Annual Report 2019
Division of Biostatistical Research
Taro Shibata, Aya Kuchiba
Introduction
Biostatisticians at the NCC have two key roles. The first is to contribute to providing the best evidence as scientists through planning and implementing statistical designs and methods, and providing statistical considerations in every stage of a subject-matter research project. The second is to develop novel statistical and mathematical methods, motivated by methodological issues that arise in various scientific disciplines. While the Biostatistics Division at the Center for Research Administration and Support makes comprehensive efforts to work on both roles, the Division of Biostatistical Research at the Center for Public Health Sciences particularly focuses on the second role and responsibility.
Research activities
Investigators in the Division of Biostatistical Research are actively working on identifying and solving statistical problems, and developing novel methodology in various research areas.
1. Epidemiology and Prevention Research Area
We are involved in the study for developing prediction models for upper aerodigestive tract cancer including gene-environmental interaction in the case-cohort study. We also perform causal mediation analysis with a potential outcome framework on cohort data to get deeper insights into the mechanism of cancer development. In addition, we have developed prediction models for major cancers in Japanese people.
2. Survivorship Research Area
We are involved in projects on breast cancer survivors: the questionnaire survey on physical activity and the development of exercise programs. We have worked on the study protocols and the study design papers.
3. Clinical Research Area
Development and utilization of clinical trial database
We are regularly updating information on ongoing clinical trials by establishing collaborative relationships with other sources providing cancer information and the project teams of cancer research.
Support system for endoscopic diagnosis with artificial intelligence
We have developed the study protocols for evaluating the diagnostic performance of support systems for endoscopic diagnosis with artificial intelligence. We have also worked on research papers for diagnostic performance evaluation using existing image data.
Reliability of real-world data (RWD) and patient registry/study design and statistical methods for utilizing RWD
In the projects supported by AMED, we provide perspectives on the requirement for reliability of RWD and patient registry data for pharmaceutical approval. We also review the study designs and statistical methods for utilizing RWD.
4. Methodological Research
Prediction performance for multi-category outcomes
We are working on developing methods to assess prediction performance for outcomes with more than two categories (e.g., breast cancer with ER+, breast cancer with ER-, no cancer) and presented preliminary results of examining characteristics of the proposed index at an international statistical conference (3rd Pacific Rim Cancer Biostatistics Conference 2019). We are working on finalizing the research paper. In addition, we have proposed the methods for interval estimate for F1 score and gave a presentation at the Japanese Joint Statistical Meeting.
Latent subgroups
We are working on the statistical method to detect latent subgroups for treatment response.
Future prospects
Biostatistics has a close connection with real applications. Working on both methodological projects and real applications is important for future developments in biostatistical research. In 2019, we continued to study the new area of dissemination and implementation (D&I) research, in addition to the above. We have opportunities to talk about the study design and statistical methodology at symposiums.
We endeavor to establish a powerful collaborative relationship with researchers throughout the NCC and to identify critical issues that the NCC needs to tackle.