Spatiotemporal analysis of regional TIA trends

Andrew Kawai; Samuel Hui; Richard Beare; Velandai K Srikanth; Vijaya Sundararajan; Henry Ma; Thanh G Phan

doi:10.3389/fneur.2022.983512

Spatiotemporal analysis of regional TIA trends

Front Neurol. 2022 Aug 22:13:983512. doi: 10.3389/fneur.2022.983512. eCollection 2022.

Authors

Andrew Kawai^{1

2}, Samuel Hui³, Richard Beare^{4

5

6}, Velandai K Srikanth^{4

5}, Vijaya Sundararajan^{7

8}, Henry Ma^{1

2}, Thanh G Phan^{1

2}

Affiliations

¹ School of Clinical Sciences at Monash Health, Stroke and Aging Research Group, Clinical Trials, Imaging & Informatics Division, Melbourne, VIC, Australia.
² Department of Neurology, Monash Health, Melbourne, VIC, Australia.
³ Department of Medicine, Monash Health, Melbourne, VIC, Australia.
⁴ Academic Unit, Peninsula Clinical School, Central Clinical School, Monash University, Melbourne, VIC, Australia.
⁵ Department of Geriatric Medicine, Peninsula Health, Melbourne, VIC, Australia.
⁶ Developmental Imaging, Murdoch Children's Research Institute, Melbourne, VIC, Australia.
⁷ Department of Public Health, La Trobe University, Melbourne, VIC, Australia.
⁸ Department of Medicine, St Vincent's Hospital Melbourne, Melbourne, VIC, Australia.

Abstract

Background: There has been a decline in the stroke incidence across high income countries but such knowledge exists at Country or State rather than areal unit level such local government area (LGA). In this disease mapping study, we evaluate if there are local hot spots or temporal trends in TIA rate. Such knowledge will be of help in planning healthcare service delivery across regions.

Methods: Linked hospital discharge data (Victorian Admitted Episodes Dataset or VAED) was used to collect TIA (defined by ICD-10-AM codes G450-G459) cases from 2001 to 2011. The State of Victoria is the second most populous state in Australia, with a population of 6.7 million and can be divided into 79 administrative units or LGA. The data is anonymized and contains residence of the patient in terms of LGA but not exact location. The date of the TIA event when the patient is admitted to hospital is provided in the dataset. The number of TIAs per year was aggregated for each LGA. Standardized TIA ratios were calculated by dividing actual over expected cases for each LGA per year. We used Integrated Nested Laplace Approximation (INLA) to perform spatial and spatiotemporal regression, adjusting for hypertension, sex and population, age (≥60), and socio-economic status (SES) decile within the LGA. The final model was chosen based on the lowest the Deviance Information Criterion (DIC) and Watanabe-Akaike information criteria (WAIC).

Results: Choropleth maps showed a higher standardized TIA ratios in North-West rural region. Compared to the baseline model (DIC 13,159, WAIC 13,261), adding in a spatial random effect significantly improved the model (DIC 6,463, WAIC 6,667). However, adding a temporal component did not lead to a significant improvement (DIC 6,483, WAIC 6,707).

Conclusion: Our finding suggests a statically significant spatial component to TIA rate over regional areas but no temporal changes or yearly trends. We propose that such exploratory method should be followed by evaluation of reasons for regional variations and which in turn can identify opportunities in primary prevention of stroke, and stroke care.

Keywords: TIA; areal unit; disease mapping; spatial regression; spatiotemporal regression.