Analysis of Cancer Epidemiological Data Using Biologically-Motivated Markov Models

• Main Authors: Wang, Mey, 王玫
• Other Authors: Chen-Hsin Chen
• Format: Others
• Language: zh-TW
• Published: 1998
• Online Access: http://ndltd.ncl.edu.tw/handle/93023284886846263375

博士 === 國立臺灣大學 === 公共衛生學系 === 86 === AbstractObjectives Most cancer epidemiological researches explore the determinants of adichotomous disease outcome and very few of them exploit underlying biologicaltheories of disease pathogenesis.The Armitage and Doll multistage model, which is a biologically-motivated model,h as been used for cancer risk assessment since 1960.Actually,their model is a s ingle path time homogeneous Markov process.The dissertation, in two separate s tudies, investigates more plausible andcomplicated stochastic models for cance r epidemiology.Study A, entitled"Estimation the Intensity Matrix of a Non-homo geneous Markov process via the EMAlgorithm with Application to a Breast Cancer Model", examines into therisk of developing cancers through a non-homogeneous Markov process with several change points, StudyB, entitled "Analysis of a Mu lti-path Multi-State ProgressiveDisease Process with the Consideration of Non- Susceptibility and Phase TypeDistribution", deals with a multi-path multi-stat e disease processconcerning the issues arised from a periodically cancerscreen ing program.Methods(Study A) Motivated by Pike''s "breast tissue age" model (Na ture, 1983), we introduce a time scale transformation to convert anon-homogene ous Markov process across the choronologic age to a homogeneous Markovprocess across the tissue age. By a combination of the EM algorithm and the Newton-Ra phson method, we estimate the transition intensities and the parameters ofchan ging rates.(Study B) We present aparametric Markov model which partitions the disease process into severalconfigurations and in each configuration the trans ition probabilities on all pathsare adequately assigned. The proposed multi-pa thmover-stayer model also allows aproportion of right censored observationsto be non-susceptible to any subsequent transition. Theinitial probabilities of a phase type distributionhave been used to facilitate the problem of left censo ring. Special attentionis given on the interval censored data for the possibl e transition to theintermediate state. The EM algorithm combinedwith the Newt on-Raphson method is used in the estimation procedure.Results(Study A) Simulat ion results reveal our proposed model andthe estimation algorithm are quite sa tisfactory, they works well even undera Markov process including four stages a nd three unknown changepoints.(Study B) Simulation results verify the validity of theestimation procedure under our multi-path mover-stayer model. The propo sedmethod is applied to analyze the data of a follow-up study for carriers oft he hepatitis surface antigen (HBsAg(+)) aged over 20 (Yu et al., 1997).The res ults ofdata analysis reveal about 55.5% of HBsAg carriers and 56% of chroniche patitis victims are not susceptible to develop cirrhosis. For theremaining 44 .5 % HBsAg carriers who are the movers, 64% of them wouldundergo the first pat h, and 36% the second path. Forthose being a HBsAg carrier at age 20, 14.0 % a nd 8.5% will develop chronichepatitis and cirrhosisrespectively at age 40.Conc lusions(Study A) Except for the three defined change points, theproposed model can also be incorporated with other factors such as the birthcohort, fat int ake, body weight .... etc to measure the risk effectson the incidence of breas t cancer. The method can also be applied to analyzemany cancer epidemiological studies in which the age at exposure, the timeat dose change, the time at imm igration/migration,etc., should be thoughtas the change points.(Study B) The m ulti-path mover-stayer model is a feasibleway to model a multi-path multi-stat e disease process with the considerationof non-susceptibility. The model can be extended to incorporate covariates(time-independent or time-dependent) to a ssess the risks of developinghepatocellular carcinoma for those HBsAg carriers .