Posted 1 October 2008
Interviewed by Tom Fagan
What lifestyle factors are associated with a person's risk of getting Alzheimer disease? For the public, the answer seems to depend on the latest study. One week research extols the virtues of fresh fruits and vegetables, while the next it warns about metabolic disorder. Even Alzheimer's experts find it challenging to keep track. What does epidemiology really show over dozens of studies comprising tens of thousands of people? To help answer this question, Alzforum teamed up with Deborah Blacker to develop AlzRisk, a free, Web-accessible database. Blacker holds positions at Massachusetts General Hospital, Charlestown, and the Harvard School of Public Health, Boston.
ARF: What is AlzRisk?
DB: AlzRisk is a Web-accessible database that is hosted by the Alzheimer Research Forum. It was meant to catalogue all of the major putative environmental risk and protective factors for Alzheimer disease. These include, for example, age and lower education, and most of the risk factors for heart disease such as obesity and diabetes, which are also risk factors for Alzheimer's.
ARF: What prompted you to take on this project?
DB: This began with June Kinoshita, Alzforum's executive editor, and it was based on the success of AlzGene, which is a similar database about genetics findings that was begun by Lars Bertram in our genetics group. June approached me to see if I might help develop something analogous for environmental risk factors. We thought that since the AlzGene database was becoming really established in the field, that setting up AlzRisk would be a great idea. But what was clear at the beginning is that dealing with environmental factors would be vastly more complicated.
ARF: How so?
DB: Not so much the data gathering but the cataloguing. For genetics, each genetic risk factor is one specific item that is already catalogued by genetics websites maintained by the National Library of Medicine and others. A whole set of data is available for each gene—you don't have to define the gene, because a definition already exists. So for genetics, things are quite straightforward. There is a lot more complexity to environmental risk factors, which are defined in a wide variety of ways in different studies. For example, with medications there could be different types, different doses, different periods of exposure. The other aspect is that some environmental risk factors will be things that people can do something about. At one level that's great, because you want there to be a potential benefit from what we learn, but it also means we need to be careful. Where our data suggest one thing and broader considerations suggest something else, then we need to warn people so that they don't run and change their behavior on the basis of a message that could be fundamentally misleading. So we are dealing with more complexity, and at a higher level we are also dealing with the social consequences.
ARF: Are those social consequences contraindications—advice that might be good for AD but bad in other ways?
DB: It could be that, but it is broader than that. In some cases, epidemiologic data point in one direction and clinical trial data point in another, so there are a lot of things that we have to take into consideration. So we ended up having to write a summary statement of what the findings are for each risk factor, which is something that adds additional complexity. The findings do not just speak for themselves necessarily the way they would if you were talking genetics. We have to put them in context by discussing measurement difference, for example. And these statements are going to be different for each risk factor.
ARF: How do you catalogue the data, then? What's the process?
DB: Graduate student research assistants have been extracting data from the scientific literature. They use a Web-accessible administrative database to enter the data, and, when we decide the data are ready, they are "published," or sent to the front end of the Web page. There is a lot of work involved. We're assuming it will get faster as we do more of it. We're still at the beginning, at the steep part of the learning curve.
ARF: Are you soliciting help from other epidemiologists, asking them to submit data, for example?
DB: Not yet. We had planned to work with specific cohorts first, and there's a lot you can get online without having to bother people. I'm sure we will be in touch with researchers, but we are not quite ready to do that in an official way. We still have plenty of things to work on without asking for input, but we will, I'm sure, develop a mechanism to do that.
ARF: Was there, or is there, anything like this in use or in the pipeline elsewhere?
DB: There are papers meta-analyzing individual risk factors, and there are larger compilations of multiple risk factors, but there isn't an accessible review across a full range of risk factors. The strengths of the site, we hope, will be its comprehensiveness, the availability of summary tables and summary odds ratios, and the ongoing curation that will keep it up-to-date.
ARF: How many people are working on the database?
DB: About six, none of whom is remotely full time. But we now realize we need to get grad students and others involved in data extraction.
ARF: How will people use this database, and what will be the major benefit?
DB: The database will be used mainly by researchers, but we are mindful that the general public may read it. Because it's primarily for researchers, we didn't try to keep it at a high school reading level, but we did try to make sure that it wasn't misleading, and that the findings were put in a broader context. I think the principal benefit will be that it will tell people where the field is so they can figure out what to do next—what dead horses have been thoroughly beaten versus what is preliminary but promising versus what looks pretty well established.
ARF: When people go to the database, what kind of information are they going to find that will be of benefit to them?
DB: They will find data from all of the studies done in what we call cohort studies, all the longitudinal studies done in defined populations, where there are data about exposures before the development of Alzheimer's. They will find data about the study population; in fact, there's a page for each cohort study that tells how that population was ascertained, what the gender and age makeup is, how long the subjects have been followed, what kind of follow-up, etc. Typically, each study reports on multiple risk factors, so having this level of detail will be informative for a wide variety of users. Then for each risk factor, a table (or tables) will list each specific study from each of the various cohorts, and there'll be data about the length of follow-up, the gender and age distribution, the number of cases of Alzheimer disease and total dementia that were observed, and the impact of that risk factor on risk for the disease. And then if there are at least three studies with sufficient similarity, there will be a meta-analysis on a separate page that will give you a graphical view of the outcome of each study and a summary odds ratio across all the studies. For these meta-analyses, we are grateful to Matt McQueen, who developed these methods for AlzGene. He was formerly at the Harvard School of Public Health, and is now an assistant professor at the University of Colorado at Boulder.
ARF: So the alternative to using AlzRisk is to manually go through a bundle of publications pulling out all the data.
DB: Right. And the other thing is that they will be able to see at a glance what all the studies were and in what cohorts a given risk factor has been examined. They will also be able to see what methods were used to analyze the data. The level of detail that we were able to get into a surprisingly uncluttered table is amazing, and for that the credit goes to the Web team at the Alzheimer Research Forum and to Jennifer Weuve, my main scientific collaborator on this project. Jennifer was formerly at the Harvard School of Public Health and has recently taken a faculty position at Rush University Medical School, but is able to continue to work with us thanks to phone and Internet connections.
ARF: So how far along is AlzRisk?
DB: So far, we've done vitamin E and diabetes to start, but we have another 20-30 risk factors to go. Our hope for the current year is to get the rest of them in there, but that is probably grandiose.
ARF: Are there any priorities?
DB: Not really. We are working on hypertension next, and then maybe physical activity. We've decided that graduate students are the optimal group to do our data entry because they are well trained, interested, willing to work at modest rates, but that means we need to configure the work so that it is interesting to them and they can get some educational benefit. Thus, we are trying to let each graduate assistant select a topic of interest that may help lay a foundation for the thesis, for example, or complement the thesis work.
ARF: Have any new insights or research angles come out of the analysis?
DB: I think what you get out of the analysis is a systematic look that weights them by informativeness (largely related to sample size), so you get a fair assessment of all the data at once. For the two we did so far, we did not get earth-shattering results, but we have put a more precise estimate on the impact of each of these factors. Diabetes clearly leads to a modest increase in risk for Alzheimer's, and vitamin E supplements do not offer the hoped-for benefits.
ARF: Have you gotten feedback from other epidemiologists? What's the response been like?
DB: We worked with a scientific advisory board that helped keep us focused. If we had gone overboard we could have ended up with a table that was several miles wide. So we needed advice on how much detail was useful. So our board, comprising six or seven advisors, was very helpful along the way. Overall, their responses to the enterprise and the work we've done so far have been positive. Since the database was launched on September 10, we've gotten e-mails from colleagues who want to be involved: that is encouraging in these early days.
ARF: What is the process for including more data, and what's the eventual scope of AlzRisk? Might you expand it in any way?
DB: After we have finished our big task of entering the data from existing cohort studies, the first goal is to curate the site and keep it up-to-date. Potentially, there are ways to expand, and it remains to be seen if we'll do that. One option that we've largely rejected is to extend to non-cohort studies, so-called case-control studies, but there are many more of them and the quality is less consistent, and many were conducted when diagnostic methods were less up-to-date. More likely we would expand to outcomes other than Alzheimer disease. Cognitive decline, for example, is something that everyone is interested in. But our main priority will be getting to the point where we can enter "maintenance mode" and curate the database to keep it up-to-date.
ARF: Are there technical difficulties in getting something like this off the ground and keeping it going?
DB: The daunting thing technically was to figure out how to display the information. That was a very nice collaboration between Jennifer and the folks at Alzforum. It was not trivial to balance aesthetics and readability with displaying data of this complexity. If I can paraphrase Einstein, we tried to "keep everything as simple as possible, but no simpler." The other major challenge was the epidemiology terminology. I had hoped that our typical post-college research assistants would be able to extract the data from scientific papers, but it proved too difficult, so we had to recruit graduate students. They, on the other hand, have less time, so the challenge has been to find a group of students who can each take on one or several risk factors.
ARF: Thank you for this conversation.
DB: Thank you.