Nonmedical influences on medical decision making: an experimental technique using videotapes, factorial design, and survey sampling

Health Services Research, August, 1997 by Henry A. Feldman, John B. McKinlay, Deborah A. Potter, Karen M. Freund, Risa B. Burns, Mark A. Moskowitz, Linda E. Kasten

Numerous reports suggest that medical decision making - ideally a matter of symptoms, tests, and probabilities - is in fact a social transaction prone to medically extraneous influences. These nonmedical factors include personal characteristics of both patient and physician, as well as organizational characteristics of the setting where healthcare is delivered (Clark, Potter, and McKinlay 1991; Haug and Ory 1987). For example, aggressive treatment of breast cancer is reported to be less likely when the patient is over 75 years of age (Silliman et al. 1989; Chu, Diehr, Feigl, et al. 1987; Greenfield et al. 1987), despite comparable survival rates and tolerance of chemotherapy (Yancik, Ries, and Yates 1989; Early Breast Cancer Trial Collaborative Group 1988). Aggressive treatment is also reportedly less likely when the physician is older (Belanger, Moore, and Tannock 1991).

Demonstrating such influences objectively is a difficult matter, but an important one for the field of medical decision making. This is particularly true in areas of medicine where standard practice is in flux because of new biomedical or technical developments. With some diseases, such as AIDS or breast cancer, treatment choices have acquired a quasi-political flavor, owing to the emergence of an active patient advocacy movement. The rapid evolution of new organizational contexts for medical care raises additional questions concerning the incentives, constraints, and barriers that actually operate in the doctor-patient transaction.

The difficulties of doing useful research in this area can be summarized under two general headings: (1) the multiplicity of variables and (2) the shortcomings of observational studies.

The first problem arises simply because many nonmedical influences demand attention, injecting a large and potentially unmanageable set of independent variables into any analysis plan. Moreover, many of these variables are vague in nature. On the patient's side, these include socioeconomic status, insurance coverage, physical attractiveness, assertiveness, and medical knowledge. On the physician's side, the factors include experience; specialty; practice setting; and personal attitudes, beliefs, or concerns. On both sides age, sex, and race come into play, as well as numerous intangible factors and forms of nonverbal behavior. Scoring any but a few of these variables meaningfully in an observational study, or controlling them systematically in an experimental study, is certain to present serious problems. Gathering enough cases to analyze the independent and joint influences of such a large ensemble of predictors with adequate statistical power is likewise very difficult. Nevertheless, the omission of any of these factors can stand out as a glaring oversimplification in any particular area of medical decision making.

The second major impediment to definitive research in nonmedical influences is that most of the work in this field has necessarily been observational rather than experimental. Short of a full-scale clinical trial, it is simply not practical or ethical to intervene in the presentation of patients to physicians or in the formation of diagnostic and treatment decisions. Abstracts of faits accomplis are therefore the predominant source of data on medical decision making, at least where actual patients are involved. In the field of breast cancer, for example, Chu et al. (1987) drew data from community hospitals, while Ayanian et al. (1993) combined data from the New Jersey State Cancer registry, state-mandated hospital discharge abstracts, and the U.S. Census. Similarly Samet et al. (1986) analyzed the New Mexico tumor registry, and Yancik, Ries, and Yates (1989) utilized the Surveillance, Epidemiology, and End Results (SEER) database of the National Cancer Institute. In large observational studies, the physicians' decisions cannot be examined in detail. In smaller studies, even though the circumstances of a medical decision can be more thoroughly recorded, those circumstances are not under the researchers' control. Any characterizations of patient, physician, and setting are therefore necessarily approximate. Whether large or small, all observational studies are subject to the possibility of confounding, whereby such variables as race and socioeconomic status - no matter how well defined and recorded - tend to be strongly correlated and therefore largely inseparable in interpreting the results.

In this report we describe a set of techniques designed to overcome the problems just discussed.

1. A medical appointment is depicted on videotape by professional actors. The patient's presenting complaints are relatively specific (e.g., chest pain, breast lump, dyspnea, depression) but ambiguous enough to allow a range of valid interpretation.

2. The scenario is videotaped in several alternative versions, featuring patient-actors of different age, sex, race, or other characteristics. The script does not vary except for minor modifications of language appropriate to the character of the patient.