糖心vlog

Key Points

Question听 To what degree are geographic differences in leading risk factors for morbidity and mortality in the US associated with the place where people live?

Findings听 In this retrospective cohort study of 5鈥�342鈥�207 adults, moving from a 10th to a 90th percentile zip code for a given outcome was associated with a significantly increased prevalence of uncontrolled blood pressure of 7 percentage points (from a baseline of 29%), uncontrolled diabetes of 1 percentage point (from a baseline of 7%), obesity of 2 percentage points (from a baseline of 39%), and depressive symptoms of 3 percentage points (from a baseline of 19%) among movers.

Meaning听 A substantial percentage of the change in individuals鈥� likelihood of having poor blood pressure control or depressive symptoms may be related to the place in which they live.

Importance听 The prevalence of leading risk factors for morbidity and mortality in the US significantly varies across regions, states, and neighborhoods, but the extent these differences are associated with a person鈥檚 place of residence vs the characteristics of the people who live in different places remains unclear.

Objective听 To estimate the degree to which geographic differences in leading risk factors are associated with a person鈥檚 place of residence by comparing trends in health outcomes among individuals who moved to different areas or did not move.

Design, Setting, and Participants听 This retrospective cohort study estimated the association between the differences in the prevalence of uncontrolled chronic conditions across movers鈥� destination and origin zip codes and changes in individuals鈥� likelihood of uncontrolled chronic conditions after moving, adjusting for person-specific fixed effects, the duration of time since the move, and secular trends among movers and those who did not move. Electronic health records from the Veterans Health Administration were analyzed. The primary analysis included 5鈥�342鈥�207 individuals with at least 1 Veterans Health Administration outpatient encounter between 2008 and 2018 who moved zip codes exactly once or never moved.

Exposures听 The difference in the prevalence of uncontrolled chronic conditions between a person鈥檚 origin zip code and destination zip code (excluding the individual mover鈥檚 outcomes).

Main Outcomes and Measures听 Prevalence of uncontrolled blood pressure (systolic blood pressure level >140 mm Hg or diastolic blood pressure level >90 mm Hg), uncontrolled diabetes (hemoglobin A_1c level >8%), obesity (body mass index >30), and depressive symptoms (2-item Patient Health Questionnaire score 鈮�2) per quarter-year during the 3 years before and the 3 years after individuals moved.

Results听 The study population included 5鈥�342鈥�207 individuals (mean age, 57.6 [SD, 17.4] years, 93.9% men, 72.5% White individuals, and 12.7% Black individuals), of whom 1鈥�095鈥�608 moved exactly once and 4鈥�246鈥�599 never moved during the study period. Among the movers, the change after moving in the prevalence of uncontrolled blood pressure was 27.5% (95% CI, 23.8%-31.3%) of the between-area difference in the prevalence of uncontrolled blood pressure. Similarly, the change after moving in the prevalence of uncontrolled diabetes was 5.0% (95% CI, 2.7%-7.2%) of the between-area difference in the prevalence of uncontrolled diabetes; the change after moving in the prevalence of obesity was 3.1% (95% CI, 2.0%-4.2%) of the between-area difference in the prevalence of obesity; and the change after moving in the prevalence of depressive symptoms was 15.2% (95% CI, 13.1%-17.2%) of the between-area difference in the prevalence of depressive symptoms.

Conclusions and Relevance听 In this retrospective cohort study of individuals receiving care at Veterans Health Administration facilities, geographic differences in prevalence were associated with a substantial percentage of the change in individuals鈥� likelihood of poor blood pressure control or depressive symptoms, and a smaller percentage of the change in individuals鈥� likelihood of poor diabetes control and obesity. Further research is needed to understand the source of these associations with a person鈥檚 place of residence.

The prevalence of leading risk factors for morbidity and mortality in the US varies geographically.^1-3 However, the extent that such geographic differences are related to a person鈥檚 place of residence vs the characteristics of the people who live in different places is unclear.^4-7 Identifying the contribution of place of residence to geographic health differences is vital in addressing geographically disparate declines in life expectancy¹ because policies targeting patient factors may fail to reduce differences driven primarily by contextual factors.⁸

The relationship between a person鈥檚 place of residence and geographic differences in health outcomes is challenging to study because people who live in places with dissimilar mean health outcomes likely differ in unmeasured ways that are correlated with their health and for which direct adjustment cannot be performed. To account for unobservable individual-level factors, prior studies used 1998-2008 Medicare claims to trace how health care use and diagnosis rates changed among individuals who moved to a destination with a different mean outcome level than their origin location.^9,10

However, a limitation has been the inability to observe movers鈥� health outcomes over time. Leading risk factors for morbidity and mortality, including elevated blood pressure and blood glucose levels, obesity, and depressive symptoms,¹¹ are recorded in electronic health records that do not typically follow people after they move, and are not captured by claims data. Furthermore, claims-based proxies using diagnosis codes introduce systematic, place-specific surveillance bias.¹⁰

This retrospective cohort study evaluated the outcomes of individuals who moved and were tracked before and after moving and the outcomes of individuals who did not move using the nationally integrated electronic health records of the Veterans Health Administration (VHA), thereby circumscribing the limitations of information based on claims data. The study estimated the association between the difference in the prevalence of a health outcome in a person鈥檚 destination vs origin zip code and changes in the person鈥檚 health outcome after moving, adjusting for the person鈥檚 characteristics, the time since he or she moved, and secular trends.

The study was approved by the Department of Veterans Affairs subcommittee for human studies at the Veterans Affairs New York Harbor Healthcare System. A waiver of informed consent was granted because the research involved minimal to no risk for participants and could not be practicably conducted without such a waiver.

The VHA is the largest health care system in the US. It maintains a national repository populated with information from all local VHA electronic health record platforms, which includes demographics, vitals, laboratory values, self-reported measures, and outpatient and inpatient records on all encounters with VHA clinicians.^12-14

For the primary analysis, we identified all persons with at least 1 VHA outpatient encounter between 2008 and 2018 who moved zip codes exactly once or never moved (eFigure 1 in the Supplement). Individuals could have moved within the same county, between counties in a state, or between states. For each risk factor, we required a person to have at least 1 value recorded in the electronic medical record during both the premove and the postmove period.

Moves were identified by quarterly changes to the zip codes of residence. This limited the accuracy with which the premove and postmove periods could be precisely separated (eg, if a person moved midquarter). For those who did not move, we used the midpoint quarter-year the individual was observed as the start of their postmove period.

For each outcome, exposure was defined as the difference in the prevalence of the outcome between a person鈥檚 destination vs origin zip code. The prevalence of an outcome in a person鈥檚 origin or destination zip code was calculated as the mean of all values recorded in the zip code during the study period after excluding the values of the index mover, which were left out to eliminate any mechanical correlation between a person鈥檚 outcomes and exposure.

Outcomes were defined as uncontrolled blood pressure (equal to 1 if a person鈥檚 quarterly mean systolic blood pressure level was >140 mm Hg or quarterly mean diastolic blood pressure level was >90 mm Hg and 0 otherwise),¹⁵ uncontrolled diabetes (quarterly mean hemoglobin A_1c level >8%),¹⁶ obesity (quarterly mean body mass index [BMI; calculated as weight in kilograms divided by height in meters squared] >30),¹⁷ and depressive symptoms (quarterly mean score 鈮�2 on the 2-item Patient Health Questionnaire [PHQ-2], a validated and sensitive questionnaire routinely used to screen for depression).¹⁸ Secondary outcomes included quarterly mean values of the underlying vital (blood pressure level and BMI), laboratory (hemoglobin A_1c level), or self-reported (PHQ-2 score) measures, which were recorded during routine care, as well as quarterly counts of VHA outpatient visits and inpatient admissions per person.

Outcome values were obtained from electronic health records of visits to any VHA facility via the VHA corporate data warehouse (CDW). In audit studies, CDW-based measures of uncontrolled blood pressure, uncontrolled diabetes, and BMI demonstrated high sensitivity and specificity (>95%) when using full chart review and local medical records as the reference standard.^12,19 In addition, the CDW-based PHQ-2 scores demonstrated moderate sensitivity (81%) and high specificity (100%)²⁰ for depression screening, consistent with other PHQ-2 validation studies.¹⁸ The percentage of patients without recorded values for blood pressure level was 2.8%, it was 29.3% for hemoglobin A_1c level, it was 11.4% for BMI, and it was 13.4% for PHQ-2 score. If a person had no values recorded for a risk factor during a given quarter-year, the value was left as missing and was not imputed. For outpatient visits and inpatient admissions, the absence of a recorded service was coded as no health care use.

In addition to zip code, other covariates included age, sex, race, ethnicity, marital status, co-pay exempt status, percentage of service connection (based on the extent to which a person鈥檚 injuries and illnesses were incurred or aggravated during active military service and the degree of disability), and priority group (eligibility category for VHA benefits). Race and ethnicity were self-reported during the regular course of care and were important to include given known associations with chronic disease control and to evaluate whether exposure was associated with changes in the composition of the cohort鈥檚 race or ethnicity over time.

We estimated the association between an individual鈥檚 quarterly health outcome and the interaction between an indicator variable for ever moving, a postmove indicator variable, and the difference in prevalence of the outcome between the person鈥檚 destination vs origin zip code (eFigure 2 in the Supplement). We also examined the association between outcomes over time and the new environment by replacing the postmove indicator variable with a vector of quarter-years since moving indicator variables. This generated quarterly estimates during the premove period to serve as a falsification test of whether premove trends in health outcomes were correlated with future exposure (eg, premove coefficients should be null).

All regression analyses included person-specific fixed effects that adjust for and preclude inclusion in the model of individual-level covariates that do not vary over time (eg, sex and origin zip code); quarter-year fixed effects that adjust for trends in outcomes common to all individuals in the cohort, including those who did not move; and fixed effects for each quarter-year period before and after a person鈥檚 move date. Huber-White robust standard errors were clustered at the level of origin zip code.

We conducted subgroup analyses to determine whether the association between each outcome and the difference in the prevalence of the outcomes between a person鈥檚 destination vs origin zip code differed across individuals who moved within vs across counties or states.

Several sensitivity analyses were conducted to examine the robustness of the results (details appear in eMethods in the Supplement). First, we examined the sensitivity of the results to potential bias from nonrandom missing outcomes data correlated with exposure. We repeated the analysis using a balanced 2-period model in which the dependent variable was the mean value of the outcome during each person鈥檚 premove and postmove period (all persons included had 1 value for the outcome per period). In addition, we directly tested whether data missingness (differential attrition or testing rates) or the characteristics of people without missing values over time (compositional changes) were correlated with exposure. To mitigate potential attenuation bias from high-dimensional fixed effects,^21,22 the person-specific fixed effects were removed and replaced with a set of individual-level covariates.

Second, we examined the sensitivity of the results to potential bias from unobserved factors. In addition to conducting a falsification test using premove outcomes, E-values were calculated to estimate how strong unmeasured confounding from factors correlated with both exposure and outcomes (but not already accounted for by the person-specific fixed effects, time fixed effects, or time since moving fixed effects) would need to be to explain away the observed association.²³

Third, we estimated the potential effect of unmeasured confounders related to moving or a person鈥檚 choice of destination. We applied an instrumental variable for whether a person ever moved during the study period using the mean outmigration rate of the person鈥檚 origin zip code (leaving the index mover out) and the 2007 unemployment rate of the person鈥檚 origin county, respectively. We evaluated the strength of the instrumental variables (based on the first-stage regression F statistic >10), the monotonicity assumption (based on a binned scatterplot of a binary variable for whether a person ever moved vs each instrumental variable), and the plausibility of the exclusion restriction (based on whether the instrumental variable was correlated with within-person changes in outcomes after moving).²⁴ Furthermore, we adjusted our estimate for changes to a person鈥檚 income and disability level around the time of moving that might be correlated with his or her exposure (eg, if a person鈥檚 decision of where to move to was related to a change in his or her economic or disability status). Specifically, we included individuals鈥� quarterly priority group category as an independent variable, a time-varying measure of eligibility for VHA services commonly used as a proxy for economic and health status (available quarterly for 2009-2018; eTable 1 in the Supplement). A propensity score model that used inverse probability of treatment weights to reduce imbalance in potential confounding factors between movers and those who did not move also was implemented. In addition, following the methods of Chetty and Hendren,²⁵ placebo tests were conducted to assess whether observed changes in a given outcome after moving were related to differences between zip codes for all other outcomes.

Fourth, we modified our inclusion and exclusion criteria and definition of exposure to reduce the potential for measurement error in outcomes or exposure variables and to expand generalizability.

Fifth, we ran the primary analysis at the county level (rather than the zip code level), using each person鈥檚 county of residence as the geographic unit of analysis to define place of residence, movers, and those who did not move.

Study design and reporting were based on the Strengthening the Reporting of Observational Studies in Epidemiology statement.²⁶ The statistical analyses were performed using Stata version 15 (StataCorp). Statistical tests were 2-sided with a significance threshold of P鈥�<鈥�.05. Because of the potential for type I error due to multiple comparisons, the findings for the analyses of the secondary end points should be interpreted as exploratory.

The study population included 5鈥�342鈥�207 individuals (mean age, 57.6 years [SD, 17.4 years], 93.9% men, 72.5% White individuals, and 12.7% Black individuals), of whom 1鈥�095鈥�608 moved exactly once and 4鈥�246鈥�599 never moved during the study period (eTable 2 in the Supplement). Individuals who moved were younger, were more likely to be female, and more likely to have a service connection than those who did not move (eTable 3 in the Supplement). Prior to moving, 89.7% of movers had 1 or more measures for blood pressure level recorded, 49.0% for hemoglobin A_1c level, 74.2% for BMI, and 66.7% for PHQ-2 score, of whom 29.3% had uncontrolled blood pressure, 39.0% had uncontrolled diabetes, 7.1% had obesity, and 18.9% had depressive symptoms prior to moving.

Baseline characteristics were similar across individuals from the same zip code who subsequently moved to zip codes with a lower vs higher prevalence of each risk factor (Table 1 and eTable 4 in the Supplement). Rates of loss to follow-up and demographic characteristics of the people observed over time were not significantly associated with exposure (eTables 5 and 6 in the Supplement).

We observed substantial variation across zip codes in the unadjusted prevalence of uncontrolled blood pressure, uncontrolled diabetes, obesity, and depressive symptoms (Figure 1). The mean absolute difference in the rate of uncontrolled blood pressure between destination and origin zip codes was 3.2% (median, 2.2%; interquartile range, 0.1%-4.5%), with 538鈥�758 movers (49.2%) relocating to zip codes with worse blood pressure control than their origin zip codes (Table 2). Differences across moves for other risk factors were similar in magnitude (Table 2 and eFigure 3 in the Supplement) and at most weakly correlated with one another (eTable 7 in the Supplement).

Before moving, there was not a statistically significant association between differences in the prevalence of each risk factor across destination and origin zip codes and changes in movers鈥� uncontrolled blood pressure, uncontrolled diabetes, obesity, or depressive symptom rates over time (Figure 2). For all risk factors, the adjusted estimate was not significantly different from 0 during each quarter-year before the move, suggesting any postmove differences were not driven by preexisting differences in risk factor trends across individuals from the same origin zip code who subsequently moved to a destination zip code with a different risk factor prevalence (Figure 3).

After moving, differences in the prevalence of each risk factor across destination and origin zip codes were positively and significantly associated with changes in the prevalence of the risk factor among movers (Table 3). Among the movers, the change after moving in the prevalence of uncontrolled blood pressure was 27.5% (95% CI, 23.8%-31.3%) of the between-area difference in the prevalence of uncontrolled blood pressure. Similarly, the change after moving in the prevalence of uncontrolled diabetes was 5.0% (95% CI, 2.7%-7.2%) of the between-area difference in the prevalence of uncontrolled diabetes; the change after moving in the prevalence of obesity was 3.1% (95% CI, 2.0%-4.2%) of the between-area difference in the prevalence of obesity; and the change after moving in the prevalence of depressive symptoms was 15.2% (95% CI, 13.1%-17.2%) of the between-area difference in the prevalence of depressive symptoms. For example, moving from a 10th to a 90th percentile zip code for a given risk factor was associated with a statistically significant increased prevalence of uncontrolled blood pressure of 7.4 percentage points (95% CI, 6.4-8.5 percentage points; from a baseline of 29.3%), uncontrolled diabetes of 1.0 percentage point (95% CI, 0.5-1.4 percentage points; from a baseline of 7.1%), obesity of 1.6 percentage points (95% CI, 1.0-2.2 percentage points; from a baseline of 39.0%), and depressive symptoms of 3.0 percentage points (95% CI, 2.6-3.4 percentage points; from a baseline of 18.9%). Results were similar for systolic and diastolic blood pressure levels, hemoglobin A_1c level, BMI, and PHQ-2 score and for outpatient and inpatient health care use (eTable 8 in the Supplement).

When separately estimating the association for each quarter-year period after the move, the magnitude of the estimated association jumped discretely at the time of the move and remained similar during each quarter-year period after the move (Figure 3).

For each risk factor, the association between geographic differences and postmove individual changes was significantly greater for individuals who moved between counties in the same state vs within the same county, and greater again for those who moved between states (eTable 9 in the Supplement). For individuals who moved across states, the change after moving in the prevalence of uncontrolled blood pressure was 46.9% (95% CI, 37.2% to 56.6%) of the between-area difference in the prevalence of uncontrolled blood pressure; for uncontrolled diabetes it was 10.9% (95% CI, 6.6% to 15.4%) of the between-area difference in the prevalence of uncontrolled diabetes; for obesity it was 4.9% (95% CI, 2.6% to 7.2%) of the between-area difference in the prevalence of obesity; and for depressive symptoms it was 26.3% (95% CI, 23.2% to 29.5%) of the between-area difference in the prevalence of depressive symptoms. We also observed significantly greater changes in risk factors for moves with a larger magnitude of between-area difference in prevalence.

The magnitude and significance of the results were similar for the majority of sensitivity analyses (eTables 10-14 and eFigure 4 in the Supplement). However, when using county of residence vs zip code as the geographic unit of analysis, the estimated magnitude was 51.5% vs 27.5%, respectively, for uncontrolled blood pressure, 10.1% vs 5.0% for uncontrolled diabetes, 7.4% vs 3.1% for obesity, and 28.8% vs 15.2% for depressive symptoms (eTable 14 in the Supplement).

When using the mean outmigration rate of a person鈥檚 origin zip code as an instrumental variable for whether he or she ever moved during the study period, the estimated association for uncontrolled blood pressure was 24.7% (95% CI, 20.6%-28.8%), uncontrolled diabetes was 4.5% (95% CI, 2.0%-7.0%), obesity was 2.6% (95% CI, 1.3%-3.9%), and depressive symptoms was 13.3% (95% CI, 11.0%-15.7%) (eTable 14 in the Supplement). Both instrumental variables were strong for predicting ever moving (F statistic >1000) and uncorrelated with within-person outcome changes after moving, suggesting the exclusion restriction was plausible (eTable 15 in the Supplement). Binned scatterplots of each instrumental variable vs ever moving were consistent with a monotonic relationship (eFigure 5 in the Supplement).

The E-value was 3.7 for uncontrolled blood pressure, 1.9 for uncontrolled diabetes, 1.3 for obesity, and 3.3 for depressive symptoms, indicating that an unmeasured confounder would have to be correlated with both the outcome and exposure (and not already accounted for by the person-specific fixed effects, time fixed effects, or time since moving fixed effects) with an odds ratio of 1.3 or higher to explain away some of the findings (eTable 16 in the Supplement).

In this retrospective cohort study of individuals receiving care at VHA facilities, geographic differences in prevalence were associated with a substantial percentage of the change in individuals鈥� likelihood of poor blood pressure control or depressive symptoms, and a smaller percentage of the change in individuals鈥� likelihood of poor diabetes control and obesity. The findings were consistent across alternate analyses with varied underlying assumptions that used different methods to minimize potential bias from unmeasured confounders.

These findings contribute to the literature on geographic differences in health outcomes across the US by estimating the degree to which differences are associated with the place where people live vs the characteristics of the people who live in different places. Prior cross-sectional analyses^1,2,27,28 were unable to control for unobservable individual-level factors that differ across regions. Prior longitudinal studies^9,10 were limited to using claims data, which lack the necessary information on vital signs, laboratory results, and clinical assessments to assess leading risk factors for morbidity and mortality.

The current analysis improves on the prior research by using nationally integrated electronic health records to evaluate how an individual鈥檚 risk factors changed after moving to a destination with a different prevalence of the risk factor than his or her origin location, holding the person fixed, and adjusting for secular trends using the outcomes of individuals who did not move. The results are consistent with the Moving to Opportunity randomized trial that found neighborhood effects on depressive symptoms,²⁹ and with 2 influential patient migration studies^9,10 that used Medicare claims from 1998-2008. Finklestein et al⁹ found that half of the geographic variation in health care use was associated with place, of which they estimated that 37% may be explained by changes in individual patient鈥檚 chronic conditions. Song et al¹⁰ found that rates of chronic condition diagnoses increased among beneficiaries who moved to an area where the mean diagnosis count was higher. The current study鈥檚 secondary finding that movers鈥� underlying health outcomes were not significantly associated with differences in mean health care use between their destination vs origin zip code was a key but previously untested assumption underlying the internal validity of the prior studies.^9,10

Further research is needed to understand the sources of the associations we observed between place of residence and changes in individuals鈥� likelihood of uncontrolled chronic conditions. Across all conditions, the estimated association of place was significantly larger for individuals who moved between counties or states compared with those moving within the same county. This suggests that both county-level factors and broader contextual factors may contribute to the observed changes. The larger association of place for blood pressure control and depressive symptoms vs diabetes control and obesity suggests the differential importance of individual behaviors (such as dietary and physical activity and adherence to lifestyle interventions) in patterns of geographic variation for the latter risk factors.^30,31 Consistent with this, the association of place with uncontrolled diabetes increased from 5.0% to 9.1% and with obesity increased from 3.1% to 4.1% when not adjusting for unobservable person-specific fixed effects vs from 27.5% to 29.5% for uncontrolled blood pressure and 15.2% to 17.9% for depressive symptoms. Incorporating longitudinal data on contextual factors (eg, the timing of states expanding Medicaid) into the design of future patient migration studies, and combining medical records with prospectively collected survey measures and biospecimens among participants鈥� enrolled in large cohort studies who move between areas,³² could advance understanding of how specific contextual factors contribute to different behavioral and biological processes that underlie geographic variation in morbidity and mortality.

There are several limitations to this study. First, confounding by a person鈥檚 reason for moving or choice of where they moved to cannot be ruled out. In particular, the study design was unable to separate the causes that led someone to move to a destination with better or worse health outcomes from the consequences of moving to that destination. For example, a person who moved to a neighborhood with a greater prevalence of uncontrolled blood pressure due to job loss may have poorer blood pressure control after moving due to reduced income and loss of commercial insurance coverage. Even though the findings were robust to adjustment for within-person changes in individuals鈥� economic and health status around the time of the move, to falsification tests using premove outcomes, and to using instrumental variables for moving and inverse probability weights to reduce imbalance in potential confounding factors, these sensitivity analyses do not eliminate the possibility that factors related to the cause of the move may have accounted for the findings.

Second, the E-values for uncontrolled diabetes and obesity were small, indicating that an unmeasured confounder correlated with both the outcome and exposure could account for those findings.

Third, this study used electronic health records from visits to VHA clinicians. It is possible that patients who received some of their care outside the VHA system may have had more missing outcomes. However, the results were similar after excluding individuals with any record of non-VHA insurance coverage and truncating individuals at aged 65 years (the age of Medicare eligibility) (eTable 14 in the Supplement).

Fourth, more generally, there may have been selection bias for the type of patients who had outcome variables measured. For example, 34.7% of individuals had their hemoglobin A_1c level measured both before and after moving vs 81.5% of individuals for blood pressure level (eTable 17 in the Supplement). Although the rates of loss to follow-up and changes in the characteristics of people who had outcome variables measured over time were not associated with exposure, and the results were robust when running a 2-period model without missing outcomes data, it remains possible that nonrandom missing outcomes data could bias the results.

Fifth, veterans are predominantly male and older compared with the overall adult US population, which limits generalizability to younger populations who may be more influenced by changes in area factors earlier during their life course.^33,34

In this retrospective cohort study of individuals receiving care at Veterans Health Administration facilities, geographic differences in prevalence were associated with a substantial percentage of the change in individuals鈥� likelihood of poor blood pressure control or depressive symptoms, and a smaller percentage of the change in individuals鈥� likelihood of poor diabetes control and obesity. Further research is needed to understand the source of these associations with a person鈥檚 place of residence.

Corresponding Author: Aaron Baum, PhD, Department of Health System Design and Global Health and the Arnhold Institute for Global Health, Icahn School of Medicine at Mount Sinai, Veterans Affairs New York Harbor Healthcare System, 1216 Fifth Ave, Ste 559, New York, NY 10029 (aaron.baum@mssm.edu).

Accepted for Publication: July 19, 2020.

Author Contributions: Dr Baum had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Baum, Basu.

Acquisition, analysis, or interpretation of data: Baum, Wisnivesky, Siu, Schwartz.

Drafting of the manuscript: Baum.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Baum, Basu.

Administrative, technical, or material support: Schwartz.

Supervision: Wisnivesky, Basu.

Conflict of Interest Disclosures: Dr Baum reported serving as a consultant to the American Board of Family Medicine. Dr Wisnivesky reported receiving personal fees from Sanofi, GlaxoSmithKline, and Banook; and receiving grants from Sanofi. Dr Basu reported receiving grants from the National Institutes of Health, the US Centers for Disease Control and Prevention, the US Department of Agriculture Economic Research Service, the Center for Poverty Research, the Robert Wood Johnson Foundation, Harvard University, and Stanford University; and receiving personal fees from KPMG, Research Triangle Institute, PLoS Medicine, the New England Journal of Medicine, and Collective Health. Dr Schwartz reported receiving personal fees from Veterans Affairs New York Harbor Healthcare System. No other disclosures were reported.

Funding/Support: This article is the result of work supported with resources and the use of facilities at the Veterans Affairs New York Harbor Healthcare System.

Role of the Funder/Sponsor: The Veterans Affairs New York Harbor Healthcare System had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Disclaimer: The contents do not represent the views of the US Department of Veterans Affairs or the US government.