We obtained estimates for seasonal influenza- and RSV-attributable cardiorespiratory hospitalizations by fitting negative binomial regression excess models, a method commonly used in surveillance studies.8,31,32 Candidate models that were fit and compared consisted of different linear combinations of (1) a continuous measure of week number transformed with sine and cosine harmonic terms (ie, Fourier series) to allow annual or semiannual periodicity trends; (2) polynomial terms (quadratic and cubic) for number of weeks from the first observed time point to model nonlinear, aperiodic time trends; and (3) indicators for viral activity, both influenza viral subtype and RSV, with or without 1- to 2-week lags to account for delays between viral testing and attributable hospitalizations. Model selection was guided by the Akaike information criteria, with the best model chosen on the basis of the lowest Akaike information criteria. The final model contained an annual Fourier term; polynomial week terms (quadratic and cubic); and 2-week lagged viral terms for pandemic 2009 influenza A(H1N1), human influenza A(H3N2), influenza B, and RSV; the models were offset with the natural log-transformed number of person-weeks at risk. The final binomial regression model containing polynomial terms for time trends, annual sine (sin) and cosine (cos) harmonic terms, and 2-week lagged viral terms for percentage of weekly influenza and respiratory syncytial virus (RSV) specimens with positive test results is as follows:

where E(Y) is the expected number of cardiorespiratory hospitalizations during a given week, H1N1pdm09 is pandemic 2009 influenza A(H1N1), H3N2 is seasonal human influenza A(H3N2), B indicates influenza B, and PCR is polymerase chain reaction.

The final model was fit to obtain estimated values for cardiorespiratory hospitalizations in the presence of influenza and RSV over the entire study period. A second model was fit with weekly viral terms set to 0, representing the cardiorespiratory events expected in the absence of influenza and RSV. The difference between the 2 estimates represents the influenza- and RSV-attributable cardiorespiratory hospitalizations. Weekly influenza- and RSV-attributable events and long-stay LTCF resident person-weeks at risk were summed across the entire study window. In the primary analysis, weekly estimates for negative (ie, less than 0) influenza- and RSV-attributable events were set to 0 because of the infeasibility of negative attributable events. A stability analysis was performed without setting negative weekly events to 0 to assess the robustness of the results to this assumption. Attributable cardiorespiratory events were also estimated separately for both influenza and RSV. Incidence rates were calculated by dividing the attributable events by the total person-time at risk. Incidence rates were then scaled to 100 000 person-years. This process was repeated separately in each age group.

The overall cost in US dollars, derived from the Medicare reimbursement for hospitalization, and LOS of attributable events were estimated using information from the MedPAR inpatient claims. Claim reimbursements throughout each year were converted to 2017 inflation-adjusted dollars using the gross domestic product price index.33 Within each age group, the distributions of observed cardiorespiratory hospitalizations were selected and the top and bottom 0.5% of each distribution were excluded to remove outliers. The mean costs of cardiorespiratory hospitalizations and LOS were then taken and multiplied by the attributable events within each age group. Attributable hospitalization costs and LOS were then summed across age groups to obtain overall totals. We estimated CIs for a difference in means to obtain 95% CIs for attributable events.

Data were analyzed with SAS, version 9.4 (SAS Institute Inc) and R, version 3.5.1 (R Foundation for Statistical Computing) using the package flumodelr, a publicly available R package developed by members of our team (K.W.M., R.v.A, and A.R.Z.).34 Data analysis was performed between January 1 and September 30, 2020.

Note: The content above has been extracted from a research article, so it may not display correctly.

Please log in to submit your questions online.
Your question will be posted on the Bio-101 website. We will send your questions to the authors of this protocol and Bio-protocol community members who are experienced with this method. you will be informed using the email address associated with your Bio-protocol account.

We use cookies on this site to enhance your user experience. By using our website, you are agreeing to allow the storage of cookies on your computer.