Big Data & Machine Learning in Healthcare Study

Big Data & Machine Learning in Healthcare Study

15 Episodes

To learn more about this study , read the excellent work of David Matheson. Matheson is the now retired BCG senior partner who pioneered a more data intensive way to manage healthcare programs and he basically invented the field of disease management in the 1980s and 1990s. More can be done today, by taking Matheson’s work further. Much, much further.

We help an energy giant develop the national disease management strategy for one particularly costly and burdensome disease. The disease has been singled out as the most pressing problem for this economy and is directly impacting productivity and foreign direct investment in one of the most important sectors. We want to tackle this problem by avoiding the per treatment least cost analyses approach that was pioneering in the 1980s but now outdated. Yet, it is still used today.

Most times, consulting assignments zoom in on cost spikes in the chain of treatment and try to lower them. That is the early work of Matheson at BCG. At the time it was pioneering but now that his approach has been implemented widely, doing more of the same generates little incremental improvement. Yet, it is still the gold standard.

In recent times, economists have used health exchanges to pass the burden of costs to consumers so they are incentivized to help lower the costs. The tool is new but the goal is still the same: lowering costs.

The problem is that when consumers are dis-incentivized to spend money on treatment, the consumer may lower costs by avoiding a treatment they actually need. So healthcare costs go down initially, hopefully, but the overall health of the patient and patient pool is no better off. In fact, relapses from avoiding a needed treatment could cause costs to rise in the long-term.

Doctors attack this problem from a different way. They try to analyze the pathways of a disease to figure out which treatments work best and which can be discarded. They then recommend a treatment path and insurers have to figure out how to pay for it.

So there is this big disconnect. The insurer refuses to pay for treatment x, like physiotherapy, since the benefit is not clear to the system and there is no direct benefit when measuring the impact of the treatment alone. However, the physiotherapy may be necessary for the hip-replacement to work which makes the overall operation a success. Those linkages are hard to understand. They should be understood.

To prove the concept, we built a simple but radically different way to model this problem using Big Data. Rather than being obsessed with the costs spiking, we wanted to see if the return attached to that cost outweighed the costs. The key thing is that the return is not directly linked to the one treatment point. The return for treatment x may be spread all over the system and we need to hunt it down and add it up. The hunting down part cannot be a group of consultants poring over spreadsheets. Even if they could find all the data, the consultants could not understand all the linkages.

I can say that we arrived at this very elegant, and beautiful way to understand the costs, risks, returns and business case around a disease by moving beyond simple 2-dimensional correlations – in English that means moving from plotting data on a x-y graph to modelling the relationships that existed in a 3rd or 4th dimension.

Moreover this is a powerful and simple way to communicate to the board of directors of the client. They get it immediately. Our governing hypothesis is that the wrong treatment decisions are being made since we do not correctly understand the risk, cost, return and benefit of each treatment step for the disease and the overall treatment protocol. It is a way to analyze diseases and extract new insights and recommendations that you have never ever seen before. That, we can guarantee.
Image from A Health Blog under cc, cropped, added text.

Subscribe Share
Big Data & Machine Learning in Healthcare Study