# Online Reviews of Health Care Facilities

**Authors:** Neil K.R. Sehgal, Sharath Chandra Guntuku, Lauren Southwick, Raina M. Merchant, Anish K. Agarwal

PMC · DOI: 10.1001/jamanetworkopen.2025.24505 · JAMA Network Open · 2025-08-01

## TL;DR

This study analyzes online reviews of US healthcare facilities to find language patterns linked to positive or negative patient experiences.

## Contribution

The study identifies specific words and themes correlated with patient satisfaction or dissatisfaction in healthcare online reviews.

## Key findings

- The word 'not' and administrative barriers correlate with negative ratings.
- The conjunction 'and' and support staff interactions correlate with positive ratings.
- Payment issues and poor treatment are strongly linked to negative feedback.

## Abstract

This cross-sectional study analyzes online reviews of patient experiences in US health care facilities, identifying words, linguistic categories, and themes correlated with highly positive or negative ratings.

What words, linguistic categories, and themes correlate with highly positive or negative ratings in online reviews of health care?

In this cross-sectional study of 1 099 901 online reviews posted from 2017 through 2023, 46.3% of US health care facilities were rated 1 to 2 stars and 50.1% were rated 4 to 5 stars. The word “not” and topics related to administrative barriers were significantly associated with negative ratings, whereas the conjunction “and” and words related to support staff interactions were associated with positive ratings.

Findings of this study suggest that assessing online-review language in real time could help clinicians and administrators identify possible emerging communication or access problems and target patient-centered quality-improvement interventions.

Understanding patient experience is crucial for improving health care delivery. However, language patterns and themes correlated with negative or positive ratings are not well known.

To examine online reviews of US health care facilities, identifying language patterns and themes associated with negative or positive ratings.

For this cross-sectional study, all reviews of US health care facilities offering essential health benefits, as defined by the Affordable Care Act, posted on 1 online platform (Yelp.com) under “Health & Medical” from January 1, 2017, to December 31, 2023, were obtained. Reviews are posted voluntarily with ratings (1 star = lowest, 5 stars = highest) and open-ended review narratives regarding patients’ care experiences.

The primary outcome was the correlation between n-grams (1- to 3-word sequences) and review ratings (negative: 1 or 2 stars; positive: 4 or 5 stars). Secondary measures included linguistic analysis and topic modeling based on standard machine-learning algorithms. Machine-learning and natural-language processing, including n-gram correlation, linguistic feature analysis, and topic modeling, were applied to determine correlations with review star ratings.

A total of 1 099 901 online reviews from 138 605 facilities were identified over the 7-year study period. Among these, nearly one-half (46.3%) were negative and one-half (50.1%) were positive, with a median (IQR) rating of 4 (1-5) stars. The word “not” was most correlated with negative ratings (r = 0.31; 95% CI, 0.31–0.32), whereas “and” was most correlated with positive ratings (r = 0.35; 95% CI, 0.35–0.36). Among 200 topics, the strongest negative correlations involved payment issues (r = 0.25; 95% CI, 0.25–0.25) and poor treatment (r = 0.24; 95% CI, 0.23–0.24); the strongest positive correlations involved kindness (r = 0.32; 95% CI, 0.32–0.32) and anxiety relief (r = 0.32; 95% CI, 0.32–0.32).

In this cross-sectional analysis, negative patient experiences frequently centered on quality of communication and administrative issues. Negative feedback centered on unmet expectations, whereas positive reviews emphasized supportive staff interactions. Incorporating real-time online-review data into existing quality-improvement frameworks—such as patient experience dashboards or service recovery protocols—could help clinicians, administrators, and policymakers identify emerging concerns, monitor patient sentiment, and tailor interventions that enhance patient-centered care across diverse health care settings.

## Full-text entities

- **Diseases:** COVID-19 (MESH:D000086382), LIWC-22 (MESH:C535733), anxiety (MESH:D001007)
- **Chemicals:** n-gram (-)
- **Species:** Homo sapiens (human, species) [taxon 9606]

## Full text

_Full body text omitted from this summary view._ Fetch the complete paper as Markdown: https://tomesphere.com/paper/PMC12317347/full.md

## Figures

2 figures with captions in the complete paper: https://tomesphere.com/paper/PMC12317347/full.md

## References

22 references — full list in the complete paper: https://tomesphere.com/paper/PMC12317347/full.md

---
Source: https://tomesphere.com/paper/PMC12317347