THE ACCURACY OF HUMAN DETECTION OF SUBMUCOSAL INVASIVE CANCER – ANALYSIS OF 739 INDIVIDUAL ASSESSMENTS OF LARGE NON-PEDUNCULATED COLORECTAL POLYPS USING A NOVEL CLINICAL DECISION TOOL

Aims Current tools to detect submucosal invasive cancer (SMI) within large (>=20mm) non-pedunculated colorectal polyps (LNPCPs) are complex. This potentially leads to incorrect decision-making (e.g. piecemeal resection of SMI necessitating surgery or surgery for benign disease).

An online decision-support tool was developed using the concept of a demarcated area – where a regular pit/vascular pattern becomes disordered – to search for OVERT (visible on the surface) SMI and 4 morphological characteristics of LNPCPs (Paris classification, size, colonic location, and granularity) to quantify COVERT (hidden) SMI risk.

Methods A survey was sent to endoscopists containing an educational video with 20 subsequent randomly-presented standardized videos of LNPCPs. Participants’ first impression was asked before using the tool to classify polyps as low/high (COVERT), or very high (OVERT) risk of SMI. Responses were compared to expert responses and histopathology.

Results 739 individual responses were analysed. First impression strongly predicted absence of SMI – negative predictive value (NPV) 97.5% (95% confidence interval (95%CI):95.0-99.0%) and accuracy 72.3% (95%CI:68.9-75.6%).

Absence of a demarcated area was predicted with similar NPV – 97.6%(95%CI:96.0-99.0%). Lower accuracy, 78.6%(95%CI:76.5-81.5%), resulted from participant overcalling of a demarcated area. The accuracy of participant size, Paris classification, granularity, and location (provided) determination were 66.3%(95%CI:61.8-70.5%), 66.3%(95%CI:61.8-70.5%), 64.8%(95%CI:60.3-69.1%) and 94.8%(95%CI:92.4-96.6%) respectively.

Overall accuracy was 66.2%(95%CI:62.6-69.6%), clinically resulting in correct treatment in 71.3%, undertreatment in 13.1% and overtreatment in 15.6%.

Table 1 Resultant treatment if determined by participant score from the clinical decision support tool as versus expert score. % in header of row, % in table body of column.

Expert score versus participants result in terms of treatment outcome (right) // Result of participant scoring (below)	Undertreatment n=97 (13.1%)	Correct treatment n=527 (71.3%)	Overtreatment n=115 (15.6%)
Low risk, n (%)	0 (0)	329 (62.4)	115 (100)
High risk, n (%)	87 (89.7)	60 (11.4)	0 (0)
Very high risk, n (%)	10 (10.3)	138 (26.2)	0 (0)

Fig. 1

Conclusions Endoscopists of varying experience can exclude SMI within LNPCPs from a standardized video using a decision-support tool after training. Further validation and application of this tool may prevent negative patient outcomes.

Publication History

Article published online:
14 April 2022