Refined Algorithm for Identifying Recurrence Among Patients with Non-Metastatic Colorectal Cancer Based on Danish...

In the Danish and other national health registries, colorectal cancer (CRC) recurrence is not routinely registered. Algorithms to label patients with recurrence in Denmark exist but produce cohorts with a risk of selection bias due to either pre- or postoperative exclusion criteria. In this study, we aimed to refine and increase the generalizability of an existing registry-based algorithm.

Data from 5077 patients from an institution and a regional database, encompassing several departments of surgery in Denmark, were retrieved. Patients with non-metastatic CRC were included from 2008 to 2019. Electronic health journal-based recurrence registration was used as reference for the algorithm. Patients were linked with data from the Danish Colorectal Cancer Group database, the Danish National Health Registry, the Danish Cancer Registry, and the Danish Pathology Registry. The algorithm utilized metastasis, chemotherapy, pathology, and local recurrence codes. Refinement of the algorithm included the addition of targeted and radiation therapy codes and including patients who died within 180 days after surgery, along with revising the pathology codes and removing any preoperative exclusion criteria. Performance metrics were evaluated in 10,000 bootstrapped runs, while all-stage and stage-specific cumulative incidence of recurrence and overall survival were estimated.

The refined algorithm included more patients than the conventional algorithm (4388 vs 3684) and performed marginally better in terms of sensitivity (0.92 (95% CI 0.89-0.94) vs 0.90 (95% CI 0.87-0.92)) and specificity (0.97 (95% CI 0.97-0.98) vs 0.96 (95% CI 0.95-0.96). A significant difference in cumulative incidence of recurrence for UICC stage I was detected between the conventional algorithm and reference, which was not significant when using the refined algorithm.

The refined algorithm improves identification of CRC recurrence in national data, enabling broader inclusion and better representation of population subgroups.