Tritiated thymidine labelling in vitro of human cancer of the breast: counting error and sampling error.

Tritiated thymidine labelling indices (TLIs) were determined on a number of primary cancers of the human breast. Twenty-two slides were chosen which demonstrated a wide range of TLIs and each was counted twice to assess 'counting error'. TLIs derived from successive counts of the same slide showed a coefficient of variance greater than 25% in half of the slides, but a significant difference between the two (P less than 0.05) in only 2 of 22 pairs. When TLIs derived from paired specimens taken from different sites in each of 22 tumours were compared, there was found to be a significant difference (P less than 0.05) in 17 of 22 pairs. This is 'sampling error' and is clearly a major source of inaccuracy when TLIs are derived from single small samples of heterogeneous tumours such as cancers of the human breast.