Multistate Markov Model for Individually Tailored Breast Cancer Screening

• Main Authors: Yi-Ying Wu, 吳怡瑩
• Other Authors: Hsiu-Hsi Chen
• Format: Others
• Language: en_US
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/76180962967184875869

博士 === 國立臺灣大學 === 流行病學與預防醫學研究所 === 101 === Background Individually tailored screening for breast cancer is now a panacea for reducing the concern expressed by health policy makers that the harm and cost of screening should be minimized and the benefits maximized. The two-throng problem may be solved by intensive screening on high-risk groups to reduce the false-negative cases and reducing the frequency of screen on low-risk groups to reduce the false-positive cases through the risk stratification done by the superimposition of initiators and promoters using the three-state Markov model and also the five-state Markov model, if possible, by incorporating tumor attributes. The application of the multistate Markov model to facilitate the development of individually tailored screening has been hardly addressed. The subjects of this thesis not only embrace the development of the Markov model but also demonstrate how these proposed models can be applied to individually tailored breast cancer screening. Materials and Methods The thesis begins with literature search for initiators and prompters related to the development of breast cancer in the preclinical detectable phase (PCDP) and the clinical phase (CP), respectively. We proposed a risk-score-based approach that translates state-of-the-art scientific evidence, including genomic discovery, biomarkers, and conventional risk factors, into the initiators and promoters underpinning a novel multi-factorial three-state temporal natural history model. The estimated results are used to construct the risk scores to stratify population into different risk groups and to assess the optimal age to begin screening and the inter-screening interval for each category, and to ascertain which high risk group requires an alternative imagine technique. We extended the three-state model to five-state model with the incorporation of node involvement, however the information from literature is insufficient. We propose an estimation method based on the reparameterization method to estimate the state-specific parameters. Furthermore, the empirical screening data from 1977 to 2010 in Dalarna county, Sweden was used to construct the risk assessment model. Results From the literature, we identify the initiators including BRCA gene, 7 Single nucleotides polymorphisms (SNPs), breast density, body mass index (BMI), and age at first full-term pregnancy (AP) and the promoters including BMI, AP, ER, HER-2 and Ki-67 expression. According to the simulated results based on one-million Taiwanese women, there were 2182, 2867 and 1560 prevalent screen-detected (SD) cases, incident SD cases and interval cancers. The 10-year predicted risk for the transition from Free of breast cancer (FBC) to the clinical phase (CP) was 25.83% for high risk group (75th percentile), 20.31% for intermediate risk group (50th percentile), and 13.84% for low risk group (25% percentile), respectively in BRCA-carrier. The corresponding figures were 1.55% for high risk group, 1.22% for intermediate risk group, and 0.76% for low risk group in non-carrier. This risk-score-based approach significantly reduced the interval caner rate as a percentage of expected rate in the absence of screening by 30% compared with triennial screenings, and it also reduced the false-positive cases by 8.2%. For the five-state Markov model, the estimated results of the proposed method were similar to the maximum likelihood estimates (MLE) based on full data analysis and also the true values. The simulated results show the proposed method which only used the aggregate data rather than individual data combined with the external information is valid for estimate the effects of promoters. Based on the Sweden empirical data, we identified the initiators including BMI, AP, breast density, and family history and promoters including BMI, breast density, and molecular phenotypes. Conclusions We proposed risk assessment Markov models for modeling the progression of breast cancer as a function of a constellation of initiators and promoters, and used the estimated results to construct composite risk scores for individually tailored screening. The concept and approach can be readily applied to other screening programs in other populations by tuning risk scores with their own genetic susceptibility factors, tumor phenotypes, clinical attributes, and risk factors.