Enhanced health surveillance systems for chronic disease support could mitigate factors that contribute to the incline of morbidity and mortality of diseases such as obesity. Obesity is linked to increased overall mortality and has reached pandemic proportions, being responsible for approximately 2.8 million deaths annually [1,2]. Obesity represents an excessive and abnormal accumulation of body fat, which leads to adverse health effects that impose a health and financial toll on individuals and society [2]. More than half of the US population has at least one chronic condition, and 27% are living with multimorbidity [3]. These conditions cause more than 1.7 million deaths per year in the United States, where obesity is associated with the top leading causes of death (eg, diabetes, heart disease, stroke, and cancer) [4].

Neighborhood factors such as socioenvironmental determinants of health (SDoH) significantly contribute to these statistics [5-8]. Implementation of an intelligent health surveillance platform that incorporates SDoH can improve preparedness, prevention, and management of this obesity pandemic by assisting in the implementation of effective treatment and interventions.

Health surveillance involves the “ongoing systematic collection, analysis, and interpretation of data essential to the planning, implementation, and evaluation of public health practice, closely integrated with the timely dissemination of these data to those who need to know” [9]. Researchers have aimed to develop chronic health surveillance systems to assist in chronic health decision-making [10-16]. The World Health Organization (WHO) developed a conceptual framework for an Urban Public Health Observatory (UPHO) comprised of 3 domains: mission, governance, and knowledge and intelligence, the latter of which incorporates a data management component [17]. This framework from WHO therefore provides a strategic model for health surveillance.

Many current digital health solutions and electronic health record (EHR) systems lack the ability to incorporate machine learning algorithms into their decision-making process, and even if they do, the algorithms used do not have appropriate capabilities to explain the suggested decisions and actions to human users [18]. Machine learning approaches, so-called black-box statistics, should be trustworthy, transparent, interpretable, and explainable when making decisions in the clinical or health science setting [18-20]. A system’s explanation constitutes its interpretability [18,20-22]. Explainable AI (XAI) increases the intelligence delivered to the user by providing explanations, thereby enhancing the interpretability of outcomes and findings. Researcher efforts have been shifting toward applying algorithms that can aid in explaining the results of machine learning models. For instance, the SHAP (Shapley Additive Explanations) analysis [23] is an approach that assigns each model feature an importance score for making a particular prediction. Compared to traditional feature importance analyses, the novelty of SHAP lies in its ability to assess importance at the individual patient level. In this paper, we propose a novel approach to explainability that uses knowledge graphs as a semantic infrastructure explainable by design and enriches those graphs with results from machine learning algorithms as metrics and scores. The semantic causal relationships on the graph provide contextual knowledge around a population, and the metrics support those relationships, which provides 2 levels of evidence: knowledge level and statistical level.

We implement a UPHO platform as a knowledge-based surveillance system that provides better insight to improve decision-making by incorporating SDoH and providing XAI and interpretability functions [24]. Our UPHO consists of 3 layers: data, analytics, and application. In this work, we refine the initial design by incorporating data management, knowledge, and intelligence domains (Figure 1) that are in alignment with the conceptual model by WHO and a focus on the semantics layer.

The objectives of this article are to (1) expand UPHO by incorporating a semantics layer, (2) cohesively employ machine learning and semantic/logical inference to provide measurable evidence and detect pathways that lead to undesirable health outcomes, (3) provide clinical case scenarios and design case studies on identifying SDoH associated with obesity prevalence, and (4) provide a dashboard design that demonstrates the use of UPHO in the context of obesity, using the provided case scenario.

Figure 1 shows the expansion of the UPHO to incorporate the semantics layer. In the following section, we provide a detailed description of the UPHO platform expansion design.

The data management domain comprises the data layer. The UPHO collects population-level health and SDoH data and individual-level clinical and demographic data from EHRs through regional registries.

To obtain population-level health data, we used the US Centers for Control and Prevention (CDC) 500 Cities Behavior Risk Factors Surveillance System, which includes data regarding chronic diseases and their behavioral risk factors [25]. These variables are model-based estimates of crude prevalence among adults aged ≥18 years in 2018. We extracted variables pertaining to obesity, lack of physical activity, lack of insurance, and diabetes mellitus at the census tract level.

We extracted population-level SDoH variables that pertain to food insecurity, transportation, and socioeconomic stability at zip code, census tract, census block, and census block group levels from the US Census Bureau 2018 American Community Survey [26] and the US Department of Agriculture Research Atlas [27].

The UPHO platform can be used as a basis to develop several applications, some of which we have already developed, including dashboards [24], mobile health (mHealth) apps [34], digital assistants, and recommender systems [35]. In this article, we leverage the UPHO to implement a dashboard for real-time surveillance. By accessing dynamic knowledge discovered through the UPHO, the dashboard can provide real-time early warnings that are based both on content and context. The platform is accessible to policymakers, physicians, researchers, public health officials, and the public.

The following sections present 2 clinical case scenarios that focus on a physician and a researcher as users to demonstrate the methodology used in the knowledge and intelligence domain layers and the corresponding dashboard design in the application.

Scenario 1: A physician seeks an effective intervention for an adult African American patient diagnosed with obesity. The physician focuses on how SDoH in the patient’s neighborhood can influence the doctor’s management plans.

Scenario 2: A researcher investigating the impact of SDoH on obesity seeks an effective intervention for the adult obese populations in Memphis, TN.

We trained a machine learning–based support vector regression (SVR) machine model [36,37]. We linked CDC 500 Cities population-level obesity and behavior data [25] to the population-level social DoH [26,27] data set at the census tract level in Memphis, TN. We analyzed 8 features and used a Spearman rank test to assess the positive or negative relationship between each feature. We used a variance inflation factor (VIF) to detect multicollinearity between features. To examine patient neighborhood-level exposure, we used SHAP analysis. Table 1 shows the summary statistics for features considered for this study. We trained our support vector model on the 85% of randomly selected training data and tested the model on the 15% of remaining data to ensure the generalizability of the model, and we applied the linear kernel function. We scaled our data to have a mean of zero and standard deviation of one. We applied the grid search optimization method to seek optimal hyperparameters to improve model performance using the Caret package in R software (R Foundation for Statistical Computing) [38]. In addition, to avoid overfitting 5-fold cross-validation was applied to the training data set. We used root mean square error (RMSE) and R² to evaluate the performance of the model.

We followed the following ordered steps to generate the semantics layer knowledge graph from concepts defined in our domain ontologies.

No ethics review board assessment was required for this study because we used publicly available data.

The significant Spearman rank coefficient and VIF of the 7 features included in this study are shown in Table 2. Any feature exhibiting a VIF of greater than 10 was removed. For the SVR model results, we obtained an RMSE of 0.312 for the training set and 0.203 for the test set, while the R² for the training set was 0.91 and that for the testing data set was 0.95 . Since the model provides similar results for training and test data sets, the proposed model does not overfit. The SVR feature importance results range on a scale of 0 to 100, and the greater the score, the most important the feature (Table 3). We found that the percentage of the population lacking physical activity, percentage of population below poverty level, percentage of population without high school diploma, percentage of population unemployed, and percentage of Black population were the most important variables when predicting obesity prevalence in Memphis, TN.

Figure 4 shows SHAP’s value plot of feature contribution at the patient neighborhood level (census tract: 10300), which indicates the most important features such as the percentage of the population that lack physical activity and the percentage of population below the poverty level, from the point of view of the prediction of obesity prevalence in the patient neighborhood. The lack of physical activity and poverty had the largest positive (increased) contributions to obesity prevalence. On the other hand, the population of low income and more than a half-mile from the supermarket showed a negative (decreased) contribution but was the least important variable when predicting the patient neighborhood obesity prevalence. The knowledge extracted from our analysis will be used to detect the obesity prevalence pathways, which are defined by the top 5 most important features.

The graph part of the dashboard can serve as a tool for researchers and physicians to semantically explain the recommendations that we made about a specific patient or population. The current version of the graph provides 2 different visual cues, as follows.

UPHO’s metrics can be implemented into the backend of EHR systems (eg, Epic), and the results of those metrics can be rendered on the EHR interface in the form of risk scores on dashboards with severity indicators based on thresholds. Physicians can examine these metrics at the population level or individual patient level. UPHO can alternatively be used in a standalone approach by allowing a physician to extract more details about a single patient by providing the patient’s address or a population of patients by providing their city, state, or county. The input is coded to a geographical level of granularity that can be aligned with the population-level data to gain insights into the patient’s environment.

Previous studies provided evidence that socially disadvantaged communities are disproportionately affected by chronic diseases such as obesity [5-8], which is a risk factor for developing diabetes, heart disease, and cancer. The significance of UPHO lies in its ability to provide a multifaceted surveillance system design that serves as an apparatus for actualizing effective interventions, addressing concerns in health disparities, providing awareness to the public, and equipping health officials with a surveillance system that will improve population health decision-making and planning [24]. Using the semantics layer, the UPHO platform provides contextual knowledge by reusing several ontologies focusing on public health (eg, diseases, transportation, geography).

The incorporation of semantics provides the user with an additional layer of explainability and interpretability, which could decrease errors in intervention or treatment due to misinterpretation or misunderstanding. The semantics layer can also use ontologies to overcome the challenge of scattered data sources, thereby assisting in the achievement of interoperability, which will be used to maintain features such as data integration, XAI, and interpretability. We apply logical reasoners to extract and supply knowledge despite limited data.

Similar chronic disease surveillance systems [10-16] have offered approaches to assist in the efforts of improving chronic disease surveillance. Several published systems do not incorporate SDoH data [10-14]. One did not provide an implementation of the systematic framework [16] and several of them did not include XAI as a feature.

We followed the conceptual framework for UPHOs [17] and sought to improve the quality of disease surveillance by incorporating advances in AI and Bog Data, including interactive dashboard design, explainability, data integration, and interoperability, and incorporation of multimodal SDoH data. Developing a multidimensional scalable surveillance system to monitor and detect trends and deliver rapid early warnings and recommendations could assist health officials, physicians, and researchers in mitigating a health crisis such as the ongoing COVID-19 pandemic [24].

One of the major limitations of UPHO is that it collects population data so that neighborhood or population assumptions are made for an individual in a clinical setting. For instance, individuals or patients who live in a particular population or neighborhood might not have the same characteristics as other individuals residing in the same neighborhood or population. However, our platform provides an end-to-end approach to examining the environment one resides and incorporates information that is important for the implementation of effective interventions for a given disease.

The future work will be focusing on the further development of the UPHO platform, so it can enable timely, insight-driven decisions and inform immediate or long-term health policy responses [15] to current and future public health crises.

This study leveraged semantic technology and presented a proof-of-concept prototype design for our knowledge-based surveillance system, UPHO, which aims to reduce health disparities and improve population health. The expanded feature incorporates another level of interpretable knowledge needed to inform physicians, researchers, and health officials’ decision-making process at the community level. Incorporating XAI helps with the explainability and interpretability of the relevant data, information, and knowledge. Users who are not equipped with domain knowledge could extract common sense knowledge from a system that incorporates XAI [35]. We as humans need a clear visualization and understanding of relationships between parameters in a system to make informed decisions. The lack of understandability and explainability in the health care and public health domains often leads to poor transparency, lack of accountability, and ultimately lower quality of care and biased health policies [39]. Thus, the incorporation of semantics and XAI can improve fairness, accountability, transparency, and trust in health care and public health.