Keywords Endoscopy Upper GI Tract - Endoscopy Lower GI Tract - Endoscopy Small Bowel
Introduction
Gastroenterologists spend long hours in the endoscopy suite performing ever-higher
volumes of complex procedures with little rest or recovery. In 2006, it was reported
that on average, endoscopists performed 12 upper endoscopies and 22 colonoscopies
per week [1 ]. More recent reports have shown that some endoscopists perform as many as 24 endoscopic
procedures per day [2 ]. Such a demanding lifestyle predisposes endoscopists to injury. In a recent survey
of 1,698 gastroenterologists, 75% reported experiencing endoscopy-related musculoskeletal
injuries (ERIs) [3 ]. This topic has garnered a significant amount of attention, including ergonomics-related
publications [4 ]
[5 ]
[6 ]
[7 ]
[8 ], instructional videos [9 ]
[10 ]
[11 ], and yearly talks at national meetings and fellowship programs. Summarizing this
work, the American Society of Gastrointestinal Endoscopy recently published endoscopy
ergonomics guidelines [12 ]. Although these efforts are essential, education and awareness alone are inadequate
for meaningful improvement in ergonomics for a variety of reasons [13 ]; citing forgetfulness, busy endoscopy schedule, and lack of insight, endoscopists
often revert to old patterns of behavior despite viewing and understanding the ergonomics
educational materials. Passive lecture-based didactics are poorly suited to break
habits formed over years of training and practice because they do not provide real-time
coaching or recurring ergonomic performance feedback.
To fill this gap, we developed ErgoGenius, a computer-vision tool that provides live
posture assessment, comprehensive posture analysis, and longitudinal feedback for
healthcare providers and trainees. The tool analyzes video capture data of endoscopist
movements, without the need for wearable sensors. Utilizing artificial intelligence
(AI)-based posture analysis, it identifies relevant joint locations to establish posture
and calculate the validated Rapid Entire Body Assessment (REBA) ergonomic score [14 ]
[15 ]
[16 ]
[17 ]; a validated surrogate for musculoskeletal injury risk. ErgoGenius provides a personalized
procedure-level ergonomics analysis of each major joint that is coupled with tailored
feedback on improving posture. ErgoGenius is deployed in real time, providing live
feedback, and offering longitudinal monitoring to follow the results of each intervention
over time. The aims of this study were to determine the feasibility of deploying ErgoGenius
in a controlled environment in the endoscopy suite, its accuracy compared with independent
human appraisers, and its ability to detect abnormal posture.
Methods
Software
ErgoGenius employs a convolutional neural network to detect body parts from still
images and is based after a combination of the BlazePose [18 ], BlazeFace [19 ], and the single-shot-detector [20 ] models. The software is designed to determine postural changes in real-time. ErgoGenius
can process videos with up to 60 frames per second. It analyses posture in each frame
separately and averages out the scores over all frames. The software has been tailored
to the types of movements often performed in the endoscopy suite such as neck flexion,
extension, rotation, trunk movements including bending and rotation, upper extremity
movements including torque, as well as lower extremity position and weight loading.
Study type
This was a prospective, open label, dual center, pilot study.
Set-up
The study took place in the endoscopy suites of two academic medical centers (University
of Texas in San Antonio and New York University Langone Health). The participants
were gastroenterology fellows in years 1 to 3 of training. The participants performed
a limited endoscopic challenge using the Olympus Thompson Box (the task consisted
of using rat-tooth forceps to pick up and transfer plastic shapes from one end of
the box to the other requiring retroflexion to place the plastic shapes on top of
plastic spikes protruding on either side of the box). Participants were filmed using
an iPhone12 rear-view camera. Each participant performed the same 2-minute challenge
in two different settings. The first was performed with the bed in optimal position
(5–10 cm below elbow height). The second was performed with the bed lowered to about
1 to 5 inches above knee height (this depended on endoscopist height and limitations
of bed descension).
Videos were collected and fed into the ErgoGenius tool. Procedure-level mean Rapid
Entire Body Assessment (REBA) scores were calculated. REBA is calculated by examining
various factors such as joint positioning and force exertion [14 ]
[21 ]. Scores are separately assigned for neck, trunk, arm, wrist, and lower limb postures
(Supplementary Fig. 1).
Outcomes
Validation
To determine the accuracy of ErgoGenius REBA calculations, we compared them to the
current gold standard (human appraisal). Four independent providers from three institutions
participated in this process. In the beginning, the participants were introduced to
the REBA scoring system, and were provided with examples of how to calculate it. They
were then given 10 randomly chosen screenshots of endoscopists performing endoscopic
procedures. They independently calculated REBA scores for the endoscopist in each
of these images (Supplementary Fig. 1). The images were later processed by ErgoGenius
with three repetitions. The similarity between ErgoGenius and human REBA scores was
compared using Spearman’s rank correlation coefficient.
Ergonomic performance based on bed position:
Our primary endoscopist outcome was the change in REBA scores with change in bed position.
A paired T-test was used to compare REBA scores for each bed position. P < 0.05 was considered statistically significant.
Results
Ten providers (five male and five female) of different heights and weights participated
in our study ([Table 1 ]). ErgoGenius was successfully deployed in a controlled endoscopy setting.
Table 1 Endoscopist characteristics.
Endoscopist
Sex
Height
Weight
1
M
185
198
2
F
165
140
3
M
180
186
4
M
193
235
5
F
160
140
6
F
165
140
7
F
168
145
8
F
160
128
9
M
185
175
10
M
167
145
Validation
Four independent appraisers participated in the validation cohort. Human appraisers
had relatively poor internal agreement on REBA scores with an average rho of 0.71,
whereas ErgoGenius had perfect internal agreement with rho of 1. The average readings
of human appraisers correlated to those of ErgoGenius with a rho of 0.987, indicating
that ErgoGenius performs at the level of human appraisers and validating its accuracy
for this task ([Fig. 1 ], Supplementary Table 1).
Fig. 1
a Correlation of human-appraiser scores with ErgoGenius-generated scores. b Correlation of average human-appraiser scores with ErgoGenius scores.
Ergonomic performance based on bed position
A significant decline in ergonomic performance (indicated by increased REBA scores)
occurred when the bed was switched to the lower position with a mean difference of
2.1 (P = 0.006) ([Table 2 ], [Fig. 2 ]).
Table 2 REBA scores based on bed height.*
Endoscopist
Bed height
Optimal
Low
*Ergonomic scores for participants at optimal bed height (bed height 5–10 cm below
elbow height) and low bed height (bed height up to slightly above knee position).
Higher scores correspond to higher risk of injury.
1
2.58
2.62
2
2.67
4.6
3
3.54
5.27
4
1.68
4.47
5
2.26
9.14
6
2.62
3.87
7
3.15
5.14
8
2.29
3.29
9
1.97
3.55
10
2.72
4.48
Mean
2.548
4.643
P value
0.0055
Fig. 2 ErgoGenius screen showing real-time comprehensive ergonomic analysis. a Endoscopist in semi-neutral posture with low-risk scores. b Endoscopist in contorted posture with crouching and twisting movements, resulting
in higher-risk scores.
Discussion
We validated a novel automated ergonomics assessment in a controlled setting in the
endoscopy suite. It was able to detect abnormal posture related to change in bed position
and exhibited human-level accuracy in calculating REBA scores that reflect risk of
musculoskeletal injury, showing promise as a novel method of automating ergonomic
assessment for endoscopists in the real world in an objective way.
Up to 75% of gastroenterologists suffer from ERI [3 ], which undermines their welfare and productivity; 28% of endoscopists reduce their
procedure volume due to pain, 25% increase time between procedures to avoid pain,
65% use nonsteroidal anti-inflammatory drugs to treat musculoskeletal pain, 37% reduce
physical activity outside of work to avoid pain, and 10% report missing days from
work due to pain (median 30 days). These numbers only reflect part of the problem
because musculoskeletal injury results in reduced access to healthcare for patients
whose providers are sidelined due to injury. A cost analysis conducted by our group
showed that a missed day for a gastroenterologist translates into $15,760 in lost
revenue, and a drop in case volume of 25% translates into $78,700 in lost revenue
per month for a gastroenterology practice. These numbers reflect the economic impact
of the problem on a gastroenterology practice. While no data currently exist on rates
of forced retirement as a result of musculoskeletal injury for endoscopists, a recent
study investigating the effects of musculoskeletal injury on interventional medical
specialties reported that 12% of physicians with work-related musculoskeletal injuries
required a leave of absence, practice restriction, or early retirement [22 ]. Existing measures to address this problem lack real-time continuous monitoring,
a key feature offered by ErgoGenius. Previous studies have shown that training programs
and personal assessment by ergonomics experts lead to lower risk of injury [6 ]
[23 ]. However, access to these experts is extremely limited. ErgoGenius offers an automated
and accessible alternative, dramatically increasing provider access to ergonomic assessment
and expertise both in urban and rural areas.
Our study was limited by the relatively small sample size and use of a Thompson box
challenge rather than a clinical patient-based procedure. However, as a pilot study,
we provide robust data showing utility and clear statistical significance despite
the small sample size. Although real patients were not included, the study was conducted
in two clinical endoscopy suites often used for clinical procedures rather than a
sim-lab, further supporting usability in real-life clinical setting. Future studies
including larger sample sizes conducted in patient-care settings are needed to revalidate
our results.
Our findings suggest that AI- and computer-vision based technologies may have a role
to play in the clinical environment in a provider-focused manner. We find that ErgoGenius
may be able to function as an ergonomics trainer, providing real-time monitoring and
risk assessment. Future iterations of this technology can focus on more detailed analysis
of the endoscopy suite environment, set-up, and equipment and identify further risks.
In addition, this technology may be used to appraise the potential ergonomic impact
of new endoscopic technologies and guide manufacturers to adopt more ergonomically
friendly designs earlier in the device development process.
Conclusions
ErgoGenius was able to detect abnormal posture and quantify ERI risk by calculating
the validated REBA score for providers in real time at the level of human appraisers.
Future studies are needed to study the effectiveness of this tool in reducing risk
of injury among providers by identifying their areas of improvement from an ergonomic
standpoint and offering real-time and longitudinal feedback.
Pilot evaluation of a novel, automated ergonomics assessment tool
Bibliographical Record
