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

General discussion
Perfect positive predictive value assumption
The positive predictive value of the notification and laboratory registers for malaria is
considered to be high. Malaria is a specific disease, often requiring a history of recent
travelling in tropical areas, and unlikely to be diagnosed without confirmation or strong
suspicion. The laboratory plays a crucial role in the diagnosis through thick and thin
smear microscopy and serological antigen tests, all with a high specificity. Malaria has a
specific ICD-9 code and although a number of patients could have been admitted to
hospital for observation after developing fever following a tropical journey without a final
diagnosis of malaria, compared to Legionnaires’ disease and tuberculosis, the positive
predictive value of the malaria hospital episode register is also expected to be high.
For the notification of Legionnaires’ disease the criteria require a clinical picture
compatible with pneumonia and a confirmed or probable microbiology laboratory
diagnosis, according to the European Working Group for Legionella Infections (EWGLI)
definition. We demonstrate after record-linkage that these criteria apparently are not
applied uniformly over all registers. More than malaria, the registers could use different,
less specific, case-definitions, resulting in a proportion of false-positive cases in these
registers, e.g. as the result of the absence of pneumonia, the low positive predictive value
of the single Legionella antibody titre test or the absence of a specific Legionnaires’
disease code in ICD-9. 6,7
For tuberculosis the number of false-positive cases is assumed to be zero in the
reference laboratory register and limited in the notification register. The latter is due to a
good organisation of tuberculosis surveillance in the Netherlands and identification of
false-positive cases with an infection caused by non-tuberculous mycobacteria or another
diagnosis than non-tuberculous mycobacteriosis or tuberculosis through our crossvalidation.
However, foreign reports indicate considerable contamination of tuberculosis
hospital registers with false-positive cases. 8,9 The results of our study support these
observations as 62.4% of the unlinked hospital cases could not be verified through crossvalidation,
compared to 7.6% of the unlinked notified cases. These possibly remaining
false-positive cases likely contribute to considerable bias in the capture-recapture estimate.
Cross-validation and identification of assumed false-positive cases could not be
performed for malaria and Legionnaires’ disease. Violation of the perfect positive
predictive value assumption results in overestimation of the number of malaria,
Legionnaires’ disease and tuberculosis patients in the Netherlands.
Absence of specific interdependencies assumption
For all three infectious diseases studied, co-operation between the conventional registers
used (notification, laboratory and hospital) is expected, resulting in positive dependence
and underestimation of the number of cases in two-source capture-recapture models.
Therefore three-source log-linear capture-recapture analysis, incorporating possible pairwise
dependencies, was selected in the study design to reduce bias.
For malaria, significant interaction between the laboratory and notification registers is not
identified and could indicate absent pre-notification of laboratory results to the Public
Health Services at the time of this study. The explicable interactions of notification by
161