Responsible & Ethical AI Utilization in Social Determinants of Health

RAIS Journal for Social Sciences | VOL. 9, No. 2, 2025 ISSN 2574-0245 (Print) | ISSN 2574-1179 (Online) | DOI 10.5281/zenodo.17689075 Responsible & Ethical AI Utilization in Social Determinants of Health Paige COLEMAN1, Atif Farid MOHAMMAD2 1Healable Inc., Rockville, MD, USA, [email protected] 2Capitol Technology University, Laurel, MD, USA, [email protected] Abstract: The impact of SDoH (Social Determinants of Health) on population health in general is a significant area of concentration. There are biased determinants and there are risk determinants, which are not always very prominent. The utilization of Responsible AI allows us to deal with the aforementioned for the decision makers to pinpoint the lacks/gaps in healthcare provision. It is vital to provide relevant resources to the people who might have the resources, however, the required ones are not given priority. By doing this provision, the service providers include the whole care continuum. This research is beneficial for CMS (Centers for Medicare & Medicaid Services), as well as for standard population health. Our research presented in this paper shows tangible results. We have found in the application of AI/ML in the vast arena of SDoH. The application of Healable AI provides the insightful decision making for the associated stakeholders with the healthcare spectrum. Keywords: AI, Healable, Healthcare, ML, SDoH Introduction Social determinants of health have emerged as critical factors shaping population health outcomes, yet their complexity often obscures the most actionable intervention points for healthcare systems. Beyond the widely recognized determinants such as income and education, healthcare disparities arise from subtle biases and less prominent risk factors that traditional analytical approaches struggle to identify systematically. These hidden determinants create gaps in care delivery that disproportionately affect vulnerable populations, even when resources exist within the system but remain misallocated due to inadequate prioritization mechanisms. The integration of responsible artificial intelligence into population health management offers a transformative approach to addressing these challenges by enabling stakeholders to identify and respond to healthcare provision gaps with unprecedented precision. Machine learning algorithms can analyze vast datasets to uncover patterns in social determinants that human analysis might overlook, while responsible AI frameworks ensure these insights do not perpetuate existing biases or create new forms of discrimination. This capability is particularly valuable for organizations like the Centers for Medicare and Medicaid Services, which must allocate resources across diverse populations while ensuring equitable access to care throughout the entire care continuum from prevention through treatment and follow-up. This research demonstrates the tangible benefits of applying AI and machine learning methodologies to the complex landscape of social determinants of health. Through the implementation of explainable AI techniques, healthcare decision-makers gain actionable insights that connect resource availability with population needs, enabling more effective interventions that address both overt disparities and subtle gaps in service delivery. The findings presented in this paper illustrate how proposed algorithm application can transform population health management 512 from reactive problem-solving to proactive, data-driven resource allocation that serves both COLEMAN & MOHAMMAD: Responsible & Ethical AI Utilization in Social Determinants of Health 513 standard populations and those facing systematic barriers to care. State of the Art Social informatics is a subfield of medical informatics that enhances healthcare by bridging technology and social contexts to better integrate social and health information (Pantell et al., 2020). It transforms how we understand and respond to social data, leveraging techniques like artificial intelligence (AI) broadly defined as computational systems performing human-level cognitive functions (Mintz & Brodie, 2019). This includes natural language processing (NLP), a specialized area of AI that computationally analyzes human language, and machine learning (ML), which employs statistical methods to identify patterns and build predictive models from data (Harrison & Sidey-Gibbons, 2021). By effectively gathering and analyzing this information, social informatics aims to improve patient care and health outcomes (Dorr et al., 2019; Conway et al., 2019). AI technologies can harness the advantages of SDoH information to recognize what patients require (Patra et al., 2021; Lybarger et al., 2023) and address healthcare system strain and medical condition intricacy (D'Elia et al., 2022; Balicer & Cohen-Stavi, 2020). A prevalent form of AI called machine learning works by identifying, forecasting, and classifying results through pattern recognition in datasets linked to verified observations or established ground truth instances (Bearse, Mohammad, & Haque, 2021). While still at an early research phase, machine learning combined with natural language processing has shown promise in extracting various SDoH elements from electronic medical record information, including early life experiences (Singh, Mohammad, & Ur Rahim, 2025), social networks, housing circumstances, work status, and forecasting health results using SDoH factors. Proposition As cost and quality continue to shape the national conversation around healthcare, I believe we often overlook one of the most important factors in how people actually experience care the human mind. Behind every data point is a person trying to navigate complex emotions, environments, and responsibilities, and that navigation is directly tied to their sense of safety and stability. When we think about Maslow’s Hierarchy of Needs, it becomes clear that no one can move toward healing, self-care, or growth if their most basic needs are unmet. If a person does not feel safe, secure, fed, or rested, expecting them to focus on health management or long-term behavior change is simply unrealistic. These are not just abstract theories; they represent the real-world barriers that keep individuals from engaging with the healthcare system in a meaningful way. This is where I see artificial intelligence and machine learning playing a transformative role, not as replacements for human care, but as tools to enhance it. If we use AI responsibly, it can help identify patterns and gaps that humans alone might not see. By analyzing data tied to social determinants of health, AI can help uncover the underlying reasons why someone struggles to follow through with treatment, miss appointments, or manage chronic conditions. It can surface early warning signs that point to issues like food insecurity, housing instability, or emotional distress all before those challenges escalate into medical crises. But for this to be ethical and effective, we must design these systems with empathy at their core, ensuring that technology amplifies human understanding instead of replacing it. We have always believed that healthcare’s biggest challenge isn’t what happens in the office, but what happens between visits. A 15-minute conversation with a doctor can be powerful, but it can’t compete with the realities patients face every day when they go home. Responsible AI integration could help bridge this gap by extending support beyond those episodic moments connecting people to resources, community programs, or care coordinators who understand their personal circumstances. By embedding social and environmental awareness into the care process, we can make the system feel more human again one that listens, anticipates, and adapts to people’s RAIS Journal for Social Sciences | VOL. 9, NO. 2, 2025 514 real lives rather than expecting them to fit into rigid healthcare structures. If we bring these two ideas together, human psychology and responsible AI, we can finally begin to close the gap between patient behavior, cost, and quality. Healthcare should not be something that happens to people; it should evolve with them. When technology is guided by compassion and understanding, it can help individuals regain control over their health and rebuild confidence in the system meant to support them. The goal is not perfection but alignment creating a space where patients can move from surviving to thriving, from being managed to being empowered. That is where we believe the future of healthcare must go: one that honors both data and humanity equally. Algorithm The given algorithm depicts the Mathematical Formulation of SDOH Algorithm for Population P₅₅₊. Notation: •P₅₅₊: Population aged 55 and older •i ∈ P₅₅₊: Individual in the target population •SDOH = {d₁, d₂, ..., dₙ}: Set of n SDOH domains •where d₁ = income, d₂ = housing, d₃ = food security, d₄ = social isolation, d₅ = transportation, d₆ = access to care Step 1: Identification and Assessment 1.1 Screening Function: S: P₅₅₊ → {0,1}ⁿ S(i) = (s₁(i), s₂(i), ..., sₙ(i)) where sⱼ(i) = 1 if individual i screens positive for domain dⱼ, 0 otherwise 1.2 Assessment Severity: For individuals with positive screening: A: {i ∈ P₅₅₊ | ∃j: sⱼ(i) = 1} → ℝⁿ A(i) = (a₁(i), a₂(i), ..., aₙ(i)) where aⱼ(i) ∈ [0, 1] represents severity level for domain dⱼ 1.3 Data Collection: D = {(i, S(i), A(i), C(i)) | i ∈ P₅₅₊} where C(i) represents community-level context data for individual i Subject to constraints: •Privacy(D) ≥ θₚᵣᵢᵥₐcy (privacy threshold) •Ethics(D) = TRUE Step 2: Understanding Interplay 2.1 Correlation Analysis: ρⱼₖ = Corr(dⱼ, hₖ) where hₖ represents health outcome k, and ρⱼₖ measures correlation between SDOH domain j and health outcome k 2.2 Vulnerability Identification: Define vulnerable subgroups: V = {V₁, V₂, ..., Vₘ} ⊆ P₅₅₊ where Vₗ = {i ∈ P₅₅₊ | vulnerability_criterion_l(i) = TRUE} Risk score for subgroup Vₗ and domain dⱼ: R(Vₗ, dⱼ) = (1/|Vₗ|) Σᵢ∈Vₗ aⱼ(i) COLEMAN & MOHAMMAD: Responsible & Ethical AI Utilization in Social Determinants of Health 515 Step 3: Intervention and Resource Navigation 3.1 Intervention Mapping: I: P₅₅₊ × SDOH → 𝒫(Resources) I(i, dⱼ) = {rₖ ∈ Resources | match(i, dⱼ, rₖ) > τ} where 𝒫(Resources) is the power set of available resources and τ is a matching threshold 3.2 Intervention Components: Define intervention vector for individual i: Θ(i) = [θ₁(i), θ₂(i), θ₃(i), θ₄(i)]ᵀ where: •θ₁(i) = resource_connection_score(i) •θ₂(i) = community_program_participation(i) •θ₃(i) = advocacy_benefit_index(i) • θ₄(i) = care_coordination_level(i) 3.3 Optimization: Maximize overall impact: max Σᵢ∈P₅₅₊ Σⱼ₌₁ⁿ wⱼ · [aⱼ(i) - aⱼ'(i)] subject to: •Budget constraint: Σᵢ∈P₅₅₊ cost(Θ(i)) ≤ B •Resource constraints: utilization(rₖ) ≤ capacity(rₖ) ∀rₖ ∈ Resources where aⱼ'(i) is post-intervention severity and wⱼ are domain weights Step 4: Monitoring and Evaluation 4.1 Impact Tracking: Define impact function at time t: H(i,t) = f(SDOH(i,t), Θ(i,t)) where H represents health outcomes 4.2 Longitudinal Assessment: ΔH(i) = H(i, t₁) - H(i, t₀) ΔSDOH(i) = ||A(i, t₁) - A(i, t₀)|| 4.3 Feedback Integration: F: P₅₅₊ × ℝ → [-1, 1] F(i,t) = feedback_score(i,t) 4.4 Adaptive Updating: Θ(i, t+1) = Θ(i,t) + α · ∇E(i,t) where E(i,t) is effectiveness metric and α is learning rate Step 5: Collaboration and Partnerships 5.1 Stakeholder Network: G = (N, E) where: •N = {healthcare, social_services, government, community_orgs, ...} •E = {(nᵢ, nⱼ, wᵢⱼ) | nᵢ, nⱼ ∈ N, wᵢⱼ = collaboration_strength} 5.2 Coordination Efficiency: CE = (Σ(ᵢ,ⱼ)∈E wᵢⱼ · success(i,j)) / |E| 5.3 Community Engagement Index: CEI = (|{i ∈ P₅₅₊ | engaged(i) = TRUE}|) / |P₅₅₊| Target: CEI ≥ β (minimum engagement threshold) Overall Algorithm Objective: Maximize: Ψ = Σᵢ∈P₅₅₊ [w₁·ΔH(i) + w₂·ΔSDOH(i) + w₃·F(i) + w₄·CEI] RAIS Journal for Social Sciences | VOL. 9, NO. 2, 2025 516 Subject to: - S(i) performed ∀i ∈ P₅₅₊ - A(i) performed ∀i where Σⱼ sⱼ(i) > 0 - Privacy and ethical constraints satisfied - Budget ≤ B - CE ≥ γ (minimum coordination threshold) where w₁, w₂, w₃, w₄ are weights reflecting relative importance of health outcomes, SDOH improvement, satisfaction, and engagement. Conclusion The implementation of this mathematically formalized SDOH algorithm offers healthcare organizations a systematic, data-driven framework to address the multifaceted social needs of their aging patient population, ultimately leading to measurable improvements in both clinical outcomes and operational efficiency. By integrating routine screening, individualized assessment, and tailored interventions with continuous monitoring and multi-sectoral collaboration, healthcare systems can proactively identify and mitigate social barriers that contribute to poor health outcomes, preventable hospitalizations, and increased healthcare costs among individuals aged 55 and older. The algorithm's optimization function ensures resource allocation is maximized within budgetary constraints while maintaining high levels of patient engagement and care coordination, thereby reducing health disparities, enhancing patient satisfaction, and improving population health metrics. Furthermore, the adaptive nature of the framework allows healthcare organizations to continuously refine their interventions based on real-time feedback and outcome data, creating a sustainable, person-centered approach that not only addresses immediate social determinants but also builds long-term community partnerships and institutional capacity for holistic care delivery. This comprehensive strategy positions healthcare organizations to meet value-based care objectives, reduce total cost of care, and fulfill their commitment to health equity in an increasingly complex healthcare landscape. References Balicer, R. D., & Cohen-Stavi, C. (2020). Advancing healthcare through data-driven medicine and artificial intelligence. In B. Nordlinger, C. Villani, & D. Rus (Eds.), Healthcare and artificial intelligence (pp. 9– 15). Springer International Publishing. Bearse, P., Mohammad, A. F., & Haque, I. (2021). Artificial intelligence use in social determinants of health: A comparative study. In Proceedings of the 2021 World Congress in Computer Science, Computer Engineering, and Applied Computing (CSCE) (pp. 9). Springer Nature. Conway, M., Keyhani, S., Christensen, L., South, B. R., Vali, M., Walter, L. C., Mowery, D. L., Abdelrahman, S., & Chapman, W. W. (2019). Moonstone: A novel natural language processing system for inferring social risk from clinical narratives. Journal of Biomedical Semantics, 10(1), Article 6. D'Elia, A., Gabbay, M., Rodgers, S., Kierans, C., Jones, E., Durrani, I., Thomas, A., & Frith, L. (2022). Artificial intelligence and health inequities in primary care: A systematic scoping review and framework. Family Medicine and Community Health, 10(Suppl 1), Article e0001670. Dorr, D., Bejan, C. A., Pizzimenti, C., Singh, S., Storer, M., & Quinones, A. (2019). Identifying patients with significant problems related to social determinants of health with natural language processing. Studies in Health Technology and Informatics, 264, 1456–1457. Harrison, C. J., & Sidey-Gibbons, C. L. (2021). Machine learning in medicine: A practical introduction to natural language processing. BMC Medical Research Methodology, 21(1), 158. Lybarger, K., Dobbins, N. J., Long, R., Singh, A., Wedgeworth, P., Uzuner, O., & Yetisgen, M. (2023). Leveraging natural language processing to augment structured social determinants of health data in the electronic health record. Journal of the American Medical Informatics Association, 30(8), 1389–1397. Mintz, Y., & Brodie, R. (2019). Introduction to artificial intelligence in medicine. Minimally Invasive Therapy & Allied Technologies, 28(2), 73–81. Pantell, M. S., Adler-Milstein, J., Wang, M. D., Prather, A. A., Adler, N. E., & Gottlieb, L. M. (2020). A call for social informatics. Journal of the American Medical Informatics Association, 27(11), 1798–1801. COLEMAN & MOHAMMAD: Responsible & Ethical AI Utilization in Social Determinants of Health 517 Patra, B. G., Sharma, M. M., Vekaria, V., Adekkanattu, P., Patterson, O. V., Glicksberg, B. S., Lepow, L. A., & Ryu, E. (2021). Extracting social determinants of health from electronic health records using natural language processing: A systematic review. Journal of the American Medical Informatics Association, 28(12), 2716–2727. Singh, B., Mohammad, A. F., & Ur Rahim, M. A. B. (2025). Leveraging NLP and large language models for clinical documentation improvement: A medical AI chatbot approach. In R. Stahlbock (Ed.), Data science: 20th International Conference, ICDATA 2024, Held as part of the World Congress in Computer Science, Computer Engineering and Applied Computing, CSCE 2024, Las Vegas, NV, USA, July 22–25, 2024, revised selected papers (pp. 9). Springer Nature Switzerland.