Robotic Surgery and Clinical Automation: Four Reviews and a Guideline Read Together

Five recent publications address automation in medicine. Three are meta-analyses of robotic surgery in thyroid, benign gynecologic and kidney transplant procedures. One is a systematic review of artificial intelligence (AI) in evidence synthesis. The fifth is an international consensus guideline for radiotherapy target delineation, which its authors also present as guiding principles for automated contouring programs ^{[1][2][3][4][5]}. The surgical reviews do not share a single comparator. The thyroid analysis set a robotic technique against open, endoscopic and another robotic approach. The hysterectomy analysis compared robotic surgery with laparoscopic, vaginal and open surgery, and the transplant analysis compared it with open surgery only ^[2][3][4]. This review covers what each source contributes, where the findings line up and where they diverge, how strong the designs are, and what the sources leave unresolved ^{[1][2][3][4][5]}.

What each source is

The thyroid systematic review and meta-analysis drew studies from six databases ^[2]. It included 22 studies with 8830 individuals ^[2]. It compared bilateral axillo-breast approach (BABA) robotic thyroidectomy with open thyroidectomy, transoral robotic thyroidectomy and BABA endoscopic thyroidectomy ^[2].

The hysterectomy systematic review and meta-analysis last searched PubMed, MEDLINE and EMBASE on 6/2/2021 ^[3]. It sought English-language randomized controlled trials (RCTs), prospective cohort studies and retrospective independent database studies published between January 1, 2010 and December 31, 2020 that reported perioperative outcomes ^[3]. It included 24 articles reporting on 110,306 robotic, 262,715 laparoscopic, 189,237 vaginal and 554,407 open patients undergoing benign hysterectomy ^[3].

The transplant systematic review and meta-analysis searched MEDLINE, Embase, Web of Science and Cochrane databases ^[4]. It collected baseline characteristics, intraoperative and postoperative outcomes, long-term renal function, and graft and patient survival ^[4]. It included 11 studies comparing 482 robot-assisted kidney transplantation (RAKT) procedures with 1316 open kidney transplantation (OKT) procedures ^[4].

The AI review searched Medline, Embase, CDSR and Epistemonikos up to April 2021 for reviews that implemented AI methods ^[1]. To qualify, a review had to use some form of AI method to enable full or semi-autonomous performance of one or more stages of evidence synthesis. The inclusion criteria gave machine learning, deep learning and neural networks as examples ^[1]. Twelve reviews were included, and together they used nine tools to implement 15 AI methods ^[1].

The radiotherapy source is a practice guideline for nasopharyngeal carcinoma that updates the first International Guideline (IG-2018) ^[5]. It addresses clinical target volumes (CTVs), the volumes radiation oncologists contour in treatment planning at different dose levels ^[5]. Its authors conducted a literature review and sent a preliminary questionnaire to international experts ^[5]. After initial voting and iterations, they prepared a final Delphi survey for consensus building ^[5].

Where the surgical findings line up

Each of the three surgical meta-analyses reports at least one outcome favoring the robotic approach over open surgery ^[2][3][4]. In benign hysterectomy, the robotic approach was associated with a shorter hospital stay (p < 0.00001) and less blood loss (p = 0.009) than the open approach ^[3]. It was also associated with fewer complications (odds ratio 0.42 [0.27, 0.66], p = 0.0001) ^[3].

In kidney transplantation, RAKT was associated with a lower risk of surgical site infection than OKT (risk ratio 0.15, p < 0.001) ^[4]. It was also associated with a lower risk of symptomatic lymphocele (risk ratio 0.20, p = 0.03) ^[4]. Other differences favoring RAKT were less postoperative pain (mean difference -1.38 points, p < 0.001), a shorter incision (mean difference -8.51 cm, p < 0.001) and a shorter hospital stay (mean difference -1.69 days, p = 0.03) ^[4].

In thyroid surgery, BABA robotic thyroidectomy showed greater cosmetic satisfaction and less transient hypoparathyroidism than conventional open thyroidectomy ^[2].

Longer-term outcomes appear in both the transplant and the thyroid analyses, measured differently ^[2][4]. The transplant meta-analysis collected long-term renal function and graft and patient survival, and found no difference between RAKT and OKT in renal function, graft survival or patient survival ^[4]. The thyroid abstract reports permanent hypothyroidism for the comparison with BABA endoscopic thyroidectomy, where the robotic approach had lower chances of it ^[2]. For the same comparison it reports better surgical completeness with the robotic approach, assessed by postoperative thyroglobulin level and lymph node removal ^[2].

Where they diverge

In hysterectomy, the robotic approach was described as mainly comparable to the laparoscopic approach ^[3]. Its main benefit over both laparoscopic and vaginal surgery was a shorter hospital stay, reported as a weighted mean difference (WMD) ^[3]. The WMD was -0.144 [-0.21, -0.08], p < 0.0001 against laparoscopy and -0.39 [-0.70, -0.08], p = 0.01 against vaginal surgery ^[3]. The authors state that other benefits were sensitive to the inclusion of database studies ^[3].

In thyroid surgery, the direction of the result depends on the alternative ^[2]. BABA robotic thyroidectomy had more drainage than BABA endoscopic thyroidectomy, yet lower chances of transient vocal cord palsy and permanent hypothyroidism ^[2]. Against transoral robotic thyroidectomy, it produced greater postoperative drainage and greater patient dissatisfaction ^[2]. The authors concluded that BABA robotic thyroidectomy is inferior to transoral robotic thyroidectomy in drainage and cosmetic satisfaction but superior to BABA endoscopic thyroidectomy in surgical performance ^[2]. Within this one analysis, one robotic technique was inferior to another robotic technique on cosmetic satisfaction ^[2].

The sources also report trade-offs differently ^[2][3][4]. The thyroid meta-analysis found a longer operation time for BABA robotic thyroidectomy than for open thyroidectomy ^[2]. The transplant abstract collected intraoperative outcomes but does not list operation time among its summary results ^[4]. The transplant authors concluded that RAKT is a safe and feasible alternative to OKT with fewer surgical complications and without compromising renal function, graft survival or patient survival ^[4]. The hysterectomy authors drew a narrower conclusion: the robotic approach confirms the classic benefits of minimally invasive surgery against open surgery and is mainly comparable to laparoscopy ^[3].

How strong the designs are

The hysterectomy meta-analysis combined RCTs, prospective cohort studies and retrospective database studies ^[3]. Its authors list four limitations: differences in outcomes by study type, a lack of RCTs for robotic versus open comparisons, learning-curve issues, and few robotic versus vaginal publications ^[3]. The robotic versus open results are reported as associations ^[3]. The authors also note that the advantages over vaginal surgery were observed in a patient population with a higher incidence of large uterus and prior surgery ^[3].

The thyroid and transplant abstracts do not state the designs of their included studies ^[2][4]. From the abstracts alone, their strength cannot be graded beyond their label as systematic reviews with meta-analysis ^[2][4]. The transplant meta-analysis reports its comparisons in associational language ^[4]. Its robotic arm totals 482 procedures, against 110,306 robotic patients in the hysterectomy analysis ^[3][4]. The thyroid abstract reports the direction of effects but no effect sizes, confidence intervals or p-values ^[2].

Automation outside the operating room

In the AI review, 11 of the 15 methods (73%) were used at the screening stage of evidence synthesis ^[1]. Two (13%) were used in data extraction and two (13%) in risk-of-bias assessment ^[1]. Ten of the 12 reviews reported advantages, but the authors describe the benefits of data extraction as ambiguous ^[1]. The authors also note that these results rely on self-reporting by the review authors ^[1]. They conclude that AI platforms have taken hold with varying success, and that extensive human validation still appears to be required ^[1]. They add that further evaluation is needed to define the overall contribution of AI to efficiency and quality ^[1].

The radiotherapy guideline documents variation in expert practice ^[5]. Its preliminary questionnaire went to 30 experts from 26 centers in 14 countries or regions ^[5]. It revealed marked variation in CTV contouring and prescribed doses ^[5]. The final Delphi survey had 58 questions ^[5]. Of these, 20 parameters (34%) attained consensus, defined as at least 75% agreement, and 32 (55%) attained agreement, defined as 60% to 74% ^[5].

The 2024 version changed or clarified 36 parameters relative to IG-2018 ^[5]. The major differences are listed below ^[5].

• Use of the post-induction chemotherapy gross tumor volume for the high-dose CTV to 70 Gy equivalent, except in patients with advanced extranodal extension ^[5].
• Stepwise refinement of elective coverage to ipsilateral anatomical structures for an eccentric primary tumor ^[5].
• Selective coverage of nodal levels ^[5].
• A lower elective dose of 50 Gy equivalent ^[5].

The authors present the guideline as a practical reference for target coverage and as guiding principles for generating automated contouring programs to enhance standardization ^[5]. They describe it as based on updated evidence and collective agreement among international experts ^[5]. They cite increasing data on locoregional extension patterns and outcomes from studies on optimizing CTV and doses as the reason an update was needed ^[5].

Both non-surgical sources tie automation to human expert input, in different ways ^[1][5]. The AI review concludes that extensive human validation of AI methods still appears to be required ^[1]. The guideline offers expert-agreed principles for automated contouring, and consensus at the 75% threshold was reached on 20 of 58 survey questions ^[5].

What remains unknown

The hysterectomy authors identify the lack of RCTs comparing robotic and open surgery as a limitation ^[3]. Learning-curve issues are named as a limitation in the hysterectomy analysis and are not mentioned in the thyroid or transplant abstracts ^[2][3][4]. The hysterectomy authors describe the robotic-versus-open advantages as the classic benefits of minimally invasive surgery, and found the robotic approach mainly comparable to laparoscopy ^[3]. The transplant review compared RAKT only with open surgery ^[4].

For AI in evidence synthesis, the authors state that the overall contribution to efficiency and quality still needs to be defined ^[1]. The review's search ended in April 2021 ^[1]. For automated contouring, the guideline abstract sets principles but reports no validation data for any automated program ^[5].

Limitations

This review relies on abstracts ^{[1][2][3][4][5]}. The thyroid and transplant abstracts omit the designs of their included studies, and the thyroid abstract omits effect estimates ^[2][4]. The sources address different specialties and outcomes, so any comparison across them is qualitative ^{[1][2][3][4][5]}. The AI review's findings rest on self-reported benefits ^[1]. The radiotherapy guideline rests on updated evidence and collective expert agreement, and its abstract reports the consensus process without summarizing the underlying outcome data ^[5].