Fentanyl–Norfentanyl Concentrations During Transdermal Patch Application: LC–MS-MS Urine Analysis

Abstract

Poklis and Backer published a survey of the concentrations of fentanyl and norfentanyl that could be expected in urine from patients using Duragesic^®, a transdermal fentanyl patch. That study employed a relatively small number of patient data points and analysis by Gas Chromatography/Mass Spectrometry. This work examines a larger population of patient positives for fentanyl and norfentanyl to determine whether more than a decade later the original report remains accurate in predicting the range and median levels of fentanyl and norfentanyl concentrations physicians can expect to see from their patients. Additionally, these data were transformed to develop a model that results in a near Gaussian distribution of urine drug test results. This retrospective approach was developed to transform and normalize urine drug testing results to provide a historical picture of expected patient values for this important analgesic. The resulting near Gaussian distribution is dose independent and as such should be of value to physicians in quickly assessing whether their patient is consistent with this historical population in the broad terms of this model. While this comparison alone is not definitive for adherence with a treatment regimen, together with patient interviews, prescription history and other clinical criteria, it can add an idea of expected patient values from urine drug testing.

Introduction

Fentanyl (Actiq^®, Duragesic^® and Sublimaze^®) is a high potency synthetic narcotic analgesic with a short duration of action. The analgesic effects of fentanyl are similar to those of morphine and other opioids but are ~100 times more potent than morphine. These transdermal patches contain 2.5–10 mg fentanyl and provide doses of 25–100 µg/h for 72 h aimed at management of chronic pain. Urine drug monitoring traditionally tests for fentanyl and norfentanyl by analogy to an intravenous dose where up to 85% are excreted in the urine over a 3- to 4-day period with 0.4–6% eliminated as unchanged fentanyl and 26–55% eliminated as the norfentanyl metabolite (1).

Patients on opioid therapy regimens are typically screened periodically to monitor compliance with the prescribed therapy because of known dependency risks (2). Poklis and Backer published a survey of the possible concentrations of fentanyl and norfentanyl in urine analyzed for patients using Duragesic^®, the original transdermal patch delivery form of fentanyl (3). Their work used Gas Chromatography/Mass Spectrometry (GC–MS) results from 546 patients who were taking different levels of fentanyl via the Duragesic^® transdermal patch. Interestingly, the formulation of Duragesic^® changed in 2009 from a “gel reservoir” to a “drug in adhesive” type of patch (4). In addition, urine drug monitoring has evolved to a greater use of Liquid Chromatography-Tandem Mass Spectrometry (LC–MS-MS) to assess drug concentrations for quantitative monitoring of patient compliance. Thus, it was decided to look at a larger body of historical patient data to determine whether (i) the drug concentration ranges in the original paper remain relevant through formulation change and analytical technique changes and (ii) a transformation of creatinine normalized data could be used to describe a more accurate statistical approach to determine the mean/median and standard deviations of these data.

Several authors have published reports of logarithmic transformations of urine drug data, as well as other matrices, to enable the use of normal (e.g., Gaussian) statistics to describe the distribution of data from drug monitoring results (5–12). In general, a simple logarithmic transformation of historical data yields a near Gaussian distribution. However, deviation from true Gaussian distribution statistics can remain unacceptably large. Application of patient-specific data such as transformation by weight, height, pH and age for example and normalization to the patient's creatinine level can improve the agreement of the distribution to the expected Gaussian statistics.

This work compares the original GC–MS work of Poklis and Backer with historical data taken by LC–MS-MS over a period of nearly 5 years (January 2011–September 2015) (3). The number of patient results examined exceeds 73,000 data points. Furthermore, this work details statistical methods developed to retrospectively transform and normalize historical fentanyl test results from urine drug testing to afford a near Gaussian distribution. The results provide a look at historical data for fentanyl that is both independent of daily dose and normalized to patient-specific parameters. The value of the methods detailed in this work is expected to extend to other opioids and other drug classes in urine drug testing.

Materials and Methods

LC–MS-MS analysis

Patient urine specimens were submitted for quantitative Liquid Chromatography-Tandem Mass Spectrometry (LC–MS-MS) fentanyl and norfentanyl analysis. The resulting data were used to compare with the original work of Poklis and Backer (3) and to develop the statistical distributions discussed in this paper. The juxtaposition of the raw testing results is used to compare concentration ranges observed in these two studies while the effort to develop the statistical distribution is intended to supplement these findings and provide a “comparison” tool that can be useful to clinicians. The limit of quantitation (LOQ) for this method is 2 ng/mL for fentanyl and 10 ng/mL for norfentanyl while the upper limit of quantitation (ULOQ) is 2,500 ng/mL for both analytes. The method was validated as described in Enders and McIntire with key parameters summarized therein (13). The data analysis and model development were conducted using R version 3.1: A language and environment for statistical computing (14). Data smoothing was completed using a kernel density process. The kernel density estimation is a well-accepted mathematical tool that “smooths” continuous data (e.g., Histograms) such that mathematical curve fitting and modelling can be accomplished. The kernel density estimation plot is simply used to “clean up” the display for inspection (15).

Model development

To ensure all the information required to adequately develop the model was available, only patients who had demographic information (gender, weight and height), a prescribed transdermal fentanyl dosage and positive fentanyl concentrations were included in the model. Further, the samples had to have acceptable creatinine, specific gravity and pH test results to be included—as established for workplace drug testing guidelines (16,17). The resulting population used to simulate the fentanyl model consisted of 73,710 independent individual patient results of which 65% were females and 35% were males. The average age of patients included in the model was 55 years old with an average lean body weight (LBW) of 53 kg. The average daily dosage of fentanyl taken by patients included in this model was 1.224 mg (with average prescribed therapeutic administration of 59 µg/h) and their median and average urine fentanyl concentrations were 37 and 88 ng/mL, respectively.

The developed model utilized a transformation and normalization of fentanyl concentrations from urine drug testing patient results. The term normalization refers to the concentration of fentanyl that has been modified to correct for one or more parameters associated with the patient. Caution was taken to ensure that units associated with all patient parameters were consistent. Furthermore, the natural log is taken for a unit-less function as described in Equation (1).

The raw fentanyl drug concentration measured in urine of the patient (F_CONC) is transformed and normalized as a function of patient prescribed drug dosage (Dose), LBW, patient urine pH and creatinine concentration (Crt) as described in Equation (1).

$${\mathrm{NORM}}_{\mathrm{CONC}}=\ln \left(\frac{F_{\mathrm{CONC}}(\frac{\mathrm{kg}}{\mathrm{L}})\times \mathrm{LBW}(\mathrm{kg})\times \mathrm{pH}}{\mathrm{Dose}\left(\mathrm{kg}\right)\times \mathrm{Crt}(\frac{\mathrm{kg}}{L})}\right)$$

(1)

This value is then transformed into its corresponding value on the standard normal distribution using Equation (2):

$$F_{\mathrm{STD}}=\frac{{\mathrm{NORM}}_{\mathrm{CONC}}-{\mu }_A}{{\sigma }_A}=Z_{\mathrm{SCORE}}$$

(2)

where ${\text{F}}_{\text{STD}}$ is the standardized normal value—referred to as the z-score for simplicity—and ${\mathrm{\mu }}_A$ and ${\sigma }_A$ are the mean and the standard deviation of the data set used to construct the model described in Equation (1). The values of ${\mathrm{\mu }}_A$ and ${\sigma }_A$ for this model are 11.781 and 1.311, respectively; the resulting mean and standard deviation of the standardized normal distribution, ${\text{F}}_{\text{STD}}$⁠, are “0” and “1”, respectively. The LBW parameter utilized in Equation (1) accounts for the sum of everything in the human body with the exception of fat including but not limited to bones, muscles and organs. The LBW is calculated using the James Formula described in the following equation (18,19):

$$\mathrm{LBW}\left(\mathrm{kg}\right)=\mathrm{fac}{\mathrm{t}}_a\times \mathrm{weight}\left(\mathrm{kg}\right)-\mathrm{fac}{\mathrm{t}}_b\times {\left(\frac{\mathrm{weight}\left(\mathrm{kg}\right)}{100\times \mathrm{height}\left(\mathrm{m}\right)}\right)}^2$$

(3)

where $\mathrm{fac}{\mathrm{t}}_a$ equals 1.1 for Men and 1.07 for Women and $\mathrm{fac}{\mathrm{t}}_b$ equals 128 for Men and 148 for Women.

Results and Discussion

Data analysis

Data from the earlier work and the current retrospective study are compared in Tables I, II and III. Overall, the fentanyl data from Poklis and Backer prove reliable across the drug formulation change and analytical testing differences. Data used in this retrospective study were accumulated over ~5 years (January 2011–September 2015) of LC–MS-MS urine drug testing. While the number of results used in each study differs, the mean values from each study are closely correlated for fentanyl. On the other hand, the mean values from each study for norfentanyl differ in a uniform manner with the LC–MS-MS values consistently higher than the GC–MS results. The observed maximum concentration values in this study are artificially restricted by the ULOQ of the LC–MS-MS analytical method for both fentanyl and norfentanyl; hence, a reliable inference from maximum values observed in the two studies (Table I) is impractical. Note, while the original work did not report data for the 12-µg/h dose, it is included here for completeness of dosage groups.

Table I.

Data from Fentanyl and Norfentanyl LC–MS-MS analysis

Dose (µg/h)	Number of specimens		Mean concentration (ng/mL)				Max concentration (ng/mL)
			Fentanyl		Norfentanyl		Fentanyl		Norfentanyl
	Current	P&B initiala	Current	P&B initiala	Current	P&B initiala	Current	P&B initiala	Current	P&B initiala
12	3,757	N/A	32	N/A	176	N/A	1,929	N/A	2,515	N/A
25	17,322	142	43	47	221	175	1,971	983	2,502	980
50	22,039	184	74	74	388	257	1,940	589	2,499	2,200
75	16,742	85	106	107	543	328	2,098	1,280	2,500	5,630
100	17,158	135	137	100	695	373	2,000	1,080	2,499	5,730
Total	77,018	546

The minimum concentrations for both Fentanyl and norfentanyl in this study were observed as being <LOQ (2 and 10 ng/mL, respectively).

^aThe results shown here are taken from the initial work published by Poklis and Backer (3), the minimum concentrations for fentanyl and norfentanyl for all dosages were reported as 0 ng/mL.

Table II.

Data from 90% of specimens tested Fentanyl and Norfentanyl using LC–MS-MS analysis: adjusted to remove potential outliers

Dose (µg/h)	Number of specimens		Mean concentration (ng/mL)				Max concentration (ng/mL)
			Fentanyl		Norfentanyl		Fentanyl		Norfentanyl
	Current	P&B initiala	Current	P&B initiala	Current	P&B initiala	Current	P&B initiala	Current	P&B initiala
12	3,381	N/A	19	N/A	149	N/A	120	N/A	2,515	N/A
25	15,590	128	30	32	195	173	161	167	2,502	980
50	19,835	166	56	58	367	251	272	250	2,499	860
75	15,068	77	84	95	525	285	385	444	2,500	1,330
100	15,442	121	139	79	680	327	494	350	2,499	1,670
Total	69,316	492

The minimum concentrations for fentanyl in this study were 1 ng/mL at 12 and 25 µg/h, 2 ng/mL at 50 µg/h, 3 ng/mL at 75 µg/h and 4 ng/mL at 100 µg/h. Minimum concentrations for norfentanyl remained 0 ng/mL for all dosage levels.

^aThe results shown here are taken from the initial work published by Poklis and Backer (3). The minimum concentration of fentanyl was reported as 0 ng/mL for all dosage levels and for norfentanyl was reported as 4 ng/mL for the 75-µg/h dosage level and 0 ng/mL for all other dosage levels.

Table III.

ANOVA table using the data from 90% of specimens tested Fentanyl and Norfentanyl using LC–MS-MS analysis

Dose (µg/h)	Fentanyl
	Source of variation	Degrees of freedom	Sum of squares	Mean squares	F_Value
25	Between studies	1	596	596.47	0.5688
	Within studies	15,716	16,480,891	1,048.67
50	Between studies	1	865	864.5	0.2588
	Within studies	19,999	66,812,389	3,340.8
75	Between studies	1	9,813	9,813.2	1.383
	Within studies	15,143	107,448,995	7,095.6
100	Between studies	1	123,379	123,379	10.456
	Within studies	15,558	16,480,891	11,800

Dose (µg/h)	Norfentanyl
	Source of variation	Degrees of freedom	Sum of squares	Mean squares	F_Value
25	Between studies	1	60,886	60,886	1.0236
	Within studies	15,716	934,785,796	59,480
50	Between studies	1	1,692,684	1,692,684	11.829
	Within studies	19,999	2,861,746,351	143,094
75	Between studies	1	3,155,732	3,155,732	13.755
	Within studies	15,143	3,474,118,676	229,421
100	Between studies	1	10,084,719	10,084,719	31.961
	Within studies	15,558	4,909,080,225	315,534

In all cases the F_Critical Value = F_crit(1, >1,000) with α = 0.05 which is 1.04. Hence, H0 is rejected if F ≥ 1.04.

In keeping with the original study, and in an effort to remove outliers and evaluate a more representative sample (3) we examined 90% of the data; i.e., removing 10% of the number of points with 5% from the low end and 5% from the high end of the data range. The fentanyl concentrations were used for this re-evaluation and 10% data removal. The results of this approach are compared with results from the original study in Table II.

Table II demonstrates better agreement between the original work (3) and the current data set analyzed in this study when compared with data presented in Table I. Even the maximum and minimum values for fentanyl show much better agreement. The mean LC–MS-MS values for norfentanyl remain consistently higher than the corresponding GC–MS values seeming to be on a higher slope of concentration in urine vs dose level than the original work as shown in both Tables I and II. Table III shows the results of the one-factor analysis of variance (ANOVA) conducted on the two data sets. The results of this analysis are used to provide statistical evidence of whether there is a significant different between the average fentanyl and norfentanyl concentrations between the two studies at the defined dosing levels. In this case, the null hypothesis is that the means for fentanyl and norfentanyl are equal at all the varying concentration levels. For example H0: mean fentanyl 25 µg/h (Poklis and Backer (3)) = mean fentanyl 25 µg/h (current study); mean norfentanyl 75 µg/h (Poklis and Backer (3)) = mean norfentanyl 75 µg/h (current study) and so on. The level of significance associated with these tests is α = 0.05. Since all the dosage level have one degree of freedom between studies Df1 = 1 and >1,000 degrees of freedom when adding the number of test subjects within both studies Df2 = >1,000, the critical F value is equivalent to F (1, >1,000) which is 1.04 and the null hypothesis will therefore be rejected if F_value ≥ 1.04 and accepted if F_value ≤ 1.04.

Table III suggests that there is statistically significant evidence to show that there is a difference in the means of fentanyl concentrations at 75 and 100 µg/h dosage levels and the means of norfentanyl concentrations at 50, 75 and 100 µg/h dosage levels. The statistically significant difference between fentanyl mean values at a dose of 75 µg/h is perplexing inasmuch as the Poklis and Backer data exceed that from this study at 95–84 ng/mL, respectively. The subsequent data points at 100 µg/h are switched with the Poklis and Backer data point being less than the mean value from this study (Table II). Furthermore, results presented in Table III show that there is statistically significant evidence to show that the means are equal for fentanyl concentrations at 25 and 50 µg/h dosage level and for norfentanyl at 25 µg/h dosage level. Overall, any significant differences between the original report of Poklis and Backer (3) appear to be at the highest dose level for fentanyl and are much more apparent for norfentanyl, at all but the lowest dose, than fentanyl. It is tempting to assign these differences to analytical technique, altered formulation (of Duragesic) or numbers of samples used. However, in broad terms, the current results demonstrate that even after more than a decade the results of the smaller study conducted by Poklis and Backer remain fairly reliable and robust for fentanyl and norfentanyl especially at lower dose levels.

Near Gaussian distributions

A Histogram of raw fentanyl concentrations for the population of patients used to develop the normalized and transformed model is presented in Figure 1. The exponential nature of the “linear” distributions of fentanyl concentrations from urine drug testing is evident. While the concept of “mean” is still the sum of all data points divided by the number of points, the concept of standard deviation as applied to a “normal” distribution is meaningless in this context. In an effort to present these data in the more common Gaussian distribution, several authors have used a logarithmic transformation (5–12). Cao et al. transformed their data into the log (base 10) space to generate means and standard deviations, which they then reverse transformed to generate the same values in the original linear space (5). However, their transformation did not take patient-specific information into account. In an effort to get these data to better fit a Gaussian distribution, we have applied patient-specific parameters as detailed in the Materials and methods section to transform and normalize these data.

Figure 1.

Histogram of the raw fentanyl urine drug concentrations observed from a body of collected urine test results used to generate the mathematically transformed and normalized standard curve for fentanyl in urine.

Fentanyl urine results, demographic information (gender, weight, height and age) and the prescribed dosage of fentanyl for 120 randomly selected patients—not included in the patient population used to design the model—were used to assess the validity and robustness of the model. A comparison of the validity assessment data set and the data set used to develop the fentanyl model is shown in Table IV.

Table IV.

Summary of Fentanyl model and validity testing data set

Total population			Model data set	Validity testing data set
			N = 73,710 Subjects	N = 120 Subjects
Data collection period			58 Months	2 Weeks
Gender distribution			F: 65%	F: 61%
			M: 35%	M: 39%
Patient population averages with standard deviations	Age		55 ± 13 Years	56 ± 13 Years
	Height		1.68 ± 0.11 m	1.68 ± 0.10 m
	Weight		83 ± 23 kg	88 ± 28 kg
	Fentanyl dosage	Daily dosage	1.224 ± 0.921 mg	1.010 ± 1.406 mg
		Therapeutic dosage	59 ± 29 µg/h	51 ± 30 µg/h
	Urine fentanyl concentration	Mean	88 ng/mL	235 ng/mL
		Median	37 ng/mL	34 ng/mL
	Lean body weight (LBW)		53 ± 11 kg	55 ± 12 kg

Figure 2 (top) shows a histogram of the results of normalizing and transforming the entirety of the fentanyl data described in this paper. The resulting graphic is consistent with a near Gaussian distribution. A true Gaussian population distribution has 68% of the data within ±1SD, 95% of the data within ±2SD and the other 5% greater than ±2SD. In the model described here, 70% of the population are within ±1SD while 96% are within ±2SD. The kernel density distribution of the fentanyl model is overlaid with a best fit standard normal distribution (Figure 2—bottom) giving a visual comparison of how well the developed model corresponds with a standard normal distribution.

Figure 2.

Histogram of the transformed, normalized and standardized fentanyl urine data (top) and the kernel density estimation plot derived from the transformed, normalized and standardized fentanyl data overlaid with the least squares minimized best fit Gaussian distribution curve (bottom).

In the broadest sense, this model provides a method of determining whether a patient's fentanyl urine drug test result is consistent with a historical population of fentanyl-positive patients. The method requires determining the concentration of fentanyl in the urine of the patient and other specific parameters including, prescribed daily dose of fentanyl, urine creatinine and pH levels as well as patient weight, height and gender. This information is used as described in Equations (1) through (3) to determine the $F_{\mathrm{STD}}$—standardized normal value or z-score—for an individual patient urine sample tested for fentanyl. Comparing a patient's z-score with the historical Gaussian distribution prepared from a body of known test patients who were both prescribed fentanyl and tested positive for fentanyl and/or metabolite in urine could provide insight as to whether the patient is “typical” when compared with a larger population.

Conclusion

Comparison of a large number of fentanyl and norfentanyl results with earlier work suggests that other than extreme outliers, the data compare favorably for fentanyl (3). Norfentanyl data are not statistically significantly different at the 50-µg/h dose level but are higher for the LC–MS-MS results than the GC–MS results at elevated doses. Transforming and normalizing these historical urine fentanyl data results in a near Gaussian distribution that can be used to assess patient consistency with this historical population.