| Summary: | This study examines the effect of length of interval between routine tests on the risk of
carcinoma in situ (CIS) of the uterine cervix using cohort data from the B.C. population.
CIS is a symptomless disease only detected by screening. Because of this, special
methods are required for estimating incidence rates. Some case-control studies have used
prevalence odds ratios to estimate the relative risk of disease, usually invasive cancer,
from length of screening interval. But duration of disease is related to interval length
and hence prevalence rates cannot be used to estimate relative risk. A multivariate model
is fit to the incidence data using Poisson regression, and prevalence rates are fit with a
logistic regression model. The results for prevalence odds ratios indicate a positive
association between screening interval length and risk of disease whereas the results for
relative risk indicate a negative relationship. Theoretical screening models are
considered to examine the consequences of a case-control paradigm in which controls are
matched with cases on the basis of having had a screen near the date of diagnosis of the
case, the matching period. As the matching period shortens, the distribution of interval
lengths for controls converges to the underlying distribution, whereas the distribution of
interval lengths for cases equals the distribution of lengths of intervals which span a point
in time. The latter distribution favours longer intervals. The difference is not due to the
sampling of controls but, rather, to the relation between interval length and duration of
disease. A matched case-control study is simulated with the cohort data, and a conditional likelihood logistic regression model is fit. The results agree with those of a
logistic regression analysis of prevalence rates indicating a positive relation between
interval length and risk of disease. When the sampling of controls is weighted by
interval length the odds ratios approximate the relative risk. A possible explanation of
the surprising result that screening interval length is inversely related to risk of diagnosis
of CIS is that more cases are cured with time by the natural regression of disease than
by treatment of earlier stages of disease. On the other hand, incidence rate is negatively
related to recency and frequency of prior negative screens, possibly because of the
occurrence of false negative tests. However, the effect of regression predominates and
the unavoidable conclusion is that less frequent screening decreases the risk of diagnosis
of CIS.
|
|---|
