Microsoft’s former executive vice president for Windows and Devices, Terry Myerson, recently tweeted about the new Truveta Language Model (TLM), which he claims is trained on the largest collection of complete and representative clinical data. According to Myerson, TLM has over 90% accuracy on medical concepts and is free from commercial bias introduced into claims data by revenue cycle systems.
The Truveta platform was launched in January 2021 as a collaborative effort among a group of major U.S. health systems including Providence, Tenet Healthcare, HCA Healthcare, CommonSpirit Health, Trinity Health, Adventist Health System/West Florida Division and Baylor Scott & White Health. The primary aim of Truveta is to leverage their vast troves of patient data to improve healthcare outcomes through research and innovation.
Truveta’s CEO Terry Morgan said that TLM was developed specifically for use with electronic health records (EHRs). EHRs contain massive amounts of unstructured text notes written by doctors and other medical professionals during patient visits. These notes often include vital information about patients’ medical histories but are difficult for computers to interpret accurately without proper training algorithms like TLM.
By training TLM on clinical data rather than claims data from revenue cycle systems used by insurance companies, the model can provide more accurate diagnoses while avoiding potential biases that may exist in claims data due to financial incentives. This could lead to better treatment decisions for patients who might otherwise receive suboptimal care based solely on insurance reimbursement policies.
According to Dr. Peter Lee, Microsoft’s corporate vice president responsible for healthcare initiatives: “We believe this technology will enable researchers across the industry to accelerate discoveries that advance our collective understanding of disease prevention and treatment.”
While Truveta promises great benefits from using TLM with EHRs – such as improved diagnosis accuracy – there are concerns around patient privacy issues given its access to sensitive personal information.
In response to these concerns, Truveta has stated that it is committed to protecting patient privacy and will use de-identified data for research purposes. Additionally, the company’s board of directors includes a number of experts in ethics and data privacy.
Truveta is not alone in developing such technology. In recent years, a number of companies have developed natural language processing (NLP) algorithms specifically for EHRs that can extract relevant medical information from unstructured text notes. Some notable examples include Epic Systems’ SlicerDicer, Google Health’s DeepVariant and IBM Watson Health’s Clinical Reviewer.
However, TLM appears to be the first NLP model trained on clinical rather than claims data. This could give it an edge over other models when it comes to identifying rare or unusual medical conditions that may not be well represented in insurance claims data.
Overall, Terry Myerson’s tweet about the Truveta Language Model highlights an exciting development in healthcare technology that could lead to better diagnoses and treatments for patients. However, as with any new technology accessing sensitive personal information, there are concerns around patient privacy that must be addressed by all involved parties.