Simplified, standardized methods to assess the accuracy of clinical cancer staging

Background: Hospitals lack intuitive methods to monitor their accuracy of clinical cancer staging, which is critical to treatment planning, prognosis, refinements, and registering quality data. Methods: We introduce a tabulation framework to compare clinical staging with the reference-standard pathological staging, and quantify systematic errors. As an example, we analyzed 9,644 2016 U.S. National Cancer Institute SEER surgically-treated non-small cell lung cancer (NSCLC) cases, and computed concordance with different denominators to compare with incompatible past results. Results: The concordance for clinical versus pathological lymph node N-stage is very good, 83.4 ± 1.0%, but the tumor length-location T-stage is only 58.1 ± 0.9%. There are intuitive insights to the causes of discordance. Approximately 29% of the cases are pathological T-stage greater than clinical T-stage, and 12% lower than the clinical T-stage, which is due partly to the fact that surgically-treated NSCLC are typically lower-stage cancer cases, which results in a bounded higher probability for pathological upstaging. Individual T-stage categories Tis, T1a, T1b, T2a, T2b, T3, T4 invariant percent-concordances are 85.2 ± 9.7 + 10.3%; 72.7 ± 1.6 + 11.3%; 46.6 ± 1.8 + 10.9%; 54.6 ± 1.6 – 20.5%; 41.6 ± 3.3 – 0.1%; 54.7 ± 2.8 – 24.1%; 55.2 ± 4.7 + 2.6%, respectively. Each percent-concordance is referenced to an averaged number of pathological and clinical cases. The first error number quantifies statistical fluctuations; the second quantifies clinical and pathological staging biases. Lastly, comparison of over and under staging versus clinical characteristics provides further insights. Conclusions: Clinical NSCLC staging accuracy and concordance with pathological values can improve. As a first step, the framework enables standardizing comparing staging results and detecting possible problem areas. Cancer hospitals and registries can implement the efficient framework to monitor staging accuracy.

Abstract licence: CC BY-NC

A person counts once a day per paper. We keep no cookie and no address: a reader is a hash that changes every day. Any month under five reads shows no number.