Is it Time to Turn Healthcare Over to the Machines?

Is it Time to Turn Healthcare Over to the Machines?

Despite all the fear and concern over smart machines ushering in a jobless, dystopian future, I think we may need to turn our health system redesign over to the machines. Humans (at least in the United States, anyway) seem incapable of agreeing on the best actions to improve our costly and inefficient health system.

Stated diplomatically, health indicators for the US are not ‘best in class.’ For example, here’s a link to a list of health statistics captured by the Organisation for Economic Co-operation and Development (OECD) that compares the US to other OECD countries. People in the US top the charts in terms of obesity, life expectancy is near the bottom, it has fewer physicians and hospital beds, and it spends the most per capita by far than other OECD nations.

Unfortunately, grand visions on health system strengthening and redesign are in short supply due to political polarization. Therefore, it may be time to ask the machines what we should do.

Getting Precise With Healthcare

This train of thought stems from a recent webinar that Ayasdi’s Chief Medical Officer, Dr. FX Campion, recorded with SCM World, detailing the important clinical impact that machine learning software is already having on healthcare.

In one case study, researchers at Mount Sinai Hospital ran Ayasdi’s topological data analytical models on the data residing in Sinai’s EHR system on patients with type 2 diabetes. From this, they discovered three subtypes of type 2 diabetes: one subtype was more susceptible to kidney disease and eye disease, the second had higher rates of cancer and cardiovascular disease, and the third had a prevalence of neurological issues. Below, we show a model that represents these subtypes.

Illustration showing the three types of diabetes.

According to Dr. Campion, if clinicians can subtype patients at the onset of type 2 diabetes, they’ll be able to map the best course of treatment. “We are at the dawn of precision medicine,” said Campion.

Machine Learning and the Supply Chain

Much of Dr. Campion’s presentation focused on clinical opportunities; however, it isn’t hard to see the role that the software can play in solving complex operational challenges.

Ironically, other industries recognize the potential groundbreaking role of machine learning in supply chain to a much larger degree than the healthcare industry. Below, we show data from our most recent Future of Supply Chain study, reflecting the percentage of respondents by industry that see machine learning as disruptive and important to their companies’ supply chain strategies.

Chart showing percentage of survey respondents (grouped by industry) who say machine learning is 'disruptive and important'.

Ayasdi’s topological data analysis capability could allow companies to get much more sophisticated with the multi-dimensional data sets within the supply chain. For instance, a traditional regression analysis may describe the rate at which the hardware of a telcom company’s customers will decline, but it wouldn’t be able to show if there were customer subgroups at different stages of decline. Today, telcom companies essentially flood the waterfront when it comes to hardware upgrades. This is wasteful, expensive, and a poor deployment of capital. A cluster graph would more effectively describe the data and outline a precise upgrade plan for the telcom provider’s customers (see the first two images in the graphic below).

Infographic illustrating machine learning topological data analysis (TDA).

Machine Intelligence to Fix the System

The health system in the US has issues both at the clinical and at the operational levels. Unnecessary variation in both clinical pathways and operational processes is at the heart of many of the problems with healthcare and is why costs are so high and quality is so low. It would be really interesting if we could use a machine-learning tool like Ayasdi to analyze not only clinical inefficiencies but operational ones as well. Doing so across the entire value chain would be even more powerful. Could we use data representative of the product, financial, and informational flows across the healthcare value chain, from manufacturer to point of use and care delivery, to highlight needless variation, redundancies, and sources of cost but not value?

The truth is, we probably know where the problems are and what we should do to fix them. However, we lack the will to make hard choices. Insights gained from machine intelligence may indicate the removal of entire supply chain nodes. Right now, any such solution would be fiercely resisted.

Maybe it’s time for us to put our trust in the machines.


Author Barry Blake

More posts by Barry Blake