Medical Care as a Commodity
 
            For the non-clinicians trying to design “rational” healthcare systems, or at least decide what constitutes care worth paying for, it is tempting to view the medical visit as a commodity—something that can be produced, assessed, and paid for in a standardized way, much as we buy a five pound bag of flour. One of the major forces behind the way computerized medical records are constructed is the desire to capture measures of this encounter which can be commoditized. The hope is that “big data” and “machine learning” will lead to more rational provision and consumption of healthcare services (commodities?)
            Chen and Asch recently published a perspective on machine learning and prediction and suggested this aspiration is keeping us from seeing where, and when, we might be able to use big data to improve clinical care.[1] Comparing current efforts to extract predictive rules from various data sources to previous efforts to construct cohorts of patients to extract predictive rules, they note the confounding, biases, and missing data encountered in these newer approaches.
 
     “Awareness of such challenges may keep the hype from outpacing the hope for how data analytics can improve medical decision making. Machine-learning methods are particularly suited to predictions based on existing data, but precise predictions about the distant future are often fundamentally impossible.”
 
            They note limitations in current efforts to do things such as predicting 30-day hospital readmission risk leads to compiling even more data, but they note that chaos theory predicts this will not lead to greater certainty. They also note that even if we create a more accurate clinical prediction rule, it may not translate into better clinical care.
 
     “An accurate prediction of a patient outcome does not tell us what to do if we want to change that outcome—in fact, we cannot even assume that it’s possible to change the predicted outcomes…Moreover, many such predictions are “highly accurate” mainly for cases whose likely outcome is already obvious to practicing physicians. The last mile of clinical implementation thus ends up being the far more critical task of predicting events early enough for a relevant intervention to influence care decisions and outcomes.”
 
            I can’t help but note many of the interventions currently recommended to primary care physicians for patients with common ailments such as hypertension and diabetes, assume such interventions will improve outcomes despite a paucity of evidence. We would surely like to believe tight control of blood pressure and diabetes prolongs life and reduces morbidity, and at the extremes we have reason to believe it does. But does that translate to the majority? Consider the persistent debate about the value of a target blood pressure of 140/90 versus 130/80, for instance. Are big data likely to tell us what to shoot for or persuade patients they should aim for?
 
     “Although predictive algorithms cannot eliminate medical uncertainty, they already improve allocation of scarce health care resources…Early-warning systems that once would have taken years to create can now be rapidly developed and optimized from real-world data…”
 
            At the macro level, the errors of big data are not as important, so it is possible to make some rational decisions. But medicine is practiced at the level of the individual, and there the problem of uncertainty continues to drive a lot of behaviors by both physicians and patients. Physicians order more lab tests than they need to make a working diagnosis, “just to make sure.” (Sometimes this behavior is based on the risk of lawsuits, as in “the lawyers demand it.”) Patients sometimes don’t want to wait and see how their problem responds to a trial of therapy—they want high-tech testing and specialty consultation. So if the goal is to commoditize medical care, some way of accounting for the problem of uncertainty must be created.
            For now, I find I spend a good bit of time talking to patients about what “we” know and don’t know about their problems and make recommendations about steps they can take to shift the odds in their favor (maybe.) But I almost always end up pointing out life is lived one day at a time, and the only thing we know for sure is right now. Maybe the solution to our problem is cultural—we must come back to grips with this ancient truth and accept that our modern technological solutions do not offer a way out. I don’t think that is something doctors are well-qualified to address. Maybe we need the preachers and teachers to engage with this.
 
13 August 2017


[1] Chen JH, Asch SM. Machine Learning and Prediction in Medicine—Beyond the Peak of Inflated Expectations. NEJM 2017;376(26):2507-2509. doi. 10.1056/NEJMp1702071.

Further Reading

Activating Patients - The Achilles Heel of Healthcare Reform?
 Studies show 25% of the population is not involved in their healthcare, but reform efforts assume wider application of evidence-based medicine is the key to better value.  Perhaps it is the Achilles' Heel?

More Data on the Value Proposition
Value-Based Purchasing" is a complex program designed to improve hospital quality and outcomes by using financial leverage. A recent study by Ryan and associates suggest it has had minimal effect.

More on Physician Work
The changing nature of physician work is decreasing the availability, and probably the quality of care at a time when demand is increasing. Two recent articles provide data supporting these effects.

The Limits of the Medical Model

The Problem of Scale
Have we lost our moral compass as medical organizations have grown larger?