This study is compliant with the Local Ethical Committee Standards of the Fondazione Policlinico Universitario Agostino Gemelli IRCCS. It was approved and registered with number 2018-04/23/2018 and was carried out in accordance with the Helsinki Declaration and EU Regulation 2016/679 (GDPR) concerning the processing of personal data. For this type of study, Ethical Committee did not foresee the need for participant consent.

A retrospective observational study was carried out in our teaching hospital in Rome, analyzing data from the 2017–2018 influenza season. This is an acute-care reference center with a 1,500-bed capacity and almost 5,500 workers employed.

Vaccination was offered to all staff as a part of the hospital prevention program; it was administered by the hospital’s Preventive Medicine Service and performed with quadrivalent flu vaccine Fluarix^® Tetra (split, inactivated, non-adjuvanted). During the 2017–2018 influenza vaccination campaign, training sessions on flu vaccination have been organized according to Academic Detailing methodology [21]. The purpose was to deliver tailored training and technical assistance to healthcare providers to help them using best practices on influenza prevention and control [21]. Vaccinated and unvaccinated hospital workers (health workers and administrative staff) and the absenteeism (number of working days lost for illness—namely, sick days—during the epidemic flu period) in both groups have been counted and compared. Based on the fact that exposure to other possible causes of illness (including circulating infectious diseases other than influenza, for example other respiratory viruses) was the same for both groups, it was assumed that the difference in absences was only due to influenza. Working days were considered lost for illness when the mandatory medical certificate for sick leave was sent by the worker to the hospital health management (his own employer), as requested by the Italian law. The epidemic period was considered in accordance to Influnet [22], an Italian nationwide sentinel surveillance network of influenza epidemiology that collects clinical and virologic information about influenza from week 42 to week 17 of the following year.

Descriptive statistic was performed analyzing mean and Standard Deviation (SD) of quantitative variables and frequencies and percentages of qualitative ones. Workers job categories have been grouped into physicians, nurses, administrative staff, other graduated and ungraduated health professionals. Moreover, t-Student test was used to assess differences in absenteeism due to illness between vaccinated and unvaccinated workers, considering all employees and then stratifying by job categories. The level of statistical significance was set at p<0.05.

To compare the impact of flu vaccination among hospital workers, in terms of both costs and sick days, a cost-consequence approach was used [23]. Through this methodology the impact of the vaccination on short-time (epidemic period) resource use and costs (productivity losses) and health outcomes (sick days) was estimated and presented in a tabular format. The analysis was performed through a 6-months horizon, no discount rate was applied, adjustment for a common base year was not necessary.

Direct medical costs referred to the acquisition costs of the vaccine (ex-factory price of 11.08 €) [24, 25] and the administration costs of the vaccination (set to 6.16 €) [24, 26, 27].

Indirect costs referred to costs falling outside the healthcare sector and concerned productivity lost (workdays/hours lost) due to influenza. Daily/hour wage was set in accordance with the national collective agreement [28].

The loss of productivity (linked to sick leave) for the society and for the company was estimated by using the human capital approach and the friction cost method, respectively.

The human capital approach estimates the value of lost production due to temporary absence from work for sick leave. The measure of income can be obtained through the sample estimate of individual patient income or, alternatively, through the remuneration received by workers classified by sector of activity and professional categories, detectable from surveys conducted by national institutions responsible for the management of tax revenue [29].

The friction cost method assumes that the value of lost production associated with a disease depends on the period necessary to retrieve the original level of production (during the absences the work can be done by colleagues—friction period). It has been estimated that the productivity reduction deriving from one day of absence from work is equal to about 80% of the sick worker's daily income [30]. Therefore, this value of productivity loss was applied.

Direct costs were considered only for the human capital approach, because of for the company perspective (friction cost method) the vaccine was provided free of charge by the Italian NHS and administered by the hospital’s Preventive Medicine Service without additional usage of resources and working hours.

Finally, the cost-outcome ratio was calculated by dividing the difference in total costs (vaccination costs, where appropriate, and loss of productivity) per capita among vaccinated and unvaccinated workers by the difference between the mean of absenteeism among vaccinated and unvaccinated workers.

The teaching hospital counted 5,483 hospital workers employed and all of them were included in the study. During the flu vaccination season 2017–2018, 538 of them (9.8%) have been vaccinated. In Table 1 the main sample characteristics are reported, according to flu vaccination status. Considering the job categories, physicians had the highest immunization coverage (21.9% of doctors have been vaccinated, compared to 8.5% of nurses, 10.2% of graduated health professionals and less than 5% of other categories).

Regarding the absenteeism during the epidemic flu period, a mean of 1.6 ± 6.0 days of absence from work for sick leave was observed among vaccinated and a mean of 3.3 ± 10.9 days among unvaccinated hospital workers (mean difference = 1.7; p<0.001) (Table 2). Moreover, nursing absenteeism was 2.8 ± 7.1 days among vaccinated and 4.2 ± 12.3 days among unvaccinated employee, while physicians’ absences were close to 0 for both groups in the epidemic period.

In Table 3 the results of the cost-consequence analysis are described.

The human capital approach estimated, during the flu epidemic period, a loss of productivity equal to 297.06 € for each vaccinated and 517.22 € for each unvaccinated hospital worker (cost-outcome ratio = 120.07 €/sick day). Referring to all employees, a total loss of productivity of 1,628,765 € avoided for vaccinated and 2,835,916 € avoided for unvaccinated hospital workers has been found.

The friction cost method estimated, during the epidemic flu period, a loss of productivity equal to 237.65 € for each vaccinated and 413.78 € for each unvaccinated hospital worker (cost-outcome ratio = 104.19 €/sick day). Referring to all employees, a total loss of productivity of 1,303,012 € avoided for vaccinated and 2,268,733 € avoided for unvaccinated hospital workers has been found.

This study did not analyze differences in absenteeism rate between the epidemic and the non-epidemic period, focusing only on the first one. This could be a limit because several studies reported an absenteeism that range between the 35% and 2-fold higher during epidemic period compared to the rest of the year [31, 32], despite other studies highlighted few differences [33]. For example, Gianino et al. [18] founded difference in absenteeism between vaccinated and unvaccinated healthcare workers also during the non-epidemic period and explained this discrepancy assuming that vaccines also protect employees from illness during non-epidemic period.

In addition, this study was based on a prevention practice conducted “on the field”, offering vaccination to all hospital workers without randomization or sample size calculation. Despite this, the hospital prevention program guaranteed free access to vaccination and allowed to have a much larger population in both groups.

Moreover, in this pilot study it was assumed that the difference in absences between groups during the epidemic period was mostly due to the flu vaccination intervention. Indeed, as mentioned above, we assumed that the exposition to other possible causes of illness during the epidemic period (including typical circulating infectious diseases of the wintertime, not vaccine-preventable, i.e. other respiratory viruses) was the same, considering the high number of people in the two groups too. Besides, the absenteeism has not been stratified by cause of illness because of the privacy obligations that do not allow the employer to know the diagnosis of sick leave for which the hospital workers employed were absent.

Furthermore, the study design is based on a static model. Despite this does not consider the indirect effect of vaccination on the unvaccinated individuals (a worker is more protected from infection if a colleague is vaccinated—herd immunity effect), as in dynamic model [34, 35], the low vaccination coverage rate among all job categories make this effect almost negligible [36, 37]. This study design limitation bounds the possibility of extrapolating the results by assuming the condition in which everyone is vaccinated because of in this case the indirect effect is no longer negligible.

Finally, although cost-consequence analysis is sometimes considered less rigorous than other economic evaluations, it is at the same time more versatile and practical being able to offer clear and simple information thus representing a valid framework for the appraisal of flu vaccination in the Italian context.