloitte Center for Health Solutions, 2025). Even when institutions commit to implementing big data and health equity initiatives, they often lack the governance structures, mechanisms, and metrics to track and encourage delivery. This is because unlike regulatory compliance, which is tied to financial or legal consequences, health equity efforts driven by big data often remain voluntary and lack clear metrics for accountability. In this regard, the KPIs described in Table 1 could prove valuable for tracking progress against key health equity metrics, facilitating accountability, and assessing the success of related initiatives.

According to one survey, more than 40% of life sciences and healthcare executives had difficulties tracking the progress of health equity initiatives (Deloitte Center for Health Solutions, 2025). Furthermore, 32% of health equity leaders had no data on the impact of health equity initiatives on their organizations' financial indicators. The same survey reported that while health equity leaders have the potential to differentiate between short- and long-term goals, relatively few are involved in decisions related to technology and IT (14%) and use of AI (12%). These findings suggest that the economic models of healthcare delivery and biomedical research may not align with the investments (and ROI) needed to use big data for equity-focused interventions. However, this may change in the future if health equity leaders can identify the factors associated with cost savings and ROI or shift to incorporating SROI and other broader value-based assessments.

Health equity leaders, in particular, are needed to direct policy and investment opportunities. Additionally, considering the lack of performance incentives and the need to incentivize health equity in resource-limited providers, the Agency for Healthcare Research and Quality's stakeholder engagement recommended the development of equity-focused evidence-based quality indicators, use of federal data to develop health equity benchmarks, and development of toolkits to assist healthcare organizations with integrating health equity metrics into their performance management (Chisolm et al., 2023). A standardized set of health equity measures would enable value-based incentive programs to reward strategies that reduce performance gaps by addressing the unique challenges faced by disadvantaged populations, rather than assuming that improvements in overall population outcomes will automatically benefit at-risk groups (Chisolm et al., 2023). Looking ahead, further investigation is needed to understand which incentives have the greatest impact, as well as which groups of stakeholders are best positioned to deliver health equity improvements. A tiered approach to performance incentives has also been proposed to ensure efforts that fall short of key benchmarks are still recognized as progress (Chisolm et al., 2023).

As proposed by Aluko et al. (2023) investment in longer-term commitments, such as BDA, will be important to ensure the success of health equity initiatives. However, investment in AI-powered BDA platforms capable of tracking disparities over time is challenging owing to difficulties in collecting sufficient, reliable, and up-to-date information on health disparities. For example, understanding health disparities requires careful consideration of confounding factors, such as healthcare insurance in the US, and the selection of appropriate research questions and populations. To address the unmet need for frequent and granular data collection, particularly regarding SDoH, Sabet et al. discussed the potential benefits of a large national public database of anonymized patient data capable of collecting diverse metrics based on equitable data collection strategies (Sabet et al., 2023). To ensure that the database captures data from marginalized populations, these groups should be included in the process from the early design stages, the design should be adapted for those with low literacy or limited technological proficiency, and investment should be made in technology infrastructure and staff training to prepare for comprehensive data collection (Sabet et al., 2023). Further, recommendations and guidelines are needed to progress the field in an ethical and collaborative manner to ensure data collection and storage methodologies adhere to ethics regulations and data privacy laws, and that findings can be effectively translated into clinical decision-making (Gossec et al., 2020).

To address these challenges, several data standards and principles have been developed, such as the Clinical Data Interchange Standards Consortium and the FAIR Guiding Principles for Scientific Data Management and Stewardship (Wilkinson et al., 2016; cdisc, 2025). As mentioned previously, success of such a database would be predicated on the development and achievement of predefined KPIs (see Table 1). It would also need sufficient data to address the lack of information on rare immune diseases, which would benefit from consolidating information from multiple sources (Peng et al., 2021). A holistic approach to health equity remains difficult due to the fragmentation of patient data across EHR systems, insurance databases, and research cohorts, hindering the development of comprehensive, equity-driven insights. Without institutional commitment to data sharing, achieving health equity will be challenging. Promoting cross-sector collaboration and using data dashboards to deliver insights to researchers and policymakers could be one solution to expedite investment (Sabet et al., 2023) in big-data platforms for immune disease.

A potential solution to these issues is to enhance data accuracy by increasing patient ownership, allowing patients to review, edit, or validate their personal information. In the global shift toward paperless healthcare, patient data are increasingly accessible through online portals and mobile applications (e.g., MyChart). These platforms typically incorporate multiple features that have been shown to encourage patient ownership, including multilingual support, consolidation of data across multiple connected systems to prevent inefficiencies (e.g., entry of similar data across multiple platforms), and protection of confidentiality (Peng et al., 2021; Vishwanatha et al., 2023). However, there are potential limitations—while these platforms may allow patients to view their data, they can lack functionality to directly edit or correct inaccuracies related to ethnicity. Additionally, they may introduce errors due to a limited set of standardized ethnicity response options. Feedback from service users and advocacy groups could help refine these systems; however, a global framework for standardizing ethnicity categories may be needed to support future data integration and better identify disparities.

A key concern with engaging patients in data collection is that access to digital health technologies (DHTs) and overall digital literacy, which are key digital determinants of health, can create a digital divide that impacts the representativeness of big data (Ibrahim et al., 2020; Eruchalu et al., 2021; Campanozzi et al., 2023; Chidambaram et al., 2024) and may affect efforts to increase the diversity and accuracy of patient-reported data. For example, despite DHTs being increasingly used by patients and physicians in the management of asthma, their usage in smartphone applications has been shown to widen the digital divide by SES, as not all individuals own smartphones (Kaplan et al., 2023). While such technologies can be used to facilitate earlier diagnosis of asthma (Al Meslamani, 2023), and also atopic dermatitis (Yanagisawa et al., 2023), there is potential for outcome disparities to arise due to earlier diagnosis/DHT use and treatment in groups of higher vs. lower SES. It is therefore important to engage communities, implement strong governance, and enhance public digital literacy to ensure that the digital divide is minimized rather than widened by the adoption of DHTs (Fernandes et al., 2024).

In addition to providing ownership and an infrastructure for patients to validate their data, the AIM-AHEAD (Artificial Intelligence/Machine Learning Consortium to Advance Health Equity and Researcher Diversity) US-based stakeholder listening sessions identified the need to engage communities locally, obtain buy-in for each population, and ensure algorithms are transparent and easily understood (Vishwanatha et al., 2023). Transparency in practices could go a long way—especially in community engagement—toward building capacity and readiness among those who industry needs as volunteers in medical product development. To facilitate this, patients should be provided with multilingual educational materials on how their data are used, who has access, and the short- and long-term benefits of participation and data sharing to optimize BDA outputs and health equity. Short-term benefits include improved patient trust and generalizability of clinical findings. Long-term benefits include greater innovation, improved access to effective medical interventions, reduced health disparities, and lower economic costs (Bibbins-Domingo and Helman, 2022b). This education could also highlight the different disease prevalences among different racial and ethnic groups for relevant immune diseases (Goonesekera et al., 2024), and the aims of health equity to ensure equitable access to care and outcomes. In parallel with initiatives aimed at improving patient engagement and reducing barriers to clinical trial participation, frameworks such as the Clinical trial Diversity Rating should be used to ensure that key stakeholders and regulatory bodies have the data and oversight needed to address remaining challenges (Agboola and Wright, 2024).

With the advent of natural language processing, AI can help improve the quality of patient educational materials by allowing near instant translation across multiple languages and by simplifying content to improve quality and readability, maintain or improve understandability, and improve actionability (Saatçi et al., 2024; Will et al., 2025). Table 2 presents a selection of case studies on the application of AI and BDA in immunology; however, while they have the potential to improve health equity, no evaluations were conducted—highlighting the need to track equity-related KPIs in future studies. For example, an ML model scouring EHR data for immune-driven traits has been used to identify patients in need of further testing—potentially accelerating diagnosis and treatment (Forrest et al., 2023), and achieving cost savings with earlier diagnoses (Wylezinski et al., 2019). This model also identified a high-risk subgroup that would likely be underdiagnosed based on a lack of testing (Forrest et al., 2023), which is especially useful given the high prevalence of misdiagnoses in immune diseases (Goonesekera et al., 2024). Additional positive examples of AI and BDA being applied to increase health equity in immune diseases, together with appropriate ways of assessing how effective the initiatives have been, would be helpful to increase community engagement.

While community engagement is essential to increase the diversity of data and ensure the damaging effects of bias can be identified and mitigated, robust data collection standards are needed to account for bias and employ tools that address physician biases in diagnosis and measurement of patient outcomes. To expand and plug information gaps, it is essential to collect race, ethnicity, sex, gender, and other social risk factor data from diverse sources and educate stakeholders on the importance of maintaining accurate and complete records. Reviewing supplier data procurement contracts and incorporating bias-handling clauses may help ensure that disparities are actively mitigated. To facilitate improved data quality, a steering committee or leadership structure could be established by thought leaders or a governmental organization, such as the HSS. This committee would inform data collection practices and create a gold standard for representativeness, enabling the assessment of underrepresented intersectional subpopulations within big data. Collaboration is also crucial for improving access to healthcare data and sharing it within diverse communities; there is enormous potential for sharing data within a nationwide database (Sabet et al., 2023).

To ensure RWD from multiple sources (e.g., claims, Centers for Medicare and Medicaid Services, EHR, and demographic data) are useful, it will be important to harmonize and standardize the information by implementing paperless systems, standardizing metrics, and building an infrastructure for sharing data. Universal standards, such as Health Level Seven—Fast Healthcare Interoperability Resources (HL7 FHIR), should be applied to provide a standardized way of formatting and exchanging healthcare data, making it easier for different systems (e.g., EHRs, apps, hospitals, and insurers) to communicate and share data consistently and securely (HL7 FHIR, 2023). However, there are barriers to implementation because of fragmented data, inconsistent coding practices, and interoperability gaps across institutions. Implementing FAIR data principles and adopting standardized vocabularies, such as those from Observational Health Data Sciences and Informatics, can address these issues by enabling consistent data integration, improving usability, and data quality (Pezoulas and Fotiadis, 2024). Data harmonization can, however, be operationally complex and costly due to the need to manage data privacy, consent, and compliance with regulations like the Health Insurance Portability and Accountability Act and EU General Data Protection Regulation (GDPR) (Pezoulas and Fotiadis, 2024).

Considering the complexities of data protection and regulation, robust governance is needed to ensure that these frameworks enhance the potential of AI and BDA to address health equity, rather than becoming a significant barrier to progress. In an article by Murdoch (2021), regulation was highlighted as a key issue due to EHR data being among the most private and protected forms of information. The article cautioned that regulation and oversight risk falling behind BDA and emphasized the need for technologically enabled methods of communicating and obtaining patient consent, as well as improved data protection and anonymization (Murdoch, 2021).

An example of the challenges faced when developing and accessing robust EHR data, especially at scale, is the digital transformation of the UK National Health Service (NHS), which is one of the largest employers in the world and received an annual budget exceeding £180 billion in 2023 (NHS Confederation, 2023). Despite its best efforts to go paperless with a digital EHR system, the NHS has failed to meet targets of 100% digitization by 2018–2024, with the last target of March 2026 scrapped and “no set date” now in its place (Clews, 2024; Lovell, 2025). While 90% of NHS trusts have achieved digitization—a notable accomplishment given the organization's scale—the process has been marked by significant delays attributed to difficulties in harmonizing data across primary and secondary care, the slow pace of digital adoption, and lack of fresh thinking and decisive action (Lovell, 2025)—all suggesting potential issues with leadership, strategy, and potentially technical and financial barriers. In a related incident, General Practitioner leaders voiced concerns about patient consent and data governance in an NHS-funded AI model designed to improve predictive healthcare related to COVID-19 vaccinations, ultimately leading to its termination (Colivicchi, 2025). It is therefore clear that, even with significant investment and time, there are issues to overcome regarding the handling and governance of data; however, such challenges are not insurmountable. For example, AI has also been piloted in the NHS to help identify patients who require proactive outreach to address the risk of non-attendance. This approach aims to help patients from marginalized communities get an appointment that works for them, and in doing so improves their outcomes, reduces health inequalities, and lowers costly inefficiencies stemming from missed appointments, all while respecting GDPR and protecting patient data (Deep Medical, 2025). However, as highlighted by Xiao et al. (2025), the potential of AI technologies remains limited by concerns around data privacy, the lack of data-sharing infrastructure, and inconsistent policies, which underscores the need for secure and shared data environments. A privacy-by-design approach has also been recommended by the European Alliance of Associations for Rheumatology (EULAR; formerly the European League Against Rheumatism) and other organizations to ensure that privacy and data protection are embedded at every stage, safeguarding patient information and enabling ethical, compliant, and trustworthy research (Gossec et al., 2020).

To avoid exacerbating existing biases, it is essential to engage data scientists and subject-matter experts collaboratively to ensure fairness in big data and AI-powered BDA (Boykin et al., 2021). Part of this is to improve the reliability and accuracy of data through systematic identification of data quality issues using anomaly detection techniques (Gaspar et al., 2011; Churová et al., 2021), along with addressing data blind spots (e.g., biased proxy variables) that perpetuate inequities (Obermeyer et al., 2019).

Detecting systemic and harmful biases in data, models, and outcomes is also critical (Schwartz et al., 2022). This involves understanding and addressing non-random reasons for missing data, implementing methods to identify non-equitable outcomes, and developing decision-support tools to detect biases in patient-generated data, such as EHRs, predictive models, and decision-making processes (Parikh et al., 2019). Fair ML approaches with RWD, including bias mitigation techniques with supervised models (Hardt et al., 2016; Huang et al., 2024), and techniques to detect and correct biases across intersectional subpopulations (Zhang and Neill, 2016; Kearns et al., 2018) should be utilized to mitigate bias. These techniques help identify metrics leading to equitable outcomes, and assess fairness at each step of the algorithm (Suresh and Guttag, 2021; Black et al., 2023).

As an example adapted from the fair ML literature, consider a case where an algorithmic decision-support tool is used to predict the progression of immune disease, and both false positives (incorrectly predicting an increase in severity) and false negatives (incorrectly predicting that severity will not increase) are harmful to patients. Given a concern that a specific protected class defined by a sensitive attribute, such as race or gender (e.g., female patients), are receiving lower-quality predictions, a typical approach (Barocas et al., 2023) is to compare false-positive and false-negative error rates for the protected and non-protected class and identify any statistically significant discrepancies. If the affected subpopulation is not known a priori, or there is a concern that the bias may be affecting subgroups defined by multiple data dimensions (e.g., older Black male patients), then techniques such as a bias scan (Zhang and Neill, 2016) can efficiently search across subgroups defined by multiple attributes (race, gender, age, etc.) and identify the subgroups with the most significant error rate imbalances. If biases are detected, approaches for mitigation include adjustment of decision thresholds to balance the error rates (Hardt et al., 2016), resampling or reweighting the data (Kamiran and Calders, 2012), and relearning of the predictive models with additional constraints or penalties to reduce error rate imbalance (Kamishima et al., 2012; Zafar et al., 2017). Alternatively, the detected biases may inform system-wide changes, such as increasing the amount and quality of data collected for population subgroups for whom the algorithmic decision-support tool is performing poorly.

Together, these approaches help to identify and mitigate bias and to develop fairer ML models by balancing error metrics across subgroups. However, even when accounting for these sources of bias, people can misuse data from algorithms in decision-making by discounting algorithmic recommendations in favor of their own judgment, showing tolerance for algorithmic errors, and struggling to evaluate algorithmic performance accurately. They may also be influenced by irrelevant information, trust inaccurate algorithms, and apply algorithms in ways they were not designed for (Green and Chen, 2019), exemplifying the need for increased participation and improved education of subject-matter experts to ensure that AI and BDA are not misused.