Redefining Therapy Documentation with AI & Enhancing Clinician Productivity

To address this challenge, KidsAbility has partnered with Dr. Bryan Tripp and a team of biomedical engineering graduate students at the University of Waterloo to explore the use of AI and Large Language Models (LLMs) in pediatric rehabilitation. These AI tools and LLMs have the potential to drastically reduce the amount of indirect time required for clinical documentation.   

Our approach has involved testing two different AI documentation tools. The first approach, a prompt-based LLM, generates visit notes based on brief text inputs from therapists. While it demonstrated a reduction in documentation time, therapists found it challenging to adopt. This is mostly because therapists don’t typically create the type of rough notes that the model requires as an input. As a result, a net new set of behaviours and workflows are required for this type of tool to work. Building that new workflow was difficult to justify given the marginal time savings the AI produced.  

The second approach, a transcript-based LLM, uses audio to autonomously transcribe therapy sessions. These transcripts can then be used to generate a range of documentation types. Clinicians were more receptive to this approach, particularly in fields like physiotherapy and occupational therapy. For therapists in these disciplines, instead of visit documentation taking 15min per session, they were able to complete the task in just 5min. Further, because the entirety of a therapy encounter is used to produce draft documentation, the scribes often catch details that a therapist might otherwise have missed. This results in some cases in more accurate notes.  

Speech-Language Pathologists (SLP), however, had different experiences. They experienced challenges with this type of AI because SLP therapy involves sounds and word fragments that aren’t recognized by the transcription tools. If you have non-sensical transcript, you also get a non-sensical draft note. This finding has led to a new project with Dr. Tripp, where we are now investigating whether phoneme (distinct speech sound) data can be used to improve the accuracy and fidelity of SLP encounter transcripts. If successful, this would enable scribe-based AI tools to work for all of KidsAbility’s therapy disciplines.