Risk prediction models versus simplified selection criteria to determine eligibility for lung cancer screening: an analysis of German federal-wide survey and incidence data

Anika Hüsing,Rudolf Kaaks

doi:10.1007/s10654-020-00657-w

Abstract

As randomized trials in the USA and Europe have convincingly demonstrated efficacy of lung cancer screening by computed tomography (CT), European countries are discussing the introduction of screening programs. To maintain acceptable cost-benefit and clinical benefit-to-harm ratios, screening should be offered to individuals at sufficiently elevated risk of having lung cancer. Using federal-wide survey and lung cancer incidence data (2008–2013), we examined the performance of four well-established risk models from the USA (PLCOM2012, LCRAT, Bach) and the UK (LLP2008) in the German population, comparing with standard eligibility criteria based on age limits, minimal pack years of smoking (or combination of total duration with average intensity) and maximum years since smoking cessation. The eligibility criterion recommended by the United States Preventive Services Taskforce (USPSTF) would select about 3.2 million individuals, a group equal in size to the upper fifth of ever smokers age 50–79 at highest risk, and to 11% of all adults aged 50–79. According to PLCOM2012, the model showing best concordance between numbers of lung cancer cases predicted and reported in registries, persons with 5-year risk ≥ 1.7% included about half of all lung cancer incidence in the full German population. Compared to eligibility criteria (e.g. USPSTF), risk models elected individuals in higher age groups, including ex-smokers with longer average quitting times. Further studies should address how in Germany these shifts may affect expected benefits of CT screening in terms of life-years gained versus the potential harm of age-specific increasing risk of over-diagnosis.

Highlights

Following results from the earlier US National Lung Cancer Screening trial (NLST) [1], recent findings from the DutchBelgian NELSON trial [2] and five smaller randomized trials in Italy, Denmark and Germany [3,4,5,6,7] confirm thatElectronic supplementary material The online version of this article contains supplementary material, which is available to authorized users.An alternative to concise inclusion criteria is the use of more refined models for the prediction of an individual’s1 3 Vol.:(0123456789)A
We focused on model comparisons with the NLST/United States Preventive Services Taskforce (USPSTF) and NELSON eligibility criteria, which empirically were at opposite ends on the scale of inclusiveness when applied to the German population (GEDA) (Supplemental Table 1)
The models differ with regard to the shape and strength of age-related increase in risk, inclusion of a gender effect, and variables and model coefficients accounting for smoking history (Supplemental Table 2b)

Summary

Introduction

Based on risk models, calibrated-in-the-large to federal-wide annual lung cancer incidence (Fig. 3a), we find that among ever-smokers aged 50–79 about 39–62% (NELSON), 37–61% (USPSTF), 32–51% (NLST) and 23–43% (55-80-30-15) of the lung cancer incidence are expected to occur in individuals eligible by these respective criteria, with estimated average 5-year risks of, respectively, 1.3–2.1%, 2.2–3.6%, 2.0–3.2% and 2.5–4.6%.

Results

Conclusion