B. Evidence Suggesting the Operation of Physician Bias

1. Challenges of Research Design and Interpretation

Having defined the type of physician bias under examination, we move now to inspect research suggesting that these biases sometimes operate in medical treatment decisions. Before describing specific studies, however, a brief examination of the challenges inherent in effectively designing and accurately interpreting studies involving disparities in health care services is in order. First, the temptation to equate a finding of a lower rate of health services received by a particular group with discrimination against that group must be resisted. A number of possible explanations--other than discriminatory motive or biased decision making--may explain disparities in utilization. For example, if one demographic group has a higher incidence of a particular condition or tends to be afflicted with a more severe form of the condition, a higher per capita utilization rate for that group (and likewise a lower per capita rate for the less afflicted group) is to be expected.

*206 In addition, a wide array of factors may play a role in a physician's choice of intervention for a particular patient, and thus may contribute to disparities between groups. Deciding the optimal treatment in any case requires the exercise of complex judgment, taking into account the stage and severity of the patient's condition, any other conditions (or “comorbidities”) that affect the patient's baseline health status, the patient's social situation (to the extent it may affect the patient's ability to comply with or benefit from a particular therapy), the patient's values and preferences and whether a patient's insurance will cover a particular therapy. If one group of patients tends to share clinical characteristics rendering a particular treatment medically inadvisable, then that group should have lower utilization rates for the inadvisable procedure. The multifactorial nature of medical judgment challenges researchers to control for these factors in an attempt to isolate the role that a group-related, but clinically irrelevant, characteristic plays. Thus, the complex nature of medical decision making renders bias in the process difficult to prove conclusively. By the same token, it also renders bias difficult to disprove.

Isolating the role of and the implications of bias is particularly challenging when the treatment at issue is deemed discretionary, or not clearly in the patient's best medical interests. When a physician must decide between more than one possible intervention, and none of the interventions is conclusively indicated for the patient, factors that are not purely medical may be more likely to sway the physician's choice of intervention. In that circumstance, the physician may rely on factors commonly deemed legitimate, such as the patient's preferences, or he may be guided by assumptions about the patient based upon his membership in a particular *207 group. In addition, the social policy implications of biased medical decision making are less clear when discretionary treatments are involved. If medical science cannot clearly say that “x rather than y is the preferred treatment for condition z,” then it is not at all clear that providing a patient with treatment y harms the patient in any concrete sense. In other words, we should be most concerned about disparities in the rates of health services provided to specific groups when those disparities can be linked to a difference in health outcomes for members of those groups.

Thus, the value of research suggesting the operation of bias in medical decision making--at least from a policy perspective--depends on researchers' ability to link the decisions made to particular health outcomes. Patients are tangibly injured by biased decisions only if a preference for treatment y based on race (or gender, etc.) causes patients to suffer worse outcomes than if treatment x had been provided. This point is particularly relevant in combating the common assumption that when it comes to medical care, “more is better.” A higher rate of utilization of a particular procedure by members of one group (e.g., whites), as compared to members of another group (e.g., African Americans), may represent underuse of the procedure by members of the latter group. It may also, however, represent overuse of the procedure by members of the former group. *208 The disparity's appropriate characterization depends on whether the procedure at issue can be shown to improve the health outcomes of those who receive it. If not, the group receiving what some might characterize as a “lower” level of care in fact may have received better quality care because group members have not been subjected to the risks and burdens of the inappropriate therapy. In other cases, a group's high utilization rates for a procedure may suggest that group members received lower quality care prior to resorting to the procedure. For example, one study suggests that African-American Medicare beneficiaries are more likely than white beneficiaries to undergo an amputation of all or part of a lower limb resulting from complications of diabetes because African Americans are less likely to have their diabetes optimally managed in order to prevent complications. Additionally, the women's health movement has questioned the assumption that more medical care is better, particularly with respect to medicine's intervention into women's reproductive functions.

Accordingly, for a study to find meaningful evidence suggestive of physician bias in a disparity in the rates of interventions provided to different patient groups, the researchers must control or adjust for the influence that other variable factors may have on a treatment decision. *209 Otherwise, determining whether the disparity flows from bias in the decision making or from other factors is impossible. Furthermore, for the evidence of bias to stimulate any strong remedial effort, the biased decision making should be connected with either actual or probable adverse health outcomes.

Although many published studies demonstrate disparities by race or gender in the rates of particular interventions, relatively few of these studies have met both of the conditions just indicated. The purpose of many of the studies was simply to demonstrate the disparity, not to search for its causes. But even those researchers who have sought to plumb the sources of disparities encounter difficulties in adequately controlling for potentially confounding variables. One cause of this difficulty is the limited demographic information contained in the large administrative databases of patient records that many researchers draw their raw data from. These databases typically contain information provided to third-party payers, such as information obtained by Medicare as part of its claims process. The patient record will include demographic information such as the patient's age, gender, and race and medical information such as the plaintiff's primary diagnosis and all treatments billed for. The record will likely not include information about comorbidities, severity of the condition, patient preferences or social factors affecting treatment choices. As a result, some researchers have urged caution in relying too heavily on the results of research based on these administrative databases.

Despite these difficulties, a number of studies extensively control for these confounding variables and find that a group-related characteristic, unrelated to patients' need for medical treatment, persistently correlates with a disparity in treatment rates. Even the authors of these studies, however, have refrained from concluding that bias based on group membership is the clear or exclusive cause of the disparity. Instead, their *210 conclusions have been more circumspect, suggesting perhaps that the difficult to control for variable of patient preference may be at work, or that the dynamics of the physician-patient interaction need to be examined more closely. Finally, many of the researchers caution that their findings of disparity based on a group-related characteristic may not be generalizable. In other words, the appearance that something questionable is going on in the doctor-patient relationship in the 200 or 20,000 cases the researchers studied does not necessarily suggest that similar dynamics pervade medical practice more broadly. To use a medical metaphor, because some number of physicians are infected with group-based biases does not mean that the condition is epidemic or even endemic.

This subpart seeks to highlight some of the complexities in designing and interpreting research studies that address the causes and effects of disparities in health services utilization. This cautionary note is critical to the Article's purpose. If bias does indeed play a role in some medical decisions in a way that may lead to worse patient health outcomes, then the question of whether the legal system provides any effective response becomes quite pressing. By contrast, if studies are not carefully designed or are interpreted in an oversimplified fashion that misidentify physician bias as a primary contributor to harmful disparities in health care utilization rates, then parties rightfully concerned about the disparities may spend their energies barking up the wrong remedial tree and fail to identify or address the true roots of the disparity.

Unfortunately, the extant research provides no conclusive answer regarding the role that physician bias plays in producing disparate rates of *211 health services consumption. Nonetheless, the studies described in the following sections, taken together, strongly suggest that physician bias, whether conscious or subconscious, infects some number of medical decisions and are thus sufficient to compel a serious inquiry into possible legal responses.

2. Evidence of Bias Based on Race

Our review of the evidence suggesting physician bias begins with research into racial disparities because researchers' efforts have concentrated in this area. Over the past fifteen years, a burgeoning number of published articles have examined racial disparities in the receipt of health services. Many of these articles have sought simply to identify the existence and magnitude of disparities between racial groups and have not sought to isolate the causes of the disparities. As such, these articles do not suggest physician bias as the particular source of disparity over other potential sources. A number of studies, however, have attempted to control or adjust for as many clinical and nonclinical variables as possible in order to determine whether racial disparities persist. Although the results of recent studies are not entirely consistent, several of these studies do find that race continues as a determinative factor regarding the receipt *212 of health services even when adjustments are made for potentially confounding variables. This subsection will describe a handful of these recent studies strongly suggesting that some dynamic in the patient-physician interaction--possibly including physician bias--plays a significant role in contributing to disparities.

The point of this description bears emphasis. The goal is not to try and prove that medical decisions by doctors in all contexts are racially biased; that ambitious effort--if relying on existing evidence--would be doomed to failure. Instead, the description's more modest purpose is to provide a sufficient evidentiary basis for concluding that race apparently does play a role in influencing physicians' medical treatment decisions in at least some cases. That assertion, if accepted, justifies an examination of potential legal responses.

a. Disparities in Drug Therapies for HIV

In a study published in 1994, researchers from Johns Hopkins examined whether racial differences in prescribed drug use at the time of patients' presentation at an urban HIV clinic could be attributed to differences in demographic factors other than race. Specifically, the authors of the study analyzed whether patients--all of whom had been diagnosed with HIV--had been prescribed either antiretroviral therapy or prophylactic therapy for Pneumocystis carinii pneumonia (PCP), before coming to the clinic. The authors noted that the optimal use of these prescription drug therapies is clearly defined in clinical guidelines issued by public health agencies; thus, the drugs should not be seen as discretionary therapies.

The results of the authors' analysis showed that black patients were significantly less likely to have been prescribed either antiretroviral therapy or PCP prophylaxis prior to their presentation at the clinic. Upon *213 further analysis of the data, the authors concluded that neither data on the patients' insurance coverage and income, nor such behavior as injection-drug use, nor demographic (sex and age) or socioeconomic (level of education and place of residence) factors explained the racial disparities. The analysis did show that black patients were less likely than whites to have had a usual source of medical care through which they could have received appropriate drug therapy before being referred to the clinic. But, the data also indicated that even among patients who identified a usual source of care, whites were more likely than blacks to receive drug therapy. The authors suggested that the disparities might flow from barriers to the spread of medical information that affect blacks more than whites, misconceptions about HIV and AIDS that are more common among blacks and a distrust of health authorities in the black community. The authors, however, also identified a final potential barrier to appropriate care for blacks: physicians' prescribing habits. After citing to studies suggesting that race may influence a physician's treatment recommendations, the authors emphasized that no evidence-based reasons existed for racial disparities in prescribing drug therapies for persons infected with HIV. In other words, if race influenced physicians' likelihood of prescribing either antiretroviral therapy or PCP prophylaxis for patients in the study, it was not because race was medically relevant to the patients' need for the therapy.

b. Disparities in Coronary Revascularization Procedures

In a study published in 1997, a group of researchers from Duke University undertook to advance an already large body of research reporting racial disparities in the use of cardiac procedures. They examined whether clinical features of patients' heart disease could explain racial differences in the use of two cardiac procedures by patients following cardiac catheterization. The researchers analyzed the medical records of patients who had been diagnosed with cardiac disease after a cardiac catheterization, both to discover whether coronary angioplasty or bypass surgery had been used and to determine whether the patient's age, sex, insurance status or a variety of clinical factors were significant predictors of which therapy would be chosen. The analysis revealed that the severity of the patient's disease had the strongest influence on which treatment was used, with the more severe disease being more likely to be treated with bypass surgery rather than angioplasty. After adjusting the rates of usage for all *214 of the clinical and nonclinical variables employed, the researchers found that while blacks were only marginally less likely than whites to receive angioplasty, blacks were thirty-two percent less likely to undergo bypass surgery. Moreover, the racial disparity in the utilization of bypass surgery was more marked among patients with more severe disease. Finally, by conducting follow up with the patients studied, the researchers found that black patients were eighteen percent more likely to die in the five-year period following their catheterization, even after adjustment for baseline prognostic factors. Given this linkage to unfavorable outcomes, the researchers concluded that revascularization procedures may have been underused in treating black patients.

In discussing their finding that race significantly affected both the chances of receiving bypass surgery and health outcomes, the authors characterized their results as “disturbing” because they were not attributable to differences in clinical characteristics. Discussing the few remaining explanations for the racial disparity, the authors raised the possibility that the patient's or the physician's preference for one cardiac intervention over another might vary according to race and concluded that “physician-patient interactions become key to understanding practice patterns.” Thus, a study that links racial differences in utilization to less favorable health outcomes for black patients and that eliminates a variety of clinical and demographic factors as possible explanations, strongly suggests that some part of the dynamic that occurs between physician and patient acts to the disadvantage of some black patients.

c. Disparities in the Treatment of Lung Cancer

Another study also identified a clear linkage between a lower rate of surgical interventions and worse outcomes for African-American patients. Researchers engaged in a population-based study of ten thousand patients who had been diagnosed with early-stage, non-small-cell lung cancer to estimate the differences in the rates of surgical treatment of those patients. For patients whose lung cancer is diagnosed early, surgical resection*215 offers a substantial cure rate; by contrast, patients who fail to receive the intervention have a median survival of less than one year. Thus, surgical resection is clearly the optimal treatment for early-stage non-small-cell lung cancer. The study's findings revealed, however, that black patients were less likely (by 12.7 percentage points) than white patients to undergo surgical resection, even when the researchers controlled for age, sex, stage of disease, type of insurance coverage, availability of care, income and coexisting illness. Further, the researchers estimated that the difference in surgery rates resulted in significantly lower survival rates for black patients.

In discussing their research results, the authors distinguished this study from others that demonstrated variations in practice patterns. Unlike the therapies considered in many other studies, surgical resection of early-stage non-small-cell lung cancer is not discretionary--instead it is unambiguously the optimal therapy. The researchers disclaimed any ability to explain why blacks received lower rates of resection, but identified as possible explanations racial differences in patient preferences or physicians' failure to offer black patients the optimal treatment as frequently as it is offered to whites. Again, while this study does not prove that physician bias accounted for the differential rate of surgeries for black patients, by disproving the role of a number of other clinical and nonclinical factors, it leaves physician bias as one of the few remaining possible explanations.

d. Disparities in Pain Relief Practices

A study published in 2000 describes how African-American patients in a hospital's emergency department (ED) were less likely than white patients to receive medications for pain relief. The researchers conducted *216 a record review for all black and white patients that presented with isolated long-bone fractures at the urban hospital's ED over a forty-month time period. They used a multiple logistic regression to establish race's independent effect on analgesic use, while controlling for variables including time since injury, time in the ED, shift of presentation, need for fracture reduction and payer status. After controlling for these variables, the researchers found that black patients were more likely than whites to receive no analgesia for pain relief in the ED.

In discussing their results, which were similar to those produced by an earlier study comparing the rates of analgesic administration to white and Hispanic patients, the authors discounted the possibility that the differences in pain medication ordered could have resulted either from racial differences in the pain experienced by patients or from racial differences in the assessment of the pain by health professionals. Instead, the researchers indicated that their findings suggested that patient race affected physicians' decisions to administer analgesics independently of any clinical criteria. In two editorials accompanying the study, emergency physicians noted the existence of potentially confounding covariables that the study had not controlled for. The editorialists, nonetheless, emphasized the need not only to conduct further studies on the existence and causes of racial and ethnic disparities in pain management, but to confront and respond to the real possibility that physician bias contributes to the disparities, particularly in light of other studies with similar results.

e. Disparities in Treatment for Stroke

Most recently, researchers have reported that black patients who have suffered a stroke are significantly less likely than white patients to receive *217 tissue-type plasminogen activator (tPA) (an effective therapy for ischemic stroke if it is initiated within three hours of the onset of stroke symptoms). Although tPA is a relatively new therapy, published consensus guidelines recommend its use, and the authors of the study describe it as the “first proven therapy for acute ischemic stroke.” The researchers examined the medical records of patients admitted for ischemic stroke at forty-two academic medical centers, comparing the rates of tPA use for white patients and African-American patients. When the researchers limited their analysis to patients who were candidates for tPA (i.e., those patients who had arrived at the hospital within three hours of symptom onset and who did not have any other contraindication to tPA use), they found that black tPA candidates were only about one third as likely to receive tPA as white candidates, even after adjusting for age, gender, insurance status and stroke severity.

After noting that the magnitude of the disparity found in the study was quite large in comparison to documented racial disparities for other procedures, the researchers considered possible explanations. They indicated that neither the failure of blacks to seek medical care quickly nor ethnic differences in stroke etiology or severity appeared to explain the disparity in tPA usage. They considered the possibility that blacks' greater distrust of the medical system and lack of confidence in treatment recommendations could contribute to the disparity, but noted that no black patients in the study had been documented as declining offered tPA treatment. Finally, the researchers considered the possibility that biases based on ethnicity and socioeconomic status explained the disparities. While calling for more detailed analyses of factors leading to treatment and acknowledging their inability to definitively identify racism as the primary cause of the disparity, the researchers concluded that racism likely contributed to the disparity in tPA treatment for ischemic stroke and called on practitioners to “carefully examine their own motivations when withholding this proven therapy.”

*218 3. How Might Race Play a Role in Treatment Decisions?

Taken together, the foregoing studies--along with other studies documenting racial disparities in the use of therapeutic interventions --suggest that doctors, as a group, may be less likely to provide black patients than white patients with aggressive therapies for life-threatening conditions, effective preventive care and effective pain relief. These differences cannot be explained away by clinical or nonclinical factors independent of the patient's race. Thus, we are left to confront the inevitable conclusion that a patient's race may play some role in a physician's decision of what therapeutic intervention to provide.

I hasten to interject that by asserting that race plays some role in medical decision making, I am not asserting that physicians in the U.S. are bigoted or malevolent racists who believe that African-American patients should receive a lower level of care than whites. While it is certainly plausible that cases of conscious, deliberate decisions to treat blacks less favorably than whites occur in medical practice (as they do in other settings), it seems far more likely that a patient's race influences treatment decisions in a more subtle, less overt fashion. One possible explanation is that physicians may unconsciously employ race-based assumptions or stereotypes in making judgments regarding therapeutic options.

The subject of how physicians make decisions and the possible operation of stereotypes has been examined far less often than the existence of racial disparities in utilization rates. Some evidence, however, exists that begins to suggest how patient race may affect physicians' perceptions *219 about patients. For example, one recent study examines how patient race and socioeconomic status affect physician beliefs about patient personal and psychosocial traits, patient behavior and patient role demands. The researchers used survey data to examine these effects during encounters between physicians and patients following an angiogram. After controlling for a number of physician characteristics and patient characteristics such as patient age, sex, frailty and social assertiveness, the researchers found that the survey results indicated that physicians tended to perceive black and lower socioeconomic patients more negatively on a number of levels. With respect to race, physicians were more likely to view black patients as being at risk for substance abuse and for noncompliance with cardiac rehabilitation and as having inadequate social support. Physicians were also less likely to report feelings of personal affiliation with black patients and judged black patients to be less intelligent than white patients, even when patient age, sex, income and education were controlled for.

In discussing their results, the researchers considered the possibility that physicians' perceptions were simply accurate observations of individual patient differences that, when aggregated, resulted in group differences. They rejected this possibility as a complete explanation because physicians rated black patients as less educated than white patients, when in fact the two groups were similarly educated. The alternative explanation offered by the researchers was that physicians incorporated epidemiologic evidence regarding population-based likelihoods into their general belief systems. Even though patient-specific information disconfirmed the *220 likelihoods, the physicians applied them to individual patients. In this fashion, epidemiologic information could effectively function as stereotypes and bias the way that physicians processed information about individual patients. The authors identified these differences in perceptions and feelings of affiliation as possible explanations for the racial treatment disparities documented in the medical literature.

A number of commentators have also suggested that physicians may unconsciously employ racial stereotypes or race-based assumptions in clinical decision making and thereby contribute to racial disparities in receipt of health services. Marian Gornick, who has extensively studied disparities among Medicare beneficiaries, describes how physician perceptions may operate to influence treatment choices:

[T]he course of treatment physicians . . . recommend to their patients may be influenced by stereotypical beliefs about the behavior of their patients. Physicians . . . may believe that poor and minority patients are more likely to break appointments and to misunderstand complex information, and less likely to adhere to their orders. These perceptions may affect--perhaps subconsciously--the decision-making process and lead physicians to refrain from orders that require patient compliance and to hesitate before recommending certain procedures if they assume the patient*221 does not live in an environment that is conducive to the aftercare needed for the best outcomes of the procedure . . . . Other commentators have described the operation of racial bias as “subtle” or “nuanced” and not perceived by the physician.

Although patients' preferences for therapies that are less aggressive, less expensive or more “tried and true” is always a potential (though extremely difficult to document) explanation of treatment disparities, a particular aspect of the oft-cited preference of black patients for less aggressive or less risky therapies bears noting at this point. It is conventional wisdom now that African Americans, as a group, have a higher than average level of distrust towards doctors. This distrust is typically traced to a history of racist medical practices in the United States. Although the history of segregated and racially discriminatory medical research and treatment practices both pre- and post-dates it, the Tuskegee Syphilis Study-- in which the U.S. Public Health Service sponsored research over four decades that examined the progression of untreated syphilis in poor black men in Alabama--is often cited as the paradigmatic example of racist medicine. Although many aspects of the study are disturbing, two of the most egregiously unethical facts are that the researchers never obtained any kind of informed consent from the black men infected with *222 syphilis to be part of the research, and as effective treatments for syphilis were developed, the researchers deprived participants of those treatments.

In light of the history of institutionalized racism in American medicine and the Tuskegee study in particular, it is hardly surprising that black Americans may be wary of treatment recommendations made by doctors and suspect that the health care establishment is pursuing goals other than the black patient's best interests. Indeed, some studies indicate that blacks may be less likely than whites to consent to some aggressive therapies. Based on this evidence and perhaps their own practice experience, “clinicians may believe that whites are more likely than blacks to prefer intensive medical treatment or surgical therapy” and consequently may be less likely to offer black patients aggressive therapy. Thus, the history of racial abuses in American medicine may have had the effect of putting into motion a vicious cycle: The history of discrimination causes blacks, as a group, to distrust white doctors; because blacks distrust doctors, they are generally more likely to decline aggressive or risky medical treatment; since blacks as a group are more likely to decline aggressive treatments, doctors (employing stereotypes) assume that individual black patients will prefer less aggressive treatment; and because doctors make this assumption, they are less likely to offer aggressive treatment to their black patients.

This scenario, if accurate, raises thorny questions about the appropriate response by medical practitioners. Does a physician treating black patients satisfy her ethical and legal obligations simply by making sure that she makes decisions for each patient solely on the basis of that patient's individual needs and that patient's medical information, without employing assumptions or stereotypes? If so, that physician's treatment recommendations will be unbiased, but resulting treatment patterns may still reflect some residual racial disparity because of the greater tendency of blacks, as a group, to decline aggressive treatments. The residual disparity could be characterized as reflecting patient preferences, a characterization that would allow the disparity (and any resulting adverse health outcomes) to be deemed acceptable out of respect for patient autonomy, and not the product of inequitable or biased medical judgments. Surely this complacent *223 conclusion cannot end the analysis, for the patients' preferences themselves are the product of African Americans' experiences of racist medicine. Thus, the question would become whether medicine needs to go beyond cleansing clinical decision making of all hints of bias, and additionally whether it needs to take affirmative steps to re-assure black patients of physicians' trustworthiness in an attempt to mold the patients' preferences. While possible legal remedies for biased decisions will be discussed below, this question of American medicine's moral obligation to make amends for past discrimination ties intimately into an assessment of how race plays a role in physicians' medical treatment choices.

4. Evidence of Bias based on Gender

When our attention shifts to sex-based differences in health services utilization, the evidence of disparities is less voluminous and the challenge of isolating bias as a cause of the disparities is more challenging. Some research into the existence of sex-based disparities in utilization has found either that no disparity exists or that the disparity exists only in limited areas. Moreover, other research suggests that sex-related disparities in treatment may sometimes reflect more appropriate care for women than men. Nonetheless, the cumulative evidence of disparities raises serious concerns that gender bias sometimes may lead to potentially harmful treatment differentials, in the form of either more or less frequent interventions for women.

One difficulty in assessing the meaning of sex-based disparities lies in determining when disparities in health care usage reflect sex-based differences *224 in medical needs. Unlike race, which is increasingly recognized as being a social construct rather than a biological fact, biological differences do exist between the sexes. These biological differences lead to differences between the sexes in disease incidence, disease manifestation and effective disease treatment, and consequently play some role in contributing to differences in the diagnostic and therapeutic interventions provided to men and women. Indeed, one criticism that women's health advocates level at the medical establishment is that medical research historically has failed to include women as subjects. As a result, there is a dearth of scientifically validated knowledge about sex differences and their clinical implications. Because researchers excluded women, and particularly pregnant women, from research into a variety of conditions, doctors often do not know whether the findings from male-subject research are equally applicable to their women patients. This knowledge gap may force practitioners either to treat women as if they were men (which should not lead to disparities in use, but in some cases may be ineffective or dangerous because of biological differences) or to treat women differently based on their un-validated assumptions about sex differences (which would lead to disparities in usage, but still may not be effective care if the physician's assumptions are inaccurate).

Despite these complexities in determining which disparities in medical care utilization flow from biological differences in women patients and which flow from other causes, some research suggests that gender bias plays a role in producing some utilization disparities between women and *225 men. Before reviewing this evidence, however, a few broad distinctions regarding gender differences in health care utilization, as compared to racial differences, bear noting. The research shows that--in broad terms--utilization of health care services by African Americans lags behind that of whites with respect to most services, including primary care, preventive care and therapeutic interventions for treating acute and chronic illnesses such as cancer, heart disease and HIV/AIDS. Blacks, however, are more likely to be treated for preventable conditions and to be hospitalized for treatment of a condition (both measures indicate receipt of less than optimal levels of preventive or primary care). By contrast, women have been shown generally to consume more health care services than men, in terms of visits to doctors, laboratory tests and number of prescriptions written.

Thus, in many instances, sex-related disparities in health care services tilt in favor of women receiving more care than men. With respect to treatment of certain conditions including heart disease, renal disease and lung cancer, however, studies have shown women to receive fewer interventions. In addition, one study finds that, among patients with advanced illness, women (or their proxy decision makers) are twice as likely as men (or their proxies) to perceive that their doctor had recommended only “comfort care,” as opposed to more aggressive treatment of their illness. But “comfort care” should not be confused with appropriate pain relief. Data from a number of studies indicate that women's complaints of pain are more likely to be discounted by health care providers and less likely to be treated adequately. Thus, in examining the evidence of possible gender bias in the receipt of health care services, it is particularly vital to remain attuned to the fact that more medical care is not always better *226 care and to focus on the appropriateness of care provided by seeking to link differences in the receipt of care to actual or probable differences in outcome.

That said, let us briefly examine several areas in which studies have documented gender disparities in receipt of health services and have suggested that physician bias plays a role. Again, while the published literature is far more extensive than the findings described here, I chose the following two examples of disparity as particularly illustrative and suggestive of bias.

a. Disparities in Physician-Prescribed Activity Restrictions

The first example of disparity does not involve women utilizing a procedure at a lower rate than men, but instead involves women being prescribed a particular intervention more often than men. In a study published in 1997, a team of researchers tried to identify the underlying reasons for physicians' more frequent imposition of activity restrictions on women than men. In their study, the researchers found that--even when controlling for differences in patients' socio-demographic characteristics, health profile, main role responsibilities, patient illness behaviors and physician characteristics--the odds of a physician directing a woman to restrict her activity was 3.6 times higher than for a man.

The researchers explored four competing hypotheses seeking to explain gender differences in medical care to determine whether any of the hypotheses could account for the study findings. The researchers concluded *227 that neither differences in the health profiles of the men and women patients nor their role responsibilities could explain the difference in rates of activity restrictions, but that both gender differences in illness behavior and physician gender bias appeared to contribute to the differential. The authors interpreted their findings that a larger differential existed among patients of male, older and subspecialist physicians as particularly suggestive of a role for attitudinal biases, and then explained how gender bias might operate in this context:

[A] societally based view of women as weaker and more vulnerable than men could incline physicians who have consciously or unconsciously adopted this view to advise activity restrictions for female patients more readily than for males. A devaluation of female role responsibilities might also make physicians less concerned about advising temporary activity suspension . . . . Attitudinal biases such as these may subtly and unconsciously influence physicians' perceptions of patients' preferences . . . . Thus, the authors concluded that their findings suggested that physicians' attitudinal biases influenced how the physicians interpreted patients' illness behaviors.

b. Disparities in Cardiac Care

Differences in the rates at which physicians have provided diagnostic and therapeutic interventions relating to heart disease for women, as compared to men, have been under study since the mid-1980s and disparity *228 in cardiac care appears to represent the single most closely examined area of sex-related disparities. Although the evidence here is inconclusive regarding the frequency and pervasiveness of disparities and their causes, recent studies have identified sex-related disparities in cardiac care and have suggested a possible role for physician bias in contributing to those disparities.

In a recently published study, researchers examined the rates at which men and women diagnosed with coronary artery disease were provided with lipid-lowering therapy (LLT) for high cholesterol. The researchers found overall a low rate of LLT among study participants, but found that “[t]he overall reduced rates of therapy were particularly noteworthy in women,” despite similar baseline and arteriographic characteristics. Moreover, the research showed that significantly fewer women than men in the study succeeded in lowering their cholesterol levels. The researchers concluded that their results “provide evidence of considerable sex bias at academic medical centers in the United States and Canada.”

Another study demonstrates an apparent interaction between gender and race in influencing physician decisions regarding cardiac care. Researchers in a widely publicized and controversial study used video recordings of actors portraying patients complaining of chest pain in a medical interview setting in order to assess physicians' recommendations for managing the chest pain. The researchers found that the hypothetical patients *229 (who all had identical histories, read their scripts verbatim and were directed to express a consistent range of emotions in their presentation) who were black women received referrals for cardiac catheterizations at a lower rate than white male patients, black male patients or white female patients. In discussing their results, the researchers characterized their findings as suggesting that “a patient's race and sex [or, more precisely, the combination of race and sex] may influence a physician's recommendation with respect to cardiac catheterization regardless of the patient's clinical characteristics,” but speculated that any bias was more likely to be the product of physicians' “subconscious perceptions rather than deliberate actions or thoughts.”

c. What to Make of the Disparities: Does Gender Play a Role?

A statement made by the AMA's Council on Ethical and Judicial Affairs in 1991 still rings true today: “Available data do not conclusively demonstrate a connection between gender bias and gender disparities in the provision of health care. Designing a study that can control for the myriad social, economic, and cultural factors that might influence decision making in a clinical context has proved extraordinarily difficult.” Despite the lack of conclusive evidence, however, the suggestion of gender bias in clinical decision making is quite compelling and cannot be ignored. As with race, I do not suggest that all doctors are sexist and misogynist and deliberately provide different and less effective care for women. Instead, as with race, any bias is far more likely to be subconscious and unperceived by the practitioner, and is likely to be present in some providers but not in others. The effects of allowing patient gender (when gender is not related to medical need) to influence treatment decision making are real, however, and may operate to the detriment of either women or men. While the operation of subconscious gender bias seems likely to mimic that of unconscious race bias, researchers in this area have proposed that two factors may be particularly likely to interact in the context of treatment decisions for women.

*230 First, as noted above, because women were historically left out of medical research, physicians treating women for many conditions have less scientifically based guidance than they do for their male patients. Thus, unless a physician simply assumes that findings from male-subject research are equally applicable to female patients, the physician is left to exercise his best judgment with respect to treatment of his female patients. In other words, because of the paucity of scientifically valid research regarding females with a variety of conditions, the treatment of female patients is more likely to be discretionary. And, as noted above in the discussion of racial disparities, situations in which treatment choices are discretionary or diagnostic determinations are uncertain create greater opportunities for a physician's own subjective perceptions or societal stereotypes to influence the decision made.

Second, there is no shortage of fairly consistent social stereotypes pertaining to women. Although women have made huge strides in the past four decades towards achieving greater equality and eroding gender-based stereotypes regarding their “place,” gender-based norms and expectations persist quite robustly in American society. In the medical community, these expectations may be embodied in physicians' assumptions that women are less likely to choose aggressive interventions, assumptions that women are less likely to have demanding social or career roles, attributions of women's physical complaints to emotional or mental causes and devaluations of women's contributions to society. Each of these gender-based stereotypes could infect a physician's judgment regarding a female patient's diagnosis and her optimal treatment. While doctors are probably unlikely to flatly deny a female patient an aggressive and potentially effective therapy that she actively seeks, the physician's counseling of the patient to follow the course that the physician believes is in the patient's *231 best interest could easily provide the vehicle through which the bias operates.

5. Evidence of Other Biases

In addition to evidence strongly suggesting the existence of race and gender bias in clinical decision making, some (albeit much sparser) evidence exists indicating that other patient characteristics unrelated to a patient's medical need--such as age, sexual orientation, disability or obesity--may influence physicians' treatment decisions. This subpart will briefly describe some of this evidence in order to reinforce the concept that physician bias is not monolithic, but instead can wear many faces and present varying issues as the nature of the bias varies.

For example, a number of studies indicate that older patients are less likely to receive aggressive medical care than younger patients, even after controlling for the severity of the patients' illness. This age-based disparity appears particularly strong in the context of aggressive treatments for advanced illness. One study found that among seriously ill, hospitalized adults, older patients were more likely than younger patients to have treatments such as surgery, dialysis and ventilator support withheld, even after adjusting for patients' preferences for life-extending care. Another study found that patients over the age of seventy (or their proxy decision makers) were almost four times more likely than younger patients to believe that their doctor had recommended only comfort care for them, rather than life-sustaining treatment. Yet another study suggests a patient's advanced age would not by itself bias a doctor's treatment decision for a patient who had been functioning well mentally and physically before the onset of an illness, but that an older patient with pre-illness disabilities would likely be treated less aggressively than a younger patient with the same illness and similar pre-illness disabilities. Similarly, a survey of physicians who were questioned regarding their agreement with treatment choices for hypothetical patients revealed that a significant proportion of the respondents concurred in judgments to treat an older patient *232 less aggressively than a younger patient, even when those patients' likelihoods of survival were identical.

Researchers have advanced several possible explanations for these age-related differences in care. One team suggests that physicians may recommend less aggressive treatment for older patients based on an erroneous belief that older patients are more likely to suffer poor outcomes or are less likely to benefit from aggressive treatment; this team, however, also recognizes the alternative possibility that the differences in care may reflect “a covert, and perhaps unintentional form of health care rationing.” Other researchers point to the possibility of “ageism, or discrimination based on chronological age” and suggest that this bias may reflect either society's values or the physician's own personal values and preferences. Bolstering these suggestions are attitudinal surveys demonstrating that physicians subjectively associate “old age” with sickness and inactivity and indicating that students in their first year of medical training have already formed negative impressions about older patients and how aggressively they should be treated.

What is perhaps most distinctive about apparent age-based bias in medical decision making is that some commentators view patient age--unlike patient race or gender--as a potentially legitimate consideration in making health care treatment decisions, even when age is unrelated to the individual patient's medical need or prognosis. Several scholars have argued that advanced age is a legitimate moral basis for rationing health care resources. Others have responded that chronological age is an unjust and arbitrary basis for allocating resources. It should be noted, however, that most proponents of age-based rationing argue for the adoption of explicit social policies implementing such a scheme; they do not advocate that individual physicians should covertly seek to ration care through their treatment decisions for older patients. Nonetheless, the unresolved *233 debate regarding age-based rationing highlights the complexity of the issues that age bias raises and suggests that physicians' choices of less aggressive care for elderly patients may reflect their personal belief in the legitimacy of age-based rationing.

A patient's sexual orientation may also bias a physician's medical judgment. Homosexuality remains socially stigmatized, and evidence suggests that some physicians share commonly held negative attitudes towards gay men and lesbians. Surveys indicate that many physicians feel uncomfortable treating gay men and lesbians and may even view homosexuality as an illness. These negative attitudes may bias a doctor's medical decisions so that a patient's sexual orientation influences the treatment provided even when sexual orientation is unrelated to the patient's medical needs. A 1994 study polled the membership of the American Association of Physicians for Human Rights (today called the Gay and Lesbian Medical Association) regarding how sexual orientation affects patients and doctors. Almost two-thirds of the physician respondents thought that a patient who disclosed his or her homosexuality to a doctor risked receiving substandard treatment, and about half had witnessed their colleagues (other physicians) providing reduced care or denying care to gay or lesbian patients. A patient's sexual orientation may influence a physician's clinical judgment not only by making the physician less willing to pursue aggressive treatment on the patient's behalf, but also by leading the physician to focus inappropriately on the patient's sexuality, thus interpreting the patient's problems in sexual terms rather than considering a full range of diagnoses.

Out of concern for the perceived likelihood of disapproval and substandard treatment, many lesbians and gay men are reluctant to disclose their sexual orientation to their physicians. Herein lies an important distinction between bias based on gender, race or advanced age and bias based on sexual orientation. Because a patient's sexual orientation may not typically be a trait that is readily visible, patients must decide whether to divulge their sexual orientation to their physician and risk receiving *234 biased treatment. This fear regarding the consequences of disclosure appears to lead many gays and lesbians to respond in one of two ways, each of which may be quite detrimental to their health. Some may seek medical care but remain silent regarding their sexual orientation. While this silence may prevent a physician from inappropriately considering the patient's sexuality in exercising medical judgment, it also precludes the physician from considering the patient's sexual orientation in an appropriate (i.e., a medical-needs-related) fashion. Gay men and lesbians do have certain health needs that differ from those of the heterosexual population. A physician who simply assumes that all her patients are heterosexual unless the patient discloses otherwise may thus “misdiagnose conditions, provide inadequate treatment, offer irrelevant health teaching, lecture needlessly about birth control, ask insensitive or biased questions, and make sexist remarks.”

The other way that gay men and lesbian women may try to avoid biased treatment is to avoid medical care altogether. For example, one study showed that almost half of lesbians rarely or never visited a gynecologist. Gay men may also be less likely to seek preventive care than their heterosexual counterparts. Thus, when we examine the problem of sexual orientation bias in medical care, we see that physician bias (particularly the more overt, expressed sentiments of disapproval and hostility that gay men and lesbians may face) can be pernicious for the subjects of the bias not only when the bias operates to corrupt medical judgment, but also when the subjects of the bias seek to avoid its operation. Effectively, the potential for bias can function as a barrier to accessing effective care for gay men and lesbians.

Two remaining patient characteristics that may influence medical decision making beyond any clinical relevance of the characteristic are patient disability and patient obesity. Although some evidence suggests that the process of medical education may improve students' attitudes towards persons with disabilities, physicians at times still tend to focus their attention *235 on a patient's disability, rather than addressing her health needs more generally. In addition, the well-publicized case of Sandra Jensen illustrates that physicians may be reluctant to pursue aggressive medical treatment for some persons with disabilities. Physicians at two academic medical centers in California initially refused to place Sandra Jensen on a waiting list for a heart and lung transplant because they believed that a person with Down syndrome would not be able to comply with the rigorous post-transplantation regimen. Thus, physicians may allow the mere existence of a disability to influence their medical judgment without examining how the individual patient's disability affects her actual medical needs.

As with physical and mental disabilities, obesity is a patient characteristic with potential health consequences that a physician may need to consider in order to address appropriately a patient's medical needs. Evidence suggests, however, that obesity in a patient may also stimulate strongly negative and judgmental reactions from physicians, just as it may from members of the general public. One result of such openly biased behavior is that obese persons, like gay men and lesbians, may be reluctant to seek medical treatment. Beyond overt and conscious negative reactions to patients who are obese, however, physicians may also subconsciously incorporate negative social stereotypes regarding obesity into their medical decisions. A physician recently reflected in The New York Times on the possibility that his treatment of a patient with heart disease had been influenced by the patient's obesity. His words reflect the struggle involved in recognizing “camouflaged” bias:

*236 It did not appear that our prejudice, if it existed, had been conscious. Granted, the patient's obesity had been on our minds in deciding on his treatment. But had we been hypersensitive about his obesity, to his detriment? Had we made a value judgment that because of his weight, surgery would be wasted? Or worse, that he was somehow less deserving of surgery because he was unable or unwilling to control his weight? These words provide a fitting conclusion to our survey of the evidence regarding various forms of physician bias in medical practice. They begin to suggest the variety of ways that a characteristic of a patient may influence a physician's medical judgment beyond the characteristic's relevance to the patient's medical needs. The following section will consider what to make of the evidence that has been surveyed.