Electronic Health Records

An enormous potential for surveillance exists in electronic health records (EHRs) from clinical providers. These records provide the most detailed health care data available, including examination and testing outcomes, diagnoses, procedures conducted, and services utilized (Rein, 2015). Public health programs in particular have the opportunity to benefit greatly from this newly available electronic health information, which has the potential to become a crucial component of surveillance. Public health practitioners need to leverage this electronic revolution by being innovative in their use of new and existing electronic data sources. For health departments, the use of EHR systems may be the next frontier in chronic disease surveillance (Maylahn, 2013). However, while EHRs present a huge opportunity to understand local conditions, access to these databases is often limited and can be proprietary. Local health department staff may not be able to make use of the data or may not have staff who can spend time analyzing the data.

The American Academy of Ophthalmology launched an ophthalmology electronic clinical registry, IRIS^TM (Intelligent Research in Sight), in 2014. The IRIS registry collects uniform data, both clinical and patient-reported information, that can be analyzed to inform clinical decision making. IRIS captures data from 80 percent of U.S. ophthalmology practices (Parke, 2015). As of January 2016, the IRIS registry represented data from 14 million unique patients (Parke et al., 2016).

Similarly, the American Optometric Association's Measures and Outcomes Registry for Eyecare (MORE), a clinical data registry launched in 2015, integrated EHR data from multiple systems to facilitate reporting for Centers for Medicare & Medicaid Services' (CMS's) Physician Quality Reporting System, and created an evidence base for future clinical decisions and research directions (AOA, 2015).

CMS has approved both IRIS and MORE as Qualified Clinical Data Registries, making substantial data available not only for quality reporting and CMS initiatives such as the EHR incentive program or the Value-based Payment Modifier program, but also for research utility through tracking outcomes and comparing clinical practice data (CMS, 2016a). However, Rein has cautioned that differences may exist between the clinical practices represented in the database and other populations of interest (i.e., populations without access to treatment) (Rein, 2015). In addition to IRIS and MORE, there are other commercial and not-for-profit sources of EHR data available, such as Epic Systems, though few of the data are related to vision or eye health (Rein, 2015).

Administrative Claims

Administrative claims from private insurers, Medicare, or Medicaid are a source of surveillance data about vision and eye-related health status and care utilization. Claims data rely on billing codes that indicate whether a service or procedure was performed or whether a patient received a diagnosis. These data are limited by the lack of detail inherent in billing codes and the possibility of human error in coding (Rein, 2015). Diagnostic codes may be too broad or too narrow, leading to under- or over-reporting of eye conditions (Elliott et al., 2012). However, administrative claims have many advantages versus other data sources. Data may be available over time for the same patient, provided that he or she remains enrolled in the same insurance plan (Rein, 2015). Large care providers, such as CMS or the U.S. Department of Veterans Affairs, have data that are representative of the subpopulations they serve (Elliott, 2012). Data from private insurers are not necessarily representative of the general population, but there are a few sources of data that include many patients. These include the MarketScan^TM research databases, which are commercial products offering fully integrated patient-level data from more than 170 million patients, and claims from Vision Service Plan (VSP) vision and eye care insurance, which include acuity readings and cover 77 million patients (Truven Health Analytics, 2016; VSP, 2016).

Medicare data in particular have utility for surveillance. Medicare covers nearly all of the over-65 population. Medicare claims provide longitudinal data, because people generally stay enrolled in Medicare from age 65 on. The large amount of data available in Medicare claims allows for examination of lower-incidence conditions (Elliott et al., 2012). Medicare data have been used for a number of studies, including an analysis of 9 years of data to estimate how many beneficiaries with diabetes and chronic eye diseases receive annual eye exams (Lee et al., 2003).

Patient information from EHRs or administrative claims as a source of surveillance data has an advantage over data from surveys or research studies because it is real-world data collected in real time: it captures diagnostic information, how care is actually being provided, and the outcomes of that care (Elliott et al., 2012). However, these data are only collected when patients present themselves for care and therefore do not include information about the people who cannot or will not use clinical care. In addition, no one source of patient data covers the entirety of eye care. Services may be provided by optometrists, ophthalmologists, or general practitioners, and vision care is often paid for out of pocket or with separate vision insurance (Wittenborn and Rein, 2016). This division among multiple health and payment systems means that in order to get the full picture of eye care, patient data will have to be gathered from multiple sources.

Scope and Quality of Data

Vision surveillance suffers from the limited scope of data that are available at the national, state, and local levels. National and state surveys and patient data systems collect some information about the presence of eye disease and vision impairment, but they fail to measure many other factors linked to visual impairment or eye disease. None of the surveillance methods discussed above measures the total burden of vision impairment and blindness, which includes the extent of visual acuity loss; the presence and stage of specific eye disease, comorbidities, and health-risk behaviors; social or physical limitations imposed by low vision; quality of life (QOL) measures; and the existence of barriers to accessing health care and rehabilitation services. Most surveillance focuses primarily on prevalence and incidence, and instruments do not usually measure factors such as access to care, barriers to the use of care, health literacy, or access to assistance, all of which may greatly affect the health and QOL of a person with visual impairment or eye disease (Wittenborn and Rein, 2016).

The lack of standardized measures of vision impairment, eye health, and related health indicators also impairs the collection and analysis of surveillance data. No gold standard method of measuring vision impairment or the presence of eye disease has been developed or evaluated (Crews, 2015). Furthermore, current evaluation methods fall short of providing a comprehensive understanding of the impact of vision impairment and blindness. For example, the most commonly used instrument to assess vision-related QOL in the eye health literature is the National Eye Institute Visual Function Questionnaire (VFQ-25), although many other measurement approaches and instruments are in use (Heintz et al., 2012; Hirneiss, 2014; Mangione et al., 2001). Psychometric flaws have been noted concerning the VFQ-25 in its original structure, and other versions have been developed (Kowalski et al., 2012; Pesudovs et al., 2010). While generic QOL instruments capture some aspects of vision-related QOL, scales that only include items related to visual acuity or central vision are unlikely to capture the QOL impact of visual field deficits.

Consequently, surveys do not use the same metrics, and the ones that they use are not necessarily correlated with the clinical definitions coded in EHR and claims data. Standardized measures are difficult to implement in part because vision loss is not an all-or-nothing concept. The loss of sight occurs across a continuum, with varying degrees of impairment and functionality. Vision may be assessed with any number of measures, including acuity, contrast sensitivity, visual field, or night vision (Wittenborn and Rein, 2016). There is no consensus on how vision impairment should be defined, so it is measured in different ways in different surveillance instruments (Wittenborn and Rein, 2016). Likewise, eye conditions can vary from minor to fatal and can be short-lived or chronic; this complexity makes it difficult to identify which conditions to include in surveillance and how. Because visual impairment and eye disease are often undiagnosed, a surveillance instrument that relies on self-reported diagnosis to identify people with eye disease may underestimate prevalence (Wittenborn and Rein, 2016). One population-based study analyzing the agreement between self-reported data and medical record for prevalence of eye disease and eye care utilization suggested national estimates tended to underestimate rates of eye disease and overestimate eye care utilization (MacLennan et al., 2013).

Limited Local Data

Although national surveillance collects information about vision impairment and eye health, there are limited data available at the local level, where interventions and policy decisions are often developed and implemented. The BRFSS provides state estimates, but the data are not broken down further by county or municipality, while other national surveys such as NHIS and NHANES provide only national or regional estimates (IOM, 2011). Increasingly, states and local communities are investing resources for data collection at the county level or in local jurisdictions. This trend has accelerated with the new requirements in the Patient Protection and Affordable Care Act of 2010 that tax-exempt hospitals must conduct an assessment of the health needs of their catchment areas and begin to address them (Rosenbaum and Margulies, 2011). This requirement builds on current best practices of hospitals and hospital systems, which include the strategic investment of resources and the building of community partnerships, including local groups and leaders in public health, with the goal of improving overall community health. Every 3 years, hospitals must conduct community health needs assessments (CHNAs). These coincide with assessments done by local health departments and others in the community. The goals of these assessments are to “identify existing health care resources and prioritize community health needs” (Stoto, 2013, p. 4). Each hospital also must develop an implementation strategy to meet the needs identified by its CHNA and a set of performance measures to track progress (Stoto, 2013). These actions must be reported to the Internal Revenue Service using Schedule H of Form 990 as an obligation for tax exemption. At the same time, the Public Health Accreditation Board established prerequisites of national accreditation for local health departments, including having completed a community health assessment, community health improvement plan, and an agency-wide strategic plan within the past 5 years (CDC, 2016d; PHAB, 2016; Riley et al., 2012). Data about the risk factors for eye disease and vision loss could be collected as part of these assessments, but eye-related issues must compete against the constellation of other public health problems present in communities.

One opportunity to improve the availability of local data involves combining data from community surveys with data from state and local surveillance systems to characterize patterns of risk, health outcomes, and relevant determinants. An example is the public health surveillance of cancer, in which data about cancer risk factors can be supplemented with information from cancer registries. Using geographic information system methods, data from different sources can be linked to help guide local cancer screening and treatment programs targeting high-risk or underserved communities (Birkhead and Maylahn, 2010). For vision surveillance, survey estimates of vision indicators can be augmented with clinical information about eye disease to produce a more complete picture of the eye health in a local community.

Availability of Data

Because public health surveillance is motivated by the public's concerns about a health issue, its impact on various groups, and what can be done to address it, surveillance data need to be made available, analyzed, and disseminated in a timely manner. However, as discussed above, some surveys collect and disseminate data infrequently. Furthermore, data on eye and vision health are often inaccessible to those who are not subject-matter experts, and it may be difficult for a policy maker or member of the public to easily find accurate estimates of vision impairment or eye disease (Wittenborn and Rein, 2016).

Populations Not Captured

The surveillance of visual impairment and eye disease, whether conducted by surveys or patient information databases, inevitably fails to include or represent some groups of people. For example, surveillance that relies on patient data, as found in EHRs or administrative claims, excludes people who are outside of the traditional clinical health system and who may be uninsured, underserved, or part of at-risk populations. People who cannot speak English may be underrepresented in surveys. While most national surveys that are conducted via interviews accommodate non-English speakers through the use of translators or bilingual interviewers, many of these services are restricted to the Spanish language or are only available in areas containing large numbers of non-English speaking respondents (Islam et al., 2010). Low-resource communities—including rural, homeless, and undocumented populations—may be excluded from surveys and studies because they are difficult to contact. For example, the BRFSS, because it collects data by telephone, necessarily samples only those people who have access to a telephone. There are myriad ways in which surveillance methods fail to capture some populations; new or improved surveillance instruments should be designed to mitigate this as much as possible.

Lack of Integration Between Clinical and Public Health Data

Data from surveys and other population-based surveillance conducted by federal, state, and local public health agencies can contribute relevant information about eye health, especially for the country as a whole. Likewise, there is an abundance of data collected during patient health care visits that can be accessed from EHRs or administrative claims. Integrating these surveillance systems would allow for a more comprehensive picture of visual health. However, major constraints prevent this information from being aggregated or shared to create richer and more useful data. For example, there are technical constraints, including software incompatibility and lack of interoperability of datasets. Commercial and legal constraints affect data sharing and availability. Perhaps most importantly, coordination and collaboration among surveillance actors are limited. As a result, the nation lacks standard metrics for measuring vision impairment and eye health, there are limited sources for baseline data that can be used in surveillance and research, and significant hurdles stand in the way of integration or sharing between surveillance systems. Data that could be used to improve public health practice and clinical care are instead fragmented and siloed (CDC, 2011a).

Modest Funding Support

Vision and eye health surveillance efforts have received only modest attention and support from federal agencies. Several federal programs focus on eye health, including the National Eye Institute (NEI) and the National Eye Health Education Program (NEHEP) at the National Institutes of Health (NIH), and the Vision Health Initiative (VHI), which is housed in the Division of Diabetes Translation at the Centers for Disease Control and Prevention (CDC). These programs tend to emphasize research and education rather than surveillance activities. More recently, the NIH has begun to focus more heavily on genomics and personalized medicine than on population-level health. While these NIH research priorities are valuable in generating new knowledge, they may not offer the same potential impact as population-wide interventions and more traditional public health approaches to disease, which include the monitoring and tracking of diseases to identify populations at risk and how interventions affect subgroups.

Over the past several years, funding for eye-related activities at the NIH and the CDC has remained fairly constant or has dropped slightly. The NEI funding,¹ representing about 2 percent of all NIH funding, has remained around $700 million since 2010 (NIH, 2015). At the CDC, since 2010 the annual budget for the VHI has declined in both total amount and as a percentage of the overall CDC budget. The proportion of the VHI budget allocated to glaucoma-specific activities has remained relatively constant—at 86 percent—since 2011 (CDC, 2011b, 2012, 2013, 2014, 2016b). Funding and staffing for community surveillance activities is limited as well. Funding for local health departments has been cut on the federal, state, and local levels in recent years, and as a result, thousands of public health jobs have been eliminated (NACCHO, 2015).

However, integrating vision and eye components into the Multi-Ethnic Study of Atherosclerosis (MESA) was a way to “conduct surveillance by leveraging a relatively small investment of funds to build on an extensive existing study infrastructure” (Cotch, 2016, p. E1). MESA was originally designed as a multicenter community-based study on cardiovascular disease, risk factors, and outcomes in four different ethnic and racial groups (MESA, 2016a). Retinal photography and a refraction assessment were added to the study, and the inclusion of these components introduced access to the data collected from the rest of the study on comorbid conditions, genetic profiles, behaviors, and multiple health outcomes, with the potential of providing a broader, comprehensive picture of health (Cotch, 2016; MESA, 2016b). Research from the initiative included a study that published prevalence rates of AMD by age, gender, and four racial and ethnic groups using objective, clinical measures as the means to collect data (Klein et al., 2006).

Analysis of Surveillance Data

Surveillance data, such as data from surveys, can be used to answer research questions by combining data from studies that compare vision and eye disease to other factors such as socioeconomic status, race and ethnicity, or geographic area. For example, Chou and colleagues (2012) used several years of BRFSS data in order to assess the prevalence of annual eye care among diabetic adults who were visually impaired, analyzed by state, race and ethnicity, education, and annual income. Participants were defined as “visually impaired” if they indicated that they had moderate or severe difficulty reading print or recognizing a friend across the street. Zhang and colleagues (2012) combined data from NHANES and NHIS in order to explore disparities in vision health by race and ethnicity, education, and economic status. Kirtland and colleagues (2015) analyzed American Community Survey data to describe the geographic pattern of severe vision loss and to assess its association with poverty level. These types of analyses allow researchers to stretch the utility of survey data by combining and comparing data across years and surveys. However, the ability to conduct meta-analyses is constrained because of the limitations of surveys discussed above: metrics that vary from survey to survey, reliance on self-reported data, and infrequent data collection and reporting.

Population-Based Community Studies

Population-based community studies are the “backbone of eye disease epidemiological knowledge” (Wittenborn and Rein, 2016). This type of research collects detailed information about a specific population. The population can be defined in a variety of ways—for example, by geographic area (e.g., Framingham Heart Study or Baltimore Eye Survey), ethnic group (e.g., Chinese American Eye Study), or occupation (e.g., Nurses' Health Study)—and the population may be studied either in its entirety or by a representative sample. Population-based community studies can be either ongoing and long-term, or one-time assessments of a population.

Ongoing population-based community studies can provide valuable insight into the life course of a disease and the relationship of risk factors to disease outcomes. These prospective cohort studies collect a wide variety of data over a period of time from the same population. One of the longest-running eye studies is the Beaver Dam Eye Study, which began in 1988 with baseline examinations of nearly all of the 43- to 84-year-old residents of Beaver Dam, Wisconsin (about 5,000 initial participants) (Beaver Dam Eye Study, 2014). Researchers conducted follow-up examinations every 5 years on all cohort members (more than 1,900 in each follow-up period), resulting in more than 300 publications of the study findings. The study observed the natural history of several eye diseases, tracked the decline in vision as participants aged, and measured the relationship between eye conditions and long-term exposures such as blood pressure levels (Beaver Dam Eye Study, 2014). The Rotterdam Study in the Netherlands, another population-based community cohort study, has been collecting information on thousands of residents of Rotterdam since 1990 and is focused on cardiovascular, neurological, endocrine, and ophthalmological diseases (Hofman et al., 2007).

In addition to these long-term community studies, researchers also conduct one-time assessments of specific populations. For example, the Chinese American Eye Study sought to understand the prevalence of visual impairment and eye disease in Chinese Americans living in Monterey Park, California. Each of the 4,570 participants completed a questionnaire about health- and eye-related behaviors, risk factors, and their QOL, and underwent a clinical examination (Varma et al., 2013).

The main limitation to population-based community studies is the potential for differences between the study population and the general population and thus a lack of generalizability (IOM, 2011). Because the population studied is usually a specific group of people, rather than a representative sample of the general population, data about prevalence or risk factors may not be applicable to people outside the group. Moreover, respondents may differ from nonrespondents even in the population studied. However, these types of studies also have significant benefits. The data are usually based on examinations rather than on self-reporting, so prevalence and incidence numbers may be more accurate (Wittenborn and Rein, 2016). Using a prospective cohort allows researchers to draw correlations between disease outcomes and behavioral or environmental factors, while minimizing selection and recall bias (Kukull and Ganguli, 2012). In addition, in those cases when the participation of the cohort is ongoing, researchers can return to the cohort repeatedly for clinical examinations, biomarker measurements, or new areas of investigation. This study method can be optimal for collecting information about a minority group. For example, because the Chinese American Eye Study selected its participants from the U.S. city with the highest percentage of Chinese Americans, it was less time- and resource-intensive than drawing a representative sample of Chinese Americans from across the United States. While the results may not be generalizable to the entire U.S. population, the information may be used to estimate prevalence and health issues in the U.S. Chinese American population using census data (Rein, 2015).

Meta-Analysis of Population-Based Studies

The meta-analysis of population-based studies can provide insight into links between vision and eye health and other factors. The findings of individual population-based studies may not be applicable to the general population; however, if the results can be replicated across multiple studies, the validity and generalizability are strengthened (Kukull and Ganguli, 2012). In order to replicate findings, the studies must use the same standardized measurements and outcomes so that the data can be compared. The Beaver Dam Eye Study developed an imaging system and a standardized scale for the severity of certain eye conditions that were subsequently used in several other population-based community studies. As a result, it has been possible to compare data among these multiple studies and draw more robust conclusions from the findings. For example, the Beaver Dam Eye Study found an association between smoking and the development of cataract and AMD. Two other ongoing studies using the same system and scale—the Blue Mountains Eye Study in Australia and the Rotterdam Study in the Netherlands—found the same correlation, despite differences in the populations (The Board of Regents of the University of Wisconsin System, 2014). Similarly, a collaborative study between the Multi-Ethnic Pediatric Eye Disease Study and the Baltimore Pediatric Eye Disease Study used the same protocol and measurements to collect information on vision impairment from more than 10,000 children. The population studied lived in Los Angeles or Baltimore and included large percentages of minority children (44 percent African American and 32 percent Hispanic) (NEI/NIH, 2011). By using the same protocol on a large, diverse study population, researchers increased the generalizability of the results.

Vision Problems in the U.S. (VPUS), a project of Prevent Blindness America, is a meta-analysis of 12 population-based studies, including the 3 discussed above. VPUS uses the data from these studies, along with census data, to develop age-, race-, and sex-specific prevalence estimates on both the national and state levels (VPUS, 2012). The VPUS online database allows users to customize a research inquiry; for example, a user could compare glaucoma rates in black females between three different states, or look at vision impairment by age in California. VPUS has estimates for four eye conditions (AMD, cataract, diabetic retinopathy, and glaucoma) and four categories of vision (vision impairment, blindness, hyperopia, and myopia). While VPUS provides valuable, easy-to-understand information about vision and eye disease, there are several limitations to the data. The studies on which VPUS relies may not be representative of the current U.S. population—all of the study populations were from small geographic areas and were not probabilistically sampled, 5 of the 12 studies are international, and some of the data are up to 30 years old (Wittenborn and Rein, 2016). Furthermore, the scope of the VPUS data is limited, with data on only four eye disorders and no information on rates of under-diagnosis of eye conditions (Wittenborn and Rein, 2016).

Population Health Administrative Databases

Databases that link information from different sources at an individual level can be a useful tool for population-based research. Population health databases are used to track the health of a population of interest based on agreed-upon metrics. Although these databases rarely include measures specific to eye health, they may contain measures that are risk factors for poor eye health. For example, America's Health Rankings includes indicators on diabetes, high blood pressure, and smoking, all of which may affect eye health. These types of databases, which also include Healthy People 2020, Community Health Status Indicators, and data collected by the Trust for America's Health, can be linked to eye-specific databases in order to answer research questions about relationships between eye health and environmental, behavioral, or other risk factors.