Table of Contents
Aims and objectives
This study aimed to estimate total hospitalizations and the hospital costs borne by the Italian National Health Service (NHS) associated with influenza at the national and regional levels, from season 2008/09 to 2018/19.
More specifically, the project had the following objectives:
-
i.
To estimate the total number of hospitalizations associated with, including those recorded as such and the estimates of excess hospitalizations associated with influenza, in all Italian regions, that is 19 regions and two Autonomous Provinces (AP) i.e., Bolzano and Trento from the Trentino Alto Adige region, (a total of 21 territorial units, which characterize the Italian territorial organization, hereafter named regions), and by age groups.
-
ii.
To estimate the economic burden (in terms of hospital costs from the Italian healthcare system) of hospitalisations associated with influenza, at the national level.
-
iii.
To estimate the in-hospital mortality associated with influenza.
Study design
This is a 11-year excess modelling study that used data administrative data collected retrospectively, and included all patients admitted to a hospital for influenza or causes associated with influenza from season 2008/09 to 2018/19, each season going from week 27 of the first year and week 26 of the following year, for a total of 574 weeks.
Subjects
The study included all patients, of all ages, who had been admitted to any Italian hospital with a valid ICD-CM-9 diagnosis corresponding to influenza or related diagnoses. Hospital admissions were identified based on selected ICD-CM-9 codes (see Table A1) and aggregated for each epidemiological week, region, age groups of 0–4, 5–14, 15–64, and ≥ 65 years, and sex. The choice of age categorisation was dependent on the data structure of influenza incidence (see “Data” section).
Data
Hospital discharge data
Access to Hospital Discharge Records (HDR, Scheda di Dimissione Ospedaliera [SDO]) requires authorisation from the Ministry of Health. Following the introduction of the 2016/679 European Union (EU) Regulation (known as General Data Protection Regulation (GDPR)), the Ministry of Health has adopted a new procedure for the extraction of such data. The authorisation request was submitted to the Ministry of Health on December 2021, requesting the extraction of weekly SDO data for the period 2008–2019, including the information listed in Table A2 of the Additional file 1.
We considered only those hospitalisations with a length of stay > 24 h and primary diagnosis codes related to influenza, based on literature (see for example [21, 23,24,25,26]) and further validated by an expert in influenza epidemiology. These comprised: i) influenza (ICD9 code 487): the hospitalizations with primary code 487 are named “observed” throughout the paper, as they represent the known burden of disease; ii) pneumonia other than influenza (ICD9 480–486); iii) respiratory diseases (codes) other than influenza, including 31 respiratory diseases (ICD9 460–466, 480–487, 490–496, 500–508, 510–516, 518); iv) cardiovascular diseases including 17 diagnoses (ICD9 410–414, 422, 427, 428, 430–435, 437, 438, 440); v) other possible associated diagnoses including diagnosis codes related to the nervous system and sense organs (ICD 322, 323, 341, 357, 382); infectious and parasitic diseases (ICD) 040); endocrine, nutritional and metabolic diseases, and immunity disorders (ICD9 250); diseases of the musculoskeletal system and connective tissue (ICD9 728, 729); and congenital anomalies (ICD9 747). All ICD-9-CM codes and corresponding diagnoses are listed in Table A1 of the Additional file 1. We also considered subgroups of these codes related to influenza in a stepwise approach, to enhance comparability with existing published results based on data and techniques similar to those adopted here, in which multiple case definitions and discharge diagnoses have been used to identify potential influenza-associated hospitalizations [27]. We looked first at the codes of pneumonia and influenza (ICD9 480–487), then these codes plus those of the remaining 31 respiratory diseases (ICD9 460–466, 480–487, 490–496, 500–508, 510–516, 518), finally including 17 cardiovascular diseases (ICD9 410–414, 422, 427, 428, 430–435, 437, 438, 440).
For each hospitalization, date of admission and discharge are recorded in the HRD. We, therefore, estimated the average length of stay for admissions with primary code of influenza and other groups of diseases, for each season and for the 11- and latest 3-season average, as shown in the result “Hospitalisations associated with influenza” section.
Hospitalization-related costs are encoded according to the national diagnosis-related group (DRG) ICD-9-CM version for hospitalizations. This system is currently employed in Italy as an instrument for financing the hospital structures in the national health system. Each DRG is associated with a tariff that reflects an estimate of the average cost of each admission, and we used this information to estimate hospitalization costs.
The HDR also contain information on the type of discharge, from which it is possible to know whether the hospitalization ended due to an in-hospital decease.
Influenza activity data
The data on the flu incidence at a regional level and by age group were obtained from the national epidemiological and virological surveillance system for influenza (InfluNet), coordinated by the Istituto Superiore di Sanità (ISS), in collaboration with the Interuniversity Center for Influenza Research (CIRI) of Genoa and the support of the Ministry of Health [28]. The incidence measures influenza-like illness (ILI) and is expressed as the number of influenza syndromes (cases) per 1,000 patients per epidemiological week. The system of weekly monitoring is based on syndromic surveillance relying on a network of sentinel physicians consisting of General Practitioners (MMG) and primary-care Paediatricians (Pediatri Libera Scelta [PLS]), recruited according to regions, who report cases of influenza syndrome observed among their patients. InfluNet defines an individual with "influenza syndrome" as any person who presents with a sudden and rapid onset of either general symptoms (fever, feverishness, headache, malaise, exhaustion, or muscle aches) or respiratory symptoms (cough, sore throat, or shortness of breath). Data on the incidence of influenza syndrome are available for the period between the 42nd week of a year and the 15th/17th week of the following year (except for 2009, in which the survey was conducted throughout the year during the pandemic influenza due to the H1N1 virus). We attributed a value of 0 for the ILI incidence of the missing weeks, following the approach previously adopted in a study using the same data [22].
Temperature data
Temperature trends are among the factors that contribute to the spread of the influenza virus [29] and have been considered predictors of hospitalisations associated with influenza in other studies [30]. Data on average temperature were obtained from the National System for the Collection, Processing, and Dissemination of Climate Data of Environmental Interest (Sistema nazionale per la raccolta, l’elaborazione e la diffusione di dati Climatici di Interesse Ambientale [SCIA]) monitoring system coordinated by the Higher Institute for Environmental Protection and Research (Istituto Superiore per la Protezione e Ricerca Ambientale [ISPRA]) [31]. The national and regional weekly average temperatures were estimated as the average of temperatures measured by about 1,000 weather stations belonging to the monitoring system network.
Demographic data
Finally, to compute the rates as cases per 100,000 people, data on age-specific seasonal resident population estimates were downloaded from the Institute of National Statistics (ISTAT)’s website [32].
Outcomes
According to the research objectives, the outcomes of the study are the following:
-
Total number of hospitalizations associated with influenza: this is given by the sum of “observed” hospitalizations due to influenza, i.e., those recorded with primary ICD9 code 487, and “excess” hospitalizations, that are those estimated by the model described in “Data analysis” section, that are hospitalizations recorded with primary ICD9 codes different from 487, but considered associated with influenza based on the model assumptions.
-
Cost of hospitalisations associated with influenza: this is an approximated estimate since it refers to excess (i.e., estimated) rather than observed hospitalisations. Hence, we hypothesised a range of possible attributable costs, ranging from prudential estimates, i.e., attributing to hospitalizations associated with influenza the same cost of hospitalizations coded with primary diagnosis influenza, to wide scope-estimates considered more realistic, where the average cost of hospitalizations for cardio-respiratory diagnoses is attributed to hospitalizations associated with influenza. See “Economic burden of hospitalisations associated with influenza” section.
-
Total in-hospital mortality associated with influenza: this is given by the sum of “observed” in-hospital deaths for hospitalizations due to influenza, i.e., those deaths recorded in hospitalizations with primary ICD9 code 487, and in-hospital deaths attributable to influenza, that are those estimated by the model described in “Data analysis” section, that come from hospitalizations recorded with primary ICD9 codes different from 487, but considered associated with influenza based on the model assumptions.
Data analysis
Hospitalisations associated with influenza
For the hospitalisation model, we relied on the approach developed in existing literature applying regression modelling techniques to estimate influenza-associated hospitalizations and mortality. Such approach was first developed for estimating influenza-associated mortality in the United States (US) [33] and has been widely applied ever since [24, 34, 35], then modified to estimate the numbers and rates of influenza-associated hospitalisations [23]. The most common regression techniques used for analysing the count of rare, repeatable and independent events are Poisson and negative binomial models [27]. Here we used an age-specific negative binomial regression model with a log link. At the national level, the age-specific negative binomial regression models that we used can be written as
$$Log(E(Y))= {\upbeta }_{0}+{\upbeta }_{1}[{\text{t}}]+{\upbeta }_{2}[\mathrm{sin }(2\mathrm{t\pi }/52.18)]+{\upbeta }_{3}[{\text{cos}}(2\mathrm{t\pi }/52.18)]+{\upbeta }_{4}[{\text{I}}]+{\upbeta }_{5}[{\text{T}}]$$
(1)
where Y represents the number of hospitalisations during a particular week for the group of diagnoses associated with influenza. Specifically, hospital admissions with ICD-9-CM diagnosis code of influenza (code 487) were excluded from the model (parameter Y), as such hospitalisations are assumed to be associated with influenza, and thus, should not be used for modelling [22, 23]. t is the number of weeks in a time series from October 2008 (week 27, numbered 1) to April 2019 (week 26, numbered 574); I is the incidence of influenza syndrome expressed as the number of cases per 1000 patients per week; and T is the average weekly temperature. The estimated coefficients are as follows: β0 is the intercept; β1 accounts for linear time trends (in weeks); β2 and β3 account for the seasonality of hospitalisations; and β4 and β5 are coefficients associated with the incidence of influenza and temperature, respectively (coefficient estimates of the negative binomial regression for all ICD9 codes considered, all ages, at national level are available in Table A3 of the Additional file 1).
The model fitting included testing the various lags and moving averages of influenza indicators to account for delays between disease onset and hospitalisation. We selected the best model based on the Akaike Information Criterion (AIC) [34] (see Additional file 1 Table A4) estimated for the models considering all-age hospitalisations for all selected ICD9 codes associated with influenza, at the national level.
Thereafter, we applied the selected negative binomial model to the age-specific (0–4, 5–14, 15–64, and ≥ 65 years) and region-specific time series (i.e., considering age-specific and region-specific hospitalizations and ILI incidence time series). Furthermore, the models were replicated considering as Y various diagnosis subgroups of hospitalisations based on the ICD-9-CM code described above (not including ICD9 code 487), specifically respiratory (ICD9 460–466, 480–486, 490–496, 500–508, 510–516, 518), and cardiorespiratory diagnosis groups (ICD9 for respiratory diseases plus ICD9 410–414, 422, 427, 428, 430–435, 437, 438, 440).
Outcome measures obtained from the model include numbers, representing the number of excess hospitalizations associated with influenza, and rates, expressed per 100,000 population, where population is the age- and region-specific population in each season. Specifically, to estimate the number of hospitalisations not directly associated with influenza, the hospitalisations predicted using the model(s) just described were subtracted from those estimated with the same model in which the variable relating to the incidence of the influenza syndrome was set to 0 (baseline). In this way, an attempt was made to separate hospitalisations for respiratory, circulatory and other pathologies from those related to patients not infected by the influenza virus [22, 33].
Finally, as a sensitivity analysis, we replicated the above model by additionally including admissions with ICD-9-CM primary diagnosis code of influenza.
Economic burden of hospitalisations associated with influenza
To assess costs related to influenza-attributed hospital admissions, we approximated their estimate since we deal with estimated rather than observed hospitalisations.Footnote 1 We estimated a range of costs for hospitalisations associated with influenza, going from conservative to more wide scope estimates. Prudential estimates were meant as the minimum attributable cost and estimated applying the mean hospitalisation cost for influenza (ICD-9-CM code 487) to the estimated number of influenza-associated excess hospitalisations. Then, a wide-scope value was estimated by applying the mean hospitalisation cost for cardiorespiratory diseases. With either assumption, the total cost of hospitalisations for influenza was defined as the sum of the cost of observed hospitalisations with a primary diagnosis of influenza and the cost of excess hospitalisations associated with influenza, for which we considered the two approximations just described, using the mean hospitalization cost for influenza, likely to underestimate the real economic burden of hospitalizations, and a wide scope estimate, considering mean hospitalization cost for cardiorespiratory hospitalizations.
In-hospital mortality associated with influenza
Influenza-associated death have mostly been studied in literature based on national mortality data, which refers to the overall mortality associated with influenza rather than the in-hospital mortality. One study [36] considered in-hospital mortality as a type of influenza-associated critical illness hospitalisation and provided an estimate by re-adapting the abovementioned method.
In our study, we estimated Eq. (1) with Y representing the number of in-hospital deaths recorded for admissions with primary codes of influenza and influenza-related diseases as listed in Table A1. All other parameters are the same as described above. Consistently, to estimate the number of in-hospital deaths not directly associated with admissions due to influenza, the deaths predicted using the model just described were subtracted from those estimated with the same model in which the variable relating to the incidence of the influenza syndrome was set to 0 (baseline).
As for excess hospitalizations associated with influenza, outcome measures of excess in-hospital mortality associated with influenza include numbers, representing the number of in-hospital deaths associated with influenza, and rates, expressed per 100,000 population.
All statistical analyses were performed using Stata 17 software (StataCorp LP, College Station, TX, 2015).
Patient and public involvement
Patients or the public were not involved in the design, conduct, reporting, or dissemination of our research.
