Two-phase designs to evaluate the prognostic value of a biomarker on a survival outcome

Date:

Monday, 15 April 2019

Time:

10.00am – 11.00am

Venue:

Seminar Room 1, Level 8
Tahir Foundation Building (MD1)
Saw Swee Hock School of Public Health
National University of Singapore
12 Science Drive 2, S(117549)

Synopsis:

In many epidemiological and clinical studies, some information (e.g., exposure or outcome variables) may not be available for the whole cohort because the additional information is ascertained in a selected subsample for feasibility and economic reasons. These studies are part of a broad category of study designs called two-phase designs. Common examples of such designs include the nested case-control and the case-cohort designs. We applied the two-phase design on a cohort of children with acute lymphoblastic leukemia from clinical trial (AIEOP-ALL2000) with stored biological samples at diagnosis to assess the prognostic value of novel biomarkers. In this context, we compare the performance of different sampling designs by exploring the use of different types of variables (e.g. surrogate, risk factor…) as strata in the clinical trial cohort, in order to determine the optimal subcohort to measure the biomarker. Moreover, since no estimator of the cumulative incidence that accounts for competing events is available in the two-phase context, we developed a nonparametric estimator for it. This is relevant in the presence of multiple types of events, where the estimation of each specific event type is needed for a comprehensive evaluation of the performance of the biomarker. The proposed estimator handles a general sampling design by using the weights that are related to the sampling probabilities. The variance of the estimator is derived from the influence function of the subdistribution hazard. An R function is available to apply such estimator within the survey package. The proposed method has good performance from the simulations and was applied to estimate the incidence at different relapse sites by the biomarker value in children with acute lymphoblastic leukemia. Moreover the estimator was applied to account for lost to follow-up in the evaluation of the number of HIV patients who stop to access care after starting antiretroviral therapy (ART) in resource limited settings.

About the speaker:

She has a Master in Biostatistics and PhD in Biomedical Statistics from the University of Milano. Her research interests include the development of statistical methods for survival outcome, especially the two-phase designs on genetic/biological data nested in large clinical or epidemiological cohorts, joint modeling of longitudinal measures and survival outcome and evaluation of time-dependent treatment effect. She collaborates in the design and analysis of several clinical studies, such as, the evaluation of the prognostic value of biomarkers in childhood acute lymphoblastic leukemia, evaluation of self-care in diabetic patients, evaluation of the anticoagulant treatment in end-stage renal disease patients with atrial fibrillation, and population-based studies on cancer risk. She has published 52 papers, mainly as first or second author, in clinical, epidemiologic, nursing and statistical peer-review journals and presented her work in 30 international conferences.