Page 55 - Read Online

Goyal et al. Art Int Surg. 2025;5:450-64 https://dx.doi.org/10.20517/ais.2024.108 Page 458

Table 4. Recognition of surgical phases across studies
Ortenzi et al. Takeuchi et al. Choksi et al. Takeuchi et al. Zang et al. Zygomalas et al.
Authors [10] [20] [21] [8] [9] [7]
(2023) (2023) (2023) (2022) (2023) (2024)
Phase 1 Balloon dissection Pubic symphysis Adhesiolysis Preparation Adhesiolysis Mesh
Phase 2 Access to the Direct hernia Peritoneal Scoring Peritoneal flap incision Peritoneal Scoring Grasp
preperitoneal space orifice
Phase 3 Preperitoneal Cooper ligament Preperitoneal Peritoneal flap dissection Preperitoneal Bowel
dissection dissection dissection
Phase 4 Hernia sac reduction Iliac vein Reduction of Hernia dissection Reduction of Needleholder
hernia hernia
Phase 5 Mesh placement Doom triangle Mesh positioning Mesh deployment Mesh positioning Umbilical ligament
Phase 6 Mesh fixation Deep inguinal ring Mesh placement Mesh fixation Mesh placement Inf Epi Vessel
Phase 7 - Iliopsoas muscle Positioning suture Closing of peritoneal flap, Positioning suture Spermatic Vessel
additional closure
Phase 8 - - Securing mesh No step Primary hernia Cooper
repair
Phase 9 - - Primary hernia - Catheter insertion Bipolar
repair
Phase 10 - - Catheter insertion - Peritoneal closure Vas Deferens
Phase 11 - - Peritoneal closure - Transitory idle Scissors
Phase 12 - - Transitory idle - Stationary idle -
Phase 13 - - Stationary idle - Out of body -
Phase 14 - - Out of body - Blurry -
Phase 15 - - Blurry - - -

Accuracy and performance metrics across studies
The accuracy and performance metrics of the included studies are summarized in Table 3. Overall, there
was variability in reported accuracy and performance across studies, influenced by methodological
[10]
differences and dataset characteristics. Ortenzi et al. and Zygomalas et al. demonstrated the highest
[7]
accuracy, exceeding 87%, while Choksi et al. reported lower accuracy with considerable variation across
[21]
cases. F1 scores, reported in a subset of studies, further highlighted predictive performance, with Zygomalas
et al. achieving the highest score at 82%. Precision metrics were also noted, with values ranging from 51%
[7]
[20]
to above 50% in studies by Takeuchi et al. and Zygomalas et al. .
[7]
Risk of bias
The risk of bias for all six included studies was assessed using the Newcastle-Ottawa Scale (NOS) across
three domains: Selection, Comparability, and Outcome Assessment. Three independent reviewers
conducted the assessment, and consensus scores were derived after resolving discrepancies. Two studies
[10]
(Ortenzi et al., 2023 ; Takeuchi et al., 2023 ) were classified as high quality, while the remaining four were
[20]
rated as moderate quality based on their total NOS scores. The scores for each domain and their totals are
presented in Table 5.

To evaluate the reliability of these assessments, inter-rater agreement was calculated using Fleiss’ kappa for
each domain. The selection domain exhibited the strongest agreement, with κ = 0.82 (95% CI: 0.74-0.90),
reflecting high consistency across reviewers. Comparatively, the comparability domain demonstrated
moderate agreement, with κ = 0.49 (95% CI: 0.37-0.61), while the outcome domain showed fair agreement,
with κ = 0.44 (95% CI: 0.32-0.56). The overall agreement was 0.89, 0.73, and 0.67 for the selection,
comparability, and outcome domains, respectively, with corresponding expected agreement values of 0.39,
0.47, and 0.41. Notably, Choksi et al. (2023) had the lowest inter-rater agreement within the outcome
[21]
domain (κ = 0.33), highlighting variability in the reviewers’ interpretation of the NOS criteria for this study.