Robust risk prediction with biomarkers under two-phase stratified cohort design

Conditional concordance-assisted learning under matched case-control design for combining biomarkers for population screening

Incorporating promising biomarkers into cancer screening practices for early-detection is increasingly appealing because of the unsatisfactory performance of current cancer screening strategies. The matched case-control design is commonly adopted in biomarker development studies to evaluate the discriminative power of biomarker candidates, with an intention to eliminate confounding effects. Data from matched case-control studies have been routinely analyzed by the conditional logistic regression, although the assumed logit link between biomarker combinations and disease risk may not always hold. We propose a conditional concordance-assisted learning method, which is distribution-free, for identifying an optimal combination of biomarkers to discriminate cases and controls. We are particularly interested in combinations with a clinically and practically meaningful specificity to prevent disease-free subjects from unnecessary and possibly intrusive diagnostic procedures, which is a top priority for cancer population screening. We establish asymptotic properties for the derived combination and confirm its favorable finite sample performance in simulations. We apply the proposed method to the prostate cancer data from the carotene and retinol efficacy trial (CARET).

Evaluating Discrimination of a Lung Cancer Risk Prediction Model Using Partial Risk-Score in a Two-Phase Study

BACKGROUND

Independent validation of risk prediction models in prospective cohorts is required for risk-stratified cancer prevention. Such studies often have a two-phase design, where information on expensive biomarkers are ascertained in a nested substudy of the original cohort.

METHODS

We propose a simple approach for evaluating model discrimination that accounts for incomplete follow-up and gains efficiency by using data from all individuals in the cohort irrespective of whether they were sampled in the substudy. For evaluating the AUC, we estimated probabilities of risk-scores for cases being larger than those in controls conditional on partial risk-scores, computed using partial covariate information. The proposed method was compared with an inverse probability weighted (IPW) approach that used information only from the subjects in the substudy. We evaluated age-stratified AUC of a model including questionnaire-based risk factors and inflammation biomarkers to predict 10-year risk of lung cancer using data from the Prostate, Lung, Colorectal, and Ovarian Cancer (1993-2009) trial (30,297 ever-smokers, 1,253 patients with lung cancer).

RESULTS

For estimating age-stratified AUC of the combined lung cancer risk model, the proposed method was 3.8 to 5.3 times more efficient compared with the IPW approach across the different age groups. Extensive simulation studies also demonstrated substantial efficiency gain compared with the IPW approach.

CONCLUSIONS

Incorporating information from all individuals in a two-phase cohort study can substantially improve precision of discrimination measures of lung cancer risk models.

IMPACT

Novel, simple, and practically useful methods are proposed for evaluating risk models, a critical step toward risk-stratified cancer prevention.