Recent methods development for cancer screening
Presenter: Theodore Holford
When: Tuesday, July 12, 2016 Time: 2:00 PM - 3:30 PM
Room: Oak Bay 1-2 (Level 1)
Session Synopsis:
A Simulation Model for the Lifetime Exposure to Cigarette Smoking among Individuals Born in the US from 1890-2000
Cigarette smoking remains as a leading preventable cause of mortality in the US. The Smoking History Generator (SHG) was developed by the Cancer Intervention and Surveillance Modeling Network (CISNET) to provide a simulation model that can be used to design effective interventions for controlling tobacco related disease. The US National Health Interview Surveys (NHIS) are a series of cross-sectional surveys of a probability sample of the US, which have been conducted since 1965 and these have included information on current, former and never smoking status, smoking intensity, as well as occasionally obtained retrospective detail on age at initiation and cessation. Age-period-cohort models are used to obtain single year of age and birth cohort parameter estimates adjusting for bias in the surveyed population that results from differential mortality due to smoking history. These yearly estimates of the conditional probability of smoking initiation and cessation, and the distribution of smoking intensity are incorporated into SHG, thus providing a simulation model designed to replicate the experience of the US population. By modifying these fundamental parameters, an investigator can evaluate the public health impact of interventions designed to change these fundamental parameters. In addition, it makes available a way of characterizing the distribution of exposure by enabling one to assess effects of control strategies that are not directly related to tobacco but dependent on differing risk resulting from cigarette smoking, e.g., cancer screening. The method is illustrated by estimating the impact of tobacco control strategies on US mortality.
Recent methods development for cancer screening
Presenter: Summer Han
When: Tuesday, July 12, 2016 Time: 2:00 PM - 3:30 PM
Room: Oak Bay 1-2 (Level 1)
Session Synopsis:
Simulating risk factors for lung cancer to optimize lung cancer screening guidelines
Lung cancer (LC) causes significant mortality worldwide. How to guide screening for individuals with potential risks of developing cancer is a critical public health concern. While current national guidelines for lung screening are based on smoking and age, the importance of risk-based screening was recently reported by several studies. The evidence showed that incorporating various risk factors for LC (such as family history and COPD) into selection criteria for screening provides efficiency in detecting LC cases. However, these results are based on specific trial populations such as the National Lung Screening Trial, and there is a need to evaluate the effectiveness of risk-based screening in the general population. As part of the Cancer Intervention and Surveillance Modeling Network (CISNET), microsimulation models were developed for identifying optimal lung screening strategies in the general population. While CISNET used an algorithm called Smoking History Generator (SHG) to simulate general population-level data for evaluating smoking-based screening strategies, there is currently no available population-level data of risk factors for lung cancer. Our aim is to develop an algorithm for simulating a set of correlated risk factors for LC in the general U.S. population. In developing the algorithm, we will consider the issues of: (i) exploiting smoking history data obtained by SHG, (ii) keeping correlation structures among risk factors as observed in the data available, and (iii) incorporating the prevalence information of risk factors in a given U.S. birth cohort. We propose to estimate the joint distribution of risk factors conditioning on smoking using one of the largest lung screening data from the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial. The joint distribution will be used to predict risk factor values for each individual within a given birth cohort, whose lifetime smoking histories are generated using SHG. A set of conditional distributions is derived using a chain rule of conditional probabilities to estimate and simulate from the joint distribution. We calibrate the estimated models using external prevalence data from U.S. Census and National Health and Nutrition Examination Survey. We demonstrate how the proposed algorithm can be used for answering important public health questions such as estimating the proportion of screen-eligible population in the U.S., which can help estimate national costs of lung screening.
Recent methods development for cancer screening
Presenter: Carolyn Rutter
When: Tuesday, July 12, 2016 Time: 2:00 PM - 3:30 PM
Room: Oak Bay 1-2 (Level 1)
Session Synopsis:
The Impact of Time to Diagnostic Assessment on Screening Efficacy
This presentation will demonstrate the use of microsimulation models to evaluate the clinical relevance of study findings. Observational studies of screening demonstrate that the time from receipt of a positive screening test to completion of recommended diagnostic follow-up ranges from 2 weeks (for breast cancer screening) to 3 months (for cervical and colorectal cancer screening). We used simulation models to predict the clinical impact of time to follow-up (T2FU). Models simulated a cohort of individuals who underwent guideline-based screening for breast, cervical, and colorectal cancer, with T2FU set to 0, 3, 6, 9 or 12 months. We then compared the predicted efficacy of screening at each T2FU, relative to no screening, with efficacy measured by the shift to an earlier stage at diagnosis, cancers prevented (possible for cervical and colorectal cancers), and life years gained (LYG). The hypothetical maximum achievable benefit occurs when T2FU=0. We found that the benefit of screening declined as T2FU increased. However, when T2FU is 3 months, screening retains much of the maximum possible benefit (breast: 82%, colorectal: 96%, cervical: 99%). Thus, screening remained highly efficacious across all cancers, even without immediate follow-up. These conclusions were bolstered by the use of multiple independently developed simulation models.
Recent methods development for cancer screening
Presenter: Nilanjan Chatterjee
When: Tuesday, July 12, 2016 Time: 2:00 PM - 3:30 PM
Room: Oak Bay 1-2 (Level 1)
Session Synopsis:
A tool for Individualized Coherent Absolute Risk Estimation (iCARE):
This talk with describe a R package, called the Individualized Coherent Absolute Risk Estimation (iCARE ), which allows researchers to quickly build models for absolute risk, and apply them to estimate an individual's risk of developing disease during a speci ed time interval, based on a set of user dened input parameters. An attractive feature of the software is that it gives users exibility to update models rapidly based on new knowledge of risk factors and tailor models to dierent populations. The tool requires three input arguments be specied: (1) a model for relative risk (2) an age-specic disease incidence rate and (3) the distribution of risk factors for the population of interest. The tool handles missing risk factor information for individuals for whom risks are to be predicted using a coherent approach where all estimates are derived from a single model after appropriate model averaging. The software allows single nucleotide polymorphisms (SNPs) to be incorporated into the model using published odds ratios and allele frequencies. We discuss the statistical framework, handling of missing data and genetic factors, and provide real data examples that demonstrate the utility of iCARE for building and applying absolute risk models, using breast cancer as an example.
