Sir,

Richtig and associates raised the question, whether calculated tumour volume would be a better prognostic indicator of survival of patients with choroidal melanoma than the largest basal tumour diameter (LBD) and height.1 They answered in the positive and also suggested that tumour volume be calculated in daily routine.

We tested their hypothesis with independent, consecutive, clinically unselected, and population-based data of 289 patients with choroidal and ciliary body melanoma with long-term follow-up.2 A Cox regression multivariate model that combined LBD (mean 13 mm, range 3–25) and tumour height (mean 7.8 mm, range 1–20), fitted to survival data significantly better (P=0.0031, difference between models; Table 1) than a model based on tumour volume as calculated by Richtig et al.1 Of models that included only one size parameter (LBD, height, and volume), the one based on LBD fitted to the survival data best and was superior to the one based on volume (P=0.020, Table 1).

Table 1 Comparison of Cox proportional hazards regression models, based on tumour dimensions and volume, as predictors of survival of patients with choroidal and ciliary body melanomaa
Full size table

The model that combined LBD and height was somewhat more strongly associated with survival than the model based on LBD alone (P=0.045).

The range of tumour dimensions in Richtig's study was more limited (mean LBD 10.4 mm, range 4.1–18.9, and mean height 5.7 mm, range 1.7–14.9). We consequently delimited our data to correspond to their LBD and tumour height limits (mean 12.5 mm, range 6–18 and mean 7.3 mm, range 2–14 mm, respectively). The statistical associations among this subset of 237 patients did not change (LBD plus height vs volume, P=0.0094; LBD vs volume P=0.0028, Table 1). The model which combined LBD and height, however, no longer differed statistically from that based on LBD alone (P=0.81).

Our unselected data set, which was larger than Richtig's (145 vs seven tumour deaths), showed that replacing tumour LBD as a prognostic indicator with tumour volume probably is not worth the effort in daily practice. A caveat in Richtig's study is that when multivariate analyses are applied to small samples—and the sample size in survival analysis is the number of events, not the number of patients who enter the study—a model which cannot be generalised is easily obtained.

All equations used for calculating tumour volume so far are rough approximations.1, 4, 5, 6 If true tumour volumes will be reliably obtained by imaging in the future, our conclusion should be reassessed.