In the News: Closing the Stable Door After the Horse has Bolted

Thanks to Google Alerts, I recently came across an article published by the folks at MedPage Today reporting on an algorithm developed by Dr. Cromwell at The University of Iowa  Hospital and Clinics touted to cut surgical site infections (SSI) by 58%. Results of that magnitude certainly pique my interest, so I did some internet sleuthing to learn more, and eventually came across a more thorough review on the WSJ CIO Blog and an unpublished manuscript describing their approach.

Although predictive analytics for surgical site infection isn’t in itself novel (NSQIP developed a patient risk assessment tool, for example), the novelty of Cromwell’s approach seems to be rooted in the strength of their data analytics and committment to incorporate detailed intraoperative factors into the assessment.

From the WSJ post:

“During surgery, as the surgeon closes up the patient’s abdomen, the circulating nurse logs onto a Web portal for the software and enters real-time data such as patient blood loss during the operation, the wound classification and whether it was contaminated, said Dr. Cromwell.”

The theory is that if caregivers are armed with this information before patients leave the operating room, doctors can create a plan to reduce the subsequent risk of infection by “altering medication or using different techniques in treating the wound.” The reader is left to wonder what those medications or techniques might be, especially when the physiology of wound healing dictates that infection prevention strategies are severely impaired after the incision is closed due to fibrosis and encapsulation of infectious material (Surgical Infections, March 2013).

Regardless, I think Cromwell is on to something here, because the events of operation itself represent the best opportunity to reduce risk. But assessing these predictive variables at closure is a bit like closing the stable door after the horse has bolted. A better approach would be to make sure the stable door is closed to begin with.

As an illustrative example, consider the Cromwell’s results: factors associated with increased risk of infection include certain patient zip codes, ostomy creation, higher wound class (i.e. wound contamination), higher ASA score, higher total # of procedures (per patient), surgical apgar score, and open procedural approach. We highlighted the importance of wound contamination in an earlier post, and would further maintain that of these factors, wound class (i.e. wound contamination) is the only one under direct control of the surgeon. So why aren’t we doing more to prevent wound contamination? We could start by directing powerful analytical tools like those developed by Cromwell towards the operative field.

Is Surgical Site Infection a Unifactorial Problem?

A quick look at the search term “surgical site infection” on Google Trends reveals that has trended positively relative to the Health category  over the past decade. And no doubt a large proportion of the search results have used the term “multifactorial” to describe the problem (see below; blue line = “surgical site infection;” black line = Health category).

Google Trends time ssi

But is SSI really a multifactorial problem? In an era of unfettered access to data, we risk measuring the impact of too many variables–jeopardizing the core assumption of independence central to the statistical techniques employed to analyze the data.  As a result of this “data fatigue,” over-fitting and information gain are real concerns. Furthermore, this data rich environment can make us lazy–after all, why bother collecting your own data if others have already curated a ton of it for your consumption?

What if this focus on collecting more data distracted us from collecting the right data? When we started exploring surgical site infection as Fellows in the Stanford Biodesign Program, we were immediately drawn to the link between wound contamination and SSI. Nothing new here–we recognized that this was well-established in the literature over the years, quantitatively by bacterial concentration risk thresholds of 10^4 CFU/g  (Krizek, 1975), and qualitatively through use of the NRC Wound Classification. But nonetheless, we we struck by the flood of results over the years demonstrating a strong, consistent link between the perceived (“subjective”) degree of wound contamination and the observed SSI rate.

wound class

Alas, the NRC Wound Classification is a bit too subjective and categorical for detailed risk model development, so we dug deeper. A few researchers (Fa-Si-Oen, 2005, Horiuchi, 2010, and Waldron, 1983) went as far as to measure the presence (and sometimes degree) of bacterial contamination present at the wound edge at the time of closure. The result of these studies are pooled and summarized below.

contamination graph

The implications are sobering:

  • 50% of abdominal wounds are contaminated during surgery
  • 20-33% of contaminated wounds result in infection
  • Remarkably, when there was no bacterial contamination present, there is correspondingly no risk of subsequent infection. Perhaps SSI is indeed a unifactorial problem!

Around the same time, the results of the CHIR-Net study were published, representing one of the most comprehensive studies evaluating the effect of wound protection devices on SSI rates. Might bacteriology results be predictive of these SSI results? Fortunately, a study by Mohan et al in 2012 evaluated the rate of bacterial contamination in a similar patient population on the exposed and protected surfaces of the wound protection device. As illustrated below, Mohan’s contamination rates, in view of the contamination studies mentioned above, effectively predicted the results of the CHIR-Net study.

alexis back calculated

To be entirely fair, post-hoc analyses like this can be subject to bias, but the results are highly suggestive, and certainly worthy of follow-up consideration and research. To that end, Prescient Surgical is currently sponsoring one of the most comprehensive studies of wound bacteriology to date, and we look forward to learning how the results expand our scientific knowledge in this arena. I look forward to updating this community with the results this summer.