Identifying dementia cases with routinely collected health data: A systematic review

Area of Science:

• Epidemiology
• Health Informatics
• Gerontology

Background:

• Prospective, population-based studies are crucial for dementia research.
• Follow-up often relies on linkage to routinely collected, coded health-care data.
• The accuracy of these data sets for dementia case identification requires evaluation.

Purpose of the Study:

• To evaluate the accuracy of routinely collected health-care data for dementia case identification.
• To systematically review literature comparing dementia coding in routine data to expert standards.
• To assess positive predictive value (PPV) and sensitivity of dementia coding.

Main Methods:

• Systematic literature review.
• Inclusion of studies comparing routine data dementia coding to expert-led reference standards.
• Recording of study characteristics, PPV, and sensitivity.

Main Results:

• 27 eligible studies were identified.
• PPVs for all-cause dementia ranged from 33%-100% (16/27 >75%). Sensitivities ranged from 21%-86%.
• PPVs for Alzheimer's disease were generally higher than for vascular dementia.

Conclusions:

• Linkage to routine health-care data can yield high PPV and reasonable sensitivity in specific contexts.
• Significant heterogeneity in accuracy estimates across studies was observed.
• Cohorts should perform setting-specific validation of routine data for dementia research.