More than patient data: What else is needed to accurately predict surgical outcomes and reduce complications?
Incorporating procedure complexity and surgical skill into outcome models will help reduce patient complications.
Estimates are that within 30 days of general surgery, anywhere from 5.8% up to 43.5% of surgical procedures will have complications.[1] The uncertainty around this value has a lot to do with the type of procedure, where it was performed and how a “complication” is defined, but no matter how you look at it, surgical complications aren’t very rare.
The number of surgical procedures being performed isn’t going down either. An estimated 234 million procedures were performed worldwide in 2004 and this increased to 313 million in 2012.[2] In 2018, 19 million outpatient surgeries (not requiring an overnight stay in a hospital) were performed within US community hospitals.[3]
Together, this means that the number of complications coming from surgery is massive and continuing to grow.
So how can we reduce surgical complications?
There are lots of ideas and approaches out there for how to reduce complications. An approach based on the use of statistical or machine learning (ML) methods is trying to predict, for a specific patient, what the chances are that they will have a complication. As there could be multiple types of complications associated with a surgery, the challenge is to also try and predict the odds for each of the possibilities.
Being able to make good predictions on surgical risks could have a huge impact on improving patient outcomes. Using this information you could be better prepared for specific problems before they happen - perform additional tests, extra monitoring, or more frequent followups. In some cases it may even be possible to take preemptive steps to potentially reduce the chances of certain complications happening in the first place.
Patient data obviously forms the core of the models that could help make such predictions, but not all information is on the patient-side. While hard to measure, variables that capture the skill of the surgeon can make better predictive models. This is something we recently stumbled across for ourselves.
We were trying to make simple ML models for predicting three surgical procedure-specific complications: anastomotic leak following colectomy, bile leak following hepatectomy, and pancreatic fistula following pancreaticoduodenectomy.[4] Without going into the details, these complications can be serious and the surgical procedures they are associated with are involved, with plenty of opportunities for problems to arise. We were comparing our models to logistic regression (LR) models that are the standard method underlying current prediction tools used by surgeons.
From data covering ~200,000 patients, we found that the ML methods did better than the standard LR approach, though marginally so. Not huge differences, but enough to want to look at making your models using something other than logistic regression.
However, what was really interesting is that models that included intra-operative information performed better than those that did not.
To be clear, intra-operative information is data that is acquired at the time of surgery. It includes such information as the surgical approach used, whether reconstruction of part of the tissue/organ was performed, the texture of the organ as observed during the procedure, and similar information. It can even include such things as how many staples were used during the procedure.
So leveraging intra-operative information in models improved our ability to predict adverse outcomes. The complexity of these procedures means that the skill of the surgeon, as well as other external factors, can make a significant difference in how successful the procedure is.
The role of surgical skill has probably always been recognized, but it is now starting to be looked at more directly, with attempts to quantify it. An early example is a 2013 study that ranked the skill of 20 bariatric surgeons. Ranking was done through peer evaluation of a video for each surgeon showing them performing a gastric bypass procedure.[5] The authors then looked at the relationship between these skill ratings and complication rates in their surgical patients. They found that the bottom quartile of surgeons had higher rates of complication and mortality, along with longer operations and higher rates of reoperation and readmission.
Beyond the procedure itself, processes and procedures within the particular hospital, along with how well they are implemented, also factor in to at least some post-operative complications. The skill of the supporting surgical staff are also part of the equation.
Skill makes a difference.
So the real trick is how to capture that information for use in tools that help calculate patient risk, along with coming up with ways to improve the surgeon’s and supporting staff’s skill level throughout their careers. For those in the translational AI space, ways to augment the abilities of healthcare providers throughout the patient’s journey through the hospital and beyond are truly needed.
References & Notes
All references are Open Access and free to the public.
[1] Tevis SE, Kennedy GD. 2013. Postoperative complications and implications on patient-centered outcomes. The Journal of surgical research 181:106–113.
[2] GP Dobson GP. 2020. Trauma of major surgery: A global problem that is not going away. International journal of surgery. 81:47–54.
[3] McDermott KW, Liang L. 2021. Overview of Major Ambulatory Surgeries Performed in Hospital-Owned Facilities, 2019. Available at https://www.hcup-us.ahrq.gov/reports/statbriefs/sb287-Ambulatory-Surgery-Overview-2019.pdf.
[4] Chen KA, Berginski ME, Desai CS, Guillem JG, Stem J, Gomez SM, Kapadia MR. 2022. Differential Performance of Machine Learning Models in Prediction of Procedure-Specific Outcomes. Journal of gastrointestinal surgery: official journal of the Society for Surgery of the Alimentary Tract. DOI: 10.1007/s11605-022-05332-x.
[5] Birkmeyer JD, Finks JF, O’Reilly A, Oerline M, Carlin AM, Nunn AR, Dimick J, Banerjee M, Birkmeyer NJO, Michigan Bariatric Surgery Collaborative. 2013. Surgical skill and complication rates after bariatric surgery. The New England journal of medicine 369:1434–1442.