Toxicological Sciences 43:1-9; 1998

Toxicological Sciences 43:10-18; (1998)

ERRATA

Igor Linkov#, Richard Wilson# and George M. Gray*

This is more than the usual erratum. We address both technical errors in our papers and several scientific issues raised at a meeting, of the Arkansas Medical Society by Dr. Joseph Haseman of the National Toxicology Program (NTP). The errors make a small contribution to the quoted results and the conclusions are unaltered. Most importantly, Dr. Haseman has disagreed with the decision of the authors, following a recommendation of a reviewer, to present the analyses unadjusted for survival and instead to simply compare cumulative distributions of anticarcinogens and non-anticarcinogens as a function of survival (in figure 7 of paper 2). In that plot there is a significant difference between anticarcinogenic and non-anticarcinogenic chemicals. About 20% of all rat anticarcinogens had a survival loss of more that 10% vs. 1-5% for non-anticarcinogens. A less dramatic difference was observed for mice anticarcinogens. We pointed out this difference and intended to convey that these 20% were probably spurious anticarcinogens for this reason. This was supported by an unpublished analysis in which the data were "survival adjusted", according to a simple method used by NTP in early reports, leading to a number of chemicals no longer being classified as "anticarcinogens". The survival adjustment did lead several additional chemicals not previously called anticarcinogenic being identified as such. In this adjustment, any animals not surviving until the appearance of the first tumor, of the type being analyzed, in control animals are removed from the analysis. After this adjustment the curves for anticarcinogens and non-anticarcinogens were very nearly superimposed. Unfortunately space did not permit of presenting the results in both ways. This survival adjustment also means that 2 of the 8 chemicals in Table 1 of Paper 2 are no longer classified as anticarcinogens.

In addition, there is a misprint in the first paper (the numbers in column 2 and 4 of Table 6 were about 50% too large) as well as computer errors and problems induced by failure to understand some of the limitations of the CBDS database of the NTP. None of these changes the number of "anticarcinogens" by more than a few percent. These points are:

(a) There was an error in our coding of the Cochran-Armitage trend test. Correcting this changes the fraction of "anticarcinogenic" chemicals from the published (22.8)/(25)/(47.8)/(44.6)% to (23.7)/(24.7)/(45.2)/(43.9)% where we list the changes for the 4 combinations of species and gender in the order (MM)/(FM)/(MR)/(FR).

(b) For some chemicals and tumor types only a few of the animals were examined at the intermediate dose. There is some inconsistency in the NTP publications about which sites were examined, and some tumors are noted in sites reported as not subject to histopathology. These two factors resulted in incorrect calculation of the number of animals at risk (denominator). The number of chemicals with at least one "anticarcinogenic" response changes from the published (22.8)/(25)/(47.8)/(44.6)% to (21.8)/(22.8)/(42.3)/(41.7)% after the correction for the reduced denominator and dose trend test.

(c) There are a few discrepancies between what is on the tape provided to us in 1987 (and we were told to and do trust pending resolution of the discrepancies) and in the published and numbered Technical Reports (TR) for a specific bioassay. We have identified only four major discrepancies between the CBDS tapes and the technical reports - all in the early stages when the reports were issued by NCI. Correction for this changes the percentage of anticarcinogenic chemicals by a small amount from (22.8)/(25)/(47.8)/(44.6)% to (22.4)/(25.3)/(47.7)/(44.2) if we remove these chemicals from consideration.

In addition, following legitimate scientific questions were raised. These have no clear "correct" answer, but below we present results when we make changes suggested by Dr. Haseman.

(d) Dr. Joseph Haseman objected to the tumor and site classification scheme used by us (and in several papers previously starting with that of Bailar et al., Risk Analysis 8:485-497;1988). In addition, he suggested that inconsistencies in the way NTP (or contractor) pathologists recorded tumors or actual changes in tumors with dose, might lead to spurious carcinogenic and anticarcinogenic effects. Specifically, in our classification "neoplastic liver nodules" were assigned to liver carcinomas (our classification 64) not liver adenomas (our classification 7) as indeed was done in some early NTP reports. Assigning the neoplastic nodules to classification 7 reduces the number of "anticarcinogenic" chemicals from (22.8)/(25)/(47.8)/(44.6)% to (22.4)/(24.4)/(47.4)/(44.2)%. Dr. Haseman also pointed out a probable inconsistency in the pathologists' assignments when pituitary adenomas were seen. Sometimes they were called pituitary adenomas (our classification 10) and sometimes chromophobe adenomas (our classification 36). Combining pituitary tumors in one classification reduces the number of "anticarcinogenic" chemicals from (22.8)/(25)/(47.8)/(44.6)% to (22.4)/(24.4)/(46.5)/(43.3)%. We have searched for other possible places where this might have been a cause of problems, but found none. It is important that to have a spurious effect it is not enough to have inconsistency between pathologists or between experiments but inconsistency must exist between dose groups in a single experiment.

(e) There were a number of cases where the number of liver adenomas was reduced with chemical dosing in the presence of an increase in carcinomas. This apparent anticarcinogenic response could be spurious for either one of two reasons. There is an obvious suspicion that the pathologists only choose the most aggressive tumor in any one organ. Alternatively many scientists believe that liver adenomas sometimes, if not always leads to liver carcinomas. This would lead the number of recorded liver adenomas to go down as the number of liver carcinomas goes up. If we omit from the list all chemicals called "anticarcinogenic" because of falling adenoma rates the number of anticarcinogens drops less than 1%.

(f) Some of the chemicals and substances in the computer database were never the subject of an NTP technical report and Dr. Haseman suggested these were not appropriate studies for analysis. It is an important question whether or not to include them. On the one hand it might seem that NTP do not trust the results if there is no hard copy, and on the other hand if no clear reason is stated for the distrust any exclusion might be an undesirable bias. We did the analysis both with and without these "unpublished" chemicals. When we restrict the analysis to those for which NTP assigned a number we increase the percentage of "anticarcinogenic" chemicals and the conclusions of the papers are very slightly enhanced. In the published papers we used (and still use) all data available to us. But the numbers of the more limited group (and modified figures that would be appropriate thereto) are also available.

The fraction of "anticarcinogenic" chemicals changes from the published (22.8)/(25)/(47.8)/(44.6)% to (19.8)/(22.1)/(40.9)/(38.6)% when the effects (a), (b), (c), (d) ,and (e) are included and is reduced the further 20% to (14.0)/(17.5)/(35.7)/(29.2) % if the survival the survival adjustment is also made The final number of "anticarcinogenic" chemicals (for p<0.01) is still considerably larger than the fraction of anticarcinogenic chemicals suspected to be random responses (9.4)/(9.1)/(16.7)/(12.6) % found by the Monte Carlo program. Thus the principal conclusion of our two papers that neither random effects nor weight depression accounts for all the "anticarcinogenic" effects remains, albeit with somewhat less significance. We emphasize that the large (30% - 50%) random contribution, and other non-random errors makes it impossible to claim that any particular chemical is truly anticarcinogenic without further information. We also repeat that the random responses may be a slighter overestimate because as seen in figures 1 & 2 of paper 1 the model overestimates the number of chemicals for p < 0.1 when the random effects dominate. The authors are studying other possible reasons why the results may be spurious, including a variation in control tumors.

Further analyses using other tumor classifications are underway and will be presented in due course. Copies of all figures and tables with and without the various corrections are available on request from the second author (RW)

The authors are grateful to Dr. Haseman for his careful criticism and to Mr. Azat Shagiakhmetov for his help in performing the calculations herein.