How Google is Working to Harness AI to Address Physicians’ EHR Documentation Burden

Physicians’ electronic health record (EHR) documentation processes are a subject of much debate. Physicians’ focus on EHR charting during the office visit is believed affect provider-patient communication and also clinical outcomes. Moreover, EHR data entry is associated with physician stress and burnout. Medical transcription outsourcing helps reduce the documentation burden to a great extent, but continuous efforts are being made to improve the medical charting process for physicians. The latest development is Google’s attempts to harness artificial intelligence (AI) to improve note-taking for physicians.

EHR Documentation

Components of EHR Charts and Medical Notes

Systematic documentation of a patient’s medical history, diagnosis, treatment and care is necessary for accurate and complete EHR notes. Medical notes are a record of everything related to the patient, including diseases, major and minor illnesses, and growth milestones. Information is the EHR chart will include:

  • Surgical history
  • Obstetric history
  • Medications and medical allergies
  • Family history
  • Social history
  • Habits
  • Immunization records
  • Developmental history
  • Demographics
  • Medical encounters

During the medical encounter, the provider has to enter all the information relevant to the patient’s care such as:

  • Chief complaint
  • History of the present illness
  • Physical examination
  • Evaluation, diagnosis, and treatment plan
  • Orders and prescriptions
  • Progress notes
  • Laboratory and imaging test results

There are basically two formats for physician note-taking: the narrative style and the bullet-point/checklist style. Notes may be also a combination of these formats. The narrative style, which describes “what happened?” and “what is going on?, is suitable for the history of present illness section. The narrative style offers a clear picture, but can be lengthy and time consuming. On the other hand, the bullet-point style allows clinicians to list the relevant information and symptoms without much detail or context.

In digital patient records, much of the notes are limited to checklists and bullets. Though this improves efficiency, saves time, and supports medical billing, it can make notes difficult to comprehend. Moreover, EHR checklists tend to oversimplify serious problems

Automating the Physician EHR Note-taking Process

According to a 2016 study:

  • For every hour physicians spend with patients, they spend about two additional hours on EHR and desk work within the clinic day.
  • Physicians spend nearly half of the total office day on EHR and desk work and less than one third on direct clinical face time with patients.
  • Physicians spend another 1 to 2 hours of personal time outside office hours doing additional computer and other clerical work.

The use of dictation and medical transcription services as well as scribes help clinicians to reduce time spent on note-taking. Google believes that physicians’ EHR documentation processes can be improved using AI tools, according to a research paper cited in a TechXplore report.

Peter Liu, a researcher at Google Brain, proposes a new language modeling task that can predict the content of new notes by analyzing demographics, laboratory measurements, medications and past notes in patient medical records. Published on arXiv, the paper proposes automated EHR note-taking as the solution to physicians’ EHR documentation burden.

The focus of the paper was on building language models for clinical notes. Liu put forward two language models: a transformer architecture model for shorter notes and a transformer-based model for longer sequences. The models were able to correctly predict a lot of the content of physicians’ notes. According to Liu, these models could help in the creation of more advanced spell-check and auto-complete features, which could be integrated into tools to support clinicians in performing their administrative tasks.

“We find that much of the content can be predicted, and that many common templates found in notes can be learned,” Liu writes in the paper. “Such models can be useful in supporting assistive note-writing features such as error-detection and auto-complete.”

Liu points out that there are challenges to be overcome before these models can be put to practical use. Limitations include the insufficiency of context provided by the EHR such as the lack of imaging data and lack of information about the latest patient-provider interactions. The researchers say that future work could involve combining EHR data with information from outside a patient’s medical record, such as imaging data or transcripts of patient-physician interactions.

Medical Transcription Outsourcing is still Relevant

Till such new models are perfected, the services provided by experienced medical transcription companies will continue to be relevant to ensure complete and accurate medical charts. With expert support, health care providers do not have to worry about incomplete or inaccurate medical charts. They can focus on patient care as their EHR documentation needs are taken care of by their reliable medical transcription service provider.