Get Newsletter
AlzRisk Methods

1. Introduction

Alzheimer disease (AD) is a complex disorder estimated to affect over five million people in the United States alone (Alzheimer's Association 2012). The public health impact of AD is substantial, and will continue to grow with increasing longevity and the aging of the post-World War II baby boom generation, 11 to 16 million of whom are expected to develop AD in their lifetimes (Alzheimer's Association 2012). However, consistent with other disorders of complex etiology, establishing risk factors for AD remains a challenge. While the precise etiologic mechanisms remain unknown, a variety of risk factors including age, family history, and the e4-allele of the APOE gene have been identified. Multiple lines of evidence indicate that additional genetic factors play a role in the development of AD, and our sister project, AlzGene, (available from aims to catalog findings regarding putative AD genes. However, much of that same body of evidence indicates that non-genetic risk factors have a role as well. These non-genetic risk factors are critical to prevention efforts, as many of them are modifiable.

Cohort studies of aging populations have contributed valuable information on the impact of various risk factors on AD risk, and more recently, additional data have become available as a substantial number of well-established cohorts originally designed to characterize non-AD outcomes (e.g., cardiovascular disease) are collecting data on cognitive status. Collectively, these studies have a wealth of primary information collected many years before participants were at risk for AD. However, while some risk factors have been replicated across multiple studies (e.g., increased risk among those with type 2 diabetes [Cukierman 2005]), quantitative estimates of the impact of these factors have varied widely. And for other risk factors, studies are less consistent. Moreover, the volume of published results is substantial, and many more results are anticipated in the coming years, so tools to organize and summarize the relevant findings will be increasingly valuable. The development of the database is ongoing, with new and updated risk factor content expected to be added over the coming year and beyond.

2. Overview of Database Organization and Methods

The AlzRisk database aims to provide a comprehensive, unbiased, centralized, publicly available and regularly updated collection of epidemiologic reports that evaluate environmental (i.e., non-genetic) risk factors for Alzheimer disease (AD) in well-defined cohorts. Eligible publications are identified through a systematic review of the literature (see Literature Review Methods). For each risk factor, we provide a summary table or tables of all peer-reviewed studies (Risk Factor Overview). For each study, the tables show (or link to) data on the cohort, the distribution of the risk factor (the "exposure"), the analytic methods, and the effect size and statistical significance of the observed association. Where there are enough results for a given risk factor, we group them by the definition of the risk factor (e.g., diagnosis of hypertension vs. continuously measured systolic blood pressure). Within each table, where comparable data on the exposure (e.g., risk ratio for hypertensives vs. non-hypertensives, risk ratio per 10 mm increase in systolic blood pressure) are available or can be derived in at least four samples, we present random-effects meta-analyses (see Meta-Analysis Methods). Our methods are structured to meet the guidelines outlined in the Meta-analysis of Observational Studies in Epidemiology (MOOSE; Stroup 2008) and where possible, recommendations outlined by the Cochrane Collaboration (Higgins & Green 2011), the Center for Reviews and Dissemination (CRD 2009), and published literature (Egger 2001).

To ensure the highest degree of objectivity regarding the posted information, only peer-reviewed studies from well-defined cohorts with appropriate methods are included. In addition, there must be sufficient information for us to discern fundamental details about the study and the results. Abstracts presented at scientific meetings or findings reported in non-peer-reviewed publications are not considered for inclusion. The criteria are designed to identify high quality studies and ensure that they provide adequate data for meta-analyses. We have found that this approach works as well as applying explicit quality ratings (e.g., Wells 2000) with a threshold, and is more efficient.

3. Literature Review Methods

The review process for risk factors published on before January 2012 followed our previous systematic review methods outlined below, and we are now in the process of updating each of these using our updated systematic review methods. All risk factors published after January 2012 are reviewed using this updated systematic approach with additional handsearching and contact with study investigators where appropriate.

Previous Systematic Review Methods

Previously, we identified publications through searching electronic databases, checking reference lists of relevant review and primary literature, and publication lists of major cohort studies. We used standard inclusion criteria to identify high quality studies and ensure that they provide adequate data for meta-analyses. Non-English articles were not considered during this phase.

Updated Systematic Review Methods

Our updated electronic search strategy and enhanced documentation are designed to comply with standards for systematic review and meta-analyses of observational studies (MOOSE; Stroup 2008; CRD 2009). All searches are conducted by AlzRisk team members, typically by a single reviewer. Medical Subject Heading and EMBASE EMTREE trees, along with relevant article abstracts and metadata, are used to build lists of index and free text terms, their synonyms, abbreviations, and potential misspellings. Conceptually, these search terms map to four domains: 1) Alzheimer disease or dementia, 2) risk or onset, 3) epidemiologic study design, and 4) the specific risk factor under review. These search terms are combined to build similar search strategies for MEDLINE and EMBASE. Both databases provide excellent coverage of the world's leading biomedical journals (Robinson, 2005). Reviews for psychological, sociological, or behavioral risk factors may additionally involve other databases such as PsycINFO or CSA Sociological Abstracts. For this updated strategy we no longer exclude non-English publications, but instead set each one aside to be screened for inclusion by an epidemiologist fluent in the article’s language (who would then work with our staff to extract the requisite data for any papers deemed eligible).

The search results from each database are combined and duplicate citations are removed. Titles and abstracts are then reviewed against our inclusion criteria. We review the full text of any citation whose title or abstract suggests an epidemiologic study of risk factors for dementia onset. Reference management software, such as Endnote or RefWorks, is used to manage the citation review process. For each risk factor, we provide the search strategy used along with a flowchart describing the number of retrieved citations, the number of excluded citations, and the general reasons for exclusion. We also provide a list of all articles that narrowly miss meeting our inclusion criteria, and the reasons for our decision to exclude them.

To assess the reliability of the review process, a reviewer not involved in the original search for physical activity articles drew a 25% random sample of unique candidate citations, and then independently screened these citations using the same procedures as the original review. A Kappa statistic was used to evaluate reviewer agreement for inclusion (Κ=1). We found perfect agreement between the original and independent reviewers. A second reliability study is being conducted for the Hormone Therapy systematic review.

Inclusion Criteria

Our systematic review is designed to identify and summarize peer-reviewed epidemiologic literature on associations between non-genetic risk factors and Alzheimer’s disease. An article qualifies for inclusion in our summary tables and meta-analyses when it meets all of the following criteria.

1) Evaluates Alzheimer’s disease as an outcome
2) Analyzes data from a well-defined cohort (including nested case-control studies)
3) Clearly defines the exposure under study
4) Appropriately adjusts for age and sex by regression, stratification, matching or restriction
5) Is published in peer-reviewed journals (not just as an abstract)
6) Includes sufficient detail about the study design, analysis, and results
7) Provides an estimate of association and at least one corresponding measure of statistical uncertainty such as a p-value, standard error, or confidence interval

4. Meta-analysis Methods

For all risk factors for which comparable exposure data are available in four or more independent samples, estimates of association (e.g., hazard ratios [HRs], odds ratios, cumulative incidence ratios) and 95 percent confidence intervals (CIs) are recorded or calculated from the reported data in the report. Summary association estimates and 95 percent CIs are calculated using the DerSimonian and Laird (1986) random effects model, which utilizes weights that incorporate both within-study and between-study variance. This procedure is done including all studies irrespective of publication order (denoted by "All Studies" on the meta-analysis figures) and repeated after exclusion of the initial report ("All Excl Initial Report"). Overlapping samples (of which typically only the one with the largest sample, longest follow-up, or optimal analytic strategy is included) and studies with missing data are indicated on the meta-analysis forest plots. Occasionally, when findings are available from only three independent samples and we anticipate that findings from more samples are forthcoming, we will conduct meta-analyses of the existing three sets of findings. Please note that when only a few studies are included in the meta-analyses (i.e., less than ~10), the random effects model may yield summary HRs and confidence bounds that are slightly anti-conservative.

To allow a visual assessment of the presence of publication bias (or other sorts of reporting bias), we use a Begg modified funnel plot, which depicts the exposure level-specific HR (on a logarithmic scale) against its standard error for each report (Egger 1997). Note that the power to detect deviations from a symmetrical distribution is limited, especially for analyses based on less than ~20 individual studies.

5. References

Alzheimer’s Association. 2012 Alzheimer’s Disease Facts and Figures. Alzheimers Dement;8(2):131-168. Abstract

Centre for Reviews and Dissemination. Systematic Reviews: CRD’s guidance for undertaking reviews in health care. York, UK: 2009. Available from:

Cukierman T, Gerstein HC, Williamson JD. Cognitive decline and dementia in diabetes–systematic overview of prospective observational studies. Diabetologia. 2005;48: 2460–2469. Abstract

DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials. 1986 Sep;7(3):177-188. Abstract

Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-analysis detected by a simple, graphical test. BMJ. 1997 Sep 13;315(7109):629-634. Abstract

Egger M, Davey Smith G, Altman D, editors. Systematic Reviews in Healthcare: Meta-analysis in context. London, UK: BMJ Publishing Group; 2001.

Higgins JPT, Green S, editors. Cochrane Handbook for Systematic Reviews of Interventions Version 5.1.0 [updated March 2011]. 2011. Available from:

Robinson KA. Should meta-analysts search Embase in addition to Medline? J Clin Epidemiol. 2005;58(3):320 Abstract

Stroup DF, Berlin JA, Morton SC, Olkin I, Williamson GD, Rennie D, Moher D, Becker BJ, Sipe TA, Thacker SB. Meta-analysis of observational studies in epidemiology: a proposal for reporting. Meta-analysis Of Observational Studies in Epidemiology (MOOSE) group. JAMA. 2000 Apr;283(15):2008-2012. Abstract

Wells G, Shea B, O’Connell D, Peterson J, Welch V, Losos M, Tugwell P. The Newcastle-Ottawa Scale (NOS) for assessing the quality of nonrandomised studies in meta-analyses [cited 2012 Mar 28]. Available from: