Annals of Internal Medicine March 04, 2014

Benefits and Harms of Computed Tomography Lung Cancer Screening Strategies: A Comparative Modeling Study for the U.S. Preventive Services Task Force (CISNET / MISCAN-Lung)

de Koning HJ, Meza R, Plevritis SK, ten Haaf K, et al. (CISNET Lung Cancer Working Group)

Bottom Line

A landmark comparative microsimulation modeling study (featuring the MISCAN-Lung model) that extrapolated National Lung Screening Trial (NLST) data to determine that annual LDCT screening for individuals aged 55 to 80 years with heavy smoking histories optimizes the balance between mortality benefit and screening harms.

Key Findings

1. Evaluated 576 distinct screening scenarios using 5 independent microsimulation models (including the well-known MISCAN-Lung model) calibrated to the incidence and mortality outcomes of the NLST and PLCO trials.
2. Determined that annual low-dose computed tomography (LDCT) screening from age 55 to 80 years for individuals with a ≥30 pack-year smoking history (and ≤15 years since quitting) yielded the most efficient balance of benefits and harms.
3. This optimal strategy reduced lung cancer mortality by an estimated 14% (range across the 5 models: 8.2% to 23.5%) over a lifetime.
4. The screening program averted an estimated 521 lung cancer deaths per 100,000 persons (range: 176 to 1,058) while generating an estimated 50,911 LDCT screens per 100,000 persons.
5. Estimated that overdiagnosis—the detection of cancers that would not have caused clinical symptoms during a patient's lifetime—accounted for 11.9% (range: 8.3% to 24.6%) of all screen-detected lung cancers.

Study Design

Design
Microsimulation Modeling Study
N/A
Sample
100,000
Patients
Duration
Lifetime
Median
Setting
Simulated U.S.
Population Simulated hypothetical U.S. cohorts (e.g., 1950 birth cohort) generated using the Smoking History Generator, matching the demographics, overall mortality, and lung cancer risk profiles evaluated in the National Lung Screening Trial (NLST).
Intervention 576 simulated low-dose computed tomography (LDCT) screening strategies varying by age of initiation/cessation, smoking history (pack-years and years since quitting), and screening interval (annual vs. biennial).
Comparator No screening (baseline natural history of lung cancer without LDCT intervention).
Outcome Mean reduction in lung cancer mortality, absolute lung cancer deaths averted, life-years gained, number of LDCT screens performed, false-positive screens, and overdiagnosis rate.

Study Limitations

Microsimulation models inherently rely on complex mathematical assumptions regarding the unobservable natural history of lung cancer, such as the dwell time of preclinical disease and tumor doubling times.
The models assumed 100% population adherence to annual screening and diagnostic follow-up algorithms, which significantly overestimates real-world clinical compliance and actual population benefit.
The simulations were heavily calibrated to the NLST population, which was a highly selected, predominantly white, and generally healthier cohort, potentially limiting the generalizability of the harms and overdiagnosis rates to a broader, sicker population.

Clinical Significance

This was the foundational modeling analysis that translated the 3-year data from the NLST into a lifetime public health strategy. By establishing the risk-benefit thresholds for different ages and smoking histories, this paper directly informed the influential 2014 USPSTF lung cancer screening guidelines (which originally set the screening stopping age at 80, expanding on the NLST's cutoff of 74). Note: 'MILES' in the query is interpreted as a reference to the MISCAN-Lung model or the broader CISNET microsimulation modeling analyses of the NLST.

Historical Context

The 2011 NLST proved unequivocally that LDCT screening could reduce lung cancer mortality by 20% compared to chest radiography. However, the trial was limited to three annual screens and capped enrollment at age 74. Policymakers required long-term projections to design a population-level screening program. The USPSTF commissioned the CISNET consortium—utilizing advanced models like the Erasmus MC MISCAN-Lung model—to project lifetime outcomes, optimizing the age parameters, intervals, and smoking eligibility criteria for national implementation.

Guided Discussion

High-yield insights from every perspective

Med Student
Medical Student

Why is low-dose computed tomography (LDCT) effective for lung cancer screening compared to traditional chest radiography, and what is the primary clinical rationale for targeting the 55-80 age group with a heavy smoking history?

Key Response

LDCT detects smaller, early-stage (Stage I) nodules that CXR misses, enabling curative surgical resection. The age and smoking criteria target the population with the highest pre-test probability of lung cancer, maximizing positive predictive value and minimizing harms like radiation exposure and false positives in low-risk individuals.

Resident
Resident

When evaluating a 60-year-old patient who quit smoking 10 years ago after a 35 pack-year history, what are the primary harms of LDCT screening you must discuss during a shared decision-making visit, as highlighted by the MISCAN-Lung modeling study?

Key Response

Shared decision-making is a guideline requirement. Residents must discuss high false-positive rates leading to downstream invasive procedures and anxiety, incidental findings, radiation exposure (quantified by the model as radiation-induced cancers), and overdiagnosis (treating indolent cancers).

Fellow
Fellow

The CISNET modeling study highlighted the risk of overdiagnosis in lung cancer screening. How do we clinically define overdiagnosis in this context, and how do models attempt to quantify it compared to the observed NLST data?

Key Response

Overdiagnosis refers to detecting a histologically malignant tumor that would not have caused symptoms or death in the patient's lifetime. Models estimate this by comparing the expected incidence of lung cancer in an unscreened cohort with the incidence in a screened cohort, adjusting for lead time and competing mortality such as cardiovascular disease.

Attending
Attending

The USPSTF recommendations heavily relied on the CISNET microsimulation models extrapolating NLST data. In clinical practice, how do real-world patient demographics and adherence rates challenge the mortality benefit-to-harm ratio projected by these models?

Key Response

Trial populations (like NLST) are generally healthier, younger, and highly adherent to annual screening and follow-up. Real-world populations have more comorbidities, lower adherence, and community radiology may have lower specificity than specialized academic centers, potentially increasing false positives and diminishing the projected mortality benefit.

Scholarly Review

Critical appraisal through the lens of expert reviewers and guideline development

PhD
PhD

The study utilized the MISCAN-Lung microsimulation model to extrapolate NLST findings. What are the methodological advantages of using a continuous-time semi-Markov model for this task, and what structural uncertainties are introduced when estimating natural history parameters?

Key Response

Microsimulation models individual patient trajectories with competing risks, vital for an older smoking population. Structural uncertainties arise because unobservable natural history (e.g., transition from preclinical to clinical stage) must be calibrated using observable trial data; different structural assumptions can yield divergent optimal screening strategies.

Journal Editor
Journal Editor

As a peer reviewer assessing a comparative modeling study like this CISNET report, how do you critically evaluate the calibration of the models against trial data, specifically regarding the assumption of perfect versus real-world screening adherence?

Key Response

Reviewers must flag that models often assume 100% adherence to annual screening (a maximum efficacy scenario) to compare strategies, which vastly overestimates real-world effectiveness. Evaluating whether authors included robust sensitivity analyses for non-adherence and loss to follow-up is critical for determining the study's external validity.

Guideline Committee
Guideline Committee

The 2014 USPSTF guidelines based on this CISNET modeling recommended screening for ages 55-80 with a 30 pack-year history. How did subsequent evidence drive the USPSTF in 2021 to update these criteria to ages 50-80 and a 20 pack-year history, and what was the anticipated impact on health equity?

Key Response

The 2021 update lowered the age and pack-year thresholds to increase sensitivity and directly address racial disparities, as data showed Black and female smokers develop lung cancer at lower pack-year exposures. Updated CISNET modeling confirmed this broader strategy maintained an acceptable trade-off between life-years gained and screening harms.

Clinical Landscape

Noteworthy Related Trials

2011

National Lung Screening Trial (NLST)

n = 53,454 · NEJM

Tested

Low-dose computed tomography (LDCT) screening

Population

High-risk current and former smokers aged 55 to 74 years

Comparator

Single-view posteroanterior chest radiography

Endpoint

Lung cancer mortality

Key result: LDCT screening resulted in a 20% relative reduction in lung cancer mortality compared to chest radiography.
2019

Multicentric Italian Lung Detection (MILD) Trial

n = 4,099 · Ann Oncol

Tested

Prolonged low-dose CT screening

Population

Current or former smokers aged 49 to 75 years with at least 20 pack-years

Comparator

No screening

Endpoint

Overall and lung cancer specific mortality

Key result: Prolonged LDCT screening beyond 5 years achieved a 39% reduction in lung cancer mortality at 10 years compared to the control arm.
2020

NELSON Trial

n = 15,792 · NEJM

Tested

Volume-based low-dose CT screening

Population

High-risk current and former smokers aged 50 to 74 years

Comparator

No screening

Endpoint

Lung cancer mortality

Key result: Volume-based LDCT screening led to a 24% reduction in lung cancer mortality in men and a 33% reduction in women compared to no screening at 10 years.

Tailored to your role

Want this tailored to you?

Add your specialty or training stage to get role-specific takeaways and more questions.

Personalize this analysis