Evaluating the effectiveness of population-based breast cancer service screening: an analysis of parsimonious patient survival information with the time-varying Cox model

Background This study is aimed at estimating the unbiased effectiveness of population-based breast cancer service screening based on case survival information alone rather than large-scale individual screening data pursuant to the intention-to-treat principle of a randomized-controlled trial. Methods A novel time-dependent switched design with two modalities of cancer detection (screen-detected vs clinically detected) was proposed to evaluate the effectiveness of breast cancer screening. We used data on 767 patients from Kopparberg in the Swedish Two-County trial and on 78 587 patients in the Taiwan population-based service screening. We estimated the relative rate of the screen-detected vs the clinically detected with adjustment for both truncation and lead-time biases. The absolute effectiveness in terms of the number needed to screen (NNS) for averting one death from breast cancer was estimated. Results The relative rate of effectiveness was estimated as 33%, which was consistent with the 37% reported from the original Swedish randomized-controlled trial. The corresponding estimate for the Taiwan screening programme was 42%, which was also very close to that estimated using individual screening history data (41%). Both relative estimates were further applied to yield 446 and 806 of NNS for averting one death from breast cancer for the corresponding two data sets. Conclusion The proposed time-dependent switched design and analysis with two modalities of case survival information provides a very efficient means for estimating the unbiased estimates of relative and absolute effectiveness of population-based breast cancer service screening dispensing with a large amount of individual screening history data.

Automated summary

This study proposed a novel time-dependent switched design with two modalities of cancer detection to evaluate the effectiveness of breast cancer screening. The relative rates of screen-detected vs clinically detected were estimated as 33% for the Swedish trial and 42% for the Taiwan screening program, with corresponding estimates of 446 and 806 people needed to screen for averting one death from breast cancer. The proposed design provides an efficient means for estimating the unbiased estimates of relative and absolute effectiveness of population-based breast cancer service screening without the need for large amounts of individual screening history data.