Plasma-only ctDNA-Guided MRD Detection in Patients with CRC—Letter

Below we express our concerns about the article authored by Parikh and colleagues, investigating the performance of a tumor-uninformed ctDNA assay (Guardant Health, Reveal) for MRD detection in patients with colorectal cancer (1).

The authors present a “landmark analysis” that unconventionally includes both postsurgical and postadjuvant ctDNA assessments, when the objectives of these two timepoints clearly differ; postsurgical assessments aid in determining whether or not a patient may benefit from adjuvant treatment, whereas postadjuvant assessments help determine who could benefit from further treatment. The clinical relevance of this combined “landmark analysis” can be debated. Furthermore, a direct comparison to previous reports (2, 3) that analyzed these two timepoints separately is inappropriate and misleading.

Besides the uncertain relevance of a “landmark analysis,” our major concerns lie in the apparent selection bias associated with the “longitudinal analysis” and the “surveillance analysis.” First, two patients that tested positive but did not recur (false positives) were excluded from the analysis despite patient TPS1473 showing sufficient follow-up of a year (Fig. 2B). Second, “longitudinal timepoints” were defined as patients who had subsequent draws after their “landmark” timepoint. However, all of the longitudinally negative patients that did not recur (n = 37) had only one draw, which makes the reader wonder whether these patients would have had reversal of calls if more than one draw was performed. Thus, challenges remain in assessing the true specificity and positive predictive value (PPV) in this setting.

Finally, in their “surveillance analysis,” the authors incorrectly reference the methods employed by Reinert and colleagues. As authors of this publication, we confirm that we have never reported a surveillance analysis defined by a draw obtained within 4 months of clinical recurrence (2). Moreover, the resulting exclusion of 7 false negative patients in the “surveillance analysis” artificially inflates the sensitivity from 55.6% to 90.9% (Fig. 3B).

In addition, we urge readers to consider presurgical detection rates as a positive control when considering the performance of any MRD test. In their cohort, even after excluding neoadjuvant treated patients, a ctDNA detection rate of 45.7% was observed, which is surprisingly low compared to the 55.6%, 69.0%, and 90.9% reported for landmark, longitudinal, and surveillance analyses, respectively. For all the aforementioned reasons, the clinical utility of a plasma-only ctDNA assay in patients with colorectal cancer needs to be further validated and carefully designed studies are warranted to make comparisons to established methods (2).

See the Response, p. 6614