Skip to main content

Lund University Publications

LUND UNIVERSITY LIBRARIES

Temporal clustering of ANCA-associated vasculitis occurrence

Coq, Matthieu ; White, Arthur ; Gisslander, Karl LU orcid ; Bordignon Draibe, Julianna ; Nic an Riogh, Eithne and Little, Mark (2024) p.471-471
Abstract
Background: We hypothesised that observed annual incidence rates of ANCA-associated vasculitis (AAV) are driven by
sporadic short-term, high-intensity rates characterised as clusters of diagnosis events in time. Using data from Ireland, Sweden
and Spain, we performed a changepoint analysis to discover time intervals with high and low rates of vasculitis diagnosis.
Methods: We recruited 417 patients from the RITA-Ireland Vasculitis Registry. Adults (>16 years) diagnosed with AAV after
1/1/13 were eligible. Inclusion criteria were a definite diagnosis of microscopic polyangiitis or granulomatosis with polyangiitis.
Patients with positive anti-GBM antibodies were excluded. The rate of diagnosis was modelled using a... (More)
Background: We hypothesised that observed annual incidence rates of ANCA-associated vasculitis (AAV) are driven by
sporadic short-term, high-intensity rates characterised as clusters of diagnosis events in time. Using data from Ireland, Sweden
and Spain, we performed a changepoint analysis to discover time intervals with high and low rates of vasculitis diagnosis.
Methods: We recruited 417 patients from the RITA-Ireland Vasculitis Registry. Adults (>16 years) diagnosed with AAV after
1/1/13 were eligible. Inclusion criteria were a definite diagnosis of microscopic polyangiitis or granulomatosis with polyangiitis.
Patients with positive anti-GBM antibodies were excluded. The rate of diagnosis was modelled using a Poisson process changepoint
model, which divides the timeline into intervals. Within each interval the time between diagnosis follows an exponential distribution
with rate µ. Intervals with larger values of µ will have higher diagnosis rates. The number of changepoints was selected using
cross-validation. We also applied this method to independent validation cohorts in Skåne (n = 351, Jan 1997 – Dec 2019, thus not
including the pandemic period) and Barcelona (n = 82, Jan 2013 - present).
Results: A five changepoint model fit the Irish data best (Figure). We observed several diagnosis incidence rates, ranging from
µ=22/year over 42 months (pandemic period) to µ=60/y over 41 months (2015-2018). We also observed a recent post-pandemic
reset to a higher diagnosis rate. No clusters of diagnosis events in time were clearly identified by the model. No changepoints were
identified for the Skåne data, with a consistent diagnosis rate of µ=16/y across the cohort time frame. For the Barcelona data, we
observed one changepoint in 2021, with associated rates of µ=5/y and µ=10/y. Observed intervals between diagnoses closely
matched simulations generated by the estimated models for the Irish data (Figure). However, very short-term intervals (<5 days)
were underestimated by the model, possibly suggesting that these short intervals occur more frequently than expected by chance.
We observed similar short interval results in the Skåne and Barcelona registries.
Conclusions: We observed little clear evidence for short-term, high-intensity diagnosis incidence rates across 3 independent
cohorts. For both Barcelona and Ireland (inception cohorts without 100% ascertainment), changes in incidence rate can likely
be explained by changes in registry recruitment strategy, particularly over the pandemic. Our findings suggest that much of the
perceived volatile behaviour of diagnosis events are consistent with the expected rates using well-established Poisson process
models. However, our models failed to fully account for very short-term diagnosis intervals.
Disclosures: None. (Less)
Please use this url to cite or link to this publication:
author
; ; ; ; and
organization
publishing date
type
Contribution to conference
publication status
published
subject
pages
471 - 471
DOI
10.5281/zenodo.11068008
language
English
LU publication?
yes
id
6c0e92fb-ac5c-4608-9b8d-558cc17e3c70
date added to LUP
2024-09-13 17:13:46
date last changed
2025-04-04 13:57:49
@misc{6c0e92fb-ac5c-4608-9b8d-558cc17e3c70,
  abstract     = {{Background: We hypothesised that observed annual incidence rates of ANCA-associated vasculitis (AAV) are driven by<br/>sporadic short-term, high-intensity rates characterised as clusters of diagnosis events in time. Using data from Ireland, Sweden<br/>and Spain, we performed a changepoint analysis to discover time intervals with high and low rates of vasculitis diagnosis.<br/>Methods: We recruited 417 patients from the RITA-Ireland Vasculitis Registry. Adults (&gt;16 years) diagnosed with AAV after<br/>1/1/13 were eligible. Inclusion criteria were a definite diagnosis of microscopic polyangiitis or granulomatosis with polyangiitis.<br/>Patients with positive anti-GBM antibodies were excluded. The rate of diagnosis was modelled using a Poisson process changepoint<br/>model, which divides the timeline into intervals. Within each interval the time between diagnosis follows an exponential distribution<br/>with rate µ. Intervals with larger values of µ will have higher diagnosis rates. The number of changepoints was selected using<br/>cross-validation. We also applied this method to independent validation cohorts in Skåne (n = 351, Jan 1997 – Dec 2019, thus not<br/>including the pandemic period) and Barcelona (n = 82, Jan 2013 - present).<br/>Results: A five changepoint model fit the Irish data best (Figure). We observed several diagnosis incidence rates, ranging from<br/>µ=22/year over 42 months (pandemic period) to µ=60/y over 41 months (2015-2018). We also observed a recent post-pandemic<br/>reset to a higher diagnosis rate. No clusters of diagnosis events in time were clearly identified by the model. No changepoints were<br/>identified for the Skåne data, with a consistent diagnosis rate of µ=16/y across the cohort time frame. For the Barcelona data, we<br/>observed one changepoint in 2021, with associated rates of µ=5/y and µ=10/y. Observed intervals between diagnoses closely<br/>matched simulations generated by the estimated models for the Irish data (Figure). However, very short-term intervals (&lt;5 days)<br/>were underestimated by the model, possibly suggesting that these short intervals occur more frequently than expected by chance.<br/>We observed similar short interval results in the Skåne and Barcelona registries.<br/>Conclusions: We observed little clear evidence for short-term, high-intensity diagnosis incidence rates across 3 independent<br/>cohorts. For both Barcelona and Ireland (inception cohorts without 100% ascertainment), changes in incidence rate can likely<br/>be explained by changes in registry recruitment strategy, particularly over the pandemic. Our findings suggest that much of the<br/>perceived volatile behaviour of diagnosis events are consistent with the expected rates using well-established Poisson process<br/>models. However, our models failed to fully account for very short-term diagnosis intervals.<br/>Disclosures: None.}},
  author       = {{Coq, Matthieu and White, Arthur and Gisslander, Karl and Bordignon Draibe, Julianna and Nic an Riogh, Eithne and Little, Mark}},
  language     = {{eng}},
  month        = {{04}},
  pages        = {{471--471}},
  title        = {{Temporal clustering of ANCA-associated vasculitis occurrence}},
  url          = {{http://dx.doi.org/10.5281/zenodo.11068008}},
  doi          = {{10.5281/zenodo.11068008}},
  year         = {{2024}},
}