“Number of medical students pursuing surgery specialty drops by half”

That was the headline in a September 10 Becker’s ACS Review article. The first sentence of the piece was more specific, “Only 4 percent of medical students surveyed in 2018 said their chosen medical specialty is general surgery, compared with 8 percent in 2016, according to Medscape's Medical Student Life & Education Report 2018.”

This caused some consternation among general surgeons on Twitter. I tweeted, “Interesting. Lifestyle is finally catching up to us. I think it will get worse.”

How to interpret the literature: A new series of posts

The Salty Statistician will be a recurring feature of this blog wherein we ask statisticians in medicine to break down articles from the surgery literature and assess whether the reported conclusions are supported by the data. Let’s look at this study:

Groh MA et al. Is Surgical Intervention the Optimal Therapy for the Treatment of Aortic Valve Stenosis for Patients With Intermediate Society of Thoracic Surgeons Risk Score? Annals of Thoracic Surgery.

The authors attempted to address the question of whether aortic stenosis patients deemed “intermediate risk” [IR] for surgical aortic valve replacement [AVR] are best treated with open surgery or transcatheter AVR. The authors looked at 1,144 patients who received surgical AVR from 2008-2014 at a single center focusing on the 620 “intermediate risk” patients. At the end of the follow-up period, 72 had died.

Unfortunately, major methodological issues undermine the paper’s conclusions. Fortunately, this provides an excellent teaching opportunity.

First, the authors inappropriately used logistic regression to analyze independent predictors of mortality. Logistic regression treats the outcome as a simple “Yes” or “No” variable, while ignoring the time-at-risk. This study included patients treated over a six-year period (2008-2014) who therefore have substantial differences in the amount of time at risk. Consider the following hypothetical patients.

Patient A treated in 2008 and died in 2014 surviving six years after surgery. The logistic regression model simply counts patient A as “dead.”

Patient B treated in 2014 and alive in 2017 but dies in 2018, after the data were analyzed and the paper published. He survived four years after surgery and in the logistic regression model, counts as “alive” since data were analyzed in 2017.

Patient A lived for six years after surgery, but counts as “worse” in the analysis than Patient B who only lived for four years because of the time at which the data were “frozen” and analyzed. Of course, this is unavoidable in long-term outcomes studies, but one must choose an appropriate statistical method that accounts for time-at-risk.

Cox proportional-hazards models are more appropriate for a long-term survival outcome than logistic regression. When building a Cox model, one specifies both the current status (i.e., alive/dead) as well as an amount of follow-up time. For example, Patient A is “dead” with six years of follow-up; Patient B is “alive” but with only three years of follow-up. This provides a proper assessment of how strongly the independent variables are associated with risk of mortality while accounting for the unequal follow-up time.

Second, the authors state their data supports the conclusion that “SAVR is the optimal therapy for most of the patients” in the IR group in comparison to TAVR. However, their paper lacks any data on outcomes in IR patients who were treated with TAVR. Why the authors believe presenting data from a series of SAVR patients is sufficient to claim that SAVR is the “optimal therapy” absent any comparison data on patients treated with TAVR is unclear. Randomized controlled trials have more appropriately compared SAVR and TAVR in the IR population. Link here and here.

Which patients should receive surgical AVR versus transcatheter AVR is a good question, but to answer it, the paper used an incorrect approach.

Final Rating (1-5 Scalpels): 1 Scalpel - significant methodological issues

This issue of the Salty Statistician was written by Andrew Althouse (@ADAlthousePhD), currently an Assistant Professor of Medicine at the University of Pittsburgh as well as Statistical Editor of Circulation: Cardiovascular Interventions.

We intend this series to focus on work that is perceived to have a high impact on clinical practice, so we welcome reader suggestions. If you have a paper that you would like to see reviewed as part of the Salty Statistician series, please tweet @Skepticscalpel or @ADAlthousePhD or email SkepticalScalpel@Hotmail.com. We cannot promise that all submissions will be reviewed in this space, but we will do our best.