Rob van Hest Capture-recapture Methods in Surveillance - RePub ...

Summary
1019-1715). To reduce possible bias due to heterogeneity among the patients, a covariate
capture-recapture model was specified, i.e. a capture-recapture model including
categorical covariates associated with the probability of capture in a register, because we
expected and observed regional differences in the incidence rate of Legionnaires’ disease.
The covariate model including “region” as covariate estimated 886 Legionnaires’ disease
patients (95%CI 827-1022), resulting in an estimated completeness of notification of
42.1%. The notified, ascertained and estimated average annual incidence rates of
Legionnaires’ disease were 1.2, 2.4 and 2.8 per 100 000 inhabitants respectively but higher
in the southern region of the Netherlands. We conclude that covariate capture-recapture
analysis, acknowledging regional differences of Legionnaires’ disease incidence, appears to
reduce bias in the estimated national incidence rate in the Netherlands. In Chapter 6 we
describe a systematic process of record-linkage, cross-validation, case-ascertainment and
capture-recapture analysis to assess the quality of tuberculosis registers and to estimate
the completeness of notification of incident tuberculosis cases in 1998. A saturated loglinear
capture-recapture model initially estimated an unexpectedly high number of 2053
(95%CI 1871-2443) tuberculosis cases, resulting in an estimated completeness of
notification of 63.2%. After adjustment for possible imperfect record-linkage and
remaining false-positive hospital cases a more parsimonious and better fitting capturerecapture
model estimated 1547 (95%CI 1513-1600) tuberculosis cases, resulting in a
completeness of notification of 86.4%. Truncated population estimators gave similar
results. In this chapter we demonstrate the possible impact of violation of the perfect
record-linkage and perfect positive predictive value assumptions on capture-recapture
estimates.
In the context of the second research question of this thesis, after chapter 6
describing capture-recapture analysis to estimate completeness of notification of
tuberculosis during one year at the national level in the Netherlands, in Chapter 7 we
estimate tuberculosis incidence and completeness of tuberculosis registration systems
during one year at the regional level in the Piedmont Region of Italy. A parsimonious
capture-recapture model estimated 704 (95%CI 688-728) tuberculosis patients, resulting
in an estimated completeness of notification of 79.1%. The overall estimated tuberculosis
incidence rate in the Piedmont Region is 16.7 cases per 100 000 population but varies
between different subsets of the population. We conclude that when multiple recording
systems are available, record-linkage can improve case-detection and capture-recapture
analysis can be used to assess tuberculosis incidence and the completeness of notification,
contributing to a more accurate surveillance of local tuberculosis epidemiology. In
Chapter 8 we estimated the completeness of tuberculosis notification in England at the
national level for four years to assess the performance of the Enhanced Tuberculosis
Surveillance (ETS) system, introduced in 1999. Due to the scale of this study (28 678
observed patients), for record-linkage of the hospitalised tuberculosis cases sophisticated
record-linkage computer software was required and the proportion of false-positive cases
among the unlinked hospital-derived tuberculosis records was estimated through a logistic
regression population mixture model. According to a saturated capture-recapture model
the estimated completeness of notification is 56.2%, highly inconsistent with prior
estimates. A truncated population estimator, Zelterman’s truncated Poisson-mixture
model, estimated the completeness of notification at 79.5%. We conclude that record-
176