Prospective study of computer-aided detection of colorectal adenomas in hospitalized patients

Abstract Background Adenoma detection with polypectomy during total colonoscopy reduces the incidence of colorectal cancer (CRC) and colorectal cancer-associated mortality. The adenoma detection rate (ADR) is an established quality indicator, which is associated with a decreased risk for interval cancer. An increase in ADR could be demonstrated for several artificially intelligent, real-time computer-aided detection (CADe) systems in selected patients. Most studies concentrated on outpatient colonoscopies. This sector often lacks funds for applying costly innovations like CADe. Hospitals are more likely to implement CADe and information about the impact of CADe in the distinct patient cohort of hospitalized patients is scarce. Methods In this prospective, randomized-controlled study, we compared colonoscopies performed with or without computer-aided detection (CADe) system (GI Genius, Medtronic) performed at University Medical Center Schleswig-Holstein, Campus Luebeck. The primary endpoint was ADR. Results Overall, 232 patients were randomized with n = 122 patients in the CADe arm and n = 110 patients in the control arm. Median age was 66 years (interquartile range 51–77). Indication for colonoscopy was most often workup for gastrointestinal symptoms (88.4%) followed by screening, post-polypectomy and post-CRC surveillance (each 3.9%). Withdrawal time was significantly prolonged (11 vs. 10 min, p = 0.039), without clinical relevance. Complication rate was not different between the arms (0.8% vs. 4.5%; p = 0.072). The ADR was significantly increased in the CADe arm compared to the control (33.6% vs. 18.1%, p = 0.008). ADR increase was particularly strong for the detection in elderly patients aged ≥50 years (OR 6.3, 95% CI 1.7 − 23.1, p = 0.006). Conclusion The use of CADe is safe and increases ADR in hospitalized patients.


Introduction
Total colonoscopy is the established method for screening colorectal carcinoma (CRC) [1].Colonoscopic polypectomy reduces the incidence of CRC and CRC-associated mortality [1,2].Nevertheless, 25% of colorectal neoplasias including benign or premalignant adenomatous polyps are missed during screening colonoscopy and may result in interval cancer, which is defined as colorectal adenocarcinoma diagnosed between the time of screening colonoscopy and post-screening examination [3].Therefore, several quality-assessing indicators have been established in order to ensure high-quality examinations [3].Next to the cecal intubation rate, the adenoma detection rate (ADR) is most commonly used as a quality indicator [3].Within the German guidelines, the endoscopist′s ADR within screening colonoscopies is required to be more than 20% [1,[3][4][5].
In order to increase ADR, several artificially intelligent, real-time computer-aided detection (CADe) systems have been developed.High-definition colonoscopy with real-time CADe systems significantly increased ADR by 6-16% in selected patients undergoing colonoscopy within seven prospective, randomized controlled trials [6][7][8][9][10][11][12][13].However broad implementation of CADe in the outpatient sector is unlikely due to insufficient funding.Patients undergoing colonoscopy during hospital stay differ significantly in terms of age and co-morbidities from patients undergoing out-patient colonoscopy for primary screening or post-polypectomy surveillance.Information about risks and benefits in patients attending diagnostic colonoscopy is scarce and recommendations regarding CADe use are frequently derived from primary screening studies.These considerations prompted us to analyze ADR with and without CADe in patients undergoing colonoscopy in our tertiary hospital center.

Study population
This study was a prospective, open, randomized-controlled study per formed at University Medical Center Schleswig-Holstein, Campus Luebeck (CompuColo study 21-119), Germany.Patients, who were planned for colonoscopy were included in this study from 09/2021-11/2021.Indications for colonoscopies included i.) primary screening, ii.) post-polypectomy surveillance, iii.) tumor follow-up and iv.) work-up for gastrointestinal (GI) symptoms such as bleeding and anemia, inflammatory bowel disease, diarrhea and tumor search, further referred to as diagnostic colonoscopy.Inclusion criteria were age ≥ 18 years and obtained confirmed consent.Exclusion criterion was pre-planned partial colonoscopy or planned repetition of colonoscopy due to an unprepared colon.The study protocol conformed to the ethical guidelines of the 1975 Declaration of Helsinki as reflected in a priori approval by the appropriate institutional review committees.The study was registered within the German Clinical Trial Registry (DRKS00024943, Deutsches Klinisches Studienregister (DKRS)).

Procedures
High-definition colonoscopy was performed using a regular orthograde endoscope (Gif-Q180H, Gif-Q190; Evis X1, Olympus) by trained endoscopists.Endoscopists were trained for the ability to identify and localize one or more adenomatous lesions in real-time colonoscopy using a CADe system (GI Genius, Medtronic).The output appears on the same screen of the endoscopy system without affecting the routine technique of the examiner.Bowel preparation was evaluated and graded by the endoscopist performing the exam, using the Boston bowel preparation score (BBPS).A score of 6 and above (each segment ≥2) was considered adequate [14].Intubation and inspection time during withdrawal was measured using a stopwatch.All polyps were classified according to their location, size and morphology according to the Paris classification [15].Polypectomies were performed using forceps biopsy for very small polyps (<3mm), cold snare (Steris Exacto ® cold snare 9 mm) for medium polyps (3 mm-5mm) and diathermic snare (medwork UltraSnare 15 -40 mm) for larger lesions (>5mm).Randomization of the patients was carried out by alternate use of CADe.Patient and exam data were prospectively evaluated.Age, gender, indication for colonoscopy, BBPS, the endoscopist's experience (having performed >200 colonoscopies annually and >50 polypectomies anually within the past three years was defined as an expert), number and size of polyps according to Paris classification, complication rate, withdrawal time and histopathological results were documented.

Outcome measures
The primary outcome was the ADR, which was defined as the proportion of colonoscopies with at least one histologically proven adenoma or adenocarcinoma according to all colonoscopies performed.Secondary outcomes were polyp detection rate (PDR), carcinoma detection rate (CDR), withdrawal time and complication rate.PDR and CDR were defined as the proportion of colonoscopies with at least one detected polyp or adenocarcinoma according to all colonoscopies performed.All resected specimens underwent histopathological examinations.

Sample size
The sample size was calculated based on the evaluation of the primary outcome ADR.Baseline ADR of 20% was estimated after a review of previous data from our colonoscopy unit.An increase in ADR of 15% up to a total of 35% was expected due to prior data on CADe performance [6,7].Using a two-arm binomial one-sided calculation, at an alpha level of 5% and a Power of 80% we estimated a sample size of 109 patients per arm.With an expected drop-out rate of 10%, 240 patients were calculated for the study (Statistical Tools (stattools.crab.org))[16].

Statistical analysis
Statistical analyses were performed using SPSS 27.0 (SPSS Inc., Chicago, IL, USA).Continuous variables were represented as medians and interquartile ranges (IQR).Medians were compared using the Wilcoxon signed-rank test.Differences between categorical variables were calculated using Pearson′s Chi-squared test.Posthoc subgroup and association analyses of the impact of CADe on ADR using binary logistic regression were performed for the following subgroups: gender, expert colonoscopies and adequate preparation, each as defined above, complete colonoscopy, indication for colonoscopy (diagnostic versus (vs.) others) and age of at least or above screening recommendation (≥50 years).

Study population and procedures
Overall, 301 patients were considered eligible, but 18 examinations were excluded due to repetition for unprepared bowel and 51 were excluded because total colonoscopy was not intended.
Patient-and procedure-specific characteristics were similar between the arms (Table 1).

Polyp-related outcomes
The total number of detected polyps was 273 and was higher in the experimental arm compared to the control arm (n = 172 (1.41 per colonoscopy) vs. n = 101 (0.92 per colonoscopy)).The total number of polypectomies was 198 and accordingly higher in the experimental arm (128/172 = 74.4% vs. 70/101 = 69.3%).In total, n = 75 polyps were left due to macroscopic identification as hyperplastic polyp (NICE-class 1), inadequate bowel preparation, coagulation disorders or lack of therapeutic consequence.The median size of the polyps was 5 mm (3-10, p > 0.05) in both arms.

Subgroup analysis
In patients at age above screening recommendation (≥50 years) the ADR was significantly increased by the use of CADe with 42.7% vs. 20.2% in the control arm (p = 0.001), whereas in younger patients (age <50 years) ADR was comparable in both arms with 9.1% vs. 9.5% (p > 0.05).In the subgroup of adequate bowel preparation ADR did not differ significantly with 36.6% vs. 25.8%(p > 0.05), between both groups, whereas in patients with inadequately prepared bowels (BBPS <6 or one segment <2) ADR was significantly higher with 29.4% vs. 6.8% (p = 0.005).In the group of diagnostic colonoscopies (n = 205), the difference in ADR was comparable to the whole cohort but lacked significance with 28.3% vs. 18.2% (p = 0.087), whereas in the minority of the non-diagnostic colonoscopies CADe usage showed an impact with 68.8% vs. 18.2% (p = 0.010).The difference of ADR the arms was higher in non-expert endoscopies with 34.0% vs. 15.4% (p = 0.029) compared to expert examinations with 33.3% vs. 20.7%(p = 0.109).Cecal intubation was achieved in 211 (90.9%) of the examinations, therefore the difference in ADR resembles that of the whole cohort with 34.5% vs. 21.1% (p = 0.031).In incomplete colonoscopy, no adenoma was detected in the control group and ADR was 16.7% in CADe group (p > 0.05).The ADR was higher in males when using CADe compared to the control group with 40.4% vs. 21.4% (p = 0.030) but showed only a trend towards a higher detection rate in females with 27.7% vs. 14.8% (p = 0.091).Results are summarized in Table 4. Differences in ADR according to indication could only be detected in patients with a clinical/radiological suspicion of tumor (48.7% vs. 19.0%,p = 0.005) (see Supplement Table 2).
Regression analysis is plotted in Figure 2.

Discussion
In this prospective, open, randomized-controlled study, ADR was significantly increased by 16% using CADe during colonoscopy in hospital patients.Importantly, the use of CADe was neither associated with an increase of complication rate nor a clinically relevant prolongation of withdrawal time.So far, seven randomized-controlled trials compared colonoscopy with or without real-time CADe systems, all achieving an increase of ADR ranging from 6% to 16% using CADe [6][7][8][9][10][11][12].In accordance with our findings, ADR was increased by 14% and 9% within two prospective, randomized controlled studies, the AID-1-and AID-2-trials, using the same CADe system as in this study (GI Genius, Medtronic) [6,7].In the AID-1 study, 685 patients underwent outpatient screening colonoscopies with or without the CADe system for CRC, post-polypectomy surveillance, or workup due to positive results from a fecal immunochemical test or signs or symptoms of CRC.The ADR was significantly higher in the CADe group than in the control group (54% vs. 40%) without a significant increase in withdrawal time similar to our results [6].The indications for colonoscopy, however, differed significantly from the indications in our study, in which only a minority of patients underwent screening colonoscopy.
The non-inferiority AID-2-trial addressed a very important issue of a potential improvement of colonoscopy quality performed by non-experts by the help of CADe [7].Endoscopist's experience is an established quality indicator for colonoscopy, which ensures high-quality examinations.In clinical practice, however, non-expert endoscopists are frequently involved particularly in a university hospital with training character [3].Within the AID-2 trial, non-expert endoscopists (<2000 colonoscopies) performed colonoscopies in 660 patients with or without the CADe system achieving a significantly higher ADR in the CADe group (53.3% vs 44.5%) [7].A pooled analysis of the AID1 and −2 data, however, could not show an association of ADR increase with the examiner′s experience (relative risk 1.02, 95% CI 0.89-1.16)[7].In contrast, here, the increase of ADR was dependent on the examiner′s experience as a stronger benefit in non-experts was detectable (19%, p = 0.029).Accordingly, a recent prospective German real-world study could not demonstrate any improvement of ADR by the use of CADe in colonoscopies, which were all performed by experts with or without CADe using two opposing screens [17].Here, the high ADR in the control arm was not further increased using CADe (52.0% vs. 50.7%).Nevertheless, in our study, 44% of the examinations were performed by non-experts, which reflects a real-life situation in hospitals and strongly suggests an integration of CADe in clinical practice to improve colonoscopy quality.
We found that the benefit in terms of ADR increase due to CADe was specifically pronounced for colonoscopies in patients aged 50 years or older (OR 6.3, p = 0.006).Additionally,  we showed that the detection of small (≤10 mm) and sessile adenomas was significantly improved by the use of CADe.This finding is comparable to the results of two previous, randomized controlled studies.However, the prognostic significance of a higher detection rate of small polyps remains unclear and needs to be addressed in future studies.An improvement in the detection of larger polyps (>9-10 mm) with the help of CADs has not been shown so far [7,9,18].In our study, control ADR was relatively low at 18%, which was most probable due to the rather old and frail patient cohort in our center.Overall, 40.1% of the patients were not able to prepare their bowel sufficiently due to advanced age and co-morbidities (BBPS of <6 or one segment <2).In these patients, ADR was 18.9% compared to 31.4% in patients with sufficient bowel preparation, confirming that insufficient bowel preparation negatively correlates with ADR [19,20].Compared to the AID-1-and −2-trials, our study population was older with a median age of 66 years in contrast to median ages of 61 and 62 years, respectively [6,7].Moreover, the majority of our patients underwent diagnostic colonoscopy for GI symptoms (88.4%) with a large proportion of non-expert colonoscopies, whereas the majority of patients of the AID-2 trial underwent colonoscopy for primary screening and post-polypectomy surveillance (68.2%) [7].However, other randomized-controlled trials which, similarly to our study, included unselected patients undergoing diagnostic colonoscopy achieved comparable or even lower control ADRs of 8%-20% [9,10,12].
Limitations of our study are the single-center design, the large proportion of non-expert colonoscopies and missing information about frailty.In point of fact, we believe that our study population reflects a real-life, in-hospital situation and our results provide valuable evidence for the use of CADe in hospital practice.If a patient is likely to benefit from polypectomy, we encourage the use of CADe in routine use.
In summary, the application of CADe was associated with an increase in ADR in hospital patients undergoing colonoscopy.The improvement of adenoma detection was particularly high in patients aged 50 years or older.The use of CADe may also be promising for other applications such as surveillance in IBD or for real-time characterization of polyps and assessment for polypectomy, which is beyond the scope of this study and needs to be addressed in further analyses.However, the integration of CADe in routine clinical practice is challenging due to regulatory and reimbursement issues, which urgently need to be addressed in the near future.Recent data have confirmed the importance of screening colonoscopy and polypectomy in CRC prevention, but the extent of benefit fell short of expectations [21].Therefore, it is even more important to investigate the possible effect of the use of CADe on the risk of interval CRC and CRC mortality in large patient cohorts with long follow-up periods.

Table 1 .
characteristics of the study population and procedures.n.a.: not applicable; n.s.: not significant.

Table 2 .
Patient-and procedure-related, primary and secondary outcomes.n.s.: not significant.