Cost-effectiveness of Oral Phenytoin, Intravenous Phenytoin, and Intravenous Fosphenytoin in the Emergency Department

Transcription

1 NEUROLOGY/ORIGINAL RESEARCH Cost-effectiveness of Oral Phenytoin, Intravenous Phenytoin, and Intravenous Fosphenytoin in the Emergency Department Maria I. Rudis, PharmD Daniel R. Touchette, PharmD Stuart P. Swadron, MD Amy P. Chiu, PharmD Michael Orlinsky, MD From the Department of Pharmacy, School of Pharmacy (Rudis, Chiu), and the Department of Emergency Medicine (Rudis, Swadron, Orlinsky), Keck School of Medicine, University of Southern California, Los Angeles, CA; and the Department of Pharmacy Practice, Oregon State University, Portland, OR (Touchette). Dr. Chiu is currently affiliated with the Department of Pharmacy, John Muir Medical Center, Walnut Creek, CA /$30.00 Copyright 2004 by the American College of Emergency Physicians. doi: / j.annemergmed See editorial, p Study objective: Oral phenytoin, intravenous phenytoin, and intravenous fosphenytoin are all commonly used for loading phenytoin in the emergency department (ED). The cost-effectiveness of each was compared for patients presenting with seizures and subtherapeutic phenytoin concentrations. Methods: A simple decision tree was developed to determine the treatment costs associated with each of 3 loading techniques. We determined effectiveness by comparing adverse event rates and by calculating the time to safe ED discharge. Time to safe ED discharge was defined as the time at which therapeutic concentrations of phenytoin ( 10 mg/l) were achieved with an absence of any adverse events that precluded discharge. The comparative cost-effectiveness of alternatives to oral phenytoin was determined by combining net costs and number of adverse events, expressed as cost per adverse events avoided. Cost-effectiveness was also determined by comparing the net costs of each loading technique required to achieve the time to safe ED discharge, expressed as cost per hour of ED time saved. The outcomes and costs were primarily derived from a prospective, randomized controlled trial, augmented by time-motion studies and alternate-cost sources. Costs included the cost of drugs, supplies, and personnel. Analyses were also performed in scenarios incorporating labor costs and savings from using a lower-urgency area of the ED. Results: The mean number of adverse events per patient for oral phenytoin, intravenous phenytoin, and intravenous fosphenytoin was 1.06, 1.93, and 2.13, respectively. Mean time to safe ED discharge in the 3 groups was 6.4 hours, 1.7 hours, and 1.3 hours. Cost per patient was $2.83, $21.16, and $175.19, respectively, and did not differ substantially in the Labor and Triage (lower-urgency area of ED) scenarios. When the measure of effectiveness was adverse events, oral phenytoin dominated intravenous phenytoin and intravenous fosphenytoin, with a lower cost and number of adverse events. With time to safe ED discharge as the outcome measure, the incremental cost-effectiveness ratios were $3.90 and $ per hour of ED time saved for oral phenytoin versus intravenous phenytoin and for intravenous fosphenytoin versus intravenous phenytoin, respectively. Conclusion: Oral phenytoin is the most cost-effective loading method in most settings. Intravenous phenytoin is preferred if one is willing to pay an additional $ ANNALS OF EMERGENCY MEDICINE 43:3 MARCH 2004

2 to $44.25 per patient and willing to have more adverse events for a quicker average time to safe ED discharge. It is unlikely that intravenous fosphenytoin is justifiable in any setting. [Ann Emerg Med. 2004;43: ] INTRODUCTION Capsule Summary What is already known on this topic The emergency department (ED) management of patients who have suffered a seizure and have subtherapeutic serum levels of phenytoin is common and time consuming. Fosphenytoin has been proposed as a safer and more efficacious preparation of phenytoin to use in these situations. What question this study addressed The authors analyzed the cost-effectiveness of oral phenytoin, intravenous phenytoin, and intravenous fosphenytoin in this scenario, accounting for a broad variety of costs. What this study adds to our knowledge Cost-effectiveness was determined by comparing the net costs of each loading technique required to achieve the time to safe discharge, combined with the cost per adverse event avoided. Costs were primarily derived from a randomized controlled trial, augmented by time-motion studies and alternate costing sources. Oral phenytoin was the most cost-effective loading method in most settings. How this might change clinical practice Intravenous fosphenytoin is unlikely to be cost-effective in any setting. Intravenous phenytoin decreases time to discharge but increases costs and adverse effects compared with oral dosing. Phenytoin remains one of the most commonly used antiepileptic medications in the emergency department (ED). For patients presenting to the ED with seizures and subtherapeutic serum concentrations of phenytoin, it is common practice to use a loading dose of phenytoin with the aim of achieving therapeutic serum phenytoin concentrations. Patients with subtherapeutic phenytoin concentrations are often noncompliant and consume significant and valuable health care resources. 1-3 Methods for loading phenytoin both orally and intravenously vary widely. 4-7 Each loading technique has significant disadvantages: the adverse events of intravenous loading, the potential delayed effectiveness of oral phenytoin, and the expense of fosphenytoin. In determining the most cost-effective, safe, and rapid method of preventing seizures and achieving therapeu- tic concentrations (10 to 20 mg/l),published data suggest that drug formulation and pharmacokinetic differences may indeed account for differences in adverse events and clinical efficacy There is a paucity of published data evaluating the cost-effectiveness of various methods of phenytoin loading in the ED. 7,12-14 The existing literature has used pharmacoeconomic modeling outside the scope of a clinical trial, 13,14 has overestimated costs attributable to adverse events, 12 or has not focused solely on the ED patient population. 13,14 Furthermore, none of these studies has included oral loading techniques in a costeffectiveness analysis. 7 The purpose of our study was to determine the cost-effectiveness of 3 methods of loading phenytoin in the ED. We determined effectiveness by comparing adverse event rates and by calculating the time to safe ED discharge. Time to safe ED discharge was defined as the time at which therapeutic concentrations of phenytoin ( 10 mg/l) were achieved with an absence of any adverse events that precluded discharge. The comparative cost-effectiveness of alternatives to oral phenytoin was determined by combining net costs and number of adverse events, expressed as cost per adverse event avoided. Cost-effectiveness was also determined by comparing the net costs of each loading technique required to achieve safe ED discharge, expressed as cost per hour of ED time saved. The data on effectiveness were reported in a previous article. 15 In this article, we report on the cost-effectiveness data. METHODS This analysis was based on our prospective, randomized, clinical trial comparing oral phenytoin (n=16), intravenous phenytoin (n=14), and intravenous fosphenytoin (n=15) in 45 patients admitted to the ED after seizure. Written informed consent was obtained from all patients. A detailed description of the clinical trial is published elsewhere. 15 In brief, patients were enrolled in the study if they were receiving maintenance phenytoin therapy for an existing seizure disorder and presented to the ED with a seizure in the past 48 hours and a serum phenytoin concentration less than 5 µg/ml. After their ED evaluation, patients were transferred to the General Clinical Research Center unit for phenytoin loading and subsequent observation before discharge. Patients were observed for a period of 24 hours from the time of initiation of study medication for drug administration, pharmacokinetic sampling at predetermined times, and observation of adverse MARCH :3 ANNALS OF EMERGENCY MEDICINE 387

3 events. Oral phenytoin was administered in increments of 400 mg every 2 hours until a dose of 20 mg/kg was reached. Dosing for the intravenous routes was 18 mg/kg for phenytoin and 18 mg/kg phenytoin equivalents for fosphenytoin. The infusion rates were initiated at 50 mg/min and 150 mg/min for each drug, respectively, and adjusted downward according to a preset protocol. Adverse events were also managed according to a standardized algorithmic approach (Table 1). To determine the cost-effectiveness, we used a simple decision tree to determine the total treatment costs associated with each of the 3 loading techniques (Figure 1). Outcomes considered important and included in the analysis were time to achieve therapeutic phenytoin concentrations, incidence of adverse events, the calculated ED length of stay, and cost of care. Seizure recurrence was also collected in the study but was not included in the pharmacoeconomic analysis because a significant difference was not observed in this outcome and none were attributed to subtherapeutic concentrations. The pharmacoeconomic analysis was performed from the perspective of the institution, a 750-bed, university-affiliated, county hospital in an urban setting. Sample size calculations were performed by the General Clinical Research Center Department of Biostatistics at our institution. These calculations were based on the characteristics required to conduct a costeffectiveness study. The pharmacoecomonic characteristics used in the determination of sample size included the cost of drug and other products used in the loading process, the costs of nursing and medical personnel time, and rates of adverse events as estimated from a Table 1. Adverse event treatment guidelines for intravenous phenytoin and intravenous fosphenytoin. Adverse Event Arrhythmia/ataxia/ disorientation/ dizziness/nystagmus Hypotension Pain/phlebitis Pruritus Treatment Guideline Stop intravenous infusion; restart infusion at slower rate; monitor; switch to oral regimen if necessary Administer 250 ml of normal saline solution; monitor Stop intravenous infusion; apply ice pack; administer 250 ml of normal saline solution; restart new peripheral line; switch to oral regimen if necessary Restart infusion at slower rate; monitor medical record review. According to the total estimated costs for each arm, the sample size was calculated to detect a minimum difference of 30% in cost between the 3 groups, with 80% power. A total sample size of 180, with 60 patients per group, was conservatively chosen on the basis of this analysis. An interim analysis was performed after the enrollment of the first 45 patients, at which time significant differences in costs were found among the 3 groups. The model was populated with data from several sources. All of the characteristics for calculation of the time to safe ED discharge were obtained directly from the clinical trial data, namely, the time at which therapeutic concentrations were achieved, the time the patient was free of adverse events precluding discharge, and, in the case of the intravenous arms, the time at which the infusion was complete. According to the institutional perspective, costs were defined as those incurred by the hospital only. Cost of care and salary data were obtained from hospital administrative databases and applied as outlined below. The cost of care included cost of drug administration and management of adverse events in the clinical trial. The cost of care was estimated by applying microcosting methods. The resources use for intravenous phenytoin and intravenous fosphenytoin administration was obtained through time and motion studies. The time and motion studies were performed by randomly selecting 6 nurses from the ED staff and timing their performance of starting an intravenous line, setting up the cardiac monitor and intravenous pump for drug administration, and preparing the medication for administration. A research assistant not involved in the care of the patient timed the procedures from the moment the nurse initiated the first activity to the time the drug infusion was started. To minimize the chance of introducing timing bias, 4 (66%) of the 6 timed events were randomly selected, and the mean time per event was calculated for purposes of estimating the cost per event. The mean nurse salary and benefits of $0.66 per minute was derived from a mean of 2,080 hours worked each year at an annual salary plus benefits of $75,900. The cost per adverse event was calculated by multiplying the time to adverse event resolution, measured in the clinical trial, by the nurse salary for that period. Physician time for adverse event management was similarly estimated with time and motion studies by applying a salary plus benefits cost of $87,377 ($0.70 per minute), 388 ANNALS OF EMERGENCY MEDICINE 43:3 MARCH 2004

4 Figure 1. Tree structure (mathematical model) used in the decision model. The tree structure was too large to demonstrate using a single figure and has been divided up into 4 sections labeled a, b, c, and d. The 3 therapies evaluated are presented after the decision node, indicated by a square, in section a. Each possible adverse event (eg, ataxia, disorientation) is presented on a tree branch after a chance node, shown as a circle on the tree. The probability of the adverse event occurring is indicated by a variable beneath the appropriate decision tree branch (eg, p_disorientation_pop for the probability of experiencing disorientation while on oral phenytoin). The complete tree is symmetrical beginning from the left of section a and moving to the right, through to sections b, c, and eventually d. The costs and outcomes were entered at the payoff node, indicated by a triangle at the end of the model in section d. An individual simulated patient can be followed through the tree by starting with the treatment choice in section a and moving through the tree (to the right). The patient will have the potential to develop adverse events based on the probabilities indicated at each chance node, which were derived from the observed adverse events in the clinical trial. At the end of the tree (in section d), the total cost for that simulated patient is calculated on the basis of the drug therapy received and adverse events experienced. The expected cost of each therapy, a weighted average of all simulated patients, was determined by multiplying through the probability of progressing down a particular branch by the payoff for that branch, to determine the weighted cost for that branch, and then adding all of these weighted branch costs together. PO, Oral; IV, intravenous. MARCH :3 ANNALS OF EMERGENCY MEDICINE 389

5 which was derived from salary information of the current mix of attending physicians and residents employed in our ED. Because oral phenytoin requires minimal preparation time and its adverse events are typically transient and do not require additional supplies or physicans orders, only nursing time was included in the analysis. For the base case (Base), physician and nursing time for drug preparation and adverse event management were omitted. The base case is likely the most accurate for our ED because the number of personnel is unlikely to be influenced by phenytoin loading strategy. The personnel costs to the institution remain constant, regardless of the time required to treat individual patients because seizure patients do not compose a large portion of the ED population. As a result, only direct medical variable costs were included in the Base analysis. To apply our results to different practice settings and different beliefs on how to value employee contribution, an alternative Labor analysis was developed. This Labor analysis takes into account the cost associated with physician and nursing time for drug administration and adverse event management. In this analysis, time saved by treating patients with medications that were either faster to administer or resulted in fewer adverse events allowed clinicians to focus on other duties that would be cost-saving to the institution, completely offsetting their salaries. The true costs to any health care system associated with treating postseizure patients lies somewhere between the 2 extremes of the Base and this Labor analysis. A second alternative, Triage (lower-urgency area of ED) analysis, was developed to examine the costs and effect of treating patients receiving oral phenytoin in a triage or minor care area, rather than having them assigned to a monitored bed. Because cardiac and intravenous infusion monitoring is not necessary when oral phenytoin is used, this method may be useful in a very busy ED by freeing a monitored bed, which would also allow the hospital to charge for having treated an additional patient in this setting. In this analysis, the Base costs were used to estimate the costs of drug administration and adverse events (ie, direct medical variable costs were used). However, the reimbursement associated with a county hospital visit ($300) was subtracted from the cost to allow for the increase in efficiency from having a free emergency bed for another patient. For all 3 scenarios (Base, Labor, and Triage), we reported the costs and outcomes separately. The costs and outcomes were then combined to create the following incremental cost-effectiveness ratios: cost per adverse event avoided and cost per ED hour saved. The incremental cost-effectiveness ratios were estimated with the following equation: (C 1 C 2 )/(E 2 E 1 ) where C 1 and C 2 are the costs of the 2 therapies being compared and E 1 and E 2 are either the number of adverse events or estimated ED length of stay. Sensitivity analyses are performed because there is always uncertainty in decision analysis. This uncertainty results from several sources. First, because the clinical and administrative research studies used in decision models sample from a larger population, there is variability around the mean values derived from these samples. Another source of variability is present in decision models because the studies and database information incorporated in the analysis differ in enrollment criteria, treatment administration, and study setting. Combining different studies and data from other sources into a single estimate of cost-effectiveness carries with it a considerable degree of uncertainty that must be addressed. A third source of uncertainty comes from the application of studies from a clinical trial to that of clinical practice. The decision analysis provides only singlepoint estimates (mean cost, mean effectiveness, and mean cost-effectiveness) without any measure of the variance or uncertainty associated with it. Sensitivity analyses are a collection of methods designed to provide insight into the effects of this uncertainty on the results of the model. Sensitivity analyses were performed to ensure that our results are applicable throughout a wide range of variability and costs in different practice environments. In our analysis, we used 2 methods of sensitivity analysis to the decision model to characterize this uncertainty: one-way sensitivity analysis and Monte Carlo simulation. One-way sensitivity analysis is a procedure in which each variable in the model is varied, one at a time, between 2 extreme but believable values. The model results are then recalculated with the new inputs and recorded. The process is repeated for all the variables in the model. Variables that result in a change in the preferred treatment option or that result in a substantial impact on the costs, effectiveness, or cost-effectiveness estimates are reported. Oneway sensitivity analyses alter only a single model input at a time and inform readers of which variables are most important in the model but do not provide information on the overall uncertainty associated with the model. We also used a second approach, Monte Carlo simulation, which is a method of varying some or all of the decision model variables simultaneously to assess the overall variability in the model. 16,17 With this method, 390 ANNALS OF EMERGENCY MEDICINE 43:3 MARCH 2004

6 and the results were plotted on a cost-effectiveness plane. Labor, drug, and supply costs were held constant. For example, the incidence of an adverse event was allowed to vary according to the binomial distribution from the clinical trial in which the estimate was obtained. The results of the Monte Carlo simulations were used to construct acceptability curves. Acceptability curves characterize the proportion of time one option is cost-effective for various thresholds of what is considered cost-effective, called a ceiling ratio. Acceptability curves use the concept of net benefits to reorganize the incremental cost-effectiveness ratio according to the following equation: NMB 1 NMB 2 = (E 1 E 2 ) (C 1 C 2 ) where NMB 1 and NMB 2 are the net benefits of therapies 1 and 2, respectively, E 1 and E 2 are the effects, C 1 and C 2 are the costs, and is the ceiling ratio. Rearranging the cost-effectiveness equation in this way overcomes many of the methodologic difficulties of estimating CIs surrounding a ratio and is therefore becoming a preferred method of handling uncertainty in cost-effectiveness analyses. This method has been described in greater detail by Briggs and Fenn 18 and Briggs et al. 19 All calcuthe distributions associated with model variables, usually obtained from a clinical trial or other database, are entered into a computer program. For a single Monte Carlo simulation run, the program uses the entered distributions to randomly select values for each variable and calculates the model results. This process is repeated a predetermined number of times, and the results are used to demonstrate the model variability. All of the sensitivity analyses presented in this study were performed on all 3 scenarios. One-way sensitivity analysis was performed on all variables included in the model. Adverse event rates and time to safe ED discharge were varied by their 95% confidence intervals (CIs), whereas cost inputs were varied by 50% to 200% of their Base values. In additional one-way sensitivity analyses, adverse event costs were treated as a single variable and varied by 50% to 200% of their Base values. We chose this arbitrary range because these values should incorporate the vast majority of believable values for costs. As mentioned previously, a Monte Carlo simulation was also conducted for each scenario. Adverse event rates and calculated time to safe ED discharge were altered according to distributions from the clinical trial, Table 2. Model inputs and range tested in one-way sensitivity analyses. Treatment Option Variable Oral Phenytoin Intravenous Phenytoin Intravenous Fosphenytoin Adverse event rates, % (range tested) Ataxia 25.0 ( ) 14.3 ( ) 0.0 ( ) Disorientation 12.5 ( ) 14.3 ( ) 13.3 ( ) Dizziness/headache 37.5 ( ) 28.6 ( ) 40.0 ( ) Hypotension NA 14.3 ( ) 6.7 ( ) Pruritus NA NA 80.0 ( ) Nausea/vomiting 12.5 ( ) 0.0 ( ) 26.7 ( ) Nystagmus 18.8 ( ) 42.9 ( ) 20.0 ( ) Phlebitis NA 78.6 ( ) 20.0 ( ) Tachycardia NA 0.0 ( ) 6.7 ( ) Differential costs * (range) Administering drug, labor costs excluded (Base) $2.83 ($ ) $19.28 ($ ) $ ($ ) Administering drug, labor costs included (scenario 1) $2.83 ($ ) $31.06 ($ ) $ ($ ) Mean time to safe ED discharge, h±sd ll 6.4± ± ±1.0 NA, Not applicable (adverse event would not be expected to occur in this group [attributable to diluent or drug-specific property]). * Differential costs are those costs that occur for at least 1 of the therapies and not for another. Costs that are similar between all 3 therapies were excluded from this analysis to maintain the model s simplicity. The drug and supply costs of administering oral phenytoin include the cost of the drug ($2.83). There were no differential labor costs involved. The drug and supply costs of administering intravenous phenytoin include the cost of the drug ($6.35), bag of normal saline solution ($0.66), and other supplies such as needles, syringes, gauze, and bandages ($12.93). The labor costs included drug preparation for infusion ($1.98), starting an intravenous line ($3.30), priming the intravenous pump ($3.30), and preparing the cardiac monitor ($2.54). The drug and supply costs of administering intravenous fosphenytoin include the cost of the drug ($162.26), bag of normal saline solution ($0.66), and other supplies such as needles, syringes, gauze, and bandages ($12.93). The labor costs included drug preparation for infusion ($1.98), starting an intravenous line ($3.30), priming the intravenous pump ($3.30), and preparing the cardiac monitor ($2.54). ll Not tested in one-way sensitivity analyses. Time to safe ED discharge for oral phenytoin was statistically different from that with intravenous phenytoin and intravenous fosphenytoin (P<.001). MARCH :3 ANNALS OF EMERGENCY MEDICINE 391

7 Table 3. Time to adverse event resolution with range tested in oneway sensitivity analyses and mean calculated costs associated with treating the adverse event (labor included). Mean Adverse Event Costs for the Labor Scenario Time to Adverse Event Resolution, Oral Intravenous Intravenous Adverse Event Min (Range) * Phenytoin Phenytoin Fosphenytoin Ataxia 35.0 ( ) $26.60 $28.10 $28.10 Disorientation 58.5 ( ) $42.11 $42.11 $42.11 Dizziness/ 63.4 ( ) $45.34 $45.34 $45.34 headache Hypotension 7.5 ( ) NA $9.95 $9.95 Pruritus 18.3 ( ) NA NA $14.18 Nausea/vomiting 18.5 ( ) $15.58 $17.21 $17.21 Nystagmus 45.9 ( ) $33.79 $35.29 $35.29 Phlebitis 19.9 ( ) NA $19.21 $19.21 Tachycardia 5.0 ( ) NA $6.80 $6.80 NA, Not applicable (adverse event would not be expected to occur in this group [attributable to diluents or a drug-specific property]). * Mean time to adverse event resolution obtained from clinical study. Range calculated from 95% CIs in the study. Adverse event costs included monitoring of the adverse event by a nurse for the entire period of the event at an average salary of $0.66 per minute, physician time involved in evaluating the adverse event as derived from a time and motion study (3 to 5 minutes per adverse event) at an average salary of $0.70 per minute, and drug and supply costs for managing an adverse event ($0.00 to $3.98 per adverse event). Adverse event occurred in only 1 patient; therefore, a range for the time to adverse event resolution for use in the sensitivity analyses could not be determined from 95% CIs and is arbitrary. lations were performed with DATA Professional Release 2 (TreeAge Software Inc., Williamstown, MA). RESULTS The model structure that was used is shown in Figure 1. The estimates that were used to populate the model are shown in Tables 2 and 3. The adverse event rates, total number of adverse events, and time to safe ED discharge for each of the therapies, estimated from the clinical trial, are shown in Tables 2, 3, and 4. Seizures occurred in 5 patients during the study; however, none was a result of the loading process. In 4 patients, seizures occurred several hours after the attainment of therapeutic phenytoin concentrations, and in 1 patient, a seizure occurred after enrollment but before the initiation of the study protocol. The results of the Base scenario (Table 4), a scenario in which labor costs were not included, showed that oral phenytoin was the least expensive option, costing $2.83 per patient. Intravenously administered phenytoin was the next least costly option, at $23.48 per patient, followed by intravenous fosphenytoin at $ per patient, for an incremental cost difference compared with oral phenytoin of $20.65 and $173.96, respectively. The Labor scenario (Table 4) included the labor costs associated with drug administration and adverse event Table 4. Predicted incremental cost-effectiveness of oral phenytoin, intravenous phenytoin, and intravenous fosphenytoin (model results). Outcome Scenario Oral Phenytoin Intravenous Phenytoin Intravenous Fosphenytoin No. of adverse events per person All Time to safe ED discharge All Total cost, $ Base (labor costs excluded) Total cost, $ Labor (labor costs included) Total cost, $ Triage (labor costs excluded; oral * phenytoin patients treated in triage) Cost per adverse event avoided All Dominates Dominated by oral Dominated by oral phenytoin phenytoin Cost per hour of ED time saved Base (labor costs excluded) Least costly $4.39 Compared with oral $ Compared with intraalternative phenytoin venous phenytoin; $34.11 compared with oral phenytoin Scenario 1: Labor (labor costs included) Least costly $9.41 Compared with oral $ Compared with intraalternative phenytoin venous phenytoin; $38.94 compared with oral phenytoin Scenario 2: Triage (labor costs excluded; Least costly $81.46 Compared with oral $ Compared with intraoral phenytoin patients treated in triage) alternative phenytoin venous phenytoin; $ compared with oral phenytoin * In the triage case, oral phenytoin would be expected to generate a revenue for the institution. The cost of giving oral phenytoin was estimated at $2.83, whereas the ability to admit another patient to an ED bed would generate a $300 charge, the estimated opportunity cost of treating patients in a monitored ED bed. A therapy dominates another therapy when it is more effective and less costly. 392 ANNALS OF EMERGENCY MEDICINE 43:3 MARCH 2004

8 management. In this scenario, oral phenytoin remained the least costly option, with an expected cost of $40.06 per patient. Intravenous phenytoin cost $84.31 per patient and an incremental cost of $44.25 compared with oral phenytoin. Intravenous fosphenytoin was again the most expensive agent in this analysis, with an expected mean cost of $ per patient and a difference of $ compared with oral phenytoin. The Triage scenario (Table 4) examined the cost differences associated with providing oral phenytoin in a triage or urgent care area. In this scenario, oral phenytoin resulted in a cost-savings to the system of $297.17, whereas intravenous phenytoin and fosphenytoin cost $23.48 and $176.79, with differences of $ and $473.96, respectively. Combining costs and adverse events rate in the Base scenario revealed that oral phenytoin was cheaper and caused less morbidity in the Base and both scenario analyses, dominating the other 2 therapies (Table 4). Likewise, intravenous fosphenytoin was dominated by intravenous phenytoin, although the differences in the total number of adverse events for these 2 therapies were not statistically significant in the original study. Combining costs and time to safe ED discharge resulted in incremental gains in effectiveness as the cost of the regimen increased. For the Base scenario, intravenous phenytoin had an incremental cost-effectiveness of $4.39 per hour of ED time saved compared with oral phenytoin. Intravenous fosphenytoin had an incremental cost-effectiveness of $ per hour of ED time saved compared with intravenous phenytoin. Compared with that of oral phenytoin, the incremental cost-effectiveness of intravenous fosphenytoin was $34.11 per hour of ED time saved. In the Labor scenario, the incremental cost-effectiveness of intravenous phenytoin compared with oral phenytoin was $9.41 per hour of ED time saved. The incremental cost-effectiveness of intravenous fosphenytoin compared with intravenous phenytoin was $ per hour of ED time saved and compared with oral phenytoin was $38.94 per hour of ED time saved. In the Triage scenario, the incremental cost-effectiveness of intravenous phenytoin compared with oral phenytoin was $81.46 per hour of ED time saved. The incremental cost-effectiveness estimates of intravenous fosphenytoin compared with intravenous phenytoin and oral phenytoin were $ and $ per hour of ED time saved, respectively. The impact of each model input on costs was examined by using one-way sensitivity analyses. The sensi- tivity analyses were performed on all model inputs, including adverse events rate estimates from the clinical trial, adverse event treatment and drug administration cost inputs, and time and salary inputs from the time and motion study. No single input resulted in a change in the preferred agent. Oral phenytoin always remained the least costly option, followed by intravenous phenytoin and then intravenous fosphenytoin. The effectiveness measure of the adverse event rate was not examined with one-way sensitivity analyses, because it was obtained directly from the clinical trial. In the clinical trial, there was a statistically significant difference in the total number of adverse events between oral phenytoin (17 adverse events in 16 patients) and the 2 intravenous regimens (intravenous phenytoin: 27 adverse events in 14 patients; intravenous fosphenytoin: 32 adverse events in 15 patients) in favor of the oral regimen (Table 2). No significant differences in the total number of adverse events were seen between intravenous phenytoin and intravenous fosphenytoin. With respect to calculated time to safe ED discharge, there was no difference observed between intravenous phenytoin (1.7 hours) and intravenous fosphenytoin (1.3 hours). However, there was a significantly reduced time to safe ED discharge when either of the 2 intravenous regimens was compared with oral phenytoin (6.4 hours; difference 4.7 hours versus intravenous phenytoin and 5.1 hours versus intravenous fosphenytoin). Because oral phenytoin dominated the decision model when costs and adverse events rate were considered, the Monte Carlo simulations and acceptability curves did not provide much additional insight into this measure of the cost-effectiveness and these results are not shown. The results of the Monte Carlo simulations are shown for the comparison of costs and time to safe ED discharge. In the Monte Carlo simulations, there was no discernable overlap in the cost estimates, regardless of the scenario (Base, Labor, or Triage). However, there was considerable overlap in the effectiveness measure: time to safe ED discharge. The results of the Base and Labor scenario Monte Carlo simulations are shown on a cost versus effectiveness graph in Figure 2. For our analysis, the ceiling ratio for each hour of ED time saved was the main outcome of interest and constitutes the x axis of the acceptability curves. For the Base scenario (Figure 3), a ceiling ratio of less than $2 per hour of ED time saved results in oral phenytoin being the preferred therapy more than 95% of the time. When the ceiling ratio is between $2 and $14 per hour of ED time saved, either oral phenytoin or intravenous phenytoin may be MARCH :3 ANNALS OF EMERGENCY MEDICINE 393

9 preferred, depending on patient- or institution-specific details. Beyond $14 per hour of ED time saved, intravenous phenytoin is preferred in more than 95% of all trials. Within the ceiling ratio range that we have shown, fosphenytoin did not become a preferred therapy. When the ceiling ratio was $100 per hour of ED time saved, fosphenytoin was preferred in 1.6% of the trials. A similar pattern develops for the Labor scenario, although there is more uncertainty associated with this model. Oral phenytoin is preferred in 95% of the trials or greater, up to a ceiling ratio value of approximately $1 per hour of ED time saved. Intravenous phenytoin becomes preferred more than 95% of the time beyond $30 per hour of ED time saved. The results are shown in Figure 3. The Triage scenario (results not shown) favors oral phenytoin more than the previous 2 scenarios. Oral phenytoin is preferred more than 95% of the time, up to a ceiling ratio of $34 per hour of ED time saved. Neither intravenous phenytoin nor intravenous fosphenytoin achieve preference greater than 95% of the time within the ceiling ratio range shown on the graph. LIMITATIONS There are several limitations to our study. The first is that we could not adequately quantify the incremental costeffectiveness of each therapy in preventing a seizure because the clinical study was not designed to detect differences in seizure rates. Given that we observed no seizures in patients with subtherapeutic concentrations, a large sample size would likely be needed. We are unaware of any other published studies that compare the relative seizure rates among the 3 loading techniques. Second, not factored into the triage scenario is the calculation of the incremental cost of a seizure in triage versus a seizure in the monitored area. This incremental cost is essentially the cost of transferring the patient into the monitored area. However, we believe that this cost is negligible. The third limitation is the fact that the time to safe ED discharge number does not take into account other inefficiencies and administrative duties associated with a normal ED encounter. Given the inherent inefficiencies of a large and busy ED, issues such as creating the patient s medical record, waiting time for room assignment and physician evaluation, and drug administration and reevaluation before discharge all may take a much longer time. In fact, in a recent quality assurance project, we found that the average time from patient presentation to the triage area to the start of phenytoin loading was more than 8 hours (n=27). By this measure, had all patients been treated before the availability of a bed in the monitored area, they may have been ready for discharge at the point at which they would otherwise only be starting the loading process. DISCUSSION Our cost-effectiveness analysis determined effectiveness by examining adverse events and time to safe ED Figure 2. Base case and Labor case costeffectiveness plane. This figure shows the results of the Monte Carlo simulation on a cost-effectiveness plane, giving an estimate of the variability in the model. Each dot represents one of the Monte Carlo runs. There is little or no overlap between the costs of therapy. However, there is considerable overlap between intravenous phenytoin and intravenous fosphenytoin in time to safe ED discharge. 394 ANNALS OF EMERGENCY MEDICINE 43:3 MARCH 2004

10 discharge. Time to safe ED discharge was, in turn, determined by the time to a therapeutic level and absence of any adverse events that required ED care. We performed the cost-effectiveness analysis from the perspective of a health care system and, more specifically, from the perspective of a county health care system. As such, our costs may not reflect the actual costs in other systems. However, the Base analysis was robust to sensitivity analyses on costs, indicating that although the estimates of expected costs and cost-effectiveness may differ under different settings, the choice of preferred Figure 3. Acceptability curve for the Base scenario (cost of labor excluded). The ceiling ratio in this graph refers to the maximum cost that the institution is prepared to pay to invest to discharge a patient one hour earlier from the ED. The proportion of trials from the Monte Carlo simulation in which intravenous phenytoin is the preferred agent over oral phenytoin is plotted against the ceiling ratio. Intravenous phenytoin rapidly becomes the preferred agent as the ceiling ratio changes from $2 (where oral phenytoin is preferred >95% of the time) to $18 (where intravenous phenytoin is preferred >95% of the time). Intravenous fosphenytoin does not become a favorable therapy within this ceiling ratio range. In the Labor scenario (cost of labor included), oral phenytoin is the preferred drug in >95% of the trials to an approximate ceiling ratio of $4 per hour of ED time saved. Intravenous phenytoin is preferred >95% when the willingness to pay is $46 per hour of ED time saved. Intravenous fosphenytoin does not become a favorable therapy within this ceiling ratio. A similar pattern was observed in the Triage scenario (results not shown). For all 3 scenarios (Base, Labor, and Triage), oral phenytoin was the preferred agent if the institution was less interested in investing money to discharge patients quicker from the ED, which may be appropriate for many patients. In certain circumstances or for certain patients, it may be more desirable to pay a greater amount to discharge patients from the ED earlier. In these situations, intravenous phenytoin would likely be the preferable therapy. Proportion of trials Ceiling ratio (cost per h of ED time) Preferred option PO Phenytoin Base case IV Phenytoin Base case IV Fosphenytoin Base case PO Phenytoin Labor case IV Phenytoin Labor case IV Fosphenytoin Labor case agent is not greatly affected. We chose not to do the analysis from a societal perspective, a common recommendation in pharmacoeconomic guidelines. 20,21 Societal costs would be extremely difficult to obtain in this study setting and would add little information for our intended audience: those within a health care system charged with selecting agents for use in the ED. We deliberately chose our scenarios to represent current costing practices. Our base scenario, by exclusion of labor costs, reflects current pharmacoeconomic methods. Labor costs were excluded regardless of which therapy was administered. Thus, there is no cost benefit to the institution for using a medication that is easier to administer or has fewer adverse events. The recently renewed interest in employee productivity as a component of cost-of-illness analyses led us to develop the Labor analysis in our study. 22 Theoretically, these time savings could be applied to other work that would benefit the institution, possibly generating revenue to offset the labor costs. In our Triage analysis, we examined the impact of using a lower-urgency area of the ED for oral loading, allowing for even further improvement in resource use. Our cost-effectiveness study extends the findings from other published data comparing the pharmacoeconomics of phenytoin and fosphenytoin in the ED by including an oral arm and by performing advanced sensitivity analyses. 7,12-14 Three studies have been performed with adult patients. 7,12,13 An additional study performed in a pediatric population will not be discussed in this article. 14 Two studies in adult patients suggested that fosphenytoin was more cost-effective than phenytoin for use in the ED, 12,13 whereas the other concluded that intravenous phenytoin was more cost-effective. 7 The incidence of adverse events in our study was lower than that reported by Marchetti et al 12 and Armstrong et al 13 and higher than that documented by Touchette and Rhoney. 7 In addition, the rate of adverse events in the intravenous fosphenytoin group was higher than was reported previously. Differences may have occurred as a result of variations in study methodologies. Marchetti et al 12 used clinical results from one multicenter trial and abstracted financial data from a different institution. 12 The rate of adverse events to intravenous phenytoin was high in this study, and the costs attributed to the management of adverse events were overestimated. The authors assigned a length of time to resolution of these adverse events according to a survey rather than quantifying real time to resolution. They also assumed that all adverse events resulted in a MARCH :3 ANNALS OF EMERGENCY MEDICINE 395

11 prolongation of length of stay in the ED. In clinical practice, most adverse events caused by intravenous phenytoin are usually managed during the intravenous infusion and do not result in an increased length of stay in the ED. 23 In addition, the attempted blinding in this study may have been unsuccessful because of the higher infusion rate of fosphenytoin, potentially causing a bias against documentation of adverse events to fosphenytoin. Furthermore, the investigators used average fixed and variable costs to calculate the cost of care for ED treatment instead of using only variable or marginal costs of care, as is recommended. 24 Our results, together with those of Touchette and Rhoney, 7 demonstrate the effect of using variable costs on the final decision of the analysis. In their analysis, Armstrong et al 13 used a questionnaire administered to nurses from critical care, emergency, and neurology services in 3 local hospitals to obtain effectiveness data, frequency of adverse events, and methods of treating the reactions, which may have resulted in an overestimation of rate of adverse events and may have unduly affected the analysis because expert opinion overestimates time and costs of managing adverse events. 25,26 In our study, we quantified the actual time required for clinical care and management of adverse events by performing a time-motion study, which is a preferred costing method. 24 In contrast to the 2 aforementioned studies, that of Touchette and Rhoney 7 found a lower rate of adverse events. Again, this difference may be explained by differences in the methods of quantifying adverse events. Our study used an initial infusion rate of 50 mg/min, with downward adjustment for adverse events according to a standardized algorithmic approach. The study on which Touchette and Rhoney based their analysis initiated infusion rates at 20 mg/min. 23 That study quantified the rate of adverse events through retrospective medical record review and was not designed to evaluate delayed adverse events. 23 In our study, the presence of adverse events was assessed not only during the infusion but also at predetermined intervals thereafter for a 24-hour period by specialized nurses in a research unit. This method may have overestimated the rate of adverse events requiring intervention and, hence, may have overestimated costs. Despite these differences, our conclusion that phenytoin is more costeffective for intravenous loading is the same. 7 Our pharmacoeconomic analysis suggests that oral phenytoin is the least expensive method of loading phenytoin and results in the fewest number of adverse events. However, if rapid discharge from the ED is desired, it may be preferable to use intravenous phenytoin at an estimated additional average cost of $20.65 (Base) or $44.25 (Labor) and with an estimated mean time savings of 4.7 hours per patient. Several patient factors, such as occupation, income, previous adverse events with phenytoin loading in the ED, and frequency of seizures are likely to influence their preference of loading technique. For patients who are not employed, the inconvenience of remaining longer in the ED for oral loading may be minor compared with the discomfort of adverse events associated with intravenous loading. For the patient who experiences frequent or severe seizures when subtherapeutic, an intravenous route may be preferable. Cost-effectiveness analyses typically focus on generalizable outcomes such as years of life saved or qualityadjusted life-years saved. We did not expect to see differences in quality of life between the 3 methods of phenytoin administration and are unaware of any studies demonstrating such differences. Therefore, we evaluated the 3 regimens by using 2 outcome measures of incremental cost-effectiveness: cost per adverse events avoided and cost per hour of ED time saved. With the cost per adverse event avoided measure, oral phenytoin dominated both intravenous methods of administration. Intravenous phenytoin also resulted in fewer adverse events and a lower cost compared with intravenous fosphenytoin. Only one identified study, by Armstrong et al, 13 estimated the cost-effectiveness of phenytoin by using cost per adverse event avoided. This study estimated that intravenous fosphenytoin use resulted in an incremental cost-effectiveness of $96.91 per adverse event avoided. Other studies have neither found differences in adverse events rates, making incremental analysis impractical, 7 nor performed incremental analysis. 12 One limitation of using these nongeneralizable outcomes is that an institution s consideration of what is cost-effective for a reduction in adverse events or early discharge from the ED has not been quantified. We attempted to overcome this limitation by using acceptability curves. 18,19,27 The acceptability curves demonstrate the likelihood that one therapy is preferred over another at various levels of cost-effectiveness. Decisionmakers within institutions can plot their own beliefs of what is considered cost-effective on this graph and determine the optimal therapy and certainty with which it is preferred. However, there are several caveats to acceptability curves. First, the institution may not know or be able to ascertain their beliefs on what is considered cost-effective (ie, the ceiling ratio) for early ED 396 ANNALS OF EMERGENCY MEDICINE 43:3 MARCH 2004

12 discharge. Second, the ceiling ratio for ED time saved may actually vary considerably among institutions, depending on the institutions workload, finances, and other factors. In an ED with a high workload, the ceiling ratio may be significantly higher for ED time saved. It may also be higher in a competitive market, where time to ED discharge may be a measure of work efficiency and a marketing tool to the public. As a result of our study, we now recommend in our ED that oral phenytoin loading be considered in cases in which patients do not otherwise require monitoring. In many cases, the oral phenytoin loading process can be performed in triage, freeing monitored beds for other patients. With respect to fosphenytoin, we continue to restrict its use to status epilepticus or cases in which the time required to load must be kept to the absolute minimum (eg, patient going immediately to the operating room). In summary, our study demonstrated that oral phenytoin is the most cost-effective method of loading phenytoin in the ED in most circumstances. Intravenous phenytoin is preferred if one is willing to pay and willing to have more adverse events because of a quicker time to safe ED discharge. It is unlikely that intravenous fosphenytoin use is justifiable in any setting. The additional cost of ED time saved by using the faster intravenous routes would cost an additional $4.39 per hour and $ per hour for intravenous phenytoin and intravenous fosphenytoin, respectively. We are indebted to the faculty, nurses, and residents of the Department of Emergency Medicine at Los Angeles County+University of Southern California Medical Center and the staff of the General Clinical Research Center; to Gregory Garcia, MD, Christian Zimmerman, MD, and John Murray, MD, who helped with patient recruitment and data collection; and to Jorge Avila who provided quality assurance data discussed in the manuscript. Author contributions: MIR, DRT, SPS, APC, and MO collaborated on all aspects of the study, from the initial hypotheses through the pharmacoeconomic analysis and the preparation of the manuscript. APC was also involved in carrying out the time-motion component of the study. DRT carried out the modeling for the cost-effectiveness analysis and performed the statistics. MIR takes responsibility for the paper as a whole. Received for publication January 9, Revisions received June 27, 2003, and October 2, Accepted for publication October 8, Presented as an abstract at the American College of Emergency Physicians Research Forum, October 2002, Seattle, WA. Supported by grants from the National Institutes of Health General Clinical Research Center (grant M01 RR-43), the University of Southern California Health Research Association, the University of Southern California Medical Faculty Women s Association Fund, and the University of Southern California School of Pharmacy Drug Development Research Center. Address for reprints: Maria I. Rudis, PharmD, Department of Pharmacy, University of Southern California School of Pharmacy, 1985 Zonal Avenue, PSC 700, Los Angeles, CA 90033; , fax ; rudis@usc.edu. REFERENCES 1. Krumholz A, Grufferman S, Orr ST, et al. Seizures and seizure care in an emergency department. Epilepsia. 1989;30: Brokaw J, Olson L, Fullerton L, et al. Repeated ambulance use by patients with acute alcohol intoxication, seizure disorder, and respiratory illness. Am J Emerg Med. 1998;16: Simonson HC, Pelberg AL. Unnecessary emergency transport and care of grand mal seizures. Am J Med Qual. 1993;8: Osborn HH, Zisfein J, Sparano R. Single-dose oral phenytoin loading. Ann Emerg Med. 1987;16: Dela Cruz FG, Kanter MZ, Fischer JH, et al. Efficacy of individualized phenytoin sodium loading doses administered by intravenous infusion. Clin Pharm. 1988;7: Johnson J, Wrenn K. Inappropriate fosphenytoin use in the ED. Am J Emerg Med. 2001;19: Touchette DR, Rhoney DH. Cost-minimization analysis of phenytoin and fosphenytoin in the emergency department. Pharmacotherapy. 2000;20: Meyer MC, Straughn AB. Biopharmaceutical factors in seizure control and drug toxicity. Am J Hosp Pharm. 1993;50(12 Suppl 5):S Fierro LS, Savulich DH, Benezra DA. Safety of fosphenytoin sodium. Am J Health Syst Pharm. 1996;53: Earnest MP, Marx JA, Drury LR. Complications of intravenous phenytoin for acute treatment of seizures. Recommendations for usage. JAMA 1983;249: Bourgeois BF. Pharmacokinetic properties of current antiepileptic drugs: what improvements are needed? Neurology. 2000;55:S Marchetti A, Magar R, Fischer J, et al. A pharmacoeconomic evaluation of intravenous fosphenytoin (Cerebyx) versus intravenous phenytoin (Dilantin) in hospital emergency departments. Clin Ther. 1996;18: Armstrong EP, Sauer KA, Downey MJ. Phenytoin and fosphenytoin: a model of cost and clinical outcomes. Pharmacotherapy. 1999;19: Graves N. Pharmacoeconomic considerations in treatment options for acute seizures. J Child Neurol. 1998;13(Suppl 1):S Swadron SP, Rudis MI, Aziman K, et al. A comparison of phenytoin-loading techniques in the emergency department. Acad Emerg Med. 2004;11: Doubilet P, Begg CB, Weinstein MC, et al. Probabilistic sensitivity analysis using Monte Carlo simulation: a practical approach. Med Decis Making. 1985;5: Critchfield GC, Willard KE. Probabilistic analysis of decision trees using Monte Carlo simulation. Med Decis Making. 1986;6: Briggs A, Fenn P. Confidence intervals or surfaces? Uncertainty on the cost-effectiveness plane. Health Econ. 1998;7: Briggs AH, O Brien BJ, Blackhouse G. Thinking outside the box: recent advances in the analysis and presentation of uncertainty in cost-effectiveness studies. Annu Rev Public Health. 2002;23: Siegel JE, Weinstein MC, Russell LB, et al. Recommendations for reporting costeffectiveness analyses: Panel on Cost-Effectiveness in Health and Medicine. JAMA. 1996;276: Hjelmgren J, Berggren F, Andersson F. Health economic guidelines: similarities, differences and some implications. Value Health. 2001;4: Osterhaus JT, Gutterman DL, Plachetka JR. Healthcare resource and lost labour costs of migraine headache in the US. Pharmacoeconomics. 1992;2: Coplin WM, Rhoney DH, Rebuck JA, et al. Randomized evaluation of adverse events and length-of-stay with routine emergency department use of phenytoin or fosphenytoin. Neurol Res. 2002;24: Weinstein MC, Siegel JE, Gold MR, et al. Recommendations of the Panel on Costeffectiveness in Health and Medicine. JAMA. 1996;276: Evans C. The use of consensus methods and expert panels in pharmacoeconomic studies; practical applications and methodological shortcomings. Pharmacoeconomics. 1997;12(2 pt 1): Evans C, Crawford B. Expert judgement in pharmacoeconomic studies: guidance and future use. Pharmacoeconomics. 2000;17: Willan AR, O Brien BJ, Leyva RA. Cost-effectiveness analysis when the WTA is greater than the WTP. Stat Med. 2001;20: MARCH :3 ANNALS OF EMERGENCY MEDICINE 397