Estimating Effect of Antiviral Drug Use during Pandemic (H1N1) 2009 Outbreak, United States

From April 2009 through March 2010, during the pandemic (H1N1) 2009 outbreak, ≈8.2 million prescriptions for influenza neuraminidase-inhibiting antiviral drugs were filled in the United States. We estimated the number of hospitalizations likely averted due to use of these antiviral medications. After adjusting for prescriptions that were used for prophylaxis and personal stockpiles, as well as for patients who did not complete their drug regimen, we estimated the filled prescriptions prevented ≈8,400–12,600 hospitalizations (on the basis of median values). Approximately 60% of these prevented hospitalizations were among adults 18–64 years of age, with the remainder almost equally divided between children 0–17 years of age and adults >65 years of age. Public health officials should consider these estimates an indication of success of treating patients during the 2009 pandemic and a warning of the need for renewed planning to cope with the next pandemic.

From April 2009 through March 2010, during the pandemic (H1N1) 2009 outbreak, ≈8.2 million prescriptions for infl uenza neuraminidase-inhibiting antiviral drugs were fi lled in the United States. We estimated the number of hospitalizations likely averted due to use of these antiviral medications. After adjusting for prescriptions that were used for prophylaxis and personal stockpiles, as well as for patients who did not complete their drug regimen, we estimated the fi lled prescriptions prevented ≈8,400-12,600 hospitalizations (on the basis of median values). Approximately 60% of these prevented hospitalizations were among adults 18-64 years of age, with the remainder almost equally divided between children 0-17 years of age and adults >65 years of age. Public health offi cials should consider these estimates an indication of success of treating patients during the 2009 pandemic and a warning of the need for renewed planning to cope with the next pandemic.
F rom April 23, 2009, through April 10, 2010, it is estimated that pandemic (H1N1) 2009 virus caused ≈61 million cases of infl uenza (range 43-89 million cases), ≈270,000 related hospitalizations (range 195,000-403,000 hospitalizations), and ≈12,500 deaths (range 8,900-18,300 deaths) in the United States (1). Even before the impact was fully known, the Centers for Disease Control and Prevention (CDC) recommended prompt empiric treatment with infl uenza antiviral drugs, principally the neuraminidase-inhibiting infl uenza antiviral drugs oseltamivir and zanamivir, of persons with suspected or confi rmed infl uenza and who also met >1 of the following conditions: 1) illness that required hospitalization; 2) progressive, severe, or complicated illness, regardless of previous health status; and 3) risk for severe disease (e.g., patients with asthma, neurologic and neurodevelopmental conditions; chronic lung or heart disease; blood, endocrine, kidney, liver, and metabolic disorders; pregnancy; and those who were old or young) (2). The primary goal of these recommendations was to reduce the number and severity of pandemic (H1N1) 2009 cases, especially hospitalizations.
We present estimates of the number of pandemic (H1N1) 2009-related hospitalizations, by age group, averted because of use of antiviral drugs given to treat clinical cases of infl uenza. These results can be used by public health policy makers to plan and prepare for the next pandemic. For example, these estimates can be used to help evaluate the policy option of replenishing state and federal infl uenza antiviral drug stockpiles

Methods
We developed a spreadsheet-based model to calculate the number of pandemic (H1N1) 2009-related hospitalizations averted because of treatment with the neuraminidase-inhibiting infl uenza antiviral drugs oseltamivir and zanamivir (online Technical Appendix, www.cdc.gov/EID/content/17/9/110295-Techapp.htm). The risk for hospitalization (and thus potential benefi t from antiviral drugs) differed by age groups (1). Therefore, we estimated the reduced number of hospitalizations separately for 3 groups: persons 0-17 years of age, 18-64 years of age, and >65 years of age. We calculated the hospitalizations averted by using the following general equation: no. hospitalizations averted (by age group) = [no. prescriptions written -estimated no. written for prophylaxis, stockpiling, or incomplete adherence to drug regimen] × age groupspecifi c risk for hospitalizations caused by pandemic (H1N1) 2009 × age group-specifi c effectiveness of drugs in preventing hospitalizations.

Prescriptions Filled
We used the number of prescriptions fi lled for these drugs for weeks ending April 24, 2009, through March 26, 2010 (Table 1), collected from the IMS Health (IMS) Xponent proprietary prescription database (IMS Health, Norwalk, CT, USA) (3). This database contains all retail prescriptions fi lled from a representative sample of 35,000 (73%) of ≈50,000 US-based retail pharmacies, including independent pharmacies, chain pharmacies, pharmacies in discount outlets, pharmacies in food stores, mail order pharmacies, and pharmacy benefi t management companies. IMS then proportionately extrapolates their data on the basis of populations served by the included pharmacies to provide weekly estimates of all prescriptions fi lled in the United States for these drugs. The Xponent database does not track prescriptions fi lled by in-hospital pharmacies. Therefore, in-hospital prescriptions are not part of our calculations.
The IMS Xponent database captures all fi lled prescriptions related to infl uenza antiviral drugs within its sample pharmacies. However, it does not identify the source of the drugs. During 2009, there were 2 main potential supplies for the antiviral drugs-the regular commercial supply system and state and federal government-maintained drug stockpiles. The IMS database does not track medications dispensed from public domains, such as public health departments. Furthermore, the federal and state stockpiles of antiviral drugs were meant to supplement the commercial supply chain in times of drug shortages anticipated to occur during a pandemic emergency. As of August 2010, the estimated total amount of antiviral drugs managed by states throughout the pandemic was 38 million treatment regimens. This estimate includes antiviral drugs purchased by states (26 million treatment regimens) plus ≈12 million treatment regimens distributed early in the pandemic to states from the CDC Strategic National Stockpile (SNS). Preliminary reports from state public health departments to the CDC show that most SNS product was either retained by the health departments or deployed at the local level (to dispensing sites such as drug stores and health departments). Sites received directions that the SNS-provided supplies were to be dispensed if commercial supplies could not keep up with demand or used to treat uninsured or underinsured persons who could otherwise not afford treatment. Preliminary data reported to CDC through SNS show that minimum quantities of stockpiled antiviral drugs were actually dispensed to patients. Because the commercial supply chain for antiviral drugs remained relatively robust, most states did not need to use stockpiled antiviral drugs. Therefore, we did not include any estimates of impact on antiviral drugs dispensed from these government stockpiles.

Prescriptions by Age Group
IMS collects for fi lled prescriptions deidentifi ed data regarding age of patient from the pharmacy systems. We thus divided the total number of prescriptions given into 3 age groups (0-17 years, 18-64 years, >65 years) by using age-specifi c data from IMS that covered prescriptions written for oseltamivir from October 9, 2009, through March 26, 2010. The percentages were as follows: 0-17 years, 38.6%; 18-64 years, 53.4%; >65 years, 5.3% ( Table  2). Note that ≈3% of prescriptions fi lled during this period did not have the age of the patient recorded. Therefore, we did not include those prescriptions in our analysis.

Prescriptions over Time
We plotted the total number of prescriptions fi lled per week, from the IMS database, against the weekly number of estimated pandemic cases for April 24, 2009, through March 26, 2010. Estimates of cases for April through the end of July 2009 are not available on a weekly basis. Thus, all cases were combined into a single estimate for that period (1). We combined for the same period all fi lled prescriptions and directly compared cases and prescriptions. A notable divergence in the correlation between plots of cases and prescriptions over time would indicate the possibility of prescriptions being fi lled for reasons other than the immediate treatment of infl uenza-related illness (e.g., stockpiling or use for prophylaxis).

Percentage of Prescriptions Written for Prophylaxis
We assumed in the absence of any data that 10% of all prescriptions for these antiviral drugs were written for Assumption: Not all patients will adhere with the drug regimen as prescribed. Also, some prescriptions were for personal stockpiles Antiviral drug effectiveness against hospitalization, by age group, y ‡ Literature review (see Table 3) 30%-50% Median (range) risk for hospitalization, given pandemic (H1N1) 2009-related illness, by age group, y § †These inputs were subjected to sensitivity analyses (see Table 4). ‡Effectiveness estimate assumes that the patient follows the drug regimen, i.e., these estimates do not allow for those who do not take the complete course. Failure to follow prescribed drug regimen was assumed to have 0% effect on reducing risk of hospitalization. This assumption was accounted for in a separate input. §Risk of per-person hospitalization, given symptomatic illness caused by pandemic (H1N1) 2009 virus.
prophylaxis. This assumption was subject to sensitivity analyses (described below). We further assumed that such prescriptions essentially had no impact on reduction of hospitalizations (Table 2).

Adherence to Drug Regimen and Stockpiling
We also assumed that a total of 20% of all prescriptions were for either personal stockpiles (i.e., not written for a clinically ill patient at time of prescription) or patients who did not suffi ciently follow the recommended drug regimen so that the prescription had no impact on risk of hospitalization caused by nonadherence (Table 2). A study conducted in the United Kingdom during the (H1N1) 2009 pandemic found that 76%-80% of the patients did complete the full course of prescribed antiviral drugs (5). Another study among schoolchildren in London, UK, that examined adherence among those offered oseltamivir for prophylaxis found that 89% actually took >1 dose and 66%  of this group completed (or said they would complete) a full 10-day prophylaxis course (6). One of the drug effectiveness studies that we reviewed (discussed below) and used for model input values asked patients to selfrecord adherence; it found that ≈90% of enrolled patients were fully compliant (7). Our assumption that 20% of prescriptions were for either stockpiling or nonadherence was subject to sensitivity analyses (described below). This allowance for nonadherence also acts as a proxy for those who may have started the treatment too late. To maximize drug effectiveness in alleviating the duration of symptoms, it is recommended that antiviral drug treatment start <48 hours after onset of clinical symptoms (2).

Risk for Hospitalization Given Clinical Case of Pandemic (H1N1) 2009
We used the risk for hospitalization by age group, given clinical illness caused by pandemic (H1N1) 2009, from Reed et al. (4) ( Table 2). We identifi ed 17 published studies that evaluated the effectiveness of neuraminidase inhibitors given infl uenza-induced clinical illness (7,8-21; Table 3). Although many studies were random placebo-controlled trials, the studies did not use hospitalizations averted as a measured endpoint (13,(15)(16)(17). We identifi ed only 4 studies that specifi cally evaluated the impact of the antiviral drugs on risk for hospitalization, given clinical illness. One study provided an estimate of 50% reduction in the probability of infl uenza-specifi c hospitalizations (no confi dence interval was published) (7). Three retrospective studies, using health insurance claims data, reported effectiveness in reducing hospitalizations (any cause) that ranged from 22% to 59%, with some variation by age (8)(9)(10). For each age group, we used lower and upper estimates of effectiveness, from a lower estimate of 22% reduction for children 0-17 years to an upper estimate of 50% for adults (Table 2).

Calculating Ranges and Sensitivity Analyses
For each level of antiviral effectiveness (lower, upper), and for each age group, we calculated the median and lower and upper estimates of hospitalizations averted. We also conducted sensitivity analyses by altering from 0% to 30% the assumed percentages of prescriptions written for prophylaxis, personal stockpiles, and patients who did not adhere to the drug regimen.

Results
Pandemic infl uenza vaccine became available in week 40 of 2009 (near the peak of cases). We hypothesized that before this date is when doctors would have been most likely to try to protect patients by prescribing prophylactic courses of antiviral drugs. However, the plot of the prescription data against estimated cases over time shows a close correlation between the occurrence of pandemic (H1N1) 2009 clinical cases and fi lled prescriptions (Table  1; Figure). This comparison suggests that antiviral drugs were mostly prescribed to treat the occurrence of clinical cases of pandemic (H1N1) 2009.
The total number of prescriptions fi lled before adjustments was 8.2 million (Table 1). After removing the prescriptions presumed fi lled for prophylaxis and for patients who failed to adhere to the drug regimen or had prescriptions fi lled for personal stockpiles, 5.7 million prescriptions were fi lled that may have reduced hospitalizations (Table 4). Most (97%) were fi lled for oseltamivir, and ≈55% of all prescriptions fi lled were for persons 18-64 years of age, and ≈40% were fi lled for children 0-17 years of age.
We estimated that the median number of hospitalizations averted ranged from 8,427 (lower 6,961; upper 9,479) to 12,641 (lower 10,442; upper 14,219) ( Table 5). Approximately 60% of averted hospitalizations were among persons 18-64 years old. The estimated hospitalizations averted in children and adults >65 years of age (Table 5) were similar. Although adults >65 years of age received only ≈5% of fi lled prescriptions (Table 4), these prescriptions had a relatively substantial impact in averting hospitalizations because the risk for hospitalization is higher in this age group than the other risk groups (Table 2).
Doubling the assumed percentages of fi lled prescriptions for prophylaxis and personal stockpiles/ nonadherence from 30% to 60% (i.e., a 100% increase) produced only a 40% reduction in median hospitalizations averted, from ≈12,600 to 7,200 (Table 6). Thus, the major factors infl uencing hospitalizations averted were total prescriptions fi lled and (assumed) effectiveness of the drugs in preventing hospitalizations.

Discussion
The close correlation between estimated pandemic infl uenza cases and fi lled prescriptions (Figure) can be used as evidence that antiviral drugs were mostly used to treat those who were clinically ill (i.e., recommendations regarding use were essentially followed). Restricting the use of antiviral drugs to treating the clinically ill meant that preventing clinical cases from deteriorating into severe cases requiring hospitalizations was likely to have been among the major effects of antiviral drug use. By our estimates, this strategy worked; ≈8,000-13,000 hospitalizations were averted (Table 5). This reduction is equivalent to ≈4-5% of the total estimated pandemic (H1N1) 2009-related hospitalizations (1).
We found no other studies with which to compare our methods and results. We compared the accuracy of the IMS database using unpublished data from the Behavioral Risk Factor Surveillance System (BRFSS), conducted in 49 states (excluding Vermont, the District of Columbia, and Puerto Rico). From September 1, 2009, through March 31, 2010, adults (>18 years old) responding to the BRFSS telephone survey were asked whether they had infl uenzalike illness (ILI) (defi ned as having had a fever with cough or sore throat) in the month preceding the interview. They were also asked if they sought medical care for their ILI condition and if they were prescribed antiviral drugs to treat their illnesses. Extrapolating the results to the national level in the period covered by the survey, we found that ≈54 million adults reported having ILI symptoms. Of those who reported having ILI and sought medical care, 4.1 million adults reported they were prescribed infl uenza antiviral drugs (oseltamivir or zanamivir) during August 2009-March 2010. The IMS database recorded 6.86 million prescriptions in the same period (Table 1); ≈40% for those 0-17 years of age (Table 2), leaving ≈4.1 million fi lled prescriptions for adults. This estimate is close to the number recorded by the BRFSS survey and further supports the idea that few prescriptions were for prophylaxis or personal stockpiles.
There are many limitations to this study; the biggest is the uncertainty regarding the effectiveness of the drugs in preventing hospitalizations. The effectiveness of the drugs in reducing risk for hospitalization caused by pandemic (H1N1) 2009 may vary considerably from estimates reported for nonpandemic strains of infl uenza virus. The data are also limited in that we cannot verify if those persons who fi lled a prescription were actually clinically ill from pandemic (H1N1) 2009 or to what extent they adhered to the drug regimen. We addressed this issue by allowing a wide range in drug effectiveness and a relatively large percentage of prescriptions fi lled for conditions other than direct treatment of pandemic (H1N1) 2009.
We were unable, because the available literature did not contain suffi ciently reliable estimates of effectiveness of antiviral drugs against death, to estimate the number of deaths averted by treatment with antiviral drugs. Shrestha et al. (1) estimated that deaths caused by pandemic (H1N1) *These antiviral drugs were prescribed in a variety of forms (e.g., capsules, tablets, syrup, and inhaled powder). The estimated numbers came from the IMS database (3), which records 73% of all prescriptions filled by >50,000 US-based retail pharmacies. IMS then proportionately extrapolates their data, based on populations served by pharmacies, to provide weekly estimates of all prescriptions filled in the U.S. for these drugs. The IMS Health Xponent database does not cover in-hospital prescriptions. †These subtotals, by age group, are the estimates of prescriptions filled to treat pandemic (H1N1) 2009-related clinical illness, after removing the prescriptions filled for prophylaxis and for patients who failed to adhere to drug regimen or prescriptions filled for personal stockpiles (see Table 1). The total number of prescriptions filled, before adjustments, was 8,177,542 ( Table 1). Note that 3% of prescriptions filled during this period did not have age of patient recorded, and we omitted those prescriptions from our calculations. ‡These subtotals, by age group, were the estimates used to calculate the hospitalizations averted as shown in Table 5.  Table 4). If during the next pandemic there is a desire to produce better quality estimates (perhaps even produce estimates at regular intervals during the event), then additional data collection systems must be developed to overcome some of these limitations. For example, measuring the number of prescriptions fi lled for prophylaxis or personal stockpiles or degree of adherence can only reliably be conducted by interviewing patients and physicians. Improving estimates of impact of fi lled prescriptions in reducing adverse health outcomes during an event will require a large case-control study. Policy makers will have to determine if the value of such information warrants the investment in such data collection systems.
Our results also highlight how the use of infl uenza antiviral drugs during a pandemic is likely to be benefi cial, notably through a presumed reduction in the demand for hospital-based resources. Reduced demand will also reduce costs of hospitalizations. Assuming a cost per infl uenza-related hospitalization of US$5,000-$7,000 per patient admitted (adjusted to 2009 dollars) (22)(23)(24)(25)(26), averted hospitalizations saved ≈$42 million to $88 million (based on median values of hospitalizations averted; Table 4). A detailed cost-effectiveness analysis, including an in-depth consideration of the costs of hospitalizing pandemic (H1N1) 2009 patients, is the subject of a separate analysis.
If the next infl uenza pandemic causes greater numbers of severe cases and hospitalizations than in 2009, there may be an increased demand for antiviral drugs for treatment and prophylaxis. Such increased demand could overwhelm the existing commercial distribution chains. Therefore, public health offi cials should consider these estimates as an indication of success of treating patients during the 2009 pandemic and a warning for the need for renewed planning to cope with the next pandemic. *Baseline data used displays 10% for prophylaxis and 20% for personal stockpiling and non-adherence. This baseline assumption was used to generate results in Table 5. †Results of sensitivity analysis were calculated by using the upper median estimates of antiviral effectiveness in preventing hospitalization among the clinically ill (Tables 1, 2).