Enhancing information on stage at diagnosis of cancer in Africa.

Background/purpose: Stage at diagnosis is an important metric in treatment and prognosis of cancer, and also in planning and evaluation of cancer control. For the latter purposes, the data source is the population-based cancer registry (PBCR), but, although stage is usually among the variables collected by cancer registries, it is often missing, especially in low-income settings. Essential TNM has been introduced to facilitate abstraction of stage data by cancer registry personnel, but the accuracy with which they can do so is unknown.

Methods: 51 cancer registrars from 20 countries of sub-Saharan Africa (13 anglophone, 7 francophone) were tasked with abstracting stage at diagnosis, using Essential TNM, from scanned extracts of case. The panel comprised 28 records of each of 8 common cancer types, and the participants chose how many to attempt (between 48 and 128). Stage group (I-IV), derived from the eTNM elements that they assigned to each cancer, was compared with a gold standard, as decided by two expert clinicians.

Results: The registrars assigned the correct stage (I-IV) in between 60 and 80% of cases, with the lowest values for ovary, and the highest for oesophagus. The weighted kappa statistic suggested a moderate level of agreement between participant and expert (0.41-0.60) for 5 cancers, and substantial agreement (0.61-0.80) for three, with the best for cervix, large bowel, oesophagus and ovary, and the worst (weighted kappa 0.46) for non-Hodgkin lymphoma (NHL). For all except NHL, early stage (I/II) and late stage (III/IV) was correctly identified in 80% or more of the cases.

Conclusions: A single training in staging using Essential TNM resulted in an accuracy that was not much inferior to what has been observed in clinical situations in high income settings. Nevertheless, some lessons were learned on how to improve both the guidelines for staging, and the training course.