Models for dementia risk prediction: so much activity brings a need for coordination and clarity

Dementia Risk Models in an Australian First Nations Population: Cross-Sectional Associations and Preparation for Follow-Up

Background

Reducing the burden of dementia in First Nations populations may be addressed through developing population specific methods to quantify future risk of dementia.

Objective

To adapt existing dementia risk models to cross-sectional dementia prevalence data from a First Nations population in the Torres Strait region of Australia in preparation for follow-up of participants. To explore the diagnostic utility of these dementia risk models at detecting dementia.

Methods

A literature review to identify existing externally validated dementia risk models. Adapting these models to cross-sectional data and assessing their diagnostic utility through area under the receiver operating characteristic curve (AUROC) analyses and calibration using Hosmer-Lemeshow Chi².

Results

Seven risk models could be adapted to the study data. The Aging, Cognition and Dementia (AgeCoDe) study, the Framingham Heart Study (FHS), and the Brief Dementia Screening Indicator (BDSI) had moderate diagnostic utility in identifying dementia (i.e., AUROC >0.70) before and after points for older age were removed.

Conclusion

Seven existing dementia risk models could be adapted to this First Nations population, and three had some cross-sectional diagnostic utility. These models were designed to predict dementia incidence, so their applicability to identify prevalent cases would be limited. The risk scores derived in this study may have prognostic utility as participants are followed up over time. In the interim, this study highlights considerations when transporting and developing dementia risk models for First Nations populations.

Development of a dementia prediction model for primary care: The Hisayama Study

INTRODUCTION

We aimed to develop a risk prediction model for incident dementia using predictors that are available in primary-care settings.

METHODS

A total of 795 subjects aged 65 years or over were prospectively followed-up from 1988 to 2012. A Cox proportional-hazards regression was used to develop a multivariable prediction model. The developed model was translated into a simplified scoring system based on the beta-coefficient. The discrimination of the model was assessed by Harrell's C statistic, and the calibration was assessed by a calibration plot.

RESULTS

During the follow-up period, 364 subjects developed dementia. In the multivariable model, age, female sex, low education, leanness, hypertension, diabetes, history of stroke, current smoking, and sedentariness were selected as predictors. The developed model and simplified score showed good discrimination and calibration.

DISCUSSION

The developed risk prediction model is feasible and practically useful in primary-care settings to identify individuals at high risk for future dementia.

Plasma and cerebrospinal fluid ABeta42 for the differential diagnosis of Alzheimer's disease dementia in participants diagnosed with any dementia subtype in a specialist care setting

BACKGROUND

Dementia is a syndrome that comprises many differing pathologies, including Alzheimer's disease dementia (ADD), vascular dementia (VaD) and frontotemporal dementia (FTD). People may benefit from knowing the type of dementia they live with, as this could inform prognosis and may allow for tailored treatment. Beta-amyloid (1-42) (ABeta42) is a protein which decreases in both the plasma and cerebrospinal fluid (CSF) of people living with ADD, when compared to people with no dementia. However, it is not clear if changes in ABeta42 are specific to ADD or if they are also seen in other types of dementia. It is possible that ABeta42 could help differentiate ADD from other dementia subtypes.

OBJECTIVES

To determine the accuracy of plasma and CSF ABeta42 for distinguishing ADD from other dementia subtypes in people who meet the criteria for a dementia syndrome.

SEARCH METHODS

We searched MEDLINE, and nine other databases up to 18 February 2020. We checked reference lists of any relevant systematic reviews to identify additional studies.

SELECTION CRITERIA

We considered cross-sectional studies that differentiated people with ADD from other dementia subtypes. Eligible studies required measurement of participant plasma or CSF ABeta42 levels and clinical assessment for dementia subtype.

DATA COLLECTION AND ANALYSIS

Seven review authors working independently screened the titles and abstracts generated by the searches. We collected data on study characteristics and test accuracy. We used the second version of the 'Quality Assessment of Diagnostic Accuracy Studies' (QUADAS-2) tool to assess internal and external validity of results. We extracted data into 2 x 2 tables, cross-tabulating index test results (ABeta42) with the reference standard (diagnostic criteria for each dementia subtype). We performed meta-analyses using bivariate, random-effects models. We calculated pooled estimates of sensitivity, specificity, positive predictive values, positive and negative likelihood ratios, and corresponding 95% confidence intervals (CIs). In the primary analysis, we assessed accuracy of plasma or CSF ABeta42 for distinguishing ADD from other mixed dementia types (non-ADD). We then assessed accuracy of ABeta42 for differentiating ADD from specific dementia types: VaD, FTD, dementia with Lewy bodies (DLB), alcohol-related cognitive disorder (ARCD), Creutzfeldt-Jakob disease (CJD) and normal pressure hydrocephalus (NPH). To determine test-positive cases, we used the ABeta42 thresholds employed in the respective primary studies. We then performed sensitivity analyses restricted to those studies that used common thresholds for ABeta42.

MAIN RESULTS

We identified 39 studies (5000 participants) that used CSF ABeta42 levels to differentiate ADD from other subtypes of dementia. No studies of plasma ABeta42 met the inclusion criteria. No studies were rated as low risk of bias across all QUADAS-2 domains. High risk of bias was found predominantly in the domains of patient selection (28 studies) and index test (25 studies). The pooled estimates for differentiating ADD from other dementia subtypes were as follows: ADD from non-ADD: sensitivity 79% (95% CI 0.73 to 0.85), specificity 60% (95% CI 0.52 to 0.67), 13 studies, 1704 participants, 880 participants with ADD; ADD from VaD: sensitivity 79% (95% CI 0.75 to 0.83), specificity 69% (95% CI 0.55 to 0.81), 11 studies, 1151 participants, 941 participants with ADD; ADD from FTD: sensitivity 85% (95% CI 0.79 to 0.89), specificity 72% (95% CI 0.55 to 0.84), 17 studies, 1948 participants, 1371 participants with ADD; ADD from DLB: sensitivity 76% (95% CI 0.69 to 0.82), specificity 67% (95% CI 0.52 to 0.79), nine studies, 1929 participants, 1521 participants with ADD. Across all dementia subtypes, sensitivity was greater than specificity, and the balance of sensitivity and specificity was dependent on the threshold used to define test positivity.

AUTHORS' CONCLUSIONS

Our review indicates that measuring ABeta42 levels in CSF may help differentiate ADD from other dementia subtypes, but the test is imperfect and tends to misdiagnose those with non-ADD as having ADD. We would caution against the use of CSF ABeta42 alone for dementia classification. However, ABeta42 may have value as an adjunct to a full clinical assessment, to aid dementia diagnosis.