A Novel Correction Equation Avoids High-Magnitude Errors in Interpreting Therapeutic Drug Monitoring of Phenytoin among Critically Ill Patients

Abstract

Background:

Phenytoin has a narrow therapeutic index and the potential of under-treatment or toxicity. Available equations are used to correct for the impact of hypoalbuminemia on unbound (free) phenytoin levels. The authors aimed to determine the accuracy of equations used to estimate free phenytoin in hospitalized patients and assess the impact of using additional clinical data.

Methods:

Concurrently measured total and free phenytoin levels in hospitalized patients (2014-2018) were retrospectively evaluated, excluding those from patients on renal replacement therapy and valproic acid. Differences between actual and estimated free phenytoin levels by the original (Original WTZ), Anderson-modified, and Kane-modified Winter-Tozer equations were assessed using Pearson correlations and Bland-Altman analysis. Thereafter, a population-derived formula was developed and validated in a testing cohort.

Results:

In the 4-year training cohort (n=81), the Original WTZ equation had the smallest mean difference of all equations. A higher mean difference (−0.362 μg/mL [95% CI −0.585, −0.138] vs. −0.054 μg/mL [95% CI −0.186, 0.078]) was observed in intensive care unit (ICU) patients compared to non-ICU patients. A cross-validated multivariable model improved the accuracy of free phenytoin estimation in ICU and non-ICU patients, even in the separate testing cohort (n=52) with respective mean differences of −0.322 μg/mL [95% CI −0.545, −0.098] and −0.025 μg/mL [95% CI −0.379, 0.329] and was superior to the Original WTZ (mean difference −0.858 μg/mL [95% CI −1.069, −0.647] vs. −0.106 μg/mL [95% CI −0.362, 0.151], respectively).

Conclusions:

Free phenytoin levels in hospitalized patients cannot be accurately determined using available estimation equations, particularly in critically ill patients. Combining ICU status and other available clinical data can improve therapeutic drug monitoring and prevent high-magnitude errors, particularly when free phenytoin assays are not readily available.

Background

Phenytoin remains an important anticonvulsant that is commonly used in the hospital setting and widely implemented as part of treatment algorithms for status epilepticus.¹ Although phenytoin is highly efficacious, close therapeutic drug monitoring of serum levels is required because of its complex pharmacokinetics and drug-drug interactions, which increase the risk for both drug toxicity and under-dosing.² Extensive protein binding, non-linear hepatic clearance, and considerable inter-individual variation in metabolism contribute to dynamic serum drug levels in the setting of acute illness.³ Furthermore, phenytoin exists in an equilibrium between an active free (protein unbound) drug and an inactive protein-bound drug. Free phenytoin is considered to be a pharmacologically active component and is typically thought to encompass approximately 10% of the total drug levels in patients with normal albumin levels.⁴

In patients with a critical illness, variations in phenytoin pharmacokinetics have been observed in the setting of altered protein binding.⁵ Case reports of critically ill patients with phenytoin toxicity have shown that free phenytoin levels are disproportionately higher than the values expected according to the total serum concentration.^6–8 As the pharmacologically active component, free phenytoin concentrations determine both efficacy and potential toxicity. Total phenytoin levels greater than 20 μg/mL or free levels > 2 μg/mL have been associated with a range of toxicity from ataxia to coma, arrhythmia, and death.⁴

Because the assay to determine total phenytoin levels is readily available at most institutions, a variety of equations have been developed to estimate free phenytoin from the measured total phenytoin level. The most well-known of these estimates, the original Winter-Tozer (Original WTZ) equation⁹ was derived from outpatients with epilepsy; however, this equation has been found to yield inconsistent results in different hospitalized patient populations.^10–12 Additional modified equations based on the Original WTZ have been developed to estimate free drug levels in hospitalized and critically ill patients.¹³ However, it is unclear which equation has optimal performance in these settings.

We aimed to evaluate the performance of several free phenytoin estimation equations in critically ill and non-critically ill hospitalized patients at our institution. Furthermore, we developed and validated a novel multivariable equation to determine whether a further improvement in the accuracy and precision of free phenytoin level prediction could be achieved.

Methods

Study Design

Herein, we retrospectively assembled a cohort of consecutive patients admitted to an academic medical center and treated with phenytoin or fosphenytoin therapy. Patients with measured free phenytoin serum levels were identified through the hospital’s electronic health record database. Patients 18 years or older whose free and total phenytoin serum concentrations were measured within 1 h of each other were included in the analysis. Free phenytoin results were excluded if the patient: (a) had plasma albumin and blood urea nitrogen retrieved more than 48 h from the time free and total phenytoin serum levels were obtained, (b) was administered concurrent valproic acid therapy, (c) was administered renal replacement therapy, or (d) had an estimated creatinine clearance <10 mL/min based on the Cockcroft-Gault equation. We evaluated existing free phenytoin estimation methods and developed a novel model using demographic, laboratory, and intensive care status data obtained between January 2014 and December 2017 (training cohort) (Supplemental Digital Content Figure e-1). In the training cohort, estimated free phenytoin using existing correction equations were compared to concurrently measured free phenytoin levels to derive the agreement in both critically ill and non-critically ill patients. The performance of a novel multivariable model incorporating additional clinical data was subsequently validated in an independent and separate cohort of data acquired between January 2018 and December 2018 (testing cohort) and compared to existing equations. The study was approved by the institutional review board and formal consent was waived.

Laboratory Assays

Samples for total and free phenytoin levels were collected within 1 h of each other for analysis. Total phenytoin serum concentrations were measured by collecting blood samples in a serum separator test tube (Gold-top; BD Becton, Dickinson and Company, Franklin Lakes, NJ, USA) for immunoassay at 37 °C. Free phenytoin samples were analyzed by the Mayo Clinic Laboratory (Rochester, MN) by using the blood samples retrieved in a silicone-coated empty plastic tube (Red-top; BD Dickinson and Company) and immunoassayed at 37 °C. Free phenytoin samples were subjected to ultra-filtration to retrieve the protein-free filtrate that was analyzed according to the kinetic interaction of microparticles in a solution (KIMS) technique. Free phenytoin samples were stored at 25 °C for a maximum of three days prior to transport to Mayo Clinic Laboratory. The reference range for total phenytoin was 10-20 μg/mL while that of free phenytoin was 1-2 μg/mL.

Evaluation of the Test Performance of the Available Methods: ICU and Non-ICU Patient Populations

Herein, the Original WTZ, Anderson-modified Winter-Tozer (Anderson WTZ), and Kane-modified Winter-Tozer (Kane WTZ) were the phenytoin correction equations evaluated.^9,14–15 As phenytoin is highly protein bound (~90%), the estimated free phenytoin was calculated with 10% of corrected phenytoin using commonly utilized correction equations.⁴

$$\begin{gathered} \text{Original WTZ}:\text{Estimated free phenytoin}=(\text{Total phenytoin}∕([{0.20}^{\ast }\text{Albumin}]+0.1){}^{\ast }0.1 \\ \text{Anderson WTZ}:\text{Estimated free phenytoin}=(\text{Total phenytoin}∕([{0.25}^{\ast }\text{Albumin}]+0.1)){}^{\ast }0.1 \\ \text{Kane WTZ}:\text{Estimated free phenytoin}=(\text{Total phenytoin}∕([{0.29}^{\ast }\text{Albumin}]+0.1)){}^{\ast }0.1 \end{gathered}$$

Pearson correlations and Bland-Altman plots were assessed for each equation and stratified by the intensive care unit (ICU) and non-ICU status for agreement. Fisher’s exact test was used to determine whether patients with supratherapeutic free phenytoin levels (>2 μg/mL) were more likely to require vasopressors for hemodynamic instability.

Development, Validation, and Testing of Novel Methods: ICU and Non-ICU Patient Populations

Stepwise linear multivariable regression with the K-fold cross validation (5 folds) and least squares method was used to formulate a population-derived estimation equation with data obtained in the training cohort. By using this cross-validated estimation equation in the training cohort, we could predict free phenytoin levels in a separate external validation testing cohort. Pearson correlations and Bland-Altman plots were used to evaluate the performance between the population-derived estimation equation and free phenytoin in the testing cohort. JMP Pro, version 13.0 (SAS Institute, Cary, NC, USA) was employed to conduct the statistical analysis.

Results

A total of 133 phenytoin results from 88 unique patients were retrieved over the 5-year study period (Table 1). The median free phenytoin level was 1.9 μg/mL (interquartile range (IQR) 1.4 to 2.7) and the median total phenytoin level was 11.1 μg/mL (IQR 7.3 to 14.9).

Table 1.

Patient characteristics

Characteristic	4-Year Training Cohort [1/2014 – 12/2017] (N=81)	1-Year Testing Cohort [1/2018 – 12/2018] (N=52)
Female, n (%)	40 (49.4)	15 (28.8)
Age, years a	64.1 (24.5 – 94.5)	64.2 (36.2 – 93.7)
Patient care unit, n (%)
Neurology	49 (60.5)	9 (17.3)
Medicine	23 (28.4)	37 (71.2)
Cardiology	5 (6.2)	1 (1.9)
Surgery	4 (4.9)	4 (7.7)
Oncology	0 (0)	1 (1.9)
Admission to ICU, n (%)	38 (46.9)	40 (76.9)
Creatinine clearance, mL/min a	83 (15.1 – 239.7)	55.9 (13.8 – 287.9)
Actual body weight, kg a	72.7 (39.2 – 128)	70.0 (45.2 – 113)
Body mass index, kg/m2 a	25.7 (13.9 – 51.7)	25.4 (17.7 – 45.6)
Albumin, mg/dL a	3.2 (1.1 – 4.8)	2.6 (1.3 – 4.6)
Blood urea nitrogen, mg/dL a	17 (3 – 88)	26.5 (6 – 81)
AST, units/L a	30 (11 – 116)	39 (10 – 114)
ALT, units/L a	24 (8 – 138)	28.5 (11 – 204)
Total bilirubin, mg/dL a	0.3 (0.1 – 41.4)	0.5 (0.1 – 4.4)
Concurrent beta lactam use, n (%)	28 (34.6)	22 (42.3)

^aData are presented as median (range)

ALT, Alanine aminotransferase; AST, Aspartate aminotransferase; ICU, Intensive Care Unit

Test Performance of the Available Methods in the Training Cohort

In the training cohort, 81 phenytoin levels were obtained from 64 unique patients (Table 1). The median free phenytoin level was 1.7 μg/mL (IQR 1.3, 2.4) and the total phenytoin level was 12 μg/mL (IQR 8.4, 16.1). Although the Original WTZ equation had a strong correlation (r=0.85), a substantial mean difference (−0.198 μg/mL; 95% CI −0.326, −0.070) was identified relative to the measured free phenytoin levels, p=0.003. Both the Anderson Winter-Tozer equation and Kane-modified Winter-Tozer equation resulted in larger mean differences than the Original Winter-Tozer equation, with notable differences between ICU patients and non-ICU patients, respectively (Table 2, Figure 1). There was no statistically significant difference in vasopressor use for patients with measured free phenytoin levels >2.0 μg/mL (21.9%; 7/32) versus those with measured free phenytoin levels ≤ 2.0 μg/mL (10.2%; 5/49), p=0.20. Similar inter-patient and intra-patient variability was observed for difference in estimated free phenytoin using the Original WTZ equation and measured free phenytoin (Supplemental Digital Content Figure e-2).

Table 2.

Performance of the free phenytoin estimation equation model.

	Correlation	Mean difference μg/mL	95% CI	R2	P-value
4-Year Training Cohort (January 2014 – December 2017)
All Results in the 4-Year Training Cohort (n=81)
Original WTZ	0.85	−0.198	−0.326, −0.070	0.678	0.003
Anderson WTZ	0.85	−0.515	−0.643, −0.388	0.483	<0.0001
Kane WTZ	0.85	−0.698	−0.831, −0.566	0.269	<0.0001
Population-Derived a	0.88	−0.008	−0.123, 0.107	0.767	0.893
ICU Results in the 4-Year Training Cohort (n=38)
Original WTZ	0.77	−0.362	−0.585, −0.138	0.459	0.002
Anderson WTZ	0.77	−0.665	−0.886, −0.445	0.180	<0.0001
Kane WTZ	0.77	−0.842	−1.065, −0.619	−0.077	<0.0001
Population-Derived a	0.80	−0.009	−0.227, 0.208	0.603	0.932
Non-ICU Results in the 4-Year Training Cohort (n=43)
Original WTZ	0.92	−0.054	−0.186, 0.078	0.847	0.414
Anderson WTZ	0.92	−0.383	−0.520, −0.245	0.714	<0.0001
Kane WTZ	0.92	−0.572	−0.723, −0.420	0.529	<0.0001
Population-Derived a	0.95	−0.007	−0.118, 0.105	0.893	0.905
1-Year Testing Cohort (January 2018 – December 2018)
All Results in the 1-Year Testing Cohort (n=52)
Original WTZ	0.81	−0.684	−0.875, −0.493	0.253	<0.0001
Anderson WTZ	0.82	−0.958	−1.160, −0.755	−0.154	<0.0001
Kane WTZ	0.82	−1.117	−1.329, −0.905	−0.459	<0.0001
Population-Derived	0.84	−0.253	−0.441, −0.065	0.589	0.009
ICU Results in the 1-Year Testing Cohort (n=40)
Original WTZ	0.86	−0.858	−1.069, −0.647	0.075	<0.0001
Anderson WTZ	0.86	−1.120	−1.349, −0.890	−0.400	<0.0001
Kane WTZ	0.87	−1.273	−1.515, −1.031	−0.737	<0.0001
Population-Derived	0.83	−0.322	−0.545, −0.098	0.538	0.006
Non-ICU Results in the 1-Year Testing Cohort (n=12)
Original WTZ	0.93	−0.106	−0.362, 0.151	0.840	0.383
Anderson WTZ	0.93	−0.418	−0.712, −0.123	0.629	0.010
Kane WTZ	0.93	−0.598	−0.926, −0.270	0.398	0.002
Population-Derived	0.87	−0.025	−0.379, 0.329	0.715	0.88

^aLow mean differences <0.01 during the 4-year training cohort for the Population-Derived Method reflect this epoch which serves as the training set.

Figure 1.

Performance of the free phenytoin estimation equations in ICU and non-ICU patients.

Performance of the free phenytoin estimation equations in the ICU and non-ICU patient populations in the 4-year training cohort and 1-year testing cohort. The estimated free phenytoin equations assessed to determine the extent of agreement were the Original Winter-Tozer (Original WTZ), Anderson-modified Winter Tozer (Anderson WTZ), Kane-modified Winter-Tozer (Kane WTZ), and the population-derived correction equations. Mean difference between estimated free phenytoin and actual free phenytoin (μg/mL) by Bland-Altman analysis is displayed.

Development and Testing of the Novel Methods: ICU and non-ICU Patient Populations

We evaluated the baseline characteristics associated with the 81 phenytoin levels obtained from 64 unique patients in the training cohort to derive their significance and impact on the estimation of free phenytoin levels. Overall, age, albumin, blood urea nitrogen, and ICU status in addition to albumin and total phenytoin level were deemed significant factors for estimating free phenytoin. As a result, the free phenytoin estimation equation was derived from the population (Table 3).

Table 3.

Population-derived prediction model build: stepwise regression with K-Fold cross validation

Variable	Estimate	Standard Error	P-value
Intercept	1.836	0.522	0.0007
Total PHT level, μg/mL	0.139	0.010	< 0.0001
Age, years	−0.008	0.004	0.0374
Albumin, g/dL	−0.424	0.116	0.0005
BUN, mg/dL	0.010	0.004	0.0124
Intensive care unit status – yes/no	0.144/−0.144	0.071	0.0458
Population Derived Estimation Equation Estimated Free PHT = 1.69 + 0.139*(Total PHT) − 0.008*(Age) − 0.424*(Albumin) + 0.010*(BUN) + 0.288*(Critically Ill [yes, 1; no, 0])

Novel Population-derived estimation equation incorporating additional clinical data including age (years), albumin (g/dL), and blood urea nitrogen (BUN, mg/dL) to improve the estimation of free phenytoin (PHT, μg/mL) levels in ICU and non-ICU hospitalized patients

Validation of the Population-Derived Prediction Equation in the Testing Cohort

In an independent testing cohort, we applied our equation to 52 phenytoin levels from 24 unique hospitalized patients (Table 1). The median free phenytoin level was 2.1 μg/mL (IQR 1.5 – 3.2) and the total phenytoin level was 8.6 μg/mL (IQR 4.9 – 12.6).

In the testing cohort, the Original WTZ equation displayed a strong correlation (r=0.81); however, a large mean difference (−0.684 μg/mL; 95% CI −0.875, −0.493; p<0.0001) was identified relative to measured free phenytoin levels. In addition, larger mean differences were obtained with the Anderson-modified Winter-Tozer equation and Kane-modified Winter-Tozer equation than the Original Winter-Tozer equation, with more pronounced differences in ICU patients than non-ICU patients (Table 2, Figure 1).

The cross-validated population-derived equation incorporating serum total phenytoin, age, albumin, blood urea nitrogen, and ICU admission at the time of lab collection was further validated in a separate 1-year testing cohort. The novel population-derived equation could improve the estimation of free phenytoin levels in both ICU (mean difference −0.322 μg/mL; 95% CI −0.545, −0.098; p=0.006) and non-ICU patients (mean difference −0.025 μg/mL; 95% CI −0.379, 0.329; p=0.880) (Table 2, Figure 1). Although both inter-patient and intra-patient variability in the estimated and measured free phenytoin was found (Supplemental Digital Content Figure e-2), differing results were not achieved when a repeat analysis with a single sample per subject was carried out in the testing cohort for both ICU (mean difference −0.3014 μg/mL; 95% CI −0.520, −0.088); p=0.004) and non-ICU (mean difference 0.026; 95% CI −0.326, 0.379; p=0.56) patients.

Misidentification of Supratherapeutic Free Phenytoin Using Traditional Correction Methods

In hospitalized patients, the population-derived equation resulted in the lowest mean difference in the estimated and actual free phenytoin. This was followed by the Original WTZ equation, Anderson-modified Winter-Tozer equation, and the Kane-modified Winter-Tozer equation. Over the 5-year study period, the magnitude of the mean difference between Original WTZ estimated and actual free phenytoin levels (μg/mL) increased with corresponding fluctuations in baseline clinical variables (Supplemental Digital Content Figure e-3).

Overall, 58 (43.6%) of the phenytoin levels were supratherapeutic when directly measured during the study period (2014–2018), 20 (34.5%) of which were identified as subtherapeutic or therapeutic when predicted with the Original WTZ correction methods (Figure 2). For the phenytoin levels obtained during ICU admission, 48.7% (19/39) of the supratherapeutic free phenytoin levels was not identified as supratherapeutic with the Original WTZ correction equation. Notably, two patients admitted to the ICU were found to have supratherapeutic levels of 4.0 and 4.2 μg/mL, while the predicted levels were 1.04 and 2.13 μg/mL, respectively. Comparatively, in non-ICU patients, 5.3% (1/19) of supratherapeutic free phenytoin levels were not identified with the Original WTZ correction equation. Median free phenytoin fraction was 0.17 [IQR 0.12 to 0.24] during the study period. In ICU patients, free phenytoin fraction was higher (median 0.21 [IQR 0.16 to 0.31]) than that in non-ICU patients (median 0.13 [IQR 0.11 to 0.16]).

Figure 2.

Misclassification using the clinically relevant target free phenytoin ranges.

Mislabeling of the estimated free phenytoin levels (μg/mL) using the Original Winter-Tozer correction equation and actual free phenytoin levels (μg/mL) among ICU and non-ICU samples in the 5-year cohort (A) and the use of the population-derived estimation equation in the 1-year testing cohort (B)

In the testing cohort that comprised 52 phenytoin levels, 26 (50%) of the free phenytoin levels were supratherapeutic when directly measured. However, only 1 (3.8%) was identified as subtherapeutic or therapeutic when predicted using the population-derived correction methods compared to the 10 (38.5%) estimated with the Original WTZ equation.

Discussion

Estimations of free phenytoin levels are unreliable across patient populations, with the most prominent impact observed in the ICU patient population. Herein, we could develop and validate a novel, multivariable population-derived equation for free phenytoin prediction (incorporating age, ICU status, and blood urea nitrogen), which has a greater accuracy and precision, particularly in critically ill patients, than the traditionally used formulas.

Therapeutic drug monitoring for phenytoin is essential to ensure optimal efficacy and safety during acute illness. Although the therapeutic range for total serum phenytoin is 10 – 20 μg/mL (free phenytoin 1 – 2 μg/mL), target serum total phenytoin concentrations of 15 – 20 μg/mL (free phenytoin 1.5 – 2 μg/mL) may be required to manage status epilepticus to achieve improved seizure control.^2,16 Total phenytoin assays are readily available; however, the assay to determine free phenytoin levels may not be readily available at many institutions.

Multiple free phenytoin estimation equations have been proposed in the literature. The Original Winter-Tozer equation for correction of phenytoin in the setting of hypoalbuminemia was developed in outpatients with epilepsy, assuming a normal plasma albumin using an immunoassay temperature of 37 °C.⁹ A revised Winter-Tozer equation was proposed by Anderson et al., which proposed an interaction coefficient with albumin of 0.25 instead of the traditional 0.2; this occurred after the authors determined that the traditional equation overpredicted the normalized phenytoin concentration in elderly nursing home patients and critically ill trauma patients.¹⁴ Kane and colleagues proposed that an albumin interaction coefficient of 0.29 may be a better predictor of unbound phenytoin concentrations in a cohort of neurointensive care unit patients.¹⁵

In our study, the Winter-Tozer and derivative equations underestimated the actual free phenytoin levels in hospitalized and critically ill patients. Further, the misestimation was an order of magnitude larger in critically ill patients than those admitted to a non-ICU unit. Although this inaccuracy is consistent with prior reports,^10,15,17,18 the magnitude is demonstrated to result in meaningful differences in the ability to detect supratherapeutic levels in critically ill patients. The Kane-equation developed specifically in the neurointensive care unit patient population had the lowest agreement in our study. Although free phenytoin fractions may be constant in the outpatient setting, physiological changes during acute illness may contribute to the higher than predicted and variable free phenytoin fractions in hospitalized patients.

The unique pharmacokinetic alterations in the ICU patient population may contribute to an unexpectedly higher free to total phenytoin ratio.¹⁹ The important clinical mechanisms for the alterations in phenytoin protein binding include hypoalbuminemia, hepatic or renal impairment, uremia, and competition with other highly protein-bound medications that can cause a higher free fraction of phenytoin.⁵ Indeed, our multivariable regression model revealed that age, albumin, blood urea nitrogen, and ICU status are factors that influence phenytoin binding and the resulting free fraction. Based on our findings, current correction equations are unreliable for critically ill patients. Critically ill patients with significant metabolic derangements, hypoalbuminemia, and acute renal insufficiency and treated with multiple highly protein bound medications might be at the highest risk of increased free phenytoin exposures.

Although the best performance was achieved with the Original Winter-Tozer equation out of the three available equations at our institution, there were considerable differences in the actual and predicted free phenytoin level that may impact clinical decision making. Approximately 35% of supratherapeutic levels were not predicted to be supratherapeutic by the Original Winter-Tozer equation, thereby highlighting the need for more accurate free phenytoin prediction models to prevent unrecognized supratherapeutic levels and potential toxicity. A multivariable prediction model improved the estimation of free phenytoin levels in our patient population.

Clinical Implications

Although clinical decision making should not be conducted on serum drug concentration alone, and patient-specific phenytoin serum concentration goals may fall above or below free phenytoin levels of 1 – 2 μg/mL or total phenytoin levels of 10 – 20 μg/mL, the tendency for the current equations to underpredict free phenytoin levels should be considered during dose adjustment to maintain a therapeutic target; this is because of the increased risk of adverse side effects with higher serum concentrations. The improved estimation of free phenytoin levels can be obtained through direct measurement of free phenytoin or the utilization of the multivariable equations. Institutions with readily available assays for total and free phenytoin should consider monitoring free phenytoin in all critically ill patients and those with significant renal impairment or drug-drug interactions. If cost limitations are applicable, the total phenytoin immunoassayed at 37 $°$C corrected using the Original Winter-Tozer equation or population-derived multivariable equation may be appropriate in non-ICU patients without metabolic derangements or complex drug-drug interactions when clinical concerns of toxicity are absent. In institutions without access to readily available assays for free phenytoin, total phenytoin correction using the population-derived multivariable equation is recommended to improve the accuracy of free phenytoin estimation, particularly when phenytoin exposure is monitored in critically ill patients. Patients with clinical concerns regarding the adverse side effects of phenytoin therapy, such as those with altered mental status of unclear etiology, should have their free phenytoin level monitored to rule out phenytoin toxicity, regardless of the total phenytoin serum concentrations reported.

Furthermore, the clinical significance of this study may not be limited to hospitalized patients receiving phenytoin therapy. A discordance between the estimated and measured free serum valproic acid in critically ill and hospitalized patients was found for valproate, another highly protein bound (≥90%) anticonvulsant, thereby resulting in higher than expected free valproate concentrations and increased exposure.^20,21 Further data are however required to evaluate the impact of alterations in protein binding on the clinical outcomes and adverse events in patients receiving anticonvulsant therapies with narrow therapeutic indexes.

Limitations

The present study was retrospective and free phenytoin levels were measured at the discretion of the clinician. Therefore, free phenytoin may have been measured during a clinical suspicion of free phenytoin misestimation; however, the high volume of monitoring suggests that these levels were routinely measured. Due to the retrospective nature of this study, we could not differentiate between the levels obtained at request due to toxicity concerns and those obtained through routine monitoring to evaluate whether a difference existed in the predictive performance of existing equations. Nevertheless, an increase in the frequency of free phenytoin level assays was identified over time, suggesting an increase in use, including in the ICU population where clinical toxicity might be difficult to assess. A total of 53% of the free phenytoin levels evaluated during the training cohort were retrieved from non-critically ill patients while 77% of the levels in the testing cohort were retrieved from critically ill patients. Differences in clinical laboratory results (e.g., albumin, creatinine clearance, blood urea nitrogen) support the finding that the testing cohort had a greater illness severity, and the population-derived estimation equation from the training cohort was sufficiently robust for generalization to a population of different levels of illness severity and more frequent testing.

In addition, we could not correlate the disagreement between free and total phenytoin with clinical outcomes related to efficacy or toxicity due to the at-will reporting of medication-related adverse events at our institution. Herein, patients with end-stage renal disease on hemodialysis, those with an estimated creatinine clearance <10 mL/min, or on concurrent valproate therapy were not included in the analysis as errors are commonly encountered when estimating free levels in these populations. Therefore, our results are conservative and errors may be even greater in the general ICU population. Finally, laboratory analysis of free and total phenytoin was carried out at 37 °C at our institution. As a result, the findings herein may differ from those of other institutions that perform the phenytoin immunoassays at 25 °C.

Conclusions

Currently available correction equations poorly estimate free phenytoin levels in critically ill patients. The magnitude of differences is clinically large and misclassification often occurs whether a patient falls within the goal range that is clinically used for TDM. A novel population-derived formula that incorporates additional clinical data improves the estimation of free phenytoin levels in critically ill patients and is comparable to the Original WTZ equation in non-ICU patients. The accuracy achieved was also identified in a separate testing cohort. Owing to the potential failure to detect supratherapeutic drug levels because of the large errors encountered with traditional equations, centers should either pursue access to tests for the timely determination of free phenytoin levels and implement free phenytoin monitoring protocols or consider this novel formula for estimating free phenytoin instead of the Winter-Tozer and derivative equations, particularly in critically ill patients.

Supplementary Material

supplemental 1

Figure e-1. Study Schematic

The study schematic employed to evaluate the predictive performance of available free phenytoin estimation equations and the development and validation of a multivariable estimation equation.

^A The available free phenytoin estimation equations included the Original Winter-Tozer Equation, Anderson-Modified Winter-Tozer Equation, and Kane-Modified Winter-Tozer Equation.

Abbreviations: ICU, intensive care unit

supplemental 2

Figure e-2. Intra-patient and inter-patient variability in the estimated and measured free phenytoin samples.

A total of 81 free phenytoin levels in 64 unique patients were evaluated during the 4-year training cohort (2014 – 2017). Interpatient variability in the estimated free phenytoin using the Original Winter-Tozer equation and the measured free phenytoin is depicted across the patient cohort. Intrapatient variability in the estimated and measured free phenytoin is depicted in patients with more than one free phenytoin sample. Similar interpatient and intrapatient variability was observed.

supplemental 3

Figure e-3. Mean difference between the estimated and actual free phenytoin levels overtime.

Over the 5-year study period, the magnitude of the mean difference between Original Winter-Tozer (Original WTZ) estimated and actual free phenytoin levels (μg/mL) increased. The Original WTZ equation underpredicts the actual free phenytoin levels. The additional significant baseline clinical variables over time are presented as a median value for the patient population of the designated period.