The Impact of Slavery Legacy on the Ecological Framework of Health: A Lasting Power Dynamic Impacting Present-Day Health

Abstract

Drivers of population-level health in the United States (U.S.) are complex. This analysis refines an ecologic framework of health, employing artificial intelligence modeling to estimate the impact of slavery on present-day, population-level lifespan. This study utilized several U.S. county-level datasets with more than 40 predictive variables representing the ecological framework of health. A non-linear artificial intelligence statistical approach was used to assess the ability of these variables to predict county-level life expectancy, years of life lost and death rate. The R² values demonstrated that the overall predictive performance of the models for life expectancy, years of life lost and death rates were all very strong, with mean R² ≥ 0.71 in all 3 prediction models. The number of predictor variables retained in the models ranged from 31 to 46, with measures from each category of the ecological framework of health being retained. The historical prevalence of slavery on a county level was a significant, repetitive interactive term. Unhealthy lifestyle behaviors are a primary driver of chronic disease, ultimately leading to diminished quality of life and lifespan. These health challenges are not insurmountable if the true root causes and forcing factors of health are acknowledged, studied, and addressed.

“The order of variable importance clearly demonstrates the complexity of U.S. health in the context of chronic disease and shortened life expectancy.”

Introduction

The drivers of population-level health trajectories in the United States (U.S.) are complex, with multiple factors interacting in ways yet to be fully appreciated.^1,2 Traditionally, in the context of the U.S. chronic disease crisis, the path to decreased life expectancy arising from one or more chronic disease diagnoses begins from the lens of unhealthy lifestyle behaviors. This approach fails to appreciate the upstream forcing factors of unhealthy lifestyle behaviors, population factors that drive these behaviors and subsequently lead to increased risk for and diagnosis of chronic disease. Pronk et al ² has proposed an ecological framework of health, identifying culture, politics, policy, and social, physical, and economic environments as upstream forcing factors for unhealthy lifestyle behaviors, chronic disease risk factors and diagnosis, and ultimately a decreased life expectancy (please see figure from previous publication). The framework was subsequently validated by artificial intelligence (AI) analyses, demonstrating the importance of the identified forcing factors for population health.^1,3

The ecological framework of health also posits that power dynamics interact with the model and significantly impact health outcomes. Power dynamics characterize the “way different people or different groups of people interact with each other and where one of these sides is more powerful than the other.” ⁴ In this context, structural racism, defined as “the totality of ways by which intrinsically linked and mutually reinforcing cultural domains and social/political institutions work in concert to disenfranchise racially/ethnically marginalized populations,” ⁵ may be considered an important power dynamic to consider in the ecological framework. There is a longstanding and growing recognition of the detrimental impact of structural racism on health outcomes, particularly in black individuals and communities in the U.S.^6-9 Moreover, legacies are created when events from the past are so powerful that they continue to impact the environment of the present day. As such, the legacy of slavery in the U.S. created a power dynamic whose effects are still felt today in the context of structural racism on numerous fronts, including health disparities.^7,10-12

The current analysis embarks on an expanded exploration of the ecologic framework of health, employing AI modeling to determine the interactive impact of the legacy of slavery as a present-day power dynamic on the model. The primary hypothesis of this analysis is the power dynamic created by the legacy of slavery will significantly interact with the ecological framework of health and shed further light on the predictors of U.S. lifespan.

Methods

This analysis utilized U.S. county-level datasets to achieve the study objectives. Table 1 describes source data for independent and dependent measures used in the current study. Source data were linked through zip-code identifiers to create a merged dataset for analysis. Apart from the American Nations ¹³ dataset, all other datasets are publicly available. The defining characteristics of the four-level American Nations model listed in Table 1 are defined as follows: Aggressively communitarian regions have built strong institutions, social services, and regulatory environments paid for by higher taxes on wealth, income, property and business, while the passively communitarian ones did so to a lesser degree. The aggressively individualistic regions have fewer and weaker institutions, regulatory regimes, and taxes and provide markedly fewer public services. Passively individualistic regions are less rigid on these matters but still seek to have a lean public sector.

Table 1.

Source Data for Independent and Dependent Variables.

Independent Variable Grouping	Specific Measures	Source
Culture	American Nations Model 4 level grouping: • Aggressively Individualistic (Deep South and Greater Appalachia) • Passively Individualistic (Far West) • Passively Communitarian (Midlands, El Norte) • Aggressively Communitarian (Yankeedom, New Netherland, Left Coast, and First Nation.)	https://www.nationhoodlab.org/the-american-nations-and-the-50-states/
Politics	2022 MRP Ideology Index Percentage of citizen population aged 18 or older who voted in the 2020 U.S. Presidential election.	https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/BQKU4M https://www.countyhealthrankings.org/health-data/methodology-and-sources/data-documentation
Policies	Gross Domestic Product 2019-2022 Percentage of all households that self-responded to the 2020 census 2020/2021 Civic Opportunity Index	https://www.bea.gov/data/gdp/gdp-county-metro-and-other-areas https://www.countyhealthrankings.org/health-data/methodology-and-sources/data-documentation https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/TCXRTM
Social, Physical and Economic Environments	Social Vulnerability Index Subtheme Scores 1-4 (2022 Release) 2022 Social Association Rate: Number of membership associations per 10 000 population	https://www.atsdr.cdc.gov/place-health/php/svi/index.html https://www.countyhealthrankings.org/
Behaviors	2022 Age Adjusted Prevalence of No Leisure-Time Physical Activity, Smoking, Bing Drinking, and Short Sleep Duration (2024 Release)	https://www.cdc.gov/places/
Health Risk Factors	Prevalence of Obesity, Diabetes, Hypertension, High Cholesterol, Depression, Arthritis, Cognitive Disability, Any Disability, Frequent Mental and Physical Distress, and Fair to Poor Health Rating (2024 Release)	https://www.cdc.gov/places/
Chronic Disease	Prevalence of Coronary Heart Disease, Stroke, Cancer, and Chronic Obstructive Pulmonary Disease (2024 Release)	https://www.cdc.gov/places/
Power Dynamic	Census of the Enslaved Population: 1860 Number of slaves in 1860 per 100 000 of current county-level population calculated	https://usa.ipums.org/usa/slave/slave.shtml
Dependent Variable Grouping	Specific Measures	Source
Outcomes	Life expectancy, age-adjusted premature mortality, and years of potential life lost before age 75 per 100 000 population - 2025 County Health Rankings Release	https://www.countyhealthrankings.org/

Subject Protection

HealthPartners Institute Research Subjects Protection Program determined that this study is exempt from IRB review and ongoing oversight under 45 CFR Part 46 as it involves the analysis of existing, publicly available data sets.

Machine Learning

Preprocessing

A series of essential preprocessing steps was conducted on the dataset prior to model training. These preprocessing procedures included the handling of missing data, encoding categorical variables, and scaling numerical features (i.e., predictors).

Missing values were detected in both predictors and outcomes. Since our sample size can be considered sufficient (i.e., observations per feature >40 for all outcomes), we dropped the sample with any missing values in either predictors or outcomes using dropna() function in Pandas, leading to the final sample size of 2964 for life expectancy, 2807 for years of life lost rate, and 2982 for death rate.

For the predictor SLAVES PER 100K, we convert it to a categorical variable (Slavery), with 3 county-level categories: 1 = no slavery (value = 0), 2 = low prevalence (non-zero values < median), and 3 = high prevalence (non-zero values ≥ median). Approximately two-thirds of the samples (i.e., counties) had no history of slavery; hence, the median value was calculated from non-zero values. The percentage of each level can be seen in Figure 1. Since the focus of this study is to examine the interaction between the slavery-related predictor and other variables, we assigned the “no slavery” group a category value of 1, rather than 0, to avoid misleading the model during interaction term generation. Interaction between Slavery and other features were calculated, increasing the total features from 36 to 71. For all predictors including the interaction, we employed the StandardScaler technique to ensure uniformity in data scale, which is essential for optimizing model performance.

Figure 1.

The sample size for each level in the slave category predictor.

Model Selection

To determine the best-performing regression model for all outcome prediction, we evaluated 6 widely used machine learning regressors: Support Vector Machine (SVM) regressor, Adaboost regressor, Random Forest regressor, XGBoost regressor, LightGBM regressor, and Extra trees regressor^14-20 using all the 71 features. Our preliminary results showed that Extra Trees Regressor and LightGBM demonstrated the best performance in terms of lowest mean absolute error (MAE). However, Extra Trees Regressor exhibited signs of overfitting, with a larger discrepancy between training and test performance. Therefore, LightGBM was selected in this study. LightGBM is a fast, efficient, and high-performance gradient boosting framework based on decision tree algorithms. It is widely used for tasks such as ranking, classification, and regression. Compared to other gradient boosting models like XGBoost, LightGBM offers superior speed, lower memory usage, and often better accuracy. Moreover, LightGBM is an interpretable ML model, which could quantify the importance of the input variables.

Model Training

The dataset was initially split into a training set (70%) and a test set (30%). The training set was further divided into a training subset (80%) and a validation subset (20%) using a 5-fold cross-validation method (cv = 5). Hyperparameter optimization for the LightGBM was performed using randomized search with cross-validation, tuned parameters include number of trees, maximum depth of trees, learning rate, number of leaves, L1 regularization, subsampling of data instances, and subsample ratio of columns. The last 3 parameters could help to prevent overfitting. Each model was trained using a backward feature selection approach. Starting with all 71 features, we iteratively removed the 5 features with the lowest importance scores until only 6 remained. This backward elimination strategy allows the model to retain the most predictive features while progressively reducing dimensionality, helping to improve model generalizability and interpretability. To enhance the robustness of our results, each model was trained 4 times using different random state values for train-test splitting (0, 1, 2, and the default one 42), allowing us to account for performance variability related to data partitioning.

Model Evaluation

To evaluate model performance, MAE, root mean square error (RMSE), and determination coefficient (R²), were calculated. The equations for computing MAE, and RMSE are presented below:

$$MAE=\frac{1}{n}{\displaystyle \sum }_{i=1}^n\left|y_{pred,i}-y_i\right|$$

$$RMSE=\sqrt{\frac{1}{n}{\displaystyle \sum }_{i=1}^n{\left(y_{pred,i}-y_i\right)}^2}$$

Where y _pred,i is the predicted value for the i-th observation, y _i is the actual value, and n is the total number of samples. In this work, MAE is used as the primary metric to evaluate and compare model performance across feature sets (6 to 71 with interval of 5), and R² was calculated to facilitate comparisons across models, as it is independent of the value ranges and distribution of the target variable. The optimal model across different number of features was selected using the feature–MAE curve. A lower MAE indicates better predictive performance, allowing us to identify the optimal feature that achieved the highest accuracy and lowest complexity.

Results

Table 2 lists the number of U.S. counties with data available to assess outcomes of interest and descriptive statistics for each dependent variable.

Table 2.

Detailed Statistics of Outcomes.

	Life Expectancy	Years of Potential Life Lost Rate	Age-Adjusted Death Rate
Original Sample Size	3075	2912	3095
Available Sample (No Missing)	2964	2807	2982
Mean	75.14	10 495.32	499.77
SD	3.61	3786.85	161.87
Min	53.98	3315.25	139.76
Median	75.27	9941.42	479.85
Max	94.22	46 417.85	1948.61

The feature-MAE curves demonstrated that the performance of LightGBM was affected by the number of input features, increasing the number of features led to reduction in MAE, indicating improved model performance (Figure 2). However, there is an elbow point for each model, after which adding more features resulted in minimal improvement or even worse performance, hence the model at the elbow point was selected as the optimal one. For life expectancy, the optimal model included 31 features, the best-performing model for years of life lost rate used 46 features, and the 38-feature model was identified as the optimal model for death rate.

Figure 2.

The changes in MAE with different number of features for: (A) life expectancy; (B) year of life lost rate; and (C) death rate prediction. Mean and SD values for each feature were calculated across models with different random state values for train-test splitting (0, 1, 2, and the default one 42), the optimal model for each outcome was highlighted in red.

The R² values demonstrated that the overall performance of death rate prediction model was the best with highest R², followed by life expectancy prediction model, and the years of life lost rate showed the lowest R² (Table 3). Nevertheless, all 3 optimal models demonstrated strong predictive capability, with average R² ≥ 0.71. Similar findings are demonstrated in Figure 3, which compares the predicted and actual values based on the optimal models. Life lost prediction models showed the greatest discrepancies between the predicted and actual values, especially when the life lost rate is extremely large (i.e., >200 000).

Table 3.

Model Metrics for all Prediction Models, and the Model With the Highest R² are Bold.

Model		Life Expectancy	Year of Life Lost Rate	Death Rate
Feature #		31	46	36
R 2	Mean	0.732	0.716	0.753
	SD	0.035	0.034	0.033
	Max	0.773	0.745	0.778
	Min	0.695	0.674	0.705
MAE	Mean	1.329	1386.6	55.94
	SD	0.021	46.8	1.33
	Max	1.349	1441.0	57.42
	Min	1.303	1327.0	54.23
RMSE	Mean	1.847	2056.7	79.41
	SD	0.115	113.3	6.28
	Max	1.950	2159.2	88.53
	Min	1.712	1954.2	74.14

Figure 3.

Comparison of actual outcomes and predicted outcomes for the best-performing model of: (A) life expectancy with 31 features; (B) year of life lost rate with 46 features; and (C) death rate prediction with 36 features.

The contribution of each predictor to the performance of the optimal prediction models is listed in Table 4. Although the optimal models have different outcomes and different number of input features, their feature importances were highly consistent. Notably, 9 out of the top 10 predictors were the same across these models. Among these common predictors, only one was an interaction term (interaction between Slavery and Census participants), which ranked 7th in both life expectancy and life lost rate prediction models and ranked 9th in the death rate prediction model. The remaining common predictors included Social Vulnerability Index-SubTheme2, Census participants, Arthritis, Voter Turnout, High blook pressure, Social Vulnerability Indext-SubTheme1, smoking, and non-Hispanic black population. Among them, Social Vulnerability Index-SubTheme2 and Census participation were the most important predictors for all the models. For life expectancy and death rate prediction, Social Vulnerability Index-SubTheme2 had the greatest impact (importance = 106 and 105), In contrast, for life lost rate prediction, Census participation had the greatest impact (importance = 97), and it ranked 2nd and 3rd in other 2 models. For other interaction terms, such as the interaction between Slavery and Gross Domestic Product, consistently ranked around 20^th across all models, suggesting a modest but stable contribution to prediction performance.

Table 4.

Top 30 Importance (Imp) of Predictor Variables for Life Expectancy, Year of Life Lost Rate, and Death Rate. All Interaction Terms Were Bold.

#	Life Expectancy		Year of Life Lost Rate		Death Rate
	Feature (n = 31)	imp	Feature (n = 46)	imp	Feature (n = 36)	imp
1	SVI-Subtheme 2	106	Census Participation	97	SVI-Subtheme 2	105
2	Arthritis Prevalence	99	Stroke Prevalence	92	Census Participation	79
3	Census Participation	91	Arthritis Prevalence	86	Arthritis Prevalence	78
4	Voter Turnout	86	Smoking Prevalence	79	Voter Turnout	77
5	High BP Prevalence	76	SVI-Subtheme 2	75	Smoking Prevalence	74
6	SVI-Subtheme 1	73	Voter Turnout	69	High BP Prevalence	69
7	Slavery*Census Participation	71	Slavery*Census Participation	62	Stroke Prevalence	68
8	GDP-PERCH 2021-22	68	SVI-Subtheme 1	60	SVI-Subtheme 1	62
9	Smoking Prevalence	68	High BP Prevalence	58	Slavery *Census Participation	58
10	Non-Hispanic Black Population	64	Non-Hispanic Black Population	53	Non-Hispanic Black Population	57
11	SVI-Subtheme 1	59	Frequent Physical Distress	48	Frequent Physical Distress	49
12	GDP-PERCH 2019-20	59	Depression Prevalence	48	GDP-PERCH 2021-22	46
13	MRP Ideology Index	55	SVI-Subtheme 3	43	MRP Ideology Index	44
14	Civic Opportunity Index	55	SVI-Subtheme 4	42	High Cholesterol Prevalence	43
15	Depression Prevalence	53	GDP-PERCH 2021-22	41	SVI-Subtheme 4	43
16	Stroke Prevalence	52	Frequent Mental Distress	41	American Nations 4 Level Grouping	43
17	Social Association Rate	51	Cancer Prevalence	40	No Leisure Time Physical Activity Prevalence	42
18	Self-Rated Fair to Poor Health Prevalence	50	Coronary Heart Disease Prevalence	35	Cognitive Disability Prevalence	41
19	GDP-PERCH 2020-21	46	Slavery*Social Association Rate	34	Cancer Prevalence	39
20	Slavery*GDP-PERCH 2020-21	46	Slavery*GDP-PERCH 2019-20	33	COPD Prevalence	38
21	SVI-Subtheme 4	40	No Leisure Time Physical Activity Prevalence	32	Slavery*Binge drinking prevalence	37
22	Slavery*GDP-PERCH 2019-20	36	COPD	32	Slavery*GDP-PERCH 2019-20	37
23	No Leisure Time Physical Activity Prevalence	34	Disability Prevalence	31	GDP-PERCH 2019-20	36
24	High Cholesterol Prevalence	33	Slavery*American Nations 4 Level Grouping	30	Depression Prevalence	34
25	Slavery*Civic Opportunity Index	32	Slavery*Voter Turnout	29	SVI-Subtheme 3	33
26	American Nations 4 Level Grouping	28	Slavery*GDP-PERCH 2020-21	29	Social Association Rate	31
27	Frequent Physical Distress	28	MRP Ideology Index	29	Frequent Mental Distress	29
28	Binge Drinking Prevalence	23	High Cholesterol Prevalence	28	Disability Prevalence	28
29	Slavery *High Cholesterol Prevalence	23	GDP-PERCH 2019-20	26	Coronary Heart Disease Prevalence	27
30	Slavery *Cognitive Disability Prevalence	20	Civic Opportunity Index	26	Poor Sleep Prevalence	26

Legend: BP, Blood Pressure; COPD, Chronic Obstructive Pulmonary Disease; GDP-PERCH, Gross Domestic Product Percent Change; SVI, Social Vulnerability Index.

Of note, obesity prevalence ranked 31^st in the life expectancy model with an importance of 19.5 (last predictive variable retained—see Figure 2). However, obesity prevalence ranked 50^th in the life lost rate model with an importance of 4.75 and ranked 40^th in the death rate model with an importance of 10.75—in both instances it was not retained in the final predictive models.

Discussion

Why is the U.S., with sufficient financial resources and capability to innovatively address challenges, faltering in addressing the chronic disease crisis and diminished life expectancy compared to peer countries. ²¹ Perhaps a core reason is the all-to-often initiation point of assessing and intervening upon chronic disease—that being the incidence and prevalence of unhealthy lifestyle behaviors. While it may be unintentional, the conversation often begins with unhealthy lifestyle behaviors driving chronic disease risk factors, chronic disease diagnoses and ultimately a shortened lifespan. From a physiological perspective, the pathway from unhealthy lifestyle behaviors to a shorter lifespan is correct.^22,23 Moreover, statistical modeling indisputably demonstrates unhealthy lifestyle behaviors increase the prevalence of chronic disease risk factors and diagnoses as well as lead to a shorter lifespan. ²⁴ These are now considered facts—as a society, it is known that adopting a healthier lifestyle leads to a myriad of improved health outcomes. The U.S. certainly has the financial strength and innovative capability to significantly increase healthy living at a population level. And yet, as a country, a failure to sufficiently address and begin to reverse the chronic disease crisis continues. There is an insufficient understanding and acknowledgment of the upstream forcing factors of unhealthy lifestyle behaviors. Rather, public health messaging and health care delivery, covering both primary and secondary prevention, predominantly follow universal approach of healthy living is good for everyone and therefore everyone should adopt behaviors that are consistent with this messaging. ²⁵ Behavior choices are, however, not that simple, nor is modification of unhealthy lifestyle behaviors merely through a message that such behaviors are bad for your health effective. Understanding the complex milieu of forcing factors that set the stage for healthy or unhealthy lifestyle behaviors and subsequent downstream effects is essential to make a large-scale impact on U.S. population health and health outcomes.

Pronk et al ² recently proposed an ecological framework of health in the context of the chronic disease crisis in the U.S.—proposing forcing factors that set the stage for lifestyle behaviors. This framework was founded on a series of focused and rather straightforward correlative and group comparison analyses.^26-31 Subsequently, the framework was assessed through complex AI modeling, findings that further support model validation.^1,3 The ecological framework also proposes that power dynamic relationships amongst individuals or groups can influence all levels. ³² The results of the current analysis demonstrate the historical prevalence of slavery at a county level was a significant interactive term in AI models predicting life expectancy, years of life lost and death rate. These findings highly suggest the legacy of slavery and its historical prevalence (i.e., greater number of slaves per county in 1860), viewed from the lens of a lasting power dynamic, significantly and negatively influences the ecological framework of health proposed by Pronk et al. ²

Previous studies have performed geohistorical analyses in the context of the history of slavery. Rebbeck ³³ illustrated geohistorical links between cancer mortality and U.S. county slavery density. Kramer et al ¹¹ demonstrated black populations have a 17% slower decline in cardiovascular mortality over approximately 50 years (1968 to 2014) in counties with the highest vs lowest slavery concentration. Only Southern U.S. counties with a history of slavery were included in this analysis. From previous literature, those authors described the concept of “place-based legacy of slavery.” In this concept, the history of a location (e.g., county, community, etc.) is lastingly entrenched and perpetuated by the “ongoing interaction of actors and institutions, making places durably distinct from one another.” ¹¹ The authors go on to state that evidence supports the lasting socioeconomic and political legacy that can be caused by a “historical initial condition,” ¹¹ such as slavery. The findings of the significant disparity in cancer mortality as well as reductions in cardiovascular mortality in previous studies support the importance of the place-based legacy of slavery on health outcomes. These previous studies, while justifiably puts forth complex in theoretical rationales, used concentrated datasets centered on specific chronic conditions and health outcomes to demonstrate their link to slavery concentration without analyzing how the legacy of slavery interacts with other forcing factors of health. The current analysis both aligns with and supports the conclusions of previous studies as well as significantly advances our understanding of the importance of structural racism on population health in a broader, more comprehensive context, demonstrating the significant interaction of slavery legacy with other county-level forcing factors within the ecological framework proposed by Pronk et al. ²

The current findings demonstrate the ecological framework predicts lifespan-related endpoints with high precision, as supported by model R² ≥ 0.71 for all 3 analyses. The predictors found to be significant in Table 4 cover all categories included in the ecological framework (i.e., culture, politics, policy, socioeconomics, lifestyle behaviors, risk factors and chronic disease) as well as the slavery legacy power dynamic. The order of variable importance clearly demonstrates the complexity of U.S. health in the context of chronic disease and shortened life expectancy. By leveraging machine learning methods, we can model complex, non-linear interactions and quantify the specific contribution of the slavery factor within a high-dimensional ecological framework, which is beyond the scope of traditional methods. To ensure that our findings were not driven by overfitting, we applied 5-fold cross-validation, an independent holdout set, and model-based feature importance analyses, all of which confirmed that the influence of slavery legacy is robust. Furthermore, to account for variability introduced by random train-test splits, each model was trained 4 times with different random seeds (0, 1, 2, and 42); the results indicate the stability of our model’s performance. Given the protections from prediction error in this statistical approach, the fact that location-specific history of slavery was identified as an important driver of current health in the U.S. should be viewed as highly compelling. This finding is of particular importance because its assessment as a potentially important health determinant in conjunction with a number of other measures has not been extensively explored.

The findings of the current study have important implications for public health messaging and initiatives moving forward. One implication of the current paper is that the history of slavery should be considered in developing and implementing evidence-based interventions, so they can be tailored to local contexts. Another implication of the current paper is that major determinants of health cannot be identified unless they are investigated. Social determinants of health and structural racism are now important focal points for public health, but this did not happen until substantive research documented the importance of race, ethnicity, income, education, etc. for health outcomes. Moving forward, there should be a greater focus on collectively examining the broader interactive landscape of forcing factors for downstream health behaviors and outcomes, including social determinants of health and structural racism. The disconnect between the apparent complexity of the issue as supported by the findings of the current analysis and the simplicity by which public health initiatives and health care delivery commonly address the issue may be a primary reason the nation has been unsuccessful in improving lifestyle behaviors and the downstream effects of healthy living. Until there is a realization that the imperative is to look upstream from lifestyle behaviors and confront the unfavorable forcing factors, no matter how unpleasant that may be from a historical context, the poor health phenotype and outcomes will persist and likely worsen. Specifically, a focus should be on the following: (1) Understanding how distinct cultural belief systems influence health decisions and behaviors and subsequently align health messaging and interventions with an individual’s or community’s distinct cultural beliefs; (2) use population health as a political unifier as opposed to a platform to propagate divisive messaging and blame; (3) create policies that promote health that transcends cultural and political differences; (4) assess and accommodate for socioeconomic barriers to better health behaviors and outcomes; and (5) determine, acknowledge, and ameliorate unfavorable power dynamics that disproportionately effect the health of vulnerable groups. The ecological framework proposed by Pronk et al ² may prove useful in illustrating the complexity of these issues and establishing future research and public health initiatives.

Unhealthy lifestyle behaviors are a primary driver of chronic disease risk factors and diagnoses, ultimately leading to diminished quality of life and lifespan—unhealthy lifestyle behaviors and chronic disease are, in fact, present-day pandemics.^34-36 These health challenges are not insurmountable if the true root causes and drivers of health in the U.S. are acknowledged, studied and become primary targets for future public health campaigns and ingrained into healthcare delivery. True health equity for the entire U.S. population should be a unifying goal.

So What?

What is Already Known on This Topic?

Unhealthy lifestyle behaviors are a primary driver of chronic disease, ultimately leading to diminished quality of life and lifespan.

What Does This Article Add?

The current findings demonstrate an ecological framework predicts lifespan-related endpoints with high precision. The predictors found to be significant cover all categories included in the ecological framework (i.e., culture, politics, policy, socioeconomics, lifestyle behaviors, risk factors and chronic disease) as well as the slavery legacy power dynamic.

What are the Implications for Health Promotion Practice or Research?

These health challenges associated with unhealthy lifestyle behaviors are addressable if the true root causes and drivers of health are acknowledged, studied and become primary targets for future public health campaigns and ingrained into healthcare delivery.

ORCID iD

Ross Arena https://orcid.org/0000-0002-6675-1996