A modeling analysis to compare eligibility strategies for lung cancer screening in Brazil.

BACKGROUND

Country-specific evidence is needed to guide decisions regarding whether and how to implement lung cancer screening in different settings. For this study, we estimated the potential numbers of individuals screened and lung cancer deaths prevented in Brazil after applying different strategies to define screening eligibility.

METHODS

We applied the Lung Cancer Death Risk Assessment Tool (LCDRAT) to survey data on current and former smokers (ever-smokers) in 15 Brazilian state capital cities that comprise 18% of the Brazilian population. We evaluated three strategies to define eligibility for screening: (1) pack-years and cessation time (≥30 pack-years and <15 years since cessation); (2) the LCDRAT risk model with a fixed risk threshold; and (3) LCDRAT with age-specific risk thresholds.

FINDINGS

Among 2.3 million Brazilian ever-smokers aged 55-79 years, 21,459 (95%CI 20,532-22,387) lung cancer deaths were predicted over 5 years without screening. Applying the fixed risk-based eligibility definition would prevent more lung cancer deaths than the pack-years definition [2,939 (95%CI 2751-3127) vs. 2,500 (95%CI 2318-2681) lung cancer deaths], and with higher screening efficiency [NNS=177 (95%CI 170-183) vs. 205 (95%CI 194-216)], but would tend to screen older individuals [mean age 67.8 (95%CI 67.5-68.2) vs. 63.4 (95%CI 63.0-63.9) years]. Applying age-specific risk thresholds would allow younger ever-smokers to be screened, although these individuals would be at lower risk. The age-specific thresholds strategy would avert three-fifths (60.1%) of preventable lung cancer deaths [ = 2629 (95%CI 2448-2810)] by screening 21.9% of ever-smokers.

INTERPRETATION

The definition of eligibility impacts the efficiency of lung cancer screening and the mean age of the eligible population. As implementation of lung screening proceeds in different countries, our analytical framework can be used to guide similar analyses in other contexts. Due to limitations of our models, more research would be needed.