Page 411 - 2014 Printable Abstract Book
P. 411
associations in models stratified by gender and separately for each disease considering both exposures
types (continuous, categorical scale). Odds ratios (OR) and 95% CI, concordance and ROC area were
calculated for genes significantly associated with the disease and the exposure using logistic regression
analysis. Altogether 12 mRNAs and 9 microRNAs appeared to be significantly associated with 6 diseases,
e.g. thyroid diseases (3 genes, OR: 1.2-5.1, concordance: 71-78%), atherosclerotic diseases (4 genes, OR:
2.5-10, concordance: 70-75%), kidney diseases (6 genes, OR: 1.3-8.6, concordance: 69-85%), cholelithiasis
(3 genes, OR: 0.2-0.3, concordance: 74-75%), benign tumors (1 gene [AGAP4], OR: 3.7, concordance: 81%)
and chronic radiation syndrome (4 genes, OR: 2.5-4.3, concordance: 70-99%). Further associations were
found for systolic blood pressure (6 genes, OR: 3.7-10.6, concordance: 81-88%) and body mass index (1
gene [miR-484], OR: 3.7, concordance: 81%). All associations were gender and exposure type dependent.
Gene expression changes observed after occupational prolonged radiation exposures seem to increase
the risk for certain chronic non-cancer diseases.
(PS7-59) Application of simulation-extrapolation (SIMEX) to the Radiation Effects Research Foundation
Life Span Study data. Munechika Misumi, Radiation Effects Research Foundation, Hiroshima-shi, Japan
Errors in radiation dose estimates are commonly accounted for in radiation epidemiological
studies. Studies at the Radiation Effects Research Foundation (RERF) have only applied a regression
calibration, to deal with so-called classical type error assuming a multiplicative error model, although
implementation of new adjustments taking both classical and Berkson errors into account has been
considered. Since regression calibration needs an assumption on the distribution of unobservable true
radiation dose, the robustness of the adjustment method to possible violations of that assumption should
be considered. However, this has been very difficult in RERF studies because the complicated nature of
the survival data creates problems in complete simulation of datasets. The SIMEX method was originally
designed to obtain approximately unbiased risk estimates in problems with additive measurement error.
The underlying idea is that the effect of measurement error can be corrected for by simulating doses with
various amounts of added error, starting from the available observed doses, which are assumed to contain
a certain amount of error, performing a risk regression for each set of doses with the other data
unchanged, and extrapolating the results back to the case of no measurement error. This enables us to
consider applying a functional approach despite the difficulty in analytical derivation of error structures
due to the cross-classification in the Life Span Study data analysis. Because it does not need an assumption
on the true dose distribution, it might be considered as a sensitivity analysis of the results obtained by
regression calibration. In the session, the results of a SIMEX application will be shown and a comparison
of the results to those previously reported will be discussed, along with results of numerical studies.
(PS7-60) Evaluation of medical radiation exposures among atomic bomb survivors. Atsuko Sadakane;
Ritsu Sakata; Eric J. Grant; Mai Utada; and Kotaro Ozasa, Radiation Effects Research Foundation,
Hiroshima, Japan
Background: When estimating A-bomb dose-response, it is important to consider medical
radiation exposures as they may confound or modify relationship. From 1964 to 1982, exposure to
medical radiation was assessed by interview as part of periodic examinations of the Adult Health Study.
In addition, doses from diagnostic and therapeutic x-ray were estimated based on a series of field surveys
types (continuous, categorical scale). Odds ratios (OR) and 95% CI, concordance and ROC area were
calculated for genes significantly associated with the disease and the exposure using logistic regression
analysis. Altogether 12 mRNAs and 9 microRNAs appeared to be significantly associated with 6 diseases,
e.g. thyroid diseases (3 genes, OR: 1.2-5.1, concordance: 71-78%), atherosclerotic diseases (4 genes, OR:
2.5-10, concordance: 70-75%), kidney diseases (6 genes, OR: 1.3-8.6, concordance: 69-85%), cholelithiasis
(3 genes, OR: 0.2-0.3, concordance: 74-75%), benign tumors (1 gene [AGAP4], OR: 3.7, concordance: 81%)
and chronic radiation syndrome (4 genes, OR: 2.5-4.3, concordance: 70-99%). Further associations were
found for systolic blood pressure (6 genes, OR: 3.7-10.6, concordance: 81-88%) and body mass index (1
gene [miR-484], OR: 3.7, concordance: 81%). All associations were gender and exposure type dependent.
Gene expression changes observed after occupational prolonged radiation exposures seem to increase
the risk for certain chronic non-cancer diseases.
(PS7-59) Application of simulation-extrapolation (SIMEX) to the Radiation Effects Research Foundation
Life Span Study data. Munechika Misumi, Radiation Effects Research Foundation, Hiroshima-shi, Japan
Errors in radiation dose estimates are commonly accounted for in radiation epidemiological
studies. Studies at the Radiation Effects Research Foundation (RERF) have only applied a regression
calibration, to deal with so-called classical type error assuming a multiplicative error model, although
implementation of new adjustments taking both classical and Berkson errors into account has been
considered. Since regression calibration needs an assumption on the distribution of unobservable true
radiation dose, the robustness of the adjustment method to possible violations of that assumption should
be considered. However, this has been very difficult in RERF studies because the complicated nature of
the survival data creates problems in complete simulation of datasets. The SIMEX method was originally
designed to obtain approximately unbiased risk estimates in problems with additive measurement error.
The underlying idea is that the effect of measurement error can be corrected for by simulating doses with
various amounts of added error, starting from the available observed doses, which are assumed to contain
a certain amount of error, performing a risk regression for each set of doses with the other data
unchanged, and extrapolating the results back to the case of no measurement error. This enables us to
consider applying a functional approach despite the difficulty in analytical derivation of error structures
due to the cross-classification in the Life Span Study data analysis. Because it does not need an assumption
on the true dose distribution, it might be considered as a sensitivity analysis of the results obtained by
regression calibration. In the session, the results of a SIMEX application will be shown and a comparison
of the results to those previously reported will be discussed, along with results of numerical studies.
(PS7-60) Evaluation of medical radiation exposures among atomic bomb survivors. Atsuko Sadakane;
Ritsu Sakata; Eric J. Grant; Mai Utada; and Kotaro Ozasa, Radiation Effects Research Foundation,
Hiroshima, Japan
Background: When estimating A-bomb dose-response, it is important to consider medical
radiation exposures as they may confound or modify relationship. From 1964 to 1982, exposure to
medical radiation was assessed by interview as part of periodic examinations of the Adult Health Study.
In addition, doses from diagnostic and therapeutic x-ray were estimated based on a series of field surveys
