Paper detail

Statistical Problems in a Paper on Variation In Cancer Risk Among Tissues, and New Discoveries

Tomasetti and Vogelstein (2015) collected data on 31 different tissue types and found a correlation of 0.8 between the logarithms of the incidence of cancer (LCI), and the estimated number of stem cell divisions in those tissues (LSCD). Some of their conclusions however are statistically erroneous. Their excess risk score, "ERS" (log10 LCI x log10 LSCD), is non-monotonic under a change of time units for the rates, which renders meaningless the results derived from it, including a cluster of 22 "R-tumor" types for which they conclude, "primary prevention measures are not likely to be very effective". Further, r = 0.8 is consistent with the three orders of magnitude variation in other unmeasured factors, leaving room for the possibility of primary prevention if such factors can be intervened upon. Further exploration of the data reveals additional findings: (1) that LCI grows at approximately the square root of LSCD, which may provide a clue to the biology; (2) among different possible combinations of the primary data, the one maximizing the correlations with LCI is almost precisely the formula used by Tomasetti and Vogelstein to estimate LSCD, giving support to stem cell divisions as an independent factor in carcinogenesis, while not excluding other such factors.