OUP user menu

Personalized Oxycodone Dosing: Using Pharmacogenetic Testing and Clinical Pharmacokinetics to Reduce Toxicity Risk and Increase Effectiveness

Oscar A. Linares MD, David Daly MBA, JD, Annemarie Daly Linares MD, JD(c), FACP, Darko Stefanovski PhD, Raymond C. Boston PhD
DOI: http://dx.doi.org/10.1111/pme.12380 791-806 First published online: 1 May 2014

Abstract

Objective To develop a framework for integrating pharmacogenetics with clinical pharmacokinetics for personalized oxycodone dosing based on a patient's CYP2D6 phenotype.

Design Randomized, crossover, double-blind, placebo-controlled. Subjects were genotyped as CYP2D6 ultra-rapid metabolizer, extensive metabolizer, or poor metabolizer phenotypes. Five subjects from each phenotype were randomly selected for inclusion in our study.

Setting Studies were performed in silico.

Subjects The subjects were male, age 26 years, height 181.2 cm, and weight 76.3 kg. They were healthy without comorbidities, and their medical examinations were normal.

Methods The trajectories of phenotype-specific plasma oxycodone concentration-time profiles were analyzed using weighted nonlinear least-squares regression with WinSAAM software. A global two-stage population-based model data analysis procedure was used to analyze the studies. Clinical pharmacokinetics were calculated using the R package cpk, eliminating the need to perform hand-calculations.

Results Our study shows how clinicians can reduce risk and increase effectiveness for oxycodone dosing by 1) determining the patient's likely metabolic response through testing a patient's CYP2D6 phenotype, and 2) calculating clinical pharmacokinetics specific to the patient's CYP2D6 phenotype to design a personalized oxycodone dosing regimen.

Conclusions Personalized oxycodone dosing is a new tool for a clinician treating chronic pain patients requiring oxycodone. By expressing a patient's CYP2D6 phenotype pharmacokinetically, a clinician (at least theoretically) can improve the safety and efficacy of oxycodone and decrease the risk for iatrogenically induced overdose or death. Pharmacokinomics provides a general framework for the integration of pharmacogenetics with clinical pharmacokinetics into clinical practice for gene-based prescribing.

  • Oxycodone
  • CYP2D6
  • Metabolism
  • Pharmacogenetics
  • Pharmacogenomics
  • Clinical Pharmacokinetics
  • Pharmacokinomics
  • Clinical Pharmacology
  • Mathematical Medicine

Introduction

Pain is one of the most significant health care crises in the United States [1]. Nearly half of 308 million Americans see a clinician each year primarily because of pain [2]. Yet the magnitude of the problem is much greater because a substantial number of people with pain do not consult a clinician [3], the result of which is increased mortality [4,5].

The Institute of Medicine reported to Congress that chronic pain affects 100 million American adults. By contrast, more than 16 million Americans have pain that lasts for weeks. Thus, over 116 million Americans suffer from pain, as those figures do not include pain in children or people in long-term care facilities, the military, or prison. This is more than the total affected by heart disease, cancer, and diabetes combined [6]. Furthermore, pain costs the nation $560–635 billion each year in medical treatment, lost productivity, and missed work [6,7].

Little is written about two potential science-based solutions, clinical pharmacokinetics [8] and pharmacogenetics (PGt) [9], that could mitigate opioid overdose deaths. Clinical pharmacokinetics is widely used for designing dosage regimens for potentially lethal medicines in anesthesia [10]. Thus, opioid clinical pharmacokinetics has a potential role in ameliorating or preventing opioid overdoses because it aims to provide safe and effective opioid prescribing. PGt takes into account patient's genetics for individualizing treatment. In addition, pharmacogenetic testing allows clinicians to better predict patient responses to targeted treatment [11]. Consequently, the challenge is how to integrate these technologies to deliver their benefits to patients.

Clinical pharmacokinetics [12–19] is the study of the relationships between drug-dosing regimens and drug concentration-time profiles, i.e., their absorption, distribution, metabolism, and elimination. The four fundamental disposition parameters that govern these relationships are: the first-order elimination rate constant (ke)—fractional elimination of drug from tissues in the body per unit time; clearance (Cl)—volume of fluid completely cleared of drug per unit time; volume of distribution (Vd)—apparent volume into which the drug distributes to produce the measured concentration; and elimination half-life (t1/2)—the time it takes for 50% of the drug to be eliminated.

A drug's Vd can be used to calculate a loading dose to achieve a target concentration quickly, while Cl can be used to calculate the dose rate required to maintain a target therapeutic concentration (TTC). Half-life can be defined as the time required for the concentration of drug in the blood to reach 50% of its concentration, and it takes five half-lives to reach steady state; t1/2can also be used to calculate the optimal dosing interval to produce a target peak-to-trough difference [10,20]. At this level, an individual-based model is a useful framework for understanding drug concentration-time profiles by providing a means for estimating a drug's pharmacokinetic parameters [21].

At the next level, population-based model data analysis assists in individualizing dosing regimens. However, due to its mathematical unfamiliarity and potential for complexity [22], its clinical usefulness is often overlooked or even avoided. A key reason to conduct population-based model studies is to investigate the potential benefits of individualizing dosing based on patient-specific characteristics, e.g., CYP2D6 phenotype. Dose individualization is achieved by understanding the onset, magnitude, and duration of drug effects that result from a given dose and dosing regimen in the target population with the patient-specific characteristic under investigation, and how these effects vary over the target population. Thus, a major aim of population-based model data analysis is to generate overall population predictions based on sampling a limited number of individuals. It is essential because it quantifies individual components of variation, which may be important contributors of pharmacological variability within individuals [21].

We performed this study to develop a framework for the integration of clinical pharmacokinetics with PGt, and we applied it to oxycodone (OC) for personalized OC dosing based on a patients' CYP2D6 phenotype. This approach allows preemptive assessment of phenotype-specific OC therapeutic responses in silico, which forms a basis for a modern “precision medicine” approach to tailoring therapy for gene-based prescribing. Table provides a glossary of symbols.

View this table:
Table 1

Glossary of symbols

graphic
graphic
graphic
graphic
graphic
graphic
graphic
SymbolismDefinitionUnits
OCOxycodoneμg/L
OCMOxymorphoneμg/L
NOCNoroxycodoneμg/L
NOMNoroxymorphoneμg/L
EMExtensive metabolizer phenotype
PMPoor metabolizer phenotype
UMUltra-rapid metabolizer phenotype
MSCMaximum safe concentrationμg/L
MECMinimum effective concentrationμg/L
TTCTarget therapeutic concentrationμg/L
Phenotype-specific OC dose ratemg/h
ClPsPhenotype-specific OC clearance rateL/Kg/h
FBioavailability
Phenotype-specific OC maintenance dosemg
τDosing intervalhour
ARPsPhenotype-specific OC accumulation ratio
Phenotype-specific OC elimination half-lifehour
Phenotype-specific steady-state plasma OC concentrationμg/L
Phenotype-specific maximum (peak) plasma OC concentrationμg/L
DpoOral OC dosemg
VdThe apparent volume into which OC distributes to produce the measured concentrationL
Phenotype-specific first-order elimination rate constant, which quantifies the fractional elimination rate of OC from tissues in the body per unit time/hour
Population phenotype-specific first-order elimination rate constant. i indexes individuals and j phenotypes/hour
τmaxMaximum length of time for a repetitive dosing interval to maintain plasma OC concentrations between MEC and MSChour
Effective DpoAn oral dose that results in a plasma OC level > MECmg
Safe DpoAn oral dose that results in a plasma OC level < MSCmg
CVCoefficient of variation%
PGtPharmacogenetics
PGxPharmacogenomics
  • Unless otherwise indicated in the text.

  • Units of μg/L (instead of ng/mL) are used to keep units consistent mathematically (see text). However, 1 μg = 1,000 ng and 1 L = 1,000 mL; converting mL to L gives 1 μg/L/1 L = 1,000 ng/1,000 mL, so that 1 µg/L = 1 ng/mL.

Materials and Methods

Study Data

Subjects were genotyped as CYP2D6 ultra-rapid metabolizer (UM), extensive metabolizer (EM), or poor metabolizer (PM) phenotypes. Five male subjects from each phenotype were randomly selected for inclusion into our study. Subjects were non-smoking, opioid-intolerant, with no history of illicit drug use, who were not taking any medications, and had no contraindication to opioids, including specifically OC. Subjects had no comorbidities, and medical examination was normal. Medical examination included electrocardiogram, liver, and kidney function tests. The mean age of the subjects was 26 years, mean height 181.2 cm, and mean weight 76.3 kg.

Plasma samples for measurement of OC concentration were taken at 0.5, 1, 1.5, 2, 3, 6, and 24 hours.

For each of the five subjects from each of the three phenotypes, the mean ± standard deviation (SD) 24 hours OC plasma levels were 0.41 ± 0.06 μg/L (EM), 0.22 ± 0.03 μg/L (UM), and 1.39 ± 0.11 μg/L (PM). Although these 24 hours OC sample values were significantly different from each other (analysis of variance [ANOVA] P < 0.05), we excluded them from the population-based model data analysis because they fell below the detection limit of 10 μg/L for a highly sensitive and specific high-performance liquid chromatographic assay [23]. Therefore, for population-based model data analysis, we only used the 0.5, 1, 1.5, 2, 3, and 6 hours plasma OC concentration measurements.

Clinical Pharmacokinetics for Personalized Dosing

How can a clinician prescribe an oral OC dose to safely and effectively treat a patient's pain? In practical terms, a safe dose means a dose that results in a plasma OC level less than the maximum safe concentration (MSC), which aside from polymorphic differences and drug interactions is in part based on opioid tolerance. An effective dose means a dose that achieves a plasma concentration greater than the minimum effective concentration (MEC). Combining the concepts of safe and effective, the therapeutic range is the range of plasma levels between the MEC and MSC.

An average MEC for an average size patient receiving OC for chronic pain at a doses of approximately 20–30 mg/day in divided doses equals 20 μg/L; the MSC equals 50 [24]. Therefore, the therapeutic range for OC is any plasma concentration between 20 and 50 μg/L (±7.5 SD). Opioid-tolerant patients will likely require higher concentrations for analgesia because of physical tolerance.

In order to calculate a safe and effective dose, the clinician may 1) choose a TTC that lies within the therapeutic range, and then 2) work backward to determine the oral dose needed to achieve that TTC at steady-state. In order to maximize the likelihood of achieving the twin goals of prescribing a dose that is both safe and effective, it is reasonable to calculate an initial TTC that lies midway between the MEC and the MSC. The TTC that represents the logarithmic average of the MEC and MSC is expressed by the following equation: Embedded Image 1 where MSC = 50, MEC = 20, TTC = 33, and ln is the natural log. Units of μg/L (instead of ng/mL) are used in this equation to keep units consistent mathematically. However, 1 μg = 1,000 ng and 1 L = 1,000 mL; converting mL to L gives 1 µg/L/1 L = 1,000 ng/1,000 mL, so that 1 µg/L = 1 ng/mL.

The dose rate necessary to achieve the TTC depends upon the patient's CYP2D6 phenotype. The phenotype-specific OC dose rate (Embedded Image) to achieve the TTC is calculated as follows: Embedded Image 2 where ClPs is the phenotype-specific clearance rate and F is OC's bioavailability. Then, the clinician selects a dosing interval, τ, to calculate OC's maintenance dose (Embedded Image): Embedded Image 3

A clinician typically prescribes oral OC for a patient in a daily dose, divided such that the previous dose has not resulted in “end-of-dose” pain, when the patient takes the next scheduled dose. In order to achieve the TTC at steady state, the clinician needs to consider the OC accumulation in the patient's body. This accumulation depends upon the rate at which the patient's body eliminates OC. This elimination rate, Embedded Image, depends on the patient's CYP2D6 phenotype. In order to calculate the patient's steady-state OC plasma concentration, we must first calculate OC's accumulation ratio: Embedded Image 4

Because AR depends on Embedded Image, which is phenotype-specific, AR is phenotype-specific. We calculate ARPs according to the following equation: Embedded Image 5 where Embedded Image is OC's phenotype-specific elimination half-life [25]. When OC is administered as a repetitive-dosing regimen, OC's expected phenotype-specific steady-state plasma concentration can be calculated from the following equation: Embedded Image 6

The plasma OC concentration will also fluctuate between a phenotype-specific maximum (peak) and minimum (trough) value: Embedded Image 7 and Embedded Image 8

The maximum length of time for a repetitive dosing interval to maintain plasma OC concentrations between MSC and MEC, τmax, can be obtained by solving Equation for τ: Embedded Image 9

Computational Methods

Phenotype-specific plasma OC concentration-time profiles were analyzed using an individual-based open two-compartment oral dosing model (Figure ; left panel). Model fits to the data were performed using weighted nonlinear least-squares regression with WinSAAM software [21,26]. Population data analysis was performed using WinSAAM's EMSA facility [27]. The model diagram for the population-based model data analysis is displayed in Figure (right panel).

Figure 1

The individual-based model (left panel) represents the base model, which is the model that best fits the experimental data in an individual. The base model parameter estimates and variance-covariance matrices are utilized, in a first step, as input into the population-based model (right panel). For population-based model data analysis, the subscript i refers to individuals, and j refers to their corresponding CYP2D6 phenotype. The transparent models represent the population-based model (right panel). The difference between the individual-based model and the population-based model is that the population-based model references both subjects i and phenotypes j across all studies, simultaneously. The triangle cuts into the sampled compartment.

Statistics

We present all values reported in the results as mean ± SD, unless otherwise stated. All individual-based model parameters were accurately estimated by WinSAAM with coefficients of variation of less than 25% [28]. Goodness-of-fit was assessed using R2 [29,30]. We performed data management and statistical analyses using R version 3.0.1 (R Foundation for Statistical Computing, Vienna, Austria) and clinical pharmacokinetics using the R package cpk version 1.3 (Comprehensive R Archive Network, 2013).

We amplified the vector of mean OC population phenotype-specific pharmacokinetic parameter estimates (see Results, Table ) for statistical analyses using the bootstrap resampling technique [31,32]. Bootstrap samples are similar to drawing a new sample from the population and provide good estimates of what would happen if we really were able to draw new fresh samples from the population. We used the bootstrapped samples to construct confidence intervals for the population parameters. The bootstrap approach is the computer simulation counterpart of DNA amplification where single-copy genomic sequences can be amplified by a factor of more than 10 million [33].

View this table:
Table 2

Oxycodone bootstrapped population pharmacokinetic parameter estimates by CYP2D6 phenotype

Parameters(N = 10,000)t1/2 (hours)ke (/hour)Cl (L/h/Kg)Vd (L/Kg)
CYP2D6 Phenotype
UM3.3 ± 0.5 (2.4, 4.3)0.2078 ± 0.0320 (0.1451, 0.2705)0.59 ± 0.12 (0.35, 0.83)2.8 ± 0.7 (1.4, 4.2)
EM3.8 ± 0.8 (2.2, 5.4)0.1824 ± 0.0380 (0.1079, 0.2568)0.51 ± 0.13 (0.26, 0.76)2.8 ± 0.9 (1.0, 4.6)
PM5.2 ± 0.2 (4.8, 5.6)0.1333 ± 0.0051 (0.1233, 0.1433)0.37 ± 0.06 (0.25, 0.49)2.8 ± 0.5 (1.8, 3.8)
  • One-way analysis of variance P < 0.01. Post-hoc Tukey tests revealed that only Vd was not significantly different between phenotypes.

  • t1/2 = elimination half-life; ke = fractional elimination rate of oxycodone from tissues in the body per unit time; Cl = clearance rate; Vd = apparent volume into which oxycodone distributes to produce the measured concentration.

  • UM = ultra-rapid metabolizer; EM = extensive metabolizer; PM = poor metabolizer.

We performed groupwise inferential statistics using one-way ANOVA and Tukey's honest significant differences for multiple comparison of means. Homogeneity of variances was checked using the Fligner–Killeen test. A significant difference or relationship was established when the probability of rejecting the null hypothesis was 0.05 or less.

Results

Pharmacokinetic Modeling

Figure shows the plasma OC concentration-time profiles digitally extracted from the publication by Samer and coworkers [34] for CYP2D6 EM, UM, and PM phenotypes. This approach has been successfully used to study the dynamic relationship between glucose, insulin and non-esterified fatty acids [35,36]. Subjects received a dose of OC equal to 0.2 mg/kg. The mean dose administered was 15.3 mg. Figure shows the fits (curved lines) of the individual-based model (Figure , left panel) to the actual plasma OC concentration-time data for each of one representative individual-based OC phenotype-specific modeling study. Figure also shows that the individual-based model was able to accurately describe the dynamic features present in the OC concentration-time profiles. We repeated this process for each individual study (N = 15 studies) and found a close fit with R2 values of 0.88 ± 0.01 for EMs, 0.74 ± 0.01 for PMs and 0.86 ± 0.01 for UMs, all P < 0.05.

Figure 2

Digitally reconstructed mean ± standard deviation oxycodone (OC) plasma concentration-time profiles from the data in Figure 2 (Panel B) in the study by Samer and coworkers [34]. OC plasma concentration-time profiles are plotted according to CYP2D6 extensive metabolizer (EM), poor metabolizer (PM), and ultra-rapid metabolizer (UM) phenotypes.

Figure 3

Plot showing weighted nonlinear regression model fits of oxycodone plasma concentration-time profiles for one of each CYP2D6 extensive metabolizer (EM), poor metabolizer (PM), and ultra-rapid metabolizer (UM) phenotypes. For each individual study, R2 values 0.88 ± 0.01 for EMs, 0.74 ± 0.01 for PMs and 0.86 ± 0.01 for UMs, all P < 0.05.

Table shows OC's population pharmacokinetic parameter estimates obtained using the population-based model data analysis (Figure , right panel). We performed a one-way ANOVA to compare the three CYP2D6 phenotype groups' individual-based model parameter estimates (P value <0.01). Post-hoc Tukey tests revealed statistically significant differences between the t1/2, ke and Cl for all three CYP2D6 phenotypes, but no significant difference in their Vd.

We found that the canonical order for OC's t1/2 was PM > EM > UM (P < 0.01). For ke, it was UM > EM > PM (P < 0.01), and for Cl, it was UM > EM > PM (P < 0.01).

In order to check the models external consistency, we used it to estimate OC's bioavailability (F) using each subjects plasma OC concentration-time data. The result was an overall F of 63% ± 5% with a coefficient of variation of 13% ± 1%. This value corresponds to our previously reported value of 63% ± 3% with a coefficient of variation of 5% ± 3% [37], which was estimated using entirely different data and technique. The F value obtained using the model also corresponds with the range of values for OC's bioavailability reported in the literature (60–87%) [38–40].

Dose-Response of UM Phenotype on Oxymorphone Levels

We determined the dose-related response that CYP2D6 mediated O-demethylation of OC could have on plasma oxymorphone (OCM) measurements in UM phenotypes in silico. We used a subject weighing 86 kg, which is the average weight for a man [41]. We used the individual-based model (Figure , left panel) with repetitive oral dosing equal to 5, 10, 15, and 20 mg OC every 4 hours, F equal to 0.74, and we parameterized the model using Embedded Image and Embedded Image population parameter estimates (Table ). The ke is the first-order elimination rate constant, which quantifies the fractional elimination rate of OC from tissues in the body per unit time. The Vd represents the apparent volume into which OC distributes to produce the measured concentration. When samples are not available, simulations using population parameter estimates are preferred [42]. Simulations were performed with the model constrained so that 2.9% of OC would be converted to OCM [43,44] to test the effect repetitive dosing increases would have on plasma OCM concentrations.

Results show that the dose-related plasma OCM concentrations were 6% (5 mg), 13% (10 mg), 19% (15 mg), and 25% (20 mg) below OC's MEC of 20 μg/L (Figure ). These results are consistent with findings in previous studies that show that OCM levels associated with oral OC dosing do not exert any central nervous system analgesic [43] or psychomotor effects [45] because OCM values derived from oral OC are too low in vivo [46].

Figure 4

Repetitive dose-related response that CYP2D6 mediated O-demethylation of oxycodone (OC) has on plasma oxymorphone (OCM) measurements in an ultra-rapid metabolizer phenotype. OCM values are consistently below OC's minimum effective concentration of 20 μg/L, indicating that they would not exert any central nervous system analgesic effect.

Personalized OC Dosing Algorithm

  • Step 1: Determine Patient's CYP2D6 Phenotype. Test a patient's CYP2D6 phenotype to classify the patient's clearance rate into one of three categories based on their CYP2D6 phenotype, i.e., either as an EM, PM, or UM.

  • Step 2: Determine Target Therapeutic OC Concentration. Determine or choose the target therapeutic plasma OC desired concentration for the patient.

  • Step 3: Use Phenotype-Specific Clinical Pharmacokinetics to Determine Dose. Determining a patient's CYP2D6 phenotype and choosing a target therapeutic OC concentration allows the clinician to use phenotype-specific clinical pharmacokinetics (Table ) to design a personalized OC dosing regimen.

Applications

Example 1

A clinician treats an 86 kg man with severe pain using OC (The NHANES 1999–2002 mean weight for men in the United States is 86.1 ± 0.4 [standard error of mean] kg). How does pharmacogenetic (PGt) testing and phenotype-specific clinical pharmacokinetics allow the clinician to determine a personalized OC dose?

  • Step 1: The clinician orders PGt testing and determines the patient's CYP2D6 phenotype is EM.

  • Step 2: The clinician chooses an OC TTC (Equation ) based on OC's therapeutic range of 50 to 20 μg/L [24]: Embedded Image 10

  • Step 3: The clinician uses phenotype-specific clinical pharmacokinetics (Table ) to determine a personalized dose rate (Equation ): Embedded Image 11

We used the median value of OC F = 0.74 for the calculation [47]. The clinician then determines a suitable dosing interval to keep the plasma OC concentration between 50 (MSC) and 20 (MEC). As: Embedded Image 12 Embedded Image 13

Taking the ln of both sides Embedded Image 14 and substituting the phenotype-specific value for Embedded Image from Table gives Embedded Image 15

The Greek symbol τ (tau) means dosing interval. So τmax represents the dosing interval to maintain plasma OC concentrations between MEC and MSC.

Finally, solving for τmax gives: Embedded Image 16

Equation gives the same result as above and is straightforward. The τmax of 5 hours means that the longest dosing interval that can be selected for the patient is 5 hours; otherwise, OC may not properly accumulate in the patient's body. But, because OC administration every 5 hours is not practical, a dosing interval (τ) should be selected from one of the following practical values: 4, 6, 8, 12, or 24 hours. In this case, a τ of 4 hours is the best choice.

Now, the phenotype-specific maintenance dose (Equation ) is calculated as: Embedded Image 17 which can be administered as OC 7.5 mg [48].

As the administered dose (7.5 mg/4 hours = 1.8 mg/h) is less than the calculated dose, back extrapolation can be performed to check that the administered dose predicts a plasma OC concentration that will fall within the desired TTC or within the desired therapeutic range: Embedded Image 18 which is sufficiently close.

Figure displays simulated OC concentration-time trajectories comparing the delivery of OC 8 mg every 4 hours with the orally administered OC 7.5 mg every 4 hours. The 7.5 mg dose provided peak (44 μg/L) and trough (21 μg/L) OC concentrations, which fall within OC's analgesic range of 20–50 µg/L [24].

Figure 5

Simulated oxycodone (OC) concentration-time trajectories in a CYP2D6 EM-phenotype 86 kg man. The delivery of OC 8 mg every 4 hours is compared with orally administered OC 7.5 mg every 4 hours. Both regimens maintain steady-state OC levels within the therapeutic range.

Example 2

A clinician treats an 86 kg man with severe pain using OC. How does PGt testing and phenotype-specific clinical pharmacokinetics allow the clinician to determine a personalized OC dose?

  • Step 1: The clinician orders PGt testing and determines the patient's CYP2D6 phenotype is PM.

  • Step 2: The clinician chooses an OC TTC of 20 μg/L based on OC's analgesic range of 20–50 μg/L [24].

  • Step 3: The clinician uses phenotype-specific clinical pharmacokinetics (Table ) to determine a personalized dose rate (Equation ): Embedded Image 19

As OC's half-life for a CYP2D6 PM patient is about 5 hours, the clinician chooses a dosing interval of 4 hours and confirms the dosing interval is suitable to keep the plasma OC concentration between Embedded Image (Equation ) and Embedded Image (Equation ): Embedded Image 20 and Embedded Image 21

Now, the phenotype-specific maintenance dose (Equation ) is calculated as Embedded Image 22

This can be administered as OC 7.5 mg [48] half-tablet (3.75 mg) every 4 hours. Percocet 7.5/325 mg is manufactured by Endo Pharmaceuticals, Inc. (Newark, DE, USA) and is readily available. It is also readily available in generic equivalent as Acetominophen/OC 325/7.5 mg, manufactured by Mallinckrodt Pharmaceuticals (St. Louis, MO, USA).

However, a resident clinician on the consult service indicated that this dosage regimen is substantially more inconvenient than giving 5 mg every 6 hours. But, the clinician indicated that the population OC t1/2 value for a PM is 5.2 hours. This means that a dosing interval greater than the OC t1/2 may not allow adequate OC accumulation on repetitive dosing. To verify, the resident clinician performed a simulation that supported the calculated prescribed dose (Figure ).

Figure 6

Simulated oxycodone (OC) concentration-time trajectories in a CYP2D6 poor metabolizer (PM)-phenotype 86 kg man. The delivery of OC 5 mg every 6 hours is compared with orally administered OC 3.75 mg every 4 hours. The OC 5 mg every 6 hours regimen produced an OC concentration-time trajectory that was below OC's minimum analgesic concentration most of the time. This was because the regimen failed to allow adequate accumulation of OC due to the dosing interval being too wide, i.e., it exceeded the OC PM t1/2 of about 5.2 hours.

Figure shows that OC 5 mg every 6 hours yields a steady-state OC MEC for less than half the 6 hours dosing interval. In contrast, the 3.75 mg every 4 hours regimen produced higher OC levels over almost the entire dosing interval (over 3 hours). The pharmacokinetics of extended release OC (OxyContin) are biphasic, which indicates that it undergoes transit delays in the circulation. This can lead to unpredictable pharmacokinetics and dangerous accumulation that can result in death.

Back calculation was performed to double-check whether the administered dose predicted a steady-state plasma OC concentration that fell within the desired TTC or within the therapeutic range: Embedded Image 23

This concentration was at the chosen target.

Figure shows simulated OC concentration-time trajectories comparing the orally administered dose of OC 3.75 mg every 4 hours, with delivery of the clinical pharmacokinetics-based maintenance dose of OC 3.5 mg every 4 hours. The OC 3.75 mg dose predicted a peak OC of 26 μg/L and an OC trough of 15 μg/L. This regimen predicts maintenance of OC concentrations >20 μg/L for over 3 hours or over two-thirds of the dosing interval. If the patient reports pain about 3 hours or more into the dosing interval, the OC dose should be increased by repeating the processes mentioned earlier.

Figure 7

Simulated oxycodone (OC) concentration-time trajectories in a CYP2D6 poor metabolizer (PM)-phenotype 86 kg man. The delivery of OC 3.75 mg every 4 hours is compared with orally administered OC 3.5 mg every 4 hours. The OC 3.75 mg every 4 hours regimen maintains OC plasma levels within OC's minimum effective concentration over three-fourths of the dosing interval or about 3 hours.

Example 3

A clinician is consulted to treat an 86 kg man with severe second-day postoperative hernia repair pain using oral OC. The patient has a history of increased nausea and constipation with previous opioid use. PGt testing was performed at that time. How does PGt testing and phenotype-specific clinical pharmacokinetics allow the clinician to determine a personalized OC dose?

  • Step 1: The patient's CYP2D6 phenotype is known to be UM.

  • Step 2: The clinician chooses an OC TTC of 50 μg/L based on OC's analgesic range of 20–50 μg/L [24]. The clinician reasons that the patient's ultra-rapid metabolism status would significantly reduce OC levels and thereby reduce its analgesic effect.

    Nonetheless, the clinician takes a preemptive step for patient safety and determines the dose-related impact OCM levels formed from O-demethylation of OC would have on OCM plasma levels in silico. It takes the clinician less than a minute to enter the patient's information into the software and 0.01 seconds for the computer to solve the problem.

    The results indicate that steady-state peak OCM levels would be 3.748 μg/L; which, contributes less than 8% to circulating OC levels. Therefore, the clinician decides to proceed with dosage regimen design using OC rather than use an opioid that is not metabolized via CYP2D6 such as buprenorphine, fentanyl, morphine, methadone, OCM, or hydromorphone [49].

  • Step 3: The clinician uses phenotype-specific clinical pharmacokinetics (Table ) to determine a personalized dose rate (Equation ): Embedded Image 24

As OC's half-life for a CYP2D6 UM patient is about 3 hours, the clinician chooses a dosing interval of 4 hours. Now, the phenotype-specific maintenance dose (Equation ) is calculated as: Embedded Image 25

Practically, this dose can be administered as OC 5.0 mg three tablets (15 mg) every 4 hours. However, because the maximum daily acetaminophen dose is 3 g daily, and the calculated dose prescribed as Percocet® would provide 3.9 g of acetaminophen; the doses are administered as Roxicodone® 5 mg tablets, which do not contain acetaminophen.

Because a dosing interval was chosen that is greater than the OC UM t1/2, back calculation is performed to check that the administered dose predicts a steady-state plasma OC concentration that falls within the desired TTC or within the therapeutic range: Embedded Image 26

The predicted steady-state concentration is essentially on target.

Example 4

A clinician is consulted to treat an 86-kg man with hepatorenal failure using oral OC. The patient has 8/10 pain from abdominal distention. He has a history of alcoholism. The clinician notes that currently, there is no quantitative approach to dosing OC in hepatorenal failure (a complex disease). She recommends PGt testing and starting an oral dose of OC without acetaminophen 5 mg every 4 hours. In addition, she orders therapeutic drug monitoring with OC plasma levels to be drawn at 2 000 and 2 400 hours after starting the repetitive dosing schedule. She choose the initial dosing assuming a UM phenotype (Table ). This regimen provides the fastest OC clearance. The patients pain improves to 3/10.

Genotyping results categorized the patient as CYP2D6 UM phenotype. Trough OC levels were 70 μg/L at 2 000 hours and 72 μg/L at 2 400 hours. Expected trough levels for a healthy UM would be 27 μg/L. The clinician calibrates the model using the measured OC trough levels. Figure shows the expected OC levels in a healthy UM (solid line) and the calibrated both phenotype-specific and patient-specific OC levels (dashed line).

Figure 8

Oxycodone (OC) concentration-time trajectories in a CYP2D6 ultra-rapid metabolizer (UM)-phenotype 86 kg man with hepatorenal failure. The scalloped solid line is the anticipated healthy population-based response. The two black triangles are the measured responses: OC trough level of 70 μg/L at 2 000 hours and 72 μg/L at 2 400 hours. The scalloped dashed line represents the calibrated model response to the measured OC trough levels. Note the significant, but accurately predictable, deviation of OC concentration-time trajectory in hepatorenal failure compared with the healthy population-based response.

The revised patient-specific estimate of Embedded Image was equal to 0.1413 ± 0.0018/h (estimate ± SD of estimate), with a difference of 0.0665/h compared with the population Embedded Image value of 0.2078/h (Table ). The fractional SD of that difference was significant at 0.001. The revised patient-specific estimate of OC ClUM was 0.396 (CV 0.4 %) L/h/Kg, with a Δ of 33% compared with the population value of 0.59 L/h/Kg (Table ). Hence: Embedded Image 27 and Embedded Image 28

Thus, due to hepatorenal disease, this UM patient's clinical pharmacokinetics behave like those of a CYP2D6 PM phenotype (see Example 2 earlier).

Discussion

The challenge has been how to integrate pharmacogenetic and clinical pharmacokinetic technologies to deliver their combined benefits to patients. We integrated PGt with clinical pharmacokinetics using pharmacokinetic modeling to allow a clinician to anticipate the relationship between OC dose, expected plasma levels, and therapeutic range. We call this integration, pharmacokinomics. Pharmacokinomics can assist in designing OC dosing regimens that maintain OC levels both at or above its MEC, and at or below its MSC. This may allow treatment at the boundary between toxicity and maximal OC effectiveness in severe complex pain. Moreover, pharmacokinomics may be helpful in the design of pain treatment strategies in patients with complex diseases, such as obstructive sleep apnea [50].

Neither CYP2D6 genotyping nor clinical pharmacokinetics alone allow personalized OC dosing because these two approaches are different. On the one hand, OC genotyping provides qualitative information about a patient's metabolic category. For example, current systems inform users that a CYP2D6 PM patient should receive a lower dose of OC than a CYP2D6 UM patient [51,52]. But, these systems do not provide dosing assistance. On the other hand, clinical pharmacokinetics provides useful quantitative information for dosing but cannot identify phenotypes. By contrast, pharmacokinomics quantitatively expresses a patient's phenotype (metabolic category) pharmacokinetically. Thus, by expressing an individual's phenotype pharmacokinetically, a clinician (at least theoretically) can improve the safety and efficacy of OC and decrease the risk for iatrogenically induced overdose or death.

Genetic factors determine opioid pharmacokinetics and contribute to their variability among individuals. In particular, OC pharmacokinetics are determined, in part, by the polymorphic CYP2D6 gene that regulates OC's O-demethylation to its minor and active metabolite OCM [53,54], and by OC's N-demethylation, regulated by CYP3A4/5 gene expression, to its predominant and inactive metabolite, noroxycodone (NOC) [43].

We assumed that the majority of OC is metabolized in the liver because the organ that expresses the highest levels of these genes is the liver, where the overwhelming majority of OC is metabolized [55,56]. We also assumed the lung does not contribute significantly to OC metabolism, although the lung contains CYP2D6 [55]. In addition, we assumed the gastrointestinal tract's metabolism of OC is negligible. The esophagus lacks both CYP3A4 and CYP2D6 gene expression. The CYP3A4 gene is expressed in the stomach, small intestine and colon in negligible amounts [55]. Violation of any or all of these assumptions would lead to underestimation of Cl, which would translate clinically into undertreatment of pain (see Equation ).

Heiskanen and coworkers [45] directly addressed the potential effects of OC metabolites in humans and found that although OCM has been suggested to be an active metabolite of OC that mediates, at least in part, analgesia analogous to morphine in codeine analgesia, this suggestion was not supported in their randomized, double-blind, crossover study. On the contrary, they found that after an oral dose of 20 mg OC, measured OCM levels were negligible. In addition, they found psychomotor function was unimpaired.

Their finding of negligible OCM levels after oral administration of OC in vivo is consistent with the results of Lalovic and coworkers [43]. Lalovic and coworkers found that only 2.9% of a 15 mg orally administered dose of OC undergoes O-demethylation to OCM via CYP2D6 gene expression. Lalovic and coworkers also found that 74% of the dose underwent N-demethylation to NOC, an inactive metabolite, via CYP3A4/5. What is more, they found that 21% of the dose was converted to the active metabolite noroxymorphone (NOM), from NOC, but at a rate 20- to 30-fold less than its N-demethylation to NOC.

More recently, Klimas and coworkers [46] demonstrated that OC itself is responsible for its analgesic effects. They found that although OCM and NOM have much higher affinity than OC for the μ-opioid receptor, the OCM and NOM metabolite concentrations at the site of action are very low. This indicates that any analgesic effects from these metabolites would be negligible. Therefore, there is currently no evidence that neither OCM nor NOM derived from oral OC administration exert any effects in adult humans.

Our study underscores that “One-Size-Fits-All” therapeutics, whose effects are predictable only in terms of probabilities, gives way to personalized therapy with entirely predictable properties. Our study shows that we can preempt individual pharmacotherapeutic responses to OC, a priori, via calculation by hand, in memory or in silico, which aids designing safe and effective OC treatments.

Most prescribing practitioners [57], with the exception of pharmacy clinicians receive limited training in basic pharmacokinetic principles, and consequently, they hesitate to use pharmacokinetic techniques for dose calculation across many important pharmacological classes [58–61]. Although this hesitation likely arises from deficient pharmacokinetic expertise, the mathematics can seem quite intimidating and practical applicability at the point of care seems daunting at best. Hesitation may also stem from an opinion that pharmacokinetics is not particularly valuable. Heretofore, dosing for most drugs is driven by the general approach that if a drug is not working, more should be given or, conversely, if it is producing toxicity, less should be given. A rational therapeutic design would logically be to tailor the concepts and techniques of clinical pharmacokinetics into specific dosing regimens using “pharmacokinomics.”

A limitation of the present study is that we only considered CYP2D6 phenotypes. OC's metabolism is also regulated by genetic expression of CYP3A4/5. The impact these may have on OC pharmacokinomics, and their implications for OC dosing regimen design and adjustment are currently unknown and require further study. But, considering that approximately 85% of all CYP450 metabolism is 3A4/5, this is less likely to contribute to polymorphic variability compared with 2D6 [62,63].

In addition, appropriate dosing adjustments have to be made in renal failure and hepatic failure (see Example 4 in Results), and in the elderly. Renal impairment is common among elderly patients. Unfortunately, we currently do not know enough about pharmacokinomic OC prescribing in these medical conditions or in special populations to give general advise [64,65]. For example, African Americans exhibit alterations in endogenous neurotransmitter modulated pain regulatory mechanisms, which may contribute to the greater rate of clinical pain symptoms they experience [66]. Also, there is evidence that African Americans have at least a 23% increase in hepatic metabolism of the G-protein ligand propranolol compared with Caucasians [67]. OC is also a ligand for a G-protein [68], and metoprolol is a phenotyping ligand for CYP2D6 activity [69]. These findings suggest that our results may represent underestimates in African Americans, which can result in undertreatment of their pain. Table tabulates prevalence of CYP2D6 UM metabolizers in different ethnic populations [70,71].

View this table:
Table 3

Prevalence of CYP2D6 ultrarapid metabolizers in different ethnic populations in the US

Ethnic PopulationPrevalence (%)
American 4.3
Northern European 1–2
African American 3.4–6.5 (12.1)
African/Ethiopian29
Caucasian 3.6–6.5 (69.1)
Hispanic 1.7 (12.5)
Spanish10
Greek 6
Saudi Arabian21
  • Prevalence is the proportion of a population found to have a condition.

  • Values in parentheses represent ethnic percentages of population density from the US Census Bureau. United States Census 2000. Available at http://www.census.gov/main/www/cen2000.html.

  • Prevalence data obtained from references 70 and 71 (see text).

A further limitation is that the pharmacokinetics of most drugs theoretically follow a multicompartmental model. But, the equations that are derived from them are too difficult for a day-to-day clinical application. For small rapidly diffusible molecules, such as OC (molecular radius 52 nm), the error in our simplified equations is acceptable [72,73]. Nevertheless, we cannot expect that measured plasma OC concentrations will be identical to those predicted using our simplified approach because there will always be interindividual and intraindividual variation, and model misspecification error.

The difference between the patient and the model is the model misspecification error, differences between patients is interindividual variability, and the difference between the patients' “true” value, if it could be known, and the measured concentration is the intraindividual variability. These will account for differences in the predictions relative to measured OC concentrations.

Although pharmacokinomics can be used by clinicians at the point of care for personalizing OC dosing regimens, caution should be exercised when applying results. Large interindividual variability in responses to OC dosing suggests that the simplified approach will not work accurately in all cases. Uncertainty in patient OC dosing histories and compliance with OC dosing regimens makes all efforts at OC dosing regimen analysis tentative at best. In addition, the expected variability and occasional errors in the laboratory may confound proper interpretation. These additional sources of variation underscore the importance of using pharmacokinomics in conjunction with a comprehensive patient pain evaluation and management approach. This includes careful attention to correct diagnosis, identification of therapeutic goals and objective therapeutic end points, correct choice of pain relievers, constant assessment and reassessment of therapeutic outcomes, and, when appropriate, pharmacogenetic testing and measurement of OC plasma and urine drug levels for therapeutic drug monitoring [37].

Conclusions

Personalized OC dosing using “pharmacokinomics” is a new tool for clinicians treating chronic pain patients with OC. By expressing a patient's CYP2D6 phenotype pharmacokinetically, a clinician (at least theoretically) can improve the safety and efficacy of OC, and decrease the risk for iatrogenic-induced overdose, or death. Pharmacokinomics moves OC's PGt from description to prediction and provides a general framework for integrating PGt with clinical pharmacokinetics into clinical practice for gene-based prescribing. Finally, the concept of incorporating pharmacokinomics underscores the importance of complex considerations that could be simplified with appropriate dosing tools.

Acknowledgments

Oscar A. Linares, MD, thanks the University of Michigan Horace H. Rackham School of Graduate Studies and the Geriatrics Center. He conceived this work's pharmacokinetic modeling as a visiting scholar there while working with Dr. Loren Zech at NIH, NCI, Laboratory of Mathematical Biology. The authors are grateful to the Editor and two anonymous reviewers for their assistance in improving the quality of this work.

Footnotes

  • Funding: This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

  • Conflict of Interest Disclosure: No conflicts of interest.

References

View Abstract