Large language models (LLMs), flexible tools for language generation, have shown promising potential in various areas, including medical education, research, and clinical practice. LLMs enhance the analysis of healthcare data, providing detailed reports, medical differential diagnoses, standardized mental functioning assessments, and delivery of psychological interventions. They extract valuable information from ‘clinical data’, illustrating their possible widespread use. However, these models are often unable to use data from wearable technologies, which monitor vital health aspects, due to the data’s complex nature and storage requirements.
In light of this, a new Google study presents a Personal Health Large Language Model (PH-LLM), a model designed to extract and utilize data from wearable technologies to guide users in achieving specific health goals. PH-LLM turns acquired data into insights and actionable suggestions for improving sleep hygiene and physical activity. Tested using various sleep and fitness-related tasks, the model showed a remarkable improvement in the use of domain knowledge and the customization of relevant user data.
In their approach, the researchers used multimodal encoding and high-resolution time-series data and tested the model’s efficacy with 857 case studies collected from volunteers. The case studies centered around assessing sleep quality and fitness preparedness for workouts. The findings revealed that the model had impressive reasoning and knowledge skills, ranking high in performance across all case study responses.
In addition, the researchers created tools for automated case study evaluation, which paralleled the evaluation metrics used by human experts, improving the rating speed and consistency.
When fitting the model to predict individual’s perception of sleep disturbance and sleep impairment, the researchers found that integrating sensor data was vital.
The study encountered limitations, such as bias in case study evaluations, inhibiting clear comparison between different models. Despite these challenges, the researchers effectively applied the model to improve many personal health outcomes and offer individualized health advice. Going forward, the scientists aim to increase the datasets used to train the model to improve interaction predictions.
The Google study elevates the application of LLMs, demonstrating the potential of tailoring these models to improve personal health outcomes. As continuous monitoring and patient-centric care become more prevalent in health practice, the ability to effectively use data from wearable technologies offers exciting possibilities for patient care and self-monitoring.
