If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
The likelihood of valproate (VPA) induced thrombocytopenia increases with higher VPA levels. In critically ill patients, the biological active free VPA level cannot be predicted from the total serum level. In this study, we evaluated the relationship between trough free VPA serum levels and concomitant platelet counts and assessed risk factors for the development of thrombocytopenia with the aim of generating a formula specifying the probabilities of developing thrombocytopenia based on trough free serum VPA levels.
Methods
Trough free VPA levels and concomitant platelet counts were collected from a large cohort of patients who participated in a prospective VPA monotherapy trial. Significant variables associated with thrombocytopenia in a univariate analysis were evaluated in a multivariate model. A receiver operator curve was performed to compute the trough free VPA levels with the greatest discriminating power in predicting thrombocytopenia.
Results
844 trough free VPA levels and concomitant platelet counts obtained from 264 patients were analyzed. In a multivariate analysis, trough free VPA levels, gender, and baseline platelet counts were significantly associated with thrombocytopenia. Using stepwise regression and multivariate logistic regression analyses, we generated gender-specific formulas for predicting platelet counts and probabilities of developing thrombocytopenia. The trough free VPA with the greatest discriminating power to predict platelet values ≤ 100,000/μL was 16.65 µg/mL.
Conclusions
The generated model was based on trough free VPA levels and achieved high sensitivity and specificity. Our results are therefore generalizable and can be applied to estimate the probability of developing thrombocytopenia in critically ill patients.
Valproate (VPA) is an antiseizure medication (ASM) approved as monotherapy and add-on therapy for the treatment of focal impaired awareness seizures, as a prophylactic agent for migraine headaches, and for the treatment of acute manic episodes in patients with bipolar disorder. Like many ASMs, it is associated with idiosyncratic reactions and dose-related adverse events including weight gain, hair thinning, tremor, and thrombocytopenia [
]. Previous studies reported substantially different frequencies of VPA induced thrombocytopenia, with the discrepancies mostly due to methodological disparities [
]. More recently, we documented a linear relationship between rising trough total VPA levels and reduced platelet counts, with additional risk factors consisting of female gender and lower baseline platelet counts[7]. In that study, we determined that the probability of developing thrombocytopenia (defined as platelet counts ≤ 100,000/μL) substantially increased at trough total VPA levels above 100 μg/mL in women and above 130 μg/mL in men [
]. Those results are useful for evaluating the risk of developing thrombocytopenia in otherwise healthy outpatients treated with VPA. In those patients, VPA is expected to be highly protein bound (90–95%), mostly to albumin, when the serum level is within the therapeutic range. However, when the total VPA level exceeds 80 µg/mL, its protein binding sites get saturated which results in a progressive increase in the biologically active free VPA fraction [
]. In addition, we and others have shown that the free VPA level is nonlinearly related to the total VPA level, and can be affected by albumin levels or the presence of exogenous or endogenous substances that compete with or displace VPA from its binding sites [
]. These findings are especially pertinent to critically ill patients who are at higher risk of developing hypoalbuminemia (reduction in VPA protein binding sites) and renal insufficiency (accumulation of organic acids that compete with VPA for its protein binding sites) which will result in higher free serum VPA levels. In those patients, monitoring the total VPA level can be misleading, since the latter can be within the therapeutic range despite high concentration of the biologically active free level. These are the principal arguments behind the recommendation to monitor the free VPA level in critically ill patients when toxicity is suspected [
In this study, we evaluated the relationship between the biologically active trough free serum VPA levels and concomitant platelet counts and assessed risk factors for the development of thrombocytopenia using data from a large cohort of patients who participated in a prospective randomized, double-blind, VPA monotherapy concentration-response design trial. Our final aim was to generate a formula that would predict the probability of developing thrombocytopenia based on trough free serum VPA levels.
Methods
Study design
The data used in this study was derived from a randomized, double-blind, parallel-group, multicenter, concentration-response design trial that compared the safety and efficacy of high (80–150 µg/mL) versus low (25–50 µg/mL) target trough serum VPA concentrations in patients with focal impaired awareness seizures (FIAS) treated with divalproex sodium (DVPX) as monotherapy. Those patients were followed on an outpatient basis with clinic visits scheduled according to the study protocol [
]. The Institutional Review Board at each center approved the study protocol, which was conducted in compliance with the US Food and Drug Administration regulations. More details about the protocol design were previously published [
Patients 10 to 75 years of age with a diagnosis of localization-related epilepsy and poorly controlled focal onset seizures (at least two FIAS per month) while maintained on one ASM (carbamazepine, phenytoin, primidone, or phenobarbital) at therapeutic serum concentrations, were eligible to participate in this study. Patients with major medical illnesses, psychiatric disorders, history of psychogenic non-epileptic seizures or noncompliance were excluded.
Patients were randomly assigned in a 1:1 ratio at each center into the high (80–150 µg/mL) or low (25–50 µg/mL) trough total serum VPA concentration groups. The study consisted of a baseline phase lasting 8 to 12 weeks, and a 24-week double-blind experimental phase. The experimental phase was divided into a dosage adjustment period (first 8 weeks) followed by a 16-week dosage maintenance period. During the dosage adjustment period, the baseline ASM was tapered and treatment with DVPX was initiated. DVPX dosage was gradually titrated upward to achieve the maximum tolerated serum concentration within the targeted range for each patient. To enter the 16-week maintenance period, patients had to be completely withdrawn from their baseline ASM and treated with DVPX as monotherapy. During this phase of the protocol, DVPX dosage was adjusted based on efficacy and tolerability, while maintaining trough serum VPA levels within the targeted ranges.
Trough total and free serum VPA concentrations and platelet counts were determined from analysis of blood samples collected either eight to 15 h after the last DVPX dose, or less than one hour after the first dose of the day (taken before noon). Platelet counts were measured during the baseline phase, every 2–4 weeks during the dosage adjustment period, and every 8 weeks thereafter. Trough total and free VPA levels were measured at every visit. All VPA concentrations were assayed at a central laboratory using a commercially available fluorescent polarized immunoassay with a detection range of 2 to 150 µg/mL
For each patient, and in order to abolish any potential confounder that could affect serum VPA levels, all included paired data consisted of trough free VPA levels and concomitant platelet count. In this study, we defined clinically relevant thrombocytopenia as either platelet counts of ≤ 100,000/μL [
] since those two cutoff values were used in previous studies.
Analysis
We calculated the mean platelet counts at baseline and following exposure to DVPX, and performed a linear regression between trough free VPA levels and platelet counts.
Univariate correlations between platelet counts and multiple variables, including trough free VPA levels, patient's age and baseline platelet counts were performed. A multivariate stepwise regression analysis was done to determine the significant variables in this group.
A univariate logistic regression analysis was performed to assess the probability of thrombocytopenia using the same variables described above in addition to gender. Significant variables in the univariate analyses were entered in a multivariate logistic regression analysis with thrombocytopenia as the dependent variable. To obtain an accurate estimation of the population coefficients derived from the multivariate logistic regression analysis, we planned to calculate these coefficients using a minimum sample size of 500 [
Sample size guidelines for logistic regression from observational studies with large population: emphasis on the accuracy between statistics and parameters based on real life clinical data.
Finally, a receiver operating curve (ROC) was plotted to compare sensitivity versus 1-specificity for the probabilities of thrombocytopenia derived from the logistic regression analysis. The Youden index was used to estimate the optimal threshold between sensitivity and specificity.
Results
Demographics and platelet counts
264 patients (144 women and 120 men with a mean age of 34.5 ± 13.3 years (range: 9–77 years) were included in this study. A total of 844 trough free VPA levels and concomitant platelet counts were analyzed. The mean trough free VPA level was 13.9 ± 14.9 µg/mL (range: 2–124 µg/mL) and the mean baseline platelet count was 273,000 ± 56,000/µL (range: 157,000–478,000/µL).
The platelet count dropped to a mean of 216,000 ± 75,000 /µL (range:16,000 – 439,000/ µL) after exposure to DVPX. The mean reduction from baseline platelet count was 57,000 ± 72,000/µL (range: −122,000 - 329,000/µL). There was a significant linear relationship between the trough free VPA level and platelet counts after exposure to DVPX (r = –0.611, p < 0.0001) (Fig. 1).
Fig. 1Linear regression of platelet counts after exposure to DVPX versus trough free serum VPA levels with 95% confidence intervals. There was a significant negative correlation that best fitted the following formula: Expected platelet count = 259,000 – (3.07 ∗ trough free serum VPA level); (r = –0.611, p < 0.0001).
Trough free valproate levels and baseline platelet counts
Whether thrombocytopenia was defined as platelet counts ≤ 100,000/μL or ≤ 150,000/μL, the thrombocytopenic episodes were associated with significantly higher mean trough free VPA levels and significantly lower mean baseline platelet counts (Table 1).
Table 1Comparison of the means of trough free VPA levels and baseline platelet counts between episodes of thrombocytopenia and absence of such occurrences.
Forty-five patients (17.0%) experienced a total of 69 thrombocytopenic episodes when defined as platelet counts ≤ 100,000/µL. During those episodes, the mean platelet count was 71,780/µL (range 16,000/µL −100,000/µL) with a corresponding mean trough free VPA level of 37.9 µg/mL (range 13 µg/mL −124 µg/mL). When thrombocytopenia was defined as platelet counts ≤ 150,000/μL, 82 patients (31.1%) experienced 150 occurrences with a mean platelet count of 100,600/µL (range 16,000/µL −150,000/µL) and a corresponding mean trough free VPA level of 33.8 µg/mL (range 5 µg/mL −124 µg/mL).
Table 2 stratifies the frequencies of thrombocytopenia according to gender and various free VPA level ranges. For the whole group, the frequencies of thrombocytopenic episodes (when defined as platelet counts ≤ 100,000/µL) were 0.0%, 7.3%, 21.3%, and 36.9% for trough free VPA levels in the ranges of < 10 µg/mL, 10–20 µg/mL, 20–30 µg/mL and more than 30 µg/mL, respectively. The corresponding percentages when thrombocytopenia was defined as platelet counts ≤ 150,000/μL were 1.0%, 15.3%, 54.3%, and 68.9%, respectively (Table 2).
Table 2Frequencies of thrombocytopenic episodes stratified by trough free VPA level ranges and gender.
The thrombocytopenic episodes were significantly more common in women although the baseline platelet count was similar between men (274,000/μL) and women (272,000/μL) (t-test, P = 0.629). When thrombocytopenia was defined as platelet counts ≤ 100,000/μL, 3.6% thrombocytopenic episodes occurred in men versus 12.2% in women (P < 0.001). The corresponding percentages were 14.3% and 20.8% when thrombocytopenia was defined as platelet counts ≤ 150,000/μL (P = 0.014).
Stepwise regression analyses
Separate stepwise regression analyses for men and women were performed to calculate the expected platelet count according to the trough free VPA level and baseline platelet count with the following results:
Univariate and multivariate logistic regression analyses
A univariate logistic regression analysis showed that gender, baseline platelet counts and trough free VPA levels were significantly associated with thrombocytopenia when defined as platelet counts ≤ 100,000/μL (Fig. 2A ) or ≤ 150,000/μL (Fig. 2B). Age was not significantly associated with the occurrence of thrombocytopenia.
Fig. 2Probability of developing thrombocytopenia when defined as platelet count ≤ 100,000/μL (A) or ≤ 150,000/µL (B) and stratified by gender according to trough free serum VPA levels.
In addition, we calculated the probability of developing thrombocytopenia stratified by gender when defined as platelet counts ≤ 100,000/μL or platelet counts ≤ 150,000/μL for a given trough free VPA level with the following results:
Probability of developing a platelet count ≤ 100,000/μL
For men: Ln (p/(1-p)) = (0.082 * trough free VPA level) - (0.011* baseline platelet count/1000) −2.163 (predictive value 95.7%).
A ROC was performed to calculate the sensitivities and specificities of various trough free VPA levels in predicting the occurrence of thrombocytopenia (defined as platelet counts ≤ 150,000/µL) (Fig. 3A). Using the Youden index, trough free VPA levels greater than 16.65 µg/mL had the greatest discriminating power for the occurrence of thrombocytopenia with a sensitivity of 89.3%, a specificity of 83.6% and an area under the curve (AUC) of 0.926 .
Fig. 3Receiver operating curves showing the sensitivities and specificities of various trough free VPA levels in predicting the occurrence of thrombocytopenia when defined as platelet counts ≤ 150,000/µL (A) or ≤ 100,000/µL (B).
Similarly, an ROC to predict the occurrence of thrombocytopenia (defined as platelet counts ≤ 100,000/µL) was performed (Fig. 3B). Using the Youden index, trough free VPA levels greater than 19.95 µg/mL had the greatest discriminating power for the occurrence of thrombocytopenia with a sensitivity of 87%, a specificity of 81% and an AUC of 0.907.
Discussion
This is the largest study to date that evaluated the risk of thrombocytopenia according to trough free VPA level. Our results indicate that a rising free VPA level, female gender and baseline platelet counts are all significant variables that predict the probability of developing thrombocytopenia.
The inverse correlation between rising VPA level and decreasing platelet counts is not novel. We previously reported that the frequency of thrombocytopenia (defined as platelet counts ≤ 100,000/μL) is negligible with trough total VPA levels up to 80 µg/mL. In this study, we found that with trough free VPA levels of up to 10 µg/mL, the likelihood of platelet counts dropping below 100,000/μL or 150,000/μL is very low. However, the probability of developing thrombocytopenia (when defined as platelets counts of ≤ 150,000/μL) gradually increases to reach 69% at trough free VPA levels above 30 µg/mL.
A recent study evaluated 51 inpatients (48% of whom were admitted to the intensive care unit) treated with VPA for the occurrence of thrombocytopenic episodes (defined as platelet counts < 140,000/μL) [
]. Out of a total of 98 trough free VPA levels, 27 (27.6%) occurences of thrombocytopenia were reported. A multivariate logistic regression model found that free VPA levels, baseline platelet counts, age, and bilirubin levels were significantly associated with the likelihood of developing a thrombocytopenic episode. By using a ROC curve, the authors reported that free VPA levels greater than 14.67 µg/mL had the best discriminating power for the occurrence of thrombocytopenia with a sensitivity of 48.1%, a specificity of 88.7% and an AUC of 0.77. Using the Youden index, we found that trough free VPA levels greater than 16.65 µg/mL had the greatest discriminating power for the occurrence of thrombocytopenia (defined as platelet counts < 150,000/μL) with a sensitivity of 89.3%, a specificity of 83.6% and an AUC of 0.926. Although those results are comparable, the higher accuracy in our study is due to the substantially higher number of free VPA samples, the wider range of trough free VPA levels (most of the free VPA values in the study by Tseng et al. ranged between 2.3 and 20 µg/mL) and the fact that patients with baseline thrombocytopenia (counts as low as 66,000/μL) prior to treatment with VPA were included in their study. In addition, the authors failed to include all significant variables in their model, most importantly gender.
It is well known that critically ill patients are at high risk of developing thrombocytopenia with an incidence ranging between 14–44% for patients admitted to the intensive care unit [
]. The causes of thrombocytopenia in these patients include sepsis, disseminated intravascular coagulation, major trauma, dilution (with massive transfusion), heparin-induced and drug-induced [
]. For those patients, we derived two formulas to calculate the expected platelet counts and the probability of a VPA induced thrombocytopenia. Our large sample size of 844 observations ensured that the calculated coefficients in these equations are very close to the population parameters and therefore generalizable to patients initiated on DVPX therapy [
Sample size guidelines for logistic regression from observational studies with large population: emphasis on the accuracy between statistics and parameters based on real life clinical data.
]. Those formulas can have practical implications, such as in determining the likely causality for a VPA induced thrombocytopenia or in adjusting the daily dose of VPA if the probability of a VPA induced thrombocytopenia is deemed to be elevated. Let us consider the example of a critically ill man who developed thrombocytopenia (platelet count of 92,000/μL) for whom the treating physician is considering discontinuing VPA. For that patient, assuming a trough free VPA level of 12 µg/mL and a baseline platelet count of 250,000/μL, the expected platelet count after exposure to VPA would be 210,00/μL and the probability of a VPA induced thrombocytopenia would be 1.9%, which will prompt the treating physician to look for other causes of thrombocytopenia. On the other hand, let us take the example of a woman with a baseline platelet count of 220,000/μL, a free valproate level of 22 µg/mL and a concomitant platelet count of 130,000/μL. The expected platelet count for that patient would be 153,000/μL and the probability that VPA was causative in dropping the platelet count below 150,000/μL would be 48.6%. In that instance, the treating physician might decide to lower the daily dose of VPA or to closely monitor the platelet count if such a level was required to maintain seizure control.
In this study, we did not assess for VPA induced platelet dysfunction that could also contribute to a bleeding diathesis [
]. Previous reports noted that VPA exposure was associated with abnormal platelet adhesion, collagen-induced platelet aggregation, and prothrombin consumption during in-vivo and in-vitro tests [
The main aim of this study was to generate a model that accurately predicts the likelihood of VPA induced thrombocytopenia. This was possible due to the double-blind prospective nature of the trial that included a large dataset with a wide range of trough free VPA level and baseline platelet counts measured at a centralized laboratory. The generated model that was used achieved excellent sensitivity and specificity reaching 87%−89% and 81%−84%, respectively.
Since it is based on the measured trough free VPA levels, this model is generalizable and can be applied to estimate the probability of developing thrombocytopenia in critically ill patients [
Sample size guidelines for logistic regression from observational studies with large population: emphasis on the accuracy between statistics and parameters based on real life clinical data.
00376-9&id=gr1.jpg){kind=link}
00376-9&id=gr2.jpg){kind=link}
00376-9&id=gr3.jpg){kind=link}