For a hospital’s managing director, most crucial and expensive resource is doctors’ bandwidth. Imagine an ailing patient entering the hospital for the first time.  In US, a GP would visit him and run the first-level diagnosis. Based on his analysis a line of treatment would be indicated or the patient would be referred to a specialist.

In India, the patient would go to the the front desk and they will guide to the appropriate department based on patient’s answers.

In both cases, It is humanly impossible to remember about vast number of medical issues with correlated symptoms. This means that initial diagnosis could take a little while, sometimes at the cost of patient’s time and patience.

Now imagine that front-desk is equipped with doctor-grade AI or the GP is assisted with doctor-grade AI, like ChatGPT, trained on vast corpus of medical data and researches. It is somewhat like all the specialists are standing behind the GP to help him reach correct diagnosis at the quickest time possible.
With the help from AI, pin-pointing the right department would be faster for the front-desk assistant. AI can help the GP handle a lot more load with higher accuracy.

Medical Generalist AI

With faster initial diagnosis, hospital’s productivity increases by magnitude. The fact is technology already has reached this level. Commercialization may take another year for the regulation bodies to create guard-rails for the use of such technology ubiquitously. But advanced hospitals may start utilizing the generalist AI service much earlier.

DeepMind Med-PaLM

Particularly two companies are at the forefront now. Google’s DeepMind is almost ready with Med-PaLM. Med-PaLM is a multi-modal large language model (LLM) trained on vast medical data. In the recent report it claims to be 90% accurate in its responses. It’s repertoire is vast if not comprehensive. It can understand Clinical Languages. It can analyze images (radiology etc). It is claimed to speak Genomics too!

OpenEvidence

And there is OpenEvidence, promoted by Mayo Clinic. Theirs’ also is a multi-modal large language model but with an extra chip on its shoulder. Unlike other LLMs like ChatGPT, it is trained with real-time data. With real-time data streaming into it, it would be up-to-date about latest medical findings on every sector –or that’s what they claim when the service will be commercially ready. It could be very soon since they recently claimed[2] it to score above 90% in United States Medical Licensing Examination.

OpenEvidence AI claims to become the First AI in History to Score Above 90% on the United States Medical Licensing Examination (USMLE)

Opportunity or Threat?

GPOnline finds that England needed nearly 7,400 more full-time GPs to manage the patient care. According to a 2021 WHO report, India was just above the 2006 standard of 22.8 healthcare workers per 10,000 population. With professional doctor-grade generalist AI at disposal of every hospital, relief is not going to be quantitative alone, it can add qualitative improvements in patient diagnosis in terms of reducing diagnostic time, improving accuracy and potentially reducing number of tests for the patient.

With a professional team always monitoring the model’s answers and fine-tuning the model’s learning, the improvement will be continuous and so will be the efficacy of the system.
Following the current trend, entire service most likely will be made available as SaaS for the hospitals and clinics. SaaS model also will help bringing incremental cost of adoption down to an affordable basis.

In summary, it looks like that generalist medical AI is going to bring paradigm-shift in patient care in not a very distant future. How pervasive the adoption would be though would depend on a hospital’s patient-care economics.

Read more

1.DeepMind announces Med-PaLM
2.First AI in History to Score Above 90%
3.OpenEvidence official site
4. Med-PaLM official page