AbstractsBiology & Animal Science

Measuring the mortality reductions due to cancer screening

by Zhihui Liu




Institution: McGill University
Department: Department of Epidemiology and Biostatistics
Degree: PhD
Year: 2015
Keywords: Biology - Biostatistics
Record ID: 2059669
Full text PDF: http://digitool.library.mcgill.ca/thesisfile130321.pdf


Abstract

Evidence of benefits due to cancer screening is commonly reported as the cumulative mortality reduction over the entire follow-up window of a randomized screening trial. However, such a single number summary statistic is of limited use in projecting the timing, duration and magnitude of the mortality reductions that would be expected from a sustained screening program, of longer duration and possibly with a different screening regimen. Meta-analyses, by averaging such measures from trials with varying follow-up windows and screening regimens, have produced summaries that are even less meaningful.This thesis, composed primarily of four manuscripts, presents theoretical and methodological developments for measuring the mortality reductions due to cancer screening. Our objective is to project the time-specific reductions in mortality that would be produced by a sustained screening program, using data from randomized trials, with the aim to give policy makers and funders more accurate evidence on how effective screening programs are and could be. In the first manuscript, we propose using a mortality reduction curve to address the mortality impact of a screening program, instead of a single-number summary. We illustrate when and how such curves from randomized trials could be computed, and how they could be used to project reduction patterns expected with different screening regimens.In the second manuscript, instead of modelling the entire history of the cancer, we develop a novel probability model to address the mortality impact, by parametrizing the conditional probability of being helped by a single round of screening, given that the cancer would be fatal otherwise. We (i) show that this conditional probability can be directly interpreted as the reduction in cancer-specific mortality, (ii) suggest a parametric form for it, based on which we formulate a likelihood function, and (iii) extend this model to accommodate unequal allocation, less than full compliance, combination of information across trials with different regimens, as well as different regimens within a trial. Two case studies are presented using data from screening trials for lung and colorectal cancers.A more detailed analysis of the data from the US National Lung Screening Trial is presented in the third manuscript. We demonstrate that our model can be fitted to both individual-level data and aggregated data, with very little precision lost when using the aggregated data. All the aggregated mortality data used in this thesis were extracted via a new reconstruction technique we propose in the fourth manuscript. Using examples and an error analysis, we illustrate the precision of the information that can be recovered from various electronic formats. One advantage of our approach, compared to previous ones, is that observer variation is eliminated and thus the extraction is completely replicable. On illustre communément les bienfaits attribuables au dépistage du cancer par la réduction de la mortalité observée à travers la période de suivi d'un essai…