Quantification of surgical workflow during robotic proctectomy

Surgical workflow assessments offer insight regarding procedure variability. We utilised an objective method to evaluate workflow during robotic proctectomy (RP).


| INTRODUCTION
Assessment of surgical workflow is critical for evaluating surgeon skill, case complexity and surgery variability.The scarcity of procedure-specific annotated data hinders the quantitative analysis of surgical workflow, consequently limiting the ability to provide targeted feedback and objective insights. 1Subjective operative reports are currently the primary source of information regarding surgical workflow.As part of an overall movement towards optimised surgical care, there is a need for more objective and standardised operative reporting.
3][4][5] Using descriptors such as 'hostile abdomen', 'extensive scar tissue', or 'extremely difficult dissection' are inherently subjective.There is an ongoing effort towards more objective reporting of operative events during surgical procedures.
For example, synoptic operative reports are now required by the American College of Surgeons and National Accreditation Programm for Rectal Cancer. 6rrently, console time (CT) and total operative time (TOT) are commonly used as quality metrics to assess robotic surgeon proficiency, but these metrics do not accurately reflect other factors that influence surgical workflow.For example, CT and TOT may be affected by patient factors (male sex, obesity, prior radiation), variations in pathology (bulky tumours), surgical team expertise, or involvement of surgical trainees (dual-console procedures).There is therefore a need for more objective, surgeon-specific data pertaining to surgery workflow.Robotic surgery, in which computer hardware and software are situated between the patient and surgeon, allows a unique opportunity for recording and analysing objective data reflective of surgical workflow during live surgery.
Machine learning-enabled metrics, objective performance indicators (OPIs), can be captured from surgical video and robotic system data.8][9] A specific OPI, step duration, enables accurate time-stamping of critical steps within a given procedure and provides a tool to evaluate surgical workflow and its variability across cases.A prerequisite for step-specific analyses is the ability to annotate surgical videos so as to identify discrete, functionally relevant surgical steps.An annotation card can be used to label start and stop times for steps during videos of robotic surgery.It has been utilised by Ghodoussipor et al.
for partial nephrectomy, 10 Tousignant et al. for sleeve gastrectomy, 11 and Metchik et al. for lobectomy, 12 but no such work has been published for proctectomy.3][14] Although there is no universally accepted annotation card for robotic proctectomy (RP), we have identified 21 steps during RP, and have previously published regarding our annotation process. 9 this study, we objectively describe step-specific workflow and transitions within RP.Additionally, we identify critical steps that highly correlate with CT and TOT, as well as steps more likely to be convergent (varied steps precede, but few follow), divergent (few steps precede, but varied follow), or nodal (varied steps precede and follow), providing novel insights into workflow variabilities of RP.This segmentation and objective characterisation of surgical workflow will aid in the future development of synoptic operative reporting.

| MATERIALS AND METHODS
The protocol of this study was approved by the institutional review board (IRB) of Emory University (IRB #00111214).
Thirty-one RPs conducted at Emory Hospitals between 2020 and 2021 were included in this study.All procedures were performed by colorectal surgeons using the da Vinci Xi platform.An Intuitive Data Recorder (IDR) (Intuitive Surgical) was utilised to capture the endoscopic video synchronised to robotics system data.No changes to patient management, surgical decision, or surgeon assignment were made due to IDR recording.
A procedure-specific annotation card was developed using a Delphi method.After reviewing other scorecards for assessing colon and rectal surgery, as well as pertinent literature, our group of six colorectal surgeons went through 6 Delphi rounds to create a final annotation card.The annotation card was trialed during a video review of 4 RPs in an auditing phase, and modifications were made to improve consistency and accuracy.The final annotation card for this study was utilised by professional video annotators to determine the start and stop times of individual RP steps.
The 21 individual steps of RP include inferior mesenteric artery (IMA) dissection, descending and sigmoid colon mobilisation, splenic flexure mobilisation (SFM) and posterior mesorectal dissection.To allow for a clear beginning and end to each step, we defined 'start' and 'stop' parameters.As an example, the start parameter of IMA dissection is 'the first instrument interaction with the mesentery surrounding the IMA'.The stop parameter is an IMA that 'is skeletonised and ready for transection'.Throughout the procedure, a surgeon may skip individual steps, perform them more than once, or revisit them multiple times.
We evaluated the time spent on each step, number of times each step was visited, CT, TOT and step transitions.We classified specific steps based on the number of visits and their relationship to other steps.Nodal steps describe a step with at least 10 different steps preceding and 10 different steps following.Divergent steps were defined as a step with fewer than 10 different steps preceding and more than 10 different steps following, and convergent steps were defined as having more than 10 different steps preceding and fewer than 10 different steps following.
Statistical analyses were conducted using SAS Version 9.4.Categorical variables were presented with counts and percentages and the numeric variables with median and interquartile range or mean and standard deviation (SD), as appropriate, based on normality.Individual step time and step visit frequency were reported using mean and SD, and their associations with CT and TOT were measured using Spearman's correlation.Circos plots, widely used for visualising genome data, were employed to visually represent step transitions. 15 achieve this, we constructed a matrix in which rows denote the frequency of divergent steps, while columns represent the frequency of convergent steps.The Circos application was employed to transform this matrix into chord diagrams.These plots facilitate visual variation of large data in a relatively small space by placing steps in a circular layout and drawing unidirectional links (ribbons) that represent transitions between them.

| Step visits and step times
The mean number of step occurrences per procedure was 49.0 (�20.3).Posterior mesorectal dissection and right lateral rectal dissection occurred during all procedures and required the most step visits, with a mean of 8.7 (�5.0) and 7.6 (�5.2) steps per procedure, respectively (Table 2).On the other extreme, inferior mesenteric vein (IMV) dissection and IMV ligation occurred in only 4 procedures, with a mean of 1.6 (�0.9) and 1.0 (�0.0) step visits per procedure,

| Step correlation with CT and TOT
To elucidate the surgical steps most predictive of operative time, we calculated correlations of individual steps with CT and TOT.The following showed strong correlations with CT and step duration: IMV dissection and ligation (ρ = 0.60 for both), lateral-to-medial SFM (ρ = 0.63), left lateral rectal dissection (ρ = 0.64), and mesorectal division (ρ = 0.71).CT also correlated strongly with medial-to-lateral and supracolic SFM visit frequency (ρ = 0.75 and ρ = 0.65).With regard to the TOT, there was a strong correlation for initial exposure time (ρ = 0.60), as well as for medial-to-lateral SFM and supracolic SFM visit frequency (ρ = 0.67 and ρ = 0.65, respectively) (Table 2).

| Step transitions and circos plots
To determine the spectrum of transitions, a subset of all the RP steps that met specific requirements were defined as nodal (varied steps precede and follow), divergent (few steps precede, but varied steps follow) or convergent (varied steps precede, but few follow).Nodal steps included medial-to-lateral and lateral-to-medial mobilisation of the descending and sigmoid colon (Table 3).On the other hand, the medial-to-lateral mobilisation of the rectum was identified as a convergent step (13 types of steps precede, 7 follow), and transection of the rectum (6 types precede and 12 follow) was classified as a divergent step.
While the nodal, convergent, and divergent step categories offer a broad framework for assessing the dynamics of RP, a Circos plot can offer a visual representation and frequencies of transitions from one step to another.Figure 1 shows the Circos plots for one of the nodal steps, while Figures 2 and 3  T A B L E 2 Surgical step times and step visits, with correlations to total operative time and console time.

Step name
Step time (min) Step visits  In this study, we quantitatively captured surgical workflows, including step transitions, thereby providing novel insights into the dynamics of RP.We also correlated two objectives, scalable metrics (step time and step visit frequency) with more traditional metrics of proficiency (CT and TOT). 16,17ilising the Delphi method, we developed a procedure-specific annotation card for RP, which allowed us to delineate each of the 21 steps of the operation.9][20] An annotation card was used to objectively define F I G U R E 1 Lateral-to-medial descending and sigmoid colon mobilisation (step 11, indicated by the arrow) is observed as a nodal step across robotic proctectomy cases.Circos plot illustrates convergence and divergence of all steps with respect to step 11.It is predominantly preceded and followed by a left lateral rectal dissection (step 15).Numbers correspond to the surgical steps as described.
correlate with CT and TOT.Steps were defined as nodal, convergent, or divergent in order to understand the flow of operations and detect trends in steps as they relate to one another.
Rectal dissection in all four quadrants (anterior, posterior, right, left) required the highest number of step visits, with posterior and right lateral dissections requiring significantly more visits than the others.As the majority of these procedures were for cancer, it is not surprising that these steps required many visits.Presumably, surgeons used meticulous attention and technique during these steps, with a focus on achieving excellent radial and distal margins.
Regarding the laterality of rectal dissection, it is noteworthy that right lateral dissection visits were significantly higher than those on F I G U R E 2 Medial-to-lateral mobilisation of the rectum (step 9, indicated by the arrow) is observed as a convergent step across robotic proctectomy cases.The circular plot illustrates the convergence and divergence of all steps with respect to step 9.It is predominantly preceded by IMA dissection (step 2), and followed by IMA dissection (step 2) and posterior rectal dissection (step 12).Numbers correspond to the surgical steps as described in Figure 1.IMA, inferior mesenteric artery.Limitations to this study include a limited sample size from a single institution, the inclusion of both single and dual console cases, and the heterogenous nature of surgical indications.Even with these limitations, this study provides critical insights into the surgical workflow and dynamics of RP.As such, future studies are needed to address these limitations and build upon these promising results.
Through the utilisation of a procedure-specific annotation card for RP, our study identifies a novel approach to quantitatively define surgical workflow.We identified step time and step visit frequency as useful OPIs to track surgical workflow, and found correlations with traditional metrics, CT and TOT.We found that the duration of SFM, mesorectal dissection, left rectal dissection and IMV dissection are strong predictors of CT, while initial exposure time is a strong predictor of TOT.SFM step frequency was also a strong predictor of CT and TOT.Additionally, we identified mobilisation of the descending and sigmoid colon, medial-to-lateral rectal dissection and rectal transection as nodal, convergent, and divergent steps, respectively.These findings will be instrumental in informing future OPI research on critical aspects of robotic colorectal surgery.Our study provides an objective method to measure surgeons' progression through cases, thereby serving as a foundation for targeted interventions aimed at shortening the learning curve and enhancing clinical outcomes.

6 of 9 -F I G U R E 3
GILLANI ET AL.the left side.This is likely due to camera placement in the right lower quadrant of the abdomen during RP, leading most surgeons to dissect right to left for more than half the dissection.Colorectal anastomosis, mesorectal dissection, and posterior rectal dissection required the longest mean times.Many (71%) of our anastomoses were performed intracorporeally, which may require more CT.Mesorectal dissection requires precise use of a monopolar instrument and vessel sealer to achieve a linear, perpendicular distal margin, while avoiding bleeding or rectal wall injury.The posterior rectal dissection is a critical step during total mesorectal excision (TME).This step requires multiple camera and retraction adjustments, and careful, deliberate bimanual dissection to develop the golden plane behind the mesorectum, but anterior to the presacral nerves.CT and TOT may not accurately reflect surgeon skill as there are intra-operative confounders, including surgical team proficiency and concurrent training (dual-console procedures), that may influence surgical times.Some variations in surgical procedures can also influence CT and TOT.For instance, some of our patients did not Rectal transection (step 17, indicated by the arrow) is observed as a divergent step across robotic proctectomy cases.The circular plot illustrates the convergence and divergence of all steps with respect to step 17.It is predominantly preceded by division of the mesorectum (step 16), and followed by medial-to-lateral descending and sigmoid colon mobilisation (step 10) and colorectal anatomosis (step 21).Numbers correspond to the surgical steps as described in Figure1.requireanastomosis, while others required both anastomosis and diverting ileostomy.Additionally, patient-related factors such as obesity, adhesions, history of radiation, or more advanced pathology, can influence CT and TOT.21,22For these reasons, as well as others, we are exploring more precise and comprehensive ways of understanding surgical workflow and its effects on CT and TOT.We discovered numerous strong correlations between step visit frequency and step time with both CT and TOT.Splenic flexure mobilisation (both step frequency and step time) had a strong correlation with CT and TOT.This makes sense, as complete SFM was usually not performed, and when done, it was indicative of a more difficult reconstruction.Similarly, a longer initial exposure time had a strong correlation with TOT.Longer initial exposure times were usually due to dense adhesions from prior surgery requiring extended adhesiolysis.Mesorectal dissection and left rectal dissection both had a strong correlation with CT.These steps require precise dissection and many adjustments to achieve and maintain exposure.They are more complicated in obese patients, those with bulky tumours, a narrow pelvis, or prior radiation.Additionally, IMV dissection and ligation showed a strong correlation with CT.IMV dissection and ligation, like SFM, were not routinely performed and appear to have a wider variation in step times.Both nodal steps were related to the dissection of the descending and sigmoid colon (medial-to-lateral and lateral-tomedial approaches), with more than 10 different steps both preceding and following.Medial-to-lateral dissection of the rectum was a convergent step, with 13 different steps preceding (most commonly IMA dissection) and only seven different steps following.Rectal transection was a divergent step, with six different steps preceding, and 12 different steps following.The most common procedure following rectal transection was colorectal anastomosis.The variation in step dynamics can be due to differences in patient anatomy, varied pathology, surgeon preference, or in response to operative events.For example, some surgeons started with lateral-to-medial dissection of the abdominal colon, followed by medial-to-lateral pelvic dissection and TME.Other surgeons began with the medialto-lateral pelvic dissection, followed by IMA dissection and ligation, then TME, and lastly medial-to-lateral abdominal colon mobilisation.By precisely annotating these procedures, and correlating objectively delineated workflow with predictors (patient demographics, specific pathology, surgeon experience) and results (operative times, patient outcomes), we hope to more accurately understand what exactly occurs during these procedures, with more precise answers for when and why.We will utilise our current findings to inform future research in the field of surgical data science, an emerging area of research focused on studying OPIs during robotic surgery.By focusing on the most influential steps of a procedure, we hope to understand which OPIs (i.e., kinematic and event data) are most predictive of surgeon skill and case complexity.Subsequent studies will include a larger sample size, surgeons with different experience levels, and patients with various diseases.This work may help define stages of the learning curve across key steps of an operation.Specifically, we anticipate assigning a percentage of individual step ownership by trainees, and tracking the progression of step ownership throughout a trainee's curriculum.Ultimately, these data will be instrumental in identifying best training practices for new robotic surgeons.