Commentary on Colon Cancer Prevention Database
Fetch the database of all chemoprevention studies, or the Mirror site Cliquer pour base de données en français
Here are explained the methods. Next pages: results, and discussion including ACF MDF BCAC discussion. Back to the introduction: Home.

Tumor Free Rats: Efficacy of Chemopreventive Agents against Experimental Neoplasms in the Gut of Rodents.

How did we gather the data and build the databases?

rat All publications on the effect of dietary chemopreventive agents against colon carcinogenesis in rodents, and cited in the Current Contents Life Science from January 1989 to December 2001, were obtained from their authors. This initial database, published in "Nutrition and Cancer, 2002, Corpet and Taché" is updated each year. In addition, searches were done on the PubMed/Medline database, and on the web sites of the journals edited by the American Association for Cancer Research, and of the journals Carcinogenesis and Cancer Letters. Studies of specific agents were collected, as well as dietary studies. Each article was carefully read, and seven of them were rejected due to low plausibility. In addition, some engaging articles do not report suitable quantitative data. We did not include obvious toxic agent, e.g., methylcholanthrene. Articles reporting non-significant protection, or a promoting effect, were not included either.

We kept only the single most potent agent from each article, since our goal was to point out the most potent chemopreventive agents. However, some articles deal with many agents (e.g., Wargowich et al., 2000). From the thirty articles dealing with more than seven agents, we kept more than one agent. From each study, we decided to keep the most potent agent out of seven. In other words, we kept the top 14 percent agents from large studies. We inferred data from the height of the bars, when the original paper gives a histogram instead of a table.

In the 2002 article and database, data were gathered from 137 articles with the ACF endpoint, yielding 186 ACF preventive agents (because two or more agents were kept form articles dealing with more than seven agents). A primary ACF table (not shown) was built containing the mean number of total ACF per colon, of large ACF per colon, and of crypts per ACF in each group of rats (control and treated). In most studies, a "large ACF" contains four or more crypts. Since the database is updated each year, the number of articles and of agents increased continously.
In the 2002 article and database, data were also gathered from 146 articles with the tumor endpoint, yielding 160 tumor preventive agents or diets. A primary tumor table (not shown) was built with the incidence of tumors, the incidence of invasive carcinomas, and the mean tumor multiplicity in each group of rats (control and treated). Data from each article were double-checked by the second author, independently from the first one. Since the database is updated each year, the number of articles and of agents increased continously.

A potency index was then estimated for each agent in each study, by calculating the ratio of mean value in control rats divided by mean value in treated rats. For instance, in the Wargovich's study (27), a mean number of 193 ACF was observed in control rats, and of 28 ACF in rats given sulindac sulfide: the potency of sulindac sulfide to reduce the ACF number was calculated as 193/28 = 6.89. As an other example, in the Corpet's study (28), the incidence of tumors (confirmed by histology) was 22/27 in the control group, and 2/21 in the group of rats given polyethylene glycol (PEG): the potency of PEG to reduce the tumor incidence was calculated as (22/27)/(2/21) = 8.55. Thus, potency tells the times-fold reduction in a carcinogenesis endpoint due to the agent. When incidence is concerned, potency is the inverse of the relative risk (e.g., a potency of 8.55 corresponds to a relative risk of 0.12). The databases reports ACF and tumor potency tables. In addition, to help the reader, some endpoints were also reported as percent inhibition afforded by the agent, which is equal to 100 - 100/potency (e.g., a potency of 8.55 gives a 88.3% inhibition). When no tumor was seen in the control group, potency calculation was arbitrarily based on 0.5 tumors in the group.
      Our potency estimate does not take in account the dose used. We chose not to include the dose in the potency calculation, because most treatments were close to the maximum tolerated dose (40 or 80%): the authors often used the highest possible dose they could, to get the maximum protection. For instance, non-toxic agents like dietary fibers can be included in both rodent and human diets at a much higher level than agents that are more toxic and possibly more potent to prevent cancer (e.g., selenium, retinoids, and NSAIDs).
      The experimental design of each study was recorded, but not fully shown in the databases: the animal species and strain, the initiating carcinogen and doses, the treatment dose and duration. If the agent was given during the time of exposure to the carcinogen, the protocol was labeled "init" for initiation, if it was given after initiation, the label was "post" for post-initiation. In most studies, the agent was given to rats during both periods (during and after the carcinogen): these protocols were labeled "both." The post-initiation protocol has more clinical relevance since it may identify agents that prevent the recurrence and progression of precursor lesions for colon cancer (27). Most studies were done in rats initiated with dimethylhydrazine or its metabolite azoxymethane (237/283). We also chose to include in this review the few chemoprevention studies done with carcinogens different from dimethylhydrazine or azoxymethane, notably the heterocyclic amines like IQ or PhIP (24/283), a field much studied by Roderick H. Dashwood. And we included also the studies done in the mouse (22/283), because some promising agents were tested only in the mouse (e.g., protease inhibitor, sphingomyelin).

Each agent was classified within a class of agents, according to its chemical structure, its supposed mechanism, or its origin. We chose the following nine classes: amines modulators and DFMO; calcium and other mineral salts; fibers and bacteria; lipids; NSAIDs; PEGs; phytochemicals; vitamin A, D and retinoids; others.

Correlation between ACF and tumor enpoints was calculated and plotted, for the 57 agents that were present in both ACF and tumor tables, using the Systat 5.03 software (Systat inc., Evanston, IL). Independent studies with the same agent were not averaged, but each was included in correlation calculation. A second attempt to calculate correlation was made with the median potency obtained for each agent in different studies. However, both calculations gave exactly the same r value.

Databases (= tables) are available on websites with sorting abilities:
- a fast one at and
- a slower one, but with more frequent updates at
Large printed tables are hard to scan for a specific agent: the website helps the reader to fetch all studies on a specific agent; In addition, the electronic table enables the reader to rank the agents according to their potency to decrease any endpoint, and to sort data by class of agents or by design protocol.

Above are explained the methods. Back to main page (indroduction). Next pages: results, and discussion

Corpet DE & Taché S, 2002, Nutrition & Cancer.