Gray B, Vandergrift J, Lipner R. — American Board of Internal Medicine
Guodong G. — University of Maryland
McCoullough J. — University of Minnesota
Presented: American Public Health Association Conference, November 2013
Objective: To assess the accuracy of patient ratings of physicians obtained by Internet searches.
Method: Physician quality measures (QMs) were drawn primarily from patient charts for 1,042 internists who completed a Diabetes or Hypertension Practice Improvement Module (PIM). PIM completion is part of the American Board of Internal Medicine's (ABIM) Maintenance of Certification process and included abstracting 25 charts and collecting 25 patient survey responses for patients with the applicable chronic condition.
Chart PIM-QMs applicable to both patient cohorts included: share of patients with blood pressure/LDL controlled and providing smoking cessation advice. Patient survey PIM-QMs included patient assessment of care quality/physician self-care support rated as “very good” or “excellent.” An expert panel was convened to form a PIM-QM composite, comprising mostly of chart-based QMs. PIM-QMs mirrored quality measures applied by NCQA/CMS.
Consumer searches were mimicked by entering each internist's name and location into a Google search, extracting ratings from the first two health websites on the search list. Searches captured 1,013 ratings for 602 internists (58% of the sample) from eight websites. Ratings were normalized by dividing each rating by that website's maximum score (e.g., two out of five stars yielded a normalized rating of 40%). The normalized website rating was categorized as either top score if >=80% (59% of ratings) or bottom score if <=0% (12% of ratings). The relationship between internists' PIM-QMs and having any website rating was estimated using probit regression. Physician QMs were evaluated as dependent variables using binomial regression where a website rating was the explanatory variable controlling for sub-specialization, PIM and website. Physicians rated on multiple websites was also accounted for.
Results: Analysis indicated that patients with diabetes/hypertension obtained limited information from Internet searches regarding which physicians to avoid (bottom rating associated only with patient share with LDL controlled, P < .05) but not whom to choose (normalized rating or top rating was not associated with any PIM-QM, only six ratings per website). However, ratings from the most popular website (HealthGrades, n = 465) yielded information regarding the overall quality of physicians (normalized rating was a significant positive predictor of PIM-QM composites, P < .05). Yet, more information about quality would have been revealed had ABIM patient ratings been accessible (PIM-QM patient reviews significantly correlated with PM-QM clinical composite (P < .001) and these associations were significantly larger than website ratings associations (interaction P < .01).
Conclusion: Consumers obtain limited physician quality information from website searches and yet there exists a great potential for patient reviews to aid consumers.
For more information about this presentation, please contact Research@abim.org.