Disease risk models with low to moderate numbers of events: a case study in cancer risk and low-dose radiation exposure
碩士 === 國立陽明大學 === 公共衛生研究所 === 105 === Background and Study Objectives Excess relative risk model is commonly used to quantify the radiation effects on the risks of cancer in international studies of radiation epidemiology. However, when numbers of cancer cases are small, the coefficients (i.e. ERR)...
Main Authors: | , |
---|---|
Other Authors: | |
Format: | Others |
Language: | zh-TW |
Published: |
2017
|
Online Access: | http://ndltd.ncl.edu.tw/handle/9d9hdu |
id |
ndltd-TW-105YM005058032 |
---|---|
record_format |
oai_dc |
spelling |
ndltd-TW-105YM0050580322019-05-15T23:39:47Z http://ndltd.ncl.edu.tw/handle/9d9hdu Disease risk models with low to moderate numbers of events: a case study in cancer risk and low-dose radiation exposure 低事件數資料疾病風險評估模式探討:以低劑量輻射與癌症風險為例 Yu-Shan Cheng 鄭伃珊 碩士 國立陽明大學 公共衛生研究所 105 Background and Study Objectives Excess relative risk model is commonly used to quantify the radiation effects on the risks of cancer in international studies of radiation epidemiology. However, when numbers of cancer cases are small, the coefficients (i.e. ERR) of this types of models can be problematic. This often occurred in site-specific cancers evaluation. Cox proportional hazard models are widely used in clinical and epidemiologic studies which based on a multiplicative model assumption. The objectives of this study were to conduct a sensitivity analysis, which compares the estimates by Cox model, Poisson ERR model, and Poisson log-linear models, for quantifying the effects of low-dose radiation exposure and cancer risks. Also, absolute risks based on person years are estimated. Materials and methods The Taiwan Radiation Contaminated Building Cohort (the RCB cohort) was linked to Taiwan Cancer Registry Annual Report (1983 -2012) and Cause of Death Data (1983-2013) from Health and Welfare Data Science Center, Ministry of Health and Welfare (HWDC, MOHW). Analyses were conducted using Excess relative risk model (ERR model), Cox model and Poisson log-linear model. In account for small events, the Hazard ratio (HR) estimated by Cox model were estimated by partial likelihood based, Wald-based and Firth’s penalized likelihood. Results and Conclusions The Taiwan RCB cohort comprises 300 cancer cases as of the end of 2012, including 249 cases after taking account of the assumed minimum latent periods for solid cancer as 10 years and leukemia as 2 years. Taiwan cumulative dose (TCD) is the main predictor of interest for cancers. When the TCD was treated as a categorical indicator, the results show that ERR estimated by ERR model is equivalent to the relative risk (RR) minus 1 estimated by Poisson log-linear model based on grouped data (ERR=RR-1). An approximate ERR estimated by HR-1, however, are more conservative than those by RR-1 or real ERR. All in all, this study supports an increased risk of cancer incidence following protracted radiation exposure from RCB cohort and statistically significant risks were observed by breast cancer and leukemia. When numbers of events are small, an ERR estimated by HR-1 based on individual data would be more conservative than the ERR estimated by a Poisson linear relative risk model based on grouped data. I-Feng Lin 林逸芬 2017 學位論文 ; thesis 67 zh-TW |
collection |
NDLTD |
language |
zh-TW |
format |
Others
|
sources |
NDLTD |
description |
碩士 === 國立陽明大學 === 公共衛生研究所 === 105 === Background and Study Objectives
Excess relative risk model is commonly used to quantify the radiation effects on the risks of cancer in international studies of radiation epidemiology. However, when numbers of cancer cases are small, the coefficients (i.e. ERR) of this types of models can be problematic. This often occurred in site-specific cancers evaluation. Cox proportional hazard models are widely used in clinical and epidemiologic studies which based on a multiplicative model assumption.
The objectives of this study were to conduct a sensitivity analysis, which compares the estimates by Cox model, Poisson ERR model, and Poisson log-linear models, for quantifying the effects of low-dose radiation exposure and cancer risks. Also, absolute risks based on person years are estimated.
Materials and methods
The Taiwan Radiation Contaminated Building Cohort (the RCB cohort) was linked to Taiwan Cancer Registry Annual Report (1983 -2012) and Cause of Death Data (1983-2013) from Health and Welfare Data Science Center, Ministry of Health and Welfare (HWDC, MOHW).
Analyses were conducted using Excess relative risk model (ERR model), Cox model and Poisson log-linear model. In account for small events, the Hazard ratio (HR) estimated by Cox model were estimated by partial likelihood based, Wald-based and Firth’s penalized likelihood.
Results and Conclusions
The Taiwan RCB cohort comprises 300 cancer cases as of the end of 2012, including 249 cases after taking account of the assumed minimum latent periods for solid cancer as 10 years and leukemia as 2 years. Taiwan cumulative dose (TCD) is the main predictor of interest for cancers. When the TCD was treated as a categorical indicator, the results show that ERR estimated by ERR model is equivalent to the relative risk (RR) minus 1 estimated by Poisson log-linear model based on grouped data (ERR=RR-1). An approximate ERR estimated by HR-1, however, are more conservative than those by RR-1 or real ERR.
All in all, this study supports an increased risk of cancer incidence following protracted radiation exposure from RCB cohort and statistically significant risks were observed by breast cancer and leukemia. When numbers of events are small, an ERR estimated by HR-1 based on individual data would be more conservative than the ERR estimated by a Poisson linear relative risk model based on grouped data.
|
author2 |
I-Feng Lin |
author_facet |
I-Feng Lin Yu-Shan Cheng 鄭伃珊 |
author |
Yu-Shan Cheng 鄭伃珊 |
spellingShingle |
Yu-Shan Cheng 鄭伃珊 Disease risk models with low to moderate numbers of events: a case study in cancer risk and low-dose radiation exposure |
author_sort |
Yu-Shan Cheng |
title |
Disease risk models with low to moderate numbers of events: a case study in cancer risk and low-dose radiation exposure |
title_short |
Disease risk models with low to moderate numbers of events: a case study in cancer risk and low-dose radiation exposure |
title_full |
Disease risk models with low to moderate numbers of events: a case study in cancer risk and low-dose radiation exposure |
title_fullStr |
Disease risk models with low to moderate numbers of events: a case study in cancer risk and low-dose radiation exposure |
title_full_unstemmed |
Disease risk models with low to moderate numbers of events: a case study in cancer risk and low-dose radiation exposure |
title_sort |
disease risk models with low to moderate numbers of events: a case study in cancer risk and low-dose radiation exposure |
publishDate |
2017 |
url |
http://ndltd.ncl.edu.tw/handle/9d9hdu |
work_keys_str_mv |
AT yushancheng diseaseriskmodelswithlowtomoderatenumbersofeventsacasestudyincancerriskandlowdoseradiationexposure AT zhèngyúshān diseaseriskmodelswithlowtomoderatenumbersofeventsacasestudyincancerriskandlowdoseradiationexposure AT yushancheng dīshìjiànshùzīliàojíbìngfēngxiǎnpínggūmóshìtàntǎoyǐdījìliàngfúshèyǔáizhèngfēngxiǎnwèilì AT zhèngyúshān dīshìjiànshùzīliàojíbìngfēngxiǎnpínggūmóshìtàntǎoyǐdījìliàngfúshèyǔáizhèngfēngxiǎnwèilì |
_version_ |
1719152443147681792 |