Measuring the various dimensions of health is usually accomplished by evaluating multiple domains that represent the full spectrum of life functioning, involving physical, psychological, and social aspects. From an analytical perspective, quality-of-life measures have been used to describe a condition or state of health, provide a prognosis, establish a reference norm, or signal a change in patient functioning. The large variety of instruments and evaluation tools available for assessing quality of life in people with diabetes has been generated to meet a number of different research objectives. As such, the instruments often seem disparate, heterogeneous, and many times contradictory in their approach.

Most quality-of-life instruments are developed for a particular purpose. Some quality-of-life measures focus on describing the perceived state of health of the individual in order to understand the patient's needs, desires, preferences, and expectations so that suitable medical and support services can be provided.^2-5 Other evaluations focus on learning more about external or internal determinants of quality of life, such as socioeconomic status, gender, coping, and social support.^6-10 In addition, health-related quality-of-life assessment has gained recognition as an important research tool for evaluating the impact of new medical treatments and health care services for people with diabetes.^11-13

In the study of diabetes, the earlier quality-of-life literature focused on describing the state of health of individuals with complications and serious health conditions, particularly patients undergoing renal dialysis or kidney transplantation or those suffering from blindness or foot amputations.^14-17 Furthermore, studies of the psychological impact of diabetes and its effects on patient quality of life were considered important for understanding the patient's ability to adhere to often difficult and demanding treatment regimens.¹⁸ While earlier treatment options were relatively limited, more recent technological and pharmacological advances, such as the insulin pump and newer classes of oral hypoglycemic agents, have generated a renewed interest in evaluating quality-of-life outcomes as part of choosing the optimal therapeutic regimen.

Rubin and Peyrot have given a thorough review of the conceptual and practical issues for assessing quality of life in diabetes.¹⁹ These researchers described three categories of determinants influencing quality of life, including medical predictors (e.g., type and duration of diabetes, treatment regimen, level of glycemic control, and presence of complications); attitudinal predictors (e.g., self-efficacy, locus of control, and social support); and demographic predictors (e.g., gender, education, ethnicity, age, and marital status). When reviewing the results of quality-of-life diabetes research, it is essential that readers be made aware of the limitations and strengths of the measures and analyses used within the context of the original objectives of the research.

In this regard, Kirshner and Guyatt recognized that the validities of a quality-of-life scale have different implications across the three types of quantitative analyses for which they might be used, namely, discriminant, predictive, or evaluative.²⁰ While discriminative indices should demonstrate large and stable between-subject variation so that groupings of similar patients can be contrasted to other groupings, evaluative indices must possess high test-retest reliability and responsiveness to true changes in quality of life. Although not used quite as frequently, analyses undertaken for the purpose of prediction must also ensure that the quality-of-life scales are highly correlated with the primary endpoint of interest.

Much of the current debate surrounding the interpretation and meaning of quality-of-life data in the study of diabetes is due to applying the same performance standards appropriate for discriminant measures to evaluative or predictive measures, for which such standards are inappropriate. For example, if the purpose of evaluation is to determine which aspects of changes in quality of life influence adherence to a diabetes treatment regimen, the technique for assessment would be very different from that required for describing the level of patient functioning necessary for conducting a nursing needs assessment. Similarly, a new side effect, such as gastric upset, which might be considered a relatively small part of overall functioning in a descriptive analysis of diabetic complications, could have a dramatic impact on day-to-day quality of life and adherence to medication when contrasting two therapies as part of an evaluative analysis. The basic rule is that quality of life is not a static measure, but rather is an approach to health assessment that focuses on patient reports, feelings, and expectations.

Quality-of-life measures must meet the performance standards corresponding to the purpose of the analysis. Examples are given in Table 1. To claim that a scale has been "validated" is meaningless unless a reference standard reflecting the purpose of the analysis is provided. An evaluative analysis must employ quality-of-life scales that are responsive to changes and differences in the underlying quality-of-life construct and scaled in such a way that meaningful changes and differences can be detected. The degree of difference or change that is considered meaningful will vary depending on the prior hypothesis, the type of quantitative analysis to be undertaken, and the intent or perspective of the investigator.

It is apparent that studying the complex interaction between the effects of diabetes treatment programs and regimens, quality of life, and patient compliance requires rigorous measurement and evaluation. First, one must expand the set of measurements used in clinical research to include patient reports of health-related quality of life. Health care professionals are schooled in the biomedical measures of therapeutic research. These include physiological measures such as blood pressure, exercise tolerance, heart rate, and pulmonary function, and clinical measures such as signs and symptoms of disease, morbidity, illness, infection, tumor, and mortality. However, while health care professionals are concerned about health status and perceived quality of life, the quantification of these measurements is not as familiar.

One can measure health status in terms of objective levels of symptoms, activities, function, emotion, cognition, and an individual's ability to perform his/her job or role in society. However, patient perceptions concerning illness and treatment, including levels of worry, distress, well-being, satisfaction, and expectations can alter health perceptions at the same level of health status. The measurement of health-related quality of life can be defined as the level of health status filtered by individual patient perceptions. It is this bidimensionality of health-related quality-of-life measurement that allows researchers to understand the forces that shape patient behavior and the ability to adhere to diabetes treatment regimens.

Another area in which quality of life can play an important role in determining treatment effectiveness is by helping to answer the question of whether diabetes specialty care results in better overall health outcomes as compared to nonspecialty care. Diabetes specialty care may include diabetes nurse educators, endocrinologists trained in diabetes, dietitians, and psychologists and psychiatrists trained in the psychological and emotional consequences of diabetes and its treatment demands. A recent randomized clinical trial comparing generalist care to specialty care using a diabetes-specific disease management program demonstrated that while improved glycemic control mediated quality-of-life changes primarily through reductions in physical and psychological distress, the specialty-based diabetes management program provided an additional independent quality-of-life benefit through improved emotional and psychological well-being.²²

For people with type 1 or type 2 diabetes requiring insulin, the potential negative impact of daily insulin injection on the quality of life of the patient must also be considered. In addition to devices such as the insulin pump and the insulin pen, less invasive alternatives to insulin administration are currently under development. A new experimental device permits insulin to be inhaled rather than injected. The primary motivation behind this development was to improve the quality-of-life restrictions relating to treatment administration and timing by eliminating syringe injections and allowing for greater flexibility in meal planning and exercise. It is clear that the ultimate decision to reimburse for inhaled insulin will be made primarily on the basis of improved quality of life and patient satisfaction rather than on efficacy alone.

Determining treatment effectiveness using quality-of-life outcomes requires evaluating the magnitude and meaning of quality-of-life changes resulting from new treatment interventions. Such evaluation raises a major question for practicing clinicians, namely, what is the meaning of such changes and how are the quality-of-life results interpreted and acted on in clinical practice? A major methodological issue addressed in the remainder of this review is the interpretation of quality-of-life treatment effects.

Interpretation is challenging for a number of reasons, including the subjective nature of the quality-of-life construct, the indirect way in which it is assessed, and the lack of clearly directed therapeutic goals defined in terms of quality-of-life changes. The first important task necessary for interpretation is defining the operative range of the quality-of-life scale and the metric that delineates a meaningful clinical effect for the individual. The second deals with evaluating the population impact of a treatment effect so that it has relevance for clinical practice. Both of these issues are discussed below.

Ordered classifications of relative importance—such as substantial worsening, moderate worsening, mild worsening, no change, mild improvement, moderate improvement, and substantial improvement—are preferable over simple dichotomous ones. Decisions should be linked to these outcome categories. For example, if quality of life improved substantially following laser treatment for patients with retinopathy, the procedure more than likely would be recommended even if the financial costs were high. If visual improvement were marginal, resulting in no or little change in quality of life, the recommendation might be quite different. If the side effects of the procedure substantially worsened quality of life, laser treatment might be discontinued altogether. The search for clinical meaning arises because the original metric of the quality-of-life scale does not translate into something meaningful to the practitioner. If categories or cutoffs for quality-of-life scales establishing when to modify a therapeutic regimen were generally available, quality-of-life outcome data would be much more useful to health care providers.

When the quality-of-life metric is unfamiliar to the end user, an effort is usually made to relate the metric to another measure that is familiar. Brook and associates reported extensively on the responsiveness of measures used in the Health Insurance Study (HIS) conducted by the Rand Corporation in the late 1970s and early 1980s.²³ The HIS study was powered to detect changes in quality-of-life measures equivalent to 5 years of aging and 17 Life Change Units (LCUs) on the Holmes-Rahe stressful life events scale. Brook and colleagues decided that effect sizes of "5 years of aging" and "17 LCUs" were considered important enough not to be missed. By doing so, they defined the MCIs for these reference measures. By associating more objective changes, namely, aging and the occurrence of stressful life events, with measures of quality of life, the investigators also established a method for creating MCI ranges for the quality-of-life scales. Individuals who question the relevance of the association have criticized the use of negative life events. However, when calibrating quality-of-life scores, it is not the context of the association that is of primary interest, but the degree and strength to which changes in the reference measures correspond to changes in the quality-of-life measures.

Quality-of-life scales are said to be responsive if they change in response to the intervention of stressful life events. A quality-of-life scale that remains stable subsequent to the occurrence of negative events, such as loss of a job, divorce, or death of a spouse, might be called into question as to whether it was truly measuring what it claimed to measure. It is evident that a quality-of-life scale could be valid with respect to changes in stressful life events or differences among subgroups of patients with varying diabetic complications, but not with respect to other criteria, such as intensity of the diabetic treatment regimen.

The determination of clinical decision breakpoints will depend on the decision maker's overall perspective. The ranges of the scales that are appropriate will vary depending on whether the decision maker is interested in answering questions concerning differences among populations or changes within individuals. Absolute ranges are best utilized to reflect differences across populations with varying diseases, while relative ranges are most useful for within-disease state comparisons. Operative ranges are most appropriate for detecting longitudinal changes that are important to the individual.

Often, the perspectives of payers, providers, and patients coincide with the absolute, relative, and operative ranges as illustrated in Figure 1. The perspective used to develop MCIs might vary depending on the nature of the day-to-day decisions and choices that must be made. Caution should be taken not to use results obtained from the absolute range perspective when the question deals with changes important to the individual requiring an operative range perspective.

FIgure 1. The changing range of interest of a quality-of-life scale depending on perspective — payer, provider, or patient.

If the scale's metric is not sensitive enough to detect the breakpoints where decisions must be made, then the scale will not be valid with respect to the events associated with those decisions. For example, say that one has the choice of using either a combination of oral hypoglycemic agents or injected insulin therapy. There are efficacy effects and side effects associated with either choice. It would be useful to know how large a difference in quality of life would be required before a decision would be made in favor of one treatment regimen over another. The quality-of-life MCI breakpoint would correspond to the difference required before such a choice could be made. Practically, the difference would have to be large enough to have a functional or behavioral impact on the patient. Symptom distress due to side effects of oral medications might impair the patient's sense of well-being to a point where he or she is not able to function as expected. Similarly, an intensive insulin regimen requiring strict meal planning, daily insulin injections, and more frequent blood glucose monitoring might also cause functional restrictions. The quality-of-life metric must be sensitive enough to detect those changes in function important to the patient. If an absolute range and perspective is used, the important differences might be missed.

Health care payers deal with decisions involving the absolute range of quality of life—from individuals hospitalized with severe illnesses to the healthiest individuals, who require little or no services. In their decision making, payers must balance re-sources across a very broad and heterogeneous population. To the payer, the difference in quality of life that is important is often viewed within the context of large variability and differences in the populations.

Health care providers see a much more homogeneous group. When providing care for patients with diabetes, the majority of patients will fall into a similar range of functioning according to the severity of their disease and their co-morbid conditions. The physician is much more likely to make treatment choices and decisions based on the differences that appear important across a more restricted relative range.

On the other hand, most individuals live their lives within a much smaller operative range. The operative range is defined by the lowest level at which the individual can function as usual. In a study of operative ranges for 290 individuals with mild type 2 diabetes, patients rated their current health state at 83 (on a 0­100 health states questionnaire).²⁴ This level corresponded to individuals with minimal symptoms and no complications whose diabetes could be controlled by monotherapy on an oral hypoglycemic agent. To the question "What is the lowest level at which it would be possible for you to live and work as you currently do?" patients stated that the lowest level would be 56. This level corresponded to the health state "Diabetes controlled with diet and oral medication or insulin and early complications and moderate symptoms."

For these patients, a 26-point drop on the absolute scale would essentially correspond to "functional death" on the operative scale. Another way to consider this is that the operative range spanned only 25% of the absolute range. A meaningful drop on the operative scale for an individual would be something much less than 26—closer to a 20% loss of operative functioning, or approximately 5% on the absolute range.

It is no wonder that managed health care reimbursement decisions often seem so out of touch with what is important to patients. From one-night hospital stays for new mothers and infants to restrictions on the number of pills allowed for the treatment of sexual dysfunction, varying perspectives often conflict. The only way to determine meaningful changes for individuals is by surveying patients directly.

Another way to aid in the interpretation of the meaning of quality-of-life changes is to describe the strength of the linkage between physiological measures, clinical symptom measures, and patient self-reported measures of health status and quality of life. For example, in a study on the treatment of type 2 diabetes by Testa and Simonson,¹⁷ associations between these measures were determined using data from a double-blind, comparative clinical trial comparing diet and exercise to diet and exercise in combination with an oral hypoglycemic agent, glipizide gastrointestinal therapeutic system (GITS). After a baseline washout period, measures of glycemic control, symptom distress, and quality of life were recorded over the 15-week study period. This study found that a 1.8% decrease in HbA_1c led to substantial improvements in overall symptom distress, general health perceptions (fatigue, vitality, sleep disturbance, feelings of general illness), and cognitive symptoms (ability to reason, concentrate, remember).

Because clinicians understand the clinical meaning and relevance of a 1.8% decrease in HbA_1c, linking biomedical and symptom changes to less familiar patient self-reports of health status and quality of life aids in understanding the relative magnitude of a quality-of-life change. The ranges that are meaningful to patients have been shown to be consistent across several clinical trials of other conditions, such as hypertension.^25-27

The interpretation question can also be addressed at another level—that is, What is a clinically important treatment effect with regard to changes in quality of life for a population of patients? This question is analogous to the one asked in epidemiological studies of risk or health economic studies of cost-effectiveness and applies not to meaningful effects for an individual patient, but rather to recommendations summarizing the results of clinical trials in terms of population risk/benefit effects. This question assumes that individual breakpoints corresponding to the MCIs already have been established and that the treatment effects within a population of individuals must be explained and understood.

In epidemiology, the relative risk serves as the standard risk/benefit measure. The relative risk is defined as the incidence of an adverse event, such as myocardial infarction (MI), when exposed to a risk factor divided by the incidence when not exposed. Hence, a relative risk of 2.0 indicates that the probability of the adverse event is twice as likely with the risk factor as it is without. When determining a relative risk, a judgment is made as to what constitutes an adverse event—or an event adverse enough to be considered meaningful. The change in health status from "no infarction" to "infarction" defines an MCI. This dichotomy is clearly understood, and clinicians would uniformly agree on its relevance for treatment decisions and practices.

When comparing two treatments, the relative importance of a population difference between a relative risk for MI of 1.1 versus 1.2 is less intuitive. We know that the categories of MI and no MI are meaningful, but is a 10% difference in risk important enough to suggest a change in lifestyle or therapy assuming that there is some additional human or financial cost? Even if quality of life were measured as a simple dichotomous outcome such as "good quality" versus "bad quality" for which all third parties could agree, researchers would still be faced with the risk/benefit assessment.

The reason why the relative risk is so popular for assessing risk/benefit is that the dichotomous health outcome scale forces one to decide on MCIs. Translating any nondichotomous health outcome, whether health status, quality of life, vision or hearing impairment, cardiac functioning, HbA_1c levels, or pulmonary function, requires discrete categorization to correspond to the categorical nature of medical decisions, judgments, and choices. Most health outcomes in clinical research are either dichotomous or grouped into a few ordinal categories by employing the terms mild, moderate, or severe. Categorization is used to translate the distribution of responses of a population into easily comprehensible categories so that comparisons between treatments are relevant to clinical practice.

Quality-of-life studies in diabetes have focused primarily on describing the states of health of individuals with varying levels of symptoms and complications. However, with an ever-increasing selection of diabetic therapies, the patient's short-term and long-term quality of life depends upon the physician's understanding of the complex relationship between human behavior, quality of life, adherence, efficacy, and outcome.

The ability of each physician to use that knowledge to find optimal diabetes therapy for individual patients will depend on his or her ability to understand the meaning of quality-of-life changes. Therapeutic regimens and programs that maximize patient acceptance will result in the greatest health gains in the long run. Studies of treatment effectiveness that use quality-of-life outcomes will provide the most useful data to determine the relationship between the therapeutic regimen and patient acceptance.

The quality-of-life tools and instruments appropriate for descriptive analyses are not appropriate for the types of evaluative analyses required for treatment effectiveness research. Such studies need to be designed to evaluate longitudinal changes due to therapeutic interventions using operative ranges and diabetes- and therapy-specific measures. Evaluation of quality-of-life treatment differences using generic instruments or indices intentionally or unintentionally evades the evaluative research question by not providing the breakpoints that contain the relevant ranges of responses.

The danger for health care policy decisions is evident. If a health care provider seeks to reduce the direct costs of medical care by phasing out a diabetes nurse educator or diabetes management program, justification could be provided by "validated" quality-of-life measures that showed "no treatment differences." The fact that those measures were never developed or validated with the intention of detecting clinically meaningful differences across the operative range of individual functioning that could dramatically affect outcomes such as well-being and adherence might never be considered. There are hundreds of examples in the literature of the inappropriate use of quality-of-life measures that failed to demonstrate a treatment effect because of the inadequate measurement properties of the survey instrument relative to the evaluative aspects of the analyses.

Before conducting quality-of-life evaluations, the purpose of the research must be clearly outlined. The degree of specificity and weighting of domains with regard to the quality-of-life measurement needed to best answer the principal research objectives must be clearly understood. Finally, the scaling procedures used to define the metric of the quality-of-life measure must be relevant to the proposed research question.

It is clear that further research needs to be conducted on methods that maximize the item and scale responses in order to reveal the associations between diabetes treatment interventions and quality-of-life changes. The issues of measurement, design, and analysis are critical to translating research findings to clinical practice. However, probably the greatest contribution of the measurement and application of quality-of-life evaluation to the study of diabetes is what we have learned from the process, namely that human perceptions, expectations, and behavior are controlled by influences extending far beyond clinical medicine and management.