Autopsy Bias Adjustment for Factors Associated with Cortical Beta-Amyloid Deposition
Desta B Fekedulegn, Cecil M Burchfiel, James S Nelson. Biostatistics and Epidemiology Branch, Health Effects Laboratory Division, Morgantown, WV; Pacific Health Research Institute, Honolulu, HI
Background: In studies involving the nervous system (e.g. Alzheimer's disease), the outcome of interest is often assessed from deceased participants who undergo autopsy (the autopsy sample). The autopsy sample is not a true random sample from the target population and is often affected by potential selection mechanisms. Therefore, estimates of association between a risk factor and outcome within the autopsy sample may not accurately represent the situation in the target population, resulting in autopsy bias. Objectives of this study are to describe a method for adjustment of autopsy bias and to demonstrate application of the method using an autopsy based study.
Design: Histological sections from four regions of the brain for 231 autopsied participants of the Honolulu-Asia Aging Study (HAAS) were examined to determine extent of beta-amyloid deposition. Associations of selected risk factors with beta amyloid deposition were examined with and without adjustment for autopsy bias. Autopsy bias adjustment was accomplished first by developing probability of selection into the autopsy sample using demographic and life style variables and then estimating autopsy bias adjusted measures of association using the inverse probability of selection as weights.
Results: Adjustment for autopsy bias affected both the statistical significance and magnitude of the odds ratio for some risk factors. For example, age, years of education and waist circumference were all significantly associated with beta amyloid deposition following yet not before adjustment for autopsy bias. Other risk factors although significantly associated with beta amyloid deposition, showed substantial change in the magnitude of the odds ratio following adjustment. For example, those with coronary heart disease (CHD) were nearly 4 times more likely to have extensive beta amyloid deposition compared to those who had no CHD (OR=4.096, p=0.0085), but after adjustment for autopsy bias the comparable odds ratio was reduced (OR=2.781, p=0.0232).
Conclusions: Adjustment for autopsy bias may affect both the magnitude and statistical significance of the measure of association. The key in adjusting for autopsy bias is developing the selection model using a sufficient set of factors that influence the selection mechanism. In the absence of adjustment for autopsy bias, interpretation of results from autopsy studies pertain only to the population of autopsied individuals.
Wednesday, March 6, 2013 1:00 PM
Poster Session VI # 253, Wednesday Afternoon