Can AI Save Healthcare? (Part 3)

This picture of a modern Sisyphus pushing a boulder up the mountain is an apt metaphor for the challenges physicians face trying to do the right thing in our current healthcare system.

I’ll start this Conversation with a conclusion:

As long as healthcare is predominately a product-driven system, the increased value from “AI” will largely accrue to those promoting the products rather than to patients or those paying the bills.

The product-driven scenario in healthcare

A product-driven company is one focused primarily on developing great products for their customers. A contemporary example would be Apple with its string of blockbuster consumer products. An earlier example would be Ford and the Model T. A successful product brings significant value to both company and consumer.

How does “product-driven” work in healthcare?

In healthcare, a company can’t just make a great product and then sell it to customers. Companies first have to “sell” it to the FDA for approval, which sets the bar for quality, defines who the customers are, and how it can be marketed. The FDA then manages the company’s compliance through requests, warning letters, fines, and recalls of previously approved products.

Although there are many customers in the system – doctors, patients, families, and hospitals – this regulatory relationship with the FDA has become primary.

The quality of regulation is therefore a significant (if not the most important) factor in the value of American healthcare. Given the high cost and the overall poor status of our health, it is an inescapable conclusion that the current regulatory process is not functioning effectively.

The product-driven scenario and prostate cancer treatment

I’ll explore prostate cancer as an example of how the post FDA product-focused dynamics affect healthcare decision-making. First, a few pertinent background facts:¹

Each year over 160,000 American men are diagnosed with prostate cancer,
Annually about 27,000 die from prostate cancer, and
The 10-year survival of localized prostate cancer is 99%.

The fact that 85% of men diagnosed with prostate cancer die from something other than prostate cancer frames the approach to the disease. The basic medical algorithm for localized prostate cancer includes:

Use the most effective treatment for aggressive cancers,
Avoid over treating unaggressive cancers, and
Prevent unaggressive prostate cancers from becoming aggressive.

There certainly is not one right answer that fits every patient. A recently published study illustrates the challenges in choosing treatment in localized prostate cancer.² This study randomized patients to one of three treatment groups:

Surgical removal of the prostate (an aggressive therapy with potentially severe side effects),
Radiation therapy to the prostate (an aggressive therapy with potentially severe side effects), or
Watchful waiting (A.K.A. active surveillance) to potentially avoid aggressive therapy altogether.

After ten years, just 1% of the men had died from prostate cancer, with no significant difference in the number who had died among the three groups. However, for every twenty-five patients receiving an aggressive treatment, one cancer was prevented from progressing beyond local disease.

Since this study began, the ability to predict which cancers are more likely to progress has improved, so watchful waiting is now an even safer option for well-selected patients.

Four recent developments in the treatment of localized prostate cancer illustrate the challenges of product-driven healthcare:

Robotic assisted prostate surgery is highly touted by those who developed and sell the equipment.³ Although the clinical benefits and cost effectiveness are still questioned in the medical literature, it has been rapidly adopted.
Proton beam therapy is also highly acclaimed, although its benefits versus standard state of the art radiation therapy are still unconvincing. Installation of one of the proton beam units costs between $125 and $150 millionversus the $6 to 8 million for a state of the art radiation therapy machine.^4, ⁵
Stabilization and reversal of early prostate cancer has been demonstrated in patients randomly assigned to follow specific healthy lifestyle changes.⁶
Genetic testing of cancer biopsy material can now help stratify the risk of cancer progression and in certain situations can improve treatment decisions.⁷

To summarize these “advances” in localized prostate cancer:

The first two are based on more expensive products with uncertain benefits over previous treatment. They both have had commercial success.
The third development, adopting certain healthy lifestyles, reduces costs, avoids aggressive treatments, potentially reduces the aggressiveness of early cancers, and improves general health. Despite the evidence, this approach is not included in mainstream medicine’s options for prostate cancer. There is not a product, so who would be backing it?
Genetic profiling will continue to improve the accuracy of prostate cancer decision-making. However, the competing proprietary methods available are difficult to compare, and hype early marketing effectiveness over scientific validity.

So how does a physician, let alone a patient, decide what to do? Well, in the same situation, a surgeon more commonly recommends surgery and a radiation therapist more commonly recommends radiation.

To conclude with this example, the complexity of decision-making in prostate cancer is daunting. Add the emotional component and the challenges of assessing the validity of the advice one receives and decisions becomes become very challenging. In this situation an AI “solution” could easily seem more definitive when it is not. If the past predicts the future, the validity of the algorithm is likely to be determined by the owner of the particular decision that has the most profitable revenue stream.

AI in the Current Healthcare System and a Better Future

We’ll now return to the challenging question of this blog, “Where does artificial intelligence fit into the reality of product-driven healthcare?”

To help answer this question, I have categorized artificial intelligence (AI) into three areas:

1. Algorithms that clearly aid better decision-making –

Such algorithms (mentioned in the previous two Conversations on AI (such as reading EKG’s and X-rays and calibrating ventilator settings) have been improving the technical aspects of medical care for decades and will continue to expand.

2. Algorithms that find correlations in large data sets (the future promise of Big Data) –

While this opportunity is over-hyped, these algorithms hold promise in finding important associations that could propel development of new approaches to treatment. A problem arises that when something is found that isn’t convenient, such as the strong relationship between diabetes and organophosphate exposure, it gets ignored.

3. Algorithms with the clear purpose to enhance the attributes of products to improve marketing positioning –

This final category poses real concerns.

In his book, Bad Pharma: How Drug Companies Mislead Doctors and Harm Patients, Dr. Ben Goldacre meticulously outlines the processes companies use to sidestep scientific rigor to in order to accelerate sales. The misuse of “AI” clearly has the power to add momentum to such deception by side-stepping the standards of rigorous research.

Our dominantly “product-focused” healthcare system is not serving us well and “AI” by itself won’t fix the “product focused” flaws.

So what can be done?

At Switch we follow “decision-focused” health rather than “product-focused” healthcare. “Decision-focused” health empowers individuals by providing them with the tools and catalytic support that will enable them to make better health decisions. Switch “sells” advice with better health as the product. Any major buyer of healthcare services should look seriously at the differences between “decision-focused” and “product-focused”.

Will AI be a key driver as we empower individuals? It depends! As healthcare is currently configured, it won’t.

AI can greatly leverage the availability and customization of cost-effective knowledge. Such a focus is almost universally ignored in the product-centered healthcare system because… what would it sell?

I’ll repeat my conclusion made at the beginning of this blog. But this time, if I’ve done my job, you will understand better that without significant changes in the system, AI will likely make things worse.

Breakthrough To Better,
Carl

¹Prostate cancer statistics
²Robotic surgery for prostate cancer
³Ten year study of prostate approaches
⁴Is proton beam therapy worth it?
⁵“Proton beam therapy review
⁶Prostate cancer patients benefit from lifestyle changes
⁷Genomics help predict prostate outcomes