Title: Learning the Learning Curve of Robotic
Coronary Artery BypassAuthors : Saqib Masroor, MD,
MBA1, Abdullah Nasif, MD1
1Division of Cardiothoracic Surgery, Department of Surgery, University
of Toledo Medical Center
Toledo, OH
USA
Manuscript: The Learning Curve of Robotic Coronary
Arterial Bypass Surgery: A Report from The STS DatabaseDisclosure : NoneWord Count : 1229
Learning the learning curve of robotically assisted coronary artery
bypass grafting is important for the advancement of this technique and
the improvement in patient outcomes. There have been many reports of
single surgeon learning curves.1, 2 But one can argue
that they depict one surgeon’s journey, depicting his or her dedication
to the field and making generalization to other surgeons difficult, if
not impossible.
In this issue of the Journal of Cardiac Surgery, Patrick et al, report
on their investigation of the Society of Thoracic Surgeons (STS)
database for Robotically Assisted Coronary Artery Bypass (RA-CABG)
procedures and the beginner surgeon’s learning curve.3Between 2014 and 2018, a total of 1195 RA-CABGs were performed by 114
surgeons, with 74 surgeons performing <5 procedures and only 9
surgeons performing >25 procedures. The median number of
cases performed was 2. The patient population was younger and relatively
lower risk. The cases included single-vessel as well as multi-vessel
Minimally Invasive Direct Coronary Artery Bypass (MIDCAB) in addition to
Totally Endoscopic Coronary Artery Bypass (TECAB) and there is no
subgroup analysis reported for the different procedures. The authors
conclude that the learning curve for procedural success is overcome by
the 10th case, even though the curve for reoperation
is still steep by the 25th case. Operative mortality
however was similar in the two groups. The authors conclude that surgeon
experience is an independent predictor of RA-CABG procedural success and
that the learning curve consistently flattens after the surgeon’s
10th case. We agree with the first but not the second
conclusion. Here is why!
In 2013, Prof Mohr’s group in Leipzig reported on the learning curve of
minimally invasive mitral valve surgery at their institution over a
17-year period involving 3895 operations performed by 17 surgeons
performing their first minimally invasive procedure, using the
sequential probability cumulative sum (CUSUM) statistical
technique.4 Learning curves were then determined for
total operation times, aortic cross-clamp times, and primary outcomes.
The mean number of operations per surgeon was 189. The authors reported
a learning curve of between 75-125 procedures, with evidence that
surgeons needed to perform more than 1 cases per week to maintain good
results. Importantly however, patient mortality was not compromised
because of the learning curve.
To assess the learning curve involved in performing a task, it is
important that both the task and the tools needed for the task remain
constant. The above publication fulfills both of these criteria. 82
percent of cases were mitral valve repair and 18 percent were mitral
valve replacement. The surgical technique and technology used was nearly
identical and robotic mitral valve procedures were excluded. The
institution had the same leadership over the period, allowing for a very
stable work environment as well as a consistent approach including case
selection, operative technique etc. As much as possible, every variable
was the same, except the variable under investigation-‘the beginner
surgeon’. The same group had reported the learning curve for MIDCAB to
be between 50-100 cases for 8 surgeons at their
institution.5
Now let us analyze the report from Patrick et al.3 In
this report, the task is not the same and neither are the tools. Single
vessel RA-MIDCAB is a less challenging procedure than multi-vessel
RA-MIDCAB, with its associated variety of conduit procedures (such as
bilateral Internal Mammary Artery (IMA) grafting, Radial Artery
T-grafting from Left Internal Mammary Artery (LIMA) to the lateral wall,
or aortocoronary Saphenous Vein bypass procedures). TECAB is a totally
different beast altogether. Grouping all of them in one learning curve
is not a valid assumption. As far as the tools/technique is concerned,
some patients had beating heart surgery while others had arrested heart
procedures, exposing the Left Anterior Descending Artery (LAD) in MIDCAB
is a different task than exposing the lateral wall targets or the
stabilizing the LAD endoscopically. Each one of those steps/techniques
have their own learning curves.
Another shortcoming of this study is the relatively small experience of
most of the surgeons in the study. 74 out of the 114 surgeons in the
study had < 5-case experience. Moreover, it is not clear what
the experience of the surgeons was before embarking on this technique.
In the Leipzig study, surgeons with less than 5 cases were excluded from
analysis and the 17 surgeons had an experience of at least 40 mitral
valve procedures via sternotomy before using the minimally invasive
approach.1
Finally, the definition of procedural success can be debated. It was
defined as an inverse composite of the three primary outcomes -
conversion, re-operation, and major morbidity/mortality. While this
“procedural success” composite showed a flattening of the learning
curve at 10 cases, the reoperation rate was still improving even after
25 cases. A chain is only as strong as the weakest link. If the
reoperation rate is still improving after 25 cases, procedural success
cannot be declared to have been mastered at 10 cases. Further analysis
of the groups of surgeons with < 10 or > 10 cases
reveals the procedural success to be 72.9% and 85.3% respectively.
15% failure of procedural success would not be consistent with
overcoming the learning curve. We assume that surgeons must strive to
continue improving the procedural success until it reaches well into the
90’s percent rate, which would be required for a successful RA-CABG
program.
The major advantage of a large clinical database such as the Adult
Cardiac Surgery Database (ACSD) is the minimization of bias due to its
large number of observations. However, for rare procedures such as
RA-CABG, that advantage is lost. In fact, with such a small number of
observations over such a diverse set of procedures and institutions,
ACSD data is not granular enough to explore an individual surgeon’s
learning curve because there is no control for numerous other variables
at the departmental and institutional level that are not tracked by
ACSD. A high-volume center in a steady-state clinical work environment
controls for most variables that influence clinical outcomes. The only
variable that changes, is the beginner surgeon, and the data thus
obtained is more likely to represent the true “learning curve” of the
procedure.
It is important to have realistic expectations from new technology. Many
beginners would embark on this journey, hoping to master the learning
curve in 10 cases. And when that expectation is not fulfilled in real
life, they might give up altogether on this very useful approach. The
number and frequency of operations are important, not just for the
surgeons, but even more so, for the rest of the operating room team
including anesthesiologists and patient-side assistants. The whole team
can be feel discouraged if they continue to have a learning curve beyond
10 or even 20 cases.
In conclusion, querying the Adult Cardiac Surgery Database of STS may
not be the best way of learning the learning curve of a rare
procedure(s). There is a concern that setting an unrealistically
optimistic expectation of 10 operative cases to master the learning
curve of RA-CABG may be detrimental to the progress of this approach. A
high-volume centers’ experience with multiple beginner surgeons may be a
better representative of the learning curve of RA-CABG and that study
has not yet been done. But based on the learning curves of other similar
procedures and our own experience, it is our opinion that the learning
curve of RA-CABG would be somewhere between 50 and 100 cases for MIDCAB
and another 50-100 for TECAB.