A Textual Analysis of Socioemotional Content in Healthcare Reviews
The text you write on social media, or any other online platform can reflect many things, such as the writer’s sentiment, intention…
The text you write on social media, or any other online platform can reflect many things, such as the writer’s sentiment, intention, opinion, and most importantly, biases, if any. In past work, there have been several instances where researchers and analysts have analyzed text data (social media data, interviews, surveys, physicians’ notes, etc.) to identify hate speech, gender bias, and other forms of discrimination. Take, for example, online review platforms such as those for writing reviews about your doctor’s visit. Many of these websites are available, such as ZocDoc.com, RateMyDoctors.com, Vitals.com, and Yelp.com, among others. With the numerous text analysis tools and techniques available, text analysis has become much easier and can help us identify different components as mentioned earlier. From the analysis of online reviews written by patients for their physicians, I found some peculiarities in the language used by female physicians.
Dataset:
I used the publicly available datasets from ZocDoc and RateMyDoctors websites where I focused on the text reviews column for the analysis. The ZocDoc dataset had a total of 19,372 reviews (after removing duplicates) for 555 unique doctors from 2008 to 2015 (214 female and 341 male doctors) whereas, from RateMyDoctors, there was a total of 46,554 rows (after removing duplicates) with 3,552 male doctors and 2,647 female doctors.
Text Analysis Tool:
To perform the text analysis, I used a psycho-linguistic tool called Linguistic Inquiry & Word Count (LIWC). This tool leverages decades of research to analyze language and provide insights into psychological states, emotions, thinking styles, and social concerns. LIWC-22 includes over 100 specialized dictionaries for various psychological categories, such as cognitive processes and affiliation. It calculates the percentage of words in a text that match these categories, helping to quantify psychological phenomena. The tool’s categories are hierarchical, and words can belong to multiple dictionaries.
The main aspect I focused on in this blog is the socio-emotional content in the language.
Socioemotional Content:
What is socio-emotional content? In human interaction, socio-emotional content refers to the interaction and emotions between individuals and their relationships. Since the analysis was about the doctor-patient relationship, this was one of the obvious metrics to assess the quality of this relationship, as the online reviews were a form of patients’ emotional reactions. To assess this type of content, I measured the emotional tone from the text reviews in three ways using LIWC categories as follows:
Tone variable: This returned the standardized metric of the difference between the use of positive and negative emotions.
Positive and negative emotions score
Social variable: From the LIWC dictionary, this helped measure the extent to which patients perceived their physicians’ social interaction and consideration beyond just the specific symptoms.
Real-World Applications:
Understanding socio-emotional content in patient reviews can have several practical applications:
Improving Healthcare Practices:
Example: Healthcare providers can use these insights to train physicians to be more empathetic and socially engaged with patients, which has been shown to improve patient satisfaction and outcomes.
Application: By recognizing patterns in patient feedback, clinics can tailor their services to better meet patient needs, fostering a more supportive and effective healthcare environment.
Addressing Gender Bias:
Example: Identifying that female physicians are often reviewed as more socially involved can highlight existing implicit gender biases in patient expectations and evaluations.
Application: This awareness can lead to initiatives to reduce gender bias in patient reviews, ensuring fairer evaluations for all physicians.
Enhancing Doctor-Patient Communication:
Example: If reviews indicate that emotional content is higher for female physicians, other physicians might be encouraged to adopt similar communication styles to enhance patient relationships.
Application: Training programs for physicians can incorporate findings from text analysis to improve their communication strategies, thereby enhancing patient trust and care quality
Statistical Analysis:
There were other LIWC categories and aspects of language that were measured and analyzed in the original research, but for the sake of this post, let’s focus more on how socio-emotional content contributed to analyzing factors reflecting gender bias. For ZocDoc data, the results of socio-emotional content were as seen in the following forest plot:
From the findings and the above plot, the emotional tone leaning towards more positive emotion (posemo) was higher for female physicians as compared to male physicians. The social aspect was also higher for female physicians. The plot below is for the RateMyDoctors dataset, where we see a higher social score for female physicians and not much difference in the emotional tone for either gender of the physicians.
From both datasets, the findings suggested that the reviews for female physicians contained more social references. Additionally, the ZocDoc data showed that the emotional content in the language was higher for female physicians.
Conclusion:
According to past literature and the findings from this research, it is noteworthy that female physicians were reviewed as more socially involved with their patients than male physicians, indicating their eagerness to connect beyond just conveying treatment diagnoses. In patient care, it has been proven that establishing a connection with patients improves patient outcomes. To learn more about the rest of the analysis, stay tuned for my next post and look at the original published research here.
References:
1. ZocDoc Data: https://github.com/turbosantosh/MLReviewsToRatings
2. RateMyDoctors Data: https://github.com/Shikhar-S/Multi-Aspect-Sentiment-Classification-for-Online-Medical-Reviews/