Ronald F. Martin, MD, FACS, Consulting Editor

In some respects, the beauty of being a surgeon is that one does not always have to know exactly what is actually going on, but one must always know what exactly must be done. That simple paradigm may seem simple enough at face value; it is not.

The topic of what do we know and how do we know it was the genesis of this issue of the Surgical Clinics. It began with a conversation that I had with one of our Guest Editors, Dr Bingham. I had just returned to Madigan Army Medical Center as a civilian surgeon years after having worked there while I was still actively in the military. The conversation wandered around several topics relating to how things have changed. One of the topics that really intrigued me in that conversation was how we were handling integrating new concepts in general. In particular, large data sets and real-time analysis and implementation. This morphed into the broader question of what we know and how do we know it is true at this time. As a result of that conversation, we set out to collect a series of ideas about what is evidence, how do we analyze it, and how do we make decisions based upon that. Dr Bingham and his co-editors, Drs Eckert and Eckert, fleshed out that list of topics and have assembled a group of people very well suited to address these questions.

While not necessarily always correct, it is extremely common to view this discussion of how we surgeons as a community gather information and put it to use, particularly as it relates to the use of technology, through a generational lens. As is mentioned in the issue, that is not always a valid construct. There are certainly members of our community that have never known the “predigital” age, but many of those whose time on Earth predates the Internet have well-developed digital skills. In fact, the entire digital age platform was developed by people who predate the “digital age.”

The development of real-time searchable digital information has been enormously useful to countless persons in all stripes of life. This utility comes with some serious caveats, however. As is described in this issue, there has been a marked expansion of sources of information for the interested person to consider in just surgery alone. I would say that in many respects this mirrors the changes in broadcast capability over the same timeframe. In the 1960s and 1970s, there were 3 major national broadcasting entities in most communities; four, once public broadcasting began its mission. The obvious downside to that framework was that those few networks essentially decided what content was relevant for all (complete with commercial biases intrinsic to the model). The upside was that most people had similar streams of information to consider and a common platform from which to discuss. For viewpoints that did not “fit well” into the mainstream networks, one had a robust print world to turn to on an individual basis. Since the advent of cable, followed by Internet, the number and variety of information sources have expanded such that is unlikely that any two people have a truly common source of input. The upside, of course, is the democratization of ideas and the ability to advance them. The downside is the democratization of ideas and the ability to advance them.

The filterless expansion of what is essentially a publishing platform (reproducing, amplifying, and transmitting information for monetary gain through subscription or advertising)—amorphous though it may be—has created an environment where all can be heard; even robots. Yet, the phenomenal increase in noise along with signal within the bandwidth does not necessarily translate into the most efficient conveyance of reliable information. While this concern is most frequently discussed today in terms of political practices, the general problem applies to the scientific community as well. We have our own share of misinformation/disinformation spread as well. Even under the most ideal circumstances, there are biases regarding the analysis of data and opinions that are permitted in the “public square” that very much limit our discussions.

One of the many byproducts of the digital age, specifically related to massively enhanced processing power, is the gathering, storage, evaluation, and distillation of large data sets. As someone who financed much of my undergraduate and medical school education by developing and writing software for various research and business purposes, I greatly appreciate the ability of computing to handle large databases. One thing I took away from that phase of my life was the need to use technology to solve the problem of my customer, as opposed to writing software that I found interesting and then have the customer find a way to use it. There is absolutely room for developing technology whose use is unclear at the moment as well. However, when it comes to real-time patient care, a certain degree of pragmatism is frequently required.

Large database analysis commonly comes up against the problem of scope. The capability to “zoom in and out” frequently changes the analysis. Missing data, variably collected data, and just plain wrong data all require analytical tools that are incompletely successful at avoiding errant conclusions. There has been a huge increase in the amount of presented and published material that is based on large data sets and analysis. I have significant concerns that much of this is not well understood by the presenters. As I have written before, when large data-based presentations are made and one asks the presenter even basic questions about how to interpret to data and/or the conclusions rendered, the stock response is, “Thank you for that excellent question. However, because our analysis is based on registry data, we do not have the granularity to address your concern.” Furthermore, it not uncommon for people to present material based on data-mining for a condition that they have never ever seen or treated.

Another concern is that presentations based on large data set reviews often achieve statistical significance based on the fundamental tenets of statistics when using “large N” models. We frequently seem to forget the P value represents the likelihood that what we are seeing is the product of random chance (apologies to actual statisticians for the crudeness of the explanation), and therefore, the possibility that the correlation we infer is not accurate. Large N models frequently reduce the shape of the distribution curve and therefore reduce P value. The potential downside of that is that many large N models do not really compare things that are truly alike; as such, the actual “N” may be more elusive than one would wish. The net result of this lack of granularity and conformity of input can yield some odd results. I recall a paper that concluded that the use of antibiotics reduced length of stay for inguinal hernia repair from 14 days to 12 days. Common sense should dictate to us that any patient undergoing inguinal hernia repair in this era who had any “length of stay” beyond an outpatient procedure is already an anomaly and cannot be part of a data set to test one treatment versus another for most inguinal hernia repairs. We must understand when data analysis gives us conclusions that are incompatible with reality. In the example just given, it is fairly obvious that something is awry (other than the paper was published in the first place). In other situations, it may be far more difficult to spot an errant assessment.

For those of us who trained in surgery prior to the advent of the digital age and all its accoutrements, we have to embrace the use of data and digital options to manage our responsibilities in real time as well as educate ourselves in an efficient manner. For those who only know the Internet age, particularly those in Generation Z or I-generation, it is imperative that they also learn how to process patients in an “analog” way as well. Just as all quantum functions collapse upon observation to a single (hopefully) data point, all patients eventually narrow down to one, or at least a very small number, of surgical problems out of the universe of possibilities upon presentation. While big data is excellent at producing the list of possibilities, analog evaluation of the patient, at present, usually prevails at addressing the patient’s clinical concern. And while artificial intelligence (AI) shows promise (or dread) in many areas, the reliability factor of truthfulness—either while ascertaining subtle information or relaying information—is lacking. For some time, I think the surgical world will be safe from AI replacement, although that may not last forever.

The ability to regard data and data evaluation as a platform to support understanding rather than as a final arbiter of understanding is critical. We must always interpret results in light of cumulative experience and observable facts. That said, the earth really does look flat from certain perspectives. We don’t wish to reinvent the Flat Earth Society. Nor do we wish to throw away everything we think we know based on meta-analysis. Wisdom will more likely come from learning to embrace both.

To address the need to educate our younger colleagues in the art of patient “analog” care is a bit beyond the scope of this particular issue. However, I am deeply indebted to our Guest Editors, Jason Bingham, Matt Eckert, and Carly Eckert, for assembling the topics and the contributors to give each of us an excellent platform to understand how we gather, store, manage, and validate data as well as how we can use large set data analysis, machine learning, AI, and technology to maintain and improve our education and the care of our patients.

Article infoIdentification

DOI: https://doi.org/10.1016/j.suc.2023.01.002

Copyright

© 2023 Published by Elsevier Inc.

ScienceDirectAccess this article on ScienceDirect Related Articles