LMO (FHCC) (Historical) Fred Hutch Cancer Center

To provide a mechanistic mathematical model of lung cancer development by histology for estimating the impact of alternative screening protocols to reduce lung cancer mortality in the US. The longitudinal multistage observation (LMO) model of lung cancer development and detection includes six pathways representing distinct histological subtypes (bronchioloalveolar, adenocarcinoma, large cell, squamous, other non-small cell, and small cell). On each pathway, the model represents five stages that may occur during progression from normal stem cells to malignant, then metastatic cells. The model assumes normal stem cells in the bronchus and lung may undergo at random two successive mutations to form a premalignant cell of a specific histological type. Premalignant cells may undergo clonal expansion or extinction through cell division and death (apoptosis), with a non-linear dose-response relationship accounting for the effects of cigarette smoking on the clonal expansion rate. The premalignant cells may undergo further mutation to become malignant cells. Maligant cells also undergo clonal expansion through division and apoptosis at rates that are generally faster than for premalignant cells. Malignant clones may be detected before metastasis through a stochastic (size based) observation process, or through further mutation events may generate metastatic cells, which also undergo clonal expansion and possibly observation. The stochastic observation processes relate the size of the malignant tumor to the probability of detection and stage, with different size-based sensitivities for X-ray screening, computerized tomography (CT) screening, symptomatic detection, and death from lung cancer. The model was fit to individual longitudinal data including lung cancer screening, incidence, and death outcomes from the National Lung Screening Trial (NLST) and the Prostate, Lung, Colorectal, and Ovarian (PLCO) cancer screening trials. The probability for successive (longitudinal) events depends on prior exposures and observation outcomes. Maximum likelihood methods were used to estimate the mutation, division, death, and observation parameters for each histological subtype. The model was then applied to simulated current, former, and never smokers generated by a Smoking Simulator program provided by the National Cancer Institute to estimate the effects of alternative CT screening protocols in the US population.

Contact: William Hazelton hazelton@fredhutch.org

Model Overview

The Longitudinal Multistage Observation LMO Model (Fred Hutchinson Cancer Research Center) was developed to estimate the impact of alternative screening protocols to reduce lung cancer mortality in the US. This longitudinal multistage observation model of lung cancer development and detection includes six pathways representing distinct histologic subtypes: bronchioloalveolar, adenocarcinoma, large cell, squamous, other non-small cell, and small cell. Using maximum-likelihood methods, the LMO Model was calibrated to individual-level smoking histories and longitudinal outcomes data in the National Lung Screening Trial (NLST), including separate pathways representing the six histologic subtypes. Parameters for each pathway were estimated separately by gender. After calibrating the model, it was validated based on independent data from the Prostate, Lung, Colorectal, and Ovarian (PLCO) cancer screening trial.

The LMO Model assumes that each histologic subtype of lung cancer may respond differently to cigarette smoke (Figure 1). Normal pulmonary stem cells may undergo two pathway-specific sequential mutational or epigenetic events to generate a premalignant cell. Premalignant cells may divide and undergo apoptosis (programmed cell death) or differentiation, leading to clonal expansion, or frequently to extinction of the clone due to the stochastic nature of the clonal expansion process where cell division and death events are assumed to occur at random. Premalignant cells in the developing clone may also mutate to generate a malignant cell. Malignant cells also divide and die, typically at faster rates, causing increased rates of clonal expansion. If the malignant clone does not become extinct, it may grow large enough to be detected through screening or other observation processes. The model estimates a rate for malignant cells to become metastatic, and it estimates clonal expansion rates of metastatic cells.

Figure 1. Illustration of the LMO model

CISNET-DFCI Model: Natural History of Breast Cancer

The LMO Model employed stochastic observation processes for possible longitudinal outcomes, which include negative screening tests, screen-based detection and diagnosis or symptomatic diagnosis, and subsequent outcomes such as cure, lung cancer death, censoring, or other-cause death. Each individual outcome, such as a negative or positive screening test, conditions the probability for subsequent outcomes. Thus, for example, a negative CT screening test indicates that the multistage process does not include large tumors at the time of screening, so the probability of symptomatic detection immediately following the test is low. However, the observation probability increases with time as small tumors not detected by the screen may grow larger.

The probability of a positive outcome from an observation depends on the tumor size and the sensitivity of the observation. We used likelihood methods to estimate all of the cell kinetic rates in the LMO, assuming separate sensitivities for tumor detection by low-dose helical computed tomography (CT) and chest X-ray (CXR). If a cancer is not detected through screening, it may grow to a larger size that causes symptomatic detection, modeled as an observation process with lower sensitivity. Calibration to the stage distribution at diagnosis and subsequent survival was based on malignant tumor size and the contribution from metastatic cells. After diagnosis, the cancer may be cured with different probability depending on histology. If the cancer is not cured, it may undergo further growth and metastatic spread, ultimately causing death. The LMO Model also allows for the possibility of overdiagnosis due to growth of indolent cancers, which may be detected through screening but are unlikely to undergo further growth that would lead to symptomatic diagnosis or death (1–4).

References

Hazelton WD, Goodman G, Rom WN, Tockman M, Thornquist M, Moolgavkar S, et al. Longitudinal multistage model for lung cancer incidence, mortality, and CT detected indolent and aggressive cancers. Mathematical biosciences. 2012;240(1):20-34. Epub 2012/06/19. doi: 10.1016/j.mbs.2012.05.008. PubMed PMID: 22705252; PubMed Central PMCID: PMC3412888.
Meza R, ten Haaf K, Kong CY, Erdogan A, Black WC, Tammemagi MC, et al. Comparative Analysis of 5 Lung Cancer Natural History and Screening Models That Reproduce Outcomes of the NLST and PLCO Trials. Cancer. 2014;120(11):1713-24. doi: Doi 10.1002/Cncr.28623. PubMed PMID: WOS:000336619700021.
McMahon PM, Meza R, Plevritis SK, Black WC, Tammemagi CM, Erdogan A, et al. Comparing Benefits from Many Possible Computed Tomography Lung Cancer Screening Programs: Extrapolating from the National Lung Screening Trial Using Comparative Modeling. PloS one. 2014;9(6). doi: ARTN e99978DOI 10.1371/journal.pone.0099978. PubMed PMID: WOS:000338506400012.
Han SS, Ten Haaf K, Hazelton WD, Munshi VN, Jeon J, Erdogan SA, et al. The impact of overdiagnosis on the selection of efficient lung cancer screening strategies. International journal of cancer Journal international du cancer. 2017. doi: 10.1002/ijc.30602. PubMed PMID: 28073150.