We appreciate Dr. Hofer's thoughtful comments about our article as well as his well-researched commentary on the more general issue of the usefulness of statistical discrimination as a conceptual framework in disparities research. In response to his suggestion for additional analyses of our data, we have conducted two analyses to examine cross-level interactions between the physician estimate of coronary heart disease (CHD) prevalence and patient gender, with the outcome of certainty of the CHD diagnosis. As noted in our paper (Maserejian et al. 2009), physicians were more certain of the CHD diagnosis for male patients, even when we adjusted for their assessment of the CHD prevalence in the statistical model. We now add that a multiplicative interaction term, consisting of the three-level categorical variable reflecting the physician's belief about CHD prevalence (higher in men, higher in women, or no gender difference) by two levels of patient gender, was not statistically significant (p=.82). This nonsignificant finding supports our paper's findings that prior prevalence estimates did not explain (nor modify) the effect of gender on diagnostic certainty. We also conducted a stratified analysis, which we present in Table 1. Consistent with results of our paper, the gender effect persisted across all categories of physicians' beliefs of the CHD prevalence for men versus women. Interestingly, the gender difference slightly decreased in accordance with physicians' beliefs about the relative prevalence of CHD, but it must be noted that only a small number (n=18) of physicians assessed the prevalence as higher in women. Overall, prior belief about the relative prevalence of CHD did not meaningfully modify gender disparities in CHD certainty. Table 1 Stratified Analysis of Physicians' Assessments of Population Coronary Heart Disease (CHD) Prevalence and Patient Gender, for the Outcome of Certainty of the CHD Diagnosis Beyond our empirical analysis, Hofer comments more broadly on the conceptual framework of “statistical discrimination” that we used to frame our results, but which has its origins in previous research. Hofer expresses concern that the “statistical discrimination” terminology has disadvantages, primarily in that it lumps together beneficial and detrimental reliance on prior data. We agree that the terminology casts a broad net and may be confusing; we observe informally that the term is used in an increasingly wide array of contexts and often with different interpretations of meaning. We also share some of Hofer's concerns about the importance of identifying and specifying mechanisms and hypotheses to facilitate interpretation of results and allow a clear understanding of the roots of disparities. At the same time, we recognize that the many nuances of statistical discrimination reflect the inherently complicated process of balancing individual patient-level data with prior data (including prevalence or personal experience) when both are needed for clinical decision making. We are therefore concerned that discarding all that statistical discrimination encompasses to simplify disparities research may impede the understanding of clinical decision making processes, by obscuring the often necessary tension between sources of data and the opposing ways in which they can influence clinical outcomes. For example, relying on existing prevalence data to help diagnose a patient of a particular group may be beneficial to the patient at hand, but in the aggregate continued reliance on population-level prevalence data could eventually be detrimental to patients of that group if such reliance substitutes for greater physician effort to derive individual estimates. Hofer's suggestion that clinical decision support can help correct some of these biases points toward some important next steps for research on quality of care, while also underscoring the enduring importance of the complexity involved in the statistical discrimination concept. In a separate analysis using data from the same study (Ketcham et al., 2009), we examined how information technology (IT), including clinical decision support, feedback, and electronic medical records, was associated with variation in clinical decision making. The observed gender disparities in CHD certainty were not lessened among physicians who reported using clinical decision support aids that provide disease prevalence data in their practice (clinical decision support was not provided during the experiment itself). Overall, IT's effects on disparities in diagnostic certainty and treatment were complex and multidirectional. Thus, current policies to reduce disparities by increasing IT adoption should proceed with caution, close monitoring, and process evaluation. Although clinical decision support is intended to be a step in the right direction, if physicians' preconceptions about the distribution of symptoms and conditions—whether accurate or inaccurate—interfere with their ability to receive, interpret, and diagnose patient-specific symptoms, then a central challenge to minimizing disparities remains. The complexity of the processes simultaneously operating in clinical decision making underscores the continued need for comprehensive decision making models.
Search from 250 M+ research papersSearch from 250 M+ research papers
Jun 18, 2024
Mark J Kissler + 5
Open Access