A multi-phase stratified sampling protocol for estimating dementia prevalence across diverse populations in Aotearoa New Zealand — The Association Specialists
  1. Schedule
  2. Oral Presentations - Advancing Epidemiological Methods 1
  3. A multi-phase stratified sampling protocol for estimating dementia prevalence across diverse populations in Aotearoa New Zealand

A multi-phase stratified sampling protocol for estimating dementia prevalence across diverse populations in Aotearoa New Zealand (23934)

Claudia Rivera 1 , Sarah Cullum 2 , Gary Cheung 2 , Rita Krishnamurthi 3 , Susan Yates 4 , Jackie Broadbent 5 , Lynette Tippett 4 , Ngaire Kerse 6 , Xiaojing Wu 6
  1. Department of Statistics, University of Auckland, Auckland
  2. Department of Psychological Medicine, University of Auckland, Auckland
  3. National Institute for Stroke and Applied Neurosciences, Auckland University of Technology, Auckland
  4. School of Psychology, University of Auckland, Auckland
  5. Health NZ Te Whatu Ora-Canterbury Waitaha, Auckland
  6. Department General Practice and Primary Health Care, University of Auckland, Grafton, AUCKLAND, New Zealand

Background

Dementia prevalence in Aotearoa New Zealand (NZ) is poorly characterised, with existing estimates largely derived from international data or routinely collected health datasets that substantially under-ascertain cases. These limitations disproportionately affect ethnically diverse populations, contributing to inequities in diagnosis and service planning. The Impact of Dementia mate wareware and Equity in Aotearoa (IDEA) programme was developed to generate robust, population-based prevalence estimates across major ethnic groups.

Methods

A cross-sectional screen–interview survey was conducted using a multi-phase, stratified sampling design. In the absence of an individual-level sampling frame with ethnicity information, an area-based approach was employed. Sampling proceeded through four stages: purposive selection of Territorial Authorities, stratified probability proportional to size sampling of Statistical Area 2 units (by region and rurality), selection of meshblocks inversely proportional to land area, and door-to-door recruitment of participants aged ≥65 years. Screening was conducted using the Brief Community Screening Instrument for Dementia, followed by diagnostic assessment using the 10/66 protocol. Oversampling of Chinese and Indian populations was incorporated through a simulation-informed design to achieve a margin of error ≤3% across ethnic groups. Non-response was addressed through repeated contact attempts and will be accounted for using post-stratification weighting and calibration.

Results

Simulation studies informed the selection of 12 rural and 45 urban areas, with sampling fractions calibrated to achieve balanced sample sizes across ethnic groups. Ongoing fieldwork demonstrates the feasibility of the multi-phase design, with progressive recruitment across populations despite variable response rates. The approach enabled the inclusion of geographically and ethnically diverse communities not captured in routine data.

Conclusion

This protocol demonstrates a pragmatic and methodologically robust approach to estimating dementia prevalence. By integrating stratified multi-phase sampling with culturally responsive fieldwork, the design supports more accurate and equitable epidemiological estimates and provides a framework applicable to other settings with diverse and underrepresented populations.