The use of discriminant analysis to guide palliative treatment for lung cancer patients

The aim of this study was to develop a prognostic index for patients with inoperable non-small cell lung cancer which could predict survival to 3 months. This would enable less radiation dose to be given to patients where prognosis is limited by occult metastases, giving rise to less treatment morbidity and raising the therapeutic ratio. Data on 18 known prognostic factors were collected on 96 patients. Performance status, lymphocyte count, weight loss and extent of disease were the most predictive factors and were combined into an index. Logistic discriminant analysis was employed to give a numerical score of likelihood of survival to 3 months, ranging from 0 (not likely) to 1 (certain). In this first set of 96 patients, 16 deaths were observed before 3 months, of which 6 were predicted. There was one false positive prediction. Overall accuracy of prediction was therefore 89% with 99% specificity. The same 4 prognostic factors were measured on a second set of 80 patients. Nineteen died before 3 months, of which 5 were predicted with 2 false positives, giving an overall accuracy of 80% and 97% specificity. A probability of survival of less than or equal to 0.2, although highly specific, was only applicable to 9% of patients and this was the limiting factor in the clinical usefulness of the test. A 16-branch tree diagram allows any patient to be assigned a risk factor based on the four predictive factors at the first clinic attendance. Use of the index could encourage more rational prescribing of radiation dose.