Rob van Hest Capture-recapture Methods in Surveillance - RePub ...
Rob van Hest Capture-recapture Methods in Surveillance - RePub ...
Rob van Hest Capture-recapture Methods in Surveillance - RePub ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
Underreport<strong>in</strong>g of tuberculosis <strong>in</strong> England<br />
likely to be managed by cl<strong>in</strong>icians familiar with notification of <strong>in</strong>fectious diseases.<br />
Apart from under-estimat<strong>in</strong>g the burden of tuberculosis, the implications for public<br />
health are limited as extrapulmonary tuberculosis patients are rarely <strong>in</strong>fectious.<br />
2. We tested our data us<strong>in</strong>g Zelterman’s truncated Poisson mixture model, which is also<br />
vulnerable to possible violation of underly<strong>in</strong>g assumptions. 34 This estimator and<br />
similar ones have been used <strong>in</strong> social sciences to estimate the size of hidden<br />
populations such as illicit drug users and homeless persons. 33,35-37 A recent<br />
publication compares three-source capture-<strong>recapture</strong> model estimates with the<br />
estimates of truncated models, <strong>in</strong>clud<strong>in</strong>g Zelterman’s model, for 19 datasets of<br />
<strong>in</strong>fectious disease <strong>in</strong>cidence and discusses the conditions where these estimates are<br />
similar or dissimilar. 38 The results of this study suggest that for estimat<strong>in</strong>g <strong>in</strong>fectious<br />
disease <strong>in</strong>cidence and completeness of notification <strong>in</strong>dependent (i.e. without pair-wise<br />
<strong>in</strong>terdependencies between the data sources) and parsimonious (i.e. <strong>in</strong>corporat<strong>in</strong>g<br />
one or two pair-wise <strong>in</strong>terdependencies between the data sources) three-source logl<strong>in</strong>ear<br />
capture-<strong>recapture</strong> models are preferable. However, when saturated models are<br />
selected as best fitt<strong>in</strong>g model and the estimates are unexpectedly high and seem<br />
implausible the data should be re-exam<strong>in</strong>ed with truncated models as a heuristic tool,<br />
<strong>in</strong> the absence of a gold standard, to identify possible failure <strong>in</strong> the saturated logl<strong>in</strong>ear<br />
model. When the truncated models produce a lower and more plausible<br />
estimated number of <strong>in</strong>fectious disease patients arguments are given that the<br />
estimates of the truncated models could be preferable. Table 8.5 shows the annual<br />
and overall estimated numbers of unobserved and total tuberculosis cases. The<br />
estimated numbers of unobserved tuberculosis cases were low compared to the<br />
structural source model, especially <strong>in</strong> 1999. From 2000 onwards the estimates reduce<br />
every year. Accord<strong>in</strong>g to Zelterman’s model estimated completeness of Notification<br />
was 68.5%, 63.8%, 73.6% and 76.4% for the years 1999-2002 respectively. The<br />
confidence <strong>in</strong>tervals do not overlap with the other models but <strong>in</strong>clude expected<br />
values of under-notification <strong>in</strong> 2001 and 2002.<br />
Hook and Regal state that “In no sense is there any proof or re-assurance that<br />
application of multiple-source log-l<strong>in</strong>ear estimators for any particular observed data on<br />
real populations results <strong>in</strong> a valid estimate, nor even necessarily produce an estimate<br />
closer to the true value than some alternative approach” and “if the saturated log-l<strong>in</strong>ear<br />
model is selected by any criterion the <strong>in</strong>vestigator should be particularly cautious about<br />
us<strong>in</strong>g the associated outcome”. 12 Confidence <strong>in</strong> the validity of capture-<strong>recapture</strong> results<br />
may reflect publication bias <strong>in</strong> favour of successful capture-<strong>recapture</strong> studies rather than<br />
the <strong>in</strong>herent strength of this methodology. 39<br />
Conclusion<br />
Record-l<strong>in</strong>kage, as performed <strong>in</strong> Enhanced Tuberculosis <strong>Surveillance</strong>, improves accuracy<br />
of surveillance data as well as completeness of case-ascerta<strong>in</strong>ment of tuberculosis.<br />
Hospital-derived data added a limited number of possible true tuberculosis patients. S<strong>in</strong>ce<br />
the <strong>in</strong>troduction of Enhanced Tuberculosis <strong>Surveillance</strong> the annual observed<br />
completeness of notification has <strong>in</strong>creased. This is most likely due to improvements <strong>in</strong><br />
case-report<strong>in</strong>g comb<strong>in</strong>ed with improved data collection and record-l<strong>in</strong>kage. This study<br />
121