Meta-Analysis of Prognostic Studies Evaluating Time-Dependent Diagnostic and Predictive Capacities of Biomarkers

Prognostic biomarker studies, which examine the association between biomarkers and patients’ prognoses, have played important roles in clinical decision making. Since prognostic studies are often conducted with small sample sizes in a limited number of ce

  • PDF / 295,262 Bytes
  • 17 Pages / 439.37 x 666.142 pts Page_size
  • 55 Downloads / 185 Views

DOWNLOAD

REPORT


Abstract Prognostic biomarker studies, which examine the association between biomarkers and patients’ prognoses, have played important roles in clinical decision making. Since prognostic studies are often conducted with small sample sizes in a limited number of centers, meta-analysis is expected to be a powerful tool to obtain sound evidence on prognostic biomarkers. However, the application of meta-analysis of prognostic studies has been limited partly due to the lack of sound statistical methods. In this chapter, we introduce some recently developed methods useful for the evaluation of diagnostic or predictive capacities of biomarkers for binary or time-to-event outcomes. In addition, we newly present a novel method to estimate the time-dependent positive and negative predictive value curves based on meta-analysis.









Keywords Cutoff value Diagnostic studies Prognostic studies Meta-analysis Time-dependent predictive value curve Time-dependent receiver operating characteristics



1 Introduction Prognostic studies have been widely conducted to determine whether specific biomarkers or other demographic factors such as age are associated with patients’ prognoses. Such studies are very useful to understand disease progression and to S. Hattori (&) Department of Integrated Medicine, Biomedical Statistics, Osaka University, 2-2, Yamada-Oka, Suita, Osaka 565-0871, Japan e-mail: [email protected] X.-H. Zhou HSR&D Center of Excellence, VA Puget Sound Health Care System, Seattle, USA X.-H. Zhou Department of Biostatistics, University of Washington, Seattle, WA 98195, USA © Springer Nature Singapore Pte Ltd. 2017 S. Matsui and J. Crowley (eds.), Frontiers of Biostatistical Methods and Applications in Clinical Oncology, DOI 10.1007/978-981-10-0126-0_16

257

258

S. Hattori and X.-H. Zhou

identify subgroups of patients with poor/good prognoses. Therefore, they have played important roles in clinical decision making, healthcare policy, and patient management [18]. However, as noted by several authors [1, 19, 23, 31, 32], prognostic studies are often conducted with small sample size data from a single or a few centers. Therefore, the findings in a prognostic study should be further assessed, and meta-analysis is expected to be useful for this purpose [32]. In the context of evaluating treatment efficacy in clinical trials, meta-analysis based on multiple independent studies has been widely applied as a powerful tool to derive more reliable evidence. Findings by well-conducted meta-analyses are regarded as highly reliable evidence [2]. However, the application of meta-analyses to prognostic studies has been very limited. Among meta-analyses of prognostic studies recently reported, Becattini et al. [3] conducted a meta-analysis to examine the association between troponins and short-term death (binary outcomes) in patients with an acute pulmonary embolism. They reported the combined odds ratio of the high-expression group of troponin relative to the low-expression group for short-term death. Meta-analys