Automating clinical decisions with predictive analytics

Automating clinical decisions with predictive analytics – Twitter discussion from Terri A Lewis, PhD @tal7291

This is Dr. Lewis’ take on how Appriss is using the electronic health record (EHR) of pain patients to automatically calculate an “opioid risk score” and guide the doctor to prescribe less and/or add a naloxone prescription. (see EHR tool to assess patient risks for opioid abuse)

This kind of automated standardization flies in the face of the supposed intention to individualize treatments. Unfortunately, such personalized care is expensive, while standardization is cheap.

Allow me to point out the obvious. Med records are far too inconsistent, messy, wrong, incomplete for this to be a valid, reliable tool upon which to make clinical decisions that involve predictive analytics. JUST SAY NO.  


When deciding whether or how to TX, ideally you should decide with your Dr if the reduction in the absolute risk outweighs the risks, side-effects and costs of Tx. This should not be left to predictive analytics based on theories for which validity, reliability are not established.

  1. ABSOLUTE risk of a disease is your risk of developing a disease over a time period. We all have absolute risks of developing various diseases which can be expressed in different ways.
  2. For ex, say you have a 1 in 10 ABSOLUTE risk of developing a certain disease in your life. This can also be said to be a 10% risk, or a 0.1 risk – depending on whether you use percentages or decimals.
  3. RELATIVE risk is used to compare the risk btwn groups of people. All sorts of groups are compared to others in medical research to see if group membership increases or decreases your risk of developing certain diseases.
  4. For example, research has shown that opioid users have a higher risk of overdosing compared to (relative to) non-users. RELATIVE risk is the calculation of a ratio of 2 probabilities (2 groups).
  5. An odds ratio (OR) is a ratio of two odds (yep, it’s that obvious) whereas the relative risk is a ratio of two probabilities. The relative risk is also called the risk ratio.
  6. The odds of an event is (the number of events / the number of non-events). This turns out to be equivalent to the probability of an event/the probability of a non-event.
  7. The research does NOT suggest that opioid-dependent users as individuals have an ABSOLUTE risk of overdosing when compared to non-opioid users. If this were true the death rate of people who have their wisdom teeth removed who use analgesics for 3 days would be off the charts.

It’s a mistake to compare the relative risk of population group to your absolute risk as an individual where you have no history. Those who claim a risk score will predict your risk of experiencing difficulties compared to pop means or events that have not occurred are lying.

On the other hand, if you have trouble with an opioid once, the chance is increased that you will have trouble again. Your absolute risk is increased. Outliers are those whose experience or risks don’t predict the behavior or risk of the population averages.

Your diagnosis doesn’t necessarily predict your absolute response to a treatment, nor does the relative experience of a population with the same or a similar diagnosis. Variables within groups and your experience with tx factors must be taken into account.

If we engineer this out of the clinical practitioner:patient exchange we remove discretion and innovation.

With today’s ever stronger focus on finances and need to show a corporate profit, it will be difficult to resist the standardization being forced upon us – or rather, we are being forced into whatever shape their standardized categories take and are then treated by whatever method the algorithm decides.

It seems that algorithms have more power than doctors these days.

2 thoughts on “Automating clinical decisions with predictive analytics

  1. Pingback: Automating clinical decisions with predictive analytics | EDS and Chronic Pain News & Info – The War on Chronic Pain Patients

Other thoughts?

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.