Increased oral ganciclovir bioavailability in HIV-infected patients with chronic diarrhoea and wasting syndrome—a population pharmacokinetic study

Abstract

Aims

Despite a lack of data, the antiviral agent ganciclovir is not indicated in AIDS patients with diarrhoea because of its presumed poor oral bioavailability. To assess the effect of diarrhoea on ganciclovir intestinal absorption, we conducted a pharmacokinetic study in 42 HIV-infected patients categorized into three groups: A, HIV stage A and B (n = 15); B, AIDS stage C (n = 13); C, AIDS with chronic diarrhoea and wasting syndrome (n = 14).

Methods

Each patient was evaluated for nutritional (body mass index, albumin, transferrin serum levels), inflammatory (haptoglobin, orosomucoid), immunological (CD4 count, plasma viral load) and intestinal (d-xylose test, faecal fat and nitrogen output, intestinal permeability) status. Ganciclovir (1 g) was administered orally to fasted patients. Six blood samples were collected over 24 h. Serum was analysed for ganciclovir by h.p.l.c. Population pharmacokinetic analysis was performed using a nonlinear mixed effects modelling program, MP2.

Results

Mean intestinal permeability (lactulose/mannitol urinary ratio) was increased in group C (0.2) compared with group A (0.05) and B (0.1) patients. Drug concentration-time profiles were best described by a two-compartment model. Apparent oral clearance (CL/F) and central volume of distribution (V₁/F) were influenced by clinical status (group). For groups A and B combined, final parameter estimates of CL/F and V₁/F were 256 ± 98 l h⁻¹ and 1320 ± 470 l, respectively. Final parameter estimates for group C were 118 ± 108 l h⁻¹ and 652 ± 573 l for CL/F and V₁/F, respectively. The 95% confidence intervals on differences between A and B combined and C were statistically significant ([ + 70, + 206] for CL/F, and [+ 314, + 1022] for V₁/F). Compared with groups A and B, ganciclovir CL/F was significantly decreased in group C patients.

Conclusions

AIDS patients with diarrhoea and severe disease may benefit from ganciclovir therapy, but a dose adjustment may be required according to their digestive and immunological status.

Introduction

Ganciclovir (9-[1,3-dihydroxy-2-propoxymethyl] guanine) is a poorly absorbed acyclic guanine nucleoside analogue used in the prevention and maintenance therapy of Cytomegalovirus (CMV) retinitis in Human Immunodeficiency Virus (HIV)-infected patients without established CMV end-organ disease [1, 2]. To date, scant pharmacokinetic studies of this drug have included HIV-infected patients with diarrhoea and weight loss [1, 2]. Thus, the effect of diarrhoea (present in 10–60% of patients at various stages of disease with and without intestinal malabsorption and increased permeability [3–7]), on ganciclovir oral bioavailability is not well defined, making safe and effective dosing strategies difficult in such patients. Moreover, oral ganciclovir is generally contra-indicated in HIV-infected patients with diarrhoea, as it is assumed not to be absorbed. Pharmacokinetics of drugs have consistently been shown to be unpredictable in HIV-infected patients due to both clinical status and concomitant multiple drug intake, leading to an increased toxicity (increased drug exposure) or a lower efficacy (decreased drug exposure) [8–10]. In addition, pharmacokinetic studies are difficult to perform in HIV-infected patients, both for ethical reasons and the inability to complete a study protocol in association with a complicated medical regimen [11]. Yuen et al. [12] previously demonstrated that, following intravenous administration, patients with Acquired Immunodeficiency Syndrome (AIDS) and CMV retinitis had a serum ganciclovir clearance approximately 40% greater than HIV-positive patients who were only shedding CMV into the urine. HIV-infected patients also exhibited a serum ganciclovir systemic clearance that was four fold higher than that for patients with solid organ transplants [12].

Ganciclovir is very poorly absorbed following oral administration, with an absolute bioavailability ranging from 6 to 9% [1]. Thus a small decrease in bioavailability could dramatically alter therapeutic efficacy. In addition, over 90% of this drug is excreted unchanged through the kidneys, indicating that it is not appreciably metabolized [1, 12]. A previous study in bone marrow transplant recipients demonstrated that diarrhoea related to acute graft vs host disease of the gastrointestinal tract did not appear to adversely affect absorption of oral ganciclovir, as its absolute bioavailability remained stable (7.2%) [13]. Similar information remains unavailable in HIV-infected patients with diarrhoea and weight loss. Therefore, we conducted a prospectively designed population pharmacokinetic study to assess the influence of diarrhoea and weight loss on oral ganciclovir pharmacokinetics in HIV-infected patients who denied recent ganciclovir intake. Population methods have already been used in HIV-infected patients, allowing for a large number of patients to be included with minimal blood sampling [10, 12, 14].

Methods

Subjects

Adult HIV-infected in- or outpatients, whatever the risk factors and duration of seropositivity, and who denied ganciclovir intake during the prior month, were enrolled in our study. Exclusion criteria included neutropenia (leucocytes < 1500/mm³), severe renal failure (creatinine clearance < 20 ml min⁻¹ 1.73 m⁻²) or digestive disease that may interfere with the absorption and permeability tests (e.g. coeliac disease, inflammatory bowel disease, tropical sprue, Whipple disease). For each patient, a complete clinical examination was performed that included weight, height, gastrointestinal symptoms, and history of opportunistic infections. Concomitant medications taken at the time of the study were recorded and not discontinued, taking into account the constraints inherent to multiple drug therapy. HIV infection was confirmed by enzyme-linked immunosorbent assay (ELISA) and Western blot analysis. Each patient provided written informed consent. This study was approved by the Ethics Committee of Paris, Hôpital Saint-Louis.

Patients were divided into the following three distinct groups according to the presence of diarrhoea and/or weight loss:

Group A: HIV-infected patients without AIDS-defining illness (stage A or B),

Group B: AIDS patients (stage C) without diarrhoea and weight loss,

Group C: AIDS patients (stage C) with diarrhoea (defined as more than three loose bowel movements a day for at least 4 weeks) and/or weight loss (loss of 10% or more of body weight from baseline weight during the last year of follow-up).

In accordance with the principles of population pharmacokinetics [15] and with the oral ganciclovir pharmacokinetic profile [1], we prospectively enrolled at least 12 patients in each group. Each patient was admitted for a 24 h period to our Infectious Disease Department to complete the study. For each patient, haemoglobin serum level, complete leucocyte and platelet counts were performed. Renal function was assessed according to the Cockroft & Gault formula for estimated creatinine clearance [16]. Nutritional evaluation included body mass index (BMI) and albumin, transferrin, orosomucoid and haptoglobin serum levels. Clinical staging was determined according to the 1993 CDC revised classification [17]. Immunological assessment included CD4 cell count and plasma HIV-1 RNA serum level measurement (Reverse-Transcriptase Polymerase Chain Reaction, Roche Molecular Diagnostic Systems, Nutley, N.J., USA).

Study design

Following an overnight fast, intestinal permeability was assessed in each patient and in eight healthy control volunteers (three men and five women from our medical staff who denied any history of diarrhoea and/or increased intestinal permeability, and who gave written informed consent prior to be tested for their permeability) by the 5 h lactulose/mannitol (L/M) urinary ratio. In brief, 10 g lactulose (Lactulose Biphar®, Procter & Gamble Pharmaceuticals, Neuilly sur-Seine, France) and 2.5 g d-mannitol 20% (Pharmacie Centrale des Hôpitaux, Nanterre, France) were mixed in 250 ml of water and ingested [7, 18]. Since lactulose is minimally absorbed under normal conditions via a paracellular pathway and eliminated unchanged through the kidney, an increased L/M ratio compared with control subjects indicates a significantly increased intestinal permeability for large molecules, including drugs [7, 18]. Intestinal absorption tests included a standard d-xylose test (d-xylose serum level determined 2 h after ingestion of 25 g d-xylose) and 72 h faecal fat and nitrogen output determinations. Patients were asked to maintain their usual diets. Two stools samples were analysed for the presence of opportunistic parasites and bacteria. Oral ganciclovir was supplied as 250 mg hard-gel capsules (Roche Laboratories, Nutley, NJ, USA). After an overnight fast, each patient took a single dose (1000 mg) of ganciclovir with 150 ml water. Blood samples were collected at 0.5, 2, 4, 8, 12, and 24 h after drug administration. Adverse events (headache, behavioural change, rash, fever, nausea, vomiting) were recorded the following morning. Serum was separated from whole blood by centrifugation and frozen at −30 °C pending analysis.

Analytical methodology

Serum ganciclovir concentrations were measured by h.p.l.c. Ganciclovir was isolated from serum proteins by precipitation using 24 µl perchloric acid, followed by 15 min centrifugation (2700 g). Supernatant (300 µl) was treated with potassium hydroxide (pH 7) and centrifuged again for 10 min. Resultant supernatant (150 µl) was injected onto a Waters' symmetry C18 column (250 × 4.6 mm, particles of 5 µm, Waters, Milford, MA, USA). The mobile phase consisted of 1.5% acetonitrile (v/v), 98.5% potassium hydroxide (100 mm) and 1% (v/v) octane sulphonic acid (2.5 mm, Waters, Milford, MA, USA), pH 2.5. Quantification of ganciclovir was achieved by simultaneous ultraviolet (u.v.) at 254 nm, and fluorescence detection (excitation and emission wavelengths at 285 nm and 380 nm, respectively) [19]. This method exhibited linear ranges of 20–1000 ng ml⁻¹ (u.v.) and 20–2000 ng ml⁻¹ (fluorescence). The lower limit of quantification was 20 ng ml⁻¹ for both detection modes. Within and between-run coefficients of variation for the two modes both ranged from 2.1 to 19% over the entire concentration range. No interference was noted between ganciclovir and the antiretroviral agents saquinavir, ritonavir, indinavir, nelfinavir, zidovudine, didanosine, zalcitabine, stavudine and lamivudine.

Population pharmacokinetic modelling

Concentration-time data were analysed using a Nonlinear Mixed Effects Modelling approach, implemented in the program MP2 (Micropharm population, INSERM, Paris, France) [20]. One- and two-compartment pharmacokinetic models with first-order or saturable absorption were fitted to the data. Parameters of the final structural model included systemic clearance (CL/F), central and peripheral compartment volumes (V₁/F, V₂/F), intercompartmental clearance (Q) and absorption rate constant (ka). F denotes bioavailability. We investigated several error models (i.e. proportional error model with constant coefficient of variation and additive random effects model) to describe interpatient and residual variability.

The influence of each patient covariate on CL/F, V₁/F, V₂/F, Q and ka were tested. Such covariates included clinical status (group), gender, age, weight, serum creatinine, albumin, haptoglobin, orosomucoid, transferrin, blood haemoglobin, CD4 count, and the absorption and permeability tests. Because albumin, transferrin, haptoglobin and orosomucoid may be severely altered in patients with advanced HIV disease and active opportunistic infections, wasting syndrome, or diarrhoea (especially when intestinal inflammation is present [4–7]), their influence (positive or negative) on absorption, diffusion or elimination processes were systematically investigated. In addition, these variables are easily accessible in daily practice, unlike intestinal absorption tests and/or invasive gut biospies. The full and reduced models (less one parameter) were compared using the chi-squared test. A decrease of at least 6.63 (P < 0.01, one degree of freedom) of the objective function value compared to the basic pharmacokinetic model (with no variable) was required for the addition of a single parameter in the model. An intermediate multivariate model was then obtained including all significant covariates. To keep only those covariates with the largest contribution to predict ganciclovir pharmacokinetics in a final multivariate model, a change of 10.82 (P < 0.001) of the objective function was required for the retention of a single parameter during backward stepwise multiple regression analysis. Non-hierarchical models (models with first-order or saturable absorption) were assessed by comparison of their respective objective function values and by visual examination of residual plots. Predictive performance of the model was assessed in terms of bias (mean prediction error, ME) and precision (root mean square prediction error, RMSE) as follows:

1

2

Where PEi stands for the difference between the i^th measured and predicted ganciclovir concentration pair taken at a given time, n being the number of pairs.

Statistical analysis

Data are expressed as mean ± s.d. Comparisons between groups were tested by analysis of variance (anova) for quantitative variables, followed by Scheffe's posthoc test adjusted for multiple between-group comparisons when an overall significant difference was found (Superanova 1.1f, SAS Institute Inc., Cary, NC). Qualitative variables were compared using the chi-squared test with Yates correction. A P value less than 0.05 was considered significant.

Results

Demographic, clinical and biological data

Forty-two patients (33 men and 9 women, ranging in age from 24 to 52 years) were enrolled over a 5 month period (January to May 1998). Fifteen patients belonged to group A, 13 to group B and 14 to group C. The main clinical characteristics are listed in Table 1. The route of HIV infection included homosexual contact in 16 cases, intravenous drug abuse in 6 cases, and heterosexual contact in 20 cases. The number of homosexual, heterosexual and intravenous drug users did not differ between groups. Half of the patients in group A were of African or Asian descent (Table 1). Seven patients did not receive any combination therapy at the time of the study (two in group A, one in group B, and four in group C). Seven patients were taking two nucleoside analogues (three in group A and B, one in group C). Two-thirds of the patients in each group were on triple antiretroviral therapy (one HIV-1 protease inhibitor and two nucleoside analogues). One patient in each group was on quadruple therapy (two HIV-1 protease inhibitors and two nucleoside analogues). In addition, most of the patients were receiving concomitant medications at the time of the study, e.g. cotrimoxazole, benzodiazepines, and antidepressants. No difference in treatment regimen, especially antiretroviral therapy, was found between groups (P = 0.3, chi-squared test with Yates correction).

Table 1.

Demographic, clinical and biological data in 42 HIV-infected patients (mean ± s.d.).

	Group A	Group B	Group C
Number (male/female)	15 (12/3)	13 (9/4)	14 (12/2)
Caucasians/others	8/7	10/3	11/3
Age (years)	37 ± 9	36 ± 7	39 ± 7
Weight (kg)	66 ± 9	64 ± 11	59 ± 14
Ganciclovir dose (mg kg−1)	15 ± 2	16 ± 3	18 ± 6
Body Mass Index (kg m−2)	21.9 ± 2.5	21.4 ± 3.1	19.1 ± 3.8#
Body weight loss (%)	4 ± 4	3 ± 5	19 ± 8**
Haemoglobin serum level (g dl−1)(1)	13.6 ± 1.4*	12.2 ± 1.5	11.3 ± 1.4
Lymphocyte count (/mm3)(2)	1829 ± 771	1458 ± 571	989 ± 830
Creatinine clearance (ml min−1 1.73 m−2)	103 ± 26	107 ± 25	96 ± 28
Albumin serum concentration (g l−1)(3)	41.4 ± 9	41.9 ± 5.3	27 ± 10.1**
Transferrin serum concentration (g l−1)(4)	2.33 ± 0.5	2.36 ± 0.3	1.8 ± 0.4(§)
Haptoglobin serum concentration (g l−1)(5)	1.1 ± 0.7	1.4 ± 0.8	2.7 ± 1.6§
Orosomucoid serum concentration (g l−1)(6)	0.8 ± 0.4	1.0 ± 0.4	2.0 ± 0.9**
CD4 cell count (/mm3)	322 ± 172	300 ± 255	56 ± 82§
Plasma viral load (log10 copies/ml)	3.72 ± 1.11	4.03 ± 1.24	4.91 ± 0.95§

^#P = 0.03 vs A and B (anova and Scheffe's test).

^**P ≤ 0.001 vs A and B (anova and Scheffe's test).

^*P < 0.05 vs C (anova and Scheffe's test).

^§P < 0.05 vs A and B (anova and Scheffe's test).

⁽¹⁾Normal range, 12–16 g dl⁻¹

⁽²⁾normal range, 1500–4000/mm³

⁽³⁾normal range, 35–50 g l⁻¹

⁽⁴⁾normal range, 2–4 g l⁻¹

⁽⁵⁾normal range, 0.8–2.2 g l⁻¹

⁽⁶⁾normal range, 0.5–1.5 g l⁻¹.

Ten out of 13 patients in group B presented with at least one active opportunistic infection at the time of the study: Kaposi's sarcoma (n = 3), tuberculosis (n = 2), oesophageal candidiasis (n = 2), Varicella zoster viral encephalitis (n = 1), biopsy proven multifocal progressive leucoencephalopathy (n = 1), Toxoplasma gondii brain abcess (n = 1), and/or Mycobacterium avium intracellulare infection (n = 1). Likewise, 10 out of 14 patients in group C presented with at least one active opportunistic infection at the time of the study: Pneumocystis carinii pneumonia (n = 4), oesophageal candidiasis (n = 4), large bowel histoplasmosis (n = 1), microsporidiosis (n = 1), cryptosporidiosis (n = 1), Isospora belli (n = 1), Mycobacterium avium intracellulare infection (n = 1), and/or Toxoplasma gondii infection (n = 1).

Biological, nutritional and immunological data are presented in Table 1. As expected, BMI (normal range 20–25 kg m⁻²), albumin and transferrin serum concentrations were significantly lower in group C patients compared with groups A and B. Conversely, serum concentrations of the inflammatory proteins haptoglobin and orosomucoid was significantly higher in group C patients compared with groups A and B (Table 1). Patients in the group C were significantly more immunocompromised compared with patients in groups A and B, as indicated by a lower CD4 cell count and a higher plasma HIV-1 viral load. Renal function was normal in all of our patients and did not differ between groups. Haemoglobin serum level was highest in group A patients.

Intestinal absorption and permeability tests

Stool examination for bacteria and parasites was positive in only seven patients with diarrhoea. d-xylose test was available in all but one patient (group B). Faecal fat and nitrogen output were available in all but four patients (one in group A and B and two in group C). Intestinal absorption and permeability test results are presented in Table 2. None of the absorption tests, including d-xylose serum concentration, faecal fat and nitrogen excretion significantly differed between groups (P > 0.1 for all comparisons, anova) (Table 2). The lack of difference between groups could be explained by the large variability observed in the absorption test results. Indeed, abnormal results were present in each group, with and without diarrhoea, as previously reported [4, 6, 7]. The d-xylose serum concentration was low in 3, 5, and 5 patients in groups A, B and C, respectively. Six patients in group A, 3 in group B, and 5 in group C had abnormally high faecal fat excretion. Faecal nitrogen excretion was high in 5, 1, and 2 patients in groups A, B, and C, respectively. The lactulose/mannitol urinary ratio was available in all but 5 patients (one in groups A and B, 3 in group C). Although intestinal permeability was increased in group C patients (urinary ratio of 0.19 ± 0.28), it was highly variable, and no statistically significant difference was reached between any patient group, nor between each patient group and our group of eight healthy volunteers (anova).

Table 2.

Intestinal absorption and permeability tests in 42 HIV-infected patients with and without diarrhoea and weight loss (mean ± s.d.).

	Controls*	Group A	Group B	Group C	P value
d-xylose serum concentration (mmol l−1)(1)	–	2.14 ± 0.74	2.15 ± 1.2	2.16 ± 0.84	0.5
Faecal fat excretion (g day−1)(2)	–	5.4 ± 2.4	5.2 ± 4.8	5.1 ± 2.5	0.14
Faecal nitrogen excretion (g day−1)(3)	–	1.5 ± 0.8	1.7 ± 0.9	1.2 ± 0.6	0.27
L/M ratio**	0.043 ± 0.017	0.045 ± 0.03	0.08 ± 0.1	0.19 ± 0.28	0.32

⁽¹⁾Normal value, > 1.7 mmol l⁻¹.

⁽²⁾Normal range, 2–6 g day⁻¹.

⁽³⁾Normal range, 0.8–2 g day⁻¹.

^*n = 8.

^**Lactulose/Mannitol urinary ratio.

Population pharmacokinetics

Ganciclovir was well tolerated in all subjects, and no adverse events were reported. A total of 235 out of a possible 252 concentration-time values were included in the analysis. Seventeen of these values (12 in group A, 3 in group B and 2 in group C) were indeed discarded as they were below the lower limit of quantification of the assay. All were in the early (n = 10), or later (n = 7) phase of the concentration-time profile. Serum ganciclovir concentrations ranged from 19 to 1162 ng ml⁻¹. Only two patients in the group C had ganciclovir concentrations outside the linear range of the assay standard curve using u.v. detection (1161 and 1162 ng ml⁻¹), 2 h after the drug intake. However, since these concentrations were within the linear range in fluorescence detection, they were included in the analysis.

The various compartmental and input models were compared on the basis of chi-squared tests. A two-compartment model with first-order input adequately described the data (objective function decrease of 86 and 162 relative to those produced by the one-compartment model and the one-compartment model with saturable absorption, respectively). Inter-patient and residual variability were best described by additive error models. Residual variability included intrapatient error and error related to the assay, sampling time, and model misspecification.

In the preliminary screening phase, covariates that individually reduced the objective function by more than 7 units were clinical status (group), CD4 cell count, haemoglobin, orosomucoid, haptoglobin, and albumin serum concentrations. CL/F and V₁/F were mainly influenced by clinical status, CD4 cell count and haemoglobin serum concentration. Inter-compartmental clearance (Q) and V₂/F were both influenced by orosomucoid serum concentration, while ka was influenced by haptoglobin serum level. None of the absorption or permeability tests improved the modelling process. Nonlinear (power) models of the quantitative covariates offered no advantage over the linear models, except for log(CD4).

In the forward multivariate model building, clinical status, CD4 cell count, orosomucoid and haptoglobin could be cumulatively included with reduction in objective function by more than 7 units. In the backward elimination phase, only clinical status in CL/F and V₁/F exceeded the objective function cut-off value of 10.82 when it was omitted individually from the model. Finally, since CL/F and V₁/F estimates were quite similar for group A and B patients, common estimates for these groups did not produce a significant increase in the objective function. Figure 1 represents the mean oral ganciclovir concentration vs time profiles. Since patients in groups A and B displayed nearly super-imposable concentration-time profiles, only one common curve is shown. Individual data points are shown at each sampling time to demonstrate the large variability. Table 3 summarizes the population estimate, residual and interindividual variability for each pharmacokinetic parameter obtained from the final model. The 95% confidence intervals for the differences between groups A plus B combined and C for CL/F and V₁/F were [ + 70, + 206] l h⁻¹ and [ + 314, + 1022] l, respectively, showing that the decrease in these parameter estimates for group C patients was statistically significant. The interindividual variability in CL/F was reduced from 96% in the basic model (with no covariate) to 47% in the final model (Table 3). Since some early time samples were below the lower limit of quantification of the assay, it was not surprising that interpatient variability in ka remained high (74%, Table 3) in the final model. The high interpatient variability in Q (166%, Table 3) is likely explained by a suboptimal sample collection time of the rapid distribution phase, and some late sampling times that were below the lower limit of quantification of the assay, especially in group A (4 patients) and B (3 patients). Final bias and precision were −2 ng ml⁻¹ (confidence interval, −24–20 ng ml⁻¹) and 173 ± 250 ng ml⁻¹, respectively.

Figure 1.

Fitted concentration – time profiles for oral ganciclovir in HIV-infected patients with (top curve, group C) and without (bottom curve, groups A and B) diarrhoea and weight loss following a single oral dose (1000 mg). Triangles, crosses, and squares are observed values and represent group C, group A and group B patients, respectively. Crosses and squares are shifted slightly for visualization purposes, Timings correspond to the triangle times.

Table 3.

Summary of parameter estimates for the final model including patient covariate effects.

Parameter*	Mean value	Inter-individual variability s.d.**	CV**	Difference¶ mean, [95% CI]
CL/F (groups A, B) (l h−1)	256
		98	47	138, [+ 70, + 206]
CL/F (group C) (l h−1)	118
V1/F (groups A, B) (l)	1320
		336	31	668, [+ 314, + 1022]
V1/F (group C) (l)	652
Q (l h−1)	200	314	166	–
V2/F (l)	1127	39	32	–
ka (h−1)	0.46	0.28	74	–
Residual variability (ng ml−1)	82	34	–

^*CL/F, oral clearance; V₁/F, central volume of distribution; Q, intercompartmental clearance; V₂/F, peripheral volume of distribution; ka, absorption rate constant.

^**Represent values for the whole population.

^¶Mean difference (95% confidence interval (C.I.)) between group A/B and C for CL/F and V₁/F.

Discussion

The population approach used here had particular relevance since a limited number (six) of blood samples was available from each patient. Moreover, due to the large interindividual variability in drug disposition, early or late-time samples sometimes fell below assay limits, and thus only 3–4 concentration-time data were available for some patients. Accordingly, complete analysis of the data from such patients using a conventional approach would be difficult. Because the population approach utilizes all of the data simultaneously, all patients can be described using the same model, as missing information in some patients is ‘borrowed’ from other patients. In addition, sparse and incomplete data sets contribute to the estimation of pharmacokinetic and variability parameters [21].

A two-compartment open model with first order input best described oral ganciclovir pharmacokinetics, as observed previously [2]. At the basic step, large interindividual variability was observed in apparent oral clearance (CL/F, 96%) and central volume of distribution (V₁/F, 93%), suggesting that accounting for individual characteristics could reduce the variability and refine the predictive model. The relationship that was first identified between clinical characteristic (group) and the various pharmacokinetic parameters clearly indicated that ganciclovir disposition (i.e. CL/F) was increased more than two-fold in group C patients compared to groups A and B (Table 3). The inclusion of other covariates related to disease severity (e.g. CD4 count, haemoglobin, albumin, haptoglobin, and orosomucoid serum concentrations) in the pharmacokinetic parameters of interest (CL/F, V₁/F) also significantly improved the fit. Covariate submodelling with these continuous variables provided final objective function values comparable with those obtained with the modelling exercise using group status. However, this modelling was not retained for the following reasons:

1. The group covariate that induced the greatest decrease in the objective function value should be removed from the continuous covariate submodel,

2. There was no significant improvement of the residual plots,

3. Bias, precision, and both interindividual and residual variability were not substantially improved.

The final ganciclovir pharmacokinetic model according to group classification is based upon both statistical criteria and a principle of parsimony; that is, ganciclovir plasma clearance can be readily predicted from a very simple, reliable, and routinely performed clinical evaluation. Our findings suggest that oral clearance is likely to be reduced at least 50% in AIDS patients with severe disease and diarrhoea. This decreased clearance is clearly illustrated in Figure 1, in which greater plasma ganciclovir concentrations and exposure (area under the curve) were observed in these patients. This was not due to impaired ganciclovir elimination, as the estimated creatinine clearance was deemed normal in all patients. Indeed, the Cockroft and Gault formula [16] has been shown to give the best estimate of true creatinine clearance in HIV-infected patients [22]. Moreover, since all patients had estimated values over 95 ml min⁻¹ 1.73 m⁻² (Table 1), a putative overestimation of their true clearance would be minimal in this range of values [22]. Haemoconcentration and/or dehydration could not explain our results since albumin and haemoglobin serum concentrations were decreased in the group C patients secondary to an active inflammatory process, which was not as pronounced in the group B patients (despite a similar rate of active opportunistic infections). Blood haematocrit did not significantly differ in our three groups (data not shown). Although not retained in our final model, intestinal permeability was also higher in the group C patients. Collectively, a decreased renal elimination could not explain the reduced apparent oral clearance, but may rather be due to increased intestinal absorption, although we did not strictly assess ganciclovir bioavailability nor renal elimination (by means of intravenous ganciclovir and a 24 h urine collection, respectively).

Erratic absorption of zidovudine, another nucleoside analogue, was previously correlated with a low CD4 cell count and mild diarrhoea [23]. Whereas absorptive capacity and intestinal integrity are relatively preserved in HIV-well patients, all subgroups of AIDS-patients have a high prevalence of malabsorption, wasting syndrome, hypochlorhydria and markedly increased intestinal permeability that may be related to primary enterocyte injury [5–7,24, 25]. However some AIDS patients with severe diarrhoea have normal absorptive capacity [5]. A major aim of population pharmacokinetics is to determine which measurable pathophysiologic factor(s) cause a change in the dose-concentration relationship, and to estimate the degree to which they do so, such that an appropriate dose adjustment can be made. Therefore, if response to therapy is related to individual exposure to oral ganciclovir (i.e. oral clearance and serum concentrations), then AIDS patients with diarrhoea and severe illness should require a lower dose than that recommended on the basis of studies with HIV-well or AIDS patients with no gastrointestinal symptoms. Since apparent oral clearance in group C patients was more than 2-fold lower than that in patients without diarrhoea and weight loss, we hypothesize that the oral ganciclovir dosing regimen should be halved in AIDS patients with diarrhoea compared with HIV- or AIDS-well patients without diarrhoea and weight loss. However such dosing recommendation should be confirmed in a larger population, using either a conventional or population pharmacokinetic approach, while considering the concentration-effect relationship of ganciclovir.

Oral ganciclovir bioavailability was not altered in bone marrow transplant patients with gastrointestinal graft vs host disease [13]. Diarrhoea in these patients may have been related to specific damages in the gastric and small bowel mucosa that are distinct from those observed during HIV infection, especially during specific HIV-enteropathy that specifically involves the small intestine [3, 5, 26–28]. Therefore, these previous results, in addition to the present findings, may involve at least two different mechanisms. Ganciclovir absorption has been suggested to occur via saturable small intestinal transport [1, 29], although our data could not be ascribed to such a model. The hypothesis of a putative specific transporter for nucleoside analogues [30] is strengthened by recent findings that oral ganciclovir significantly increased maximum concentration and area under the concentration vs time curve of zidovudine and didanosine without affecting renal clearance and metabolism, reflecting a probable increased absorption of these two drugs when taken simultaneously with ganciclovir [2]. Conversely, ganciclovir intestinal absorption was decreased during sequential administration with didanosine [2].

Chronic diarrhoea in HIV infection may be associated with several histological and ultra-structural changes in the duodenal and jejunal mucosa, including villous atrophy, increased villous height/crypt depth ratio, qualitative changes in the brush border enzymes, and increased smooth endoplasmic reticulum [3, 6, 26, 27, 31]. Therefore, a functional alteration of the intestinal brush border may explain an increased intestinal absorption of ganciclovir in AIDS patients with diarrhoea and severe disease. An increased intestinal permeability, as observed in patients in group C in the current study (Table 2) using the L/M ratio may also explain an increased ganciclovir absorption via the paracellular route. However, the L/M ratio did not predict ganciclovir pharmacokinetics in the present study. The pathogenic mechanism of this clinical observation requires further investigation.

In conclusion, we found clinical status, rather than routine intestinal absorption and permeability tests, to predict oral ganciclovir pharmacokinetics in our study subjects. The present results may affect the risk/benefit ratio of oral ganciclovir in AIDS patients. A significant decrease in ganciclovir oral clearance (increased drug exposure) in severely ill patients with diarrhoea might be associated with an increase in efficacy, but adverse events may also occur more frequently. Recommendations for the use of this drug in AIDS patients should take into account the severity of the disease as well as the presence of diarrhoea and weight loss. We believe other drugs that are frequently used to treat HIV infection, especially those for highly active antiretroviral therapy, should also be evaluated to better understand the influence of diarrhoea and disease status on their intestinal absorption and oral bioavailability. The population approach used here allowed the analysis of all the available information in daily life conditions. Understanding the effect of the intestinal barrier function and identifying predictive tests should help physicians to improve therapeutic management and quality of life for HIV-infected patients who frequently take multiple oral medications.