Keywords: Alzheimer's Disease, Bias, Cox regression model, Survival, Truncation
Alzheimer’s Disease (AD) cost the nation $240 billion in 2016, with almost half the costs borne by Medicare. By 2050, it is estimated that these costs will rise to $1.1 trillion. Thus any knowledge of factors that affect disease progression in AD can greatly help reduce the cost of health care in the United States. Accurate regression coefficient estimation is therefore crucial. However due to the inaccuracy of clinical diagnosis, many studies rely on an autopsy-confirmed diagnosis. These studies result in right truncation, since individuals who live past the end of study date are excluded from the sample since they do not receive a pathological diagnosis. Studies which recruit subjects after the onset of AD may also result in left truncation. Thus double truncation, the simultaneous presence of left and right truncation, is inherent in autopsy-confirmed AD studies. Ignoring this truncation scheme in the Cox model results in biased hazard ratio estimators. We therefore propose a weighted estimating equation approach to adjust the Cox model in the presence of double truncation. This approach yields consistent and asymptotically normal hazard ratio estimators.