The latest concept in this field, disease management, refers to the use of an explicit systematic population-based approach to identify persons at risk, intervene with specific programs of care, and measure clinical and other outcomes. Disease management is not new; for centuries, clinicians have been identifying and treating patients using information gleaned from their training, the medical literature, and personal experience. What is new about the current model of disease management is that it does much more than evaluate known practices and suggest guidelines. It is a systematic information-driven approach in which clinical encounters are computerized, summarized, and shared; opportunities for intervention are not always detected solely at the bedside, but sometimes from a central data-base; practice guidelines may be implemented by computer screens in the practice setting; and initiatives are measured and compared in terms of health outcomes.

At the core of disease management are principles described as outcomes research, an evolving and somewhat embryonic field. Confusion has arisen over terminology and the application of principles to the management of patients. We identify core concepts and highlight some issues that become apparent as outcomes principles are applied to practice.

Measurement of Outcomes

An important aspect of disease management is the ability to measure and report health outcomes. The movement to collect this information has recently escalated, stimulated by federal initiatives to compare mortality among hospitals [6, 7], by comparative outcomes studies [8], and by payers seeking to improve quality and reduce costs [9]. Donabedian [10-12] described quality health care as an optimal triad of structure, process, and outcomes. Anderson and colleagues [13] extended this model by suggesting that relations among these factors should guide decision making. The outcome measures have evolved from simple dichotomous ones such as survival or occurrence of a clinical event to patient-oriented measures such as satisfaction, quality of life, and functional status.

The term "outcomes" has been linked to different measures, ranging from physiologic values to quality of life assessments. One approach to taxonomy is shown in Table 1. The reporting of outcomes may need to differ, depending on the priorities of those examining the data. A clinician is likely to be primarily interested in clinical or humanistic outcomes, whereas a health plan administrator may be interested in economic ones. Patients are interested in humanistic outcomes and have trouble interpreting clinical outcomes. As clinicians' financial risk increases, they are also developing greater interest in economic outcomes. Current disease management programs include mixtures of all measures.

Several points should be emphasized as these measures gain broader application. Selection of specific measures is currently a somewhat arbitrary process. For asthma, for example, several disease-specific measures of quality of life are available [14-17]. One way to distinguish among these measures is to examine the evidence behind the development of each measure: Has each one been validated? This is now a standardized process of item generation, item reduction, reproducibility, responsiveness, and validation [18, 19]; however, multiple questionnaires can meet these criteria and still leave the clinician confused.

Many outcomes measures were developed for use in population surveys or clinical trials and not for use in monitoring individual patients. The within-person variability can be great, but it is often offset with aggregate data; thus, in the context of disease management, some measures must be used with care. Even when measures are aggregated, statistical power may be insufficient to detect a significant improvement in outcomes. For example, in a modestly sized health plan with 25 000 members, one might estimate that 150 persons have diabetes and use insulin. Even if all 150 participated in a program that reduced the number of diabetes-related complications by 50% through improved diet, exercise, and appropriate use of insulin, the statistical power would be too low to document the program value compared with complication rates with a previous program. Thus, not all outcomes can be collected on all populations; if outcomes are measured, statistical issues need sufficient attention.

Finally, the interpretation of the change in outcomes is not yet clear for all measures. For example, what is the meaning of a 4-point change on a 100-point quality-of-life scale? Is a $0.50 increase in the per-member per-month charge enough to warrant a disease management program? Researchers attempting to define clinical significance are beginning to address these issues by anchoring degree of change to other events [20-24]. For example, Brook and colleagues [25] noted that a three-point difference on a standardized mental health scale was analogous to being laid off from a job. Considerable additional research into the meaning and application of outcomes is needed.

Effectiveness Studies

The distinction between effectiveness and efficacy is critical to disease management; treatments with proven efficacy do not always perform as well under conditions of typical clinical practice (effectiveness). In most efficacy studies, highly specialized practitioners treat selected patients on tight protocols in a setting in which both patients and practitioners have economic incentives to comply. In effectiveness studies, on the other hand, treatments can be viewed as they would be in usual settings Table 2 [26-28].

Most therapies require evaluation of efficacy before being implemented. This evaluation generally requires a priori hypotheses, randomization (to eliminate selection bias and confounding), homogeneous patients at high risk for the outcome, experienced investigators who follow a protocol, a comparative measure such as placebo (if ethical), and intensive follow-up to ensure compliance. Under these circumstances, if a treatment proves to be better than placebo (or a comparative measure), one can be reassured that the treatment can work.

However, questions may remain about the ability of the treatment to work adequately in a broader range of patients and in usual practice settings in which both patients and providers face natural barriers to care. These issues are of central importance in assessing relative cost-effectiveness in the community. To address these questions, most effectiveness studies have been done as observational studies, wherein observed groups are not randomly assigned to therapy and neither patients nor providers generally know that they are being studied (usually retrospective studies). This eliminates issues related to inclusion of special patients and providers and allows the treatment to be observed as it is typically used. Selection bias, which occurs when patients are not randomly assigned to alternative treatments, becomes potentially problematic, and adjustment for case mix and severity of illness becomes important.

In response to this problem, prospective effectiveness trials are beginning to emerge [29-31]. As in typical clinical trials, these trials include the development of a protocol with a priori hypotheses, recruitment of participants with informed consent, randomization to eliminate selection bias, and routine collection of data at prespecified time periods. Prospective effectiveness trials differ from typical clinical trials in that they enroll heterogeneous participants, use providers more similar to those who treat the disease, impose few protocol-driven interventions, and incorporate outcome measures relevant to the disease and delivery system. Such trials attempt to provide unbiased information in relation to selection factors while still addressing outcomes and effectiveness.

In disease management, clinical trials are not always possible. When a health plan has developed a new clinical program, members may not feel comfortable having an intervention withheld, and employers may be unwilling to conduct a trial before disseminating the program. However, if the program is implemented without a concurrent comparison group, what remains is a before-and-after comparison; within a rapidly changing health care environment, the aging of the population and other factors can confound results. This issue is an especially important methodologic challenge of program evaluation.

Health Economic Evaluation

The types of studies that are considered to be economic evaluations are shown in Table 3. Of these, cost-effectiveness and cost–utility analyses overlap with outcomes research because economic measures are frequently related to outcomes such as mortality, quality of life, or value assigned to years of life saved (or quality-adjusted life-years).

Suppose the intervention is designed for middle-aged persons with a disease that leads to death and only requires treatment for 3 years. If the therapy costs $3/d, the total cost of therapy per 100 persons is 100 x 3 x 365 x 3, or $328 500. If the treatment does not reduce any other component of health care, one way of viewing the program would be in terms of its net expenditure of $328 500 per 100 persons, or net use of plan revenue.

However, a cost-effectiveness argument could incorporate a noneconomic outcome such as mortality offset and be represented as the cost per year of life saved. For example, if mortality decreased from 6% to 3% over 3 years and if the average life expectancy at this age was 25 years, the number of life-years saved per 100 persons would be (0.06 – 0.03) x 100 x 25, or 75 per 100 persons. Because the net cost of therapy was $328 500, the cost would be $328 500 per 75 life-years saved or $4380 per year of life saved; this is highly cost-effective compared with other life-saving interventions [32].

The other health economic method that includes outcomes is cost–utility assessment, which is similar to cost-effectiveness but values each year of life saved as something other than 1 equals alive and 0 equals dead. In the above example, if other events (such as myocardial infarctions) are avoided with therapy and if we assume that life after each event is worth half a healthy year of life, we would add the 75 years of life to the half-years of "quality-adjusted" life to calculate the total number of quality-adjusted years. The numerator would not change, but a cost–utility ratio would be computed, reflecting the cost per number of quality-adjusted life years.

Selection of methods depends on the research question and the perspective of the target audience; this approach can lead to disparate answers. If a health care benefits manager notices a $328 500 expenditure, the fact that this amounts to a high ratio of cost-effectiveness would not necessarily be compelling enough to warrant widespread adoption of a treatment that leads to unplanned spending. On the other hand, practitioners familiar with economic evaluation may be pleased to see so low a ratio and insist on the therapy. Patients may demand treatment if they do not pay for it and understand only the difference in mortality. Thus, the approach to economic evaluation of disease management varies widely with the stakeholder.

Another important issue is inference. Many economic evaluations rely on inferences being made, because information on long-term cost and potential outcomes is not always readily available. In these instances, inferences must be made on expected event rates; even with sensitivity analyses and disclosure of all assumptions, evaluations are only as good as the models.

Outcomes Management

Outcomes management assumes that by systematically measuring outcomes and reviewing the treatment that preceded the outcomes, optimal therapy can be determined [33, 34]. This process should lead to management recommendations, followed by reevaluation of outcomes and a continuous opportunity for improved delivery of care. This is a dynamic management paradigm; it is not the same as outcomes research, which seeks to define the range of outcomes produced by alternative interventions. Outcomes management seeks to produce desirable outcomes in a clinical setting and is therefore the application of outcomes research to practice.

Outcomes management requires an infrastructure in which population-based outcomes can be readily assessed. First, a common set of outcomes measures should be endorsed and used; this is not current practice. In some cases, numerous alternative outcomes measures exist for the same disease, with no overseeing body to endorse one or another. Second, uniform collection and encoding of outcomes data in the population are necessary; this process can be difficult when care is rendered across numerous providers who are not necessarily linked through a shared (that is, computerized) information system. Finally, to effect a change, these data must be linked to information on the process and structure of health care previously delivered.

The limited training and orientation of health care professionals to this type of analysis and program development adds to the difficulty in implementing outcomes management. Although simple in theory, incorporation of outcomes management principles into medical practice requires that a complex, expensive set of processes become seamlessly integrated into health care delivery. It also requires uniform collection of data and transmission and, ultimately, consent from patients to provide the data.

Disease Management

Disease management incorporates outcomes research technology into outcomes measurement and management programs. No clear distinction can be made between outcomes management and disease management. The only distinctions may be technical; disease management does not necessarily require outcomes assessment, but outcomes management requires outcomes assessment before implementation.

The basic premise behind disease management is that there is a more optimal way to manage patients, which results in lowered costs and improved health outcomes. This premise is predicated on three basic assumptions: 1) Medical practice varies [if not, why impose "best practices" or optimal care?]; 2) variation is related to different outcomes [if not, why bother to intervene?]; and 3) it is possible to develop and implement a system of care that improves outcomes.

Sufficient research supports the first two points, but there is limited evidence that disease management systems can improve health outcomes. This is due, in part, to the recent emergence of the concept; few systems have been in place long enough to measure outcomes. Additionally, a successful disease management program is generally based on information acquired by comparing the program with other systems of care. Providers are currently implementing community-wide programs without the benefit of such comparative data.

Disease management involves a shift from the classic paradigm of an individual physician providing comprehensive, continuous, and affordable health care to patients as they present in the clinical setting to a population-based systematic approach that identifies persons at risk, intervenes, measures the outcomes, and provides continuous quality improvement. At the core is the knowledge base on which to determine how best to identify patients, intervene, and measure outcomes. The information that drives this process involves a shift from consensusbased medicine to evidence-based medicine [35-37], the emerging science of integrating information from credible clinical trials to make decisions.

Comprehensive disease management requires a deep understanding of the natural history of disease to determine where in the life cycle of disease an intervention should be implemented. Because most chronic diseases have a long natural history, the same intervention (for example, lipid lowering to benefit atherosclerosis) could be implemented for patients who have preexisting disease (secondary prevention) or for those who do not (primary prevention). The economic consequences and time frame will differ because of the frequency of incident events or ability of the intervention to work. This level of detail helps to frame the program in terms of the characteristics of persons being sought for enrollment, expected duration and type of intervention, and selection of relevant outcomes. One must have information on the optimal method for managing disease in order to implement a successful disease management program. Some of the important issues about implementation include an under-standing of the inefficiencies in health care delivery (benefits design as it drives delivery), disincentives for the patient or provider to receive or deliver the highest quality care (such as access and cost issues), relative cost-effectiveness of alternative treatments, and the success of different interventions in modifying behavior (such as compliance).

For planning purposes, the intervention itself can be encapsulated in the form of clinical protocols or pathways (based on guidelines) that must be disseminated and applied. Many efforts are currently available to deliver these messages at the point of care, but we have much to learn about the best ways to implement such programs and change provider and patient behavior.

Although overall disease management programs have not yet been evaluated, components of such programs have been studied (Table 4). For example, recent research has shown that automated telephone reminder systems can increase clinic attendance by nearly 35% [38] and that the use of home blood pressure monitoring can reduce physician office visits by 44% [39].

The old paradigm for the management of disease was a somewhat fragmented system involving the treatment of disease by individual practitioners. Treatment was often based on anecdotal or consensus information in a fee-for-service environment in which manual systems of record keeping and information storage were used. The new paradigm is population-based risk and disease assessment, systems of disease prevention and health promotion, community-based intervention and provider contacts within a framework of automated information, evidence-based medicine, and defined protocols of care, with explicit collection of outcomes information. Figure 1 shows the spectrum of disease management concepts reviewed in our paper and the manner in which they relate to each other.

View larger version (24K): [in this window] [in a new window]   Figure 1. The multidisciplinary scientific basis of outcomes research and the manner in which it flows into disease management programs. Continuous quality improvement of outcomes is the ultimate goal.

In the future, patient health will be improved by maximizing functionality; minimizing disease, disability, and death; and improving the efficiency and cost-effectiveness of health care. This improvement involves the use of effective outcomes tools, including linked databases and collaborative research efforts. It is also increasingly apparent that health care providers will need to be hands-on computer users when clinical decision making occurs with data on individual patients. Although the field of outcomes assessment and research has not been widely understood, its application to the practice of medicine through disease management has already begun.

Dr. Sherwood: Merck & Co., PO Box 4, Sumneytown Pike, West Point, PA 19486-0004.