Novel Multi-State Models for Lung Cancer Progression
Principal Investigator
Name
Ling Lan
Degrees
MD PhD
Institution
Georgia Regents University
Position Title
Assistant Professor
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-13
Initial CDAS Request Approval
Mar 20, 2013
Title
Novel Multi-State Models for Lung Cancer Progression
Summary
The overall goal of this proposal is to better understand lung cancer progression using novel nonparametric multistate regression models based on current status data from National Lung Screening Trial (NLST). The proposed models will provide both marginal and conditional estimations for the start, end and elapsed/lag time at each state of cancer development including: healthy, cancer stage I, II, III and IV, and death. Lung cancer is the leading cause of cancer-related death in the United States. Early intervention has shown improved long term survival in about 70% of cancer patients at early stage. The largest randomized screening trial to date, NLST, demonstrates screening with the use of low dose CT reduces mortality from lung cancer as compared to chest radiology among current and former heavy smokers. Clinical practitioners still have questions remain unsettled before the large scale enforcement of lung screening: What is the risk for preclinical stage II (or, I, III, IV) lung cancer for a tobacco user at age 60? How long will it naturally take for a patient to progress into locally advanced or later stage cancer from early stage disease? Should the age for initial screening be gender specific or smoking status specific (current and former smoker)? Thus, the proposed studies to answer questions mentioned above via the investigation of the time course of lung cancer development can be justified based on cancer prevention and treatment concerns.
Limited literatures are available for modeling multistate disease progression with current status data (survey data). Data collected upon screening tests in NLST are current status, which only contains information at inspection and the latent time for cancer is unobservable. However, our recent work on nonparametric estimation for marginal quantities in multistate models have decipher the natural time course for puberty development among boys and girls aged 8-18 in the US with data collected from the third National Health and Nutritional Examination survey. Thus, re-defining lung cancer progression based on the evidence based nonparametric model will serve as a bench mark to current parametric and semi-parametric regression models which are prone to model misspecification error.
Therefore, the studies proposed here are intended to fill a blank in the fields of modeling natural history of lung cancer at individual level for data with limited information. It will be useful for prioritizing individuals for screening and participation in clinical trials of chemoprevention.
Limited literatures are available for modeling multistate disease progression with current status data (survey data). Data collected upon screening tests in NLST are current status, which only contains information at inspection and the latent time for cancer is unobservable. However, our recent work on nonparametric estimation for marginal quantities in multistate models have decipher the natural time course for puberty development among boys and girls aged 8-18 in the US with data collected from the third National Health and Nutritional Examination survey. Thus, re-defining lung cancer progression based on the evidence based nonparametric model will serve as a bench mark to current parametric and semi-parametric regression models which are prone to model misspecification error.
Therefore, the studies proposed here are intended to fill a blank in the fields of modeling natural history of lung cancer at individual level for data with limited information. It will be useful for prioritizing individuals for screening and participation in clinical trials of chemoprevention.
Aims
Aim 1. Development of a nonaparametric natural history model of lung cancer with current status data incorporating the dependence of the progression rates on risk factors at individual level.
Aim 2. Cross validation of the model predictive power on the incidence of preclinical lung cancer at various cancer stages.
Aim 3. R software package will be constructed to facilitate the usage of the proposed model.
Collaborators
Ella A. Kazerooni, M.D. University of Michigan
Somnath Datta, PhD. University of Louisville
Zhonglin Hao, M.D. Georgia Regents University