Dr. Jing Qian, Assistant Professor of Biostatistics in the School of Public Health and Health Sciences at the University of Massachusetts Amherst, has received a two-year, $448,800 exploratory grant from the National Institutes of Health to investigate, with collaborators at Massachusetts General Hospital and Harvard University, new statistical methods for use in regression analysis to explore risk factors in Alzheimer’s disease (AD). In particular, Dr. Qian and colleagues will address a statistical problem brought up by incomplete family history. Their new approach is intended to better handle certain variables in AD studies.
[Photo: Dr. Jing Qian]
The outcome of interest is beta amyloid deposition, which is associated with cognitive decline and is a neuropathological hallmark of AD, as a measure of severity of dementia in cognitively normal older adults who had a parent with Alzheimer’s. “We want to evaluate the contribution of parental history of dementia on this outcome. To do this, we will use regression analysis to assess the relationship between covariates, that is, multiple explanatory variables including the parental history of dementia, and the one outcome variable,” he says.
A major statistical problem encountered in using regression analysis with such a data set is called “randomly censored covariates,” Dr. Qian adds. In this case, “censored” means incomplete. For example, for some subjects, their parent has not experienced dementia onset at the time of the child’s interview, so the parent’s age at onset is not known exactly, but the onset age is known to be greater than his or her age at the interview time. Dr. Qian and colleagues will explore the proper way to analyze such incomplete data.
In this situation where the precise value of an important explanatory variable is not known but a range can be identified, the researchers propose two “threshold” methods for regression models with covariate subject to random censoring. The first approach is called “deletion threshold regression,” and the second is called “complete threshold regression.” The researchers plan to verify the theoretical properties of each with advanced mathematical tools and evaluate the numerical properties of the new methods through Monte Carlo simulations on computers.
“Analyzing censored covariates without any adjustment yields biased results, which is misleading and not acceptable,” he explains. “There are very limited ways to analyze data with censored covariates.”
He and colleagues plan to test new techniques on data from the Harvard Aging Brain Study and the National Alzheimer’s Coordinating Center cumulative database. They hope the new methods can adjust the censored covariate and give a valid hypothesis test on whether there is a significant association between the covariate and the outcome of interest. Furthermore, the new methods will allow them to quantify the scale of the association, yielding an unbiased estimation of the effect of the covariate on the outcome.
“Eventually, we’re going to apply this new method to analyze a new study at Massachusetts General Hospital. We will translate the numbers to meaningful results, which means we may be able to help patients interpret the risk of having dementia.” The researchers point out that their results will be broadly applicable to similar problems related to other diseases.
The number of patients with the progressive, crippling dementia is expected to triple by 2050 to affect 15 million people, and pressure is mounting to find an effective prevention or treatment, says Dr. Qian. “We hope that in two or three years we will have developed an open source software package for clinical investigators who encounter this type of problem, so they can use it to analyze their data,” he adds. “We’re trying to provide an accurate estimate and a more precise confidence interval, giving a more precise statement about the effect of a parent’s age of dementia onset in order to predict the adult child’s beta amyloid deposition level.”