Ï㽶ÊÓƵ

Title:ÌýScreens per death averted: a less biased estimator of the performance of lung cancer screening programs.

´¡²ú²õ³Ù°ù²¹³¦³Ù:ÌýWilber Deck has a BSc in Mathematics at Queen's University and graduated from McGill Medical School in 1990, and continued at McGill to complete an MSc in Epidemiology and Biostatistics and specialty training as a public health doctor in 1995. Since then he has lived and practiced in Gaspé as a public health physician and in clinical oncology and has collaborated on numerous research projects, largely involving cancer screening, at the INESSS and INSPQ.

James Hanley’s website:

Introduction: Screening trials and meta-analyses have typically used the risk ratio (RR) or hazard ratio (HR), essentially the ratio of cancer death rates in the screening and control groups, but this analysis is biased toward the null hypothesis by the fact that early and late follow-up will tend to include deaths that could not be affected by screening. The use of risk difference (RD) avoids this bias, but will not be invariant with regard to risk, making it necessary to interpret RD in the context of a group’s or an individual’s risk level.

Methods: We review traditional analysis of cancer screening effectiveness and show how risk ratios are biased to the null. We use the example of lung cancer (LC) screening trials to illustrate an unbiased estimator, risk difference between control and screened groups, and use linear regression to validate the assumption that this varies with risk level. Results: Regression of RD on risk of LC death (derived from LC death experience in the control groups) tended to confirm our hypothesis of a linear relation. LC mortality results for 9 computed tomography (CT) screening trials for LC are presented, along with calculations of RD and screens per LC death averted. Overall, trials needed 950 invitations to screening to avert one LC death, or 850 screens conducted to avert one death. Adjusted for LC mortality risk, these numbers were x and y respectively. Discussion: Calculation of RD is feasible and has a natural interpretation, as long as it is adjusted for LC mortality risk level. The number of invitations to screening required to avert one death is the screening equivalent of per protocol number needed to treat (NNT), and actual screens required to avert one death is the ‘as conducted’ equivalent. These statistics avoid the bias to the null inherent in ratio statistics (RR, HR) and provide for better evaluation of a program’s effect or an individual’s projected benefit, in the context of informed consent.

Via Zoom-Please visit our website for the Zoom Link: /epi-biostat-occh/seminars-events/seminars/biostati...

Ìý