Preface:

“Poor quality health services are holding back progress on improving health in countries, at all income levels, according to a new joint report by the OECD, World Health Organization (WHO) and the World Bank.”
As per WHO report ” Health care workers in seven low- and middle-income African countries were only able to make accurate diagnoses one third to three quarters of the time, and clinical guidelines for common conditions were followed less than 45 percent of the time on average.”
“Around 15 percent of hospital expenditure in high-income countries is due to mistakes in care or patients being infected while in hospitals.”

AI community is trying address these issues for some time and Google has come up with Med-PaLM-2. It is a large language model (LLM). It is an optimized PaLM model for medical questions.

Benchmarking of Model:

An expert clinician must collect and synthesize various types of medical information from various sources, which is time consuming process. E.g., Doctor may need to collect information from images(X-Rays), Vital Signs – Current and Past (from electronic health records). The records could be from sensors, wearables and at times even from genomics. All this may lead to more delays, and expenses.

The Med-PaLM 2 gathers such information from all these discrete sources and uses “PaLM-E”. It helps doctors to provide better care by synthesizing the information gathered from various visuals such as chest-x rays, mammograms etc.  Med-PaLM was developed using a special soft prompting method combined with responses to medical prompts written by real world clinicians.

Here are few of the points used to score the correctness of Med-PaLM-2:
1. How does it relate to the current consensus of the scientific and clinical community.
2. What is the extent of the possible harm?
3. What is the likelihood of possible harm?
4. Correct reading comprehension.
5. Correct recall of knowledge.
6. Correct reasoning step.
7. Does this answer contain any content that it shouldn’t?
8. Does this answer omit any important information?
9. Does this answer provide content any information that is inaccurate for one particular patient group or demography?

Real life use cases:
A.  Kenya has been facing a shortage of workers trained to operate traditional high-cost ultrasound machines. Google has partnered with Kenya based non-profit organization “Jacaranda Health” to improve health services to mothers and babies. In current exploratory work, they are focusing ultrasound delivery with AI tool.
B.  X-rays images have proven methodology to detect breast cancer in early stages and have been proven approach to reducing mortality. In the partnership with Chang Gung Memorial Hospital (CGMH), Taiwan is exploring the possibility to use AI models for early detection of breast cancer using ultrasound.

Benefits:
Time is very crucial when providing medication:  Reading vast relevant data, Synthesizing and coming up with conclusive decision is fast with AI. With these characteristics of Med-PaLM 2, it provides quick and timely diagnosis, treatment, clinical plan to patient.
Another perspective is the cost. One of the expensive activities in providing good health care, is the number of tests to be conducted. These tests are not only expensive but at times time consuming. With AI we could make it faster,  and less expensive. AI can assess the diabetes or heart-disease risk by scanning retinas rather than administering numerous blood tests.

Pitfalls:
One of the serious pitfalls among others is, will the system be biased? However, Alan Karthikesalingam, a clinical research scientist at Google and author on the Med-PaLM 2 study says that the developers training and evaluating Med-PaLM 2 at Google are diverse, which could help the company identify and address biases in the chatbot. But he adds that addressing biases is a continuous process that will depend on how the system is used.

Conclusion:
Google is still working on it before it is made available for the widespread use. Med-PaLM has been released with limited access. However, it can be good a companion to the expert clinicians.