Four participating VISION partner healthcare systems (HealthPartners, Kaiser Permanente Northwest, Kaiser Permanente Northern California, and Kaiser Permanente Southern California) in five U.S. states (California, Minnesota, Oregon, Washington, and Wisconsin) contributed data to this retrospective cross-sectional study (Supplementary Table 1). EHR data were included from 94 urgent cares and 142 emergency departments. VISION partners performed geocoding of patient address to assign census tract-based 2020 Social Vulnerability Index (SVI) rankings.¹⁴ Contributing network partners have patient memberships that have been found to be regionally representative of the general population.^{15, 16}

These activities were reviewed and approved by institutional review boards (IRBs) at Kaiser Permanente Northwest, Kaiser Permanente Northern California, and Kaiser Permanente Southern California (45 C.F.R. part 46.101(c); 21 C.F.R. part 56). The Westat IRB determined that the activities are non-research. Patient informed consent was not required. The study followed guidelines from the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE).

Data were analyzed from the 2023–2024 influenza season among adult patients with acute respiratory illness (ARI)-associated ED/UC encounters and a positive influenza laboratory test. Participating VISION partners contributed complete influenza antiviral treatment data for prescribing and dispensing. Medical encounters were included if they occurred at participating partners and the patient met an ARI case definition based on International Classification of Diseases, 10th Revision discharge diagnosis codes (Supplementary Table 2). Three participating VISION partners required patient membership, and one partner included encounters from both member and non-member patients.

All patients received clinician-initiated testing for influenza and had a positive test result ≤2 days before or up to 72 hours after an encounter. Influenza tests were categorized as standard molecular (reverse transcription-polymerase chain reaction [RT-PCR] and multiplex assays using nucleic acid amplification and performed on platforms associated with extended test runtimes [>30 minutes to several hours]), 2) rapid molecular (isothermal nucleic acid amplification and RT-PCR performed on platforms yielding results in approximately 15–30 minutes), or 3) rapid antigen (rapid influenza diagnostic tests [RIDTs] with results available in 10–15 minutes).¹⁷ ED and UC encounters occurring within ≤7 days of one another were combined by contributing partners into a single encounter (ED/UC). Data were analyzed at the encounter level to allow for unique opportunities for providers to implement influenza antiviral treatment guidelines.

Seasonal influenza vaccination data was obtained from EHRs, state immunization systems, and claims data. A person was considered to have received a 2023–2024 seasonal influenza vaccination if they received ≥1 dose on or after August 1, 2023 and at least 14 days prior to the date of the encounter.

ARI-associated ED/UC encounters were considered eligible if the patient was aged ≥18 years. Encounters were restricted to those occurring during the 2023–2024 influenza season (October 1, 2023 to April 30, 2024).¹⁸ While the Infectious Diseases Society of America (IDSA) recommends empiric antiviral treatment for adult outpatients at higher risk of influenza complications with suspected or laboratory-confirmed influenza, this analysis only included encounters with positive influenza laboratory testing.³ To align with IDSA guidance on treating outpatients considered at higher risk of influenza complications, encounters were limited to patients meeting one or more of the following criteria (Supplementary Table 3): 1) being ≥65 years old, 2) having certain underlying medical conditions documented at the encounter (hereafter, “higher risk-underlying medical conditions [HR-UMCs]”), 3) having a body mass index (BMI) indicating severe obesity (≥40 kg/m²), or 4) being pregnant on the encounter date.³ Other persons considered at higher risk of influenza complications including those who were up to two weeks postpartum or residents of a nursing home/long-term care facility were not included in the criteria due to under-capture of post-partum data and because information on type of residence is not collected by VISION. Encounters were excluded if a patient received influenza antivirals prior to the encounter (within 14 days) for their current illness.

VISION partners obtained antiviral treatment data by performing a search for treatment codes and text strings across electronic sources (Supplementary Table 1). A patient was considered to have received influenza antivirals if on or within 14 days following the encounter date they were prescribed and dispensed a medication recommended in the United States for influenza treatment (oseltamivir, zanamivir, peramivir, or baloxavir marboxil).^{6, 19}

Among all eligible encounters, demographic and clinical characteristics overall and stratified by VISION partner were described using frequencies and proportions or medians and interquartile ranges (IQRs). To assess differences across VISION partners, Chi-square tests were performed for categorical variables, and Kruskal-Wallis or Wilcoxon rank-sum tests for continuous variables (p<0.05 significance).

To reflect similar diagnostic practices across all contributing VISION partners, the primary analysis was restricted to encounters with positive standard molecular testing. To characterize influenza antiviral treatment practices, demographic and clinical characteristics were first described stratified into binary antiviral groups: 1) among all encounters, prescribed versus not prescribed antivirals; and 2) among those prescribed antivirals, dispensed or not dispensed antivirals. Chi-squared or Wilcoxon rank-sum tests were performed to assess significant (p<0.05) differences between groups. Second, to explore characteristics with evidence of an association with antiviral prescribing and/or dispensing (p<0.05), odds ratios (ORs) were generated using logistic regression for the following outcomes: 1) among all encounters, prescribed antivirals; 2) among those prescribed, prescribed on the same day as the encounter; 3) among those prescribed, dispensed antivirals; 4) among those dispensed, dispensed on the same day as the prescription. ORs were adjusted by VISION healthcare system, encounter setting (ED, UC, or combined ED and UC), influenza season surveillance week (to account for variation in prescribing based on influenza activity), number of HR-UMC categories, age (continuous), BMI (continuous), race and ethnicity, sex, and seasonal influenza vaccination status. Categories of race and ethnicity were included as model covariates as social constructs reflecting differences in health access and outcomes driven by structural and systemic factors.

To explore patterns in prescribing and dispensing by influenza test type, the secondary analysis was restricted to the one VISION partner performing multiple influenza test types. Descriptive results were presented by antiviral group (prescribed and dispensed, prescribed only, not prescribed), further stratified into test type performed (standard molecular, rapid molecular, or rapid antigen).

Among 27,081 adult ED/UC ARI encounters with positive influenza testing and occurring from October 1, 2023, to April 30, 2024, 16,381 (60%) were excluded (Supplementary Figure 1): 16,047 (59%) were not considered at higher risk of influenza complications, and 334 (1%) received influenza antivirals within 14 days prior to the encounter. In total, 10,700 ED/UC encounters were included.

Eligible ED/UC encounters peaked during the week starting on December 24, 2023 (Supplementary Figure 2). Among 10,700 eligible encounters from four VISION partners, most occurred at urban facilities (97%) and ED settings (61%) (Supplementary Table 4). Median age was 65 years (IQR 41–74), with 59% female, 36% Hispanic, 37% non-Hispanic White, 16% Medicaid recipients, and 35% living in census tracts within the highest (i.e., most vulnerable) SVI quartile. Among the 7,138 eligible encounters with an antiviral medication prescription, 7,135 (99.96%) were prescribed oseltamivir while only three (<1%) were prescribed zanamivir (Supplementary Table 4).

Care setting varied across VISION partners (p<0.001), with one partner exclusively contributing ED encounters and three partners contributing both ED and UC encounters (Supplementary Table 4). Three VISION partners solely performed standard molecular influenza testing at all care settings. Across all partners, most (94–96%) influenza testing was performed on the encounter date (Supplementary Figure 3). Median patient age ranged from 61–68 years (p<0.001). Racial and ethnic demographics differed across VISION partners (p<0.001), with two partners having ≥65% of encounters among non-Hispanic White patients, while another partner had a higher (49%) proportion of Hispanic patients. SVI-designated social vulnerability varied across partners (p<0.001), with one having higher representation of census tracts within the lowest vulnerability quartile (38%), and another having higher representation among the highest vulnerability quartile (45%). Across partners, there was significant variation in proportion of encounters with one or more categories of HR-UMCs (37–79%, p<0.001).

Among encounters from all contributing VISION partners with positive influenza standard molecular testing results (N=5,231), 58% were prescribed antivirals (Table 1), with 67% of prescribing occurring on the encounter date (Figure 1). Among those prescribed antivirals (N=3,050), 80% had them dispensed (Table 1), with 65% of dispensing occurring on the prescription date (Figure 2). Antiviral treatment practices varied across VISION partners (47–64% prescribed, 75–91% prescribed and dispensed, both p<0.001) (Table 1, Figure 3). Encounters occurring exclusively at UCs had lower prescribing (50%) than those at EDs (61%) or combined ED/UCs (60%) (p<0.001); however, higher dispensing was found among UC (85%) than ED (78%) encounters (p<0.001).

Antiviral prescribing (p<0.001) and dispensing (p=0.01) varied across age groups (Table 1). Patient encounters aged ≥65 years had the highest proportion prescribed antivirals (63%); however, those aged 50–64 years had the highest proportion dispensed antivirals (85%) among those prescribed. Significant differences were also found in both prescribing and dispensing based on race and ethnicity, BMI, certain categories of HR-UMCs (cardiovascular disease, neurologic and neurodevelopment conditions), and receipt of 2023–2024 seasonal influenza vaccination (all p<0.05). No significant differences in prescribing and dispensing were found among encounters with positive influenza standard molecular testing based on sex, Medicaid status, SVI, or pregnancy status (all p>0.05).

In adjusted models among patients with a positive standard influenza molecular test, patients aged 50–64 years had higher adjusted odds (aOR 2.72 [1.63–4.55]) of being dispensed antivirals compared with those 18–49 years (Figure 4). No significant differences were found in odds of prescribing and dispensing for those aged ≥65 years compared to 18–49 years. However, lower adjusted odds of being prescribed antivirals on the encounter date (aOR 0.57 [95% CI: 0.42–0.78]) and dispensed antivirals on the prescription date (aOR 0.58 [95% CI: 0.36–0.94]) were found for those aged ≥65 years compared with those 18–49 years (Supplementary Table 5).

Among other demographic and clinical characteristics, no clear patterns in prescribing and dispensing were found. Slightly higher adjusted odds of being prescribed antivirals was found for encounters among Hispanic patients (aOR 1.39 [95% CI: 1.05–1.83]) (Figure 2). Using BMI 18.5 to <25 kg/m² as the reference group, those with BMI <18.5 kg/m² had lower adjusted odds of dispensing (aOR 0.22 [95% CI: 0.07–0.76]). No significant differences in adjusted odds of prescribing and dispensing were found based on number of categories of HR-UMCs.

Higher adjusted odds of prescribing (aOR 1.28 [95% CI: 1.07–1.54]) and dispensing (aOR 1.84 [95% CI: 1.30–2.60]) were found for those vaccinated against seasonal influenza; however, no differences were found in same-day prescribing or dispensing.

In a secondary analysis among 6,142 eligible encounters from VISION partner 4, multiple types of influenza tests were performed: 19% rapid antigen (10–15 minutes estimated test runtime), 70% rapid molecular (20 minutes), and 11% standard molecular (3–8 hours) (Supplementary Tables 1 and 4).^20-22 Characteristics varied across test type, notably urbanicity, encounter setting, age, race and ethnicity, BMI, and presence of underlying medical conditions (all p<0.001) (Supplementary Table 6).

The proportion of patient encounters prescribed influenza antivirals increased as associated test runtime decreased (standard molecular 58%, rapid molecular 72%, rapid antigen 84%, p<0.001, Figure 5A, Supplementary Table 7). Among those prescribed antivirals, proportion dispensed also increased as runtimes decreased (standard molecular 91%, rapid molecular 94%, rapid antigen 97%, p<0.001). These results were also consistent for same-day encounter date and prescribing (standard molecular 44%, rapid molecular 87%, rapid antigen 97%) and same-day prescribing and dispensing (standard molecular 84%, rapid molecular 90%, rapid antigen 98%) (Figure 5B).

This study has several limitations. First, this analysis reflects data captured by the VISION Network, with contributing partners predominantly comprising large, integrated managed care networks whose membership reflects populations with greater access to care. Similar analyses using expanded data sources may illuminate health disparities that are not reflected by VISION.^{10, 12, 53} Second, VISION does not capture illness onset dates, precluding assessment of guideline adherence for early treatment initiation following symptom onset. These data would be particularly relevant for patients with reduced access to healthcare and delayed treatment based on demographic characteristics. Characterizing time from illness onset to encounter date may also put into context clinical decisions driving gaps in prescribing among patients with positive influenza testing and considered at higher risk of influenza complications. Additionally, VISION does not capture the time of day that medical encounters, influenza test results, and antiviral prescription and dispensing occurred, limiting diagnostic turnaround time and time-to-treatment characterization. For standard molecular diagnostic tests with shorter associated runtimes (e.g., Cepheid GeneXpert Xpress) than multi-hour tests, describing total time associated with diagnostic activities (e.g., sample collection, transportation, laboratory backlog, results interpretation) would provide a more complete understanding of diagnostic delays. Third, as influenza testing and prescribing are clinician-driven and may be influenced by site-specific guidelines, clinical presentation, illness severity, and prior testing, variation in testing and prescribing practices across VISION partners may have occurred. However, VISION does not capture data on individual medical providers or patient illness severity at presentation, limiting description of these differences. Fourth, VISION combines ED/UC encounters occurring within ≤7 days of a previous encounter, limiting the ability to stratify all encounters by UC or ED setting. These settings may reflect unique populations in terms of access, care seeking, illness severity, and testing practices.