Clinical monitoring of digoxin plasma concentration is recommended because of both a large inter-animal variation in digoxin pharmacokinetics and the small ratio of toxic/therapeutic concentrations. Digoxin concentration data from clinical therapeutic drug monitoring (TDM) are used to adapt dosage regimes. Hierarchical mixed effects modeling of TDM data can be used to identify factors that are covariables for pharmacokinetic model parameters. Knowledge of such pharmacokinetic model covariables may help to improve dose calculations and predictions and can also help to understand the pharmacodynamics of drug action and toxicity. Such models have been developed for digoxin in humans (Sheiner et al., 1977). The pharmacokinetics of digoxin in healthy dogs has been described previously (Button et al., 1980b) and dosage regimes have been examined in dogs with induced (Button et al., 1980a) or naturally occurring congestive heart failure (Tobias et al., 1989). These previous studies showed that dose adjustments made according to assumed covariables for digoxin pharmacokinetics did not accurately achieve the desired plasma concentrations in individual animals (Tobias et al., 1989). Individualized dosage regimes using TDM are necessary to achieve desired plasma digoxin concentration in dogs (Button et al., 1980a). We applied hierarchical mixed effects modeling to examine the pharmacokinetics of digoxin in 161 cases with naturally occurring heart disease in dogs. Of these, 32 were examined prospectively and 129 retrospectively. We aimed to describe the pharmacokinetics of digoxin in dogs with heart disease and to identify covariables, which could be used to predict an individual dog’s digoxin pharmacokinetics. Samples for digoxin analysis were taken at steady-state, after known but varying doses, dose intervals and dose-to-sample intervals, from routine clinical TDM cases. These samples were analyzed using commercial kit RIA or EMIT assays. Covariable data collected included serum creatinine, serum potassium, body weight (kg) x (x=0.75, 1), estimated body surface area, age and the formulation administered (Lanoxin tablets or pediatric elixir, Glaxo Wellcome). A one compartment open pharmacokinetic model for oral administration was applied to the data using proprietary software (PPharm v 1.51, Innaphase, France). Initial modeling identified two distinct populations of cases, one with a markedly slower absorption rate constant (ka) than the other (Fig. 1). Analysis for k-means clustering based on Euclidean differences identified two populations, one of 19 cases and one of 142 cases (Systat v 8.0, SPSS Inc. Chicago, IL, USA). The means of the two sample populations for ka differed statistically using Student’s t-test (PB0.01) (Table 1.). No correlation was identified between the 19 slow absorbing cases and any known covariable, including dose formulation (Fig. 1). All 19 slow absorbers were from the 32 cases (19/32=59%) which had been admitted to the hospital, a stressful event, within 2 h prior to dosing: in contrast, the remaining 129/142 (91%) cases had been dosed at home prior to sample collection. Acute stress has been associated with release of corticotrophin releasing factor (CRF) and intracerebroventricular infusion of CRF has been shown to delay total gastric emptying in dogs (Lee & Sarna, 1997). Therefore, we hypothesize that the slower absorption of orally administered digoxin was due to delayed gastric emptying in 59% cases admitted to hospital for dosing. Stress and pain have previously been shown to alter drug absorption and metabolism (Jamali & Kunz-Dober, 1999). Knowledge of a delay in absorption of drugs administered to patients by the oral route at the time of hospitalization, if confirmed, would influence selection of the route of drug administration for such patients. Multimodal populations violate the assumption of normal/ log-normal population distribution for parametric population pharmacokinetic analysis. Therefore, all 19 cases identified to have a low ka were deleted from further analyses. For the remaining 142 cases the number of observations per subject