PHM Glossary: R


Randomization refers to the allocation of individuals to study groups according to chance. Random allocation, hypothetically, should ensure that the control and experimental groups are similar at the start of an investigation.

Randomization eliminates errors that might be introduced, such as the personal judgment and prejudices of the investigator. Random allocation is systematic and is haphazard because it follows a predetermined plan, typically with the aid of computers or a table of random numbers.

A random sample is one that is developed through the selection of samples in such a manner that each possible unit has a fixed and equal chance of selection.


Last, J. A Dictionary of Epidemiology, 4th ed. Oxford: Oxford University Press (2001).

Rare Diseases/Conditions

The Office of Rare Diseases of the National Institutes of Health defines rare diseases as those that have a prevalence of fewer than 200,000 individuals in the United States. The National Organization for Rare Disorders maintains a database of over 1,150 disorders. This definition focuses on the small subset of these disorders for which chronic care management programs have been developed and implemented. This category of rare diseases generally shares the following characteristics:

  1. Chronic, not acute
  2. Costly to treat
  3. Amenable to interventions designed to:
    • Prevent complications,
    • Delay disease progression,
    • Reduce avoidable utilization, and
    • Improve quality of life.


Chronic care management programs most frequently focus on chronic conditions that are both costly and prevalent, have evidence-based guidelines, and have proven interventions that promote empowered patients with improved self-management skills. Given these criteria, the most common programs are for:

  • Diabetes mellitus,
  • Heart failure,
  • Coronary artery disease (CAD),
  • Asthma,
  • Chronic obstructive pulmonary disease (COPD), and
  • Chronic kidney disease (CKD)/end-stage renal disease (ESRD).

In addition, care management for both depression and high-risk maternity are often provided.

There is, however, another set of costly but less prevalent chronic conditions for which chronic care management programs have been developed and successfully implemented. These rare conditions have proven interventions designed to mitigate the risk of patients developing catastrophic exacerbations. Rare conditions for which chronic care management programs have been developed include:

  • Seizure disorders,
  • Rheumatoid arthritis,
  • Multiple sclerosis,
  • Parkinson’s disease,
  • Systemic lupus erythematosus,
  • Myasthenia gravis,
  • Sickle-cell anemia,
  • Cystic fibrosis,
  • Scleroderma,
  • Polymyositis,
  • Chronic inflammatory demyelinating polyneuritis (CIDP),
  • Amyotrophic lateral sclerosis (ALS),
  • Dermatomyositis,
  • Gaucher’s disease, and
  • Hemophilia

Since chronic care management programs often 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. See the definition for false positives in this guide for a detailed discussion. Chronic care management programs 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 care management programs often develop exclusion criteria as well as inclusion criteria. Exclusion criteria attempt to identify those patients with the disease in question who would not benefit from the proposed disease management program (e.g., institutionalized patients). These criteria are generally specific to the program and not generalizable across all situations. Examples of common exclusionary criteria are provided under the definition of exclusionary conditions.

Chronic care management programs for rare conditions are designed to prevent complications, delay disease progression, reduce avoidable utilization, empower patients, and improve their quality of life. Interventions often include:

  • Assessment and monitoring of patient status;
  • Patient education designed to improve the patient’s knowledge of the condition, their self-management skills, and compliance with medical regimens;
  • Care coordination services to promote better communication among the patient, primary care physicians, and specialists; wherein Information Technology will play a pivotal role
  • 24-hour nurse or care management coordinator (especially in developing countries) support lines; and
  • Connections to patient support services in the community or via online forums and social media.

As the volume of patients covered by these programs is generally low, programs often involve more intensive interventions than for chronic care management programs for more prevalent conditions. Programs for rare conditions are frequently integrated with existing case management programs to facilitate management of patients during catastrophic exacerbations of their condition.


“CPT (Current Procedural Terminology).” American Medical Association. 1 Feb. 2006. 9 Feb. 2006 http://www.ama­

NORD Home Page. National Organization for Rare Diseases. 23 Feb. 2006

“The National Drug Code Directory.” US Food and Drug Administration. 10 Jan. 2006. Department of Health and Human Services. 16 Feb. 2006

Reachable (Reach Rate)

Reachable refers to the ability of a chronic care management program to make contact with a target individual. The existence of valid contact information (address, email, text or telephone number, depending on the nature of the program) is a necessity.


Reconciliation is the process whereby a population health management organization calculates the outcomes of a client program at the end of a period. Specifically, during the reconciliation process, the population health management organization may calculate the amount of savings or return on investment applicable for that period and compare them with any agreed contractual terms.


In the population health management industry, the term reconciliation is used to describe the annual accounting or reporting process in which the organization reports to its customers on financial, clinical, and operational measures and, to the extent that these are specified contractually, compares them against the contractual requirements.

The use of the term reconciliation to describe the annual reporting of the population health management organization’s value is specific to the population health management industry. In insurance terminology, for example, a similar process of evaluation of a client financial position is termed experience rating or accounting. In practice, despite the use of the term reconciliation, little reconciliation (in the sense that this term is understood by auditors or actuaries) may take place during the calculation of savings.

The term reconciliation in finance describes a process of ensuring the consistency between certain source data and another set of data (often derived from the source data). In population health management, given the maturity of the industry and the relative shortage of resources to complete projects (such as the reconciliation of financial outcomes to the original source data) it would appear that this function is overlooked. However, its importance in ensuring confidence among financial users of disease management data cannot be overestimated.  

The following example illustrates both the need for reconciliation to source data and the credibility problem sometimes encountered when presenting disease management results that have not been tied back to a source.



  • In the reconciliation, the vendor claims a reduction of 500 admissions due to the intervention program;
  • Total admissions for the entire population are 1,000, after interventions;
  • Chronic population of 10% of the entire population; and
  • Entire population numbers 10,000.

Admissions/1,000 rate = 150 per 1,000 per year, preintervention.

Admissions/1,000 rate = 100 per 1,000 per year, after the intervention.

Reduction in admissions/1,000 on the chronic population = 500 per 1,000 per year.  

Total admissions, preprogram are 1,500. Assuming that the program reduces admissions by 50%, 1,000 of these admissions were in the chronic population. This implies that the non-chronic population’s admissions must be 500, or 55 per 1,000.

The question is whether 55 admissions per 1,000 is a reasonable number for the client, and, conversely, whether a reduction of 50% in chronic population admissions is reasonable. The lower the reduction in chronic population admissions, however, the lower the implied admission rate in the non-chronic population. For example, if the reduction in chronic admissions is 40% instead of 50%, the admissions per 1,000 in the non-chronic population is 28 per 1,000.

A vendor can be certain that the client will be familiar with its own experience and perform reconciliation analysis such as this to test whether claims made about the financial benefits of the program appear to be reasonable. To the extent that a health plan has a different reported experience, this may be due to differences in population (the population health management organization reports only for the chronic subset) or because admissions are defined differently by one of the reporting organizations.


A reconciliation report should include the following information:

  1. Description of the population health management program, including services provided;
  2. Effective dates, program start date, measurement and baseline period start and end dates;
  3. Description of the eligible population;
  4. Description of the target population, including chronic disease(s) definitions and number/prevalence of chronic patients identified;
  5. Description of any excluded members from the population health management program (either for services or for measurement);
  6. Numbers of eligible and enrolled member months;
  7. Baseline or reference population cost per member per month for total and chronic populations;
  8. Baseline or reference population utilization per member per month (admissions; emergency room visits, bed-days, etc.) for total and chronic populations;
  9. Cost and utilization Trends (if applicable) for total and chronic populations;
  10. Cost and utilization Trends used for calculation purposes;
  11. Projected cost and utilization per member per month for chronic population;
  12. Estimated savings per member per month for chronic population;
  13. Estimated dollar savings for chronic population;
  14. Actual cost of chronic and total population for measurement period; and
  15. Population health management program cost for measurement period.

Readiness to Change

Readiness to change is an assessment of an individual’s commitment and motivation to change a problem behavior according to the Transtheoretical Model (a.k.a. Stages of Change Model).

For more information, see behavior modification - Prochaska and DiClemente Stages of Change Model.

Regional Health Information Organization (RHIO)

A health information organization that brings together health care stakeholders within a defined geographic area and governs health information exchange (HIE) among them for the purpose of improving the quality, safety and efficiency of health and care in that community. A RHIO can be organized to support a community, groups of communities, a statewide area or a region crossing state boundaries. Key stakeholders in a RHIO may include:

  • Health care institutions and personnel that render care;
  • Businesses and government agencies that reimburse for those services;
  • Researchers and professionals who are engaged in health improvement activities;
  • Public health agencies; and
  • Consumers of health care.

The RHIO enables, facilitates and fosters collaboration among stakeholders to attain a useful level of information sharing through HIE. A RHIO may operate directly or contract for HIE services.

To be designated a RHIO, an entity needs to have certain core features:

  • data-sharing participants that are separate and distinct organizations;
  • well-defined and transparent processes to facilitate the interoperable exchange of health information across the range of participating stakeholders;
  • inclusive and convene various types of stakeholders in the geographic area;
  • ability to provide additional technical and operational services supporting its primary purpose. Examples include:
    • The technology and support for physicians to create and use electronic records.
    • Electronic exchange of messages in a secure format to report and distribute medical test results.
    • Data on specific patients to first responders in a community (e.g., whether a patient has signed a DNR (do not resuscitate) order).
    • Coordinated electronic health record and personal health record platforms for the region.


The National Alliance for Health Information Technology Report to the Office of the National Coordinator for Health Information Technology on Defining Key Health Information Technology Terms. April 28, 2008.

Registry (or Patient Registry)

A database for tracking the clinical areas and outcomes of a defined patient population. Most registries are used to support care management for groups of patients with one or more chronic diseases such as diabetes, coronary artery disease, or asthma. Paper-based registries have long been used in the care of patients with chronic disease, however computerized registries provide users with an automated way to store data and to create, sort, and display lists of patients and other data for use in planning, quality improvement, reporting, and direct care delivery.


Given that diabetes affects 8.3% of the U.S. population, it is one chronic condition targeted by registries. In addition, diabetes has a target population that can be identified, and there is evidence that certain tests, like retinal exams, LDL levels, and A1C levels, can correlate with quality of care in diabetes. In a population health management program, a registry is critical to the success of surveillance strategies for a population.


(AHRQ Knowledge Library)

Centers for Disease Control and Prevention. National diabetes fact sheet: national estimates and general information on diabetes and prediabetes in the United States, 2011. Atlanta, GA: U.S. Department of Health and Human Services, Centers for Disease Control and Prevention, 2011.

Regression to the Mean (RTTM)

Regression to the mean is a statistical term meaning simply that, if left to themselves, data will move toward the mean (either increasing or decreasing).


In most population health management discussions, the concern with regression to the mean (RTTM) is the potential misrepresentation of savings because claims of the chronic population that are high one year are observed to decline the next, even in the absence of intervention. The statistical term RTTM defines regression toward the mean, and it is important to remember that claims can move up or down toward the mean. In chronic care management there is often a built-in bias toward a reduction in claims, because candidates for a chronic care management program would most likely be selected from high-utilizing, not low-utilizing, patients. The following chart illustrates the typical progress of a chronic patient’s claims over a period of 8 quarters:


Depending on when this individual’s experience begins to be tracked for the purpose of measurement, RTTM may be captured in the claims data. For example, if the identifying event for a chronic care management program is the hospitalization claim that occurred in Quarter 3, and this claim is included in the pre-experience, the tracking of the post-experience will appear to show that the chronic care management program has reduced cost, when in fact the cost reduction is the natural course of the individual’s experience. This phenomenon is illustrated below. In this example, members are identified (through claims) and enrolled in a program. The experience pre-enrollment is included in the pre-experience; the experience after enrollment is included in the post-experience.



1. Population selection – selection bias


A chronic care management program based recruitment on a recent hospitalization in the baseline cost measure of the sample. Because the hospitalization event represents the high point in patient cost for a number of months, tracking the patient’s cost of time will show evidence of RTTM. Costs for a group of patients identified through hospitalizations will inevitably show lower costs during the following (intervention) year. (There may be a few exceptions to this rule, consisting of certain diagnoses such as ESRD, HIV and others, where costs tend to increase over time.)

Solution: While recruitment for management may continue to be performed on a targeted basis, for measurement select the entire population. A set of identifying criteria that include pharmacy users, too, can ensure a population-based measurement.

2. Impact – analysis


When assessing the impact of an intervention on a defined population (e.g., asthmatics), chronic care management providers will often measure the impact of those who have participated in the intervention, which is a smaller population than the whole asthmatic population. This is a non-representative sample and will maximize RTTM.

Solution: Measure impact on the entire population identified with the condition. When the entire population consists of a uniform distribution of members in different states of claim (pre-claim, claim, post-claim), the effect of RTTM will be neutralized.


The “patient as their own control” design is a study design that follows a patient (or group of patients) over time, comparing preintervention costs with post-intervention costs. As the above examples show, any study that uses a “patient as their own control” design should therefore be viewed with skepticism, based on the effect of regression to the mean inherent in the design.


In addition to its effect at the individual level, RTTM has implications for chronic care management studies of a population. The following chart illustrates the more general impact regression to the mean (claims increasing as well as decreasing) may have on an analysis:

Distribution of Members and Claims
  Projection Period  
Historic Period Group Historic Period Cost $0-$2 42-$25 $25+ Projection
Period Cost
$324 $327
$5,658 $668
$49,032 $847
Total $1,230 $355 $5,851 $49,377 $1,581

$ = $1,000

In the above table, in which data are for the continuously enrolled members of a managed care plan for the 2 years 1997 and 1998, members are allocated into categories based on their cost category in Year 1 (historic period). The members of this population are drawn from a health plan with limited managed care interventions: preauthorization, some concurrent review and in-hospital case management, but no outpatient case management or chronic care management. One percent of members have costs in excess of $25,000, with an average paid claim cost of $49,032. The outcome of each category of members is shown in Year 2 (projection period). Ninety percent of Year 1 low-cost members remain in the same category in Year 2, with approximately the same average cost. (Note that only members who were enrolled in Year 1 are included in this analysis, so new members or members who had no claims in Year 1 are excluded.)

The third italicized line under the projection period distribution of members and costs indicates the source of that period’s membership in the prior year. For example, 64% of the intermediate group of members in Year 2 come from the prior year’s low-cost members. Regression to the mean is illustrated by the outcome of the 1% of members who were high cost in Year 1: 26% of these members are low cost in Year 2, and 46% of these members are in the intermediate group. Only 28% of the members continue to experience high costs in Year 2, while nearly three-quarters of members have costs less than $25,000. The average cost of the high-cost members declines from $49,032 in Year 1 to $21,017 in Year 2. Those readers familiar with statistical techniques will recognize that this analysis is an example of a Markov process.

The statistical artifact of RTTM exists within a population as well as at the individual level. The moderate cost group above consists largely of chronic patients. Note that in this example, if the population tracked is the Year 1 moderate cohort, the average cost is observed to fall 4.6%, from $5,658 in the baseline year to $5,398 in the intervention year, in the absence of any interventions. If the population tracked is the Year 1 moderate population compared with a similarly defined Year 2 moderate population, costs increase by 3.4%, from $5,658 to $5,851.


Dove, H., and Duncan, I. “An Introduction to Care Management Interventions and their Implications for Actuaries.” unpublished manuscript prepared for the Society of Actuaries Health Care Section.

Parke, R. “Insight into Two Analytical Challenges for Disease Management.” Seattle, WA: Milliman (2004).

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


Reliability is the extent to which a means of measurement will yield the same results time after time.

Many types of reliability are examined, such as (1) interrater reliability – the degree to which different health care professionals, interviewers, etc. will find the same result; (2) intersubject reliability – the degree to which a test on the same subject will produce the same result at different times; and (3) intermethod reliability – the degree to which different measurement techniques will produce the same result. A lack of reliability may be the result of instability of the measurement technique or instability of the attribute being measured.


Last, J. A Dictionary of Epidemiology. 4th ed. Oxford: Oxford University Press (2001).


Remote Patient Monitoring

The electronic transmission of health care data either entered directly by a patient (or his/her caregiver) or through a medical device to a clinician’s electronic health record (HER) or a patient’s personal health record (PHR). The ability for a clinician to monitor patient information captured remotely in an ambulatory setting, such as physiological, diagnostic, medication tracking, and activities of daily living (ADLs) measurements, may be the key enabler for the management of chronic health problems and initial management of new conditions.



Retrospective Study

A retrospective study uses data about past events or experiences to evaluate population health management, or other study hypotheses. A retrospective study may be jeopardized by recall bias, a systematic error due to errors in the recall of past events. Participants may not correctly remember events because they happened in the past. In addition, participants with disease tend to remember better the negative events that they believe may have caused their disease.  

Return on Investment (ROI)

Return on investment (ROI) is a metric used to assess, report, and compare the value of population health management and other intervention programs. ROI is defined as follows:

  Total savings attributable to the program  
  Total cost of the program  

Although elsewhere in other financial applications ROI is generally calculated on a net basis (Total Savings – Total Cost)/Total Cost, this is not the case in population health management, where ROI is generally reported on a gross basis.


Calculation of ROI


Rate of return on investment = i

Evaluation period = N (may be greater than or less than 1 year)

Savings attributable to the program in year t = St

Cost of the program attributable to Year t = Ct

Both costs and benefits are measured during a measurement period, although the two periods need not be of the same duration (start-up costs, for example, may be incurred before the beginning of enrollment and emergence of savings from interventions). Because costs are incurred differently in different time periods, costs may be annualized (for example, start­up costs amortized over the life of the program).

ROI (i) is found by solving for i in the following expression:

Σ St / (1 + i) t = Σ Ct / (1 + i) t
t=1 t=1

When the period of measurement is not 1 year, adjustments should be made to the formula. This expression applies equally when t < 1, although the validity of results becomes increasingly less reliable when t < 1.

In other financial applications, rate of return is generally expressed on a net basis (i.e., as the difference between savings and cost). In population health management applications, it is traditional to express the rate of return in gross terms—that is, including the cost. It is important that the user of this information clearly define and understand the basis of the calculation.


1. In assessing whether to implement a population health management program, the projected ROI is an appropriate metric. Payers should have a risk-free hurdle rate (rate of return on investment required for a viable project) against which the planned or projected return on a population health management program should be assessed. For example, this hurdle rate could be 15% post-tax, or

Pre-tax hurdle rate:

(1 – 0.35)   = 23% pre-tax, for institutions that pay a 35% corporate tax rate.

Because population health management outcomes are subject to statistical fluctuation, they do not meet the criterion of risk-free, so a risk-margin may be added to the risk-free (hurdle) rate of return when assessing a proposed project. In any event, the pre-tax, risk-adjusted hurdle rate is unlikely to exceed 30% to 35%. A project that has a planned return of more than 30% to 35% (implying a planned ROI of greater than 1.35) either has a very high risk-margin built in, or it has been planned to a suboptimal scope (additional interventions could be performed that increase savings and still meet the minimal risk-adjusted hurdle rate).

2. As a statistic for reporting the value of a program, ROI is the metric favored by the population health management community. Unfortunately, for many reasons, it can be a misleading metric, making conclusions and comparisons between programs difficult to draw.

  • Program cost (denominator). There is no single agreed definition for program costs. Generally, in a program evaluation exercise, total costs should include:
    1. Direct costs (i.e., salaries of internal staff, vendor fees);
    2. Indirect costs (i.e., internal support activities, such as information systems, mail and printing, medical director involvement, etc.);
    3. Management costs (i.e., costs of internal management involvement, including program management, medical management, and financial management);
    4. Overhead and other allocated costs (i.e., generally, expenses allocated to internal resources for overhead such as rent, employee benefits, senior management load, etc.); and
    5. Setup costs (i.e., one-time expenses that are incurred before and coincident with the start of a program). The formula above, which discounts the pattern of future emerging savings, can accommodate setup costs as an element of total costs without further adjustment.
  • Program savings (numerator) is defined elsewhere.

The ex ante or planned ROI has been defined above. The ex post, or actual measured ROI, will be subject to the operational and stochastic factors that will cause actual ROI to diverge from the planned level.


A population health management program has the following cost structure:

Vendor costs:

$1,000,000 annually

Internal staff costs (including information technology resources, data preparation, etc.):


$250,000 annually

Start-up costs:


Contract period is 3 years. Assuming that vendor and staff costs are the same amount each year, and amortizing the start-up costs over 3 years, the annual program cost is $1,300,000.

Savings emerge as follows:

Year 1


Year 2:


Year 3:


The payer has a risk-free hurdle rate of 25% pre-tax.

Should the payer invest in this program?


Measured at the end of Year 1, ROI is





The project does not meet the minimum return requirement for Year 1. (In fact, ROI < 1.0 in Year 1.)Over 3 years, the discounted present value of costs is as follows (assuming all costs are paid at the beginning of the year.)


Over 3 years, the discounted present value of costs is as follows (assuming all costs are paid at the beginning of the year.)

Discounted present value of costs is:

present value costs =

$1.4 million + $1.25 million + $1.25 million
  1.25 1.252 1.253

Discounted present value of savings is:

present value savings =

$0.5 million + $1.5 million + $2.5 million
  1.25 1.252 1.253


In this example, discounted present value of costs ($3.2 million over three years, discounted at the payer’s hurdle rate) is less than the discounted present value of savings ($3.3 million). Therefore, the project (over a 3-year horizon) meets the payer’s return requirements and is a candidate for investment.


First-year ROI is 38% (500,000/1,300,000). Over a 3-year time horizon, the ROI on the entire program is approximately 130% (the rate of return that causes the present value savings to equal the present value costs).


In the following example, the use of ROI as a measure suggests that Program 1 is the better investment. However, evaluation of program net savings as a measure suggests that ROI is a misleading measure and that the payer is better off investing in Program 2.

Which is the better population health management program?

  Program 1 Program 2
Number of Members 10,000 10,000
Number of Chronic Members 100 500
Annual Cost $50,000 $250,000
Annual Gross Savings $150,000 $400,000
ROI 3.0x 1.6x
Per member per month (net) $0.83 $1.25


Duncan, I. “It’s Time for the Industry to Move on from Return on Investment.” Disease Management. 7 (3) (October 2004).

Fetterolf, D., Wennberg, D. and DeVries, A. “Estimating the Return on Investment in Disease Management Programs Using a Pre-Post Analysis.” Disease Management. 7(1) (Spring 2004): 5-23.


The ability to withstand critical, peer scrutiny, by adhering to sound scientific, actuarial, economic, and statistical principles.

Risk Adjustment/Predictive Modeling

Risk adjustment is the process of making statistical adjustments for patient differences that affect the probability that a patient will achieve a specific outcome in a defined time period. The goal of risk adjustment is to quantify differences in health status among populations before comparing outcomes across different patients, treatments, providers, health plans, or populations.

Predictive modeling is the process of forecasting future health care expenditures or resource utilization based on differences in individuals’ health statuses. An important objective of predictive modeling is to allocate disease management resources to patients who are predicted to incur high health care expenses (in the absence of some kind of care intervention).


Risk is the probability of some event in a given time frame. In clinical medicine, the event generally is an unfavorable one, such as death. In health services research or chronic care management, risk is more commonly used in conjunction with medical care utilization (e.g., the risk that an enrollee will be hospitalized, admitted into an intensive care unit, or experience a medical complication).

Health economists, health actuaries, medical managers, and chronic care management professionals are interested in the risk that a health plan member will incur a certain level of medical expenditures or use certain medical resources such as inpatient hospital facilities, emergency rooms, and surgical or diagnostic procedures in a future time period (usually 12 months).

Risk factors are patient characteristics that affect the probability of a certain event. For example, we may be interested in assessing how age, gender, smoking history, and race affect the probability that a patient with cancer of the lung will die of that disease within 1 year (or some other period) from the first treatment.

To estimate the risk of the preceding clinical outcome, we first must establish what risk factors affect the probability of that outcome and then construct a mathematical model that explicitly quantifies those risks. To forecast the expected number of patients with lung cancer who die within 1 year, each patient’s characteristics must be collected and used in that mathematical model.

A common study application of the risk model in epidemiology and health services research is to compare the outcomes of different groups of patients. For example, the 1-year mortality rates for two surgeons treating patients with lung cancer at the same hospital may be of interest. Because the patients that each surgeon treats have different characteristics, the risk adjustment model is used to make a statistical adjustment to compare the surgeons’ clinical performance.

There are commonly five uses of risk adjustment:

  1. Adjust payments and contributions, based on health status;
  2. Provider profiling (comparing resource utilization and treatment patterns across providers);
  3. Provider reimbursement, based on patient panel health status;
  4. Underwriting and rating (setting rates, based on health status); and
  5. Identification of high-risk groups for disease and medical management.

Chronic care management firms or medical managers of managed care organizations are concerned with risk and risk adjustment because they frequently are compensated on how well they manage patients with certain diseases or characteristics that make them likely to incur high medical expenditures, be hospitalized, or use high-cost medical resources. To evaluate the clinical and financial performance of disease management firms, risk adjustment is necessary.

Health plans or providers sometimes use risk adjustment techniques to adjust payment to health plans or to providers. Accordingly, primary care physicians who are given economic incentives to control the medical costs of their patients— and who have a caseload of sicker-than-average patients—appropriately demand statistical adjustments to reflect their patients’ unfavorable health status.

Predictive modeling, a term that is closely related to risk adjustment, is the process of predicting future medical expenditures or resource utilization for individuals. Its purpose is to identify high-risk individuals for chronic care management programs and predict medical expenditures for rating or underwriting.

Risk adjustment and predictive modeling rely on claims data, which contain information on patient age, gender, diagnoses, prescriptions, observed utilization and treatments, and sometimes clinical data such as lab results.

There are some differences between predictive modeling and risk adjustment.

  • Predictive modeling is tied to medical management and interventions to change behavior; hence, predictive models may attempt to provide insight into a member's impactability.
  • Predictive models may use additional variables (such as prior utilization and compliance with prescription regimens) and plan design elements (such as level of copayments and deductibles) for prediction purposes.


The following example illustrates how predictive modeling is used in two patients: an 82-year-old man and a 72-year-old woman.

  Member ID
Risk Drivers 6238 3754
Age 82 72
Gender M F
Number of Office Visits 13 18
Co-morbidity: Diabetes 1 1
Co-morbidity: Hypertension (yes/no) 1 0
Co-morbidity: Obesity (yes/no) 1 0
Drug: Beta-blockers (yes/no) 0 0
Test: A1c (yes/no) 0 1
Previous Year's Costs $7,934 $6,074
Predicted Annual Costs $6,200 $5,850

In this example, the underlying structure or coefficients of the predictive model are not presented. But one can imagine the model would be applied on a 500,000-member health plan in which we predict annual health care expenditures for each person. If the predictive model has been properly validated, the previously cited applications are possible. In particular, if both patients were in a diabetes management program, the predicted expenditures for each patient would be useful as a starting point in evaluating program performance.

In general, risk adjustment and predictive models are proprietary and users pay a licensing fee for their use.


Cumming, R., Knutson, D., Cameron, B. and Derrick, B. A Comparative Analysis of Claims-Based Methods of Health Risk Assessment for Commercial Populations. Schaumburg, IL: Society of Actuaries (2002).

Iezzoni, L. Risk Adjustment for Measuring Health Care Outcome. 3rd ed. Chicago: Health Administration Press (2003).

Last, J. A Dictionary of Epidemiology. 4th ed. Oxford: Oxford University Press (2001).

Olenckno, W. Essential Epidemiology. Prospect Heights: Waveland Press, Inc. (2002).

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

Risk Groupers

A risk grouper is an algorithm for combining claims information, such as age/gender, International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes or Current Procedural Terminology, Fourth Edition (CPT-4) codes at the individual level to identify and group individuals on the basis of clinical homogeneity or similar amounts and types of resource utilization.

The use of a risk grouper can range from giving an overall assessment of risk for an individual (retrospective and/or prospective) to identifying specific conditions that the individual may have.

See also risk adjustment.