Skip to Main Content

An official website of the United States government

MRI screening for lung cancer: Markov modeling using the NLST CT outcomes data

Principal Investigator
Name
Mark Schiebler
Degrees
M.D.
Institution
UW-Madison School of Medicine and Public Health
Position Title
Professor of Radiology
Email
About this CDAS Project
Study
NLST (Learn more about this study)
Project ID
NLST-265
Initial CDAS Request Approval
Dec 15, 2016
Title
MRI screening for lung cancer: Markov modeling using the NLST CT outcomes data
Summary
We propose to use Markov modeling for the use of MRI in the primary screening of lung cancer to determine its range of outcomes in this setting.[1] We hope to download the actual data from NLST for all of the various nodule sizes (benign and malignant) detected at CT and then create a statistical model (Markov) that can be used to estimate what would have happened if MRI was used instead.
We will vary the efficacy of MRI for determining the presence of a malignant nodule. The published the sensitivity of MRI for lung nodules of 4-8 mm in diameter has a range of 60 – 90% and reaches 100% for lesions larger than 8 mm in diameter. (2) Recently Cieszanowski et al. compared the diagnostic performance of MRI with different T2- and T1-weighted sequences for the detection of 113 small lung nodules in 50 patients and found 1 – 48 false positive reads depending on the sequence type. In other studies the false positive rate was only 5%.[3] When projecting a MRI-based lung cancer screening trial, it should be considered to use a multiparametric approach, combining at least 2 or 3 sequences, to minimize false positive studies.

The Markov model approach to screening would try to avoid conducting a large expensive study by using existing data on - sensitivity/ specificity, lung ca incidence, cancer growth rate to predict overall survival.

In summary, this “gedank’ experiment will help to use this new statistical method to determine the expected range of utility of using MRI for this screening.
References
[1] J. Biederer, Y. Ohno, H. Hatabu, M.L. Schiebler, E.J. van Beek, J. Vogel-Claussen, H.U. Kauczor, Screening for lung cancer: Does MRI have a role?, Eur J Radiol, (2016).
[2] G. Sommer, J. Tremper, M. Koenigkam-Santos, S. Delorme, N. Becker, J. Biederer, H.-U.
Kauczor, C.P. Heussel, H.-P. Schlemmer, M. Puderbach, Lung nodule detection in a high-risk
population: Comparison of magnetic resonance imaging and low-dose computed tomography,
Eur. J. Radiol. 83 (2014) 600–605. doi:10.1016/j.ejrad.2013.11.012.
[3] A. Cieszanowski, A. Lisowska, M. Dabrowska, P. Korczynski, M. Zukowska, I.P. Grudzinski, R.
Pacho, O. Rowinski, R. Krenke, MR Imaging of Pulmonary Nodules: Detection Rate and Accuracy
of Size Estimation in Comparison to Computed Tomography, PloS One. 11 (2016) e0156272.
doi:10.1371/journal.pone.0156272
Aims

1. To build a Markov model for lung cancer screening using the NLST CT data that is based on the size of nodule detection, efficacy for maligancy determination, costs, morbidity and mortality.
2. Using this Markov model determine the expected range of outcomes using MRI efficacy data using the sensitiviy and specificity of the detection of an 8 mm nodule is 80%.
3. Determine the range of cost effectiveness of using MRI based on nodule size 1-30mm) and specificity for malignancy ( 0-90%)

Collaborators

James Carr, MD, Northwestern University, Feinberg School of Medicine, Radiology
Bradley Allen,M.D., Northwestern University, Feinberg School of Medicine, Radiology
Christopher Francois, M.D., UW-Madison School of Medicine and Public Health, Radiology
Hans-Ulrich Kauczor, M.D., Heidelberg University, University Clinic, Radiology
Gordon Hazen, PhD, Northwestern University. Department of Mathematics