Source of Clinic Data

Our research group has developed the University of Wisconsin (UW) Electronic Health Record Public Health Information Exchange (eHealth-PHINEX), an EHR data exchange between UW Departments of Family Medicine, Pediatrics, and Internal Medicine clinics (UW clinics) and the Wisconsin Division of Public Health Information Network (PHIN) of all health care visits for patients seen in the UW clinics who had at least one encounter identified by a date of service in the clinical EHR between 1/1/2007 and 12/31/2009. During this three-year time period, we follow the health records of 376,054 patients with 5 million clinical encounters and 5.6 million associated diagnoses. The database contains extracted clinical care fields, geocoding to the census block group neighborhood level, and detailed socio-demographic data. The data exchange conforms to the HIPAA limited data set privacy rule (public health is blinded to patient/provider specific information) and has been approved by the UW Institutional Review Board (IRB) Protocol # M2009-1273 and by UW Health with data use agreements. The UW eHealth-PHINEX methodology was documented previously.⁸

UW clinics are located throughout the state, but the greatest patient density is seen in South Central Wisconsin (Dane and surrounding counties, including Sauk, Columbia, Dodge, Jefferson, Iowa, Rock, Green, and Marquette). These clinics provide care for Wisconsin residents of varied socioeconomic strata in both rural and urban settings.

Source of Community Level Data

The Esri Business Analyst Premium product⁹ contains over 6,000 variables at the census block group on demographics, socioeconomic segmentation, consumer behavior, business locations and type, street data, and market potential. For this study, we examined asthma risk by median household income at the Census Block Group, as calculated by the US Bureau of the Census.¹⁰

Measures

The primary outcome of interest was current asthma prevalence. From the BRFSS, current asthma was defined by an affirmative response to the following two questions: “Have you ever been told by a doctor, nurse or other health professional that you have asthma?” and the subsequent question, “Do you still have asthma?” Patients in the UW eHealth-PHINEX dataset were identified as having current asthma by presence of the ICD-9 code 493 in either a clinic encounter diagnosis or problem list fields of their EHR.

The following covariates (pre-established risk factors for asthma) were available from both the BRFSS survey and UW eHealth-PHINEX clinical record: gender, age group, race/ethnicity, adult body mass index (BMI) in kg/m², and adult cigarette smoking. Child BMI from UW eHealth-PHINEX clinical data was categorized using BMI-for-age percentiles.¹¹ Annual household income was available from BRFSS only. Because household income was not available for the UW eHealth-PHINEX patients, we used the 2010 median annual household income estimate by census block group from ESRI⁹ in analysis.¹² A census block group is defined as a neighborhood area containing 600–3,000 people. Insurance status was available from the UW eHealth-PHINEX clinical record only. When a UW eHealth-PHINEX patient had more than one encounter in the three-year period, data were taken from the earliest encounter.

Prevalence

Child and adult asthma prevalence by select sociodemographic factors are shown in Table 2. Child asthma prevalence among UW eHealth-PHINEX patients compared to WI BRFSS respondents was not significantly different, both in terms of overall prevalence estimates (8.96 (95% CI 8.77–9.15) vs. 7.98 (95% CI 6.01–9.95) percent, respectively), as well as for the majority of the estimates by individual sociodemographic factors. However, due to the small sample size within strata, several of the WI BRFSS child asthma prevalence estimates had wide confidence intervals and relative standard error greater than 30%, which made the estimates less reliable. Smoking status, BMI and insurance status were not available for children from the WI BRFSS.

Adult asthma prevalence estimates differed significantly between UW eHealth-PHINEX and WI BRFSS data. Overall, adult asthma prevalence was lower among the UW eHealth-PHINEX population compared to WI BRFSS (7.58 (95% CI 7.48–7.68) vs. 9.41 (95% CI 8.70–10.13) percent, respectively). Males in the UW eHealth-PHINEX population had considerably lower asthma prevalence than male WI BRFSS respondents. Asthma prevalence was lower among the UW eHealth-PHINEX population's young adults (aged 18–34 years) and older adults (aged 65+years), compared to similarly aged WI BRFSS respondents. By race/ethnicity, other non-Hispanics in the UW eHealth-PHINEX population had lower asthma prevalence than WI BRFSS respondents, while asthma prevalence was similar among non-Hispanic Whites and non-Hispanic Blacks. Adult asthma prevalence within strata of household income differed only in the lowest income category. UW eHealth-PHINEX patients had substantially lower asthma prevalence in this category compared to WI BRFSS respondents (7.35 (95% CI 7.12–7.58) vs. 12.90 (95% CI 10.84–14.95) percent, respectively); however, median household income by census block group was utilized for UW eHealth-PHINEX patients, rather than individual patient household income. UW eHealth-PHINEX clinic patients covered by Medicaid had the highest asthma prevalence, compared to patients with commercial or no insurance. Insurance status was not available for adult WI BRFSS respondents.

EHR advantages

Public health data collection via telephone survey has several drawbacks in addition to low numbers and inability to assess diseases at the local level. The data is obtained by self-report which may exclude persons with undiagnosed asthma and no adjustment is made for variables related to geographic area such as race/ethnicity, which may improve disease estimates. Furthermore, low BRFSS response rates (~50%) might indicate response bias. The 2007–2009 BRFSS only sampled households with landline telephones, potentially resulting in the undersampling of certain populations due to the increasing use of cell phones. Wireless-only households tend to have younger occupants, have non-white racial backgrounds and lower incomes. Thus, the traditional public health telephone survey may not reflect the true prevalence of asthma nor highlight counties, neighborhoods (census block groups), or census tracts with the highest prevalence.

The EHR offers a rich source of high-quality population health data to study asthma or any other chronic disease. The objective diagnoses and measurements contained in clinical data can be linked with socio-demographic databases to describe risks in detail at the neighborhood level, allowing better insight to areas of interest such as where asthma is prevalent and uncontrolled. Local data can guide public health policy goals and targeting of health services delivery while providing a baseline for evaluation and quality improvement efforts.^24,25 Using the EHR can greatly increase sample size, particularly among certain age, racial and ethnic subgroups critical to community health assessments and alleviate the inherent recall and response bias of traditional telephone surveys. EHR data can also provide additional disease risk factors not found in BRFSS such as BMI, tobacco smoke exposure and insurance coverage for children.

Electronic records are readily available for epidemiological analysis to study disease control as well as perform longitudinal surveillance in a timely manner. Costs are mainly limited to disease definition, identification of outcomes, and data extraction. With the recent widespread adoption of EHR by medical providers, EHR may offer a more sustainable data source as other systems may be less available given recent and anticipated government budget cuts. Thus, clinical EHR data exchange can be a robust method of partnering public health agencies with medical care organizations to inform mutual population health priorities.^26–34 Indeed, the federal government awarded grants in 2010 to all of the states to facilitate electronic health information exchange among health care providers, hospitals, and public health.³⁵ Public health departments can work with these organizations to assure that data exchange also supports public health surveillance priorities.³⁵

EHR challenges and opportunities

There are challenges that arise in implementing a new method of disease surveillance. EHR data are limited to patients seen in participating clinics, and patients may not have a medical home within a single health system.²⁷ The EHR may have missing values and inconsistent quality, which requires use of modeling techniques to account for missing data and attention to definitions of disease used to acquire data. There is also potential introduction of bias through misclassification of patients, even when disease identification has good sensitivity and specificity.³⁶

In this study, a physician's diagnosis of asthma was the sole case definition criteria (presence of ICD-9 code 493 in encounter diagnosis or problem list fields of EHR). This may be problematic because there is no consensus on asthma diagnosis.³⁷ For example, one study compared asthma status by ICD-9 code and criteria-based medical record review. It found that ICD code-based asthma ascertainment under-identified asthma cases when compared to a `gold standard' of manual record review. The authors concluded that “ICD codes may be useful for etiologic research but may not be suitable for surveillance of asthma epidemiology.”³⁷ In our view, the problems of detection and subsequent documentation in the EHR would also likely affect self-report in the BRFSS telephone survey. In BRFSS, participants respond to the question “has a doctor ever told you that you have asthma.” But if a person is not diagnosed, it is unlikely that the physician will tell the patient that they have asthma. Thus EHR asthma cases that could found by chart review, but not ICD-9 codes, would also be cases that would be undetected by BRFSS. The BRFSS has been the mainstay for statewide surveillance of ambulatory chronic disease states. But as is the case with asthma, in many instances disease detection is dependent on self-report of physician recognition. In our study, the EHR-BRFSS prevalence estimate comparisons are for the most part remarkably similar, and the dependence upon physician recognition in both data systems may largely explain this finding. This then points to an additional advantage of the EHR and shortcoming of BRFSS. It is impossible to apply additional clinical criteria within BRFSS to find undetected cases. But along with diagnosis, other clinical indicators could be included in an EHR case definition. In this way, electronic health records may improve asthma case detection sensitivity in a way that is impossible with BRFSS. Indeed, we have a research study underway that will compare the asthma ICD-9 code only definition to one that includes additional clinical criteria present on the EHR.

Finally, EHR data are voluminous and very detailed and it is unclear how to best analyze and display these data for public health consumption.

The authors thank Brian G. Arndt, MD for his assistance in reviewing this manuscript and William R. Buckingham, PhD for generating the asthma prevalence map.

This research was funded in part with grants from The Robert Wood Johnson Foundation (Information Links; Common Ground), Centers for Disease Control and Prevention (Preparedness; Environmental Public Health Tracking; Infrastructure Improvement), and Centers for Medicare and Medicaid Services (Medicaid Transformation Grant).