# Pitfalls in estimating secular changes in incidence and prevalence of dementia from administrative datasets

## TL;DR

The study shows that changes in dementia rates estimated from administrative data may be due to changes in data sources and diagnosis practices, not actual changes in dementia occurrence.

## Contribution

The study identifies how secular trends and data source availability bias dementia incidence and prevalence estimates from administrative datasets.

## Key findings

- Prevalence and incidence estimates increased over time, with a rapid rise after 2000–2004 due to aged care data inclusion.
- A decrease in estimates from 2015–2019 coincided with the unavailability of aged care assessment data.
- Secular changes in dementia rates may reflect diagnosis trends and data availability rather than true epidemiological shifts.

## Abstract

Administrative datasets can be used to calculate population incidence and prevalence of dementia. It is unclear how changes in data sources may affect these estimates.

We obtained linked administrative health data for individuals 60 years of age or older in Western Australia for 1989–2019 (n = 893,243) including hospital admissions, emergency department, cause‐of‐death, aged care assessment (from April 2003), and mental health services.

There was a marked increase in prevalence and incidence estimates over time. There appeared to be two phases: an initial increase attenuated by 1995–1999 and a rapid increase since 2000–2004 corresponding to inclusion of aged care assessments. There was a decrease in 2015–2019 coinciding with the unavailability of aged care assessment data.

An apparent secular change in rates of dementia over 31 years may be a product of increased propensity to record dementia diagnosis and availability of additional aged care data. Consistent comprehensive data coverage is required.

Administrative datasets may be a low cost method to estimate population prevalence of dementia but it is unknown whether these estimates are stable over time.We demonstrate over a 31‐year period that incidence and prevalence rates are prone to bias from secular trends in diagnosis and changing availability of specific data sources.Using administrative datasets to estimate dementia rates over time will require consistent and comprehensive data coverage.

Administrative datasets may be a low cost method to estimate population prevalence of dementia but it is unknown whether these estimates are stable over time.

We demonstrate over a 31‐year period that incidence and prevalence rates are prone to bias from secular trends in diagnosis and changing availability of specific data sources.

Using administrative datasets to estimate dementia rates over time will require consistent and comprehensive data coverage.