Measuring and reporting on patient satisfaction with health care has become a major industry. The number of medlinearticles featuring “patient satisfaction” as a key word has increased more than 10-fold over the past two decades, from 761 in the period 1975 through 1979 to 8,505 in 1993 through 1997. Patient satisfaction measures have been incorporated into reports of hospital and health plan quality,1, 2 and armies of consultants make a good living selling software packages to health care providers eager to assess their customers' reactions by telephone, fax, and modem. Unless my own academic hospital is aberrant, reams of patient satisfaction reports sit on the desk of every health care administrator in America. Amidst this flurry of activity, three fundamental questions emerge: Is patient satisfaction worth measuring? How can it best be measured? And how are we to use the results? These three questions—one philosophical, one empirical, and one practical—form a framework for evaluating the place of patient satisfaction in the patient outcomes movement as a whole.
The answer to the first question depends on who you ask. On one side of the aisle is an uneasy alliance of consumer advocates, marketing specialists, and proponents of patient-centered care. On the other side are skeptics who believe that focusing on patient satisfaction diverts attention from what ought to be our principal concerns in an era of resource constraints: inappropriate care; underuse of necessary care; and clinical outcomes such as morbidity, mortality, and health status. These critics have a point. Compared with measures of technical quality (e.g., appropriateness criteria or adjusted outcomes models), data on patient satisfaction are easy to collect, and many health care organizations have succumbed to the temptation to stop there. Nevertheless, helping patients achieve their goals is a fundamental aim of medicine. Because patients' goals and values vary widely, are not predictable on the basis of demographic and disease factors alone, and are subject to change, the only way to determine what patients want and whether their needs are being met is to ask them. From this perspective, viewing care “through the patient's eyes” is an ethical and professional imperative.3 Individual clinicians, medical groups, hospitals, and health plans all have reason to be interested in patient satisfaction, and not only because satisfied customers add to the bottom line. Indeed, arguments over the place of patient ratings and reports in the catalog of health care outcomes usually turn not on whether measuring patient satisfaction is important, but on whether satisfaction can be measured reproducibly and meaningfully.
If patient satisfaction is to take its place alongside morbidity, mortality, and functional status, several critical measurement issues must be addressed. First, scale developers and end-users need to be clear about what they are measuring. “Patient satisfaction” is not a unitary concept but rather a distillation of perceptions and values. Perceptions are patients' beliefs about occurrences. They reflect what happened. Values are the weights patients apply to those occurrences. They reflect the degree to which patients consider specific occurrences to be desirable, expected, or necessary.
Most contemporary measures of patient satisfaction employ hybrid questions that assess perceptions and values simultaneously. An example is, “How satisfied were you with the amount of time the doctor spent with you today? (extremely, . . . not at all)?” In responding, patients must first estimate the amount of time they spent with the doctor, compare it with an internal standard, and then provide an overall judgment. Such hybrid questions have the virtue of linguistic economy but make it difficult to distinguish perceptions from values. Given these semantic vagaries, a patient who receives poor care but has low standards may report the same satisfaction as a patient who receives good care but whose standards are unreasonably high. In the ambulatory instrument developed by the Picker Institute (Boston, Mass.), patients are not asked about “satisfaction with communication,” but rather, “Did the provider explain what to do if problems or symptoms continued, got worse, or came back?” Responses to questions of this type are not readily summed or averaged, and Cronbach coefficients for the data are often low. Nevertheless, what is lost in scalability is gained in interpretability. If I were told that my patients' adjusted satisfaction score was a full standard deviation below the mean for all practitioners at my clinic, I'd be upset, but I wouldn't know what to do about it, and I probably wouldn't change how I practiced. On the other hand, if I learned that 40% of my patients didn't know what to do if their symptoms returned, I might give my approach to providing follow-up instructions some scrutiny.
Despite the advantages of disaggregating patient satisfaction into its component parts, most research studies have treated satisfaction as a “black box” that predicts certain outcomes (e.g., plan disenrollment) and is in turn predicted by certain antecedents (e.g., practice size). Opening the black box can reveal new relationships.4 In this issue of the Journal, Zemencuk et al. report on a survey of 652 patients and 105 physicians in four primary care sites in Michigan and Ontario.5 The survey asked separately about patients' desires (what they wanted) and expectations (what they thought would occur over the near term). Although there were no cross-national differences in patient desires, American respondents were significantly more likely than their Canadian counterparts to say they “expected” mammography; Pap, prostate-specific antigen, and cholesterol testing; and breast and rectal examinations. It remains for larger, more generalizable studies to explore whether these differences reflect general cultural factors or priming by experience. However, by creating definitional boundaries between patient desires and patient expectations, the authors discovered patterns in the data that would otherwise have been lost.
A second measurement issue concerns the relation between patient satisfaction, process of care, and health outcomes. The article by Covinsky et al. (also in this issue of the Journal)6 attacks this complicated problem head on. It has been thought for some time that patient satisfaction with medical care reflects satisfaction both with how care is delivered (process) and with the results of that care (outcome). For example, a patient undergoing knee surgery may be highly satisfied with the attentiveness of the surgical staff but disappointed with his rate of recovery, or vice versa. However, several studies suggest that health status per se (rather than degree of improvement in health status) also influences satisfaction.7 A study estimating the relative impact of health status as an intrinsic characteristic (like age) and health status as a result of care requires a longitudinal design. Using data from 445 elderly patients discharged from University Hospitals of Cleveland, Covinsky and colleagues found that short-term improvement in health status were not associated with higher satisfaction after controlling for health status at discharge. They concluded that the association between patient satisfaction and health status reported in many other studies represents a tendency for healthier patients to report greater satisfaction, rather than a tendency for patients whose health has improved due to medical care to report greater satisfaction.
This is a fundamentally sound study and an important result, but several limitations deserve comment. For one thing, the effect of health status change on satisfaction was estimated only crudely: confidence limits were wide, and the possibility of a large negative effect of health improvement on satisfaction could not be excluded (see Table 4 of their article). In addition, the single-item measure of health status used in this study has uncertain reliability and validity in this complex inpatient population. It is difficult to understand, for example, what the 32% of patients with “good” or “excellent” health at admission were doing in the hospital to begin with! Finally, there was no accounting for length of stay as a potential confounder of both health status and satisfaction at discharge. Nevertheless, this study will have significant impact if it stimulates other researchers to use similarly strong longitudinal designs to study key linkages between health care process, outcomes, and satisfaction.
A third methodologic question is whether to adjust for patient characteristics that influence satisfaction when comparing health care providers. The answer has practical consequences, as managed care organizations increasingly turn to patient satisfaction scores as a criterion for bonus payments. Proponents of adjustment argue for a leveling of the playing field, whereas critics point out that adjusting away a variable dooms it to irrelevancy.8 Whether to adjust or not really comes down to the purpose of the measurement. If one is comparing two different health care providers or systems in the context of a report card, fairness demands appropriate adjustment. If the aim is quality improvement, adjustment obscures what may be important problems and stifles creative solutions aimed at patient subgroups.
So now we get to the real issue, which is how to use patient satisfaction results. Many satisfaction batteries can reliably distinguish between physicians who are great communicators and those who are interpersonally challenged. Patient satisfaction is also related to a variety of downstream outcomes, such as the propensity to change health plans,9 or to sue for malpractice.10 These results are clearly of interest to managers and marketers, but their relation to clinical quality improvement is tenuous. The important question is whether information on patient perceptions and values can stimulate genuine gains in patient-centered care. Providing physicians with comparative quarterly satisfaction reports is likely to accomplish little except to fuel resentment. Even if such reports did motivate improvement among the low outliers, they would still do nothing for the average physician living within the fat part of the bell curve, where the greatest opportunity for collective improvement lies. Separating patient perceptions from patient values and using questions that focus on potentially mutable behaviors—of persons and of organizations—would help. So would increased attention to narrative: patients' detailed reports of what went well and what didn't. In the meantime, if patient satisfaction measurement is not to be dismissed as one more health care fad, many challenges—philosophical, empirical, and practical—must still be addressed.—Richard Kravitz, MD,Center for Health Services Research in Primary Care, University of California, Davis.