The Burden of Acute Respiratory Infections (BARI) study is a multidimensional real-world evidence study assessing the clinical and economic burden of acute respiratory infections (influenza and respiratory syncytial virus) in Spain and Portugal. Here we report results for the burden of severe influenza in Portugal, measured through hospitalizations and deaths.
This publication employs the following two approaches: (i) a direct method of estimating seasonal influenza incidence, based on the number of NHS hospitalizations with influenza-specific International Classification of Diseases (ICD) codes; (ii) an indirect method of estimating excess hospitalizations and deaths using broader groups of ICD codes in time-series ecological models.
Data sources
Hospital discharge data
Anonymized administrative data on hospitalizations (January 2008–December 2018) were provided by ACSS—Administração Central do Sistema de Saúde (Health System Central Administration), which collects administrative and clinical data for all hospitalization episodes in Portuguese public hospitals, including information on diagnoses and procedures performed during hospital stay, which are coded using the ICD-9-CM and ICD-10-CM/PCS. These data were also used to describe the in-hospital case fatality risk based on discharge status, corresponding to the hospitalization episodes coded as due to influenza that resulted in in-hospital death during the respective episode.
Death certificate data
For analysis of influenza-associated excess mortality, data on daily deaths listed in the death certificate per age group and cause were obtained from the Instituto Nacional de Estatística (INE, National Statistics Institute) in July 2020. The data were aggregated into weekly counts [8]. The coding for primary death from the death certificate was used.
Influenza activity data
The primary predictor of influenza excess hospitalizations and deaths was the overall weekly incidence rate of influenza-like-illness (ILI), which was obtained from the Instituto Nacional de Saúde Doutor Ricardo Jorge (INSA, National Institute of Health Doutor Ricardo Jorge) in July 2020 [9]. This rate is estimated by the INSA based on data collected by the Portuguese General Practitioners Sentinel Network (Rede Médicos Sentinela) [10].
Demographic data
Finally, in order to compute rates of cases per 100,000 people, data on age-specific annual resident population estimates were downloaded from INE’s website [8].
Statistical analysis
An ecological approach was used to estimate the number of influenza-associated excess hospitalizations, deaths, and hospitalization costs by using cyclic regression models explicitly modeling weekly morbidity and mortality data against weekly indicators of influenza activity (moving average, broken down by season, with a one-week lag for hospitalizations and 3-week lag for deaths).
Different Poisson cyclic models (time series) were used, where age- and cause-specific hospitalization and mortality data were explained by the ILI incidence [5, 11, 12], as well as time trends and seasonal terms, using a log link [13, 14]. A stepwise selection method was used to identify significant time trends and seasonal terms in the age- and cause-specific models.
Baseline hospitalization and mortality data were calculated from Poisson models as model-expected values when the ILI variable was set to zero. The weekly number of influenza-associated excess hospitalizations and mortality was estimated as the difference between expected hospitalizations and deaths estimated by the Poisson model, incorporating the ILI incidence rate as an indicator of influenza activity and the number of hospitalizations and deaths estimated by the baseline model not taking into account this indicator. The number of influenza-associated hospitalizations or deaths was defined for each epidemic season by the sum of the weekly excesses. The seasons were defined from September to June of each year, as defined for the Northern Hemisphere.
The performance of the model was measured by the correlation between the values predicted by the model and those observed using Pearson’s correlation. The mean absolute percentage error (MAPE) was also computed as the average, for all weeks, of the percentage difference between predicted and observed values.
For the excess hospitalization and mortality analysis, the hospitalizations and deaths were organized into four categories according to cause (P&I, respiratory, respiratory, or cardiovascular, and all-cause). Regarding age groups, the population was stratified thus: 0–4 years old, 5–18, 19–49, 50–64, 65–74, ≥ 75; and ≥ 65 years old.
Confidence intervals of 95% were calculated for the estimated influenza-associated excess hospitalizations and deaths by cause and age group.
The estimates of influenza-associated excess hospitalization and mortality were compared to the numbers of primary or secondary influenza‐specific diagnoses captured from NHS hospital discharge records, as well as fatalities observed during these episodes.
Cases per 100,000 people were computed by dividing the estimated influenza cases by the Portuguese population in the respective age group.
Means are reported only for nine seasons, excluding the H1N1pdm09 pandemic (2009/10).
Case definition
Hospitalizations coded as due to influenza
Data were extracted from administrative databases from 2008 to 2018 that contained all NHS hospitalization records in mainland Portugal. Influenza episodes were defined as those coded with ICD-9 487 or 488; or ICD-10 J09, J10, or J11 in any primary or secondary diagnosis field. Total hospitalizations excluded admissions related to routine birth, planned activity, and those in which the primary diagnosis was related to musculoskeletal, alcoholic, or mental disease. Additional diagnostic information during the influenza episode was also used to identify individuals who had at least one medical condition regarded as a risk factor for severe influenza (Additional file 1: Table S1) in any primary or secondary diagnosis field, considering the following conditions: pregnancy, diabetes mellitus, respiratory or lung disease, cardiovascular disease, immunocompromised, chronic liver disease, and chronic kidney disease.
Excess hospitalization and mortality
Influenza-associated excess hospitalization was computed for four groups of diagnoses [15, 16], according to the primary diagnosis, namely pneumonia or influenza (P&I, ICD-9: 480–488, 517.1; or ICD-10: J09–J18), respiratory (R, ICD-9: 460–519; or ICD-10: J00–J99), respiratory or cardiovascular (R&C, ICD-9: 390–459, 460–519; or ICD-10: I00–I99, J00–J99), and all-cause (any ICD-9/10 diagnosis). Influenza-associated excess mortality was computed for the same groups of diagnoses according to the cause of death listed on the death certificate.
Cost estimation
Hospitalizations coded as due to influenza
Only direct costs were estimated using a diagnosis-related group (DRG)-based budget allocation model. Hospitalization costs were computed by multiplying each cost weight, considering the DRG of each episode, with the Portuguese fixed cost multiplier and funding price applicable for the 2018 year, as defined by the ACSS.
In Portugal, although there is no detailed cost information at the patient level [17, 18] and the provision of health care is mostly public, with no need to create routines or billing procedures in the definition of costs per DRG, the Maryland matrix was applied [17]. This matrix makes the relative correspondence according to North American standards of care provision between the care provided during the hospitalization episode and assigns relative weights that reflect the costs per service for each DRG [18]. Although this methodology has a few limitations, such as the assumption that the pattern of resource use is similar to the American one, according to Bentes et al., it seems to be the most cost-effective method to identify costs per DRG, having been in use since the implementation of DRG in Portugal in the 1980s [19, 20].
Excess hospitalization
The cost of excess hospitalization was computed by multiplying the number of estimated excess hospitalizations by the mean cost per hospitalization, by cause and age group.
Ethical considerations
The study was conducted following the ethical principles of the Declaration of Helsinki and as per local regulations, including privacy laws. Data were provided anonymized and may be used for research purposes without the approval of an ethics committee or informed consent. In addition, the protocol of the BARI study was validated by a panel of clinical experts, classified by the Agency of Medicines and Medical Devices (AEMPS) as an observational study and approved by the Ethics Committee of Hospital Clinic de Barcelona (HCB/2020/1132), who waived the need for participant consent.