Defining and Measuring Chronic Conditions: Imperatives for Research, Policy, Program, and Practice

Current trends in US population growth, age distribution, and disease dynamics foretell rises in the prevalence of chronic diseases and other chronic conditions. These trends include the rapidly growing population of older adults, the increasing life expectancy associated with advances in public health and clinical medicine, the persistently high prevalence of some risk factors, and the emerging high prevalence of multiple chronic conditions. Although preventing and mitigating the effect of chronic conditions requires sufficient measurement capacities, such measurement has been constrained by lack of consistency in definitions and diagnostic classification schemes and by heterogeneity in data systems and methods of data collection. We outline a conceptual model for improving understanding of and standardizing approaches to defining, identifying, and using information about chronic conditions in the United States. We illustrate this model’s operation by applying a standard classification scheme for chronic conditions to 5 national-level data systems.


Introduction
Current trends in population growth, age distribution, and disease dynamics foretell rises in the prevalence of chronic diseases, other chronic conditions, and combinations of chronic conditions. Such trends threaten both the public and financial health of the United States and include the rapidly growing population of older adults, the increasing life expectancy associated with advances in public health and clinical medicine, and the persistently high prevalence of some risk factors (3).
Traditionally, medical, public health, and social programs targeting commonly defined chronic diseases have focused on individual chronic diseases without considering the broader context of multiple risk factors and multiply occurring chronic conditions. Now, however, health initiatives have begun to expand to include not only chronic disease but also chronic conditions such as functional limitations; anatomic problems that are not manifestations of physical disease but are permanent or long-standing (eg, developmental disorders, limb dysfunction, visual impairment); and a broad spectrum of behavioral health problems, some of which have traditionally not been classified as diseases (4)(5)(6).
The nation is recognizing the emerging high prevalence of multiple chronic conditions (MCC) and related implications for prevention, treatment, public health programs, and planning (5)(6)(7). People who have MCC may require increased coordination of care from clinicians, public health, and social programs to improve their overall quality of life. To coordinate a national response to issues related to MCC, in 2010 the US Department of Health and Human Services (HHS) unveiled a strategic framework on MCC (6). Focus areas include monitoring the health of people who have MCC and facilitating the increased delivery of interventions, such as improved coordination of care to improve quality of life.
Preventing and mitigating the effect of any single chronic condition, or constellation of conditions, requires improved measurement. However, 2 major barriers exist. First is the lack of consistency in key definitions (eg, chronic disease, chronic illness, chronic condition) and in diagnostic classification schemes (eg, self-report, International Classification of Diseases [ICD] coding, Clinical Classifications Software [CCS]) (1, 2,8). Second are differences in data collection methods and in the design of data sets that confound efforts to characterize the epidemiology and management of MCC in different population groups in different settings. To overcome these barriers, we need a conceptual model that includes standard case definitions for individually or multiply occurring chronic conditions and guidance for applying these definitions to systems that provide data on population health. This model would assist researchers and practitioners in monitoring and studying individual chronic conditions and MCC.
In this article, we outline such a conceptual model for improving understanding of and helping to standardize approaches to defining, identifying, and using information about multiple chronic conditions in the US population. We first provide further context regarding the lack of consistency in past definitional approaches. We then describe the conceptual model, developed by an MCC working group within the HHS Office of the Assistant Secretary of Health (OASH), and detail the working group's development of a list of selected chronic conditions. To demonstrate the opportunities and challenges associated with using this set of chronic conditions, we provide an overview of 5 data systems maintained by HHS that measure chronic conditions and illustrate the model's operation by applying a standard classification scheme for MCC to the HHS data systems. We conclude by suggesting options for policy makers, public health officials, researchers, practitioners, health plans, and others to consider for improving the collection, analysis, and use of data on chronic conditions.

Variations in Defining and Classifying Chronic Conditions
Accurate case definitions are integral to public health surveillance efforts for monitoring population health and for conducting public health and clinical investigations (9). However, definitions for chronic conditions vary widely. Selected definitions (Table 1), drawn from peer-reviewed literature and other publicly available information sources, represent approaches used in academia, government, and other settings (4-6,10-16). These definitions exhibit heterogeneity in several characteristics, such as the duration or latency, need for medical attention, effect on function, pathology, departure from well-being, noncontagious nature, multiple risk factors, and nonamenability to cure. For example, most address duration and limitations in function, but only one requires the patient to have special training for rehabilitation (10).
The heterogeneity of these definitions stands in stark contrast to the process of measuring infectious conditions using established case definitions (17)(18)(19). As a result, lists of chronic conditions vary, and the accuracy and precision of estimating the magnitude of characteristics such as occurrence, burden, and associated costs are compromised.
The classification schemes currently used for identifying chronic conditions vary in origin, scope, and composition ( Table 2 [which also includes the newly developed OASH list]), and few have been applied across multiple data systems. For example, 3 systems were developed through the combined use of expert opinion and ICD codes: the Chronic Condition Indicator suggested by Hwang and colleagues identifies 185 conditions (4); the Chronic Condition Data Warehouse, developed by the Centers for Medicare and Medicaid Services (CMS), identifies 26 conditions (21); and the Hierarchical Condition Category system identifies 70 conditions (22). In 1999, the Centers for Disease Control and Prevention (CDC) and the Council of State and Territorial Epidemiologists developed a set of 73 chronic disease indicators that later was expanded to 97 cross-cutting indicators for use by jurisdictions at different levels to "uniformly define, collect, and report chronic disease data that are important to public health practice" (20). These classification schemes have been applied to specific data systems for specific purposes, such as reporting state-level data for public health agencies. However, variations in the number of conditions and array of conditions constrain comparisons of findings that result from use of different classification schemes.

Conceptual Model for Standardizing the Analysis of Health Data Sets for Selected Chronic Conditions
To standardize the analysis of health-related data sets for chronic conditions, we propose a conceptual model that involves a classification scheme consisting of 2 related dimensions: 1) identifying and specifying conditions of interest, and 2) understanding the structure of the data system of interest. The intersection of these 2 dimensions (specifically, applying a coding scheme for the conditions of interest to the elements of a data system) allows for the production of chronic condition indicators for program, research, and policy purposes (Figure). The first dimension (identifying and specifying codes for conditions) creates a classification scheme of coding rules that enable a set of specific individual conditions to be identified in data records created in a given data system. This process initially requires the specification of criteria (eg, indicators for chronicity, need for ongoing medical management, duration of effect on function) for defining chronic conditions. These criteria may then be applied to sets of health conditions to select chronic conditions of interest. Finally, the set of chronic conditions of interest must be mapped to measures that use standard coding rules and algorithms that can be systematically applied across different data systems. The coding algorithms can be data system-specific, because they are a function of the type of data available (eg, ICD, CCS, survey responses).
The second dimension (understanding data systems) is a hierarchical model that generically describes major components of data systems. The highest level is a data system, such as a surveillance system or family of related systems. Such systems, in turn, consist of component data sets that are discrete units that can be used for analysis. In the example of surveillance systems, a data set could be the data collected for 1 year. Then data sets can be deconstructed further into data elements -that is, the individual components that form a data set, typically representing an individual person or encounter (eg, clinic visit, hospital discharge) as the unit of analysis.
The point at which the 2 dimensions intersect (ie, where the coding scheme is applied to the data elements) results in the output of an indicator of the number of chronic conditions. This indicator allows researchers and others to examine variability in a variety of outcome, cost, and use measures, including mortality, associated costs, health care use, and other parameters.

Development of the OASH List of Selected Chronic Conditions
Another key issue involves the decision basis on what to include in sets of selected conditions. An example of the ramifications is that patterns of key indicators, such as MCC prevalence, services utilization, and cost indicators may vary directly as a function of the type and number of conditions. The optimal list should comprise a number of conditions sufficient to be practically useful but not overly inclusive.
To address the need for such a list, and recognizing the need for a standard classification scheme for chronic conditions, OASH used a deliberative process involving its MCC working group subject matter experts in clinical medicine, epidemiology, and public health. The goal of this process was to develop a list that would include conditions that meet the definition for chronicity, are prevalent, and are potentially amenable to public health or clinical interventions or both. The criterion for chronicity was addressed by applying the definition of "chronic condition" used in the HHS strategic framework on MCC (6). This definition, which is based on approaches adapted from other sources, states that chronic illnesses are "conditions that last a year or more and require ongoing medical attention and/or limit activities of daily living" (such as physical medical conditions, behavioral health problems, and developmental disabilities) (4)(5)(6). To produce the OASH list, the working group applied this definition and related criteria to sets of conditions used in 3 sources: 1) the CMS Chronic Condition Data Warehouse (21); 2) the list of "Priority Conditions" identified by the Agency for Healthcare Research and Quality's Effective Health Care Program (23); and 3) the Robert Wood Johnson Foundation chart book, Chronic Care: Making the Case for Ongoing Care (5).
The result of this process was an aggregate set of 20 conditions (Table 3) -each of which was listed by at least 1 of these sources and the majority of which were drawn from at least 2 of the 3 sources -that represented a practical balance of the above criteria. Identifying a manageable number of conditions helps to ensure comparability across data systems that encompass a spectrum of populations and settings. In addition, these conditions can be identified using ICD codes and applied to various data systems (Table 3), although how the conditions are coded varies as a function of data availability.

Selected HHS Health Data Systems for Studying Chronic Conditions
The component agencies of HHS maintain many privacy-protected data systems that provide information on the health and well-being of the US population. Many of these data systems include information about MCC and use of related health resources. In consultation with HHS agencies, the OASH working group selected 5 of these data systems on the basis of key criteria, including sufficiency of sample size; suitability for providing national-level, representative data; and recentness of data collection. These systems were the National Health Interview Survey (NHIS) (24,25); National Ambulatory Medical Care Survey (NAMCS) (26,27); Medical Expenditure Panel Survey (28); Nationwide Inpatient Sample of the Healthcare Cost and Utilization Project (29-31); and Medicare beneficiary enrollment and claims administrative data from CMS (21) (Appendix) ( Table 4). Details on these 5 systems are available elsewhere (21,24-31).

Application of a Common Conceptual Model to HHS Health Data Systems
The OASH working group selected codes that could be used to link the OASH list of 20 selected chronic conditions to measures in the HHS data systems. Although the CCS codes used by the Medical Expenditure Panel Survey and National Inpatient Sample data systems are based on ICD, the ICD codes used by CMS in the Chronic Condition Data Warehouse do not completely correspond with those in the CCS. For this reason, the OASH working group identified ICD codes instead of CCS codes for the CMS Beneficiary Claims Data File. The complete list of CCS codes is maintained by the Agency for Healthcare Research and Quality (31).
Three patterns describe the specificity of the mapping for the selected conditions. The first pattern is characterized by the presence of a measure for a condition in each data system. For example, a measure for hypertension is in all 5 data systems. For this pattern, the data elements reflect various sources: for example, in NHIS, respondents provide the self -reported diagnosis for each condition, whereas in NAMCS, data are collected for both the reason for the current visit and for a checklist of ever existing conditions. However, not all data systems measure all 20 conditions: NHIS measures 10, NAMCS measures 19, and CMS measures 15; both the Nationwide Inpatient Sample and the Medical Expenditure Panel Survey measure all 20.
In the second pattern, although a 1-to-1 match was not found, related conditions could be mapped onto the same general condition described in the OASH list. For example, although NHIS does not have a specific question on chronic kidney disease, it does have questions on weak or failing kidneys, which could be mapped to chronic kidney disease.
For the third pattern, data in a given system could not be mapped to the condition identified in the OASH list. For example, data on congestive heart failure, cardiac arrhythmias, hyperlipidemia, dementia, and depression are not collected by NHIS, although data on these conditions are collected by the other 4 data systems. Other conditions for which data are not available in NHIS include autism spectrum disorder, HIV, osteoporosis, schizophrenia, and substance use. For NAMCS, data are not available for chronic kidney disease; and for CMS, for autism spectrum disorder, hepatitis, HIV, schizophrenia, or substance abuse disorders. Although claims data may be available from CMS, they are not now available in the analytic data sets.

Summary
As the prevalence of chronic conditions continues to increase in the US population, the United States will face even greater challenges in delivering care to people with MCC (32-35). Accurate, reproducible, and understandable measures of the occurrence and impact of MCC will be an important part of the solution for these challenges. Such measures can help in improving surveillance, program planning, targeting and evaluating interventions, and other essential activities. More accurate and reliable data on individual chronic conditions and on MCC are also foundational in enabling health systems and providers to target, measure, and ultimately improve population outcomes.
As this article has shown, improvements in measurement require that we first improve methods for characterizing and monitoring chronic conditions, including achieving common agreement on the meaning of the terms "chronic condition" and "multiple." Our review of existing definitions showed not only how existing definitions differ but how these differences affect research and practice, including difficulties in comparing results of studies and the prevalence of MCC across various data systems. Although some commentators have defined "multiple" as the presence of 2 or more conditions in an individual (4,5), further study of the number of conditions and specific diagnoses may be improved by setting thresholds that are tailored to inform clinical practice, public health programs, and policy strategies.
Although the absence of standard case definitions for chronic conditions poses major challenges for uses across data systems, existing definitions and classification schemes might be applied more easily across multiple population subgroups within a given data system. For example, noting in the early 1990s the inherent limitations of conditionspecific approaches to classifying chronic conditions among children, commentators associated with a research consortium on chronic illness in childhood pointed to the need for a widely applicable, but modifiable, definition of chronic conditions for use in research, program development and delivery, and development of health care policies (36). This approach, and similar conclusions by other investigators (2), although specific to children, bears relevance also to adults, even though the epidemiology of MCC varies by population group.
Our conceptual model provides a framework for more consistently applying lists of selected conditions to multiple data systems. For the OASH list of selected chronic conditions, the model explicitly documented data elements that were used to identify the selected conditions and how the data were collected and coded. This conceptual model can be used to document coding decisions that are applied to additional data sets, an especially important need when multiple data systems are used to examine the burden of chronic conditions. Although this model may be useful for improving the consistency in research and programs that address MCC, other opportunities allow for refining sets of conditions. For example, a rigorous measure development process that applied decision rules to data from multiple systems on key parameters (eg, the prevalence of different conditions and their effect on functional status, use of services, and costs) could assist in refining sets of conditions for analysis. Additional analysis to determine the optimal number of conditions also could help in refining measures of the impact of chronic conditions in the US population.
Although this article focused on consistency in defining and classifying chronic conditions, an important related issue is the coordination between essential actors involved in developing and using data, including coordination on methods for establishing classification schemes (ie, who does this, by what means, and how often). Deciding on the number of chronic conditions to include in a given list and addressing implications for key parameters (ie, measured prevalence, use, and cost) require a combination of clinical acumen and expertise in use of surveillance data. Thus, the gaps identified in this article help to sharpen focus on the need for collaboration among different organizations, agencies, and institutions at different levels (ie, national, regional, state, and local) that collect data and maintain data systems and that may benefit from using a common conceptual model and classification scheme. Beyond data managers, analysts, and researchers, other stakeholders need to engage in the process, including practitioners and policy makers, who can provide valuable input to guide analysis of the most pressing needs for data on chronic conditions. Researchers, practitioners, and policy makers can consider using the issues identified in this article as the basis for improving the collection, analysis, and use of data on chronic conditions in the United States. Foremost, the examination of different classification schemes and their application to multiple data systems suggest that the terms "chronic disease" and "chronic illness" might be supplanted by wider adoption of a functionally more accurate and inclusive term, such as "chronic conditions." Greater consistency in and more complementary use of classification schemes for chronic conditions hold the promise for improving research and generating a stronger knowledge base for policy makers and program managers.

Definition
We defined a person as having a chronic condition if that person's condition had lasted or was expected to last 12 or more months and resulted in functional limitations and/or the need for ongoing medical care.

Key components
Duration: ≥12 months Functional limitation: yes Need for ongoing medical care: yes Comments Authors noted that they defined "chronic condition" broadly for several reasons, including the following: 1) a high proportion of individuals who have a chronic condition have more than 1 chronic condition; 2) functional limitations and other consequences of health problems often are independent of specific diseases; and 3) whereas diagnoses are important for medical management, a diagnosis alone may provide incomplete information on morbidity because of variations in condition-specific severity.