The author is grateful to Antonino Bella from the National Institute of Health, for providing InfluNet data and Patrizio Pezzotti, senior researcher at the National Institute of Health, for providing precious comments on the manuscript. Regional data were driven from the repository https://github.com/fbranda/influnet/tree/main/flu-season managed by Francesco Branda, University of Calabria.

Background

1Seasonal influenza is a preventable infectious disease, mainly involving respiratory symptoms. In recent years influenza epidemics have caused a heavy burden of deaths in Europe with an estimated cumulated influenza-attributable mortality among people aged ≥ 65 years. ranging from 30 deaths per 100,000 population in the winter season 2013/14 to 185 deaths per 100,000 population in the winter season 2014/15 [Vestergaard, 2017]. In Europe high rates of mortality among elderly was also observed in 2017/18 season, and this was quite unexpected in an influenza B dominated season, which commonly are considered to cause mild illness, mainly among children [Nielsen, 2019]. The 2018/19 influenza season was dominated by influenza A (H1N1)pdm09 and was accompanied by an excess of all-cause mortality observed between the second week of January 2019 and the end of February 2019 mainly affecting people aged 65 years or older and the 15–64 years age group. Similar proportions of influenza A (H1)pdm09, A(H3) and B/Victoria were detected throughout the season 2019-20. During the period of high influenza virus circulation in early January and February, a slight excess mortality from all causes was observed, possibly related to influenza, while later in the season the excess mortality was driven by COVID-19 [ECDC, 2019].

2In Italy, influenza viruses directly cause approximately 300-400 deaths each year, with about 200 deaths due to primary viral pneumonia. The number of “indirect deaths”, due to pulmonary or cardiovascular complications, related to influenza may vary from 4,000 to over 20,000, according to the type of circulating viruses and the methods adopted for estimating the deaths attributable to influenza. The first investigation on mortality attributable to influenza conducted in Italy concerned the period 1969-2001 [Rizzo, 2007]. The highest number of excess deaths was found for the 1969–70 season (107 deaths per 100,000), that was considered a “pandemic seasons”; other seasons with considerable excess deaths were: 1981–82, 1984–85, 1986–87, 1990–91, 2000–01. The influenza seasons with higher excess deaths was characterized by a predominance of influenza A (H3N2) viruses. This study showed a similar pattern of influenza epidemics in the three Italian repartitions (North, Center, South) during the 32-year study period. In recent years, Italy has been registering again high excesses of death rates attributable to influenza. Peaks of deaths were observed in the 2010/11, 2012/13, 2014/15, 2016/17 seasons [Rosano, 2019].

3The influenza season 2020-21, characterized by Covid-19 pandemic, showed a sharp decrease in influenza activity in Europe [Adhloc, 2021], America [Olsen, 2020] and Australia [Sullivan, 2020]. COVID-19 pandemic public health and physical distancing measures may have extinguished the 2020/21 seasonal influenza epidemic with just a few sporadic detections of all viral subtypes [Greco, 2021; Adhloc, 2021].

4Italy has been the first western country hit by the Covid-19 pandemic and has registered 26,2 million documented cases and 192,406 deaths related to the infection as of November 2nd, 2023. The first documented case in Italy was reported on February 18^th, 2020. In 2020, the total of deaths from all causes was the highest ever recorded Italy since World War II: 746,146, 100,526 deaths more than the 2015-2019 average (15.6 percent excess). The total number of deaths dropped to 709,035 in 2021 and 713,499 in 2022. Standardized mortality rate per 100,000 was 933 in 2020, 876 in 2021 with an excess respectively of 9.4% and 2.3% compared to the 2015-2019 period [Fraboni, 2021].

5The objective of our study was to describe the characteristics of influenza epidemics and to estimate the influenza-attributable contribution to excess mortality during the influenza seasons from 2013/14 to 2020/21 in Italy. As Italian regions have heterogeneous demographic and climate characteristics and are likely to show different epidemic patterns of seasonal influenza virus circulation, the excess mortality was investigated also at regional level.

Methods

6Population data (resident population and weekly deaths) were provided by the National Institute of Statistics (ISTAT), data on influenza like illness (ILI) and confirmed influenza cases were provided by the National Institutes of Health.

7To determine the pattern of influenza viruses circulation in each winter season (from week 42 to week 17), we obtained relevant data from the integrated influenza surveillance system (InfluNet), established in 1999 and coordinated by the Italian National Institute of Health. Monitoring of influenza is conducted through a national network of physicians who report cases from individuals with influenza-like illness (ILI) and collect samples from a subset of patients for virological confirmation. The data on influenza sub-types were considered to interpret the results (some subtypes are more dangerous than others), but they were not used for the analysis of mortality as it was not possible to distinguish between deaths due to different influenza sub-types. The number of influenza-attributable deaths (IAD) was estimated using the FluMOMO algorithm, based on the weekly Influenza Activity (IA) and Excess Temperature (ET) [EuroMoMo, 2021]. As an indicator of weekly influenza activity (IA) we adopted the Goldstein index, which is the product of the percentage of patients seen with influenza-like illness (ILI) and percentage of influenza-positive specimens, in each week [Goldstein, 2011]. Italian temperature data were extracted from the National Oceanic and Atmospheric Administration (NOAA) database. For the analysis at country level, national weekly mean temperatures were used, for the analysis at local level regional weekly mean temperatures were used.

8Several indirect approaches have been used to estimate excess mortality attributable to influenza, but the preferred approach was to model weekly mortality data by using weekly indicators of influenza activity. The FluMOMO model is a multiplicative Poisson regression time-series model with overdispersion [Nielsen, 2018], where all-cause mortality weekly absolute number was the outcome variable and influenza activity, and excess temperatures were explanatory variables. The general model is given by the following equation:

9The model uses a log link to model the number of deaths per week. IA is the influenza activity index and ET (extreme temperatures) represents the deviation of ambient temperature from expected ones. The baseline consists of a trend and seasonality expressed as two sine waves of one year and half year periods. The model included parameters to estimate the annual trend, seasonality. Analyses were performed separately for the age groups 0–4, 5–14, 15–64 and 65+ years of age, as well as for all ages. As the dominant type/subtype of influenza circulating viruses varies from season to season, a separate effect of IA for each season was used.. The model estimated both baselines, and the effects of IA and excess temperatures, at the same time. The model includes a lag-effect (d) of the explanatory variables, that is, a carry-over effect of IA or ET from earlier weeks. Length of the lags is predefined as external parameters. Up to two-weeks-delayed effects of the explanatory variables were considered in the adopted models. Further details on applied method are available in a previous publication [Rosano, 2019].

Estimation of regional IAD

10According to the NUTS (Nomenclature of Territorial Units for Statistics) classification, the analyzed administrative regions were those classified with NUTS 2 level. Estimation of regional IAD was conducted by calculating a weekly risk function as a combination of the number of observed deaths and the Goldstein index calculated at regional level. Also, for regional estimation two-weeks-delayed effects of the risk variables were considered. Regional populations data were provided by ISTAT, regional data on influenza like illness (ILI) and confirmed influenza cases were provided by the National Institutes of Health. Confirmed cases were available only for the 2019-20 winter season, thus the regional estimates of IAD were calculated only for this season.

Results

Descriptive analysis

11During the period from 2013/14 to 2020/21, a yearly average of 5,975,000 ILI cases were estimated in Italy, corresponding to a cumulative average incidence of 9.9% in the Italian population. The 2017/18 season showed the highest estimated number of cases, with a total of 8,674,000 ILI cases. The lowest number of cases was observed in the 2020/21 season, with 2,416,000 ILI estimated cases. During the last season 2020/21 no isolation of influenza virus was reported in the 6,818 analysed samples (Figure 1), conversely in 1,470 samples (21.6%) were positive to COVID-19.

12The number of ILI cases per 1000 individuals reported to the Italian surveillance system is reported in Figure 2. The highest estimated incidence of ILI cases was observed in children younger than 5 years and in adolescents, with similar patterns over the years, except for the season 2020/21.

Estimation of IAD

13The estimated total number of excess deaths attributable to influenza in the 8-season study period was 114,112 (95% C.I. 101,231-127,985). During the 2016/17 and 2017/18 seasons, the influenza-attributable excess mortality was higher compared to the other seasons. The total number of excess influenza-attributable deaths during the 2017/18 season was 24,339 (95% C.I 21,638-27,143), five times as high compared to season 2018/19; The average annual mortality excess rate ranged from 7.9 to 45.3 per 100,000, excluding the season 2020/21 when no confirmed case of influenza (and no related death) was reported (Figure 3). In the same graph the rate of deaths attributable to extreme temperatures were also reported: it is about 12,750 deaths in the study period with a peak of 5,190 death in the winter season 2016/17.

Estimation of regional IAD

14Central regions showed the highest rates of IAD, as well as Valle d’Aosta and Basilicata. Veneto and Trentino-Alto Adige showed the lowest rates of ED, respectively 2.05 and 2.19 per 100,000 (Figure 4). In absolute terms, Lombardia showed the largest number of estimated ED due to influenza (3,303), followed by Lazio (1,749), Piemonte (1,373), Emilia Romagna (1,362), Campania (1,298) and Sicily (1,229) (data not shown). The impact of IAD on deaths for all causes was, at national level, equal to 4.4%, with higher percentages in central regions (Marche, Abruzzo, Umbria and Lazio) and Valle d’Aosta and lower ones in Liguria, Friuli-Venezia-Giulia and Veneto (Figure 5).

15Circulation of flu viruses among Italian regions is monitored by Influnet through weekly bulletins reporting the regional ILI incidence, summarized in Figure 6. ILI incidence, with the data on influenza-positive specimens, reported by weekly Flu Virologic Surveillance of ISS [ISS, 2020], constitute the reference data for the estimation of IAD. Data of the 2019/20 winter season show a higher incidence of ILI in Valle d’Aosta, Marche, Lazio, Abruzzo and Molise and lower incidence in Friuli, Toscana and Sardegna.

Discussion

16Seasonal flu-like syndromes have a regular periodic trend: the number of cases begins to increase at the end of each year, reaches a seasonal peak at the end of January of the following year and lasts until March. In our study we investigated influenza-attributable mortality in the last recent years with an impressive burden of all cause deaths. Italy faced a dramatic excess death attributable to influenza during the winter seasons 2016/17 and 2017/18. Such excesses were impressive in terms of excess death rates, respectively 41.2 and 40.6 per 100 000, corresponding in terms of absolute number of deaths to more than 24,000 deaths each season. The excesses were concentrated in a few weeks and were among the highest observed in Europe [Nielsen et al., 2019], with the highest proportion of deaths occurred among persons aged 65+ (around 95%). The variability in the IAD can be interpreted through the data of the virologic surveillance of influenza. Principal influenza strains circulate in the last years were the subtypes A/H1NI, A/H3N2 and the B subtype. As in other European countries, the excess mortality observed in Italy during the 2016/17 season could be related to the circulation of an A/H3N2 influenza virus, which is associated to a higher mortality in the elderly [Nielsen et al., 2019]. In two seasons (2014/15 and 2015/16), a co-circulation of A and B viruses was observed [ISS, 2020]. In the 2020/21 season no case of influenza has been isolated, that is the first time since the influenza surveillance has been established in Italy.

17Little evidence is available on influenza-attributable mortality estimates at country level. Previous studies conducted in Denmark have analysed influenza seasons reporting similar mortality patterns to those here reported [Nielsen, 2018]. In Sweden, during the 2016/17 season, characterized by the predominant circulation of A/H3N2, influenza-related excess mortality was higher and lasted longer than in other recent seasons [PHAS, 2017]. In the UK estimates of the annual number of deaths directly attributable to flu range from 4 to 14 000 per year, with an average of around 8 000 per year [PHE, 2014]. In recent years, excess mortality for influenza in Italy has been previously explored in a multi-country study regarding the 2015/16 season [Vestergaard, 2017], in a study including data from a limited sample of the Italian population [Michelozzi, 2016], and in a study focusing on a single Italian region [Fedeli, 2017]. The only complete study on Italian data was conducted regarding influenza seasons from 2013/14 to 2016/17 [Rosano, 2019]. The studies conducted by Vestergaard and Rosano made use of the same methodological approach, the one used in present article, while the ones of Michelozzi and Fedeli adopted generic time series models (not specifically adapted to influenza epidemics characteristics). Available previous findings focussed on 2015/16 season, only one investigated four consecutive seasons. Seasonal analysis on 2015/16 showed an excess of deaths in winter of +13% attributable to the peak in influenza rather than to low temperatures.

18In Italy in the 2019/20 season the number of excess deaths attributable to influenza was 16,987 with an overall rate of 28.5 per 100,000, equal to the average rate observed over the study period, and in impact on the total deaths equal to 4.4%. Regional data concerning 2019/20 season showed that the IAD were heterogeneously distributed among Italian regions: those more affected were Valle d’Aosta and central regions (Umbria, Marche, Lazio, and Abruzzo), which showed a higher circulation of influenza viruses. Influenza coverage in 2019/20 season was 16.8% in Italy, with higher percentage in Tuscany and Friuli (above 20%) and lower percentage in Bolzano Province (8%), Valle d’Aosta (13%), Lombardia (13%) and Sardegna (15%) [Ministero della Salute, 2023]. In comparison with previous years, an increase in coverage was observed in Sicily, Marche, and Molise in particular (increase of 12.1%, 10.3% and 6.0%, respectively); the figure has decreased, however, significantly in Bolzano Province (-15.1%), Basilicata (-8.9%) and Veneto (-3.1%) [D’Ambrosio, 2021]. Data on vaccination coverage do not fully explain the regional variability in IAD, probably because the vaccine failed to protect the elderly against mortality, possibly due to immune senescence [Rizzo, 2006]. In Denmark, the 2019/20 season was characterised by a low number of excess death attributable to influenza (697, 95% C.I. 505 to 889), which was fewer than expected, pooled over all ages, which reflects a mild influenza season [Nielsen, 2021]. In most European countries, the 2019/20 winter season was characterised by low circulation of influenza viruses, mainly A(H3N2), and a relatively high vaccine effectiveness [Rose, 2020]. Conversely, in Italy, influenza viruses circulated in 2019/20, with 7.5 million of ILI cases (corresponding with a cumulative incidence of 13.1%) with an average percentage of positive cases of 23.2%, a predominance of A type (67% of positive cases), and a concurrent circulation of B type (33% of positive cases) [ISS, 2020]. The last weeks of the 2019/20 season overlapped with the start of the Covid-19 pandemic, when the influenza epidemic was waning.

19Data for the season 2020/21 showed very peculiar evidence, as no case of influenza was detected and, consequently, no death attributable to influenza was registered. Although seasonal influenza vaccination coverage has increased compared to previous seasons it cannot be the only reason for the disappearance of the seasonal flu epidemic. Several factors contributed: the increase in vaccination coverage, an immune memory effect and, probably, a viral competition in which the new coronavirus prevailed. The disappearance of seasonal influenza is mainly due to the stringent protection measures implemented to counter the spread of COVID-19: mask-wearing, distancing, hand washing, school and business closures have been very effective in eliminating the seasonal flu epidemic [Greco, 2021]. The impact of mortality attributable to influenza in Italy has been so high in the recent years, before the irruption of COVID-19 pandemic, that severely hit Italy. For the first time since the implementation of the influenza surveillance system in Italy, in the season 2020-21 there has not been a seasonal flu epidemic and no death attributable to influenza. The massive flu vaccination campaign may also have had a role. In Italy, vaccination coverage for the 2020/2021 season increased significantly compared to the previous one (23.7% vs. 16.8%) [Ministero della Salute, 2023]. Also, possible pathogenic competition between the two viruses has been argued a further reason for vanishing influenza [Stowe, 2021], but all these factors cannot account for the total disappearance of influenza [Greco, 2021]. Italy adopted drastic restrictions protection measures to counter the spread of COVID-19, such as mask-wearing, distancing, hand washing, school and business closures [Guzzetta, 2021], that may have temporarily eliminated influenza, similarly to other countries. With the end of nearly all COVID restrictions, occurred in Italy as of June 2022, the circulation of respiratory viruses has increased. In the flu season 2022-2023 the influenza surveillance system detected about 204,000 of ILI cases, that lead to an estimation of about 13,963,000 cases at national level [ISS, 2023]. The 2022−2023 influenza season marked the return of influenza virus activity to almost pre-pandemic levels in the EU countries. There are likely multiple reasons for the elevated number of influenza cases and ICU observed in the 2022/23. The impact will have been influenced by the dominant circulating strain. Influenza A(H3N2), which was the predominant subtype in 2022-23 both in Italy and in most European countries, is known to be more severe in older age groups. In addition, there is evidence to suggest that lower population immunity due to reduced flu circulation from social distancing measures during the pandemic meant that, overall, the population was more susceptible to be affected by flu, contributing to the trend observed in the 2022-23 influenza season.

Strengths and limitations of the study

20There are many different methods used for estimating the influenza‐associated mortality [Iuliano, 2018; CDC, 2019, Muscatello, 2013; Nielsen, 2018; Reed, 2015], the FluMOMO, the one we adopted, is a valuable tool for timely risk assessment of seasonal influenza and emerging threats such as the COVID-19 pandemic, as the data are readily available in many countries, and the approach is not sensitive to coding practices in cause-of-death-registers and collection of other indicators [Schmidt 2022]. Thanks to this model we have been able to provide reliable estimations of the excess deaths due to influenza in Italy, and, for the first time, also at regional level, intended as NUTS 2 level, while previous studies [Rizzo, 2007] were conducted at repartition level, intended as NUTS 1, level, or on a limited sample of cities [Michelozzi, 2016; Fedeli, 2017]. However, regional analysis was limited to the winter season 2019/20 because of reporting problems of regional data in previous years. This study has other limitations. In Italy, the influenza surveillance system is based on voluntary reporting of ILI cases by general practitioners, who are not selected with random criteria and do not constitute a representative sample. This limitation may introduce a potential bias due to the selection of subjects under surveillance. Moreover, the adopted statistical model did not include an interaction term between temperatures and IA. This prevented to test whether the association of influenza activity with mortality varied with temperatures. This drawback of the model should be overcome in future applications. One of the major strengths of the study is the census basis of mortality data, while previous published studies [Nielsen, 2019; Michelozzi, 2016; Fedeli, 2017] were based on sample data from a limited number of cities in the country.

Conclusions

21Surveillance of the impact of mortality attributable to epidemics is crucial for public health and deserves reliable and openly accessible methods of estimation. The study provided valid tools and strategies to implement a surveillance of the estimation of excess deaths attributable to influenza in Italy