Open AccessThe relationship between risk of death from clinical stage 1 cutaneous melanoma and thickness of primary tumour: no evidence for steps in risk
Author(s) -
M. Keefe,
Rona M. MacKie
Publication year - 1991
Publication title -
british journal of cancer
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 2.833
H-Index - 236
eISSN - 1532-1827
pISSN - 0007-0920
DOI - 10.1038/bjc.1991.357
Subject(s) - melanoma , proportional hazards model , medicine , breakpoint , prognostic variable , survival analysis , stage (stratigraphy) , regression , regression analysis , linear regression , oncology , surgery , statistics , overall survival , mathematics , biology , genetics , cancer research , paleontology , chromosomal translocation , gene
Previous reports have suggested that the relationship between survival and thickness of primary cutaneous malignant melanoma is not linear, but that there are natural breakpoints at which survival worsens in a step fashion. Nine hundred and ninety-seven cases of primary cutaneous malignant melanoma less than 9.75 mm thick, excised in Scotland between 1979 and 1983 inclusive, were examined to see if this could be confirmed. An adjusted Cox's regression analysis showed that age, sex, site and thickness were all significant predictors of survival. Thickness was grouped either empirically or by the breakpoints reported by other authors. It was then entered into a model either as a regressor or as a factored variable. The ranges 0-9.75 mm and 0-2 mm were studied separately. In the 0-9.75 mm range the factored variable was a statistically significant better fit than the regressor for each set of breakpoints, including an empirical analysis with eight groups. This suggests that there is no single best fit and that a step-effect is unlikely. Across the 0-2 mm range there was no significant improvement in the fit if thickness was entered as a factored variable, again indicating that a step effect is unlikely. We argue that there is no biological or statistical evidence to support the existence of natural breakpoints.