Evaluating long term outcomes and over diagnosis in lung cancer screening using the NLST data
Principal Investigator
Name
Dongfeng Wu
Degrees
PhD
Institution
University of Louisville
Position Title
Associate Professor
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-73
Initial CDAS Request Approval
May 16, 2014
Title
Evaluating long term outcomes and over diagnosis in lung cancer screening using the NLST data
Summary
We have developed a new method for evaluating long term outcomes and over diagnosis due to regular screening. Participants in screening exams can be separated into 4 disjoint groups: True-early-detection, No-early-detection, Over-diagnosis and Symptom-free-life. All participants will eventually fall into one of the four groups based on their diagnosis status and whether symptoms would have appeared before death. Here are the definition for each group.
Group 1: Symptom-free-life. A man in Group 1 took part in screening exams, but lung cancer was never detected and ultimately he died of other causes.
Group 2: No-early-detection. A man in Group 2 took part in screening exams, but disease manifested itself clinically and was not detected by scheduled screening exam.
Group 3: True-early-detection. A man in Group 3 was diagnosed with lung cancer at a scheduled screening exam and his clinical symptoms would have appeared before his death.
Group 4: Over diagnosis. A man in Group 4 was diagnosed with lung cancer at a scheduled screening exam but his clinical symptoms would NOT have appeared before his death.
Probability of each group has been derived and proved correct; it is a function
of screening sensitivity, sojourn time distribution and transition density from disease-free to preclinical state, future screening schedule and age at first screening. Human lifetime is treated as a random variable, which is derived from US Social Security Administration (SSA) actuarial life table in this method.
We want to apply this new method to the NLST screening data, to explore how the percentage of each group changes, and to explore the percentage of over-diagnosis vs. true-early-detection among those detected by scheduled screening exams. To achieve this goal, we first need to estimate the three key parameters (sensitivity, sojourn time distribution, and transition density) using the NLST screening data, then insert the parameters into our probability formulae. Future screening schedules and initial ages to screen will be studied by extensive simulation. We hope this will provide helpful information on long term outcomes and over diagnosis using low-dose CT and/or chest x-ray in lung cancer screening.
Group 1: Symptom-free-life. A man in Group 1 took part in screening exams, but lung cancer was never detected and ultimately he died of other causes.
Group 2: No-early-detection. A man in Group 2 took part in screening exams, but disease manifested itself clinically and was not detected by scheduled screening exam.
Group 3: True-early-detection. A man in Group 3 was diagnosed with lung cancer at a scheduled screening exam and his clinical symptoms would have appeared before his death.
Group 4: Over diagnosis. A man in Group 4 was diagnosed with lung cancer at a scheduled screening exam but his clinical symptoms would NOT have appeared before his death.
Probability of each group has been derived and proved correct; it is a function
of screening sensitivity, sojourn time distribution and transition density from disease-free to preclinical state, future screening schedule and age at first screening. Human lifetime is treated as a random variable, which is derived from US Social Security Administration (SSA) actuarial life table in this method.
We want to apply this new method to the NLST screening data, to explore how the percentage of each group changes, and to explore the percentage of over-diagnosis vs. true-early-detection among those detected by scheduled screening exams. To achieve this goal, we first need to estimate the three key parameters (sensitivity, sojourn time distribution, and transition density) using the NLST screening data, then insert the parameters into our probability formulae. Future screening schedules and initial ages to screen will be studied by extensive simulation. We hope this will provide helpful information on long term outcomes and over diagnosis using low-dose CT and/or chest x-ray in lung cancer screening.
Aims
Aim 1: Apply the likelihood method we developed to the NLST data, to obtain accurate estimation of sensitivity, sojourn time distribution and transition density from disease-free state to preclinical state (Wu et al, Biometrics 2005)
Aim 2: Apply the new probability method to the NLST data by using the estimation from Aim 1, to predict the proportion of symptom-free-life, no-early-detection, true-early-detection, and over-diagnosis in the long term, and to predict the proportion of over diagnosis vs. true-early-detection among the screen-detected cases (Wu et al, Statistica Sinica 2014).
Related Publications
-
Inference of Sojourn Time and Transition Density using the NLST X-ray Screening Data in Lung Cancer.
Rahman F, Wu D
Med Res Arch. 2021 May; Volume 9 (Issue 5) PUBMED -
Evaluating Long-Term Outcomes via Computed Tomography in Lung Cancer Screening
Dongfeng Wu, Ruiqi Liu, Beth Levitt, Tom Riley, Kathy B. Baumgartner
J Biom Biostat. 2016 -
Bayesian Estimation of the Three Key Parameters in CT for the National Lung Screening Trial Data
Ruiqi Liu, Beth Levitt, Tom Riley, Dongfeng Wu
J Biom Biostat. 2015