On the analysis of interval censored data and the serological response to treatment for syphilis
Abstract (summary)
The purpose of this research was twofold. The initial purpose was to use statistical modelling techniques to examine (i) the path of the serological response over time for patients treated for syphilis and (ii) the effect of potential explanatory variables on the treatment response. The time to seroreversion is an example of interval censored data. The second purpose of this thesis was to compare these alternate methods with the Turnbull method by means of a simulation study.
When the serological response to treatment was plotted as a function of time, using a logarithmic scale on both axes, the resulting plots were linear. The slope of the line during the first year after treatment was estimated for each individual. Proportional hazards regression modelling assessed the effect of this first year slope on the hazard of seroreversion in the presence of other potential explanatory variables.
The first year slope is a significant predictor of the hazard of seroreversion. The hazard of seroreverting is always greater when the first year slope is steeper but the estimate of the relative hazard depends on age. Results from the statistical modelling suggest that when the pretreatment RPR titer is n dilutions lower the hazard of seroreversion is increased by a factor of approximately $2\sp{n}$, for a given first year slope.
The stage of disease does not affect the hazard of seroreversion for a given first year slope and pretreatment RPR titer, but the predicted first year slope for an individual with primary or secondary syphilis is two units steeper than for an individual with early latent syphilis. Illness episode did not significantly affect the prediction of the first year slope, but for a given first year slope, the hazard of seroreverting is 2.2 times higher for an individual with an initial illness episode compared with an individual experiencing a repeat illness episode.
Results of the Monte Carlo simulation study clearly indicated that the endpoint of the interval should not be used to estimate the survivor function for interval censored data. The use of the midpoint of the interval produces an approximately unbiased estimate of the survivor function but the margin of error is large.
The results from the study indicate that the survivor function for interval censored data should be estimated using the Turnbull method. (Abstract shortened by UMI.)
Indexing (details)
Pathology;
Surgery;
Medicine
0571: Pathology
0564: Medicine
0576: Surgery