If data is new oil, use it to fund healthcare

Transforming Healthcare in UP in India, a State with a Population of 240M

A few years back, we, at State Street HCL (a State Street Corp and HCL Tech JV), were collaborating with Prof. Regina Barzilay at MIT on a project. Her story of Artificial Intelligence (AI) is personal. She was diagnosed with breast cancer in 2014. This was after her scans in 2012 and 2013 had missed a small but clearly visible white mass. Maybe, it was her training as a mathematician that she could not reconcile to the fact that there was so much uncertainty about the diagnosis. She started training her software using tens of thousands of mammograms and patient records from Massachusetts General Hospital (MGH). This was to ensure that the computer could identify patterns and establish the correlation with breast cancer. Today, her software is used by the same hospital to not only diagnose but predict breast cancer up to five years in advance.

In some of the most common diagnostic areas like diabetic retinopathy, AI is already reliable. Arvind Eye Hospital in Tamil Nadu works with Google on the same for a few years now. In many other areas, progress is significant. For example, earlier this year FDA authorised marketing of first cardiac ultrasound software that uses AI to guide user, which means technicians can perform echocardiography which is currently done by radiologists.  So, use of AI in diagnostics – blood tests, ultrasound, radiology, even our poor old ECG is technology of here and now. 

The greatest challenge in AI in healthcare (or AI in most areas for that matter) is not the software but the availability of the data to train the software. Now, consider two data points:

  • UP has a population of 230 million, which would make it the 6th largest country by population, if it were a separate country. 
  • Investors put in more than $4B in AI in Healthcare in 2019 (up from $2.7B in 2018), as per a CB Insights report. 

So, UP has data of about 230 million people but needs better healthcare. The healthcare AI industry has billions, has low-cost healthcare solutions, but is looking for data.

Let us think about it in real terms. UP has 18 divisions, 75 districts and 822 development blocks. Each of these development blocks has more than one primary healthcare center (PHC) and possibly a Community Health Center (CHC), but most of them are dysfunctional. I talked to some of my doctor friends about the scope and specifications of a decent primary healthcare center. Let us call it PHC+ (smaller in scale than the CHC, but better equipped). General view was that it should have a general physician, a couple of nurses and technicians. Given the type of cases in rural and semi-urban areas, it should have a lady gynecologist, if possible. The government already takes care of staffing and with thirty (30) new medical colleges being set-up by the current government (there were a total of twelve till now), human resource problem will be addressed to a great extent.

PHC+ should have the facility to collect and store blood for tests (linked to a cold chain), ultrasound and X-ray facility and two-bed casualty with necessary monitors, CPR machines etc. I am already talking of a PHC that most people in UP do not even dream of at the block level. Since the land and building is already available, all the equipment will not cost more than a 20M INR ($250K) based on interaction with those who run small hospitals. 

This equipment can be easily funded by healthcare AI ventures in return for availability of training data. This would mean an investment of $200M by these healthcare AI companies, which at $10-15M per company would be very viable investment for the data access they get. To give you an idea, the population of some of the districts in the state (e.g. Moradabad and Azamgarh) is about the same as that of New Zealand and Ireland. For any AI company in healthcare, it will shorten its development effort by months if not years and also significantly reduce cost of data. 

Here is how it can work. An AI venture with expertise in imaging across a set of diseases funds ultrasound and X-ray infrastructure and technicians in a division. Each division has a population of more than 10M (for reference, this is one and a half times the population of state of Massachusetts where Prof. Barzilay sourced her data from Massachusetts General Hospital). UP Government contracts them for providing diagnostics services. This revenue will make the initial investment even more attractive, at the same time reducing cost of current diagnostics for the Government. For this purpose, the AI venture will have to employ Radiologists – but the cost is rationalized as many of these radiologists would, anyway, have been employed to help with training the software. 

Before privacy fundamentalists go after me, let me clarify that this would not involve any personally identifiable information (PII). No healthcare AI company is interested in your name and Aadhar number and it is easy to mask such data captured at the time of check-in, get the reports and re-link that report to the PII for the patient copy. 

And going back to Prof. Barzilay’s software, every woman in UP above forty can have a mammogram done for under hundred rupees, hopefully, paid for by the Government.



Views expressed above are the author’s own.


Show More

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Check Also
Back to top button