There has been a growing interest in understanding the factors that may influence a patient's perception of their physician and how these perceptions might in turn impact quality of care. Research has shown that patients often rely on nonverbal cues, such as a physician's attire, to form their opinions about their health care provider [1-6]. With the advent of various web-based review platforms, such as Healthgrades and Zocdoc, patients have easy access to a wealth of information about their health care providers, as well as a relatively straightforward means to express their satisfaction. Research suggests that these web-based ratings do influence patients' choices of medical providers [7-9]. On the web-based physician review platform Zocdoc, a half-star improvement in ratings on a scale of 1-5 stars leads to a 10% increase in appointments [10,11].

Past research is mixed on the factors influencing these web-based reviews. Some research suggests that patient experience and clinical quality directly affect web-based ratings [12,13]. Other research suggests that there exists minimal correlation between the quality and value of care or peer-assessed performance with web-based ratings, and that nonphysician characteristics such as staff friendliness and appointment wait times are instead the key determinants [14-16].

What remains clear, however, is that many different factors play a role [17,18]. Web-based reviews for female physicians, for example, have been found to be more emotional and informal compared to those for male physicians [19]. Moreover, female physicians receive lower web-based ratings than their male counterparts on some platforms, though this trend is not consistent across all platforms [19-23]. Other research finds that patient-physician racial concordance is associated with higher scores on internal patient satisfaction surveys [24]. One study of a direct-to-consumer telemedicine platform shows that, on average, patients report higher rates of dissatisfaction with Black and Asian physicians compared to White physicians [25].

However, the influence of physician factors on physician ratings is likely not limited to gender and race alone. Despite the growing influence of web-based review sites, much remains unknown about the various factors that may influence physician ratings. For instance, many studies have examined the association between patient attractiveness and physician care practices [26-29]. However, to our knowledge, no study has examined how physician attractiveness impacts patient satisfaction in real-world medical settings.

Other factors such as school ranking and multilingualism are similarly understudied. Existing research finds little correlation between medical school ranking and performance scores [30]. In addition, language-concordant care is associated with increased patient satisfaction [31,32]. However, the roles of ranking and multilingualism have not been examined in web-based review settings.

In this paper, we aim to explore the role of various physician characteristics, including gender, age, race, number of languages spoken in a physician’s office, education, and facial attractiveness, in shaping patient perceptions of their physicians through ratings on a web-based doctor review site.

We collected primary care doctor profile data from Zocdoc, a platform that allows patients to rate and review their experiences with doctors. Physician variables included facial attractiveness, race, age, gender, language, and education. Outcome variables included overall satisfaction, bedside manner, and wait time scores from patient reviews. Descriptive statistics, the Wilcoxon rank sum test, and multivariate logistic regression were used to analyze the data.

A total of 1521 primary care doctor profiles from Zocdoc were collected. Of these, 66 were missing an overall satisfaction rating and excluded, leaving a sample of 1455 primary care doctor profiles for analysis (Table 1). One profile lacked information on the medical school attended, one profile contained a profile picture from which facial attractiveness, age, and race could not be estimated, and one profile contained a profile picture for which only facial attractiveness could not be assessed. These 3 profiles were still included for analysis. Three cities—Houston, Chicago, and New York City—represented 49% of all primary care doctors in the sample (Table S4 in Multimedia Appendix 4). The majority of physicians were men (54%) with no nonbinary individuals represented in the study. The sample included speakers of 66 languages, with a mean of 1.73 languages spoken (Table S1 in Multimedia Appendix 1).

This study aimed to examine the impact of physician characteristics, specifically gender, age, facial attractiveness, medical school ranking, foreign degree status, and number of languages spoken, on patients' ratings of primary care physicians on Zocdoc, a platform that only publishes reviews from verified patients. Our univariate findings show that older age, lower facial attractiveness, foreign degrees, allopathic degrees, and more languages spoken in office are negatively associated with the overall rating of primary care doctors. However, the effect sizes of these factors are small and may not be clinically significant. For example, we find a 0.14 gap in median overall rating between doctors with a US degree and doctors with a Latin American degree. Prior research identifies that a half-star improvement in ratings leads to a 10% increase in likelihood that a physician will fill an appointment [11]. A back-of-the-envelope calculation suggests that this 0.14 gap is equivalent to a 2.8% increase in appointments.

To our knowledge, this paper is the first to examine the association between the number of languages, age, facial attractiveness, degree type, foreign graduate status, school ranking, and race with web-based physician reviews, and the first to examine the association between the number of languages and school ranking on patient satisfaction in general.

Our findings on the role of degree type and school ranking are generally consistent with past research. For example, a national telephone survey found that patients of osteopathic physicians generally reported higher rates of satisfaction than patients of allopathic physicians [49]. Past research has found that osteopathic physicians are more likely than allopathic physicians to call patients by their first name, provide information on the underlying causes of their illnesses, and have conversations with them about the social, family, and emotional implications of their medical conditions, all of which may contribute to higher satisfaction rates [50]. In addition, past research finds little correlation between medical school ranking and patient mortality and readmission rates, suggesting an elite ranking has negligible impact on patient care [51].

While it may be easy to hypothesize reasons for differences based on age, facial attractiveness, or foreign degree status, it is not entirely clear what could drive the differences we observe with the number of languages. We find that more languages spoken is associated with a decrease in rating. The number of languages may serve as a proxy for foreign-born status, but we cannot be certain. It is possible that physicians who speak 2 languages are more likely to be bilingual US natives whereas those who speak more than 2 languages are more likely to be immigrants. Survey data have found that patients report lower satisfaction with international medical graduates [25,52]. Older Medicare patients notably have lower mortality rates when treated by international graduates compared to US graduates [53].

In addition, we did not find any numerical rating differences by race. This is inconsistent with past work on findings on a direct-to-consumer telemedicine platform in which patients report lower satisfaction with Asian and Black physicians [25]. Moreover, we did not find any numerical rating differences by gender, which is inconsistent with some previous papers on the topic [18,19,21,22,54]. Zocdoc, however, differs from many other physician review platforms in that reviews can only come from patients after they have received care from their respective physician [33]. On third-party independent platforms like RateMDs, Google Reviews, or Healthgrades, anyone may post a review regardless of whether they have actually seen the provider. Additionally, Zocdoc solicits reviews after each appointment which may counter the typical biases of only extremely satisfied or unsatisfied consumers leaving reviews, leading to larger, more representative samples. Lastly, Zocdoc only publishes patient reviews that do not violate their community standards barring those that include personal information, pricing specifics, profanity, claims about the accuracy of a provider’s treatment or diagnosis, or promotional content.

Several limitations must be considered in interpreting the study results. First, our study only examined patient reviews of physicians on one web-based review platform and results may not generalize to other platforms. For instance, millennial women, New Yorkers, and residents of urban areas are disproportionately represented as patients on Zocdoc [55]. Moreover, while we find no differences by gender, female physicians on review platforms such as RateMDs and Google Reviews have been found to have lower numerical ratings than men [20-22]. This may be driven by a lack of verification in review postings or a lack of review moderation. Because Zocdoc moderates reviews, we cannot determine how results might change if reviews that violated community standards were included in the analysis. For example, it is possible that immigrants or women are more likely to receive lower but also more profane reviews. If Zocdoc removes these profane reviews, any disparities we are able to observe may be attenuated. In addition, because most physicians have a very high mean rating, there may be a ceiling effect that makes it difficult to discern relationships between ratings and physician factors. Second, we restrict the study to primary care physicians and the results may differ for other specialties. For example, one small study of 271 sports medicine surgeons found gender differences in ratings on 1 of 3 platform studies [23]. Third, we limit our analysis to differences in numerical ratings. However, linguistic analyses may yield different types of bias; research on Zocdoc does find that text reviews of women physicians are more informal and emotional than reviews of men [19]. Fourth, we rely on automated face classification software to classify facial attractiveness, race, and age. However, while past social science work may have relied on human raters to code unstructured information, many social scientists have fully moved toward using automated algorithmic procedures [38-43,47,48,56-59]. Moreover, while we do not have access to the ground truth data on race and age, patients most likely do not have access to this information and instead are influenced by perceived race and age. Fifth, the number of languages spoken in a provider’s office may correspond to the number of languages spoken by the physicians themselves or by their staff, and what this variable proxies is unclear. Sixth, the overall explanatory power of our multivariable models is fair and may be limited due to the many factors that play key roles in patient satisfaction that we are unable to observe. Seventh, our data do not permit any mechanistic or causal interpretations. Despite these limitations, our study represents an important step in understanding the potential biases in web-based doctor reviews and highlights the need for further research in this area.

Although our study suggests that physician factors have a real but limited impact on numerical ratings, it is important to note that bias may play out in many forms. For example, a physician's appearance or accent may impact a patient's trust, confidence, or satisfaction with their physician, which could in turn influence their take-up of preventative services and lead to either better or worse health outcomes [60]. Such a phenomenon would not be captured in our analysis of numerical ratings, but our results open the door to investigating such phenomenon across facial attractiveness, multilingualism, education, and age more deeply. In conclusion, this study provides insights into the association between physician characteristics and patients' web-based ratings of primary care physicians. Future research should consider textual analyses of reviews, investigate how factors like facial attractiveness interact with patient outcomes, and explore whether the findings of this study generalize to other medical specialties, review platforms, or patient populations. Ultimately, our findings underscore the need for greater awareness of potential biases in web-based doctor reviews and the importance of considering a range of factors in evaluating health care providers.