Modeling the potential impacts of outpatient antiviral treatment in reducing influenza-associated hospitalizations in the United States

Model framework

We developed a probabilistic model to track care-seeking, testing, antiviral prescribing, and risks of hospitalization among individuals with symptomatic influenza virus infection in the United States (Figure 1). The model was stratified into five age groups (0–4, 5–17, 18–49, 50–64, and ≥65 years). Within each age group, individuals were further stratified by risk of hospitalization based on the presence or absence of a condition associated with increased risk of influenza-associated complications⁵. Within each age group and risk stratification, individuals with symptomatic infection were partitioned into those who did or did not seek care in an outpatient clinical setting. For those who did not seek care, the probabilities of hospitalization were p_LR and p_HR for the low- and high-risk stratifications, respectively. Those that did seek care were partitioned according to whether they sought care within 48 hours of symptom onset (referred to as ‘early care-seekers’) or more than 48 hours after symptom onset (‘late care-seekers’).

• Download figure
• Open in new tab

Figure 1. Model structure within each age group and risk stratification.

The risk of hospitalization for individuals with symptomatic influenza infection depends on whether individuals seek care, the timing of care-seeking, the occurrence and result of testing, the prescribing of antiviral treatment, and the filling and completion of any treatment course. The number of individuals progressing through each pathway may change by age and/or risk stratification (see Figure S7 for a Sankey diagram showing example flows for high-risk adults ≥65 years¹³). In addition, the number progressing through each testing and prescribing pathway will vary by the timing of care-seeking (i.e., early vs. late). Similarly, antiviral effectiveness varies by the timing of care-seeking and will lead to different numbers progressing through the fully compliant to hospitalized pathway. Finally, the colors indicate compartments that are influenced by parameters relating to individual behavior and/or access to medical care and treatment (care-seeking, timing of care-seeking, and antiviral compliance) or parameters relating to clinical testing and prescribing practices. We investigate the impact of changing these parameters as part of our analyses. Abbreviations: CAR = clinical attack rate; CHR = case-hospitalization ratio.

Both early and late care-seekers had a certain probability of being tested for influenza in the outpatient setting and, if tested, a certain probability of testing positive that was determined by the sensitivity of the diagnostic assay. Early and late care-seekers were subsequently partitioned into three groups: those who tested positive for influenza, those who tested negative, and those who did not receive a test. We assumed individuals in these groups were then prescribed antivirals with probabilities determined by their test status, timing of care-seeking, and whether they were in a low- or high-risk stratification. Those who were not prescribed antivirals were hospitalized with probabilities p_LR and p_HR for the low- and high-risk stratifications, respectively. Those who were prescribed antivirals but did not fill their prescription or did not complete the full treatment course were also hospitalized with probabilities p_LR and p_HR. Those who did fill their prescription and complete the full course (referred to as ‘full antiviral compliance’) were hospitalized with reduced probabilities (1 − r) p_LR and (1 − r) p_HR, where r is the effectiveness of antivirals in reducing the risk of influenza-associated hospitalization. We parameterized r based on reported effectiveness of oseltamivir, the most commonly prescribed influenza antiviral in outpatient settings¹², and allowed it to vary with the timing of care-seeking^3,4.

Parameterization

We determined the number of individuals in each age group and risk stratification using 2022 US census data and age-specific estimates of the proportion of people at higher risk of influenza complications^14,15 (Table S1). The latter estimates included pregnant people and those with chronic medical conditions including asthma, heart disease, and lung disease, but did not include people with a body mass index (BMI) ≥40 or those living in long-term care facilities who are also considered to be at higher risk of influenza complications⁵. The number of individuals developing symptomatic influenza virus infection was approximated using age-specific clinical attack rates for influenza-associated illness during the 2022– 2023 influenza season¹⁶. Age-specific risks of hospitalization were obtained from influenza-associated case-hospitalization ratios and adjusted for risk stratification^17–19 (Table S1). Probabilities of seeking care, being an early or late care-seeker, and filling and completing an antiviral prescription were informed by seasonal influenza surveillance platforms and prior literature^7,11,20–23. Care-seeking varied by age and risk stratification, the timing of care-seeking varied by age but was constant across risk stratifications, and antiviral compliance was constant across age and risk stratifications¹¹.

The remaining parameters of testing, prescribing, and antiviral effectiveness are highly variable in outpatient settings. We therefore defined a range of scenarios that reflected plausible values for these parameters and were informed by prior literature and unpublished data from seasonal influenza surveillance platforms (Table 1). We defined three scenarios (‘low’, ‘intermediate’, and ‘high’) for the probability an outpatient with symptomatic influenza virus infection was tested for influenza^21,22. For each scenario, we assumed these probabilities were the same for all age groups and risk stratifications but were higher among early care-seekers compared with late care-seekers. To determine the proportion of outpatients testing positive for influenza, we multiplied the testing probability by estimates of test sensitivity for rapid molecular assays^24,25. We also defined two scenarios (‘low’ and ‘high’) for the probability an outpatient was prescribed influenza antiviral medications. We assumed these probabilities were constant across age groups, but varied with risk stratification, time of care-seeking, and the result of influenza testing^{7,10,21,22,26}. Finally, we explored three scenarios for antiviral effectiveness (‘low’, ‘intermediate’, and ‘high’) that were primarily informed by studies of oseltamivir but could be applicable to other existing or novel influenza antivirals given the broad range of values considered^{2–4,27–29}. We assumed effectiveness varied with the time of care-seeking but was constant across age and risk stratifications.

Model simulation

Taking all combinations of possible values for testing, prescribing, and antiviral effectiveness resulted in 18 (3 × 2 × 3) initial scenarios (Table 1). All remaining model inputs were assumed to follow independent Uniform probability distributions, except population size and hospitalization risk which were defined by age and risk-stratified point estimates (Table S1). For each scenario, we generated 1,000 independent samples for the parameters with Uniform probability distributions (Figure 2). We then simulated the model for each of the 18,000 resulting combinations and calculated the number of influenza-associated hospitalizations that occurred among all symptomatic individuals. We also compared the estimated number of hospitalizations to a corresponding baseline scenario without antiviral medications. Results are presented as the number, or percent, of all hospitalizations averted relative to this baseline scenario. Finally, we tracked the number of antiviral courses prescribed in each simulation and calculated the number of prescriptions needed to avert one hospitalization as the total number of prescriptions divided by the total number of hospitalizations averted.

• Download figure
• Open in new tab

Figure 2. Simulated parameter inputs.

Input parameter distributions were generated by taking 1,000 samples from their corresponding probability distributions. Risk stratifications refer to groups with high or low risk of influenza-associated hospitalization; full compliance refers to individuals who filled and completed a prescribed course of antiviral treatment.

Alternative scenarios

In addition to the scenarios defined above, we considered alternative scenarios for improved uptake of influenza antivirals through increasing upstream levels of care-seeking, early care-seeking, testing, prescribing for those with a positive test, and the filling and completion of antiviral prescriptions (Figure 1). For each parameter and age group or risk stratification, we increased the initial scenario or sampled distributions by 50% (Figure S1, Table S2), and investigated the impact of increasing one parameter at a time or increasing multiple parameters simultaneously. For all alternative scenarios, we used the most optimistic initial levels of antiviral effectiveness, testing, and prescribing as the starting point (i.e., the initial scenario with the highest levels of effectiveness, testing, and prescribing).

All analyses were performed in R 4.0.3 using the data.table and tidyverse packages^30,31,32.

Initial scenarios

We estimated there would be an average of 531,920 influenza-associated hospitalizations (95^th percentile: 360,353–699,923) in the baseline scenario without influenza antivirals. The use of antivirals averted an average of 1,215 (802–1,693) hospitalizations in the least optimistic scenario with low testing, low prescribing, and low antiviral effectiveness, and 14,184 (8,700–20,224) hospitalizations in the most optimistic scenario with high testing, high prescribing, and high antiviral effectiveness (Figure S2A). These numbers corresponded to a modest percentage reduction in total hospitalizations: 0.2% (0.2– 0.3%) and 2.7% (2.2–3.2%) in the least optimistic and most optimistic scenarios, respectively. The total number of prescribed antiviral courses ranged from 1,370,274 (1,073,330–1,700,729) to 2,379,603 (1,914,821–2,879,333) and the number of prescriptions needed to avert one hospitalization ranged from 176 (117–261) to 1,168 (785–1,689) (Figure S2B).

Among individuals at higher risk of influenza complications, the number of averted hospitalizations increased with increasing age group, from an average of 45 (21–75) among children 0–4 years to 8,061 (2,927–14,216) among adults ≥65 years in the most optimistic scenario with high testing, high prescribing, and high antiviral effectiveness (Figure 3). When testing and prescribing were instead fixed at their lowest levels, the number averted ranged from 15 (7–26) among children 0–4 years to 2,849 (1,044–4,996) among adults ≥65 years. Notably, the percent of hospitalizations averted was greatest for adults aged 18–49 and 50–64 years, with the latter experiencing an average reduction of 3.7% (2.7–4.7%) in the most optimistic scenario (Figure S3). In all scenarios, the number of prescriptions needed to avert one hospitalization was highest among children 5–17 years and lowest among adults ≥65 years (Figure S4).

• Download figure
• Open in new tab

Figure 3. Estimated number of influenza-associated hospitalizations averted in each initial testing and prescribing scenario compared with a baseline scenario without antivirals.

Antiviral effectiveness is fixed at its highest value (70% reduction in risk of hospitalization for early care-seekers and 35% for late care-seekers). Results are partitioned by age group and risk stratification, with points and error bars showing the mean and 95^th percentiles, respectively.

Scenarios to improve antiviral impact

Given the relatively modest estimated impact of antivirals in reducing influenza-associated hospitalizations in the scenarios above, we explored whether this impact could be improved by increasing upstream levels of care-seeking, timing of care-seeking, testing, prescribing, or antiviral compliance. First, we considered the parameters with sampled distributions that were broadly associated with individual behavior and access to care and treatment (care-seeking, timing of care-seeking, and antiviral compliance) (Figure 1). Increasing either care-seeking or antiviral compliance by 50% resulted in equal improvements in averted hospitalizations compared to the most optimistic initial scenario. While increasing early care-seeking by 50% also resulted in improvements compared to the most optimistic initial scenario, these were slightly less than those achieved by increasing care-seeking or antiviral compliance (Figure 4). Unsurprisingly, the largest improvement was achieved by increasing all three parameters simultaneously. In this case, the number of averted hospitalizations was greatest among adults ≥65 years in the high-risk stratification, reaching up to 24,220 (8,709–42,968) (Figure 4A). In contrast, the greatest percent of hospitalizations were averted among adults aged 50–64 years in the high-risk stratification, reaching up to 11.1% (8.2–14.3%) (Figure 4B). The number of antiviral prescriptions needed to avert one hospitalization decreased in all alternative scenarios compared to the initial scenario, except when only care-seeking was improved (Figure 4C). The scenarios which resulted in the greatest decrease were those with increased antiviral compliance and with all parameters increased simultaneously.

• Download figure
• Open in new tab

Figure 4. Estimated influenza-associated hospitalizations averted in select alternative scenarios compared with a baseline scenario without antivirals.

(A) Number of hospitalizations averted. (B) Percent of hospitalizations averted. (C) Number of prescriptions needed to avert one hospitalization. Scenarios shown are as follows: most optimistic initial scenario (‘initial’); 50% increase in the fraction of people seeking care (‘seek care’); 50% increase in the fraction of people seeking care within 48 hours (‘seek care early’)’; 50% increase in the fraction of people completing an antiviral treatment course (‘comply’); and 50% increase in the fraction of people seeking care, seeking care early, and completing a treatment course. Antiviral effectiveness, testing, and prescribing are fixed at their highest initial values. Results are partitioned by age group and risk stratification, with points and error bars showing the mean and 95^th percentiles, respectively.

We also considered alternative scenarios for the parameters broadly associated with clinical practices, i.e., the probability of testing and prescribing (Figure 1). Increased antiviral prescribing among outpatients with a positive test averted more hospitalizations than increased testing and required slightly fewer prescriptions to avert one hospitalization (Figure S5). Increasing both parameters averted the most hospitalizations among adults ≥65 years in the high-risk stratification (up to 12,909 [4,650– 22,834]), and the greatest percent of hospitalizations among adults 50–64 years in the high-risk stratification (5.9% [4.4–7.6%]).

Finally, to explore even greater potential improvements in antiviral impact, we considered an alternative scenario in which all five parameters were increased simultaneously. Again, the number of averted hospitalizations was greatest among adults ≥65 years in the high-risk stratification, at 40,093 (14,332–71,087), whereas the greatest percentage of hospitalizations was averted among adults 50–64 years in the high-risk stratification, at 18.5% (13.5–24.0%) (Figure S6). Across all age and risk groups, the use of antivirals averted 71,142 (43,581–102,969) hospitalizations, or 13.3% (11.1–16.2%), and the overall number of prescriptions needed to avert one hospitalization was 107 (71–157).