Distribution of survival times of 12,000 head and neck cancer patients who died with their disease.

The lognormal parametric statistical model can provide, for groups of carcinoma cervix patients, good estimates of long-term survival fractions several years earlier than would otherwise be possible. The present paper extends this model work to head and neck cancer by using a minimum chi-squared test for goodness of fit (P greater than 0-05), to study the distribution of survival times of patients who died with their cancer present. Some 12,000 case histories were available from 7 hospital registries, 4 regional cancer registries and one national registry (the OPCS). All histories were followed up for at least 10 years subsequent to treatment and could be grouped into one of 8 cancer sites: antrum, floor of mouth, larynx, nasopharynx, pyriform fossa, post cricoid, tonsil and tongue. The theoretical distributions investigated were the lognormal, negative exponential and skew exponential. The results showed that the lognormal provided the best overall fit to the data, although the range of optimum values for the lognormal parameter, S, differed with cancer site. The optimum range did, however, usually include the value S=0-45. These results will now permit the second stage of validation of the lognormal model to proceed for head and neck cancers.