- Research
- Published:
Weighted EHR-based prevalence estimates for hypertension at the state and local levels in Louisiana
樱花视频 volume听25, Article听number:听432 (2025)
Abstract
Background
Modernization of public health data systems is a national priority. Improved chronic disease surveillance can provide more timely, accurate, and local measures to inform public health policy and intervention. Although electronic health record (EHR) data have great potential for surveillance, population coverage is non-random, which may result in biased estimates. Statistical approaches are needed to adjust estimates to represent the underlying population and validate the results against independent estimates.
Methods
MENDS, the Multi-State-EHR-Based Network for Disease Surveillance, uses EHR data to calculate chronic disease and risk factor prevalence metrics. In this study, we applied post-stratification weighting to MENDS data from nearly 800,000 adults from Louisiana to estimate the prevalence of hypertension and hypertension control at the state and parish (county-equivalent) levels during February 2023. We then compared weighted MENDS hypertension prevalence estimates with measures derived from traditional public health surveillance.
Results
Weighted MENDS hypertension prevalence estimates were approximately 10% lower than crude MENDS estimates, and approximately 13% higher than hypertension awareness estimates from the 2021 Behavioral Risk Factor Surveillance System, with similar geographic and demographic patterns. Weighted MENDS hypertension prevalence estimates indicated that 43.0% of Louisiana adult residents had hypertension (versus 47.7% crude prevalence). Prevalence was higher than the overall state estimate among men (47.3% weighted; 55.9% crude), Black patients (50.2% weighted; 55.7% crude), those receiving Medicare (70.6% weighted; 76.2% crude), and individuals living in rural areas (46.1% weighted; 49.8% crude). Hypertension prevalence increased with age and with more clinical visits during the previous 2听years. Hypertension prevalence was highest in the southeastern parishes near New Orleans and Baton Rouge. Demographic and geographic patterns in prevalence of hypertension control were like hypertension prevalence.
Conclusions
Post-stratification weighting of MENDS data brought EHR-based data estimates closer to survey-based estimates of hypertension and can improve representativeness of chronic disease indicators. These estimates can provide public health organizations with timely, accurate, and local information. Further, EHR-based systems can produce unique measures, such as the prevalence of hypertension control, which can provide a more nuanced understanding of community needs and help public health agencies evaluate the effectiveness of community interventions.
Background
Modernization of public health data systems is a national priority [1]. Accurate, timely, and locally relevant public health metrics are critical for developing effective and equitable public health policies and interventions [2, 3]. Even amid the COVID-19 pandemic, heart disease remained the leading cause of mortality in the United States, and stroke remained in the top five causes of death [4, 5]. However, despite a recognized need, development of a national system for timely, detailed, and locally relevant surveillance of cardiovascular disease and other chronic conditions has faced many challenges [6].
Traditional chronic disease surveillance commonly relies on manual data collection methods such as population-based surveys. Although these approaches have greatly contributed to public health work, they can be expensive, labor intensive, subject to self-report bias, and limited in representativeness and geographic granularity. Moreover, survey data are often subject to extended delays between data acquisition and dissemination [7, 8].
Electronic health record (EHR) data can provide complementary public health surveillance information that overcomes some of these limitations [7,8,9]. Use of EHR data has the potential to modernize chronic disease surveillance systems by providing clinically detailed data on large, geographically distributed populations in a timely, automated, and sustainable manner [7,8,9,10]. Public health departments can use EHR data to complement their surveillance systems to provide a more nuanced and granular understanding of chronic disease prevalence and risk factors [9, 10]. Access to such geographically and demographically detailed information can inform targeted public health interventions to prevent or better manage chronic conditions at local levels [11, 12].
MENDS, the Multi-State EHR-Based Network for Disease Surveillance, is an innovative data modernization pilot project that fosters partnerships between healthcare organizations, clinical data aggregators, and public health departments [13]. These partnerships provide public health departments with EHR data to monitor and respond to chronic disease conditions [9, 13,14,15]. MENDS has the capability to serve as a powerful public health tool because the system incorporates rich clinical data from approximately 11 million patients across the United States [9]. However, EHR-based surveillance systems like MENDS also have limitations that require application of advanced statistical methods and validation against other data sources to obtain meaningful surveillance estimates [16,17,18,19,20,21,22]. These limitations include potential biases in representativeness and geographic coverage caused by non-random partner selection; uneven demographic distribution of patients in partner catchment areas; a limited number of data points from small geographic locales; differences in hospital services availability, access, and utilization; and incomplete data on socioeconomic status and other social determinants of health [16,17,18,19,20,21,22]. Some of these limitations can be addressed through statistical approaches that adjust for potential biases to derive more representative estimates from EHR data [10, 11, 16,17,18,19,20,21,22].
This paper reports on the application of a statistical approach called post-stratification weighting to MENDS data to generate state and local chronic disease prevalence estimates in Louisiana. We used the prevalence of high blood pressure (hypertension) and prevalence of effective clinical treatment of high blood pressure in people who have been diagnosed with essential hypertension (hypertension control) to illustrate the application of this method to calculating geographically and demographically granular estimates from MENDS data. We then compared the resulting estimates with survey-based hypertension prevalence estimates derived from self-reported awareness of having a hypertension diagnosis from the Behavioral Risk Factor Surveillance System (BRFSS) [23].
Methods
We obtained a limited dataset of patient-level sociodemographic and clinical data from REACHnet, a MENDS data contributor operated by the Louisiana Public Health Institute [24], in accordance with MENDS鈥 standard data use agreements and methodology [13, 25]. Under MENDS data governance policies, data contributors executed legal agreements, including Business Associate Agreements and Data-Sharing and Use Agreements, with the information technology vendor that supports MENDS for software use and system activities. Similar agreements were executed with the MENDS Coordinating Center (National Association of Chronic Disease Directors or NACDD) and the University of Massachusetts Lowell to permit access to limited datasets for validation and advanced analytic purposes. To support partner site governance work, the Centers for Disease Control and Prevention (CDC) provided a written determination that MENDS operates within the public health authority pursuant to the Health Insurance Portability and Accountability Act (HIPAA), excluding it from the need for ethics approval. The legislation that outlines the exclusions for public health surveillance activities from HIPAA can be found in 45 CFR 搂164.512(b). Because it is public health surveillance, MENDS is exempt from institutional review board (IRB) review, although several data contributors in the MENDS network have sought IRB approval for their own participation.
Before being made available for indicator calculation, MENDS data undergo an extensive internal validation process to ensure accuracy and completeness of records included in the MENDS database [25], in addition to the data characterization and quality checks performed by the data contributor [26]. Available data in REACHnet as of February 2023 were transferred to our analytic team at the University of Massachusetts Lowell for processing via a HIPAA-compliant secure data transfer protocol and were stored on HIPAA-compliant data servers at the University of Massachusetts Lowell. All data transmissions and storage were password protected and encrypted to adhere to established security protocols [25].
Inclusion criteria
Our analysis included adult patients aged 20 to 84听years who had at least one clinical encounter and one blood pressure measure in the outpatient setting during the 2-year period from March 2021 to February 2023. Adults under 20听years and over 84听years were excluded to align the population with age groups used in demographic stratification in the U.S. Census American Community Survey (ACS) [27]. Data generated during inpatient encounters were excluded in the analysis because those measures might not represent the normal ambulatory status of patients. Pregnant patients were also excluded. Patients whose records were missing sex, age, race and ethnicity, or 5-digit ZIP code of residence were excluded from the analysis. Finally, we limited the analysis to patients with a valid 5-digit residential ZIP code in Louisiana. This resulted in exclusion of 1,391,384 patients, for a final sample of 785,999 patients for state-level analysis (Fig.听1).
Definitions
A hypertension case was defined as an individual having any of the following: (a) diagnosed hypertension in accordance with ICD-10 diagnosis codes (ICD-10 I10 Essential Hypertension or I15 Secondary Hypertension), (b) prescription for an antihypertensive drug (with or without a diagnosis code), or (c) in the absence of a hypertension diagnosis, high blood pressure readings (systolic blood pressure鈥夆墺鈥140听mmHg systolic or鈥夆墺鈥90听mmHg diastolic on two or more readings within a 1-year period) during the preceding 2听years. This surveillance case definition was derived from a previous study that explored and validated analytic approaches to defining hypertension using MENDS data for a similar population [28]. Hypertension prevalence was calculated as the percentage of included individuals with hypertension among all patients included in the analysis at a given geographic or demographic level during the 2-year study period.
Hypertension control was defined as individuals with essential hypertension whose most recent blood pressure reading was below 140听mmHg systolic and 90听mmHg diastolic, as recommended by the American Heart Association and the Center for Medicare & Medicaid Services quality measure for hypertension control [29]. Hypertension control was only assessed in the 259,078 patients with a coded diagnosis of essential hypertension for blood pressure readings on or after the date the diagnosis was first recorded.
Post-stratification weighting
The risk of bias in prevalence estimates owing to the non-random nature of EHR data has been well documented [16,17,18,19,20,21,22]. The sociodemographic distributions of patients of Louisiana included in MENDS data [28] do not perfectly match the characteristics (sex, age group, and race) and geographic distribution of the underlying Louisiana adult population recorded by the ACS [27] for the corresponding time periods. Broadly, among all Louisiana resident patients in MENDS, the MENDS data had a lower proportion of males (44.0% versus 49.0%) and Hispanic individuals (5.5% versus 6.9%) compared with the ACS. Geographically, MENDS data were over-representative of larger, typically urban parishes. The median coverage (percentage of the total parish population included as patients in MENDS data) among parishes was 5.9%. Among parishes with coverage equal to or above this median, the mean parish population was 93,501 people compared with the mean population of parishes below the median coverage (51,517 people). Race and age distributions were similar at the state level but that did not preclude the possibility of differences among these groups at lower geographic levels.
Probability weighting approaches commonly used to adjust for bias in demographic distributions cannot be applied to EHR data because the probability of inclusion is unknown. To address this challenge, we applied post-stratification weighting [30,31,32,33,34,35] to estimate prevalences of hypertension and hypertension control that account for Louisiana鈥檚 demographic and geographic population structure. Post-stratification weighting uses census-based population data to limit bias owing to under- and over-representation of demographic or geographic subgroups.
We calculated adjusted estimates by post-stratified weighting according to sex (male versus female), age group (20鈥44, 45鈥65, and 65鈥84听years), and race and ethnicity (non-Hispanic White, non-Hispanic Black, non-Hispanic Asian, Hispanic, and Other) distributions from the ACS [27]. Effectiveness of post-stratification weighting relied on adequate sample size (defined as鈥夆墺鈥125 patients per parish or population subgroup) in a specific geographic area and sociodemographic group. Parish rurality was categorized as Mostly Urban (2010 United States Department of Agriculture Rural鈥揢rban Commuting Area or RUCA Codes 1鈥6), Mostly Rural (RUCA Codes 7鈥9), and Completely Rural (RUCA Code 10) [36].
State estimates
We developed post-stratification weights by parish and within each parish, further post-stratified by sex, age group, and race and ethnicity using the following calculations. Suppose the population of parish \(c\) in state \(s\) was stratified by sex (two strata), age group (three strata), and race and ethnicity (six strata) into 2鈥壝椻3鈥壝椻6鈥=鈥36 strata. Use \(i\) to represent sex group, \(j\) for age group, and \(k\) for race and ethnicity group; use \(\hat{p}_{ijk}\) to represent the prevalence estimate of the stratum of \(i^{th}\) sex group, \(j^{th}\) age group, and \(k^{th}\) race group; and use \(w_{scijk}\) to represent the known population proportion of the \(ijk\) stratum in parish \(c\). First, calculate the prevalence estimate of the \(ijk\) stratum of state \(s\) as: \(\hat{p}_{scijk} = m_{scijk} /n_{scijk}\), where \(n_{scijk}\) is the total number of persons in the stratum, and \(m_{scijk}\) is the number of persons with the target condition in the stratum in parish \(c\) of state \(s\).
The post-stratified prevalence of state \(s\) was \(\hat{P}_{s} = \sum\nolimits_{c = 1}^{C} {\sum\nolimits_{i = 1}^{2} {\sum\nolimits_{j = 1}^{3} {\sum\nolimits_{k = 1}^{6} {w_{scijk} \hat{p}_{scijk} } } } }\)\(\hat{P}_{s} = \sum\nolimits_{c = 1}^{C} {\sum\nolimits_{i = 1}^{2} {\sum\nolimits_{j = 1}^{3} {\sum\nolimits_{k = 1}^{6} {w_{scijk} \hat{p}_{scijk} } } } }\), and the variance was \(V(\hat{P}_{s} ) = \sum\nolimits_{c = 1}^{C} {\sum\nolimits_{i = 1}^{2} {\sum\nolimits_{j = 1}^{3} {\sum\nolimits_{k = 1}^{6} {w_{scijk}^{2} \frac{{\hat{p}_{scijk} (1 - \hat{p}_{scijk} )}}{{n_{scijk} }}} } } } - \frac{1}{N}\sum\nolimits_{c = 1}^{C} {\sum\nolimits_{i = 1}^{2} {\sum\nolimits_{j = 1}^{3} {\sum\nolimits_{k = 1}^{6} {w_{scijk}^{2} \hat{p}_{scijk} (1 - \hat{p}_{scijk} )} } } }\), where \(n_{scijk}\) is the sample size of the \(ijk^{th}\) stratum in parish \(c\) of state \(s\). When any member of the \(scjik\) stratum has zero sample size, it can be combined with an adjacent stratum. When the sample size is sufficiently large, the chance of any \(n_{scijk}\) being zero becomes very small.
Crude hypertension prevalence estimates are provided in the results for comparison among key demographic groups at both the state and City of New Orleans levels to demonstrate the variability in effects of these adjustments across geographic scales.
Parish and subgroup estimates
Post-stratification weighting was also adapted to estimate parish and demographic subgroup prevalences adjusted for other demographic groupings. For example, for males (\(i = 1\) for male; \(i = 2\) for female) in state \(s\), the prevalence can be estimated as: \(\hat{P}_{s,male(i = 1)} = \sum\nolimits_{j = 1}^{3} {\sum\nolimits_{k = 1}^{6} {w_{s,1jk}^{*} \hat{p}_{s,1jk} } }\), where \(w_{s,1jk}^{*} = \frac{{w_{s,1jk} }}{{\sum\nolimits_{j = 1}^{3} {\sum\nolimits_{k = 1}^{6} {w_{s,1jk} } } }}\) are the rescaled proportions corresponding to the age and racial distributions in the male subpopulation. The variance can be estimated as:
where \(N_{s,i = 1}\) is the size of the male population in state \(s\).
Weighting methods were expanded to generate prevalence estimates by primary insurance payer type in the most recent encounter (commercial/private, Medicaid, Medicare, or unknown) and rurality (based on the 5-digit ZIP code of latest known home address), treating each category of the variables as a subpopulation. Based on data availability, similar weighting approaches were also applied to parish-level estimates when a parish had at least 125 included patients. Weighted estimates for parishes with fewer than 125 patients were suppressed because of poor estimate precision.
Adjusted prevalence estimates for hypertension control among patients with an essential hypertension diagnosis were calculated using a similar approach.
External comparison
To assess external validity, we compared our estimates with the most similar metric available at the appropriate geographic scale鈥擟DC鈥檚 PLACES analysis of BRFSS self-reported hypertension awareness from the most recent data released at the time of this writing (2021) [37,38,39,40]. We assessed whether the MENDS estimates were similar to those reported in the most recent BRFSS results from 2021 among demographic groups where possible (crude estimates, [38]), parishes (age-adjusted estimates, [39]), and ZIP codes within the City of New Orleans (crude estimates, [40]). The variation between the estimates at the sub-state geographic level were also compared between the sources.
Analysis and visualization
Post-stratification analysis was performed in Stata MP 18 (College Station, TX, USA). Maps were made in Quantum GIS (version 3.10.6-A Coru帽a, [41]) using county-equivalent and state area base maps from Natural Earth [42] and ZIP code tabulation area base maps from the U.S. Census TIGER/Line庐 Shapefiles [43]. Maps were made using MENDS estimates from data available as of February 2023 and CDC鈥檚 PLACES hypertension prevalence estimates derived from BRFSS self-reported hypertension awareness responses on the survey collected during 2021 [37,38,39,40,41].
The percent difference between MENDS weighted and unweighted (crude) hypertension prevalence was calculated by demographic group using the formula:
A positive percent difference indicates that the weighted estimate was higher than the unweighted (crude) estimate, and a negative percent difference indicates the weighted estimate was lower than the unweighted (crude) estimate.
Percent differences were also used to compare the weighted MENDS hypertension prevalence estimates with the BRFSS hypertension awareness estimates by demographic groups and by parish at the state level and by ZIP code within the City of New Orleans. Interpretation is similar, with a positive value indicating that the MENDS weighted estimate was higher than the BRFSS hypertension awareness estimate and a negative value indicating the weighted estimate was lower than the BRFSS estimate.
Results
The MENDS Louisiana dataset included 785,999 adults aged 20 to 84听years meeting the inclusion criteria (Fig.听1). These data included 58.2% women, 24.5% aged 65 and 84听years, and a race and ethnicity composition of 61.5% White, 30.4% Black, 5.1% Hispanic, and 1.6% Asian patients. Just over a third of patients (37.0%) had private or commercial health insurance; 16.2% were Medicaid beneficiaries, 21.4% were Medicare beneficiaries, and 25.3% had self-pay or unknown insurance status. Most individuals included in this analysis resided in urban areas (94.1%), and most (84.5%) had nine or more clinical encounters during the 2-year study period (Table听1). For New Orleans, 137,455 patients were identified for analysis, and this study population subset had a demographic profile similar to the state-level data (Table听1).
State estimates
Table 1 shows the state-level hypertension prevalence estimates overall and by selected patient characteristics. The statewide weighted hypertension prevalence was 43.0% (versus 47.7% crude prevalence). Hypertension prevalence was higher among men (47.3% weighted; 55.9% crude) than women (39.1% weighted; 41.9% crude) and progressively increased with age. Compared with the overall state estimate, hypertension prevalence was notably higher among Black people (50.2% weighted; 55.7% crude) and lower among Asian people (31.0% weighted; 31.8% crude). Hypertension prevalence estimates were highest among Medicare beneficiaries (70.6% weighted; 76.2% crude), followed by Medicaid beneficiaries (41.3% weighted; 43.3% crude) and those commercially insured (37.8% weighted; 40.7% crude). Completely rural residents had a slightly higher prevalence (46.1% weighted; 49.8% crude) than those living in mostly urban areas (43.2% weighted; 47.8% crude). Hypertension prevalence estimates increased with an increasing number of clinical encounters in the past 2听years. Overall, statewide weighted hypertension prevalence estimates were 10% lower than crude estimates, with a range of 鈭15.4%鈭3.1% across demographic subgroups.
Table 2 shows statewide weighted prevalence estimates of hypertension control among patients with an essential hypertension diagnosis. Of the 259,078 patients identified with a diagnosis of essential hypertension, 68.3% were identified as having controlled hypertension based on the most recent blood pressure record. Prevalence of hypertension control was similar among male and female patients (67.8% versus 68.8%). Older patients had slightly higher control prevalence than younger patients (69.2% of those aged 65鈥84听years versus 65.6% of those aged 20鈥44听years). Black patients had a notably lower control prevalence (62.3%) compared with white (72.1%) and Asian patients (73.1%). Medicare beneficiaries (70.5%) and those with commercial or employer-based insurance (71.8%) had the highest control prevalence, whereas Medicaid beneficiaries (65.1%) and those with self-pay or unknown payer type (62.2%) had the lowest. Residents of urban or mostly rural areas tended to have higher control prevalence than those living in completely rural areas.
Parish estimates
Hypertension prevalence estimates varied geographically (Fig.听2a), ranging from 27.8% to 66.8%. Parishes in the southeastern part of the state and greater New Orleans area had the highest prevalence estimates. Southwestern parishes tended to have lower hypertension prevalences. Sample sizes were too small for calculation of reliable estimates for several parishes in the northern part of the state.
Multi-State EHR-Based Network for Disease Surveillance (MENDS) weighted prevalence of hypertension and hypertension control among adults aged 20鈥84听years compared with Behavioral Risk Factor Surveillance System (BRFSS) age-adjusted hypertension awareness among adults 18听years and older by parish鈥擫ouisiana
Figure听2 also shows parish estimates of hypertension control (Fig.听2b). Owing to limited sample sizes, these estimates were based on data from the southern part of the state. Prevalence estimates of hypertension control ranged from 52.2% to 78.9%. The variation in hypertension control among parishes (spread of 27 percentage points) was smaller than the variation in hypertension prevalence estimates (spread of 39 percentage points).
City of New Orleans estimates
Hypertension prevalence estimates for the City of New Orleans compared with statewide estimates are shown in Table听3. The citywide prevalence estimate was 48.0% (versus 48.6% crude prevalence). Prevalence estimates were higher for men (51.0% weighted; 55.9% crude) than women (45.4% weighted; 43.0% crude). Prevalence differed by age, with the highest prevalence among those aged 65 to 84听years (81.4% weighted; 81.1% crude), lower among those aged 45 to 64听years (62.7% weighted; 61.6% crude), and lowest among those aged 20 to 44听years (26.5% weighted, 26.9% crude). Black patients had the highest prevalence (57.3% weighted; 60% crude), followed by White (38.444% weighted; 38.6% crude), Hispanic (34.9% weighted; 33.4% crude), and Asian patients (32.5% weighted; 31.2% crude). Medicare beneficiaries had the highest prevalence (82.9% weighted; 83.0% crude), followed by Medicaid beneficiaries (49.7% weighted; 49.8% crude) and those with commercial or employer-based insurance (41.0% weighted; 39.9% crude). Prevalence estimates were progressively higher among patients with more clinical encounters in the past 2听years. Overall, City of New Orleans weighted hypertension prevalence estimates were 1.1% lower than crude estimates, with a range of 鈭8.8%鈭6.7% across demographic subgroups.
Table 2 shows City of New Orleans hypertension control prevalence estimates among patients with an essential hypertension diagnosis compared with statewide estimates. Of the 44,280 patients in New Orleans with a diagnosis of essential hypertension, 60.9% were identified as having controlled hypertension based on the most recent blood pressure record. Female and male patients had comparable control prevalence (61.4% versus 60.4%). Older adults (65鈥84听years) had a higher control prevalence (64.5%) than adults aged 45鈥64听years (60.1%) and aged 20鈥44听years (55.0%). Black patients had a slightly lower control prevalence (58.2%) compared with the overall population, and Asian patients had the highest control prevalence (72.7%). Medicare beneficiaries had the highest control prevalence (64.6%), while Medicaid beneficiaries had the lowest (54.7%).
The prevalence of both hypertension and hypertension control differed by 5-digit ZIP code within the City of New Orleans. Hypertension prevalence estimates ranged from 36.9% to 55.6% (spread of 19 percentage points). Prevalence of hypertension control varied from 54.3% to 69.0% (spread of 15 percentage points; Fig.听3).
Multi-State EHR-Based Network for Disease Surveillance (MENDS) weighted prevalence of hypertension and hypertension control among adults aged 20鈥84听years compared with Behavioral Risk Factor Surveillance System (BRFSS) unadjusted (crude) hypertension awareness among adults 18听years and older by parish鈥擟ity of New Orleans
External comparison
The MENDS surveillance population in this study had a slightly higher proportion of female (58.2% versus 55.1%) and Black and Asian individuals (30.4% versus 21.2% and 1.6% versus 1.0%, respectively) when compared with the 2021 BRFSS population (Table听4). The age distributions of the MENDS and BRFSS populations are more difficult to compare because of differences in both the age ranges included in the analysis and the age grouping; however, overall, the MENDS surveillance population had slightly more individuals under 44 (41% 20鈥44听years versus 24.2% 18鈥44听years) and fewer individuals aged 65听years and older (24.5% 65鈥84听years versus 33.7% 65听years and older) compared with the 2021 BRFSS sample (Table听4).
Despite these differences in sample demographics, weighted MENDS hypertension prevalence estimates overall and among demographic subgroups were slightly higher than the BRFSS hypertension awareness estimates (Table听4). BRFSS estimates for the City of New Orleans were not available for demographic subgroups; however, the overall city-level estimate of hypertension prevalence was notably higher in the MENDS analysis (43.0%) compared with the BRFSS hypertension awareness estimate (33.7%).
MENDS parish-level hypertension prevalence estimates had a larger value range compared with BRFSS hypertension awareness estimates (27.7% to 66.8%, a 39.1 percentage point spread compared with 33.0% to 51.0%, an 18 percentage point spread) and tended to be higher compared with BRFSS age-adjusted hypertension awareness estimates (Fig.听2c and d). The median percent difference between the MENDS and BRFSS estimates among parishes was 14.9% (range: 鈭21.8% to 43.3%). Although differences between estimates were relatively small in the northern and central part of the state, MENDS hypertension prevalences were notably higher in the southeastern part of the state. Similar comparisons were noted at the city level, with MENDS estimates tending to be higher than the comparable BRFSS estimate at the ZIP code level (Fig.听3c and d).
Discussion
Using routinely collected outpatient EHR data from the MENDS project, we produced weighted adult hypertension prevalence estimates for the state of Louisiana at the state, parish, and New Orleans city levels, and at the 5-digit ZIP Code level within the City of New Orleans. These estimates were reasonable for these geographies based on comparison with hypertension awareness estimates from the 2021 BRFSS and what is known about the differences in how the two sources measure hypertension. Further, we generated sex, age group, race and ethnicity, and insurance payer type specific estimates for Louisiana and the City of New Orleans with patterns that were also consistent with those seen in the hypertension awareness estimates from the 2021 BRFSS.
At the state level, weighted MENDS hypertension estimates were about 10% lower than unweighted (crude) MENDS estimates and about 13% higher than the 2021 BRFSS hypertension awareness estimates, indicating that weighting brought crude MENDS estimates closer to survey based BRFSS estimates. Notably, the differences between crude and weighted hypertension prevalences estimates were less pronounced at the local level in the New Orleans metropolitan area, where MENDS has higher and more consistent patient coverage, than at the state level. Overall, weighted MENDS estimates across geographic levels showed similar patterns to BRFSS estimates among demographic groups.
Multiple factors may contribute to these observations. First, the MENDS hypertension prevalence estimates were based on documented medical diagnosis, medications prescribed, and blood pressure readings recorded in a medical record. In contrast, the PLACES estimates based on BRFSS data reflect only individuals who were aware of and self-reported their hypertension status. Therefore, MENDS data represented a more accurate and inclusive definition of hypertension that captures both persons with confirmed and suspected hypertension. Second, inclusion criteria for age differed between the two analyses. The MENDS analysis truncated the sample population to adults aged 20 to 84听years to match census groupings needed for post-stratification analysis using multiple demographic variables. In contrast, the BRFSS estimates include all adults 18听years or older. The exclusion of younger adults (18鈥19听years) and the exclusion of older adults (over 85听years) might have impacted affect the prevalence estimates in the MENDS analysis. Finally, although inpatient records were excluded from this analysis, individuals who are in apparent good health may be less likely to visit a doctor even for routine physical examinations, further reducing the number of people without hypertension in the MENDS denominators.
The potential for estimate bias due to underlying differences in healthcare access and use is an inherent challenge in calibrating EHR-based metrics such as hypertension prevalence [44, 45]. For this reason, when developing EHR-based public health surveillance pipelines, it is ideal to validate different definitions and weighting approaches against estimates derived from more than one data source, from a similar period and population. However, such validation is challenging for chronic disease indicators owing to both the lag between data collection and release of large-scale risk factor surveys鈥攚hich may introduce differences attributable to change over time and not to methodological differences鈥攁nd the dissimilarity in methods and geographic- or population-level granularity. For example, comparison to National Health and Nutrition Examination Survey (NHANES; [46]) data shows large differences that are likely attributable to multiple methodological differences related to time, clinical definitions, and differences in population age ranges. Increasingly, EHR-based estimates from different datasets are available that can be an additional resource for cross-validating approaches. For example, modeled hypertension prevalence estimates for U.S. states calculated using a large EHR dataset also trended higher than comparable estimates from BRFSS, and the MENDS estimate fell within the mapped range of 40鈥54% for the state of Louisiana [47].
We also calculated estimates of hypertension control among adults aged 20鈥84听years with essential hypertension by the same geographic levels and demographic subgroups.
The MENDS hypertension control estimate for Louisiana (68.3%) was higher than the national 2017鈥2018 NHANES estimate of 39.6% [48]. However, the NHANES analysis defined hypertension control at a lower, more stringent threshold (<鈥130/80听mm Hg, rather than鈥<鈥140/90听mm Hg used in MENDS). Another difference between MENDS and NHANES assessment of hypertension control is that MENDS used a more permissive denominator, including both of individuals with diagnosed hypertension and those prescribed an antihypertensive medication, whereas NHANES assessed control only among those taking antihypertensive medication. The MENDS state estimate was more similar to the range mapped (61鈥64%) to Louisiana in an analysis of a large proprietary EHR dataset where hypertension control was defined as the percentage of patients whose most recent blood pressure was鈥<鈥140/90听mmHg among people suspected to have hypertension based on a combination of diagnoses, blood pressure readings, or medication prescriptions [47].
Limitations
This analysis demonstrated the feasibility of using routinely collected outpatient EHR data for population-based chronic disease surveillance. Although using outpatient EHR data for chronic disease surveillance is promising, it is not without limitations. Electronic health record data inherently represent a healthcare seeking population. Further, what information is recorded in medical records can vary among health systems [49]. While EHR data are of large scale and rich in clinical detail, they also tend to have uneven geographic distributions. The selection of MENDS data contributors and their clinical partners across the United States, including this example in Louisiana, are non-random, resulting in data that are not statistically representative of the underlying total populations of interest, particularly for small rural parishes located far from major urban centers within the state.
The likelihood of uneven distributions of persons included in the EHR data system can be partially addressed by the application of statistical models. Post-stratification weighting is a robust approach that has been used by similar local surveillance systems, such as MDPHnet and New York City鈥檚 Macroscope project [8, 50]. Small and rural county hypertension prevalence estimates might also be improved by both expanding partner networks to capture more patients from these areas and by using more complex modeling approaches that include adjustment for the number of clinical encounters to account for potential bias driven by underlying differences in healthcare access or use patterns [30, 32, 45, 51].
Conclusions
Surveillance of hypertension prevalence and control can identify populations at highest risk for cardiovascular and cerebrovascular disease and could potentially identify opportunities to improve hypertension control through focused and local interventions [52]. This work suggests that the use of routinely collected outpatient EHR-based data for chronic disease surveillance is feasible and can provide more timely public health metrics than survey-based approaches as well as metrics often not available to public health, such as hypertension control. In areas where MENDS has a larger patient volume, directly calculated estimates using post-stratification weighting are robust and identify demographic and geographic patterns that are consistent with BRFSS estimates of hypertension awareness. Further development of MENDS, including additional work on both network expansion to increase rural coverage and modeling approaches for calculating small area estimates for less populous counties, would increase the value of EHR-based metrics for providing insights into burden and planning public health interventions in local communities.
Data availability
The datasets generated or analyzed during the current study are not publicly available due to the underlying data being derived from confidential EHR records and due to the Health Insurance Portability and Accountability Act (HIPAA) that protects patients' sensitive health information.
Abbreviations
- ACS :
-
U.S. Census American Community Survey
- BRFSS:
-
Behavioral Risk Factor Surveillance听System
- CDC:
-
Centers for Disease Control and Prevention
- EHR:
-
Electronic Health Record
- HHS:
-
Department of Health and Human Services
- HIPAA:
-
Health Insurance Portability and Accountability Act
- IRB:
-
Institutional Review Board
- MENDS:
-
Multi-state EHR-Based Network for Disease Surveillance
- NACDD:
-
National Association of Chronic Disease Directors
- NHANES:
-
National Health and Nutrition Examination Survey
- RUCA:
-
Rural-Urban Commuting Area
- ZIP:
-
Zone Improvement Plan
References
Centers for Disease Control and Prevention. Data Modernization. . Published 2023. Accessed 17 Dec 2024.
Essential Public Health Services Futures Initiative Task Force. 10 Essential Public Health Services. . Accessed 17 Dec 2024.
Centers for Disease Control and Prevention. Surveillance systems: using surveillance systems to prevent and control chronic diseases. . Accessed 17 Dec 2024.
Ortaliza J, Amin K, Cox C. COVID-19 leading cause of death ranking. 10 November 2022. . Accessed 17 Dec 2024.
Ahmad FB, Cisewski JA, Xu J, Anderson RN. Provisional mortality data鈥擴nited States, 2022. MMWR Morb Mortal Wkly Rep. 2023;72:488鈥92.
Williams BA, Voyce S, Sidney S, Roger VL, Plante TB, Larson S, LaMonte MJ, Labarthe DR, DeBarmore BM, Change AR, Chamberlain AM, Benziger CP. Establishing a national cardiovascular disease surveillance system in the United States using electronic health record data: key strengths and limitations. J Am Heart Assoc. 2022;11: e024409.
Birkhead GS, Klompas M, Shah NR. Use of electronic health records for public health surveillance to advance public health. Ann Rev Public Health. 2015;36:345鈥59.
Perlman SE, McVeigh KH, Thorpe LE, Jacobson L, Greene CM, Gwynn RC. Innovations in population health surveillance: using electronic health records for chronic disease surveillance. Am J Public Health. 2017;107(6):853鈥7.
Hohman KH, Martinez AK, Klompas M, Kraus EM, Li W, Carton TW, Cocoros NM, Jackson SL, Karras BT, Wiltz JL, Wall HK. Leveraging electronic health record data for timely chronic disease surveillance: the Multi-State EHR-Based Network for Disease Surveillance. J Public Health Manag Pract. 2023;29(2):162鈥73.
Birkhead GS. Successes and continued challenges of electronic health records for chronic disease surveillance. Am J Public Health. 2017;107(9):1365鈥7.
Chen T, Li W, Zambarano B, Klompas M. Small-area estimation for public health surveillance using electronic health record data: reducing the impact of underrepresentation 樱花视频. 2022;22:1515.
Centers for Disease Control and Prevention. 500 cities: local data for better health. . Accessed May 18, 2017.
National Association of Chronic Disease Directors. Multi-State EHR-Based Network for Disease Surveillance (MENDS). . Accessed 17 Dec 2024.
National Association of Chronic Disease Directors. Chronic disease competencies. . Accessed 17 Dec 2024.
Kraus EK, Brand B, Hohman KH, Baker EL. New directions in public health surveillance: using electronic health records to monitor chronic disease. J Public Health Manag Pract. 2022;28(2):203鈥6.
Bower JK, Patel S, Rudy JE, Felix AS. Addressing bias in electronic health record-based surveillance of cardiovascular disease risk: finding the signal through the noise. Curr Epidemiol Rep. 2017;4(4):346鈥52.
Chan PY, Zhao Y, Lim S, Perlman SE, McVeigh KH. Using calibration to reduce measurement error in prevalence estimates based on electronic health records. Prev Chronic Dis. 2018;15: 180371.
Aliabadi A, Sheikhtaheri A, Ansari H. Electronic health record-based disease surveillance systems: a systematic literature review on challenges and solutions. J Am Med Inform Assoc. 2020;27(12):1977鈥86.
Casey JA, Schwartz BS, Stewart WF, Adler NE. Using electronic health records for population health research: a review of methods and applications. Ann Rev Public Health. 2016;37:61鈥81.
Tatem KS, Romo ML, McVeigh KH, Chan PY, Lurie-Moroni E, Thorpe LE, Perlman SE. Comparing prevalence estimates from population-based surveys to inform surveillance using electronic health records. Prev Chronic Dis. 2017;14:E44.
McVeigh KH, Newton-Dame R, Chan PY, Thorpe LE, Schreibstein L, Tatem KS, Chernov C, Lurie-Moroni E, Perlman SE. Can electronic health records be used for population health surveillance? Validating population health metrics against established survey data. EGEMS (Wash DC). 2016;4(1):1267.
Chiolero A, Santschi V, Paccaud F. Public health surveillance with electronic medical records: at risk of surveillance bias and overdiagnosis. Eur J Public Health. 2013;23(3):350鈥1.
Centers for Disease Control and Prevention. Behavioral Risk Factor Surveillance System (BRFSS). . Accessed 17 Dec 2024.
Research Action for Health Network. Homepage. . Accessed 17 Dec 2024.
National Association of Chronic Disease Directors. Multi-State EHR-Based Network for Disease Surveillance (MENDS) Governance Principles, Policies, and Processes. . Accessed 17 Dec 2024.
Louzao DM, Brown JS, Curtis LH, Marsolo K. Evaluating Foundational Data Quality in the National Patient-Centered Clinical Research Network (PCORnet庐). EGEMS (Wash DC). 2018;6(1):3.
United States Census Bureau. American Community Survey (ACS). . Accessed 17 Dec 2024.
Hohman KH, Zambarano B, Klompas M, Wall HK, Kraus EM, Carton TW, Jackson SL. Development of a hypertension electronic phenotype for chronic disease surveillance in electronic health records: key analytic decisions and their effects. Prev Chronic Dis. 2023;20: 230026.
Casey DE, Thomas RJ, Bhalla V, Commodore-Mensah Y, Heidrenreich PA, Kolte D, Muntner P, Smith SC, Spertus JA, Windle JR, Wozniack GD, Ziaeian B. 2019 AHA/ACC clinical performance and quality measures for adults with high blood pressure. J Am College of Cardiology. 2019;74(21):2661鈥706.
Zhang X, Holt JB, Lu H, Wheaton AG, Ford ES, Greenlund KJ, Croft JB. Multilevel regression, and post-stratification for small-area estimation of population health outcomes: a case study of chronic obstructive pulmonary disease prevalence using the behavioral risk factor surveillance system. Am J Epidemiol. 2014;179(8):1025鈥33.
Zhang X, Holt JB, Lu H, Onufrak S, Yang J, French SP, Sui DZ. Neighborhood commuting environment and obesity in the United States: an urban-rural stratified multilevel analysis. Prev Med. 2014;59:31鈥6.
Zhang X, Holt JB, Shumei Y, Hua L, Kurt J, Greenlund, Croft JB. Validation of multilevel regression and post-stratification methodology for small area estimation of health indicators from the Behavioral Risk Factor Surveillance System. Am J Epidemiol. 2015;182(20):127鈥37.
Barbour KE, Helmick CG, Boring M, Zhang X, Lu H, Holt JB. Prevalence of doctor-diagnosed arthritis at state and county levels鈥擴nited States, 2014. MMWR Morb Mortal Wkly Rep. 2016;65(19):489鈥94.
Zhang X, Onufrak S, Holt JB, Croft JB. A multilevel approach to estimating small area childhood obesity prevalence at the census block-group level. Prev Chronic Dis. 2013;10:E68.
Wang Y, Holt JB, Zhang X, Lu H, Shah SN, Dooley DP, Matthews KA, Croft JB. Comparison of methods for estimating prevalence of chronic diseases and health behaviors for small geographic areas: Boston Validation Study, 2013. Prev Chronic Dis. 2017;14:E99.
U.S. Department of Agriculture. Rural-Urban Community Area Codes. . Accessed 19 Dec 2023.
Centers for Disease Control and Prevention. PLACES: Local data for better help. . Accessed 17 Dec 2024.
Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion, Division of Population Health. BRFSS Prevalence & Trends Data. 2015. . Accessed 17 Dec 2024.
PLACES: Local data for better health, county data 2023 release. 25 Aug 2023. . Accessed 17 Dec 2024.
PLACES: Local data for better health, ZCTA data 2023 release. 25 Aug 2023. . Accessed 17 Dec 2023.
QGIS Development Team 2020. QGIS Geographic Information System. Open-Source Geospatial Foundation Project. . Accessed 6 Dec 2023.
Natural Earth: Free vector and raster map data. . Accessed 16 November 2023.
U.S. Census Bureau. TIGER/Line庐 Shapefiles: ZIP Code Tabulation Areas. . Accessed 17 Nov 2023.
Verheij RA, Curcin V, Delaney BC, McGilchrist MM. Possible sources of bias in primary care electronic health record data use and reuse. J Med Insternet Res. 2018;20(5):e185.
Goldstein BA, Bhavsar NA, Phelan M, Pencina MJ. Controlling for informed presence bias due to the number of health encounters in an electronic health record. Am J Epidemiol. 2016;184(11):847鈥55.
National Center for Health Statistics. National Health and Nutrition Examination Survey, 1999鈥2000 to 2017鈥2018. . Accessed 17 Dec 2024.
He S, Park S, Fujii Y, Pierce SL, Kraus EM, Wall HK, Therrien NL, Jackson SL. State-level hypertension prevalence and control among adults in the U.S. Am J Prev Med. 2023;000(000):1鈥9. . Online ahead of print.
Chobufo MD, Gayam V, Soluny J, Rahman EU, Enoru S, Foryoung JB, Agbor VN, Dufresne A, Nfor T. Prevalence and control rates of hypertension in the USA: 2017鈥2018. Int J Cardiol Hypertens. 2020;6:10044.
Cohen GR, Friedman CP, Ryan AM, Richardson CR, Adler-Milstein J. Variation in physicians鈥 electronic health record documentation and potential patient harm from that variation. J Gen Intern Med. 2019;34(11):2355鈥67.
Vogel J, Brown JS, Land T, Platt R, Klompas M. MDPHnet: Secure, distributed sharing of electronic health record data for public health surveillance, evaluation, and planning. Am J Pub Health Assoc. 2014;104:2265鈥70.
Douthit N, Kiv S, Dwolatzky T, Biswas S. Exposing some important barriers to health care access in the rural USA. Public Health. 2015;129(6):611鈥20.
Million Hearts 2027. Estimated hypertension prevalence, treatment, and control among U.S. adults. . Accessed 17 Dec 2024.
Acknowledgements
The work reported in this publication was conducted in partnership with Research Action for Health Network (REACHnet), funded by the Patient Centered Outcomes Research Institute庐 (PCORI Award RI-LPHI-01-PS1). REACHnet is a partner network in PCORnet庐, which was developed with funding from PCORI庐. The content of this publication is solely the responsibility of the authors and does not necessarily represent the views of other organizations participating in, collaborating with, or funding REACHnet or PCORnet庐, or of PCORI庐. The authors acknowledge the participation of REACHnet partner health systems in this project. The authors would also like to thank the following people for their contributions to this work: Robert Merritt, MA, and Elin Begley, MPH, from the Division for Heart Disease and Stroke Prevention, National Center for Chronic Disease Prevention and Health Promotion, Centers for Disease Control and Prevention, who provided oversight to this project; Amanda K. Martinez, a NACDD consultant, for her support and project management; and Emily W. Lankau and Sonal Pathak, NACDD consultants, who helped edit and format this publication.
Funding
The 鈥淚mproving Chronic Disease Surveillance and Management Through the Use of Electronic Health Records/Health Information Systems鈥 project is supported by the Centers for Disease Control and Prevention (CDC) of the U.S. Department of Health and Human Services (HHS) as part of a financial assistance award totaling $1,800,000 with 100 percent funded by CDC/HHS. Disclaimer: The contents are those of the authors and do not necessarily represent the official views of, nor an endorsement, by CDC/HHS, or the U.S. Government.
Author information
Authors and Affiliations
Contributions
WL, MZ, JC developed and applied the statistical analysis methodology. WL, MZ, JC, SJ, AV, KH created the initial draft manuscript. WL, SJ, AV, MK, TC, EN, LM, AA, JD, KH provided additional input to manuscript versions. All authors read and approved the final manuscript.
Corresponding author
Ethics declarations
Ethics approval and consent to participate
The need for ethics approval was deemed unnecessary according to national regulations. The CDC provided a written determination that MENDS operates within the public health authority pursuant to the Health Insurance Portability and Accountability Act (HIPAA). The legislation that outlines the exclusions for public health surveillance activities from HIPAA can be found in 45 CFR 搂164.512(b). As a public health surveillance project, MENDS does not require institutional review board approval.
Consent for publication
Not applicable鈥擳he manuscript does not contain data from any individual person.
Competing interests
The authors declare no competing interests.
Additional information
Publisher鈥檚 Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article鈥檚 Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article鈥檚 Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit .
About this article
Cite this article
Li, W., Zhang, M., Cheng, J. et al. Weighted EHR-based prevalence estimates for hypertension at the state and local levels in Louisiana. 樱花视频 25, 432 (2025). https://doi.org/10.1186/s12889-025-21633-7
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s12889-025-21633-7