A Framework for Classifying Disease Trends Applied to Influenza-Associated Hospital Admissions in the United States

Abstract

Using Influenza Hospitalization Surveillance Network (FluSurv-NET) data, we established thresholds for weekly hospitalization rate differences to categorize changes. We applied thresholds to National Healthcare Safety Network influenza admissions from the 2022/2023–2023/2024 seasons. This metric identified more large increases and decreases in 2022/2023 and facilitated rapid trend assessment.

Graphical Abstract

Graphical Abstract.

This graphical abstract is also available at:https://tidbitapp.io/tidbits/a-framework-for-classifying-disease-trends-applied-to-influenza-associated-hospital-admissions-in-the-united-states-a9183fa2-2a18-4c18-9bde-41e07cc3a246?utm_campaign=tidbitlinkshare&utm_source=ITP

Seasonal influenza causes substantial morbidity and mortality, and a variety of surveillance systems are used to monitor influenza activity, including the Influenza Hospitalization Surveillance Network (FluSurv-NET) and the National Healthcare Safety Network (NHSN) [1, 2]. These surveillance systems can inform where influenza activity is happening and how many people had influenza-associated hospitalizations, for example. However, these indicators often require broader context to understand the overall impact of influenza. The Moving Epidemic Method (MEM) and the Average Curve Method (ACM), among other approaches, were developed to help provide this context [3]. These tools have been used across the globe to set thresholds for determining the severity of influenza and other respiratory viruses [4–6]. The Centers for Disease Control and Prevention uses the MEM approach to help classify the severity of seasonal influenza epidemics [4].

In addition to the overall impact of seasonal influenza, a key public health question is also whether influenza activity is stable over time or changing (ie, trends in influenza activity). Quantitatively evaluating recent weeks’ trends during the current season may aid in early detection of atypical increases or unusual patterns in influenza activity and provide a strong foundation for situational awareness.

Trend metrics based on previous influenza seasons can help anticipate seasonal dynamics and inform timely in-season information on the need for heightened prevention and control measures, particularly in health care settings (eg, hospitals). Here we describe efforts to develop a method to classify trends in influenza activity based on historic data and describe influenza hospital admissions trends across the 2022/2023 and 2023/2024 seasons in the United States.

METHODS

We analyzed hospital admission data from 2 surveillance systems, FluSurv-NET and NHSN.

FluSurv-NET is a population-based surveillance system that collects data on laboratory-confirmed influenza-associated hospital admissions among children and adults [7]. Active surveillance is conducted from October 1 through April 30 every year but was extended to June 11, 2022, during the 2021/2022 season due to late influenza season activity. Detailed FluSurv-NET methods are available elsewhere [8, 9]. Weekly rates per 100 000 population are available from participating sites as well as an overall network rate [1]. As of the 2022/2023 season, FluSurv-NET includes >90 counties in 14 states and an estimated 9% of the US population.

The NHSN hospital admission data set refers to the COVID-19 Reported Patient impact and Hospital Capacity by State Timeseries (originally called HHS-Protect) [10]. During the coronavirus disease 2019 (COVID-19) pandemic, the Department of Health and Human Services mandated that hospitals report daily influenza hospital admission counts in each state and Washington DC beginning on February 2, 2022. The original mandate remained in effect through April 30, 2024 [11]. For this analysis, daily influenza hospital admission counts were aggregated by state and epidemiological week and converted to per capita rates using 2021 and 2022 US Census population estimates [12] , Supplementary Files locations_2021.csv and locations_2022.csv.

Using 12 seasons of FluSurv-NET data spanning 2010–2023 (358 epi weeks) and 3 seasons of influenza data from the NHSN spanning 2021–2024 (117 epi weeks), we calculated week-to-week differences in influenza hospital admission rates for each season. For both data sets, we began including rate differences in our observed difference distributions once there had been at least 3 consecutive weeks of nonzero rates beginning with epi week 40.

From the distribution of rate differences, we established 5 categories of change: stable, increase, large increase, decrease, large decrease. Rounded FluSurv-NET rate differences were used to specify the category thresholds. Differences greater than the 25th percentile and lower than the 75th percentile were classified as stable, differences between the ≥75th and <95th percentiles were classified as increase, between the >5th and ≤25th percentiles were classified as decrease, and differences ≤5th percentile or ≥95th percentile were classified as large decrease or large increase, respectively. For each jurisdiction, weeks with an absolute change of <10 admissions were classified as stable to limit the impact of reporting delays and random fluctuations (most NHSN influenza hospital admission counts were revised by <10 admissions per week after initial publication [13]). See the Supplementary Analysis for additional details on count threshold.

We applied the rate thresholds, developed using FluSurv-NET, to NHSN hospital admission data for each jurisdiction to characterize the changes observed throughout the 2022/2023 and 2023/2024 influenza seasons.

RESULTS

Historic Data Analysis and Threshold Determination

The median rate difference between consecutive weeks of historic FluSurv-NET and NHSN data was 0 influenza-associated hospitalizations per 100 000 individuals. The distribution of rate differences for both data sets was roughly symmetric around the median (Supplementary Table 1, Supplementary Figure 1). Supplementary Table 1 shows mean, standard deviation, median, range, and percentiles for the rate differences from FluSurv-NET and NHSN.

Based on the criteria described in the “Methods,” stable weeks were categorized as a change in weekly rate >–0.3 influenza-associated hospitalizations/100 000 but <0.3, or weeks with an absolute difference of <10 influenza hospital admissions (Figure 1). Increases were categorized as a positive change in weekly rates ≥0.3 to <1.7/100 000 influenza-associated hospitalizations, and large increases were categorized as a positive change in weekly rates ≥1.7/100 000 influenza-associated hospitalizations. Decreases and large decreases were categorized for negative rate changes using the same increase and large increase thresholds described above.

Figure 1.

Influenza rate and count thresholds for category specifications based on updated percentiles. The horizontal axis represents sequential differences in week-to-week population-based rates. The vertical axis represents absolute differences in week-to-week hospital admission counts. Gray shaded areas of the graph between –0.3/100k and 0.3/100k rate differences and count differences <10 are considered stable. Light purple indicates increases with rate differences from 0.3 to <1.7/100k; dark purple indicates large increases with rate differences ≥1.7/100k. Teal indicates decreases with rate differences from –0.3 to ≥–1.7/100k, and dark teal indicates large decreases with rate differences ≤–1.7/100k.

Framework Applied to 2022/2023 and 2023/2024 Influenza Seasons

Figure 2 A shows national weeks across the 2022/2023 and 2023/2024 seasons colored by week-to-week trend category. The 2022/2023 influenza season was characterized by a sharp peak of influenza-associated hospitalizations in early December and a rapid decrease in hospital admissions with low activity throughout the rest of the season in most of the United States. In the 2022/2023 season, 7.4% and 7.2% of weeks were classified as large decreases and large increases, and 12.6% and 12.5% of weeks were classified as decreases and increases, respectively. Figure 2B shows the proportion of weeks classified as each category. The week with the most observed increases and large increases was 11/26/2022, when 14 states observed increases and 24 states observed large increases. The week of 1/14/2023 had the most observed decreases or large decreases when 19 states observed decreases and 32 states observed large decreases.

Figure 2.

Week-to-week influenza hospitalization rate change trends for the NHSN 2022/2023 and 2023/2024 seasons. A, Weekly rates per 100k colored by trend category for the United States. Points are colored by trend category based on the rate difference from the previous week. B, The proportion of jurisdictions classified as each category by week. C, The week-to-week trend categories by jurisdiction. The dark gray at the beginning of each panel indicates that not enough data were available to categorize that week.

A similar but slightly less sharp peak occurred in late December of the 2023/2024 season, with a much slower overall decline in activity and a second peak in some jurisdictions later in the season. The 2023/2024 season had 3.3% and 4.2% of weeks classified as large decreases and large increases, respectively, and 19.4% as decreases and 16.1% of weeks classified as increases. Two weeks shared the most observed increases and large increases. The week of 12/23/2023 observed 25 increases and 18 large increases, and the week of 12/30/2023 observed 14 increases and 29 large increases. The week with the most decreases and large decreases was 1/13/24, when 30 states observed decreases and 11 observed large decreases.

Supplementary Tables 2 and 3 detail the distributions of categorized weeks by jurisdiction for week-to-week rate differences for the 2022/2023 and 2023/2024 seasons, respectively.

DISCUSSION

In this analysis, we used historic FluSurv-NET weekly data to determine thresholds for categorizing changes in NHSN influenza hospital admission trends due to seasonal influenza. We aimed to create robust classifications that enhanced our understanding of influenza activity over multiple seasons in the United States and applied these classifications to 2 seasons of NHSN data to evaluate their performance during seasonal influenza epidemics. These categories effectively characterized week-to-week trends in influenza activity. In addition, this method complemented other existing methods, such as MEM [10, 11], which require extensive epidemiological modeling and do not determine if metric changes differ from past seasons.

These metrics can be deployed in real time to enhance surveillance data interpretation, detect changes in influenza activity, and guide investigations into contributing factors. Robust and sustained surveillance systems that capture the range of activity observed over multiple influenza seasons with varying intensity are needed to create trend classifications. Utilizing larger percentiles or other surveillance indicators (eg, influenza test positivity) could improve the ability of this approach to detect more atypical changes in activity. Beyond interpreting surveillance data, the trend classifications can also be used to interpret forecasts or to directly forecast the probability of the trend categories occurring over the coming weeks [14]. This application could provide a new way to communicate forecasts, potentially enhancing their interpretability and utility within public health contexts.

Our analysis indicated a greater number of weeks classified as large increases and large decreases in the 2022/2023 season compared with the 2023/2024 season. While the absolute magnitude of hospital admissions may be lower, large week-to-week increases in hospitalizations could result in more pressure on health care systems than smaller increases in hospital admissions.

In summary, our analysis underscores the importance of integrating data from various sources to establish thresholds for classifying changes in seasonal influenza activity. Leveraging several seasons of FluSurv-NET data, alongside NHSN data, strengthened our definitions compared with using either data set alone. By developing a flexible framework accessible to public health practitioners, we aim to enhance real-time surveillance efforts, improve trend detection, and support effective public health response to influenza. While focused on hospital admission rates, we believe this framework can be adapted to other influenza surveillance metrics or diseases with robust time-series data. We also plan to refine our classification metrics through ongoing evaluation to ensure relevance amid evolving epidemiological landscapes.

Supplementary Data

Supplementary materials are available at Open Forum Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.