2007 - København - Denmark

PAGE 2007: Methodology- Other topics
Emilie Hénin

Estimation of patient compliance from pharmacokinetic samples

Emilie Hénin (1,2), Véronique Trillet-Lenoir (1,2,3), Olivier Colomban (1,2), Michel Tod (1,2), Pascal Girard (1,2)

(1) Université de Lyon, Lyon, F-69003, France ; (2) Université Lyon 1, EA3738, CTO, Faculté de Médecine Lyon-Sud, Oullins, F-69600, France ; (3) Service d’Oncologie Médicale, Centre Hospitalier Lyon-Sud, Hospices Civils de Lyon, F-69310, Pierre-Bénite, France

Objectives: Nowadays, more and more oral anticancer chemotherapies are developed either for cytotoxic or new targeted drugs. But this relatively new route of administration in oncology drives to new problems in treatment management and particularly to non-compliance, i.e. the deviance of the actual way patients take their treatment with the prescription. Several methods to measure compliance exist (pill count, patient interview, electronic monitoring ...) but there is no gold standard. Previous studies[1,2,3] aimed to estimate compliance from serum drug level. The objective of present study is to develop a methodology allowing the estimation of patient compliance defined as the immediate sequence of doses preceding the dosing interval where a blood PK sample has been taken (correct amount of drug taken or not - intake times supposed known). The method is firstly evaluated in silico. In the future, it will be applied to OCTO Phase IV study that measures the impact of non-compliance on efficacy and toxicity in cancer patients.

Methods: The idea is to estimate compliance according to a single PK concentration value measured on one dosing interval at steady state and sparse samples taken after first dose. The sparse sample after first dose are requested in order to provide the individual PK parameter estimates from which patient predictive profile will be estimated and compared to the steady state PK concentration using different metrics. In order to be able to estimate compliance using those limited data, several assumptions were to be made, that could be released later on: (i) times of all drug intakes are supposed to be known exactly; (ii) prescribed doses are assumed to be taken or not ("all-or-nothing" approach); (iii) only the previous n doses to a PK observation can be assessed, n being dependant on the half-life of the drug and our method sensitivity; (iv) there is no inter-occasion variability; (v) individual PK profiles can be derived from POSTHOC parameters estimated using sparse data sampled after first dose.
In order to decide which dose among the n previous doses taken before the SS PK sample, 2n different compliance profiles of dose taken/not taken have to be considered. The observed concentration value assumed to be taken at SS is compared to the concentration distribution predicted from in each compliance profile. Several metrics were considered for this comparison: (1) the Euclidean distance between the observed PK and the predicted ones simulated without residual errors. The compliance profile minimizing this metric is the one that is retained for the patient; (2) ratio of the density function at the observed value and at the predicted concentration. This metric had to be close to 1; (3) probability to observe a concentration value between the observed one and the predicted one. This metric had to be minimal (close to 0).

In silico evaluation: Gieschke et al.[4] proposed a population PK model of capecitabine (Xeloda®, Roche) and metabolites. Concentrations of FBAL (α-fluoro-β-alanine - capecitabine metabolite with the longest plasmatic half-life) were simulated according to Gieschke's cascade model. FBAL kinetics was correctly modelled by a one-compartment model with 1st order absorption and elimination and a lag time.
One thousand PK parameter sets were randomly drown according to their population distribution and FBAL (population half-life = 3 hours) concentrations following several compliance patterns (last 3 doses taken or not) were simulated. The use of the criteria allowed the discrimination between several compliance patterns and a correct estimation of compliance up to 2 doses in the past. As the most informative, the last dose intake was perfectly assigned. In addition, if the last dose was not taken, we were able to estimate correctly the compliance pattern.

Clinical application: A future Phase IV clinical trial (OCTO - cOmpliance to oral ChemoTherapy in Oncology) will aim to study the impact of non-compliance on efficacy and toxicity in cancer patients treated with an oral chemotherapy. Capecitabine (Xeloda®, Roche) will be prescribed to metastatic breast and colorectal cancer patients. The first administrations will be supervised in the hospital and patients will have several PK samples drawn after first dose in order to measure FBAL plasma concentrations. These data will allow the building of a population PK model of FBAL concentrations with a perfect compliance and the determination of individual PK parameters. Compliance will also be measured by electronic monitoring and the two compliance estimation techniques will be compared.

[1]. Lu J, Gries JM, Verotta D, et al. Selecting reliable pharmacokinetic data for explanatory analyses of clinical trials in the presence of possible noncompliance. J Pharmacokinet Pharmacodyn 2001; 28: 343-362
[2]. Lim LL. Estimating compliance to study medication from serum drug levels: application to an AIDS clinical trial of zidovudine. Biometrics 1992; 48: 619-630
[3]. Mu S, Ludden TM. Estimation of population pharmacokinetic parameters in the presence of non-compliance. J Pharmacokinet Pharmacodyn 2003; 30: 53-81
[4]. Gieschke R, Reigner B, Blesch KS, et al. Population pharmacokinetic analysis of the major metabolites of capecitabine. J Pharmacokinet Pharmacodyn 2002; 29: 25-47

Reference: PAGE 16 (2007) Abstr 1193 [www.page-meeting.org/?abstract=1193]
Poster: Methodology- Other topics
Click to open PDF poster/presentation (click to open)