Improving visualization and adherence by converting the Dutch colorectal cancer guidelines into decision trees: The Oncoguide project

Abstract

Background: Clinical guidelines are designed to prevent undesired practice variation where high quality evidence or expert consensus is available. However, reading and interpretation of text-based guidelines is time-consuming and might be difficult to apply in routine daily practice. Therefore, the aim of our study is to examine the feasibility of converting the Dutch multidisciplinary colorectal cancer guideline recommendations into data driven algorithms (decision trees) to facilitate guideline usage. Methods: We converted the most recent Dutch colorectal cancer guideline (published in 2014) into decision trees modelled by decision nodes representing patient or disease characteristics ultimately branching into guideline recommendations. Where not evidence-based, decision trees were discussed with an expert panel until agreement was reached. Thereafter, the developed decision trees were published in open access decision support software. Results: In total, we developed 34 decision trees driven by 101 decision nodes. Decision trees focused on recommendations for diagnostics (n=1) staging (n=10), treatment (colon: n=1, rectum: n=5, both: n=9), pathology (n=4), follow-up (n=3) and 1 overview decision tree. We identified guideline recommendation information gaps, for example specific surgical policy related to (the number of) lung metastases, a recommendation about follow-up schemes after resection or local treatment (e.g. RFA) of metastases and the period between neo-adjuvant treatment and re-staging. It was difficult to convert some of the guideline recommendations into decision trees (i.e. 'consider PET-CT scan to exclude extrahepatic metastases'), related to non-conclusive evidence on specific topics. Conclusions: Converting the Dutch colorectal cancer guideline into decision trees is feasible, but presents several challenges. Using decision trees may (I) improve guideline adherence or more conscious guideline deviation; (II) improve guideline (adherence) evaluation from cancer registries; and (III) ultimately learn from clinical cases with documented motivation for guideline deviation.