Initial and Long-Term Costs of Patients Hospitalized with West Nile Virus Disease

. There are no published data on the economic burden for specific West Nile virus (WNV) clinical syndromes (i.e., fever, meningitis, encephalitis, and acute flaccid paralysis [AFP]). We estimated initial hospital and lost-productivity costs from 80 patients hospitalized with WNV disease in Colorado during 2003; 38 of these patients were followed for 5 years to determine long-term medical and lost-productivity costs. Initial costs were highest for patients with AFP (median $25,117; range $5,385 – $283,381) and encephalitis (median $20,105; range $3,965 – $324,167). Long-term costs were highest for patients with AFP (median $22,628; range $624 – $439,945) and meningitis (median $10,556; range $0 – $260,748). Extrapolating from this small cohort to national surveillance data, we estimated the total cumulative costs of reported WNV hospitalized cases from 1999 through 2012 to be $778 million (95% confidence interval $673 million – $1.01 billion). These estimates can be used in assessing the cost-effectiveness of interventions to prevent WNV disease.


INTRODUCTION
West Nile virus (WNV), a mosquito-borne flavivirus, was first detected in the Western Hemisphere in 1999. Since then, it has become the leading cause of domestically acquired arboviral disease in the United States and is responsible for seasonal outbreaks of disease affecting all regions of the country. 1 Approximately 80% of WNV infections are asymptomatic. 2 Most symptomatic persons experience an acute systemic febrile illness, known as West Nile fever or non-neuroinvasive disease, that often includes headache, myalgia, arthralgia, or rash. 3,4 Less than 1% of infected persons develop neuroinvasive disease, which typically manifests as meningitis, encephalitis, or acute flaccid paralysis (AFP). 5 Most patients with WNV meningitis or non-neuroinvasive disease recover completely, but fatigue and malaise can linger for weeks or months. 6 -8 Patients who recover from WNV encephalitis or AFP often have residual neurologic deficits. 9 Among patients with neuroinvasive disease, the overall case-fatality ratio is~10%, but it is notably higher for patients with WNV encephalitis and AFP. 1 From 1999 through 2012, over 37,000 WNV disease cases including 1,500 deaths were reported to the Centers for Disease Control and Prevention (CDC). 1,[10][11][12] Despite the large number of cases over the last 14 years, only two studies have estimated the initial cost of WNV disease. 13,14 There are no published data on the economic burden of the specific clinical syndromes seen with WNV infections or the longer term costs of WNV disease incurred several years after the initial illness. We present data on both initial hospital costs and 5 years of follow-up care costs for persons hospitalized with different clinical presentations of WNV disease. We then use the data to estimate the total costs of hospitalized WNV disease cases and deaths reported to CDC. These estimates can be used in assessing the cost-effectiveness of various interventions designed to lower WNV disease risk.

METHODS
The analysis is divided into two parts. First, we estimate the direct and indirect medical costs of initial hospitalization and 5-year follow-up costs by WNV clinical syndrome among a cohort of hospitalized patients in Colorado. Second, we build a Monte Carlo simulation model using the findings from the cohort to estimate the costs incurred by all hospitalized WNV cases in the United States that were reported to CDC for 1999-2012.
For all cost estimates, we took a societal perspective (i.e., relevant costs regardless of who paid). We adjusted all costs to 2012 U.S. dollars (USD) using the prescription drugs part of the U.S. Consumer Price Index (CPI) for drug costs, the medical care part of CPI for medical costs, and the CPI inflation for productivity costs from the Bureau of Labor Statistics (http://www.bls.gov/). We used the productivity tables from Grosse and others 15 to derive the indirect (i.e., lost productivity) costs. Where appropriate, we discounted future costs at an annual rate of 3%.
Direct and indirect medical costs of WNV patient cohort. Study participants and data collection. In 2003, we enrolled a cohort of patients (original Colorado cohort) hospitalized with acute WNV disease at 17 institutions in northern Colorado (N = 221). 16 A participant's WNV clinical syndrome was classified as fever (i.e., non-neuroinvasive disease), meningitis, or encephalitis, with or without AFP based on their presenting signs and symptoms and a follow-up assessment completed by a neurologist (JJS). If a participant was diagnosed with 1 clinical syndrome, we classified them as having the most severe presentation for the purpose of this report (severity: fever meningitis encephalitis AFP).
A subset of survivors was followed prospectively to assess functional and neurocognitive outcomes. 17 Inpatient costs for the initial hospitalization were obtained from the respective hospitals (acute follow-up cohort). In 2008, we administered a questionnaire regarding outpatient medical and home care costs incurred during the 5 years following their initial hospitalization (long-term follow-up cohort). Participants were asked if after their initial hospitalization, they required any of the following related to their WNV illness: 1) subsequent inpatient hospitalization, rehabilitation, nursing home, assisted living, or home health care; 2) outpatient primary care, neurology, physical therapy, occupational therapy, or speech therapy visits; 3) new medications, equipment, or modifications of home or vehicle; or 4) missed work or school. The study was approved by the CDC Institutional Review Board and participants provided informed consent.
Cost data. A. Initial costs. We did not collect any outpatient costs associated with hospital-based acute care. 2. Indirect costs. We assumed for hospitalized patients who missed work that they had a work schedule of 5 out of every 7 days. We valued time lost from work, by age and sex, using estimates from Grosse and others. 15 If a participant worked part-time, 50% daily production value was used. The lost productivity for persons who died during their initial hospitalization was derived from the lifetime production value discounted at 3% their age and sex. 15 We did not assign any value to the days spent in hospital for those who died during their initial hospitalization. B. Costs incurred in the 5 years after initial hospitalization. 1. Medical appointments and institutional care costs. To determine medical appointment and institutional care costs, we used data provided by participants in the long-term followup cohort regarding the type and number of outpatient medical and home care appointments they had during the 5 years after their initial hospitalization and multiplied this by type-specific cost estimates obtained from MarketScan Commercial Claims and Encounter databases (Truven Health Analytics, Sacramento, CA). The MarketScan databases are compiled from insurance claims filed by patients employed by~100 companies; they include information on persons 65 years of age and their dependents and covered nearly 7 million lives in 2003. The databases contain both out-of-pocket payments and insurance reimbursements. Current Procedural Terminology, Fourth edition (CPT-4) service-based codes were used to extract cost data from MarketScan for medical appointments and institutional care ( Table 1). Because all the costs were positively skewed and had a high variance, we used the median costs for the United States in the calculations. We excluded any future costs for the three participants who indicated, at the time of the survey, they were still receiving care related to their WNV illness from a neurologist. 2. Drug costs. We used the average wholesale price of drugs from MicroMedex 2.0 (Truven Health Analytics Red Book online) to determine the cost of drugs that survey participants indicated that they received post-hospitalization. Unless the dosage was indicated by the participant, we assumed they received a standard dose. For as needed (prn) medicines, we assumed a normal or maximum amount for the first 3 months; the frequency was then gradually decreased over time to once a week if they were still taking the medicine 3 years after their initial hospitalization. We excluded future costs for the four partici-pants who indicated they were still taking a medicine at the time of the survey. 3. Durable medical equipment. We used the CMS 2003 floor fee schedule (https://www.cms.gov/DMEPOSFeeSched) to determine durable medical equipment costs for participants in the long-term follow-up cohort who indicated on the questionnaire that they needed specific equipment after their hospital discharge. If a specific model was not provided by the participant, the equipment was assumed to be a new basic model with the least cost. 4. Indirect costs. We used the same methodology as described previously to assign a cost for the number of days of missed employment after their initial hospitalization for participants in the long-term follow-up cohort who were working at the time of their illness. For persons who retired early as a result of WNV, we valued their indirect costs as the number of potential years and months of lost employment (65 minus age at early retirement). 15 For persons who indicated they were retired or unemployed at the time of their illness, the survey tool did not capture the number of days they were unable to perform daily household tasks. We did not, therefore, value any such losses.
Data analysis. Data were analyzed using SAS statistical software version 9.3 (SAS Institute, Cary, NC) or EpiInfo 7 (CDC, Atlanta, GA) to determine if there were differences between demographic and cost data between clinical syndromes. To assess representativeness of the follow-up cohorts, their demographics, clinical syndrome, and clinical outcomes were compared with the original Colorado cohort. Though the data came from a comprehensive cohort, we compared categorical variables using the Fisher's exact or χ 2 test and continuous variables using the Kruskal-Wallis test when the variance was not homogeneous (e.g., cost data) or analysis of variance when the variance was normally distributed (e.g., age). We did this to determine if there was sufficient difference in the variability of costs by syndrome and to inform if syndrome-specific costs were needed when extrapolating such cost to the national level (assuming such costs were representative at such level). The overall critical P value was chosen to be 0.05; however, when there were multiple tests performed, the P value was adjusted using the Bonferroni correction. 18 Cost of hospitalized WNV cases and deaths in the United States. We compared the background characteristics of patients from the follow-up cohorts to those reported nationally from ArboNET, the national arboviral surveillance system. 1 We extrapolated cost data collected from the follow-up cohorts described previously to the cumulative hospitalized WNV disease cases in the United States from 1999 through 2012 reported to CDC through ArboNET. We built a spreadsheet-based Monte Carlo simulation model using @Risk 5.7 software (Palisade Corporation, Ithaca, NY) to achieve this goal. The model used the following formula to determine the costs of WNV hospitalized cases: Total costs of hospitalized cases from 1999 to 2012 = Σ (number of hospitalized cases by year and syndrome X cost distribution by syndrome from the Colorado cohort).
For acute costs, the number of hospitalized cases and deaths by year and clinical syndrome were obtained from ArboNET. Because hospitalization was only first reported to ArboNET in 2004, cumulative proportions for 2004-2012 were used to estimate the number of cases hospitalized for 1999-2003 by clinical syndrome. To approximate the proportion of hospitalized cases in the United States who would incur long-term costs, we used a uniform distribution with a minimum of zero and a maximum equivalent to the proportion of cases in the Colorado cohort who experienced any long-term costs by clinical syndrome.
We applied the "fit distribution to data" function of @Risk software to the costs incurred by patients in the follow-up cohorts to generate the cost distribution for the model. The best distribution that fit the participant's data was selected from those provided by @Risk software based on relative frequencies, quantile-quantile graphs, χ 2 , Kolmogorov-Smirnov, and Anderson-Darling goodness-of-fit tests ( Table 2). The distributions were assigned a lower bound of zero because all costs were positive. We ran the model for 10,000 iterations and results are presented as 1) mean (mean of 10,000 iterations) with 95% confidence interval (CI) (2.5th and 97.5th percentiles of 10,000 iterations), and 2) median with range.
All deaths reported to ArboNET were assigned a lifetime productivity loss (discounted at 3%) based on the patient's age and sex. 15 For case-patients who died and where the age or sex was missing, we assumed the person to be 80 years of age and/or female because this provided the smallest estimate of lifetime productivity loss.

RESULTS
Demographic and clinical characteristics. Initial hospital cost data incurred in 2003 were obtained for 80 WNV disease cases in the acute follow-up cohort. When the demographic features, clinical syndromes, and outcomes were compared between the 80 patients in the acute follow-up cohort and those in the original Colorado cohort who were not followed prospectively (N = 141), there were no differences noted for *Medical is defined as the initial inpatient hospital costs for acute costs and as subsequent medical care costs (e.g., medical appointments, rehabilitation, medicines, durable medical equipment) for long-term costs. Productivity is defined as the value of time lost from work, by age and sex, using estimates from Grosse and others 15 assuming a 5-of 7-day work schedule.
†Cost data from the Colorado cohort were fitted using "fit distribution to data" function of @Risk software. From the suggested distributions of @Risk, the best distribution for each cost type was selected based on relative frequencies, quantile-quantile graphs, χ 2 , Kolmogorov-Smirnov, and Anderson-Darling goodness-of-fit (see text for further details).
‡ μ (σ) refer to the Mean (SD) of a logistic normal, α (β) refer to the alpha and beta of a Gamma, Weibull, or Pearson5 distribution, and β refers to the beta of an exponential distribution. AFP = acute flaccid paralysis; LT = long-term. age, sex, number of days hospitalized, and number of deaths between the groups. The acute follow-up cohort was less likely to include patients with meningitis (24%) than the original Colorado cohort (48%) (P 0.01) and more likely to include patients who were originally discharged to a chronic care or rehabilitation facility, or needing home health assistance (51% versus 31%) (P 0.01). The only difference that was observed when the acute follow-up cohort was compared with the original cohort by clinical syndrome was that patients with AFP in the acute follow-up cohort were more likely to be younger (median age 55 years, interquartile range [IQR] 44-62 years) than those who were not followed prospectively (median age 69 years, IQR 52-80 years).
In the acute follow-up cohort (N = 80), 41 (51%) were male and their median age at hospital admission was 55 years (IQR 45-66 years) ( Table 3). Twenty-two (28%) were 65 years of age or older. Case-patients were most commonly diagnosed as having AFP (34%), followed by meningitis (24%), fever (22%), and encephalitis (20%). There was no difference in the sex distribution of participants between various clinical presentations. However, median age at admission varied significantly by clinical syndrome, with encephalitis and fever patients being older than AFP and meningitis patients (P 0.001). Nine (50%) of the participants with fever were 65 years of age or older compared with 7 (44%) with encephalitis, 5 (19%) with AFP, and 1 (5%) with meningitis. Participants were hospitalized for a median of 5 days (IQR 3-11 days). The length of hospitalization was significantly different depending on the person's WNV clinical syndrome with AFP case-patients requiring longer hospitalization (P 0.001).
Six (8%) case-patients died during the initial hospitalization for WNV disease. Two of the case-patients were diagnosed with AFP and four had encephalitis; none of the case-patients with fever or meningitis died during their initial hospitalization. Case-patients who died were older (median 76 years) than survivors (median 53 years) (P 0.01). Duration of hospitalization was similar for fatal (median 7 days) and non-fatal cases (median 4 days) (P = 0.2).
Of the 80 participants in the acute follow-up cohort, 38 (48%) completed the follow-up survey of their subsequent direct and indirect medical care incurred 5 years after their hospitalization. When compared with those for whom no follow-up data were obtained (N = 141), participants in the long-term follow-up cohort were similar with regards to sex distribution, median age at hospital admission, length of hospitalization, proportions discharged to home versus a rehabilitation facility, and proportions with each clinical syndrome. The only difference was in the rates of death, where 12 (5%) of the 141 without long-term data died while no one in long-term cohort died. Furthermore, there were no differences in demographic and clinical outcomes by clinical syndrome between persons in the long-term cohort versus those with no follow-up data. When compared with those for whom only acute follow-up data were obtained (N = 42), participants in the long-term follow-up cohort (N = 38) were similar with regards to sex distribution, median age and the proportion of persons 65 years of age or older at hospital admission, and proportions with each clinical syndrome (Table 3). However, the 38 patients who provided long-term cost data had shorter hospitalizations (median 4 days; IQR 2-6 days) than the 42 patients for whom long-term costs were not available (median 7 days; IQR 4-16 days) (P 0.01).
Initial direct or indirect medical costs. Overall, persons with AFP or encephalitis had significantly higher total initial costs (i.e., initial hospital and lost productivity costs) than persons with meningitis or fever (P 0.001) ( Table 4). The median cost for a patient with AFP was $20,774 (IQR $10,749-$120,945) compared with a median cost of $4,467 (IQR $3,241-$8,433) for a patient with WNV non-neuroinvasive disease. The majority of the total costs were incurred as a result of hospital charges. However, lost productivity was also significantly higher in persons with AFP and encephalitis (P 0.001); this was caused by the fact that several persons with AFP and encephalitis died and thus had lifetime lost productivity, which contributed to a higher total lost productivity.
Overall, there was no significant difference in total longterm costs by syndrome (Table 4). Finally, there was no correlation found between initial hospital costs and any long-term costs.
Cost of hospitalized WNV cases and deaths in the United States. From 1999 through 2012, 37,088 WNV disease cases, including 16,196 (44%) neuroinvasive disease cases, were reported to CDC. Of these, 1,529 (4%) died and an estimated 18,313 (49%) of the total cases were hospitalized. When all ArboNET cases were compared with WNV cases from the follow-up cohorts, the data were not significantly different in terms of age and sex by clinical syndrome. The follow-up cohorts did have a higher proportion of AFP patients in comparison to the U.S. national data.
The total mean cost of WNV hospitalized cases and deaths as reported to CDC for 1999-2012 was~$778 million (95% CI $673 million-$1.01 billion) or an average of $56 million per year (95% CI $48-$72 million) ( Table 6). Of the overall cumulative costs, $449 million (58%) are from lifetime lost productivity caused by deaths from WNV infection. Hospitalization during the acute illness accounted for an estimated †Lost productivity costs include the lifetime lost productivity costs for six case-patients that died during the acute hospitalization, including four encephalitis cases and two AFP cases. For these cases, no value was assigned to the days spent in the hospital prior to their death. ‡Test was statistically significant at an individual test critical P value 0.0083 when comparing among clinical syndromes for that variable using Kruskal-Wallis test and adjusting for multiple tests using the Bonferroni correction.
§Additional care costs include costs experienced because of stay in an inpatient rehabilitation or nursing home facility and as a result of home care costs. ¶Medication, medical equipment, or modifications include cost of new medications after hospitalization, new durable medical equipment, or any modifications that were needed to their home or car because of their West Nile virus disease. AFP = acute flaccid paralysis; USD = United States dollars.

DISCUSSION
This study found that both acute and long-term costs varied between the different clinical presentations of WNV disease. Overall, a substantial proportion of case-patients in our long-term follow-up cohort incurred additional medical or lost productivity costs in the 5 years after their hospitalization. We estimate that tens of millions of dollars are spent each year on WNV hospitalized cases and deaths.
Traditionally, WNV neuroinvasive disease (e.g., meningitis, encephalitis, and AFP) cases have been grouped together when looking at long-term outcomes and impact of the disease. 1,13,14,19 We found, however, that persons diagnosed with AFP incurred the most expense related to their WNV disease. This is not surprising given the severity of AFP, which often  results in partial paresis to extensive paralysis that is often permanent, and similar high costs found with other conditions that can cause limb paralysis such as Guillain-Barré Syndrome. 9,20 Medical care costs, in particular the initial hospital costs, for case-patients with encephalitis were similar to that of AFP cases in our study and the costs for WNV encephalitis are similar to those previously published for other forms of encephalitis. 21 However, the lost productivity was notably lower for encephalitis cases because encephalitis tended to occur in older individuals who were retired at the time of their illness onset and thus incurred no lost productivity costs in our study. The hospitalization costs for case-patients with WNV meningitis in our study are similar to those for other causes of viral meningitis and were more similar to hospitalized fever or non-neuroinvasive cases than other neuroinvasive disease cases. 22 In fact, meningitis cases had some of the lowest medical care costs, in particular for long-term costs, which was typically associated with only requiring a new medication (e.g., analgesic) and a visit with their primary care doctor after their initial hospitalization. However, persons with WNV meningitis tended to be younger and experienced more time away from work than fever or encephalitis cases and thus they had the highest cost for long-term lost productivity. Finally, hospitalized fever patients incurred initial median costs of $4,600 and long-term median costs over $2,200. Given that one in five cases of WNV fever reported to CDC are hospitalized, we believe that previous costs associated with WNV fever cases have been substantially underestimated. 1,13 Two previous studies have estimated the total cost of WNV disease outbreaks. 13,14 Zohrabian and others 14 focused on neuroinvasive disease cases during the early outbreak of WNV disease in Louisiana during 2002. They estimated an average cost per case of approximately $27,000. Even after adjusting to 2012 USD, this estimate is lower than the cumulative (initial and long-term) costs we have for most neuroinvasive disease syndromes, in particular AFP and encephalitis. The differences found in the costs between this and our study could be the result of the longer follow-up time in our study (5 years versus 1-8 months), inclusion of medication and home modifications costs in our study, and differences in the proportion of patients with various clinical syndromes (e.g., higher proportion of AFP patients in our study). Barber and others 13 examined the cost of the WNV outbreak in Sacramento County, California during 2005 and included the costs for WNV neuroinvasive disease cases, all assumed to be hospitalized, and for WNV fever cases, all assumed to not be hospitalized. The average cost per case in that outbreak was $13,971, which is lower than most of our cumulative costs for the various syndromes after adjusting to 2012 USD. In Barber's study, however, they included a large proportion of unhospitalized fever case-patients who had an average cost of $1,170/case that likely contributed to their lower cost estimates.
Extrapolating from hospitalized case data from Colorado, the total societal costs of WNV hospitalized cases and deaths as reported to CDC for 1999-2012 was estimated to be roughly $778 million dollars or~$56 million per year. However, the annual cost of WNV disease varies substantially as the number of WNV disease cases has ranged from 21 cases in 2000 to 9,862 cases in 2003. 1 Our estimate of WNV costs is conservative as it does not account for costs incurred by nonhospitalized WNV cases or make any adjustment for the under diagnosis or underreporting of WNV disease cases. 1,23 To date, an estimated 18,775 non-hospitalized cases of WNV disease have been reported to CDC. If the cost estimate of a fever case (cost of one healthcare provider office visit, one diagnostic test, and 5 days off work) from Barber and others 13 is applied and adjusted to 2012 USD, the estimated total costs of non-hospitalized cases from 1999 to 2012 would bẽ $28 million or 3.6% of the total direct or indirect medical costs incurred by hospitalized WNV disease cases. The extrapolated costs do not include costs incurred by public health or those related to mosquito control efforts, which are often 25% to 50% of the total cost during an outbreak response. 13,14 Finally, in the model we were conservative by 1) using a uniform distribution with a minimum of zero and a maximum equivalent to the proportion of cases in the long-term followup cohort who experienced any long-term costs to approximate the proportion of hospitalized cases in the United States who would incur long-term cost, and 2) selecting distributions from those short listed by @Risk for the Monte Carlo model that fit the data but had higher probability of picking lower costs than higher costs for the same Kolmogorov-Smirnov and Anderson-Darling goodness-of-fit.
There are several limitations to our study. Participants were interviewed about long-term care and costs 5 years after their disease onset, which may have led to recall biases. We used hospital charges directly from the hospitals in Colorado and adjusted care costs for the United States to estimate other medical expenditures. In general, hospital charges in Colorado tend to be lower than the national costs. 24 Medical care costs vary by type of service, provider, and insurance coverage from state to state and within a state 24 ; thus, our cost data may not reflect the costs incurred by those in other states. As previously noted persons in the long-term cohorts represent a nonrandom sample and therefore may not be representative of persons with WNV disease in other areas. However, the longterm cohorts were similar in their age and sex distribution to both the original Colorado cohort and WNV disease cases in ArboNET and their cost data reflected the known clinical severity of the different disease presentations and cost data for like syndromes. Given this, we feel that our cost data is likely reflective of costs incurred by WNV disease cases in other locations in Colorado or the United States. By using the Bonferroni correction when multiple tests were being performed simultaneously, we took a conservative approach and may have underestimated the significance of potential differences between clinical syndromes. Patients with WNV infections often have signs and symptoms that overlap different clinical syndromes. 9 Given this, patients in the follow-up cohorts and national data could have been incorrectly categorized into one of four syndromes; this would impact the per syndrome costs and total cost estimates. There were relatively small numbers of patients with any one syndrome in the follow-up cohorts; this likely impacted the precision of the estimated costs. Some of the data (e.g., number of hospital cases for 1999-2003) used for estimating the total national cost of WNV disease was inferred, which likely impacts the precision of the total cost. Finally, we only included hospitalized patients and thus are unable to state anything about the acute or long-term costs for WNV disease case-patients who are not hospitalized; however, the economic impact of this group is likely to be of minimal significance overall.
We provided acute and long-term costs of WNV disease among hospitalized patients and estimate that tens of millions of dollars are spent each year on hospitalized WNV disease cases. We found that long-term direct and indirect medical costs after the initial illness and hospitalization are incurred by most patients and account for a proportion of the total costs of the disease. Therefore, long-term costs need to be figured into the cost of WNV disease. Additional studies are warranted among larger cohorts of both hospitalized and non-hospitalized persons with various clinical WNV disease syndromes to better understand and estimate the long-term effects and costs associated with WNV disease. Results from this study can be used in assessing the cost-effectiveness of various interventions designed to lower WNV disease risk.