Introduction

Social determinants of health (SDOH) are the conditions in which people are born, grow, learn, work, play, live, and age, and the wider set of structural factors shaping daily life conditions. These factors include social, economic, and legal policies and systems, and access to high-quality jobs, education, housing, transportation, food, healthcare, etc. SDOH have profound and sustained influences on health, functioning, and quality of life across diseases, conditions, and stages of life [1]. Different communities and populations have different SDOH exposures; some exposures are adverse and hamper health, while others are positive and thus promote health. Despite numerous calls to characterize, address, reduce, and eliminate health disparities, they persist and, in some cases, have widened largely due to the inattention to SDOH as the underlying causes of adverse health outcomes. SDOH cross multiple levels, temporal and spatial scales, and have complex and far-reaching direct and indirect consequences. Thus, traditional study methods alone are insufficient for characterizing SDOH effects, posing a significant challenge in understanding SDOH impacts on health.

Multi-scale computational modeling is a powerful set of approaches, methods, and tools that may help characterize and assess health disparities (HD) by quantifying the impact of complex SDOH interactions across scales. Computational modeling entails constructing computer-based representations of the components and processes comprising a complex system. Such modeling is multi-scale when these representations cross different spatial and temporal scales. For example, time scales can range from minutes to decades, while spatial scales can range from small (molecular or cellular scales) to large (environmental or societal scales). Although multi-scale models (MM) have been used to understand associations between adverse SDOH and health outcomes at the population scale [2,3], their potential to examine and assess the impact of SDOH on health and disease at the individual, biological scale remains largely untapped. A session at the Interagency Modeling and Analysis Group Multiscale Modeling Meeting held on the National Institutes of Health campus in Bethesda, Maryland, from June 28th to –29th, 2023, considered some potential ways MM can characterize the mechanistic biological impacts of SDOH and resulting HD. The session brought together MM and SDOH experts from academic and federal agencies to identify key opportunities and challenges to leverage MM to understand the underlying causes of poor health outcomes, which are often steeped in SDOH influences. This perspective summarizes and synthesizes the session discussions as a call to action for the SDOH and MM communities to collaborate and bridge scientific gaps to increase our understanding of SDOH impacts on individual and population health.

The impact of SDOH and resulting disparities are multi-scale issues

Factors that influence our health can span from the societal and community levels to the cellular and molecular levels. These factors are referred to as ‘multi-level’ in social and behavioral science, and ‘multi-scale’ in mathematical and computational modeling fields.

To fully understand the health impacts of SDOH and their influences at the biological (i.e., organ, tissue, cellular, and subcellular) scales, a multi-scale framework may be necessary. As we consider this framework, we need to recognize that disparities at the organ, tissue, and cellular levels stem from adverse SDOH reflective of living within health-inhibiting conditions and systems [4]. Thus, disparities in health outcomes are not purely associated with biological mechanisms, but rather reflect complex interactions between SDOH exposures and our biological responses to these SDOH across multiple scales.

Some examples of the multi-scale nature and analysis of SDOH from the societal to individual scales include an agent-based model (ABM) that quantifies the impact of neighborhood crime (community-scale) on leisure-time physical activity rates (individual behavior-scale) and obesity (physiology/biology-scale) among African American women in Washington, DC [5] and an ABM that simulates the impact of placing point-of-purchase sugar-sweetened beverage (SSB) warning labels as a policy measure (societal- and community-scales) on SSB purchases (individual behavior-scale) and subsequently adolescent overweight and obesity prevalence (physiology/biology-scale) [6].

Breast cancer provides an illustrative example of multi-scale SDOH-HD impacts from societal to molecular scales [7,8]. Although well-studied in biomedical research, many factors and interactions that drive breast cancer disparities remain unknown [9]. Starting at the highest level of influence [10], societal factors, including organization and quality of healthcare (e.g., time to treatment and follow up), healthcare policies and regulations (e.g., private insurance patients have better outcomes) [11], and residential segregation [12] are associated with breast cancer mortality. At the community level, neighborhood disadvantage affects the breast cancer stage at diagnosis and survival rates [12]. At the individual level, socio-economic status impacts breast cancer risk, time to diagnosis, and mortality [12]; and transportation access affects the time to treatment initiation [13]. At the organ/tissue level, SDOH are associated with more aggressive tumor biology [14–16] as well as elevated inflammation and macrophage recruitment to the tumor micro-environment in African American women compared to European American women [7]. At the cellular level, SDOH manifest as more tumor-permissive macrophages in Black women with breast cancer compared to non-Black women [17]. Finally, at the molecular level, the BRCA1/BRCA2 gene mutation is a risk factor mainly identified in European American women [7]. DNA methylation, strongly influenced by SDOH, namely environmental factors, is associated with cancer and is increased in African American women [18,19] and women living in neighborhoods with adverse social and environmental exposures [18,19]. In summary, the disparate breast cancer outcomes between Black and white women cannot be solely attributed to differences in ancestry, and there is increasing evidence of SDOH-HD as underlying causes of breast cancer incidence, progression, and survival [20,21]. These examples illustrate the multi-scale nature of SDOH on health and health disparities.

Challenges to characterizing and assessing the biological impact of SDOH

Table 1 lists some challenges to characterizing and assessing the biological health impacts of SDOH. One major challenge is that the biological effects of SDOH involve factors that are multi-scale and multifactorial. For example, social, educational, and economic conditions affect the ability of individuals to receive and seek care (individual level) [22], stemming from hindered access to quality education or jobs impacting health literacy and equitable care (population level). Further, the “social causation” hypothesis suggests that social interactions directly affect health outcomes and physiological function (organ or cell level) [23]. This provides evidence of a population-to-individual-to-organ level link that requires more comprehensive datasets and modeling infrastructures that can handle these nested processes. The time-lag between the initial social exposure and later biological effect (i.e., multiple temporal scales) is another challenge, as observable phenotypes may take months, years, or generations to manifest. Isolating each determinant can be difficult [24]. For instance, the “neuroimmune network hypothesis” [25] posits early-life adversity causes peripheral inflammation and increased neural stress, contributing to elevated insulin resistance, adiposity, and emotional health problems later in life. Early-life psychological stress causes dendritic atrophy in middle-aged animals [26], yet distinguishing whether other cofactors (e.g., chronic stress or diet) contribute to cognitive risk remains a challenge [27]. Early-life adverse SDOH can significantly disrupt key life-long biological and developmental pathways [28,29]. Thus, understanding the temporal impact of early-life adversity is difficult with human observational studies. Finally, SDOH is not isolated to one location, but can propagate to others. For example, individuals may relocate to seek better job opportunities and healthcare, which can improve or exacerbate SDOH in the new location. These complex interaction networks are difficult to quantify and disentangle with observational or experimental approaches alone.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Major challenges in characterizing and assessing the impact of SDOH on health and health outcomes.

https://doi.org/10.1371/journal.pcbi.1013284.t001

Second, there are significant challenges in obtaining unbiased and accurate data in SDOH research. For example, data bias can be found in ancestry (e.g., genomics), demographics, socioeconomics, or methodological issues (e.g., how disease is measured). Researchers may create bias unintentionally (54) and the range of scales and the limitless data that could be collected inevitably lead to gaps in the necessary data to assess the biological impact of SDOH. Metrics tend to be biased towards overrepresented groups and may be less useful or applicable to underrepresented groups. Thus, assessments of how SDOH manifest into physiological or biological observations may underestimate the impact on populations. MMs may overcome these obstacles by encoding physiological hypotheses about a person’s general health and then examining whether the data being provided are consistent or inconsistent with the modeling framework. Moreover, MMs may allow for in silico testing of whether interventions may correct an individual’s trajectory, and then validate these findings to follow up data. A benefit to the MM framework is the mechanisms of population, individual, or sub-individual physiology are encoded and available even when data is limited.

Finally, multiple disciplines and study/analysis methodologies (genetics, biology, physiology, epidemiology, etc.) historically tend to operate exclusively within one or a few scales. This isolation of scales hampers our ability to quantitatively assess the broader picture of how SDOH influences and affects multiple scales. This limitation can be overcome with MMs, which encode multiple scales within a single simulation framework and can be leveraged with limited data sources at single spatial or temporal scales.

How could multi-scale modeling help address these challenges?

For each challenge noted above, Table 1 lists ways MMs could help address these challenges. Many disciplines utilize MMs, including material science [30,31], fluid mechanics [32], biology [33], and epidemiology [34]—including the recent COVID-19 epidemic [35]. In application to SDOH-HD, researchers have used MM to assess US geographic variation in the factors associated with breast cancer mortality [2], evaluate the sociodemographic determinants of COVID-19 incidence rates in Oman [3], and develop an influenza forecasting tool at the state, regional, and national scales [34]. More recently, the study by Vodovotz and colleagues incorporated aspects of societal stress, inflammation, and intervention to address the multi-scale links in chronic stress [36]. The study provided interpretable links between external stimuli at a population level with their embedded biological processes at an individual level, key for MM for SDOH-HD. One of the key advantages of MMs is the ability to handle and integrate multimodal data. Data-driven models (i.e., top-down models) use dataset(s) with inputs and outputs of interest and define equations or algorithms to extract associations, correlations, and trends between certain factors and outcomes. These models can generate insights into complex relationships but do not necessarily capture cause and effect. In contrast, mechanistic models (i.e., bottom-up) aim to rebuild systems of interest and represent the mechanisms and causal pathways involved [37–39]. Traditional modeling approaches can be applied to MMs (i.e., sensitivity analyses) to identify key parameters to guide data collection or run “what-if” scenarios to generate testable hypotheses. These examples, and MM applications listed in Table 1, highlight the emergence and potential impact of MMs in SDOH-HD research.

What is needed to facilitate the use of multi-scale modeling (MM) in SDOH research?

With SDOH’s impact on HDs being inherently complex and multi-scale, the potential for MM to advance our understanding and ability to address them is immeasurable. To advance MM in SDOH research, many of the previously described challenges, like an underdeveloped workforce, scientific siloing, lack of collaborations with the local, non-academic community, and lack of funding for example will need to be addressed. Table 2 shows what is needed to facilitate the use of MMs to address SDOH along with specific ways that they can be done.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. What is needed to facilitate the use of multi-scale modeling in SDOH research?.

https://doi.org/10.1371/journal.pcbi.1013284.t002

The future of MM in SDOH research?

The use of MM to assess SDOH is at a critical inflection point. More data are available across different scales, including big data, especially with the advent of different wearables and data from social media and other Internet use [41]. The availability of sophisticated modeling approaches continues to grow as the available computational power enables the development of larger and more complex models and faster simulation times. Together, the use and value of MM in assessing SDOH are poised to grow by leaps and bounds.

Additionally, new developments are on the horizon. Artificial intelligence approaches can interface with and further expand the capabilities of MM. For example, a machine learning technique, physics-informed neural networks (PINNs), can expedite simulation time for expensive MM [42], eliminating the computational bottleneck often ascribed to MM. Other machine learning techniques can help identify missing biophysics processes in mechanistic models, providing a synergistic approach for machine learning-enhanced MM [43]. As always, artificial intelligence and machine learning should be used thoughtfully and with caution to avoid embedding and amplifying systemic biases [44–47].

Finally, effective communication and dissemination across various digital and multi-media platforms will improve MM access, understanding, and use. Developing user-friendly interactive interfaces and descriptions of such models can provide access to a wider range of users. Enhancing the accessibility of MM to multidisciplinary researchers could bolster its use for assessing SDOH. While assessing SDOH will be a great accomplishment, the goal is to use these SDOH assessments to enhance and scale up interventions, policies, and programs that reduce and someday eliminate health disparities. To that end, collaborations across fields/expertise areas will be required to make the necessary policy, law, educational, income, etc. changes effective and equitable.

Acknowledgments

Disclaimer: The findings of this article are those of the authors. They do not necessarily reflect the views of the National Institute of Nursing Research or the National Institutes of Health.