Overview: recent developments in the toxicity of environmental oxidants.

Squamous carcinoma of the head and neck: cured fraction and median survival time as functions of age, sex, histologic type, and node status Summary The multivariate lognormal survival model can be used to determine the relationship of prognostic covariates to two important parameters of malignancy. Cured fraction and median survival time among uncured patients. Analysis with this model revealed that cured fraction is primarily a function of histologic type and node status, while median survival time is primarily a function of age and node status. Patient sex was also related to likelihood of cure, but this association was of marginal significance. The symmetric impact of node status on both cured fraction and median survival time is consistent with known biologic principles. The strongly asymmetric relationships of histologic grade to cured fraction and age to survival time suggest, however, that likelihood of cure and survival time may not operate by identical biologic mechanisms. A variety of statistical models are now available for assessing survival. In order to select the appropriate model for a particular analysis, it is important to consider the time-course of the hazard under study. For example, to model the risk of developing cancer in humans or in experimental animals exposed to a carcinogen, one must select a hazard function (such as the Weibul or Gompertz) that increases over time, to match the progressive risk that is observed in these populations (Cook et al., 1969; Pike 1966). A progressive increase in risk is also required to model deaths from all causes among adult humans, since likelihood of death increases with age in this population. An altogether different model is needed, however, when studying survival with respect to a specific histologic type of tumour that has been treated by a potentially curative therapy. In order to focus on the tumour under study, distinction must be made between deaths due to this tumour and deaths due to unrelated causes. If reliable follow-up data is available, one can consider patients that died of other causes to be withdrawn alive at the time of death (Cutler & Axtell, 1969). Since risk of death from the tumour under study may eventually decline, as ascending hazard function would not give a good fit to observed survival (Figure 1). Furthermore, those patients that are cured are not at risk of death from their tumour, and thus a successful model must allow for a risk-free portion (or …