PHM Glossary: C

Care Coordination/Care Management

In population health management, care coordination (also called care management) can take on two related but slightly different meanings. The first relates to integration across the providers of care. In this construct, care coordination is the process linking those individuals with chronic health care illnesses/conditions and their families with appropriate education, health care services, and resources in a cohesive and rational way so as to maximize health care outcomes in an efficient manner.

The second meaning relates to interrelationships across the spectrum of health oriented strategies—from primary prevention, to acute care, to chronic care management to end-of-life care. Population health management programs aim to fill a vital role in the process of care coordination by adding efficiencies and connectivity to an otherwise loosely affiliated health care system.


The concepts of coordination of interventions and collaborative practice models are explicitly included in the Care Continuum Alliance definition. Within the context of Health Insurance Portability and Accountability Act of 1996 (HIPAA) and Centers for Medicare and Medicaid Services (CMS), treatment is the provision, coordination, or management of he[alth care and related services by one or more health care providers, including the coordination or management of health care by a health care provider with a third party; consultation between health care providers relating to a patient; or the referral of a patient for health care from one health care provider to another.

Chronically ill patients may see multiple practitioners who are not fully connected and therefore not completely informed of key aspects of the patient’s history or care provided elsewhere. Coordination of care for the chronically ill can be difficult because there is no central system for tracking or linking patients, their diagnoses, providers, treatments, and pharmaceutical use.

This is further complicated by funding constraints and the presence of both public and private payers; community services may also provide certain services for some patients. Socioeconomic and cultural barriers may further hinder synchronization of these disparate providers, payers, and community care givers. Hence, care management has a role to play in linking these disparate elements into a cohesive care management system.


The term care coordination has been used for more than 20 years in relation to patient care, and was conceived with the notion of improving coordination across providers and the spectrum of health care.

More recently, The Joint Commission defined care coordination as a process to manage and coordinate health care resource use in the provision of care and services. Utilization Review Accreditation Commission expanded this to describe the processes that are coordinated—“a collaborative process that assesses, plans, implements, coordinates, monitors, and evaluates options and services to meet a consumer’s health needs through communication and available resources to promote quality, cost-effective outcomes.”

The National Committee for Quality Assurance defines clinical care coordination as the mechanisms ensuring that a member and clinicians have access to and take into consideration all required information on the member’s conditions and treatments, to ensure that the member receives appropriate health care services.


While it depends upon the setting and the design of the individual chronic care management program, care managers can promote care coordination in a variety of ways. In discussions with patients, care managers identify and track health care and other services by patients.

This data can then be accumulated and communicated by the care manager to physicians and other health care providers working with the patient. Care managers often prepare patients for meetings with specialist by helping them to identify and obtain information from their primary care doctor that will be useful for the specialists. Care managers are often a source of information for patients on community services such as smoking cessation classes or exercise programs that will help patients better manage their chronic conditions.

Coordination of chronic care is based on open lines of communication and information transfer. Information technology is incorporated into many chronic care management programs with the aim of increasing communication across the health care delivery system, chronic care management program and the patient/care giver. The linkages offered by chronic care management produce efficiencies and encourage higher quality via fully informed care management.


Nurse call lines and static Web sites Disseminate Information
Interactive Web sites and telehealth Two-way communication for enrollment, education, monitoring, and creation of an information plattform
Biometric devices Home-based measurement
Handheld devices Daily monitoring - patients and care manager track daily progress
Connectivity/work flow management Real-time alerts, reminders, and guidelines delivered at point of care


“Care Coordination: Integrating Health and Related Systems of Care for Children with Special Health Care Needs.” 4 Oct. 2002. American Academy of Pediatrics: Committee on Children with Disabilities. 5 May 2006;104/4/978

“Care Management/Care Coordination.” 26 Nov. 2002. California Department of Health Services. 16 Feb. 2006

“HIPAA’s Definition of Terms.” Privacy and Research: HIPAA. 16 May 2003. Boston University School of Medicine. 16 Feb. 2006

Paschal, R., ed. “Arizona Administrative Code.” Arizona Secretary of State. 2006. 16 Feb. 2006

Care Plan/Care Path

These terms are frequently used interchangeably to refer to a series of care and care management activities, as well as the person responsible for the activity and the goal to be achieved. Care plans are derived from evidence-based medicine but are individualized for the needs of the particular patient.


Population health management programs have care plans designed to meet the needs of patients with individual or multiple chronic conditions. These plans are derived from evidence-based medicine and national or local care guidelines. Unlike critical pathways, the care plan (in a population health management program) looks at care for an individual patient over a long time frame managing multiple conditions and activities. Care managers develop individualized care plans based upon initial patient assessments and modify these plans as the patient progresses.

Population health management programs often use automated systems that direct initial assessments, stratify patients, and establish goals consistent with the severity of their disease and the appropriate practice guideline. These systems generally support the care managers in identifying interventions to help patients achieve established goals, overcome barriers, and track progress.


Taking a 60-year-old male with heart failure, diabetes, and hypertension as an example, the care manager will assess this patient for all the conditions, taking advantage of the fact that there are commonalities across all three of the diseases. Assessment information might include general data such as demographic information, current medications, allergies, diet, exercise and smoking, as well as disease-specific information such as ejection fraction for heart failure or HbA1c measures for diabetic patients.

The care manager will work with the patient to identify opportunities to improve management of these chronic conditions and develop the care plan to set goals, coordinate supportive interventions, and monitor care processes. For example, this patient may need help identifying when heart failure symptoms require immediate medical attention, overcoming barriers to following a low-salt diet; adhering to prescribed medications, and using a blood pressure cuff. The care plan might also specify other educational discussions or materials to address behaviors such as smoking or lack of exercise.

Behind the scenes, the automated system might support other guidance functions by alerting the care manager that the patient with heart failure is not taking an angiotensin converting enzyme (ACE) inhibitor (and appears to have no contraindications). A rule or algorithm would trigger an alert to the care manager or the physician and perhaps to the participant. Another rule might be triggered if this patient has not had an eye exam in the past 12 months. Care managers might be prompted to call the participant to follow up on these conversations if claims or participant-supplied evidence for an ACE inhibitor (or angiotensin receptor blocker) does not appear within 30 days.


Johnson, M., et al. (eds.) “Nursing Diagnoses, Outcomes, and Intervention.” 2nd ed. Philadelphia: Elsevier Health Services(2005).

Sox, H. Comprehensive Care Planning for Long Term Care Facilities: A Guide to Resident Assessment Protocols (RAPs) and Interdisciplinary Care Plans. Volume 1. Worthington, OH: Robin Technologies, Inc. (2004).

Case Management (Care Management)

The Case Management Society of America defines case management as a collaborative process of assessment, planning, facilitation, and advocacy for options and services to meet an individual’s health needs through communication and available resources to promote quality cost-effective outcomes.

The case management process includes assessment, problem identification, outcome identification, planning, monitoring, and evaluating.


Case management services often focus on individual patients with acute and/or catastrophic conditions. Case managers assess the patient’s needs and, based upon this assessment, identify and coordinate needed services. Case managers monitor the patient’s progress, identify barriers to quality cost-effective care, and facilitate overcoming these barriers.

For example, a patient has been hospitalized following a serious automobile accident. The case manager will assess the patient’s needs both in the hospital and upon discharge. They may coordinate home care resources and advocate for insurance benefit exceptions that might help the patient to be discharged from the hospital sooner while still receiving optimal care. There is a significant overlap between disease and case management.


Standards of Practice for Case Management. Little Rock, AR.: Case Management Society of America. (2002).

Dove, H., and Duncan, I. “Estimating Savings, Utilization Rate Changes and Return on Investment from Care Management Interventions.” An Introduction to Care Management Interventions and their Implications for Actuaries.

CMSA Home Page. Case Management Society of America. 16 Feb. 2006

SOA Home Page. Society of Actuaries. 16 Feb. 2006


Causality is the ability to clearly prove that an intervention is truly responsible for a certain outcome. While it is difficult to demonstrate causality with experimental evidence in a single study, the strength of a causal association underpins the ability to state whether the intervention (e.g., population health management program) actually caused the outcome observed.


Causality (i.e., a causal association or a cause-and-effect relationship) is assumed when one factor is shown to contribute to an outcome or result (e.g., disease or admission to hospital) and its removal (or absence) is shown to reduce the frequency of the outcome or result.

Causality is imputed (but can never be proven) when there is sufficient evidence that an intervention is responsible for a given outcome—that the outcome would not have occurred but for the intervention. Some study designs are more useful in supporting a claim that a reported association is causal.

Outcomes associated with a population health management program include reduced utilization (or increased utilization of beneficial services, such as increased use of ventilators for asthma), better clinical outcomes and/or reduced medical expenditures.

Each of these may be attributable to chance, systemic errors (bias), confounding factors, or artifactual association. Confounding factors that may affect assumptions of causality include changes in benefit design over time or different levels of disease severity between the control group and those receiving the intervention (i.e., population health management). This is a serious concern because study findings are essentially unusable if it is not possible to reasonably conclude that the outcomes being measured were actually caused by the population health management program.


The Wilson and McDowell and Care Continuum Alliance papers approach outcomes measurement from the perspective of multiple users, both scientific and business. The scientific emphasis tends to demand a higher degree of support for a causal association than is likely to be encountered in a business setting. Business users, while they demand considerable rigor in other aspects of an evaluation, may be satisfied with a demonstration or association standard, rather than a causation standard of proof required by science.

Demonstration may be satisfied with an analysis that shows association between the intervention program and a favorable outcome, together with adequate demonstration that the results are not biased or confounded by factors that could affect the result. Satisfying the scientific audience about causality, on the other hand, requires that the mechanism through which the outcome is achieved be unambiguously demonstrated. For example, if the result to be proved is savings, then a study that proves causation would have to establish missing components in the target population (for example, compliance with best-practice medical care), then show how the intervention improved compliance with care in the population, and finally, the resulting financial outcomes. Wilson and McDowell refer to different levels of causality, as follows:



From Wilson and McDowell; reproduced by permission.

It is again worth emphasizing that one can never state that a relationship is causal with 100% confidence. However, one can become sufficiently confident that a relationship is causal based on the consumer of the study, the study’s purpose, and the degree to which the study met the preestablished criteria for imputing causality. Statistical methods alone cannot establish proof of a causal relationship within an association. Generally, an experimental design is necessary to establish causality, but observational designs may be used as strong evidence, bolstering confidence that an observed relationship is causal.

Hill’s study is perhaps the most often cited description of discerning causality from noncausality. A judgment that causality exists is never straightforward but can be bolstered by the following:

  1. Consistency of the association: If different studies show similar results even with different study designs, time frames and populations, causal plausibility is bolstered. For example, case-control and cohort studies on cigarette smoking and heart disease, using different populations, cultures and time frames, all indicate increased risk of developing the illness. The plethora of similar outcomes, multiplicity of study designs, and consistency over time support a judgment of causality between cigarettes and heart disease.
  2. Strength/magnitude of the association: The degree of association between the cause and effect (as expressed by the risk ratio, odds ratio, or hazard ratio depending on the type of study) should be strong and statistically significant. The existence of a dose-response relationship provides additional evidence of causality. For example, chronic care management outcomes studies may find that the easier it is for participants to reach a medical professional 24 hours a day and 7 days a week (i.e., a nurse call line), the fewer emergency department admissions occur, particularly during hours when offices are typically closed. The reduction is mitigated, however, when nurse call lines are available only during limited hours, pointing to a dose-response relationship.
  3. Specificity of the association: This is the case when one cause leads to one, rather than multiple effects. However, because experience shows that single events have myriad outcomes, this criterion makes a supporting, rather than strong, case for causality. Moreover, epidemiologists believe that causes are usually multiple. For example, it is observed that not everyone exposed to a cold virus becomes ill. Some of those individuals who avoid the cold have previously developed immunity, but others have not. Why do some develop the cold and not others? Any factor that can be expected to alter the frequency or quality of another factor when it itself is altered may be considered a causal factor. Specificity is a necessary condition for asserting a causal link because a causal association needs to be as specific as possible.
  4. Temporal relationship of the association: The cause must precede the effect. For example, the outcomes improvement must come after the advent of the chronic care management program.
  5. Coherence of the association: The association must be plausible or credible. Some studies, for example, found an association between sunspots, hemline height, and stock market levels. If a pre-post study of a chronic care management program emphasizing use of beta-blockers after myocardial infarction (heart attack) found a 30% relative reduction in risk of a second attack within a year and 75% of the intervention group was already using beta-blockers in the preperiod, the coherence of the association would have to be questioned, because randomized controlled clinical trials show that the theoretical maximum reduction in readmissions for myocardial infarction that can be obtained in one year in a naïve population (one in which 0% are taking beta-blockers after myocardial infarction) is less than 30%.


Evaluations must carefully and accurately evaluate the plausibility and credibility of the results and impact of the population health management program’s contribution to outcomes. In doing so, look for clear specification and testing of the causal linkage between the population health management program and the desired outcomes—through the intermediate to the ultimate outcomes.

  • The intervention must be clearly specified, and it must be clear what steps were taken to ensure that the reference group did not receive the intervention (note that the intervention might be intention to treat)—that is, the intervention group could be suitable for, and invited to, the actual intervention, even though they might not have received it.
  • The population health management program must explicitly state how measurement of the hypothesized impact of the intervention (effect) is performed. For example, if the intervention is to advise a participant to discuss taking a statin with the participant’s doctor, the measurement should state how taking a statin will be assessed (claims, participant-supplied data), and in what time frame (e.g., an appropriate National Drug Code [NDC] for a statin appearing in the claims data in the six months before a particular look-back date).
  • All the considerations about good study design described elsewhere in this document apply—for example, having a comparable reference group (or one that can be made comparable by valid adjustment methodology), applying the same criteria for the outcome to the intervention and the reference groups, and using statistical methodology that is valid for the type of data (e.g., not using independent samples t-testing in a pre-post study cohort study).
  • Evaluators should acknowledge other factors that may have affected the outcome (so-called confounding factors) and, when possible, control for them. It is important to consider both observable and unobservable confounding factors. An example of the latter is socioeconomic status. There are methods for determining how great an unobserved confounding factor’s impact would have to be to negate the difference found in the study.
  • Causal linkages should be seriously considered only when the evaluator is confident that the findings are not due to artifact, chance or confounding and the results are significantly different by valid statistical tests. The best way to test causality is to see if the strength of the causal association changes when new facts are added or different groups are studied. While this does not categorically prove causality, it can support the hypothesis.


A quasi-experimental or true experimental study design that allows program participants to be compared with an equivalent group of nonparticipants over time is the preferred approach to prove causality. However, properly adjusted pre-post studies or those with propensity score matching may be good practical choices in business settings.

In some evaluations, particularly those conducted for internal business purposes, a causal link is assumed implicitly. Hence, the evaluation does not make an additional effort to test for causality. The answer is that it depends on the intended use of the evaluation.


Study DesignEffect on Causality
Follow-up Experimental follow-up designs reduce the likelihood that the intervention and reference populations are not equivalent.
Causality determination is the highest in experimental follow-up designs.
Pre-post historical
No control group
Lacks a control/comparison group against which to determine causality.
Benchmark studies Benchmarking is commonly used for business assessments.

The credibility of these designs in assessing causality is particularly dependent on equivalence between the intervention population and the control population and comparability of the metrics and methods used to assess the two groups.

The principle of comparison with a reference or control group is important for program evaluation but might be misleading for comparisons across health plans. This is because the usual care of the comparison group might differ dramatically from one plan to another. A population health management program might be successful in a plan with unmanaged usual care but not at all in another plan where usual care already incorporates many of the elements considered by others to be a component of a population health management program.


Abramson J. (1998). Making Sense of Data. Oxford: Oxford University Press.

Hennekens, C., and Buring, J. (1987). Epidemiology in Medicine. Boston: Little, Brown & Company.

Hill, A. (1965). The environment and disease: Association or causation? Proceedings of the Royal Society of Medicine. 58:295-300.

Wilson, T., and MacDowell, M. (Fall 2003). Framework for assessing causality in disease management. Disease Management, 6(3), 143-158.

Care Continuum Alliance. (2004). Disease Management Program Evaluation Guide. Washington, DC: Care Continuum Alliance.

Chronic Care Model (CCM)

The Chronic Care Model is a care delivery prototype that summarizes the basic elements for improving care for patients with chronic conditions in health systems at the community, organization, practice, and patient levels; promote one or more of the six core elements of the Chronic Care Model, namely:1) Health care organization and leadership; 2)Linkage to community resources; 3) Support of patient self-management; 4) Coordinated delivery system design; 5) Clinical decision support; 6) Clinical information systems.

Over the past few decades, chronic conditions (such as heart disease, hypertension, diabetes, asthma, and depression) have been rapidly replacing acute and infectious diseases as the major cause of death, disease, and disability in the United States. However, because the prevailing health care system is based on the diagnosis and treatment of acute illness, it is not well suited for the effective care of chronic illness.

There are many definitions of “chronic condition,” some more expansive than others. It is characterized as any condition that requires ongoing adjustments by the affected person and interactions with the health care system. The most recent data show that more than 145 million people, or almost half of all Americans, live with a chronic condition. That number is projected to increase by more than one percent per year by 2030, resulting in an estimated chronically ill population of 171 million. Almost half of all people with chronic illness have multiple conditions. As a result, many managed care and integrated delivery systems have taken a great interest in correcting the many deficiencies in current management of diseases such as diabetes, heart disease, depression, asthma and others.

Those deficiencies include:

  • Rushed practitioners not following established practice guidelines
  • Lack of care coordination
  • Lack of active follow-up to ensure the best outcomes
  • Patients inadequately trained to manage their illnesses


Overcoming these deficiencies will require nothing less than a transformation of health care, from a system that is essentially reactive--responding mainly when a person is sick--to one that is proactive and focused on keeping a person as healthy as possible. To speed the transition, Improving Chronic Illness Care (ICIC) created the Chronic Care Model (CCM), which summarizes the basic elements for improving care in health systems at the community, organization, practice and patient levels.

Short History

In the latter part of the 20th century, researchers began to develop care models for the assessment and treatment of the chronically ill. Nurse researchers, such as S. Wellard, C. S. Burckhardt, C. Baker and P. N. Stern, and I. M. Lubkin and P. D. Larson, were often on the front lines of actual care for patients with ongoing treatments for conditions such as diabetes or renal failure.

The Chronic Care Model (CCM) originated from a synthesis of scientific literature undertaken by The MacColl Institute for Healthcare Innovation in the early 1990’s.  Edward H. Wagner, M.D., Director of The MacColl Institute for Healthcare Innovation, Director of The Robert Wood Johnson Foundation national program "Improving Chronic Illness Care"(ICIC), and Senior Investigator at Group Health Research Institute in Seattle, WA developed the Chronic Care Model, or CCM.

During a 9-month project funded by the Robert Wood Johnson Foundation (RWJF), an early version of the Model underwent extensive review by an advisory panel of experts and was then compared with the features of leading chronic illness management programs across the U.S.  Subsequently, the Model was further refined and published in its current form in 1998.  Improving Chronic Illness Care (ICIC), a national program of RWJF, was launched in 1998 with the Chronic Care Model at its conceptual core.

The CCM summarizes the basic elements for improving care in health systems on different levels. These elements are the community, the health system, self-management support, delivery system design, decision support and clinical information systems. Evidence-based change concepts under each element, in combination, foster productive interactions between informed patients who take an active part in their care and providers with resources and expertise. The Chronic Care Model can be applied to a variety of chronic illnesses, health care settings and target populations. The bottom line is healthier patients, more satisfied providers, and cost savings. . 

The Chronic Care Breakthrough Series Collaboratives (BTS) began in 1999 in partnership with the Institute for Healthcare Improvement.  Using a clearly defined change package based on the CCM, the BTS Collaboratives provided participants with proven tools and information to assist them in making those changes within their system. The National Committee on Quality Assurance and The Joint Commission developed accreditation and certification programs for chronic disease management based on the CCM. More recently, the CCM serves as a foundation for new models of primary care promulgated by the American Academy of Family Practice and the American College of Physicians.

The World Health Organization approached ICIC to work on an initiative to adapt the model for developing countries.  The major product of this partnership was the WHO global report, “Innovative Care for Chronic Conditions: Building Blocks for Action”, published in 2002.  International versions of the Chronic Care Model have been developed and used in countries as diverse as UK, Canada, Denmark, Russia, China, Australia, and New Zealand.  ICIC’s work was featured in the four-part PBS series “Remaking American Medicine…Health Care for the 21st Century”. 

This television program premiered in 2006 and featured an array of pioneering individuals and institutions struggling to better health care across the nation. Providers who care for chronically ill patients can be better supported with evidence-based guidelines, specialty expertise, and information systems. Overall health care costs can be lowered through better care delivery. 


While chronic care management programs vary in design and implementation, almost all promote one or more of the six core elements of the Chronic Care Model (CCM), as a framework for guiding specific quality improvement strategies. (See Figure.)

The Chronic Care Model
  1. Health care organization and leadership: An organizational environment that systematically supports and encourages chronic illness care through leadership and incentives results in more successful quality improvement activities.
  2. Linkage to community resources: Community linkages can provide cost-effective access to services not available inside the organization, such as nutrition counseling, peer-support groups, and data for patient registries.
  3. Support of patient self-management: Individual and group interventions that emphasize patient empowerment and self-management skills have been shown to be effective in the management of diabetes as well as asthma and other chronic conditions. 
  4. Coordinated delivery system design: Innovations in delivery system design to coordinate actions of multiple caregivers of diabetics, for example, have led to significant improvements in glycemic control, patient satisfaction, and health care utilization.
  5. Clinical decision support: Incorporating evidence-based practice guidelines into registries, flow sheets, and patient assessment tools can be an effective method for changing provider behavior.
  6. Clinical information systems:  For example, with access to adequate database software, health care teams can use disease registries to contact patients to deliver proactive care, implement reminder systems, and generate treatment plans and messages to facilitate patient self-care.

The model is built on the premise that these six elements work together to create productive interactions between an informed, activated patient and a prepared, proactive practice team – which is what leads to improvements in outcomes.

According to a recent literature review and survey of reputable programs, there is substantial evidence that chronic disease management strategies “achieve better disease control, higher patient satisfaction, and better adherence to guidelines by redesigning delivery systems to meet the needs of chronically ill patients.”  For example:

  • Acute Depression:  A simple but systematic program of feedback to doctors on treatment a recommendation, supplemented with follow up and care management by telephone, was shown to significantly improve primary care treatment of patients with acute depression.
  • Diabetes:  In a randomized trial to assess the impact of primary care group visits on the process and outcome of care for diabetic patients, the intervention group receiving self-management support through “mini-clinics” involving teams of providers exhibited better outcomes (including higher patient satisfaction and HbA1c levels) than the control group.

The World Health Organization has developed an expanded version of the Chronic Care Model: the Innovative Care for Chronic Conditions Framework , designed in particular to be relevant to low and middle income countries. It broadens and reframes the Chronic Care Model by organizing the evaluation along macro (policy and financing), meso (health care organization and community) and micro (patient and family) levels of the health care system. This framework is centered in a triad of partnership between the patient, the health care team and the community. This triad is placed in the background of organized and well equipped health care teams and a positive policy environment.


“Improving Chronic Illness Care”. 22 Feb.2011

“Diabetes Action Online”. The Innovative Care for Chronic Conditions framework (ICCC) .World Health Organization. 22 Feb.2011

“AHRQ - Agency for Healthcare Research and Quality .The CAHPS Improvement Guide”. US Department of Health and Human Services . 22 Feb.2011

The McColl Institute – add cite

Chronic Condition/Disease Definitions

A chronic condition is a disease that has one or more of the following characteristics:

  • Is permanent;
  • Is progressive if unmanaged;
  • Is caused by nonreversible pathological alteration;
  • Requires special training of the patient for rehabilitation, self-monitoring, and self-management; or
  • May require a long period of supervision, observation, or care.

Diseases or chronic conditions are typically identified by scanning patient records – either medical records or administrative data. Specific International Classification of Diseases, 9th/10th edition, Clinical Modification (ICD-9-CM), Current Procedural Terminology, 4th edition (CPT-4), and Health Care Common Procedure Coding System (HCPCS) and National Uniform Billing Codes (NUBC) data codes are used to identify seven chronic conditions common to population health management programs, as well as depression and maternity. Data codes may also be used to identify common exclusionary conditions.

See rare diseases/conditions for codes for other, less-frequently encountered conditions.


Purpose of the identification: The specific definition to be used in a population health management program should be determined by the purpose. For example, identification of chronic patients as part of an outbound call program (where sensitivity is important) could use a different definition from that used to identify patients for a financial reconciliation (where specificity is important).

Population health management programs often develop exclusion as well as inclusion criteria. Exclusion criteria attempt to identify those patients with the disease in question who would not benefit from the proposed health management program (e.g., institutionalized patients). These criteria are generally specific to the program and not generalizable across all situations. However, an example of a list of common exclusionary categories is provided under the definition of exclusionary conditions.

Because most population health management organizations use health care claims for case finding, general recommendations for avoiding false positives (i.e., people with claims for the disease but who do not actually have the disease) should be made explicitly. Population health management organizations will need to determine their tolerance for false positives based on the specific nature of their programs, implementations, and measurement objectives, which will be reflected in the sensitivity/specificity trade-off in the selected criteria.

Chronic Conditions (Exclusionary Definitions)

Certain members of a population are defined as not suitable for intervention. These members may or may not be excluded from the program but irrespective of their program status, are often excluded from the measurement of program results.


Members who meet the following claims-based definitions may be excluded from a chronic care management program (and its associated measurement).

  • End-Stage Renal Disease (ESRD) (see chronic condition/disease definitions)
  • Malignant or undefined non-skin-related neoplasms: defined as member with two separate medical or pharmacy claims, with at least one medical claim, with a primary ICD-9-CM code of 140 through 171, 174 through 208 or 230 through 239; or medical claim CPT/HCPCS codes of J8510-J8999, J9000-J9999, Q0083-Q0085 or Q0163-Q0181; drug therapeutic classes of N1B (hematinics), N1C (leukocyte [WBC] stimulants), N1E (platelet proliferation stimulants), V1A (alkylating agents), V1B (antimetabolites), V1C (vinca alkaloids), V1D (antibiotic antineoplastics), V1F (antineoplastics, miscellaneous), V1K (antineoplastics antibody/antibody-drug complexes), V1O (antineoplastic LHRH [GNRH] agonist, pituitary suppr.), V1Q (antineoplastic systemic enzyme inhibitors), V1T (selective estrogen receptor modulators [SERM]) or V1U (antineoplastic antibody/radioactive-drug complexes)
  • Hemophilia: defined as member with two separate medical or pharmacy claims, with at least one medical claim, with a primary ICD-9-CM code of 286 or CPT/HCPCS codes of J7190, J7191, J7192, J7193, J7194, J7195, J7197, J7198, J7199, Q0160, Q0161, Q0187 or Q2022; drug therapeutic class M0E (antihemophilic factors)
  • HIV: defined as member with two separate medical or pharmacy claims with a primary ICD-9-CM code of 042; drug therapeutic classes of W5C (antivirals, HIV-specific, protease inhibitors), W5I (antivirals, HIV-specific, nucleotide analog, RTI), W5J (antivirals, HIV-specific, nucleoside analog, RTI), W5K (antivirals, HIV-specific, nonnucleoside, RTI), W5L (antivirals, HIV-specific, nucleoside alg, RTI comb), W5M (antivirals, HIV-specific, protease inhibitor comb), W5N (antivirals, HIV-specific, fusion inhibitors)
  • Transplants: defined as member with two separate medical claims with these codes: for ABMT, CPT codes 38230­38241; heart-lung, CPT codes 33930-33945; liver, CPT codes 47133-47136; lung, CPT codes 32850-32854; kidney, CPT codes 50300-50380; pancreas, CPT codes 48550-48556; intestinal, CPT Codes 44132-44136; general, claims with a primary ICD-9-CM code of V42. (Not all V42 codes are always excluded; for example, those codes for retinal transplant.) It should be noted that some believe that a chronic care management program for heart failure may not wish to exclude heart transplants because they can measurably affect the rate of this type of procedure.
  • Institutionalized members (long-term care facilities; hospice; psychiatric and substance abuse facilities)

Clinical Care Metrics

Clinical care metrics are the care metrics that assess the effectiveness of a clinical or chronic care management intervention program. They are proximate measures for the clinical economic outcomes of a chronic care management program.

See causality.


Johns Hopkins/American Healthways. (Fall 2003). Standard outcome metrics and evaluation methodology for disease management programs. Disease Management, 6(3), 121-138.

Clinical Data Repository (CDR)

A real time database that consolidates data from a variety of clinical sources to present a unified view of a single patient. It allows clinicians to retrieve data for a single patient rather than to identify a population of patients with common characteristics or to facilitate the management of a specific clinical department. Typical data types which are often found within a CDR include: clinical lab test results, patient demographics, pharmacy information, radiology reports and images, pathology reports, hospital admission/discharge/transfer dates, ICD-9 codes, discharge summaries and progress notes.


The Informatics Review- Sittig, Pappas, and Rubalcaba)

Clinical Practice Guidelines

Clinical practice guidelines are statements written for health care providers and patients to guide decisions about appropriate actions in specific circumstances. Guidelines accomplish this by:

  • Describing a range of approaches for the diagnosis, management, or prevention of specific diseases or conditions.
  • Defining practices that address the needs of most patients in most situations.

The statements contain recommendations that are based on evidence from a rigorous systematic review and synthesis of the published medical literature. As opposed to fixed protocols, guidelines assist clinicians and patients to develop individual treatment plans tailored to the patient. In addition, guidelines are also useful to managed care organizations and other organizations that define benefit plans or manage health care resources.


About clinical practice guidelines. National Heart Lung and Blood Institute. National Institutes of Health. Department of Health & Human Services. Retrieved from

Cohort and Cohort Study

A cohort is a group of people who share a common set of definition criteria and who are observed over time to determine their reaction to a particular intervention.

A cohort study is also called a follow-up study and has an observational analytic design. These studies include a period when the cohort is exposed to something and a period when the cohort is not exposed. The cohorts may be observed prospectively or restrospectively.


A cohort comprises a group of individuals who share a common set of definition criteria or experience, such as participation in a population health management program. For research purposes, a group (or groups of individuals) is defined according to the presence or absence of exposure to an intervention or risk factor for disease. Because time can be a defining factor for cohort identification, a set of individuals deployed to a program for enrollment beginning January 1 may constitute a different cohort than a set of individuals deployed April 1. (This may be particularly true if the claims data by which members are identified have been updated between January 1 and April 1.)

A cohort study is also called a follow-up study with an observational analytic design. These studies include a cohort that is “exposed” to an intervention (e.g., population health management) and a comparison group that is not. The comparison group can be created from data collected on the same group of members but at a different time period (e.g., pre- and post-intervention), or the comparison group can be most anything else, including another cohort. Cohort studies may be further classified as prospective or retrospective, depending on the timing of the intervention and initiation of the study. Prospective studies are implemented before the intervention (exposure) and conducted over time. Conversely, in retrospective studies, all the relevant events (outcomes and intervention) have already occurred. For additional information on comparison groups, see study design.

While the group of individuals chosen for the exposed population in a cohort study should be representative of the entire population, the choice of the cohort members is related to the hypotheses under investigation and the design of the study. For example, a cohort that is going to be enrolled in a diabetes management program may be all individuals in the health plan with that disease, or at least a representative sample of the individuals in the health plan with that disease. However, if the incidence of a rare disease associated with occupational exposure to a contaminant is being studied, a cohort of people in certain occupations should be selected.

In research studies, a group of people may be observed over time to determine their reaction to a particular intervention. Although cohort studies may be challenging due to the need to track large numbers of people over several years, this design offers a number of advantages. For example, cohort studies allow for assessment of multiple effects of a single exposure (see example below).

Presumably, the smaller the cohort, the easier it is to obtain consistency and the fewer confounding factors, but the more difficult it is to find individuals with common factors. The larger the cohort, the more confounding factors are likely to be present. The Framingham Heart Study is a cohort study that examined coronary heart disease and risk factors for that disease. The participants were divided into quintiles of exposure based on baseline levels of blood cholesterol and blood pressure (both systolic and diastolic). The results of each cohort were compared to determine the differences in risk factors.

The cohort design can also have a substantial disadvantage. The most significant drawback is that the outcomes for the cohort are susceptible to regression to the mean. Briefly, regression to the mean refers to the fact that there is a high probability that the utilization experience of members with extreme outcomes will move toward the mean even without a particular intervention. If one compares medical claims costs for the cohort before and after the start of the program, it is likely that the costs after the program start will be less than before the program.


A health plan wants to complete a pilot chronic care management program targeting chronic obstructive pulmonary disease (COPD) over a 12-month period. The pilot was designed to include two cohorts, an intervention or experimental cohort that was enrolled in the pilot program and a control group cohort that was not.

The targeted population inclusion criteria included a confirmed diagnosis of COPD and continuous eligibility from a 12-month period before the pilot and through the pilot program period, which ran from January 2002 to January 2003. The control group included a comparable subgroup of the targeted population who met the inclusion criteria but who did not participate in the pilot program, did not have telephone access, or refused to participate in the program. Administrative data were used for program analysis and were based on a set claims period from January 31 to January 31 for both the baseline and intervention periods. In this example, both cohorts are susceptible to regression to the mean.


The same health plan wants to complete an asthma pilot program but this time only one cohort is used. The single cohort is the entire targeted population. The nonparticipants, which formed the control group in the COPD pilot, were not separated. Instead, the claims experience for the entire cohort in the pilot was compared before and after the start of the pilot. The effect of the intervention can be measured by comparing the claims experience of the cohort before and after the intervention. Caution must be used when interpreting the results, because the entire cohort is susceptible to regression to the mean.

Hence, comparison groups should be another cohort as given in Example 1 for COPD. Example 2 for asthma type studies should not be taken for scientific analysis, as there is only one cohort. This will reduce bias and make the study scientifically authentic.


Barker, D., and Hall, A. (1991). Practical epidemiology (4th ed.). New York: Churchill Livingston.

Bland, M. (1987). An introduction to medical statistics. Oxford: Oxford University Press.

Care Continuum Alliance. (2004). Disease management program evaluation guide. Washington, DC: Care Continuum Alliance.

Mausner, J., and Kramer, S. (1985). Epidemiology, an introductory text (2nd ed.). Philadelphia: Elsevier.

Personal communication with David Tinkelman, M.D., National Jewish Medical and Research Center, (303) 398-1519. March 2004.

Comparison of Tax-Advantaged Health Care Spending Accounts


Comparison Group

The comparison group is the group of individuals with whom individuals in an evaluation are compared. A comparison group is also called a control group or reference group. In population health management evaluations, the comparison group often is composed of individuals who are eligible for but not enrolled in the program. To be comparable, cohorts of members (the intervention and control or comparison groups) must be similar in all ways.

See study design.

Completion Factor

A completion factor is a factor applied to immature claims data from a period to estimate the eventual total incurred or paid claims from that period. The correction implied by applying a completion factor is necessitated by the lack of sufficient claims runout in the claims database for the claims to be considered mature.


In any work based on administrative claims, an analyst has to recognize that claims data are seldom mature. For example, if the period of analysis is the 12 months from January 1, 2003, to December 31, 2003, claims are “incurred” (i.e., the services are rendered) during these 12 months. However, because of reporting lags (from the provider to the claims processor) and processing lags, the final amount of the incurred claims from that claims year may not be known for months, if not years, after the end of the claims year. At any point in time, the number of months of claims related to a particular period that have been reported and paid (for that period) is referred to as the claims runout.

While many claims are settled quickly, there will always be delays in getting a percentage of all incurred claims. Reasons that claims take a long time to be settled include the issues of primary and secondary payment, subrogation, and contested claims, all of which add considerable time to processing and can result in a previously settled claim being reopened. Runout from a particular period can last for many months, if not years.

One frequently used measure of claims completeness is the measure “Incurred 12/Paid 18,”which refers to the percentage of the ultimate claims liability for the 12-month incurred claims period that was paid and reported by the end of the 18th month (6 months after the close of the 12-month incurral period). With most managed care organizations, this percentage should be relatively high: 95% or higher is not uncommon. A completion factor under 90% should be a cause for further analysis. “Incurred 12/Paid 15” is sometimes used as a measure in organizations where the customers are requesting analysis sooner than waiting the additional 3 months (full 18 months). An “Incurred 12/Paid 15” does provide a fairly robust completion.

For financial purposes, however, reporting periods cannot be left open until the last claim has been resolved and all claims are settled. For this reason, estimates are made of the eventual amount of the claims liability for a period, often using completion factors. This can be important in population health management applications where purchasers of services often want estimates of the ultimate savings from a program during or immediately after the close of an intervention period. The only way this can be achieved is by estimating the ultimate intervention period claims, using completion factors to estimate the unknown incurred claims.

At their simplest, completion factors are the inverse of the historic percentage of ultimate claims payment that has been made after a period of time.


In claims-based work, it is often useful to display claims in the form of an incurred/paid triangle. This triangle shows the month of incurral (the month the service was performed) as well as the month of payment. It is referred to as a triangle because in this format (paid and incurred months) some month combinations will never be populated.

In this example, which has been abbreviated for simplicity, only 3 months of incurred claims have been shown. There is no runout in this example because the reporting period and the incurral period both end in March.


PAID   January February March
January 50
February 25 60  
March 10 30 75


In the example that follows, it is critical that we assume that there is no further claims incurral after 3 months. Additional payments are expected, but we do not know them at the end of March and we will use completion factors to estimate future payments. We employ a technique (in this case, a simplified version of a technique called the chain-ladder method) to estimate the amount of completion or runout that will be expected for each month of incurred claims. In the case of March’s incurred claims, because there are two missing months of runout to come, 2 months’ completion has been estimated. In the chart below, estimated months are italicized.


PAID   January February March
January 50
February 25 60  
March 10 30 75
April 0 12 38
May 0 0 0
85 102 128

The unknown (estimated) months’ payments are estimated from the pattern of known claims development. The chart below illustrates a method (the chain-ladder method) of creating the development factors that are used to estimate the unknown months’ payments.


PAID   January February March
January 50
February 25 60  
March 10 30 75
SUM 85 90 75
SUM-1 75 60
RATIO 1.13 1.50 128
COMPLETION 1.00 1.13 1.70

Estimates are frequently used by actuaries and underwriters to estimate future payments from current, known amounts for financial reporting or pricing purposes. Techniques vary from the simple one illustrated above to much more sophisticated methods.


Bluhm, W. (ed.) (2003). Group Insurance. 4th ed., Winsted, CT: Actex Publishers.

Computerized Physician Order Entry

Clinical application that allows clinicians to order and process lab tests, medications, clinical procedures, and other services electronically.


AHRQ national resource center –

Confidence Interval

A confidence interval around a point estimate of a variable is a numeric interval that is likely to contain the true value of the variable. The probability that the interval contains the true value must always be given; traditionally, this is 95% or 99%, although in some commercial applications, the value may be lower.


A common statistical task is the approximation or estimation, Ŷ, of an unknown quantity, Y. Frequently, Y is an attribute of a population that is difficult or impossible to measure directly, such as the percent of American adults who would vote for candidate A or who have disease B. The quantity Y might also be a statistical parameter, such as a regression coefficient. In any example, it is important to realize that Y and Ŷ can differ and in many contexts, will nearly always differ. Therefore, the estimate Ŷ is not meaningful without an accompanying statement of its accuracy (i.e., how close it is to the unknown value, Y).

The confidence interval is a formal statement of the accuracy of the estimate Ŷ. We say that [Ŷ -  a, Ŷ + b] is a confidence interval around Ŷ if we believe it is likely that Y is contained in [Ŷ - a, Ŷ + b]. Because this is not certain in most contexts, a confidence interval must be accompanied by (a lower bound of) the probability that Y is contained in [Ŷ - a, Ŷ + b]. By convention, this bound is often 95% or 99%, although in business applications this probability may be relaxed. Sometimes, the confidence interval is symmetric or nearly symmetric around Ŷ. In this case, we might say that the confidence interval is ±c, as shorthand for [Ŷ - c, Ŷ + c].

As an estimate, is Ŷ too high or too low? If we are interested in both possibilities, then both ends of our confidence interval are relevant. In this case, both a and b are finite numbers and we have a two-sided confidence interval. For a two-sided confidence interval, the probabilities that Ŷ is outside of the interval are usually distributed symmetrically, – i.e., for a 95% estimate, P(Y <Ŷ - a)P(Y >Ŷ+ b) ≈ 0.025 and for a 99% estimate, P(Y <Ŷ - a)P(Y >Ŷ+ b) ≈ 0.005.

Sometimes, we do not care if Ŷ is too high, we care only if it is too low. In this case (or the symmetric case in which we care only if Ŷ is too high), we use a one-sided interval, in which a = ¥or b = ¥ (but not both). If b = ¥, then our interval is [Ŷ – a, ¥]. For a 95% estimate, P(Y < Ŷ – a) < 0.05 and for a 99% estimate, P(Y < Ŷ – a)< 0.01. Similar statements hold if a = ¥.
We observe that, under the above conventions on distributing the probability that Y is outside of the confidence interval, the 99% interval is always wider than the 95% interval.

How do we compute confidence intervals? There is no single method. However, Y is frequently the expected value of a population attribute, y, and Ŷ is the mean value of y for a representative sample of the greater population. In this special case, if our greater population is infinite or very large, if the representative sample is sufficiently large, and if the distribution of y is reasonably bell-shaped (symmetric, unimodal, and continuous or nearly continuous), then we can use the simple method given below.


Suppose we have a representative sample of the greater population. Let Ŷ denote the sample mean, – i.e., the mean value of y for our sample. Let Ŝ denote the standard deviation of y over our sample. Let N denote the size of our sample. Then a 95% confidence interval is [Ŷ – 1.96 * Ŷ / N1/2, Ŷ + 1.96* Ŝ / N1/2] and a 99% confidence interval is [Ŷ – 2.58 * Ŷ / N1/2, Ŷ + 2.58* Ŝ / N1/2].

The simple method of estimating confidence intervals is powered by the central limit theorem, which states that as N grows large, the distribution of Ŷ becomes close to a normal distribution having a mean of Y and standard deviation σ/N1/2, where σ denotes the true standard deviation of y. If we substitute Ŝ for σ; this simplistic approximation is frequently close enough to be useful. We note that approximately 95% (respectively, 99%) of the outcomes of a normal distribution are within 1.96 (respectively, 2.58) standard deviations from the mean from either side; this is the origin of the constants in the formula.

A remarkable mathematical fact is that the central limit theorem holds true for any distribution of y, so long as it has a finite mean and variance. Unfortunately, the fine print to the central limit theorem is that for an arbitrary distribution of y, we do not know a priori what constitutes a sufficiently large value of N. However, as suggested above, we do have some information on when the simple method is and is not appropriate.


Physical attributes such as height, weight, and blood pressure are frequently modeled with normal distributions. As a rule of thumb, statisticians have traditionally considered 30 to be a large sample size when considering normal distributions. The simple method above can be used for normal distributions.


An experimenter measures the heights of 50 males and computes an average of 176.1 cm and a standard deviation of 5.8 cm. In this example, y is height, which we approximate with a normal distribution. Because our sample size is greater than 30, we can use the simple method. Our 95% confidence interval is [176.1 – 1.96 * 5.8/501/2, 176.1 + 1.96 * 5.8/501/2], or [174.5, 177.7]. Thus, we are 95% confident that the true average height of the greater population falls into this interval, which is approximately 3.2 inches wide.


For readers who want to skip the theory below, the following conclusions should be noted for population health management applications of confidence intervals. We cannot use the simple confidence interval calculation when:

  • The population statistic has a discontinuous or bimodal distribution (for example, if the random variable has a yes/no value, such as diabetic/nondiabetic); and
  • The population statistic is highly skewed, such as, for example, the distribution of member costs.

The reader who is not interested in the technical aspects of this topic may wish to skip the next section and proceed to the examples.


The Bernoulli distribution is the simplest statistical distribution. The random variable can take on a value of 1 or 0 according to probabilities p and 1 – p, respectively, for 0 p 1. A single fair coin flip has a Bernoulli distribution with p = 0.5. Suppose we select a person at random from a population and assign them a value of 1 if they have asthma and 0 otherwise. This is also a Bernoulli distribution. The probability p is the population’s asthma prevalence rate.

Based on the above-mentioned criteria, we cannot use the simple test if y has a Bernoulli distribution—Bernoulli distributions are bimodal and discontinuous. However, in this case, the variable Ŷ * N has a binomial distribution, the properties of which (including its relationship to the normal distribution) are sufficiently well-known to allow us to experiment with the simple method. In fact, if Ŷ has a nearly symmetric distribution and N is large, the simple method works fairly well. In this case, Ŝ = (Ŷ *(1 – Ŷ))1/2. Note that because y can only have values of 0 and 1, we have 0  Y 1. A nearly symmetric distribution of y is the same as a value of Ŝ close to 0.5.

If, on the other hand, Ŷ is close to 0 or 1, then the simple method should not be used to make precise statements because the distribution of Ŷ will be skewed. In a population health management setting, this applies to most disease prevalence rates. If precision is desired, the statistician can use more accurate methods. See confidence intervals for proportions. Alternatively, many statistical computer programs compute Bernoulli confidence intervals exactly. If the confidence interval is an end deliverable—i.e., we do not need a closed formula for substitution into other calculations, this is the best and safest method.


A population health management telephony program made calls to a population of candidates that was 50% male and 50% female. Of these candidates, 500 answered the phone and agreed to participate. Of the 500 participants, 53.2% were female (266 females). The experimenter wanted to verify that the percent of female participants (53.2%) did not differ significantly from the percent of females in the greater population of candidates. Using the simple formula, our 99% confidence interval is [47.5%, 58.9%]. Because 50% is well within this range, we can assume no sample bias. A statistical computer program calculating the Bernoulli confidence interval confirms this with an exact interval of [47.3%, 59.0%]. Note that the exact interval is not symmetric around the sample mean because of the skewness of the binomial distribution. The simple method provides a lower left bound than the exact method, which is harmless from a probabilistic point of view (though it provides a poorer estimate of Y). The simple method also provides a lower right bound than the exact method, which is incorrect, though not too dangerous in this instance.


Out of the 266 female participants above, 9.4% reported that they had asthma (25 asthmatics). The researcher believes the correct value should be 5.2%. Can this difference be attributed to statistical error? The simple formula gives a 99% confidence interval of [4.8%, 14.0%], which contains 5.2%. It would therefore seem that the difference is not statistically significant at the 99% level. However, the exact solution is [5.4%, 14.9%], which does not contain 5.2% and therefore indicates statistical significance. In this case, that hypothesis that the sample is representative of the underlying population should be rejected.

This example illustrates the weakness and strength of the simple method. The method is too imprecise to allow conclusions to be drawn regarding values near the endpoints of the confidence interval, and it is too imprecise to be used to calculate published values. Indeed, the error in the lower half of the confidence interval (5.4 – 4.8) may seem small but it is non­trivial as a percent of the width of the lower half of the confidence interval (9.4 – 5.4). However, as a tool for qualitative evaluation, the simple method provides a good and useful ballpark estimate that can be of great benefit to the researcher.

We note also that both methods agree on significance at the 95% level.

In general, values near the endpoints of any confidence interval should be scrutinized and the wise researcher should be wary of using them as the sole basis for a conclusion. Instead it is preferable, if possible, to increase the sample size, repeat the study, or find additional information that can shed light. Under this philosophy, in the above example, both methods point toward gathering more information to determine 99% significance.


Unless it is explicitly known that y has a suitable distribution, the simple method cannot be used.


A managed care organization reported a cost of $1,500 per member per year. What is a confidence interval? Here, the random variable y represents the annual cost of a member. Unfortunately, even if the sample standard deviation is known, the distribution of y is known to be highly skewed. We need more information before we can calculate a confidence interval.


Hogg, R. and Tanis, E. (2005). Probability and statistical inference. (7th ed.). Indianapolis, IN: Prentice Hall.

Spiegel, M. et al. (2000). Schaum’s outline of probability and statistic. (2nd ed.). New York: McGraw-Hill.

Conformance to Treatment and Care Plans: Compliance, Adherence, and Persistency

Compliance, adherence, and persistency are measures to describe the extent to which patients follow their care plan (or regimen) as prescribed by their health care provider. While the terms are usually used in the context of medication taking, they can be used to describe or measure conformance (the generic term) to the requirements of any regimen that can be characterized in terms of frequency, intensity, duration, and timing. We discuss two aspects of treatment conformance: measurement and improvement.


Management of a chronic disease is a long-term proposition requiring adherence to evidence-based care by health care providers as well as conformance to a variety of therapeutic, monitoring, and behavioral components by the patient. It is well recognized that many patients do not follow their prescribed care plans. While lack of conformance tends to be greatest for behavioral changes, as many as half of patients are significantly out of conformance to their medication regimens within a year. The International Society for Pharmacoeconomics and Outcomes Research (ISPOR) defines compliance or adherence as “the consistency and accuracy with which a patient follows a recommended medical regimen, usually referring to a pharmacotherapeutic regimen.”

While it is easiest to describe conformance with drug treatment in terms of frequency (e.g., twice a day), intensity (e.g., 40 mg), duration (e.g., 10 days or chronically), and timing (with meals), most treatment regimens (or care plans) can be characterized in these terms. For example, exercise can be specified as: at least 5 times a week (frequency); sufficient to get heart rate to 120 (intensity); at least 30 minutes (duration); and not within 2 hours of bedtime (timing).


“Adherence to Long-Term Therapies: Evidence for Action.” 2003. World Health Organization. 16 Feb. 2006

Berger, M., Bingefors, K., Hedblom, E., Pashos, C., and Torrance, G., eds. “Health Care Cost, Quality and Outcomes”: ISPOR Book of Terms (2003).

Chapman, R., et al. “Predictors of Adherence with Antihypertensive and Lipid-Lowering Therapy”. Arch Internal Medicine. 165 2005 1147-1152.

Haynes, R., McDonald, H., Garg, A., and Montague, P. “Interventions for Helping Patients to Follow Prescriptions for Medications.” Cochrane Database Syst Rev (2002).

Osterberg, L., Blaschke, T. “Adherence to Medication.” New England Journal of Medicine (2005) 353:487-497.

Sikka, R., Xia, F., and Aubert, R. “Estimating Medication Persistency Using Administrative Claims Data.” American Journal Managed Care. 11 (2005) 449-157.


Confounding is the distortion of an estimate as a result of the effect of other factors (e.g., variables). This particular measurement issue significantly affects the population health management industry because of the number of variables that affect an outcome.


Confounding is a distortion in the degree of association between exposure and outcome because of a mixing of effect between the exposure and an external factor known as a confounder (or confounding variable). Another way to describe confounding is the failure to adjust fully for factors related to both the outcomes and the independent variables. Confounding is an important concept because confounders can lead to an overestimation or an underestimation of the true association and may even change the direction of the observed effect. It may lead to double counting because the confounding variables are associated and overlap. Confounding is a threat to internal study validity.

Overlap = Potential Confounding

Confounding results from complex interplay among various “exposure” factors and a population health management intervention. Understanding the concept of confounding helps researchers: 1) control the study for variables that are highly correlated (e.g., physical activity level and age); and 2) accurately interpret findings of causality or association in observational and experimental studies. The ability to address possible confounders allows the researcher to state whether an observed association such as improved A1C levels in diabetic patients enrolled in a population health management program is real or due, at least in part, to the effects of differences between study groups—for example, due to changes in benefit designs.

The direction and magnitude of the effect of potential confounding factors needs to be identified and estimated. While the former depends on the interrelationship between the confounder, exposure to an intervention, and disease state, the latter depends on the strength of association between the exposure and the disease state. Positive confounding occurs when the observed association is greater than what the true association would be, and negative confounding occurs when it is less. In a study of exercise relating to diabetes, for example, age could be a negative confounder because older people tend to exercise less and have higher costs associated with diabetes than do younger ones. Hence, the group of exercisers in the study would contain a greater proportion of younger people with diabetes, and the study would tend to underestimate the protective benefits of physical activity.


A confounding factor is defined as an intervening variable that has the ability to distort observation of relationships and variables being studied. While related to the outcome of interest, confounding factors are extraneous to the study. Moreover, they are not randomly distributed across the groups that are being compared. Confounding factors can hide a true association or give the appearance of an association when none exists.

The problem is that the value of a confounding variable is inextricably linked to that of another variable, making it difficult to disaggregate the effects of the two or more variables in a study. Several methods are available to control confounding, either through study design or during the analysis of results. The methods commonly used to control confounding in the design of an epidemiological study are randomization, restriction, and matching. At the analysis stage, confounding can be controlled by stratification or statistical adjustment.


  • Generally, a confounding variable is associated with both the exposure to an intervention and the disease it addresses. (See illustration.) If not, a factor is not a confounder.
Interrelationship among Confounding Factor, Exposure and Intervention
  • The confounding factor must independently be able to predict the disease outcome from the intervention. There must be some level of association between the confounding factor and the disease even among groups that are not exposed to disease management. As an example, diabetes prevalence increases with age, so age is considered a potential confounder.
  • A confounder is not just a link between exposure and the disease. The following describes a linkage but not a confounding relationship.

Exposure to intervention → confounding factor → disease status

  • Common potential confounders include: gender, age, and income level. For example, age and gender are associated with most chronic illnesses/conditions and are related to the presence or absence of exposure to a chronic care management program. Stratifying by age can help resolve this.
  • In a study, the effect of any single confounder needs to be considered in the context of the effects of all other possible confounding factors.
  • One can test the impact of a potential confounding variable by adjusting for the variable to see if there is a change in the estimate of association between the intervention and change in disease state during data analysis.
  • Identification of a potential confounder requires more than just determining whether an association with exposure to a chronic care management program or disease state is statistically significant. In a small study, for example, a confounder can seriously skew the findings even if it is not statistically significant, and vice versa.

Most confounding factors are correlated with each other. For example, risk of myocardial infarction in 40-year-olds is attributable to both gender and smoking. Males of this age are more likely to have a myocardial infarction than women, independent of smoking. But males are also more likely to smoke. In this example, gender and smoking are potential confounders and they are correlated with each other.


Historical changes during the period that are unrelated to the intervention will affect the outcome measures for a chronic care management program. The findings show a 10% decrease in inpatient admissions for congestive heart failure, which is attributed to nurse outreach to heart failure patients and teaching them how to self-manage their condition.

During the same period, however, the health plan instituted a complex prior authorization process for hospital admissions. In the absence of a comparison group that experienced the same benefit design change, it is not possible to determine whether the 10% reduction in admissions is due to the chronic care management program or to the prior authorization process. The study was confounded by the change in benefit design.


Beaglehole, R., Bonita, R. and Kjellstrom, T. (1993). Basic epidemiology. Geneva: World Health Organization.

Hennekens, C. and Buring, J. (1987). Epidemiology in medicine. Boston: Little, Brown & Company.

Kraut, A., Mustard, C., Walld, R., and Tate, R. (April 2000). Unemployment and health care utilization. Scand J Work Environ Health, 26(2), 169 -77.

Lilienfeld, D., and Stolley, P. (1994). Foundations of epidemiology. (3rd ed.). Oxford: Oxford University Press.

Consumer-Directed Health Plan

Consumer-directed health plans (also known as consumer-driven health plans or consumer choice health plans) offer health insurance coverage that combines a high-deductible health plan (HDHP) with a health savings account or a health reimbursement arrangement to provide insurance coverage and a tax-advantaged way to help save for future medical expenses. Consumer-directed health plans are intended to provide greater responsibility, flexibility, and discretion over how individuals use their health care dollars.


The general features of an HDHP include:

  • Higher annual deductible than traditional health plans. By law, a HDHP has a minimum annual deductible of $1,050 for self coverage and $2,100 for self and family coverage (the deductible amount is indexed every year);
  • Annual out-of-pocket limits that do not exceed $5,000 for self coverage and $10,000 for family coverage; and
  • A HDHP program may be offered with a preferred provider organization, health maintenance organization, or point-of-service plan.

Consumer-directed health plans offer a variety of plan types including medical savings accounts, health savings accounts, health reimbursement accounts, and flexible spending accounts.

Consumer-Driven Health Care (CDHC)

Consumer-driven health care (CDHC) is a broad term for health benefit plan designs structured with incentives for the judicious use of services through higher deductibles and coinsurance/cost-sharing. CDHC plans include health savings accounts and flexible spending accounts. By managing their own health care expenditures, individuals are encouraged to become more engaged in decisions about who delivers their care. Current thinking suggests that CDHC plans will lead to transparency in the health care delivery system.  


National Center for Policy Analysis. Consumer-Driven Health Care (CDHC).

Nash DB, Reifsnyder J, Fabius RJ, Pracilio VP. Glossary of Population Health: Creating a Culture of Wellness. Sudbury, MA: Jones & Bartlett Learning; 2011.

Continuity of Care Record (CCR)

The ASTM Continuity of Care Record (CCR) was developed in response to the need to organize and make transportable a set of basic information about a patient's health care that is accessible to clinicians and patients. The CCR is intended to foster and improve continuity of care, reduce medical errors, and ensure a minimum standard of secure health information transportability. Adoption of the CCR by the medical community and information technology vendors will be a great step toward achieving interoperability of medical records (one of Center for Health Information Technology’s guiding principles).


ASTM. From:

Continuum/Care Continuum

The care continuum represents comprehensive, coordinated and integrated health services that improve the quality and value of care across all states of health and care settings. Care continuum services include health and wellness promotion, chronic care management, and care coordination. Providers of these services can include primary care physicians,


CCA website.

Coordination of Benefits

Coordination of benefits is the mechanism developed to prevent duplication of payment when more than one payer covers a person. It limits the total benefits received to no more than the actual amount of cost incurred for care, or to something less than the actual amount incurred for care. It does this by informing all subsequent payers in the billing chain of the benefits determined and/or paid by all previous payers.


The Health Insurance Portability and Accountability Act defines coordination of benefits as follows: (45 CFR 162.1891) Coordination of Benefits Transaction: “The coordination of benefits transaction is the transmission from any entity to a health plan for the purpose of determining the relative payment responsibilities of the health plan, of either of the following for health are: (a) Claims. (b) Payment information.”

The payer who pays first on a claim is known as the primary payer. The secondary payer will pay second. If the secondary payer integrates its payments with those of the primary payer, coordination of benefits will have occurred.

There is more than one method for calculating the secondary payer’s responsibility under a claim. The examples use the following basic data:

Claim Amount: $5,000
  Primary Payer Secondary Payer
Deductible $100 $500
Coinsurance 20% 10%

Method 1: Under Method 1, the secondary payer processes a claim under its own rules as though it were primary, calculating an allowable amount. Then it deducts any payments already made by a primary payer from the allowable amount and pays the difference.


The primary payer will pay $3,920 for this claim = ($5,000 – $100) * 0.8.

The secondary payer allowable amount will be $4,500 ($5,000 – $500). Secondary payer payable amount is $4,050, or 90% of $4,500.

Secondary payer pays $4,050 less $3,920, or $130.

Method 2: Under Method 2, the secondary payer processes the net amount of the claim under its own rules as though it were primary.


The primary payer will pay $3,920 under this claim = ($5,000 – $100) * 0.8.

Amount submitted to the secondary payer is $1,080 ($5,000 – $3,920).

Secondary payer pays: ($1,080 – $500) * 0.9, or $522.

Note: The term crossover is frequently used for payer-to-payer coordination of benefits, since the claims data cross over from one payer to another.


“Coordination of Benefits (COB).” Retrieved from


Correlation is a term describing how strongly pairs of variables are related, if at all.

With positive correlation, the value of one variable increases as the value of another variable increases; with negative correlation, the value of one variable decreases as the value of the second variable increases. The degree of correlation can be measured in statistics using various statistical correlation coefficients, including the Pearson correlation coefficient and the Spearman’s rank correlation coefficient.

These coefficients have values between –1 and +1, where –1 signifies the maximum negative correlation, and +1 signifies the maximum positive correlation. Coefficients of 0 imply a lack of linear association. The Pearson correlation coefficient is greatly affected by outlying values (values that differ greatly from the others) because it relies on actual values. The Spearman correlation coefficient is calculated based simply on the rank of the different values in relation to each other. See also association. A correlation does not imply causality. A nonsense correlation or ecologic fallacy may occur when a linear trend appears to exist between two variables but they are actually unrelated—for example, if a relationship were observed between disease management outcomes and phases of the moon.


Bland, M. (1995). An introduction to medical statistics. (2nd ed.). Oxford: Oxford University Press.

Last, J., Spasoff, R., Harris, S. and Thuriaux, M. (2001). A dictionary of epidemiology. (4th ed.). Oxford: Oxford University Press.

Cost avoidance

Cost Avoidance is a comparison of what occurs and what it costs today with what occurred before the implementation of a specific intervention. Basic example:

  1. Baseline medical service consumption at baseline rate
  2. Current medical service consumption at current rate
  3. Baseline medical service consumption at current rate
  4. Cost Avoidance (step 3 – step 2)

Dependent and independent factors that influence variability must be controlled to demonstrate comparability when using model-based cost avoidance studies. 

 Thus, adjustments to the baseline must occur to account and offset these variabilities. 

  1. Determine consumption in the baseline period.
  2. Determine consumption and cost in the current period.
  3. Adjust the baseline consumption for dependent and independent variables so that it is more closely comparable to current conditions.
  4. Calculate the cost of step 3 using today's rates.
  5. Subtract the cost associated with step 2 from the cost associated with step 4. The result is cost avoidance — what you would have spent minus what you really did spend. 

Cost-Effectiveness Analysis (CEA)

Cost-effectiveness analysis (CEA)  is a form of economicanalysis that compares the relative costs and outcomes (effects) of two or more courses of action. Cost-effectiveness analysis is distinct from cost-benefit analysis, which assigns a monetary value to the measure of effect. Cost-effectiveness analysis is often used in the field of health services, where it may be inappropriate to monetize health effect. Typically the CEA is expressed in terms of a ratio where the denominator is a gain in health from a measure (years of life, premature births averted, sight-years gained) and the numerator is the cost associated with the health gain. The most commonly used outcome measure is quality-adjusted life years (QALY). Cost-utility analysis is similar to cost-effectiveness analysis.

 There are a variety of intervention specific cost-effectiveness analyses focused on key health activities including Heart Failure, CAD, Diabetes, Asthma and Cancer.

Cost Shifting

Transfer of the payment burden from one entity to another, including to the individual patient. 

Cost shifting may occur in response to lower payments from one set of payers which results in higher payments being expected/required from the remaining payer(s)—shifting of expected burden. An example is the alleged transfer of costs by hospitals or physicians from Medicare to private payers, or the transfer of costs from state to local government. Cost shifting may also occur if lower payments are afforded to benefit one set of payers over another set of payers—shifting of previous arrangement. Cost shifting may require an extra-contractual aspect. For example, third-party reimbursement of costs that fall within an insurance contract does not constitute cost shifting.

Cross-cutting Measures

Linking traditionally separate or independent measures.


In a population health management program, there is a need for cross-cutting measures that could be used to assess impact at a population level, regardless of the differences in service offerings. It is clear that there is still a need for service- and disease-specific measures as well, but cross-cutting measures could be useful for assessing overall impact and program comparison. With this understanding in mind, a core list of measures was developed to support the overall evaluation of a population health management program:

  • Medical costs
  • Health care utilization – appropriate use
  • Health risks/behaviors
  • Quality of life
  • Health status
  • Productivity
  • Psychosocial drivers
  • Program satisfaction
  • Process/operations measures


Care Continuum Alliance. (2010). Outcomes guidelines report, vol. 5. Washington, DC: Care Continuum Alliance.

cross-cutting. (n.d.). Collins English Dictionary - Complete & Unabridged 10th Edition. Retrieved December 08, 2010, from website: