Variation is not causation

I made a rookie mistake in statistics of the “correlation is causation” genre by confusing variation for causation in the recent JAMA Surgery paper referred to in my last post. I contacted Dr. Timothy M. Pawlik, the lead author of the Johns Hopkins study, who said the following:

"The model is explaining and attributing variation in readmission and not attributing readmission itself to the different domains. The model suggested that only 2.8% of the variation in readmissions was attributable to surgeons. This is different than saying that only 2.8% were the 'fault' of surgeons. A more accurate interpretation would be that only 2.8% of the variation seen in readmissions was attributable to provider level factors. The majority of the variation in readmission was due to patient factors."

He added that some of the 82.8% variation in readmissions attributable (note: attributable doesn’t mean it’s the patient’s fault) to the patient could be modified by better medically managing patients' comorbidities or not operating on some of these patients.

That readmissions can be explained by a single domain or a single person is simplistic. Dr. Pawlik's clarification confirms my original concern that attributing differences in patient outcomes solely to differences in technical quality of surgeons is probably inaccurate, statistically speaking.

Variation is not causation but variation is still a call to action. Regardless of who is to blame for unfavorable outcomes, surgery is a team sport. The incision is just as important as the community care. In this regard, I am certain that ProPublica and I are on the same side. Let’s work together so that we see the whole story behind the numbers.