Predictive Descriptions (July 2018)

One day in November 1987, Ernest Kunnanz came to the emergency room with an excruciating pain in his side [1]. The emergency room physicians worked him up as usual, and quickly came to a diagnosis. Even a medical student would have recognized the telltale combination of colicky back pain and costovertebral angle tenderness: Kunnanz had a kidney stone.

He was referred to Dr. Stephen Edge, who offered to perform a uteroscopy to remove the stone. It didn’t go well. Dr. Edge’s operative report noted that Kunnanz’s ureter was extremely narrow and that multiple dilations were required to pass the ureteroscope up to the renal pelvis. in the process, the stone got pushed up the ureter and into the kidney, where it couldn’t be removed. Dr. Edge put in ureteral stent, which he took out two weeks later. Three weeks after that things still hadn’t gotten much better, so Kunnanz went to another urologist for a second opinion. A follow-up ureteroscopy revealed two perforations and a submucosal fistula in Kunnanz’s upper ureter. The kidney and ureter had been irreversibly damaged. The following March, about five months after his initial visit to the hospital, Kunnanz had surgery to remove the compromised kidney and ureter.

It wasn’t too long after that final operation that Kunnanz sued for medical malpratice, claiming that, among other things, Dr. Edge had negligently performed the initial ureteroscopy. The critical question, as is typical of malpractice cases, was whether Dr. Edge had breached the standard of care and caused Kunnanz’s injury. So, the plaintiff’s attorney got right to the heart of the matter. He asked the expert witness: “Did Dr. Edge’s failure to follow accepted standards proximately cause the loss of Mr. Kunnanz’s kidney?”

The defense lawyer raised an objection, and the judge sustained it. The witness was never allowed to answer the question [3].


Roughly speaking, there are two kinds of data: descriptive and predictive. The difference is pretty much just what the names say; descriptive data tells you what happened in the past, where as predictive measures give you some idea as to what’s going to happen in the future. A descriptive figure is how many people contracted H1N1 flu last year; a predictive one is the probability that H1N1 will be the dominant strain next season. One looks back, the other looks forward.

There’s the same sort of distinction between the way that lawyers and doctors think about malpractice issues. Lawyers (and judges and juries and the rest of the legal system) are exclusively focused on figuring out what exactly happened in the case being tried. There’s not too much weight given to how likely the outcome was, or whether it’s bound to happen again. Those notions are considered too speculative to definitively tell us what happened on the day in question. Stepping into the courtroom seems to turn everyone into Joe Friday: just the facts, ma’am.

Doctors, on the other hand, take a more expansive view. We’re not concerned with just one case; we’re trying to optimize performance over hundreds, if not thousands of interactions. We have the luxury — or perhaps the obligation — to recognize that any given outcome, good or bad, might just be good (or bad) luck, and that it’s what happens over the course of a larger sample size that truly counts. That’s true whether we’re talking about which patients are at greatest risk of receiving negligent treatment, or what insurance rates should be or which doctors might benefit from remediation. It’s evidence-based practice in action. Everything is future oriented.

Maybe a better way to illustrate that difference is to think about the guiding principles that animate the two professions. The courtroom has “innocent until proven guilty”, which is pretty stern guidance on how to think about an event that’s already come and gone. Doctors, meanwhile, have “first, do no harm” — a hopeful prediction for the future.

But because they’re busy looking in different directions, “innocent until proven guilty” and “first, do no harm” don’t always get along. The conflict is especially acute in the context of how malpractice trials handle evidence that the defendant doctor had previous instances of failing to meet the standard of care. Doctors understand this to be powerful data: those who’ve had malpractice events in the past are very likely to have more in the future, a very small percentage of doctors account for a staggering portion of patient claims, and on and on. The analysis deserves some nuance and risk-adjustment, but it’s certainly reasonable to think that individualized malpractice data can point the assessment of a doctor’s performance in the right direction. It’s a powerful predictive tool.

Yet that information rarely sees the inside of a courtroom. The rules of evidence that govern what testimony juries and judges can consider frame malpractice history as more likely to poison the well than uncover the truth [2]. The crux of the problem — from the legal standpoint — is that, regardless of what a series of old malpractice claims might say about a doctor’s tendencies, they don’t directly speak to what happened in the particular case before the bar. The data is descriptive, the courts seem to acknowledge — but not of the right thing.

But that’s just it. In reality, all data is both descriptive and predictive, depending on how it’s used. The distinction is entirely contextual. Malpractice data isn’t useless in the courtroom just because it can’t definitively say whether the standard of care was breached in any given case. It can still accurately describe what happened in the past, and it can also make probabilistic assessments as to the likelihood of similar events occurring going forward. Judges and juries should never be compelled to find a verdict solely on the basis of this evidence; but it’s unclear what harm would result from allowing them to consider and balance it against all the other information presented at trial. “Iinnocent until proven guilty” or “first do no harm” are both valuable conceits. Why ignore either of them?


About a week before Kunnanz came in with his kidney stone, Dr. Edge performed a ureteroscopy on another patient, Cathy Kunkel [1]. It didn’t go well, either. Much like Kunnanz, Kunkel’s ureter was perforated and she ended up needing her kidney to be removed. She later sued Dr. Edge for negligence, and ended up settling out of court. At Kunnanz’s trial, his lawyers attempted to introduce evidence of Dr. Edge’s mishandling of Kunkel’s case, but the rules of evidence prohibited it. Whatever insights the Kunkel affair might have held never reached the jury’s ears. In effect, the court’s rules ensured that that data was neither descriptive nor predictive.




[2] Federal Rules of Evidence ss. 401, 403

[3] Ginsberg, Marc. “Good Medicine/Bad Medicine And The Law Of Evidence: Is There A Role For Proof Of Character, Propensity, Or Prior Bad Conduct In Medical Negligence Litigation?, 63 SCL Rev. 367 (2011).” (2011).