top of page

Dementia Risk Prediction Tools

Part 3 of 3 - Dementia Risk Prediction Tools: Utility, Limitations, and Clinical Application for Targeting Multidomain Interventions

Hand holding brain model

Linda Keddington, DNP, APRN

Dec 20, 2025

Given that multidomain interventions appear most effective in individuals at elevated dementia risk, which validated risk prediction tools are available, how well do they perform, and how should clinicians use them in practice?


1. Rationale for Dementia Risk Stratification

Multidomain lifestyle interventions—addressing physical activity, vascular risk factors, diet, cognitive engagement, and social health—have demonstrated modest cognitive benefits in selected populations. However, accumulating evidence suggests that intervention effects are greatest in individuals with elevated baseline dementia risk, rather than in low-risk, general populations.

As discussed in Article #2 of this series, risk stratification is therefore central to targeting prevention strategies efficiently. This article reviews validated dementia risk prediction tools, their comparative performance, and practical considerations for clinical use.


2. Overview of Validated Dementia Risk Prediction Tools


CAIDE (Cardiovascular Risk Factors, Aging, and Incidence of Dementia)

Developed from a Finnish population cohort aged 39–64 years at baseline with ~20-year follow-up, CAIDE incorporates eight predictors: age, education, sex, systolic blood pressure, body mass index, total cholesterol, physical activity, and optional APOE ε4 status. It was designed specifically for midlife risk assessment and is available as a mobile application.¹²


ANU-ADRI (Australian National University Alzheimer’s Disease Risk Index)

ANU-ADRI is a self-report instrument validated in adults aged ≥65 years. It includes 11 risk factors (e.g., diabetes, smoking, depression, traumatic brain injury) and four protective factors (physical activity, cognitive engagement, alcohol intake, fish consumption).³⁴ Its self-report format makes it particularly feasible in primary care and community settings.


BDSI (Brief Dementia Screening Indicator)

The BDSI is a concise 7-item weighted tool developed for older adults in primary care. Variables include age, education, BMI, depressive symptoms, stroke, diabetes, and functional dependence with finances or medications.⁵ It is easy to administer but narrower in scope than other tools.


CogDrisk, CogDrisk-AD, and CogDrisk-ML

CogDrisk tools were developed using systematic review and meta-analysis of 17 established dementia risk factors in adults aged ≥65 years.⁶ A midlife-specific version, CogDrisk-ML, was recently developed for adults aged 40–64 years, extending risk assessment earlier into the life course.⁷


LIBRA (Lifestyle for Brain Health Index)

LIBRA was derived from systematic review and Delphi consensus and focuses on modifiable midlife risk factors.⁶ A modified version incorporating age and sex weights from ANU-ADRI has been validated, though its discriminative ability varies across cohorts.


3. Comparative Predictive Performance


Head-to-Head Validation Studies

Across three U.S. cohorts (MAP, HRS-ADAMS, CHS-CS), CogDrisk, ANU-ADRI, and modified LIBRA demonstrated similar discriminative ability, with C-statistics ranging from 0.65–0.75. CAIDE and the original LIBRA performed less well (C-statistics 0.50–0.59).⁶


In contrast, a large UK Biobank study (n ≈ 466,000) showed that all widely used risk scores—including CAIDE, ANU-ADRI, BDSI, and LIBRA—performed poorly for individualized risk prediction (C-statistics 0.59–0.73). Notably, a model using age alone achieved a C-statistic of 0.79, outperforming all multifactorial tools.³ When calibrated to a 5% false-positive rate, these tools detected only 9–16% of incident dementia cases.³


In the Norwegian HUNT study, CogDrisk demonstrated the highest discriminative ability (C-statistic 0.76), followed by LIBRA (0.75) and ANU-ADRI (0.74), while CAIDE again performed weakest (0.59). However, no tool outperformed a demographics-only model incorporating age and education.⁸


Evidence Synthesis

A 2023 Cochrane systematic review found that CAIDE is the only tool with sufficient external validation to permit meta-analysis, yielding a pooled C-statistic of 0.71 (95% CI 0.66–0.76) for incident dementia. However, confidence in these estimates was rated very low due to heterogeneity, risk of bias, and applicability concerns.¹

4. Age-Specific Performance


Tool performance varies substantially by age and cohort alignment.

  • In the Chicago Health and Aging Project, BDSI demonstrated the highest discriminative ability in older adults (C-statistics 0.79 in Black adults and 0.77 in White adults), followed by ANU-ADRI. CAIDE performed poorly in this older population (C-statistics ≤0.55).⁹

  • For midlife populations, CAIDE and LIBRA demonstrate acceptable validity, whereas for adults <80 years, LIBRA performs reasonably. Evidence remains limited for adults ≥80 years.¹⁰

  • CogDrisk-ML outperformed CAIDE in both UK Biobank and Whitehall II cohorts, with C-statistics of 0.75 and 0.70, respectively.⁷

5. Practical Application in Clinical Settings


Recommended Age-Stratified Approach

  • Ages 40–64: CAIDE or CogDrisk-ML

  • Ages ≥65: ANU-ADRI or BDSI²⁶


Using Risk Stratification to Guide Intervention Intensity

  • Low risk: Low-intensity lifestyle counseling and periodic reassessment

  • High risk: More intensive multidomain interventions and consideration of additional evaluation (e.g., APOE ε4 testing where appropriate, structural MRI, or advanced imaging to clarify vascular vs neurodegenerative contributions)²


Across all tools, the most consistently identified modifiable risk factors include hypertension, diabetes, obesity, smoking, and physical inactivity.¹

6. Critical Limitations


Several limitations constrain the use of current tools for individualized prediction:

  1. Declining accuracy over longer follow-up—discrimination weakens beyond 6–10 years.⁹

  2. Dominant role of age—simple demographic models often perform as well as or better than multifactorial scores.³⁸¹¹

  3. Limited racial and ethnic generalizability, highlighting the need for race-specific models.⁹

  4. Lack of validated prediction for incident dementia—only two multidomain trials have reported dementia outcomes, with no significant risk reduction observed (RR 0.94, 95% CI 0.76–1.18).¹

7. Emerging Directions


CogDrisk represents a promising advance due to its broader incorporation of modifiable risk factors.⁶⁷ Hybrid machine-learning models combining variables from CAIDE, ANU-ADRI, and LIBRA have demonstrated higher discrimination (C-statistic ≈0.80) in exploratory analyses, though independent validation is required.¹²

8. Clinical Takeaway


Current dementia risk prediction tools have meaningful limitations for individualized prognosis, particularly over long time horizons. Nonetheless, they remain useful for identifying higher-risk populations who may derive greater benefit from targeted multidomain interventions—especially when combined with comprehensive assessment of modifiable risk factors and clinical judgment.

References

  1. Mohanannair Geethadevi, G., Quinn, T. J., George, J., et al. (2023). Multi-domain prognostic models used in middle-aged adults without known cognitive impairment for predicting subsequent dementia. Cochrane Database of Systematic Reviews, 6, CD014885. https://doi.org/10.1002/14651858.CD014885.pub2

  2. Ranson, J. M., Rittman, T., Hayat, S., et al. (2021). Modifiable risk factors for dementia and dementia risk profiling: A user manual for brain health services (Part 2 of 6). Alzheimer’s Research & Therapy, 13(1), 169. https://doi.org/10.1186/s13195-021-00895-4

  3. Kivimäki, M., Livingston, G., Singh-Manoux, A., et al. (2023). Estimating dementia risk using multifactorial prediction models. JAMA Network Open, 6(6), e2318132. https://doi.org/10.1001/jamanetworkopen.2023.18132

  4. Anstey, K. J., Cherbuin, N., Herath, P. M., et al. (2014). A self-report risk index to predict occurrence of dementia in three independent cohorts of older adults: The ANU-ADRI. PLoS ONE, 9(1), e86141. https://doi.org/10.1371/journal.pone.0086141

  5. Barnes, D. E., Beiser, A. S., Lee, A., et al. (2014). Development and validation of a brief dementia screening indicator for primary care. Alzheimer’s & Dementia, 10(6), 656–665.e1. https://doi.org/10.1016/j.jalz.2013.11.006

  6. Huque, M. H., Kootar, S., Eramudugolla, R., et al. (2023). CogDrisk, ANU-ADRI, CAIDE, and LIBRA risk scores for estimating dementia risk. JAMA Network Open, 6(8), e2331460. https://doi.org/10.1001/jamanetworkopen.2023.31460

  7. Huque, M. H., Welberry, H. J., Eramudugolla, R., Lautenschlager, N. T., & Anstey, K. J. (2025). Development of a midlife-specific CogDrisk algorithm (CogDrisk-ML). Age and Ageing, 54(7), afaf201. https://doi.org/10.1093/ageing/afaf201

  8. Stubs, J., Langballe, E. M., Livingston, G., et al. (2025). Dementia risk prediction: Comparative analysis of five indices in the HUNT study. Journal of Prevention of Alzheimer’s Disease, 12(9), 100326. https://doi.org/10.1016/j.tjpad.2025.100326

  9. Dhana, K., Barnes, L. L., Beck, T., et al. (2024). External validation of dementia prediction models in Black and White older adults. Alzheimer’s & Dementia, 20(11), 7913–7922. https://doi.org/10.1002/alz.14280

  10. Matovic, D., Lei, X., Picard, G., et al. (2025). Systematic review of dementia risk screening tools in primary care. American Journal of Preventive Medicine, 108124. https://doi.org/10.1016/j.amepre.2025.108124

  11. Fayosse, A., Nguyen, D. P., Dugravot, A., et al. (2020). Risk prediction models for dementia: Role of age and cardiometabolic risk factors. BMC Medicine, 18(1), 107. https://doi.org/10.1186/s12916-020-01578-x

  12. Geethadevi, G. M., Peel, R., Bell, J. S., et al. (2022). Validity of three risk prediction models for dementia or cognitive impairment in Australia. Age and Ageing, 51(12), afac307. https://doi.org/10.1093/ageing/afac307

Disclaimer

Always consult your healthcare provider before making changes to medications, supplements, lifestyle habits, or treatment plans. Do not delay seeking professional care because of something you read here.

Discussions about mental and physical health are educational and not individualized medical guidance. Tools and worksheets are for general well-being and are not designed to diagnose or treat health conditions. 

The information on this site — including articles, videos, links, and downloadable materials — is for general educational purposes only and is not a substitute for medical or psychiatric advice, diagnosis, or treatment. Use of this site does not create a clinician–patient relationship.

Content in the Clinician Hub is for professional development only and is not intended as clinical directives or patient-specific recommendations.

Your use of this site is voluntary and at your own risk. Mind in Balance, New Path Psychiatric, LLC, and Linda Keddington disclaim liability for any outcomes related to actions taken based on this content.

©2017 BY NEW PATH PSYCHIATRIC, LLC. PROUDLY CREATED WITH WIX.COM

bottom of page