Surrogate Markers and the Association of Low-Dose CT Lung Cancer Screening With Mortality

Based on surrogate markers of benefit, single-arm studies projected that lung cancer screening with computed tomography (CT) would reduce lung cancer mortality by 80%.¹ Yet subsequent randomized clinical trials produced findings ranging from no reduction to a 20% reduction.² We formally evaluated the seemingly misleading surrogates.

In each of the 33 separate sites of the randomized National Lung Screening Trial (NLST)³ we assessed the relation between the magnitude of the actual benefit observed and 3 separate surrogates of benefit:

1. The proportion of stage 1 lung cancers diagnosed in the CT arm out of all lung cancer diagnoses, with a logarithmic transformation applied to improve adherence to mathematical assumptions.

2. Lung cancer–specific survival at 4 years among participants in the CT arm found to have stage 1 lung cancer (median follow-up was 53 months), with censoring was either for death from another cause or end of follow-up.

3. The ratio of the proportion of stage 1 lung cancers diagnosed between the study arms, with a logarithmic transformation applied.⁴

Using a weighted Pearson correlation coefficient (weights were the total number of lung cancer diagnoses at the study unless otherwise noted), we evaluated the relation between each surrogate and the difference in rates of death from lung cancer in the CT arm vs the chest x-ray (CXR) arm. Two sided P values <.05 were considered statistically significant. Statistical analyses were performed using R statistical software (version 3.2.2, R Foundation) including the weights and car packages. The NLST is a public use file exempt from institutional review board approval or written informed consent.

We analyzed data from 26 722 CT patients and 26 730 CXR patients, for a total of 53 452 patients. The mean (SD) age was 61 (5) years, with 31 530 (59%) men. The site-level magnitude of lung cancer mortality reduction was not associated with either the proportion of stage 1 lung cancers diagnosed in the CT arm (Pearson correlation, –0.08; 95% CI, –0.41 to 0.27; P = .65) or 4-year lung cancer-specific survival rate of participants in the CT arm diagnosed with stage 1 lung cancer (Pearson correlation, –0.28; 95% CI, –0.58 to 0.09; P = .13) (Figure). The difference in the proportion diagnosed with stage 1 lung cancer between study arms (means, CT 54% and CXR 34%; P < .001) was neither correlated with the difference in lung cancer mortality (Pearson correlation, –0.28; 95% CI, –0.57 to 0.07; P = .12) nor did it attenuate the main effect of the intervention in a regression analysis (P < .001) (Table).

Good surrogate markers of important endpoints can make studies more efficient; poor ones can be misleading. Early single-arm studies of CT screening that used surrogate markers estimated that CT screening would reduce lung cancer death by 80% or more; later randomized clinical trials estimated a benefit ranging from 0% to 20%. The recently published randomized Danish Lung Cancer Screening Trial (DLCST)⁵ illustrates the problem clearly. Although there was clear evidence that CT screening was associated with a far higher rate of stage 1 lung cancer diagnoses than the control group (50 vs 8, P < .001), lung cancer deaths in the 2 arms were essentially equal (39 screening vs 38 usual care).⁵

Some researchers have considered improvements in surrogate markers of screening benefits to be synonymous with evidence that CT screening is improving patient outcomes.⁶ Yet our analysis suggests that drawing such a conclusion is inappropriate. We conclude that these surrogate markers must not be used to evaluate the benefits of CT screening. We base this on not only our analyses, but the observation that these surrogate markers have been poor at predicting CT screening benefits both between and within studies, and the basic epidemiologic biases associated with screening, such as length biased sampling, are known to inflate the surrogates without improving outcome. A quest for more reliable predictive markers of CT screening impact on patient outcomes is warranted.

Corresponding Author: Kaitlin M. Woo, MS, Department of Epidemiology and Biostatistics, Memorial Sloan Kettering Cancer Center, 485 Lexington Ave, New York, NY 10017 (kaitlin.woo21@gmail.com).

Accepted for Publication: March 11, 2018.

Published Online: June 7, 2018. doi:10.1001/jamaoncol.2018.1263

Author Contributions: Ms Woo had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Gönen, Silvestri, Bach.

Acquisition, analysis, or interpretation of data: Woo, Gönen, Schnorr, Silvestri.

Drafting of the manuscript: Woo, Gönen, Silvestri, Bach.

Critical revision of the manuscript for important intellectual content: Woo, Gönen, Schnorr, Silvestri.

Statistical analysis: Woo, Gönen, Bach.

Administrative, technical, or material support: Schnorr.

Supervision: Gönen, Bach.

Conflict of Interest Disclosures: None reported.

Funding/Support: This work was supported by the National Institutes of Health (grant No. P30 CA008748 to K.M.W, M.G.).

Role of the Funder/Sponsor: The National Institutes of Health had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.