Population pharmacokinetics and pharmacodynamics of escitalopram in overdose and the effect of activated charcoal

Abstract

AIMS

To describe the pharmacokinetics and pharmacodynamics (PKPD) of escitalopram in overdose and its effect on QT prolongation, including the effectiveness of single dose activated charcoal (SDAC).

METHODS

The data set included 78 escitalopram overdose events (median dose, 140 mg [10–560 mg]). SDAC was administered 1.0 to 2.6 h after 12 overdoses (15%). A fully Bayesian analysis was undertaken in WinBUGS 1.4.3, first for a population pharmacokinetic (PK) analysis followed by a PKPD analysis. The developed PKPD model was used to predict the probability of having an abnormal QT as a surrogate for torsade de pointes.

RESULTS

A one compartment model with first order input and first-order elimination described the PK data, including uncertainty in dose and a baseline concentration for patients taking escitalopram therapeutically. SDAC reduced the fraction absorbed by 31% and reduced the individual predicted area under the curve adjusted for dose (AUC_i/dose). The absolute QT interval was related to the observed heart rate with an estimated individual heart rate correction factor (α = 0.35). The heart rate corrected QT interval (QT_c) was linearly dependent on predicted escitalopram concentration [slope = 87 ms/(mg l^–1)], using a hypothetical effect-compartment (half-life of effect-delay, 1.0h). Administration of SDAC significantly reduced QT prolongation and was shown to reduce the risk of having an abnormal QT by approximately 35% for escitalopram doses above 200 mg.

CONCLUSIONS

There was a dose-related lengthening of the QT interval that lagged the increase in drug concentration. SDAC resulted in a moderate reduction in fraction of escitalopram absorbed and reduced the risk of the QT interval being abnormal.

WHAT IS ALREADY KNOWN ABOUT THIS SUBJECT

• Escitalopram is a selective serotonin re-uptake inhibitor and the S-enantiomer of racemic citalopram. Previous studies have shown that the most clinically important effect of escitalopram overdose is QT prolongation with the associated risk of torsade de pointes. It remains unclear at what dose the risk of QT prolongation is important and whether decontamination will reduce this risk.

WHAT THIS STUDY ADDS

• Escitalopram overdose causes a dose-related increase in the QT interval that lags the increase in drug concentration. In escitalopram overdose, single dose activated charcoal reduces the fraction absorbed and reduces the risk of QT interval prolongation.

Introduction

Escitalopram is one of the newer selective serotonin re-uptake inhibitors (SSRIs) and the S-enantiomer of racemic citalopram. Escitalopram has an increasing share of the antidepressant market and it is important to determine if it has similar toxicity to citalopram in overdose. Reports of escitalopram in overdose suggest that one of the most clinically important effects is QT prolongation with the associated risk of torsade de pointes (TdP) [1]. However, it remains unclear at what dose the risk of QT prolongation is clinically significant.

Defining the pharmacokinetics (PK) and pharmacokinetics-pharmacodynamics (PKPD) of drugs in overdose is important to understand the dose–concentration effect relationship and whether interventions such as decontamination may influence this relationship. This has been done for a limited number of drugs [2–6] and it demonstrates that decontamination has variable effects on different drugs in overdose and that this is not always translated into a clinically important effect. Central to the management of drug overdose is a risk assessment based on the drug taken, the dose and the potential toxicity. Therefore PKPD models that establish the relationship between the amount taken in overdose and the toxicity can be used to develop clinical guidelines for the risk assessment of drugs in overdose [7].

We have previously developed clinical guidelines for citalopram based on the PKPD of citalopram in overdose. These guidelines identify at what dose there is a risk of QT prolongation and how long patients should be monitored as well as the likely benefit of decontamination [8]. Although escitalopram may be similar to citalopram in overdose, it has a higher clearance than its R-enantiomer at therapeutic doses [9]. In addition the effect of the escitalopram on the QT interval may differ from the R-enantiomer making the risk of TdP different from that with citalopram.

This study aims to investigate the PKPD of escitalopram in overdose and the relationship between dose and abnormal QT using the QT nomogram [10] as an indicator of TdP risk. In addition, the study will determine whether SDAC has a beneficial effect on the absorption and clearance of escitalopram and whether this reduces the risk of an abnormal QT. This will provide a basis to develop treatment guidelines for escitalopram overdose.

Methods

Patient data

All presentations of escitalopram overdoses to a tertiary toxicology unit were recruited to the study. The toxicology unit accepts primary referrals for a population of about 300 000 people and provides a 24 h on-call toxicology service. The study was approved by the local human research ethics committee. Written informed consent for participation in the study, collection of blood samples and recording of multiple electrocardiograms (ECG) was obtained for all patients. Patient treatment, including the use of decontamination, was according to the admitting clinical toxicologist. Single dose activated charcoal (SDAC; 50 g with no additive) was the only form of decontamination administered in this study.

During the period from October 2003 to September 2008 clinical data were available for 78 presentations of escitalopram overdose by 68 patients. Plasma concentrations were available in 34 overdose events from 29 patients and a total of 104 samples were collected with a median of three samples per occasion (range 1 to 9) taken 1.5 to 47.5 h after the time of ingestion. A total of 232 ECGs were available for 77 presentations from 68 patients and 182 were recorded after escitalopram overdoses (median 2 per patient, range 1 to 8) between 1 and 54 h after the reported time of overdose. The remaining 50 ECGs were from 18 of the 77 patients from either another overdose occasion where they ingested a non-cardiotoxic drug or from an admission when the patient did not take an overdose.

Patient demographics, therapeutic use and dose of escitalopram, other therapeutic drugs used by the patient known to induce CYP2C19, CYP2D6 and CYP3A4, reported dose, veracity of the dosing history, co-ingested drugs and administration of SDAC are provided in Table 1 for the 34 overdoses included in the PK analysis. SDAC was administered in six overdoses after a reported median dose of 350 mg (range 70 to 450 mg) and a median time of 1.9 h after ingestion (range 1.8 to 2.5 h). No blood sample was taken prior to SDAC administration.

Table 1.

Characteristics for the 34 presentations of escitalopram overdose in 29 patients where plasma concentration–time data were available

	Median (range)	Number of patients	Number of events
Age (years)	27 (16–51)
Males/Females		5/24	5/29
Reported overdose (mg)	135 (10–450)
Escitalopram prescribed	10 (5–40)*	22 (76%)†	26 (76%)
Prescribed dose (mg)
Veracity grade
0			3
1			12
2			15
3			4
Co-ingested drugs			29†
2C19, 2D6 or 3A4 substrate			11
2C19, 2D6 or 3A4 inhibitor			4‡
Therapeutic CYP3A4 inducer		1§	1§
Activated charcoal		6 (21%)	6 (18%)
Escitalopram concentration	12–520 µg l–1

^*On at least one occasion

^†information for 26 overdose events where escitalopram was prescribed

^‡fluoxetine (2), esomeprazole or clarythromycin

^§phenytoin.

Table 2 includes data for the 78 presentations used for the PKPD analysis, although an ECG was not available in one patient with drug concentration–time data. Data include demographics, reported dose, co-ingested drugs known to affect the QT interval and administration of SDAC. SDAC was administered on 12 of the 77 occasions after a reported median dose of 280 mg (range 70–560 mg), compared with 130 mg (range 10–440 mg) for cases where SDAC was not given. The median time between the overdose event and the administration of SDAC was 1.7 h (range 1–2.6 h). The majority of ECGs were taken after SDAC administration. Fifty-eight of the patients used escitalopram therapeutically.

Table 2.

Characteristics for the 77 presentations of escitalopram overdose in 68 patients where ECG data was available

	Median (range)	Number of patients	Number of events
Age (years)	31 (15–59)
Males/Females		16/52	16/62
Reported overdose (mg)	140 (10–560)
Co-ingested drugs with effect on QT		3 (4%)	3 (4%)
Activated charcoal administered		12 (18%)	12 (15%)
Absolute QT interval (ms)	400 (280–650)
RR interval (ms)	770 (350–1580)

The ingested dose of escitalopram was estimated from (i) patient history or history from significant others, (ii) tablet counts and (iii) any other available information (e.g. ambulance record). Uncertainty in dose was included by assessing the ‘veracity’ of the history as previously described [2]. Veracity was graded on a 5-point categorical scale: 0 excellent history (recall exact numbers of tablets with supporting information, e.g. empty packets), 1 good history (good estimate of tablets taken with some supporting information), 2 less reliable history (less reliable estimate of dose due to sedative co-ingestants but able to give a history on admission), 3 poor history (poor or no recall of overdose due to sedative co-ingestants) or 4 very poor history (completely unreliable history with no supporting information). Only the categories 0 to 3 were used in this study (Table 1). For overdoses with a veracity of 0, exact dosing was assumed. In overdose events with a veracity grade of 1, 2 or 3, increasing additive and proportional error components were allowed to modify the uncertainty of the dose. The additive component was related to tablet size (e.g. multiples of 10 mg).

Time of overdose was taken from the patient history and a possible range in this time was provided by the earliest possible ingestion time (such as contacts with family members) and the latest possible ingestion time (such as the ambulance call time). This was used to account for uncertainty in timing, as previously described [2]. In this study the range varied from 5 min to 4 h.

Drug analysis

Escitalopram was quantified by high performance liquid chromatography. Standard curves (range 45–335 µg l^–1) had a correlation coefficient of greater than 0.999. Dose-normalized plasma concentration–time profiles are presented in Figure 1. Measured concentrations ranged from 12 to 520 µg l^–1. No concentration was below the limit of quantitation of 1 µg l^–1[11].

Figure 1.

Observed dose-normalized concentrations (to the average therapeutic dose of 20 mg) vs. reported time of administration. Samples from the same dose event are connected. The symbols denote if the patient was not administered SDAC (•) or was administered SDAC (×)

Electrocardiograph measurements

All QT intervals were measured manually in the 12-lead ECGs using a previously described approach [12]. The QT interval was measured from the beginning of the Q wave up to the point where the T wave returned to baseline. QT intervals were measured in six leads (three chest and three limb leads) and the median interval calculated. The RR interval was measured from the R-wave in one complex to the next complex for at least six cycles in lead II. The median RR interval was used for the analysis.

Pharmacokinetic analysis

A PK analysis was undertaken first to establish the best PK model for the data and then the same structure of the PK model was retained for the PKPD model. A fully Bayesian population PK analysis was undertaken as previously described for the 34 occasions with plasma concentration–time data [2, 4, 5]. WinBUGS 1.4.3 was used for the pharmacokinetic data modelling using a three-stage hierarchical model. First order absorption without a lag time and first order elimination was assumed, based on visual inspection of the concentration–time data (Figure 1) and citalopram overdose pharmacokinetics [2]. Additive, proportional and combined residual error models were investigated.

Prior means for clearance (CL), volume of distribution (V_d) and absorption rate constant (K_a) were elicited from a previous pharmacokinetic study of healthy volunteers [9], using the back analysis technique [13]. Weak priors were used for between-subject variability (BSV) because only one previous PK study was available. For all patients taking escitalopram therapeutically a baseline concentration was estimated using both the extrapolation subtraction method (ESM) and concentration–time method (CTM) as previously described [14]. Uncertainty in dose and dose time was investigated by incorporating veracity and modtime_i,k in the models [2]. The variable modtime_i,k represents the difference in time estimated from the model of the executed dose time from the nominal dose time for the kth occasion for the ith individual.

Covariates were assessed after evaluating baseline concentration and dosing uncertainty, including age, sex, co-ingested drugs, therapeutic drugs affecting CYP2C19, CYP2D6 or CYP3A4 and the influence of SDAC. Patient weight is practically difficult to obtain in overdose patients and could not be included. The influence of phenotypic covariates was assessed initially by visual inspection and if there appeared to be an influence the covariate was included in the model. The effect of SDAC was evaluated by modifying the individual clearance by a factor called f_CL-charc and the individual fraction absorbed by a factor of f_F-charc. If a patient was not given SDAC, the factor was fixed to 1. Differences were considered clinically significant if they were greater than 20%.

Pharmacokinetic–pharmacodynamic analysis

All PK and PD parameters were simultaneously estimated and all 78 occasions were included. The final PKPD model was based on a previous population PKPD analysis for citalopram which has been described in detail previously [3]. Heart rate dependence was assumed to have a power relationship shown in equation 1 and each individual was allowed to have an individually heart rate corrected QT interval (QT_c,i), with α_i indexed to the ith individual.

(1)

Here we denote the jth measurement on the ith patient for RR_ij and QT_c,_ij. The predicted escitalopram concentration C or concentration in the effect compartment (C_e) was used to drive the change QT_c,_ij from the baseline QT_c,j0 using equation 2. Although we show the model for the effect compartment concentration of escitalopram, this is also part of model development. Previous modelling [3] with citalopram found a linear function provided the best description of QT_c,_ij, as per.

(2)

The error term ε_ij was assumed to be normally distributed. The between subject differences in the parameters were assumed to be log-normally distributed.

The prior distribution for the PK parameters was the same as the PK model. The elicitation of priors for the PD analysis of the QT interval has been described in detail previously and the prior distributions of the PKPD analysis for citalopram were assumed to be applicable for escitalopram [3]. The QT interval was derived from a previous study of patients taking non-cardiotoxic drug overdoses [10]. All parameters were assumed to be log normally distributed, QT_c,j0 with a mean of 423 variance of 0.0004, slope_i with a mean of zero (no effect) and variance of 3.57, α_i with a mean of 0.333 and variance of 0.09 and finally the rate constant of elimination from the effect compartment (k_eo) was assumed to have a mean of 1 and variance of 0.676.

The only covariates available for the PD parameters were sex, age and co-ingested drugs. There were only three occasions where drugs affecting the QT interval were co-ingested so these were not considered. The influence of sex was evaluated by modelling different values for QT_c,j0 for each sex and age was allowed to influence the QT_c,i0, as per the citalopram PKPD [3].

Model development and estimation of parameters

WinBUGS was first used to model the PK data for patients with concentration–time data. Then the PK and PD data for the full dataset were modelled simultaneously, WinBUGS imputed the missing concentration–time data for those patients who had only PD measurements. Posterior means and credible intervals (CrI) were generated using Markov chain Monte Carlo (MCMC) simulation methods. The CrI of the posterior distribution is the interval covering 95% of the MCMC samples after the ‘burn-in’ period. Initially modelling was performed with two chains and 10 000 iterations post burn-in. The burn-in was assumed to be complete by 4000 iterations. The final model was run for 200 000 iterations, keeping every 10^th sample.

Model convergence and selection

Gelman–Rubin diagnostics, as available in WinBUGS, were used to check if convergence was achieved. For both the PK and the PKPD analysis model selection decisions were made based on a reduction in the deviance information criterion (DIC), the posterior mean of the BSV in the parameters, the residual error magnitudes and the magnitude of the estimated effect for covariate inclusions, such as SDAC. Mixture models were used for crucial modelling steps or when the previous criteria were inconsistent as previously described [2, 4, 5]. In this approach two competing models are fitted to the data simultaneously and the preference for a particular model is given by the posterior distribution of a mixing parameter that represents the relative contributions of each model [15].

A sensitivity analysis was conducted to assess the influence of the priors on the PK model on the posterior distributions by decreasing the precision of the priors by a factor 10 and decreasing the number of degrees of freedom of the Wishart distribution for BSV parameters from 11 to 9. Similarly in the PKPD analysis all prior precisions were decreased by doubling the standard deviation (so the precision was reduced by 4). The number of degrees of freedom of the Wishart distribution for the omega matrix was reduced.

Prediction of risk for having abnormal QT/RR combinations

The final model was coded in MATLAB (Version 2010b, The MathWorks, Natick, MA) using the means of the posterior distributions of the parameters from the final model as point estimates of the parameters. The QT interval was assumed to be abnormal and associated with an increased risk of developing TdP if the QT,RR pair was above the upper bound of the ‘cloud’ defined by Fossa [16], which is equivalent to the QT,HR pair being above the QT nomogram [10]. For the 50^th percentile of the observed RR intervals in our study of 762 ms the ‘threshold’ QT interval was determined to be 447 ms. Using the final model parameters 5000 QT, RR intervals were simulated for a specified covariate set (30-year-old females), who were taking escitalopram therapeutically and had taken the dose amounts of 5, 10, 20, 30 and 40 times the defined daily dose of 10 mg. From these simulated values the proportion of patients with an abnormal QT was calculated. The simulations were done with and without SDAC. In this instance we simulated from point estimates of the posterior distributions to reduce computational burden. Evidence from the work of Yano et al. [17] did not show a difference in the inference based on simulations from the full posterior compared with from point estimates when forming predictive distributions.

Using the results with and without SDAC the instantaneous hazard was estimated as the fraction of patients at time t who had a predicted QT interval above 447 ms. The cumulative hazard was then calculated as the integral of the instantaneous from 0 to 96 h. The relative decrease in cumulative hazard associated with administration of SDAC was then computed.

Results

Pharmacokinetic analysis

The best structural PK model was a one compartment first order input model without lag time. A combined error model was selected based on a reduction in DIC. The inclusion of baseline concentration for patients taking escitalopram therapeutically using the CT method improved the model. Inclusion of veracity improved the model but the inclusion of modtime in addition to a baseline concentration did not improve the model.

Administration of SDAC decreased the fraction absorbed by 31% and the posterior probability of an effect greater than 20% was 0.76. There was no effect of SDAC on clearance. There was no clear relationship between age, sex and co-ingestion of drugs that inhibited CYP2C19, CYP2D6 or CYP3A4 and any parameters, so these covariates were not included. Therefore, the final PK model had a combined residual error model and included a baseline concentration, uncertainly in dose based on veracity and SDAC as a covariate affecting fraction absorbed (f_F-charc).

There was no correlation between dose and CL or V_d. Decreasing the precision of the priors did not influence the estimates of mean CL and V_d (–5.8% and +1.9%, respectively). The estimate for K_a increased from 5.97 to 52 450, indicating that the data contained very little information on this parameter. The median half-life over individual occasions was 34 h (interquartile range [IQR] 28–38 h). The individual predicted area under the curve adjusted for dose (AUC_i,k/dose_i,k) was reduced in patients given SDAC with a median of 0.018 (IQR 0.16–0.036) compared with 0.040 (IQR 0.027–0.042) in patients not given SDAC (Figure 2). The individual predicted CL was different for patients given SDAC with a median of 40.6 l h^–1 (IQR 26.4–47 l h^–1) compared with 27.7 l h^–1 (IQR 24.6–40.4 l h^–1) in patients not given SDAC.

Figure 2.

Comparison of the individual predicted dose normalized area under the curve (AUC_k/dose) estimations for patients not receiving SDAC (○) vs. patients receiving SDAC (•); median indicated by horizontal line

Pharmacokinetic–pharmacodynamic analysis

The final PKPD model was the same as for citalopram and included an individual heart rate correction factor, α_i, a linear concentration–effect relationship (Equation 2), an effect compartment model and an age and sex-specific value of QT_c,j0. An additive error model was used for the QT interval data. The model described the data well and the final parameters are included in Table 3. There appeared to be good convergence of the MCMC process as seen visually and with Gelman-Rubin diagnostic plots (Figures S1 and S2). There was not a large effect of age or sex on QT_c,j0. The observed values for CL/F, V_d/F, K_a and baseline concentration were comparable with the PK analysis (Tables 3). The data contained information on all parameters based on the posterior means being shifted from the prior values (data not shown).

Table 3.

Means (95% CrI) of the estimated parameter distributions for the final pharmacokinetic–pharmacodynamic model

	Population mean (95% CrI*)	BSV (CV%)†
CL/F (l h–1)	33.5 (20.1, 50.3)	74.4
Vd/F (l)	1285 (950, 1715)	63.2
Ka (h–1)	8.0 (1.1, 36)	111.4
Baseline concentration (mg l–1)	0.014 (0.003, 0.045)	46.3
QTc,j0 men (ms)‡	424 (411, 436)	6.2§
QTc,j0 women (ms)‡	423 (414, 432)	6.2§
α	0.353 (0.306, 0.402)	29.2
teq (h)	1.01 (0.05, 5.6)	–
Slope [ms/(mg l–1)]	87 (27, 164)	87.4
Proportional residual error (mg l–1) on Cp	0.237 (0.191, 0.291)	–
Additive residual error (mg l–1) on Cp	0.002 (0, 0.006)	–
Proportional residual error (ms) on QT	16.6 (14.8, 18.6)	–

^*CrI, Credible interval of posterior distribution, i.e. the interval covering 95% of the MCMC samples.

^†Approximated to the square root of the estimated between subject variance (BSV).

^‡For a 30-year-old patient.

^§Forced to be the same value for men and women. Cp, plasma concentration.

For an individual with typical PK and PD parameters the maximum effect on QT was predicted to occur 8.5 h after the overdose compared with the peak concentration occurring 0.7 h after overdose (Figure 3A). For every 1 mg l^–1 increase in the drug concentration, the QT interval was estimated to be prolonged by 87 ms (i.e. slope, Table 3). The model also predicted a significant effect of SDAC on the QT interval which is shown in Figure 3B for a typical individual.

Figure 3.

Simulated plasma concentrations (—) and QT intervals (- - -) vs. time in a patient with typical PK and PD parameters for an escitalopram overdose without SDAC (A). Simulated QT intervals vs. time without (—) and with (- - -) SDAC (B). The dose was 300 mg and the RR interval 770 ms (HR = 78 bpm) for both panels

Prediction of abnormal QT and the effect of SDAC

Figure 4 shows the fraction of patients with an abnormal QT value after escitalopram overdoses for the median RR value of 762 ms (79 beats min^–1) with and without SDAC. The administration of SDAC reduced the fraction of patients having an abnormal QT (Figure 4), and the relative decrease in the cumulative hazard for an abnormal QT was approximately 35% for overdoses of over 200 mg after administration of SDAC (Figure 5).

Figure 4.

Plots of the fraction of patients with an abnormal QT (QT >447 ms, RR = 762 ms) vs. time for doses ranging from 50 mg to 400 mg. A) shows the fraction of patients without SDAC and B) the fraction of patients with SDAC. Five thousand patients were simulated and assumed to be 30-year-old women taking escitalopram therapeutically

Figure 5.

Plot of the relative decrease in cumulative hazard for having an abnormal QT interval (≥447 ms, RR = 762 ms) when SDAC is administered for doses ranging from 50 mg to 400 mg

Discussion

We developed a population PK and PKPD model of escitalopram in overdose to define the dose-exposure-effect relationship. Similar to previous PKPD studies of drugs in overdose, including information about uncertainty in the dose and baseline concentrations improved the model and reported dose was a reliable measure of drug exposure [7, 18]. SDAC reduced the fraction absorbed which was seen in the reduction in the individual patient AUCs given SDAC (Figure 2). This translated as a reduced proportion of patients given SDAC developing an abnormal QT and with a relative risk reduction of 35% in patients taking 200 mg or more.

We were unable to demonstrate an effect of SDAC on clearance in the pharmacokinetic model of escitalopram which is surprising considering that SDAC had a clinically important effect on clearance in citalopram overdose. Although escitalopram is more rapidly metabolized than the R-enantiomer, this is insufficient to explain the loss of charcoal effect on clearance, because the half-life of escitalopram was still 34 h. A more likely reason is that the PK study of escitalopram overdoses was smaller, with only six of 34 episodes being given SDAC. The model was less able to provide tight estimates of the clearance and therefore detect a difference between clearance in those with and without SDAC. This is demonstrated in Figure 6 which shows the clearance for patients with and without charcoal, comparing predicted individual clearances for citalopram vs. escitalopram.

Figure 6.

Predicted individual clearance values for citalopram (grey lines), with and without charcoal, compared with values for escitalopram (dark points), with and without charcoal. Clearance values for citalopram were predicted using the previously published PKPD model of citalopram [2]. Escitalopram (♦); Citalopram (◊)

The study adds to the increasing evidence from pharmacokinetic studies that reported dose is a useful measure of drug exposure in overdose patients [2, 4, 5]. Although there is uncertainty in the reported dose the inclusion of veracity on the dose improved the model. This means that reported dose is a useful predictor of an abnormal QT with an increased proportion of patients developing an abnormal QT with increasing dose. In addition, for doses of 200 mg or more, SDAC had a significant impact with a relative risk reduction for an abnormal QT occurring of 0.35. Therefore a careful history of the dose ingested by the patient is crucial to both risk assessment and in some cases, such as escitalopram, the use of SDAC.

Based on our previous study of citalopram overdose, where doses associated with greater than 10% of patients having an abnormal QT were taken as clinically important [8], it could be suggested that SDAC be used for ingested escitalopram overdoses above 300 mg from Figure 4 (top panel). Similarly, patients ingesting over 300 mg not given SDAC or patients ingesting over 400 mg and given SDAC should be monitored for 12 h or until the QT,HR pair is under the nomogram line. These dose cut-offs for SDAC and monitoring are approximately equivalent to previous dose cut-offs developed for citalopram overdose [8].

The inability to include phenotypic covariates in the PK and PKPD model was a limitation of the study and was possibly due to the size of the dataset. Similar to previous population PK studies of overdose patients it was practically impossible to record weight in the majority of patients so weight could not be included in the covariate analysis. Although it is likely that the inclusion of weight or lean body mass as a covariate would have reduced the BSV on CL and V_d, this may have not improved the usefulness of the model for future predictions because weight is rarely measured in overdose patients.

Another problem was that although the QT interval is normally higher in females and increases with age, this was not found in our PKPD model. The mean QT_c,i(0) was slightly higher for men than for women and there was 50% chance of at least a 5 ms difference between the youngest and the oldest individuals for both men and women. This again may have been due to the small numbers in the study but this relationship of the QT interval has been seen in other studies of drug induced QT prolongation [19].

Fully Bayesian analyses like this are sometimes criticized for including subjective information in the prior distributions. In this study we used a previously developed objective method for eliciting informative priors from the literature [2, 13]. Unfortunately, there were no previous population analyses of escitalopram available. The only information that we could use to elicit prior distributions was a single PK study of healthy volunteers taking therapeutic doses [20]. We were therefore only able to include informative priors on V_d and CL, and used weakly informative, biologically plausible priors for the BSV on the parameters (as per Friberg et al. [2]). When we relaxed the priors in the sensitivity analysis, the estimated values of CL and V_d were minimally changed, indicating that the data (and not the priors) were the most important source of information for the model. However, there were marked changes in the estimate for K_a when we relaxed the priors. This is probably a result of the limited number of blood samples in the absorption phase. The estimated values of K_a are therefore difficult to interpret in a meaningful way. The estimated values for CL and V_d in overdose patients were similar to the PK study of therapeutic use of escitalopram [20]. It appears that the PK of escitalopram in overdose are similar to therapeutic use.

The PKPD of escitalopram overdose were well described. There was a clear relationship between reported dose and the proportion of patients developing an abnormal QT. SDAC reduced the fraction absorbed which translated into a significant decrease in the risk of an abnormal QT. There appears to be benefit in administering SDAC for ingested doses greater than 300 mg.

Competing Interests

There are no competing interests to declare.

Funding: Geoff Isbister is funded by an NHMRC Clinical Career Development Award ID605817.

Supporting Information

Additional Supporting Information may be found in the online version of this article:

Figure S1

MCMC plots for three parameters (CL, V_d andQT) from the final pharmacokinetic–pharmacodynamic model.

Figure S2

Gelman–Rubin diagnostic plots for three parameters (CL,V_d and QT) from the finalpharmacokinetic–pharmacodynamic model.