C. Della Pepa, M. Costa#, R. Canaparo, L. Serpe, GP. Zara, F. Casale, E. Berno and M. Eandi.
Department of Anatomy, Pharmacology and Forensic Medicine; University of Turin. #Department of Electronics, Po litecnico of Turin. Italy.
5-Fluorouracil (5-FU) is a commonly used anticancer agent for treatment of solid tumours. Certain studies have reported c onflicting results between toxicity and/or therapeutic effects and individual 5-FU plasma concentrations. For this reasons some authors prop osed to evaluate the plasma levels of 5-fluorouridine (5-FURD), 5-fluoro-2′-deoxyuridine (5-FdURD) and 5-fluoro-5,6-dihydro-uracil (5-FDHU). We have developed and validated a new method to simultaneously determine by high performance liquid chromatography plasma concentrations of 5-FU and its three metabolites. We measured plasma samples of eighteen patients with a colonrectal cancer treated with 5-FU 400 mg/m2 i.v. bolus. The concentrations of 5-FDHU and 5-FU were detectable in all the subjects but 5-FURD and 5-FdURD concentrations were present in only eight subjects.
To correlate the plasma concentrations of 5-FU, 5-FURD, 5-FdURD, 5-FDHU, the toxicity and therapeutic effects we used the software Autoclass, developed by NASA, based on the bayesian approach.
The data concentrations, with its associated measurement error, form the database of the available experimental evidence. On the base of this definition, the software defines the specific structure of the basic functions that will be used. This determines the statistical significance of each of the free parameters, allowing finding an appropriate functional shape for the “a priori probability”. Autoclass completes the calculation of the “a priori probability” choosing (in the beginning randomly) the number of basic functions, all structurally identical. All modifiable parameters of the mixture are also started randomly. The algorithm EM modifies the values of the parameters in successive iterations, bringing the likelihood value to a maximum.
Some prototypes emerge from the analysis of the plasma concentrations measured in a specific time. The program puts the plasma concentrations of 5-FU and its metabolites in order of global influence, which is the role played by the attribute in determining classes. This means that the most influent attributes are those that mostly determinate the typical patterns in which the analysed phenomenon appears. Furthermore, the program puts the attributes in order of local influence inside of each class. Finally, the probability for each sample to be extracted from every single class was calculated. This is one of Autoclass222 most interesting features: while most of the other approaches assume from the beginning to be able to put all samples in the found classes, here this possibility emerges only at the end of the process, and, if it does emerge, it shows that the prototypes are actually well distinguished.
FdURD concentrations measured at 30 and 45 minute are most important to classificate a patient group with best therapeutic effect of 5-FU instead the FURD concentrations at 0.25 hour are related with a most toxic effects.
Reference: PAGE 9 (2000) Abstr 98 [www.page-meeting.org/?abstract=98]
Poster: poster