Natural language processing in plastic surgery patient consultations

Information extraction and summarization

Plastic surgery is a highly specialized discipline. Plastic surgeons operate across the whole body. Unsurprisingly, our experts require very specific knowledge about patients and their medical history. This includes how well their comorbidities are controlled (i.e., diabetes) and information about their surgical history (i.e., history of abdominoplasty). Electronic health records (EHRs) contain a wealth of this information. However, most patient EHRs contain hundreds of clinical documents. Manually searching for information can be tedious due to the volume and jargon. For example, in a provider note, “PT” can mean “patient”, “physical therapy”, “posterior tibial artery”, “posterior tibialis”, “prothrombin time”, or “part-time”.

LLMs that comprehend clinical documentation should be able to understand the context and easily extract information. The most basic task is “named entity recognition” - deriving the names of patients, medical procedures, and medications directly written in a document^[1]. Plastic surgeons might use named entity recognition to identify key surgical information, such as what tissue was resected, what type of mesh/implant was used, and what flaps were used in the reconstruction.

Summarization, on the other hand, is a more complex and generative task. It requires a holistic understanding of a text. Here, the generative LLMs could help plastic surgeons by processing several clinical documents in the EHR to summarize a patient’s surgical history (i.e., all previous breast procedures) or overall health status. This would help plastic surgeons prepare before or during a consultation to guide operative planning. A group from Stanford applied eight different LLMs for clinical summarization and found several untrained LLMs more completely summarized patient history than humans and had fewer errors^[2]. Importantly, their study highlighted the importance of “prompt engineering”, that is, phrasing questions so the model can generate useful summaries. Their study also found that GPT-4 had the best performance compared to other models. Indeed, industry leader Epic © has already announced a forthcoming integration of GPT-4 into its platform^[3]. This would enable surgeons to use Epic to instantaneously extract and summarize information for patient care. It would also facilitate powerful analytics and predictive modeling.

Ambient transcription and coding

Provider burnout is a well-described phenomenon in plastic surgery and healthcare. While the problem is multifactorial, several studies have described administrative burden as a root cause, including documentation^[4-6]. The ability of NLP systems to generate clinical documentation could ease this burden through ambient transcription. This function could be integrated into plastic surgery consultations, which involve documentation of patient history, treatment plans, and consent forms. Traditionally, this process is time-consuming as surgeons must manually record patient information during or after consultations. Additionally, the current medico-legal milieu holds our discipline to the highest standard of accurate documentation. NLP can automate this process by transcribing spoken consultations into written records. Voice recognition systems can capture and transcribe the conversation between the patient and surgeon, reducing the administrative burden on the surgeon and allowing for greater attention toward the patient. Indeed, several startups are building and field-testing models to perform these tasks, including Ambience ©, Nabla ©, and others^[7,8].

Once a patient’s consultation is transcribed, clinicians should aspire to use NLP to automatically assign diagnosis and procedural codes. An editorial from Venkatesh et al. describes the current state of innovation in automated clinical coding (ACC), and the challenges facing companies^[9]. While the need is apparent, it is difficult to integrate clinical data and documents into a few codes. Startups like AKASA © have made strides in building an ACC tool that performs as well as human coders^[10]. In the future, an ACC tool could radically streamline the workflow for plastic surgeons.

In summary, the promise of NLP in plastic surgery documentation is widespread: improved accuracy and consistency of patient records, minimizing the risk of errors from manual entry, and decreased administrative burden on plastic surgeons.

Patient goals, PRO

Plastic surgery is uniquely focused on patient goals and understanding. This focus arises from the nature of our services, which primarily aim to augment quality of life. It is no surprise our discipline has pioneered modern PRO measures^[11]. To this end, NLP systems should help us better understand patients and their goals. Ambient transcription systems will be able to identify important themes within a consultation and highlight what the patient perceives as important. For example, if a patient consistently expresses concerns about scarring, the surgeon can address this issue more thoroughly during and after the consultation.

NLP systems may also complete a PRO questionnaire during the consultation, alleviating the burden on patients. Having a preoperative BREAST-Q or Michigan Hand Questionnaire, for example, would help surgeons understand a patient’s well-being in a more objective way. This would also facilitate postoperative patient tracking to measure a patient’s progress using the same PRO instruments.

Translation and health literacy

LLMs should also assist in translating clinical materials, both in conversation and clinical documents. This would allow non-English speakers to more readily access care. This is particularly useful in plastic surgery, where a surgeon may need to communicate surgical details to a patient who is more comfortable with another language. Beyond language translation, LLMs can also address varying levels of health literacy within the same language. Studies have found translating clinical documents for geriatric patients may improve patient-provider interaction^[12]. More generally, translating plastic surgery jargon like “NAC” and “IMF” into more familiar terms will aid patient understanding. For example, NLP can analyze the complexity of the language used in consent forms and provide suggestions for simplifying the text to match the patient’s comprehension. Furthermore, NLP can be used to generate personalized educational materials tailored to the patient.

Chatbot

While NLP systems can facilitate conversations between plastic surgeons and patients, they may also conduct their own conversations to provide patients with real-time support and information. “Question answering” is a discipline in NLP involving systems that answer questions posed by humans, using a relevant context and in a natural language. Several foundation models, including GPT-4 and Med-PaLM 2, have shown promise in the medical question answering space^[13,14].

Question answering chatbots will enable patients to access care on demand 24/7. After a consultation, patients can use a chatbot to answer common questions (i.e., about postoperative showering or dressings), provide reminders for medication or follow-up appointments, and even alert the surgeon if the patient reports any concerning symptoms. These systems can help optimize patients’ conditions before and after their surgery, such as by advising on food and activities.

Still, recent otolaryngology and neurosurgery literature suggests we should temper our expectations. A study by Gajjar et al. compared versions of GPT in answering neurosurgery-specific patient questions. They found that, while factually accurate, the responses lacked readability and were not rated as highly beneficial^[15]. Another study by Karimov et al. found ChatGPT was inferior to UpToDate in answering several otolaryngology-specific patient questions^[16]. However, both these studies used untrained versions of GPT. Future investigations should attempt to train plastic surgery-specific chatbots to improve performance.