AI-based Analyses and Teaching of Clinical Communication Skills

Butow, P., Hogue, E. (2020). Using artificial intelligence to analyse and teach communication in healthcare. The Breast, 50, 49-55. https://doi.org/10.1016/j.breast.2020.01.008

What this research was about and why it is important

This research is based on the following premises on communication in healthcare:

• Clinical communication skills are essential to determine a patient´s correct diagnosis, to decide on appropriate treatment and to achieve optimal patient outcomes.
• There is a close link between a healthcare professional’s communication skills and a patient’s capacity to understand, recall and follow medical recommendations.
• A clinician’s ability to listen and empathize with a patient’s emotions can have a profound effect on a patient’s psychological and functional outcomes, their experience of care, and their satisfaction with it.
• A clinician’s experience of and confidence in their communication with patients can impact their own levels of occupational satisfaction, stress and burnout.
• Healthcare professionals with poor communication skills face a higher risk of being sued by dissatisfied patients.

The researchers believe that the training of clinical communication skills should be theoretically- and empirically-based, and trainees deserve accurate feedback based on reliable and valid assessment methods. They compared the approach of human-based communication skills assessment techniques with AI-driven training systems. The main advantages of AI-driven systems are their availability at any time (human auditors might be busy with other tasks) and the elimination of trainees´ hesitations to seek help and accept feedback from a  colleague. Using AI-generated teaching resources and avatars in communication skills training can be helpful in building trust with patients, showing empathy and providing arguments and recommendations for treatment procedure in challenging situations. However, despite the promising developments of AI-based training models, the researchers recommend using the technology as an adjunctive assessment approach and teaching method.

What the researchers did

• They reviewed literature which described the role of good clinical communication skills and interaction analysis systems.
• They described the characteristics of human-operated interaction analysis systems and machine learning algorithms.
• They provided a summary of some key developments in the field of AI-driven applications for clinical communication skills assessment.
• They suggested objective metrics which could be identified by AI, e. g. word/sentence length and structure, use of clinical jargon; turn-taking in interaction between a healthcare professional and a patient, their intonation, pitch and pace, and then used this to provide feedback and comparison with peers to trainees, in a safe, confidential setting.

What the researchers found

• They pointed out the controversy in promoting modern approaches such as shared-decision-making and patient-centred care, in opposition to a doctor-centred communication style. The studies revealed that many patients don´t want to share decisions when they feel they have inadequate expertise and are vulnerable and in need of reassurance rather than autonomy. In other studies, patients preferred a paternalistic style of communication when they were explained the diagnosis and treatment procedures, and a more patient-centred style when discussing emotive issues such as prognosis.
• They presented the most common interaction analysis systems typically used to describe task-oriented and socio-emotional behaviour (e. g. giving lifestyle-related information, showing reassurance, agreement), but the systems differ in clinical focus and communication modes (verbal and/or non-verbal communication). In general, the selected systems showed reliable and valid data about clinical communication. However, the data still need to be processed by a human auditor, either with or without computerized support.
• They found out that AI applications have established moderate to good reliability of machine learning algorithms, comparable with human coding (or better). Recent advances in machine learning have allowed accurate textual transcription from speech, recognition of psycholinguistic features such as word level prosody, pauses, energy, intonation, understanding of the semantics and sentiment of the utterance, emotions and communication style both on document and sentence level.
• They suggest using avatars instead of simulated patients in clinical communication training, which demonstrated promising outcomes in some studies. An avatar can help in teaching empathy and dealing with ethical dilemmas, giving arguments and information to a patient to make decision on treatment.

Things to consider

• To date, most applications have been limited, targeting only a few concepts, and not yet investigating complex inter-relationships between variables. It remains important to generate evidence about the feasibility, reliability, acceptability and effectiveness of AI applications before their broad implementation.
• Developing the AI algorithms is promising because they make data generating and processing stages of the audit cost-effective, faster and more available.
• The biggest future challenges of developing AI applications used for data collection and analysis include creating more accurate algorithms for coding socio-emotional moves in human interaction and non-verbal behaviour, mainly facial expressions, gestures, and body movement.
• Further research will be also needed in exploring how authentic the experience of obtaining computerised feedback is to learners, and whether machine learning can produce insights that reflect the true complexity of doctor-patient communication.

Read the full-text here: https://www.sciencedirect.com/science/article/pii/S0960977620300096