The Fall of Babylon

In August 2023, an innovative telehealth company named Babylon Health made some headlines in the digital healthcare and artificial intelligence (AI)-driven healthcare space. The company was a darling of the venture-backed digital healthcare startup marketplace and once valued at $2 billion. It was backed by the founders of DeepMind and large health insurance companies. Almost overnight, its U.S. shares lost nearly all of their value, and the U.K. subsidiary of the business formally went into “administration.” The administrators sold a large amount of the assets of the company to eMed Healthcare, UK, a new subsidiary of the U.S. company eMed. This overnight reduction of riches to rubble left many people speechless; albeit the trouble in the business model of the company was not completely unknown. Therein lies the crux of this column. While the hyper-excitement around AI in healthcare is taking everyone by storm, the fall of Babylon became a stark reminder that risks abound in this space and it would be wise to tame this giant with caution and adequate wisdom.

Blitzkrieg, Opportunism or Just Plain Greed?

In October 2021, Babylon declared itself the world’s “fastest-growing digital healthcare company.” It piggybacked on a new concept called “SPARC” deals that took the initial public offering market by storm, in which smaller companies could combine themselves with larger entities and go public. Babylon went to the market combining itself with Alkuri Global Acquisition and started to trade under the ticker symbols of BBLN and BBLN.W. The deal size was more than $4 billion. They moved to the U.S. market and stirred up the AI-enabled digital healthcare marketplace. They were received as the poster child of what the new generation of AI-enabled healthcare would look like. Babylon’s claims were as follows:

• Created a world-class data infrastructure to deliver a holistic single view of each individual’s health graph and added AI capabilities into a single platform to enable them to be quickly embedded across their products.
• Integrated more than 100 data sources resulting in access to 80 billion data points, fueling their AI efforts.

Very impressive claims, one has to agree. More data points and more diverse data sources should, in theory, unleash the superior algorithmic power of AI. One might ask: What exactly was Babylon doing for their customers? Who were their customers? Babylon promised to close gaps in care. In the U.S., healthcare is a misnomer because it is about caring for the sick, not keeping people healthy. Little is being done in the way of preemptive and preventive care to keep the larger population from getting sick. The COVID-19 pandemic clearly showed how fragile our overall system of care is despite being a trillion-dollar industry. Babylon promised that it could regularly capture the necessary vital signs of consumers from monitoring devices, add alerts based on the appropriate threshold, make it easy to arrange a telehealth visit with a physician or clinician as needed, and even make a referral to a network of specialists with a fast turnaround and allow consumers to place prescription orders to pharmacies. They could even add care advisors who could create personalized care plans for consumers based on their data. Voila! All gaps in healthcare are closed! If one thinks about the fragmented healthcare system, where we as consumers do not know when to consult a doctor, need to wait for three months to see a specialist after receiving a referral and do not have any personalized care plan, this sounds heaven sent!

Moreover, all of this was free to the consumer because Babylon was tying up deals with large insurance companies, and as long as consumers have insurance coverage, they are under the safety umbrella of Babylon. The powerful AI could make all of this happen without any human intervention. Of course, physicians or clinicians would talk to consumers, but the rest of the workflow was all automated via AI engine and machine learning. A story so compelling that the market, insurers, consumers and NHS in the U.K. all drank the Kool-Aid. Then came the bad news.

In the U.S., 60% of healthcare payments are made by the government, not by commercial insurance companies. Medicare and Medicaid cover 60% of the insured lives in the country. A Pew Research Center survey from 2018 shows that Americans feel the government should be responsible for their health insurance coverage (Figure 1).

So, to expand their business in the U.S., Babylon struck large deals with several health insurance plans, including managed care Medicaid health plans in 12 states, including California. For those who do not understand how managed care Medicaid works, let me explain. Managed care is a payment model in which healthcare expense is capped per capita. Health insurance companies pay a negotiated fixed per capita amount to providers to take care of their members. Providers take the risk. The more they see their patients, they absorb the loss. Depending on the provider’s population health management strategies and preventive or preemptive interventions, these risk-bearing contracts could become profitable. But if not properly managed, this could sink a provider’s business. Babylon Health took such risk-bearing value-based contracts in 12 states from managed care health plans. They believed their AI-enabled product would be able to do provide preventive and preemptive intervention at the right time, and thereby would be able to keep people away from multiple clinics or hospital visits, which are typically high-cost places of service.

This did not turn out to be true. The AI-enabled model for Babylon, which according to them, worked well for the population in the U.K. under NHS, did not perform for the Medicaid population in the U.S. Their risk-bearing “value-based care” contracts drowned them in heavy losses and, eventually, the ship sank. What went wrong? Was their previous claim simply an example of blitzkrieg (as VCs would like to call rapid growth and expansion), outright lies and greed, or simply not understanding the limits of the technology? After all, we have seen such a movie play out before in the form of IBM Watson, where the limits of the AI technology were overestimated by the company. Personally, I think it was all of the above.

The Strange Case of the U.S. Medicaid Population

Medicaid enrollees in the U.S. primarily include five categories: children, pregnant people, low-income adults, elderly adults and people with disabilities. Most of the time, the individual enrollees have complex issues, such as undetected behavioral health conditions, substance use disorders and various other social determinants of health issues, including food insecurity, transportation issues, lack of stable housing, environmental issues (e.g., proximity of toxic pollutants near their homes), etc. These conditions complicate their physical health needs and often exacerbate their disease acuity. The AI model of Babylon and the corresponding cost models they used to derive their risk exposure most likely failed to consider such complexities. From my experience, I can tell that most algorithmic risk stratification of patients failed to consider the plethora of social determinants of health, which are often time deterrents to health, into modeling. This could be partially mitigated by looking at their past healthcare utilization costs, if available. Although that is a necessary criterion for risk determination, it is not sufficient, especially when the parameters are so complex. Therefore, one could argue the business acumen of the leaders of Babylon. Didn’t they know the shortcomings of their AI models? Didn’t they study the idiosyncrasies of the population they are trying to project risks for? It would be difficult to know unless the company or its leaders come forward and divulge their full story. For the time being, they have transferred the responsibility of their patients back to their customers – the health plans – and asked their consumer subscribers to contact their health plan member services for next steps. 

What Does the Fall of Babylon Teach Us?

The fast failure of Babylon in the U.S. surfaces a key problem of Silicon Valley’s famous mantras – move fast and break things or fake it ‘til you make it. AI is a potent solution in the armament of healthcare technologists. It could have transformative effects on our fragmented siloed system of sick care. It could improve efficiency in workflows, perhaps dramatically in certain scenarios such as patient throughputs, and reduce if not eliminate physician burnout owing to documentation burdens of arcane electronic health record (EHR) systems and overly complicated patient privacy concerns. But it has to go many miles and understand many nuances before it could become as effective as digital healthcare companies would like.

(Figure 2 shows that, of the top 50 emerging digital health technology trends, AI in medical decision-making is still in the early phase, and only moderate progress is expected in the coming years.)

We need to remember that the “worried well” and well-off population in this country are not causing the highest healthcare utilization and hence cost. For AI to work effectively in relatively upstream tasks of risk stratifying the population with complex needs, it will have to learn how to weave in many complex parameters of social determinants of health, which will take time for technologists. Let the fall of Babylon be that wake-up call for all of us, and caution, not skepticism and eagerness to learn not turning profit from AI, become our new mantra.