We conducted our study in people living or working in Berlin during the influenza season of 2006/07, using a multicentre, prospective cohort design.
There were 11 study sites: three hospitals, two administrative centres (the Robert Koch Institute and Vivantes Healthcare administrative centre), four blood-donation centres and two colleges. HCWs and some non-HCWs were sought from the hospitals, but only non-HCWs were sought from all other sites.
We recruited participants through occupational health services in the hospitals and one administrative centre; through direct recruitment at blood donation centres; and through active recruitment during site visits at the other study sites (non-healthcare workers from one administrative centre and two colleges).
On recruitment before the influenza season, consenting participants gave a single serum sample and completed an exposure questionnaire. Details recorded included age, sex, type of employment, risk factors for influenza, smoking status, and vaccination status. Where participants had been vaccinated fewer than 14 days before the sample was taken, or were found to have been vaccinated shortly after the initial sample was taken, a second sample, taken at least 14 days after vaccination, was sought through direct recall or through occupational health.
Blood samples were refrigerated, then transported within three days to the national reference laboratory for influenza at the Robert Koch Institute in Berlin, where they were centrifuged and frozen.
Inclusion and exclusion criteria
HCW in our context were defined as people working on a daily basis with unwell patients in an acute hospital setting, including nurses (trainee and qualified) and doctors. Non-HCW were those working or studying at the study sites, or attending the blood donation centres, who did not fit the definition of HCW.
Exclusions (applied to both HCWs and non-HCWs) were: people with patient contact in the community (such as community doctors and nurses, dentists, and pharmacists); people working in care homes; laboratory workers who had contact with respiratory samples or with influenza virus; people who planned to be away from Berlin for more than two weeks during the projected season (January to April); and people who did not wish to be contacted weekly by mobile phone Short Message System (SMS) or email.
For the analysis of serologically-confirmed infections, we excluded vaccinated individuals where the baseline serum sample was taken fewer than 14 days after vaccination, or where a later second sample was not obtainable (see recruitment above). These participants were excluded from the serological analyses as any titre rise could have been caused by vaccination and not infection.
Because pigs can carry influenza viruses, and participants with pig contact were associated with SCII, participants reporting contact with pigs were excluded.
Active surveillance for respiratory infections during influenza season
In order to document weekly occurrences of respiratory infections during the influenza season, we contacted all participants weekly through SMS or through email, asking them if they had experienced a "new" respiratory infection during the previous 7 days. Where participants answered "yes", they were contacted by telephone and details of their infection were obtained using an illness questionnaire.
Weekly surveillance for respiratory infections covered the period from January 13, 2007 to March 30, 2007, a period chosen to coincide with the influenza season. This strategy was employed in order to maximise the predictive value of a positive answer.
Postseason follow-up
After the influenza season, participants were recalled by SMS or email. They gave a second serum sample and completed a further questionnaire, including repeat questions on vaccination status and employment type; number of patient contacts on a typical day between January 13 and April 6, 2007; broad age classification of patient contacts (adults (>17 years) and children (<18 years); clinical specialty and usage of facemasks (for HCWs); daily professional and household contacts; use of public transport and car ownership; contact with pigs (for veterinary students); and vaccination in previous years. Participants were asked again if they had had respiratory infections over the period January 13, 2007 to 6th April 2007, extending the period of surveillance to one week after the last weekly SMS was sent.
Contact was defined as either touching or having a two way conversation with someone close by, or (for patient contact only) examining or giving care to a patient. Participants were asked to estimate the number of contacts made during a typical day in household and work settings.
Laboratory investigations
The paired blood samples were defrosted and antibody titres were determined on the same day using the haemagglutination inhibition test to determine infection in any of the two A subtypes or the two B lineages. We tested for antibody to the following strains:
A/H1: A/New Caledonia/20/1999 (H1N1)
A/H3: A/Wisconsin/67/2005 (H3N2), and A/California/07/2004 (H3N2)
B, Victoria lineage: B/Malaysia/2506/2004
B, Yamagata lineage: B/Jiangsu/10/2003.
Titres of below 10 were assigned the value of 5 in order to allow calculation of the titre rise. SCII was defined as a fourfold or greater titre rise between pre- and post-season samples, with a postseason titre of at least 40. As A/Wisconsin/67/2005 (H3N2) and A/California/07/2004 (H3N2) were closely related a titre rise to either of these strains was considered as a single SCII due to A/H3N2.
Outcomes
The primary outcome was evidence of SCII by any of the above strains.
The clinical outcomes were influenza-like-illness (ILI), defined as an illness with an acute onset, self-reported fever, cough, and head or body pains; and acute respiratory infection (ARI), defined as any reported infection with coryza (nasal discharge) or cough. Clinical outcomes were based only on completed illness questionnaires or postseason illness reports, not on SMS or email replies, the latter being used only as the prompt for collection of illness data.
Where illness was reported over more than one week, symptoms for each week were combined to produce a single illness episode. To produce an epidemic curve of SCII with any ILI, or in its absence, another ARI, the respective dates of illness onset were plotted. Where more than one illness episode was reported, we used the onset date of the episode closest to the peak, on the assumption that this episode was the most likely to be due to influenza virus.
Statistical analysis
We undertook bivariate analyses for all binary exposure variables and calculated risk ratios (RR), their 95% confidence interval and p-values. For the comparison of the exposure groups (HCW vs. non-HCW) continuous variables were analysed using the Kruskal-Wallis test. For the analysis of the association with SCII continuous variables were explored by grouping them in categories. We regarded a p-value of less than 0.05 as statistically significant. We then constructed a multivariate logistic regression model (logistic command, STATA [StataCorp. 2007. Stata statistical software: Release 10. College Station, TX: Statacorp LP]) using variables which were associated with the outcome with a p-value of less than 0.1 in the bivariate analysis. Variables with p values between 0.1 and 0.2, along with healthcare worker status, were also tested in the final model.
In order to determine whether the site of recruitment had any group-level effects on the model, we constructed a random-effects logistic regression model with the same variables as the standard model plus study site as the grouping variable, and a likelihood ratio test for the proportion of variance attributable to the group level (rho) was performed (xtlogit, STATA).
Data protection and ethical approval
The data protection protocol was approved by both state and national data protection offices in Berlin, and shared with all study partners. Ethical approval for the study was obtained from the University of Berlin, faculty of medicine (Charité) ethics committee.