Jump to start of content

Breadcrumb trail:

Evaluating website ratings: How well do patient ratings of physicians predict their quality of care?


Gray B, Vandergrift J, Lipner RS. — American Board of Internal Medicine

Gao G. — University of Maryland

McCullough J. — University of Minnesota

Presented: AcademyHealth Annual Research Meeting, June 2014

Objective: Considering the growth in use of the Internet, consumers need to know the degree to which Web-based patient ratings of physicians predict care quality. Understanding this relationship is particularly important for patients with chronic conditions who have the potential to benefit greatly from high-quality primary care. This study evaluated the degree to which Web-based ratings of a physician predict care quality for their patients with diabetes or hypertension.

Study Design: The study included 1,299 internists who completed a diabetes or hypertension practice improvement module (PIM) between 07/2011 and 11/2012. PIMs are part of the American Board of Internal Medicine's 10-year Maintenance of Certification process. Completing these PIMs included abstracting 25 charts and collecting 25 patient survey responses for patients with the applicable chronic condition. Applying data from PIM chart abstractions, the following physician-level clinical composites were constructed by applying weights and scoring algorithms from an expert panel of internists: (1) overall clinical composite, (2) process of care composite and (3) an intermediate outcome composite. Measures of patient satisfaction were derived from survey responses. Consumer Internet searches were mimicked by entering each internist's name and location into a Google search. Searches captured 1,573 ratings for 792 internists (61% of the sample) from eight health websites. Ratings were normalized (rating divided by website maximum) and a website rating was categorized as being the top if the normalized rating was greater than or equal to 80% (59% of ratings). To assess the ability for website ratings to predict care quality, each clinical composite and measure of patient satisfaction was regressed against normalized ratings. Regressions, estimated using a binomial-regression (Logit-link) which accounted for physicians being rated on multiple websites, included controls for sub specialization and PIM completed. Sensitivities considered risk adjustments.

Population Studied: Mid-career board certified internists who routinely treat patients with diabetes or hypertension.

Principal Findings: With effect sizes close to zero and p-values greater than 0.15, website rating measures were unrelated to clinical composites. Website rating measures were statistically significant (ps<0.05) predictors for the two patient satisfaction measures but effect sizes were also small. A 20 percentage point change in normalized rating was associated with at most a 1.7 percentage point increase in the share of patients with very good or excellent assessment of care quality (p<0.05).

Conclusion: Overall, little evidence was found showing that website ratings predict care quality. Website ratings were not related to clinical composites and only weakly related to the two measures of patient satisfaction.

Implications for Policy or Practice: There is a growing demand by consumers to be able to easily obtain quality of care information about providers. This is particularly an issue for patients with chronic conditions that benefit greatly from high-quality primary care. This demand is being met through online sources. Yet this study suggests that these sources do not yield valid information about clinical quality and little information about patient satisfaction. Thus, to make health care markets function in a way that encourages quality improvement, policy makers need to make available the kind of validated quality of care measures applied in this study.

For more information about this presentation, please contact Research@abim.org.