Skip to main content
NCBI home page
Population Health Management logoLink to Population Health Management
. 2021 Aug 16;24(4):454–462. doi: 10.1089/pop.2020.0170

Operationalizing the Population Health Framework: Clinical Characteristics, Social Context, and the Built Environment

Scott D Siegel 1,2,✉, Madeline Brooks 1, Frank C Curriero 3
PMCID: PMC8403180  PMID: 34406088

Abstract

As a framework, population health emphasizes health outcomes for entire populations, the broad range of determinants of these outcomes, and the comparative effectiveness of medical and public health interventions. In practice, however, many contemporary population health programs instead focus on small subsets of patients who account for a disproportionate share of health care utilization, often with disappointing results. The authors proposed a new approach to operationalize population health in clinical settings, with the example of tobacco use. Electronic health record (EHR) data from a mid-Atlantic health system were used to: (1) define and describe a hospital-based population of current smokers, (2) analyze the demographic characteristics of the population to consider how the social context may impact treatment, and (3) join EHR data with public licensing data on tobacco retail locations to assess the relationship between the built environment and smoking status. Out of a total of 20,310 unique adult admissions to the health system, 3749 (18.5%) were current smokers. Compared to never smokers, current smokers were significantly younger, more likely to be male, more likely to be Black/African American, less likely to be Hispanic/Latino/a, and more likely to be on Medicaid or be self-pay. Current vs. former smokers had significantly higher exposure to tobacco retail locations, even after adjusting for demographic and other covariates. By defining populations around leading modifiable medical determinants of health, and accounting for the larger context of sociodemographic factors and the built environment, health systems can invest in comprehensive programs designed to produce the greatest population health returns.

Keywords: population health, smoking, social context, built environment

Introduction

In the 30 years since Evans and Stoddart1 first proposed a framework for what would develop into population health,2 the general concept has largely taken root in health care settings.3 Compared to more traditional medical models that focus on the treatment of disease in individual patients,4 population health places a greater emphasis on the distribution of population-level health outcomes, the broad range of determinants that interact to influence these outcomes, and the set of medical and public health interventions that have comparatively different impacts on these determinants.2 It is now widely recognized that the determinants of health include the social context and the built environment, which typically cannot be modified by medical interventions alone.5 Thus, investing resources in those medical interventions that make the greatest impact on health in tandem with public health interventions better designed to address the larger set of “upstream” health determinants should, in theory, produce the greatest population health returns.

In practice, however, many population health approaches have veered from this framework. Rather than focusing on the comparative effectiveness of specific interventions to address the range of health needs for entire populations, many health care systems and payers have instead targeted cost containment for comparatively small subpopulations. That is, contemporary versions of population health programs often concentrate on the relatively few patients who use a disproportionate share of health care resources at a given time. These “super-utilizer” or “hotspotting” programs persist even though utilization rates can fluctuate significantly over time and, therefore, represent an unreliable metric for cohorting patients6,7 and can even exacerbate disparities in care delivery.8 Ultimately, well-designed evaluations of super-utilizer interventions have produced disappointing results.9

In the wake of these negative findings, there have been calls to revise contemporary population health approaches in ways that would be more consistent with the framework that Evans and Stoddart first articulated.1 For example, Cantor10 argued in favor of matching intensive care management interventions to subpopulations with persistent complex care needs for which the evidence is most compelling; that is, taking an approach based on clinical need and comparative effectiveness rather than targeting utilization per se. In order to evaluate the potential of population health as it was originally conceived, it will be necessary to first operationalize key constructs to facilitate program implementation within health care settings.

Toward that end, the research team proposes adapting the Centers for Disease Control and Prevention (CDC) population health and prevention framework,11 which classifies interventions into clinical, community-based, and population-level categories. First, health systems can identify opportunities to define patient populations around leading modifiable determinants of health.12 More specifically, priority should be given to comparatively effective clinical interventions that can be readily implemented in health care settings to address leading determinants of health, but where the uptake of these interventions has been suboptimal. Electronic health record (EHR) data can be exploited to define and characterize clinical populations in this fashion, which would establish the scale of the opportunity and clarify what additional clinical and data infrastructure may be necessary. Second, health systems should consider the demographic characteristics of this patient population to better understand the larger context of the social determinants of health at work. The results of this type of analysis can guide community-based efforts to modify existing care delivery in ways that promote access and treatment effectiveness for vulnerable subpopulations. Third, by joining EHR data with publicly available data sets, health systems can conduct spatial analyses to better understand the impact of the built environment on the defined population. Findings from this type of analysis may, in turn, motivate health systems to develop new partnerships to target upstream determinants for the benefit of entire populations.4,11

To illustrate the proposed approach with a practical example, the research team will define a population based on an exemplar modifiable determinant of health: cigarette smoking. Smoking remains the leading preventable cause of morbidity and mortality in the United States12 and claims more than 400,000 lives each year, predominantly from lung cancer, heart disease, and pulmonary disease.13 Incidentally, these conditions account for a large share of hospital expenditures in the US and are often the focus of super-utilizer interventions. Pharmacological therapies for smoking, supported with behavioral treatments, are comparatively effective relative to making an unaided quit attempt: the number needed to treat is 18 smokers to prevent a death.14 More generally, clinical interventions for smoking are notably cost-effective compared to more commonly practiced preventive services (eg, mammography).15 Nevertheless, even though nearly 70% of smokers are interested in quitting, have regular contact with health care settings, and most have made an attempt in the past year, more than 40% of smokers were not advised to quit by a health care professional and only about 30% used evidence-based treatments in their quit attempts.16 Thus, implementing models of care that result in advising more patients to quit and improve access to smoking cessation treatments has the potential to improve population health significantly.

Despite overall reductions in smoking rates over the last 50+ years,17 progress has been much slower for those with low socioeconomic status (SES), racial/ethnic and sexual/gender (ie, LGBTQIA+) minorities, and persons with mental illness.18 Low-SES smokers are no less interested in quitting and make just as many quit attempts as more advantaged smokers but face systematic barriers to accessing treatment and are less likely to achieve abstinence.19 Black smokers are less likely to be advised to quit by health care professionals.20 Similarly, persons with mental illness are less likely to be assessed for tobacco use and offered treatment even though the evidence is clear that smoking cessation interventions are effective in this population for both reducing smoking rates and improving mental health.21 Thus, to produce the greatest population health impact, SES, race/ethnicity, sexual orientation/gender identity, and mental illness would all need to factor into any program designed to improve access to smoking cessation care and treatment effectiveness.

Tobacco control efforts that were implemented after publication of the landmark 1964 Surgeon General's report contributed to smoking rates being cut by more than half.17 The evidence shows that restrictions on marketing and increases in the price of tobacco products through taxation significantly contributed to this progress. The tobacco industry has adapted by increasing marketing and price-reduction promotions at the point of sale in retail locations.22 Tobacco retail locations are much more abundant in low-SES communities and neighborhoods with higher proportions of racial and ethnic minorities and persons with mental illness. Living in areas with elevated tobacco outlet density is associated with fewer quit attempts23 and ultimately lower rates of smoking cessation.24 Accounting for this feature of the built environment can help explain the poorer treatment outcomes observed among low-income and racial minority smokers and points to potential upstream interventions for further evaluation, such as regulating the density of tobacco retail locations within at-risk communities.25

To demonstrate this proposed approach to operationalizing population health in a health care setting, the research team will use EHR data from a mid-Atlantic health system to: (1) define, quantify, and describe the clinical characteristics of a hospital-based population of current cigarette smokers relative to former and never smokers, (2) analyze the demographic characteristics of this population to consider how social context may influence access to care and treatment effectiveness, and (3) join EHR data with public licensing data on tobacco retail locations in the community surrounding the health system to better appreciate the impact of the built environment.

Methods

Setting

The Christiana Care Health System is headquartered in New Castle County, Delaware and includes 2 acute care hospitals located in Wilmington (urban) and Newark (suburban), together accounting for 1227 inpatient beds. These 2 hospitals provide 88% (45,278 Christiana Care hospital discharges/51,262 total discharges) of non-veteran adult acute care in New Castle County.26 This study was reviewed and approved by the Christiana Care Health System Institutional Review Board.

Data

This study drew on Christiana Care Health System EHR data for 20,310 unique adult residents of New Castle County who were admitted to either the Wilmington or Newark hospitals between July 1, 2018 and June 30, 2019. Patient smoking status, residential address, and demographic and clinical data were extracted from the EHR. Patient addresses were manually cleaned and geocoded using ArcGIS 10.6 (Esri, Redlands, CA), yielding a match rate of 98% (20,310/20,706). Of the 396 unmatched patients, 362 had no address information and 34 had addresses that were not locatable. Tobacco retail address data were extracted from a public state business license database.27 This study included all establishments with a tobacco retail license as of April 17, 2019, excluding cigarette affixing agents, wholesalers, internet retailers, and tobacco manufacturers because these entities do not sell directly to consumers in physical storefronts. Records were reviewed to exclude duplicates and update addresses to correspond to the storefront address. Tobacco retail locations also were geocoded using ArcGIS 10.6 with a match rate of 100% (N = 642) and used to generate estimates of tobacco retail location exposure.

Measures

Patient characteristics

Smoking status was initially coded into 8 categories based on a standardized nurse-administered interview conducted at admission and documented in the EHR, consistent with Meaningful Use program requirements.28 The Centers for Medicare & Medicaid Services does not provide specific guidance regarding cigarettes smoked per day or time since cessation for purposes of defining current and former smokers; therefore, patients self-identified the most appropriate category, which were consolidated as: “current every day smoker,” “current some day smoker,” “heavy tobacco smoker,” and “light tobacco smoker” were all coded as current smokers; “former smoker” as former smokers; “never smoker” as never smokers; and “smoker, current status unknown” and “unknown if ever smoked” were excluded. For patients with multiple admissions, the last known smoking status was used. Demographic measures included age, sex (male or female; no data were available for sexual orientation or gender identity), race (Black/African American, White, Other), ethnicity (Hispanic/Latino/a or not) and payer status (commercial, Medicaid, Medicare, or self-pay) as a proxy for SES in lieu of other available data,29 which were all collected from registrars and recorded in the EHR. Clinical characteristics were summarized with the Elixhauser comorbidity index,30 a method for categorizing comorbidities based on International Classification of Diseases diagnosis codes abstracted from the EHR. The total index score was used as an overall measure of disease burden in addition to categories for specific comorbidities associated with smoking, including coronary, pulmonary, metabolic, and psychiatric conditions.

Spatial characteristics

Tobacco retail location exposure was quantified with 2 measures. First, proximity was determined by calculating the Euclidean distance in miles to the nearest tobacco retail location from each patient's home address. Second, density was determined by calculating the number of tobacco retail locations within a half-mile buffer of each patient's home address. Hospital exposure was determined by calculating the Euclidean distance in miles to the Newark and Wilmington hospitals from each patient's home address. Finally, to compare smoking prevalence rates in the hospital-based population to estimates of smoking prevalence in the surrounding population, census tract-level small area estimates for the city of Wilmington were used from the 500 Cities Project.31 Data for areas of New Castle County outside Wilmington were not available.

Analyses

Descriptive statistics were used to characterize the current, former, and never smoker groups for the demographic, clinical, and hospital and tobacco retail location exposure variables. Never smoker data were used to estimate the prevalence of smoking among the hospital-based population and for drawing clinical comparisons with current/former smokers, which was conducted with unadjusted and adjusted linear, logistic, and multinomial regression models. For all subsequent analyses, the research team focused only on current and former smokers to better understand which factors were associated with smoking cessation.

The team next conducted analyses to determine if areas with higher rates of tobacco retail locations were associated with a greater ratio of current to former smokers, which would imply lower rates of smoking cessation. Prior to conducting these analyses, the team evaluated the potential selection bias introduced by using a hospital-based population of smokers: if a greater density of tobacco retail locations was observed in areas proximal (vs. distal) to the hospitals and smokers who live in proximity to a hospital are more likely to use that hospital for acute care needs (vs. distal hospitals), then observed associations between smoking status and tobacco retail location exposure could be spurious. A series of hierarchical (ie, nested) logistic regression models was used to evaluate whether tobacco retail location exposure was associated with smoking status after adjusting for hospital proximity and demographic variables. To visualize these findings, probability maps were created as follows. A fine grid of spatial locations was generated across New Castle County, and tobacco retail location exposure and hospital proximity variables were calculated for each location. Predicted probabilities from the fitted logistic regression models that just contained the tobacco retail location exposure and hospital proximity variables were then generated for each location and mapped as a continuous surface to show spatial variation in predicted smoking status based on distance to the 2 hospitals before and after accounting for the role of tobacco retail exposure.

Finally, the team compared the hospital-based population of smokers to the larger community-based population of smokers in the City of Wilmington. First, the team visualized rates of smoking prevalence for all admitted patients (ie, the percentage of all unique adult admissions who were smokers) and CDC small-area smoking prevalence estimates for Wilmington at the census tract level using choropleth maps with natural break classifications. Second, the team created a scatterplot to assess the relationship between hospital-based smoking prevalence and CDC-estimated smoking prevalence to determine if more smokers were admitted from census tracts with higher rates of smoking prevalence. Third, the team calculated a census tract-level ratio of hospital-based smoking prevalence to CDC-estimated smoking prevalence to identify areas in which the hospital-based population over- or undersampled smokers. Fourth, the team visualized these ratios with a choropleth map, manually setting 4 class breaks based on the distribution of the data to represent undersampling (ratio <0.833), proportional sampling (ratio = 0.834–1.167), oversampling (ratio = 1.168–1.335), and high oversampling (ratio >1.335).

Results

There was a total of 20,310 unique adult admissions to the Christiana Care Health System hospitals between July 1, 2018 and June 30, 2019, of whom 3749 (18.5%) were current smokers, 6368 (31.4%) were former smokers, and 10,193 (50.2%) were never smokers. Table 1 displays the demographic, clinical, and built environment exposure characteristics for these groups. Compared to never smokers, current smokers were significantly younger, more likely to be male, more likely to be Black/African American, less likely to be Hispanic/Latino/a, and more likely to have Medicaid or be self-pay. Former smokers, compared to never smokers, were older, more likely to be male, less likely to be Black/African American, less likely to be Hispanic/Latino/a, and more likely to be on Medicare. Regarding clinical variables, after adjusting for age and other demographic variables, current smokers did not differ overall on the Elixhauser index compared to never smokers but had significantly higher rates of chronic pulmonary disease, alcohol and other substance use disorders, and psychoses (Table 1). By contrast, current smokers had lower rates of hypertension and diabetes than never smokers. Former smokers had significantly higher rates of hypertension, congestive heart failure, chronic pulmonary disease, diabetes, and alcohol and other substance use disorders relative to never smokers (Table 1). Comparing the groups on built environment exposures, current smokers lived significantly closer to their nearest tobacco retail location and had double the number of tobacco retail locations within a half mile of their home relative to never smokers. Former smokers did not differ from never smokers on either of these measures. Current smokers also lived significantly closer to both hospitals than never smokers; former smokers only lived closer to Wilmington Hospital (Table 1).

Table 1.

Characteristics of Current, Former, and Never Smokers

  Current smokers (N = 3749) Former smokers (N = 6368) Never smokers (N = 10193) Total (N = 20310)
Demographic Characteristics and Health Care Utilization
Age, median 55** 71** 66 65
Male, n (%) 2005 (53.5)** 3197 (50.2)** 4085 (40.1) 9287 (45.7)
Race, n (%)
 Whitea 2426 (64.7) 4847 (76.1) 6754 (66.3) 14027 (69.1)
 Black/African American 1167 (31.1)** 1314 (20.6)** 2715 (26.6) 5196 (25.6)
  Other Race 156 (4.2)** 207 (3.3)** 724 (7.1) 1087 (5.4)
Hispanic/Latino, n (%) 172 (4.6)* 198 (3.1)** 594 (5.8) 964 (4.7)
Payor, n (%)
 Commerciala 927 (24.7) 1331 (20.9) 3102 (30.4) 5360 (26.4)
 Medicaid 1474 (39.3)** 585 (9.2) 1300 (12.8) 3359 (16.5)
 Medicare 1307 (34.9)** 4430 (69.6)** 5734 (56.3) 11471 (56.5)
 Self-pay 41 (1.1)** 22 (0.3) 57 (0.6) 120 (0.6)
Length of stay, median 3.9** 3.7* 3.6 3.7
Elixhauser Comorbidity Index and Selected Comorbidities
Elixhauser index, median 8 13†† 10 11
Hypertension (uncomplicated), n (%) 2449 (65.3)† 5326 (83.6)†† 7521 (73.8) 15296 (75.3)
Congestive heart failure, n (%) 689 (18.4) 1932 (30.3)†† 2348 (23.0) 4969 (24.5)
Chronic pulmonary disease, n (%) 1875 (50.0)†† 3115 (48.9)†† 3320 (32.6) 8310 (40.9)
Diabetes (uncomplicated), n (%) 1074 (28.6)†† 2505 (39.3)†† 3594 (35.3) 7173 (35.3)
Alcohol use disorder, n (%) 1464 (39.1)†† 1070 (16.8)†† 944 (9.3) 3478 (17.1)
Substance use disorder, n (%) 1621 (43.2)†† 798 (12.5)†† 823 (8.1) 3242 (16.0)
Psychoses, n (%) 431 (11.5) †† 322 (5.1) 526 (5.2) 1279 (6.3)
Tobacco Retail Location Exposure
Miles to nearest tobacco retail location, median 0.26** 0.37 0.37 0.35
Tobacco retail locations within a half mile of home, median 4** 2 2 3
Hospital Exposure
Miles to Wilmington Hospital, median 5.5** 6.7* 7.3 6.7
Miles to Newark Hospital, median 6.2* 6.0 6.0 6.0
Open in a new tab
a

Reference group for unadjusted multinomial regression.

*

Significant difference relative to never smoker group (P < 0.05).

**

Significant difference relative to never smoker group (P < 0.001).

†

Significant difference relative to never smoker group after controlling for age, sex, race, ethnicity, and insurance (P < 0.05).

††

Significant difference relative to never smoker group after controlling for age, sex, race, ethnicity, and insurance (P < 0.001).

Table 2 displays results from the hierarchical regression analyses that evaluated whether tobacco retail exposure was associated with current smoker status after adjusting for hospital proximity and demographic variables. In Model 1, proximity to the Wilmington Hospital, but not the Newark Hospital, was associated with current smoking status. For every mile increase in distance from Wilmington Hospital, there was an associated 4% lower odds of being a current smoker (OR = 0.96, 95% CI: 0.96, 0.97, P < 0.001), which is visualized in Panel A of Figure 1. This relationship held in Model 2 when demographic variables were included; younger age, Black/African American race, and noncommercial insurance (ie, Medicaid, Medicare, self-pay) also were associated with current smoking status. In Model 3, proximity to the nearest tobacco retail location was associated with current smoking status, and proximity to the Wilmington Hospital remained significant. In the final model (Model 4), tobacco retail location density was associated with current smoking status; proximity to Wilmington Hospital was no longer significant. Panel B of Figure 1 visualizes the relationship between tobacco retail location exposure and current smoking status after adjusting for hospital proximity, providing strong evidence that this relationship is not spurious.

Table 2.

Hierarchical Regression Analysis of Predictors of Current Smoking Status

Predictor variables Model 1
 Model 2
 Model 3
 Model 4
AOR (95% CI) AOR (95% CI) AOR (95% CI) AOR (95% CI)
Miles to Christiana Care Wilmington Hospital 0.96 (0.96, 0.97)** 0.98 (0.97, 0.98)** 0.99 (0.98, 1.00)* 1.00 (0.99, 1.01)
Miles to Christiana Care Newark Hospital 1.00 (0.99, 1.01) 1.00 (0.99, 1.01) 1.00 (0.99, 1.02) 1.00 (0.98, 1.01)
Age   0.94 (0.94, 0.95)** 0.94 (0.94, 0.95)** 0.94 (0.94, 0.95)**
Sex (male = reference)   0.88 (0.80, 0.96)* 0.87 (0.79, 0.95)* 0.87 (0.79, 0.96)*
Race (white = reference)
 Black   1.14 (1.02, 1.27)* 1.09 (0.97, 1.22) 0.96 (0.85, 1.08)
 Other race   0.79 (0.58, 1.06) 0.78 (0.58, 1.05) 0.77 (0.57, 1.04)
Hispanic/Latino ethnicity (non-Hispanic/Latino = reference)   0.81 (0.60, 1.09) 0.79 (0.59, 1.06) 0.73 (0.54, 0.98)*
Insurance (commercial = reference)
 Medicaid   2.33 (2.03, 2.68)** 2.28 (1.99, 2.62)** 2.21 (1.92, 2.54)**
 Medicare   1.15 (1.01, 1.31)* 1.13 (1.00, 1.29) 1.12 (0.98, 1.27)
 Self-pay   1.91 (1.11, 3.37)* 1.90 (1.10, 3.35)* 1.93 (1.12, 3.41)*
Miles to nearest tobacco retail location     0.71 (0.62, 0.81)** 0.78 (0.68, 0.89)**
Tobacco retail locations within half mile of home       1.02 (1.01, 1.03)**
Open in a new tab
*

Significant at P < 0.05.

**

Significant at P < 0.001.

AOR, adjusted odds ratio; CI, confidence interval.

FIG. 1.

FIG. 1.

Open in a new tab

Maps of predicted probabilities of current smoker status based on hospital proximity, before and after adjustment for tobacco retail exposure.

Figure 2 visualizes census tract-level smoking prevalence estimates in the City of Wilmington for the hospital-based population (Panel A) and the general population based on the CDC estimated smoking prevalence (Panel B). In general, prevalence estimates were higher for the hospital-based population, as would be expected given the risks for hospitalization that smoking imparts. There was a large correlation between the 2 prevalence estimates (rs(22) = 0.8, P < .001), indicating that hospital-based smokers were spatially representative of the general population of smokers (Panel C). Notably, the slope of the (dashed) regression line in Panel C is steeper than the (solid) line representing what would be a 1:1 relationship between the 2 prevalence estimates, which provides evidence that smokers who live in census tracts with higher prevalence rates (red dots) are more likely to be admitted than smokers from census tracts with lower prevalence rates (blue dots). These findings are further visualized in Panel D (with color scheme matched to that from Panel C), displaying the spatial (census tract level) variation in the ratio of the 2 prevalence rates.

FIG. 2.

FIG. 2.

Open in a new tab

Comparison of census tract-level smoking prevalence in Wilmington, Delaware based on hospital inpatient data and estimates from the CDC. CDC, Centers for Disease Control and Prevention.

Discussion

Results from this study of a hospital-based population of smokers illustrate a pragmatic approach to operationalizing the population health framework first articulated by Evans and Stoddart in 1990.1 Nearly 19% of hospital admissions in a 12-month period, more than 3700 patients, were current smokers. Half of this population had chronic pulmonary disease, approximately 40% had an alcohol and/or other substance use disorder, and more than 10% had a serious mental illness. Sociodemographic findings were consistent with prior research18: current smokers were more likely to be younger, male, Black/African American, and of low SES. With regard to the built environment, compared to former smokers, current smokers had significantly higher exposure to tobacco retail locations, even after adjusting for demographic and other covariates. This study further found that the hospital-based population of smokers was spatially representative of the larger community-based population of smokers. Preliminary evidence suggests that smokers from communities with the highest levels of smoking prevalence are at greater risk of hospital admission than smokers from other communities, possibly pointing to the additional risks associated with secondhand tobacco smoke exposure.

These results should be interpreted in the context of several important limitations. First, smoking status was based entirely on self-report and therefore may represent an underestimate of the true prevalence of smoking and bias associations with sociodemographic and clinical characteristics. Second, potential inaccuracies in race and ethnicity EHR data and the lack of data on sexual orientation and gender identity may further bias association estimates. Third, although the relationships between measures of tobacco retail location exposure and smoking status remained after adjusting for hospital exposure, the research team cannot eliminate the possibility of other unmeasured confounders or specifically address the question of causality. Fourth, despite spatial representativeness, these findings apply to smokers who met criteria for hospital admission and do not necessarily generalize to all smokers.

Taken together, these results point to several opportunities to improve access to care and treatment effectiveness for smoking cessation. At a minimum, the evidence strongly supports health systems implementing clinical programs with the capacity to ensure that the nearly 1 in 5 hospital patients who are smokers are advised to quit and offered evidence-based treatment,32 including for the large subset of smokers with an alcohol/substance use disorder or another mental health condition. That said, common systemic barriers to implementing such programs would need to be addressed, including inadequate clinician training, EHR systems that have not been optimized to facilitate identifying smokers and tracking treatment, and constraints on clinician time and reimbursement.21 Furthermore, given the evidence on disparities in smoking cessation care delivery related to sociodemographic characteristics, EHR systems should be further leveraged to track treatment rates and outcomes by these characteristics. This could create yet more barriers. For example, as was the case here, few health systems routinely collect data on sexual orientation and gender identity. In traditional health care settings, or those that implement super-utilizer approaches, these barriers will continue to stymie progress, perpetuate disparities, and contribute to avoidable morbidity and mortality. For health systems that embrace a population health framework, improving training, redesigning care delivery, optimizing EHR systems, and partnering with payers to clear these barriers would represent strategic priorities.

Broadening the focus to include the larger social context, health systems can go even further to improve access to treatment for underserved groups through innovations in care delivery. For example, lessons learned from partnering with barbershops and faith-based organizations to improve access to treatment for hypertension in Black communities can be applied to smoking cessation.33 Similarly, community-based participatory research has provided evidence of the feasibility of delivering smoking cessation services in food pantries for low-SES smokers.34 Additional multivariate, multilevel analyses also can inform the development of tailored approaches for smokers who contend with intersectional forms of disadvantage.35 For example, low-SES smokers with mental illness who also face discrimination for being racial/ethnic and sexual/gender minorities may benefit from different treatment approaches than other smokers who face comparatively little sociodemographic disadvantage. More recent innovations in personalizing pharmacological36 and behavioral37 smoking cessation interventions to individual patient characteristics can factor into these tailored approaches.

Present study results add to a growing body of evidence24,38 to suggest that greater exposure to tobacco retail locations undermines efforts to quit smoking and remain abstinent. However, additional research is necessary to evaluate the impact of policies that regulate tobacco retailers.39 New regulations could produce unintended consequences that might disproportionately harm already marginalized communities, such as the generation of illicit markets, violence, and higher rates of incarceration.40 From a methodological perspective, study findings showed that both proximity to tobacco retail locations and measures of density were each uniquely associated with smoking status. Novel measures that incorporate both sources of information could represent important metrics for the evaluation of policy interventions. Health systems can support these evaluations by forging strong partnerships with the community and leveraging EHR data. Furthermore, although not a role that traditional health systems typically play, advocating for evidence-based public policies that promote equity would be entirely consistent with a population health approach.

Implementing a population framework that defines populations around modifiable determinants of health, while also addressing the social context and built environment, would in many cases represent a major undertaking for health systems that operate primarily in a fee-for-service environment or have invested in super-utilizer approaches.7 It has often been suggested that when health systems adopt value-based payment models, this would create a “burning platform” and generate enough motivation to overcome institutional barriers to change.41 However, even value-based payment models prioritize cost containment (eg, per member per month targets). As such, rather than engaging in major undertakings to redesign systems of care, which would include addressing the specific barriers raised herein (eg, EHR systems, clinician training, modifying delivery models), it could be argued that health systems respond rationally by focusing on what appears to be a much more manageable task: reducing utilization for the relatively small group of patients who account for disproportionate shares of care. Furthermore, there is limited evidence that the inclusion of quality measures in value-based payment models leads to improved population health outcomes and may even lead to unintended consequences, such as increases in low-value care and widening disparities.42 If, instead, health systems were evaluated and reimbursed primarily on their ability to equitably improve the health of the population,43 then health systems would have a true incentive to redesign health care delivery systems to emphasize comparatively effective interventions, partner with the community to improve access to treatment, and advocate for health-promoting policies.

Conclusion

This paper proposed a new approach to operationalizing the population health framework first offered by Evans and Stoddart,1 employing the example of tobacco use. By defining populations around leading modifiable medical determinants of health and considering the larger context of sociodemographic factors and the built environment, health systems can invest in developing comprehensive programs that produce the greatest population health returns. This approach can be replicated for other leading modifiable determinants of health (eg, hypertension, prediabetes, opioid use disorder) and provide an alternative to traditional health care models and super-utilizer approaches.

Acknowledgments

We thank Robert A. Schnoll and Lisa Maxwell for their helpful feedback on this article. We also thank Bayo M. Gbadebo and James T. Laughery for their assistance in creating the data set.

Author Confirmation Statement

Scott D. Siegel: Conceptualization, Writing – Original Draft, Project Administration, Funding Acquisition. Madeline Brooks: Formal Analysis, Data Curation, Writing – Original Draft, Visualization. Frank C. Curriero: Methodology, Software, Writing – Review & Editing, Supervision.

Author Disclosure Statement

The authors declare that there are no conflicts of interest.

Funding Information

This project was supported by the Delaware INBRE program, with a grant from the National Institute of General Medical Sciences – NIGMS (P20 GM103446) from the National Institutes of Health and the State of Delaware.

References

Articles from Population Health Management are provided here courtesy of SAGE Publications

RESOURCES