本文選題：基準(zhǔn)劑量 + BMDS軟件�。� 參考：《山西醫(yī)科大學(xué)》2016年碩士論文

【摘要】：目的:本文將介紹兩種估計(jì)基準(zhǔn)劑量的非參數(shù)貝葉斯方法:基于加權(quán)過程的非參數(shù)貝葉斯方法和基于隨機(jī)過程的非參數(shù)貝葉斯方法,將這兩種方法相互比較后再與參數(shù)模型作比較,研究其在不同劑量反應(yīng)數(shù)據(jù)情形下的表現(xiàn)。方法:分別介紹常用的參數(shù)模型、基于加權(quán)過程的非參數(shù)貝葉斯方法和基于隨機(jī)過程的非參數(shù)貝葉斯方法的基本原理。通過不同的參數(shù)設(shè)置構(gòu)建8種情形的劑量反應(yīng)數(shù)據(jù),采用R軟件生成模擬數(shù)據(jù)并用介紹的兩種非參數(shù)方法做模擬分析,以BMD估計(jì)值與真實(shí)值的差距以及BMDL的覆蓋率為評(píng)價(jià)指標(biāo),比較兩種方法的表現(xiàn)。從文獻(xiàn)中選取9組癌癥數(shù)據(jù),分別用兩種非參數(shù)方法和BMDS軟件中的9種參數(shù)模型進(jìn)行估計(jì),比較各方法估計(jì)結(jié)果的差異。結(jié)果:兩種非參數(shù)方法NPB1和NPB2的估計(jì)結(jié)果都與真實(shí)BMD值較為接近。當(dāng)多階段模型參數(shù)設(shè)置為MS(0,1,1,3),劑量組數(shù)為6,劑量間距為對(duì)數(shù)間距時(shí),NPB1方法的BMDL覆蓋率明顯低于正常水平,BMD估計(jì)值有較大偏倚,這種較大的偏倚可能是由相對(duì)高的后驗(yàn)寬度參數(shù)值帶來(lái)的過度擬合而造成;兩種方法的RMSE值均較小;NPB2方法的BMDL覆蓋率更接近真實(shí)水平,保守的來(lái)講,NPB2方法較NPB1方法更為可取。實(shí)例中與不同的參數(shù)模型結(jié)果相比可知,兩種非參數(shù)方法得到的BMD估計(jì)值都落入或非常接近由參數(shù)模型得到的BMD估計(jì)值的范圍;而在可信區(qū)間方面,非參數(shù)方法估計(jì)的BMDLs值大多比參數(shù)模型算得的BMDLs值要小。結(jié)論:兩種非參數(shù)貝葉斯方法在基準(zhǔn)劑量估計(jì)過程中都可提供合理的擬合值,尤其是在傳統(tǒng)的參數(shù)模型無(wú)法提供合理的擬合值的情況下,也可很好的應(yīng)用;而NPB2方法在估計(jì)結(jié)果和軟件運(yùn)算速度方面均略優(yōu)于NPB1。非參數(shù)方法在模型擬合過程中很靈活,應(yīng)用范圍廣泛,在今后基準(zhǔn)劑量估計(jì)研究中,是一種非常有用的方法。
[Abstract]:Objective: this paper will introduce two kinds of nonparametric Bayesian methods for estimating baseline dose: nonparametric Bayesian method based on weighted process and nonparametric Bayesian method based on stochastic process. The two methods were compared with each other and then compared with the parametric model to study their performance under different dose-response data. Methods: the basic principles of parametric model, nonparametric Bayesian method based on weighted process and nonparametric Bayesian method based on stochastic process were introduced respectively. The dose-response data of 8 cases were constructed by different parameter settings. The simulated data were generated by R software and simulated by two non-parametric methods introduced. The difference between the BMD estimation value and the real value and the coverage of BMDL were taken as the evaluation index. Compare the performance of the two methods. Nine groups of cancer data were selected from the literature and estimated by two non-parametric methods and nine parameter models in BMDS software. Results: the estimation results of two nonparametric methods, NPB1 and NPB2, are close to the real BMD values. When the multistage model parameters were set to MS0 / 1 / 1 / 1 / 3 / 3, the dose group number was 6, and the dose spacing was logarithmic interval, the BMDL coverage of the method was significantly lower than that of the normal level BMD estimation. The results showed that the BMDL coverage of the NPB1 method was significantly lower than that of the normal level (P < 0.05). The larger bias may be caused by the over-fitting caused by relatively high posterior width parameters, and the RMSE coverage of both methods is closer to the true level than that of the RMSE value of the two methods, and the conservative method is more preferable than the NPB1 method. Compared with the results of different parametric models, the BMD estimates obtained by the two nonparametric methods fall into or are very close to the range of the BMD estimators obtained by the parametric models, but in the confidence interval, Most of the BMDLs values estimated by the nonparametric method are smaller than the BMDLs values calculated by the parametric model. Conclusion: the two non-parametric Bayesian methods can provide reasonable fitting values in the process of baseline dose estimation, especially when the traditional parametric model can not provide reasonable fitting values, and can also be applied very well. The NPB2 method is slightly better than NPB1 in estimating results and computing speed of software. Non-parametric method is very flexible and widely used in the process of model fitting. It is a very useful method in the future study of baseline dose estimation.
【學(xué)位授予單位】：山西醫(yī)科大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2016
【分類號(hào)】：R181.2

【相似文獻(xiàn)】

相關(guān)期刊論文 前10條

1 田琳,路小婷,金泰^，

本文編號(hào)：1817779