ABSTRACT
References ............................................................................................................393
Biomarkers that are both sensitive and specific for tumor regrowth or disease progression are increasingly becoming available and routinely monitored during the course of regular follow-up of patients treated for cancer. Some specific examples are prostatespecific antigen (PSA) for prostate cancer, the CA19-9 antigen for pancreatic cancer, and the CA125 antigen for ovarian cancer. Discovery of new biomarkers through use of modern high-dimensional proteomic techniques is rapidly gaining speed (Early Detection Research Network, 2004). Obtained by a simple blood test, biomarkers are a less expensive and invasive alternative to other clinical monitoring modalities such as ultrasound or bone scan. Retrospective studies for several cancers and affiliated biomarkers have found that increases or changes in the biomarker can precede clinical evidence of recurrence of disease by time periods ranging from 1 month to 2 years depending on the context.1-3 Following some therapies, biomarkers initially decline, and the rates of such initial declines provide early independent prognostic indicators of
clinical outcome; for examples in PSA following chemotherapy for advanced hormonerefractory prostate cancer see Ref. 4. Optimized predictions of clinical event times then should incorporate prognostic information from the full posttherapy biomarker sequence. Such methods should also utilize the repeated measures to adjust for measurement error inherent in a biomarker process. Joint statistical models for longitudinal and event time data provide promising tools that satisfy these criteria.
