Consensus double reading of baseline results of lung cancer screening provided no statistically significant benefit over a single reader when using a nodule management strategy based solely on semi-automated volumetry, according to a study published in the January issue of Radiology.
There has been little agreement on the value of a double reading in the detection of cancers, according to the study authors, including Ying Wang, MD, of University Medical Center Groningen, the Netherlands, and colleagues. Studies in the detection of breast cancer, for example, have shown that radiologists can differ substantially in the interpretations of mammograms and that a double reading is recommended for breast cancer screening. While some studies have shown that double reading increased the breast cancer detection rate by 6-15 percent compared with single reading, other investigators found much more modest cancer detection rate gains of just 2-5 percent after double reading.
“Efforts to improve accuracy and to reduce variability in the interpretation may potentially increase the effectiveness of a screening program,” wrote the authors, who added that, to their knowledge, the effect of double reading on lung cancer detection in CT screening had not been investigated.
With this in mind, the researchers retrospectively evaluated the performance of consensus double reading compared with single reading in the baseline period of the low-dose NELSON [Nedelands-Leuvens Longkanker Screenings Onderzoek] CT lung cancer screening trial.
CT images of the 7,557 participants were interpreted by readers at one of four screening sites before a subsequent reading of the same images by an experienced reader at a centrally located site. Second readers were not blinded to the results of the first reading. If there was a discrepancy in interpretation, the first reader was notified and the image was re-evaluated in an attempt to reach consensus. If no consensus was reached, arbitration from a third experienced radiologist was used to form the final interpretation.
The interpretations were compared against the reference standard of a retrospective analysis of the serial CT scans performed in participants diagnosed with lung cancer within two years after the baseline screening.
Results showed that double reading did not significantly increase the detection rate or recall rate. Double reading led to the detection of 2.7 percent more subjects with lung cancer and a 0.2 percent increase in recall rate. Differences in sensitivity, specificity, positive predictive value and negative predictive value were similarly minor, if they existed at all.
Double reading, however, did lead to a statistically significant 19 percent increase in the number of detected nodules. While it might seem logical that cancer detection would increase with the detection of more modules, this was not the case.
“The reason for this discordance could be explained by the fact that first readers pay more attention to the larger, suspicious nodules and potentially neglect the smaller ones,” wrote the authors. “In our nodule management protocol, the test result was based on the highest nodule category. Therefore, the detection of additional nodules in the vast majority of the participants did not change the test result and cancer detection.”
The authors acknowledged that a limitation of the study was the low power of analysis due to the 0.9 percent cancer detection rate that was the basis of the study. With this rate, more than 80,000 study participants would be needed to get a statistically significant result on the effect of a double reading.
“Furthermore, even if this significance could be achieved, the human cost of one extra detected early stage–lung cancer corresponds to 7,557 second readings, which is equivalent to 253 working days for a radiologist with a throughput of 30 readings per day,” wrote Wang et al. “Therefore, after weighing the advantages and the cost of consensus doubling reading, we do not recommend consensus double reading in lung cancer screening with the use of our nodule management strategy based on semiautomated volumetry.”