Organisations increasingly use Twitter to communicate with service-users, including hospitals.

This study from the U.S. looked at whether a content analysis of Twitter-generated data can draw useful insights into how to improve patient experience of hospital care.

Methods

The authors started by identifying all hospital Twitter accounts in the USA. Outsourced workers were used to complete the task.

Only 2,349 of 4,679 hospitals had a Twitter account – that’s 50.2%. 404,065 tweets were retrieved from those accounts covering a one year period (1st October 2012 to 30 September 2013). The tweets were then cleaned to make them easily searchable.

To search the tweets, a specific software programme needed to be generated. To do that, researchers manually looked at a sample of tweets to see if they were about patients’ experiences. The process was repeated until there was reasonable agreement between the manual and automated classification. This allowed the content analysis of the tweets to be automated.

Once the database of tweets was generated, two different measures were used in the data extraction:

1. Sentiment calculation (authors’ labelling). This classified the tweet with either a positive or negative sentiment using a scale from -1 to +1. Any tweet with a score of 0 was discarded.
2. Manual classification. The objective was to classify the tweets into topics generated from the data (e.g. food, room condition, medication instructions, communication etc). The topic classification was applied to a subset of only 7,511 tweets, as agreed upon by two reviewers.

Only 297 hospitals were included at this stage as they had at least 50 tweets on patients’ experience.

The authors also looked at three other parallel subjects:

1. Hospital characteristics (region, urban-rural status, bed count, nurse to patient ratio, profit status, teaching status, percentage of patients of medicare/medicaid)
2. Hospital survey: researchers emailed the hospitals to ask for feedback on how they use their Twitter account (e.g. does your hospital monitor Twitter activity and do you follow-up with patients about comments?)
3. Comparison of the sentiment calculation with a validated measure of quality of care called HCAHPS (Hospital Consumer Assessment of Healthcare Providers and Systems). They also compared the sentiment calculation to the Hospital Compare 30-day hospital re-admission rate.

Results

• Only 50% of all hospitals had a Twitter account.
• The mean number of patient experience tweets was 43 per hospital.
• Hospitals with more tweets were likely to be below the national average for proportion of Medicare patients, and above national median for nurse/patient ratio, which suggests that Twitter use is more prominent in affluent areas.
• The link between the sentiment score and the validated measures was discordant. There was no association with HCAHPS and only a weak correlation with 30-day readmission.

Conclusion

The authors were quite innovative in trying to capture a non-traditional source of information. As the science develops around this kind of research we might be able to capture relevant and pertinent information in a usable manner that could hopefully somehow translate into better patient care.

Twitter content analysis might be another way to capture important topics for patients, but for now can be used to complement other tools.

Since the study, using social media and Twitter has become more common, so we might expect that this type of research be more relevant. Twitter appears to have potential as an important tool for listening and engaging with patients.

94% of psychosocial interventions were rated as low or insufficient

Limitations

A small number of tweets were included, but how these tweets were selected is not very clear. Only hospitals with 50 or more tweets were included, which limited the sample to a small fraction of all hospitals.

The classification scheme failed to categorise most Twitter topics, labelling the majority of tweets as ‘general’.

We weren’t clear on how the methods were validated and the terminology used to describe the method was also difficult to understand. We had questions around whether the context of tweets could be accurately captured using automated analysis. The pros and cons are that a huge amount of data can be gathered and cleaned, but how is the context and ‘meaning’ of the tweet captured? Additionally, Twitter users may not accurately represent the broad patient base.

Implications

Twitter may be an important way to engage patients and health organisations may need to develop their capacity to use this tool, potentially with a dedicated role to manage Twitter accounts and acting appropriately in a timely manner.

An important consideration is the ethical and legal implications of using data acquired through social media; health organisations need to develop policies and procedure to address these concerns.

Links

Primary paper

Hawkins JB, Brownstein JS, Tuli G. et al (2016) Measuring patient-perceived quality of care in US hospitals using Twitter. BMJ Qual Saf 2016;25:6 404–413 Published Online First: 13 October 2015doi:10.1136/bmjqs-2015-004309

Photo credits

• Howard Lake CC BY 2.0
• Bernard Goldbach CC BY 2.0