Study design and study population

Eskilstuna, an industrial town in central Sweden with around 98,000 inhabitants, was chosen as the study site due to its circumscribed and stable population who both reside and work in the community, and for its variety of manufacturing industries. Inclusion criteria were, besides living in Eskilstuna, being gainfully employed and aged 25–64 years. A representative sample of this particular target population was created by Statistics Sweden through record-linkage between its Labour Market Register and the Total Population Register. To achieve a premeditated sample size of 2,200, based on the low observed participation rates in earlier population-based infectious disease surveillance studies with a similar commitment [21], postal invitations were sent to an age- and sex-stratified random sample of 14,008 individuals in August 2011. In particular, Bexelius et al. [21] found a moderate under-representation of men, people with low education, and people representing one-person households, and a more marked under-representation of the age stratum 18–39 years. For this reason, men and subjects aged 25–44 years were over-sampled in our study. Fig 1 shows an overview of the study design, including the infectious disease surveillance scheme.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Overview of the study design, including the infectious disease surveillance scheme.

https://doi.org/10.1371/journal.pone.0217012.g001

The invitees registered themselves as participants on the study’s website, via a specially designed interactive voice response telephone service (IVRS), or by sending in a postal response form. Individuals who stated, in response to the invitation (n = 119) or the first questionnaire (n = 69), that they were not currently working (unemployed, retired, students, on long-term sick or parental leave) were excluded due to non-eligibility. After one reminder, the final cohort comprised 2,237 participants. Fig 2 shows the study flow diagram of recruitment, enrolment and follow-up.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Study flow diagram of recruitment, enrolment and follow-up.

https://doi.org/10.1371/journal.pone.0217012.g002

The study protocol was approved by the Regional Ethics Review Board in Stockholm, Sweden (dnr 2011/360-31/2). All participants gave their written informed consent.

Assessment of exposures

Participants completed five questionnaires (Table 1) covering a wide range of suspected determinants for exposure to and transmission of common viral infections, as well as factors tentatively linked to individual susceptibility and propensity for reporting. The English version of the questionnaires is included as supplementary material (S1–S5 Files). The first four were offered in both formats, either web or paper. Data on dietary intake and physical activity were assessed in the fifth questionnaire, only available on the web; for this purpose the two validated interactive questionnaires, Active-Q and MiniMeal-Q were used [22, 23]. All questionnaires were available immediately after entering into the study for subjects choosing to respond on the web; for subjects who preferred the paper version, the first four questionnaires were sent out monthly one-by-one over the first half of the follow-up period. After enrolment, an experienced occupational hygienist classified the occupation using the Swedish Standard Classification of Occupations from 1996 (SSYK 96 on 4-digit level, corresponding to the International Standard Classifications of Occupations 1988) and classified the industry using the Swedish Standard for Industrial Classification (SNI 92 on 3-digit level). An occupational classification is based on jobs and tasks performed whereas an industrial classification is based on economic sector of the company [24].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Description of the 5 questionnaires sent to the 2,237 participants enrolled in the cohort and their response rate.

https://doi.org/10.1371/journal.pone.0217012.t001

Follow-up and outcome assessment

The participants were instructed to self-report, on their own initiative, all onsets of fever, upper respiratory tract infection, and gastroenteritis immediately as they occurred during the entire study period (August 22, 2011 until May 31, 2012). Frequent reminders, together with monthly newsletters encouraged high compliance. The newsletters contained feedback about observed trends in occurrence of the investigated viruses, information about the study and the tasks to be carried out by the participants, along with plenty of “infotainment” related to infectious diseases. Passive follow-up of the kind practiced in the present study, relying on self-initiated, event-driven infectious disease reporting, has been thoroughly evaluated from various aspects, including validity and completeness of reporting [21, 25]. Despite substantial under-reporting, this seems to be remarkably constant over time so that adjustments using a constant correction factor would potentially restore validity of incidence rates. In additions, epidemic curves derived from self-reporting correlate well with curves obtained from the sentinel surveillance curves [21].

Reported infections and PCR-based analyses of 14 viruses

In total, 1,733 disease reports were recorded in the present study and 1,843 nasal swabs were received, of which 48% tested positive for one or more of the analyzed viruses [32]. In total, 1,461 sick reports could be merged with the laboratory data.

It is not surprising that the number of nasal swabs exceeded the number of disease reports. Some individuals, indeed, enjoyed the possibility to have a quick feedback on their test results and to get to know more about the virus description and its associated disease but they did not contribute to the study with their disease reports. This explains part of the 382 swabs that could not be matched, the remaining part is due to timing issues: when the number of days between arrival of the specimen to the lab and disease onset was greater than 15 or smaller than -5 (case where the specimen was sent before the registered onset of disease) laboratory results were not linked to symptoms reported. On the other hand, 272 sick reports could not be merged to nasal swabs. Apart from the same issue of timing, this fact might also be explained by certain symptoms reported. The absence of a nasal sample correlates well with the presence of vomiting and the absence of runny nose.

Further details on specimen results can be found in our Eurosurveillance paper [32]. Briefly, when analyzing the seasonality, we found that the peak in the number of returned swabs was reached in the last week of September 2011, but the proportion of positive tests increased from mid-November until April. Among all samples tested, rhinovirus was the most common virus diagnosed (20.8%), followed by viruses belonging to coronavirus group (16.2%) and influenza (4.8%).

The average number of reported infections and virally diagnosed infections during the 9-month follow-up is reported in Table 3 stratified by the 10 largest occupational sub-major groups and gender. The differences in the number of reported infections were larger among the different occupational sub-major groups than between men and women. The occupational sub-major groups that reported the highest number of infections were those including health care workers (SSYK96 code 32) and preschool (SSYK96 code 33) and primary teachers (SSYK96 code 23) working with younger children. Corporate managers (SSYK96 code 12) reported instead the lowest numbers of infections.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Descriptive data on the number of women and men in the 10 largest occupational sub-major groups as well as the average amount of reported infections and virally diagnosed infections during the 9-month follow-up.

https://doi.org/10.1371/journal.pone.0217012.t003

Diet and upper respiratory tract infections

In one paper, published in 2017, we investigated the relationship between diet and upper respiratory tract infections (URTI) [33]. Using data from the web-based food frequency questionnaire we found an inverse association between intake of vitamin C, vitamin E, docosahexaenoic (DHA) and arachidonic acid (AA) and risk of URTI among women, while intake of vitamin E and zinc was associated with an increased risk of URTI among men [33].

Physical activity, sleep, and upper respiratory tract infections

In a recently published analysis, we used data from the SWEDE-I cohort to investigate the relationship between sleep, physical activity and occurrence of URTI [34]. The fitted hurdle regression models did not provide support for an association with either physical activity or sleep habits.

Strengths

(I) Advantage was taken of a newly developed and validated method for infectious disease surveillance through passive follow-up of a representative cohort, with self-initiated, event-driven infectious disease reporting via automated channels. In addition, the follow-up covered the entire Swedish season of respiratory infections. (II) Viral testing was done on a large scale. Unique information was obtained about the dynamics of viral agent-specific outbreaks/epidemics occurring during nine months in a representative population. Long-term storage of the collected samples allows possible subsequent in-depth molecular studies. (III) Unlike traditional approaches in infectious disease epidemiology, where focus is placed on identifiable encounters with infected individuals, resulting in dichotomous viral exposure assessments, the assessments in this study are probabilistic in nature, based on observed, period-specific incidence rates in the population and self-reported contact patterns. (IV) The extensive set of covariates not only enhances control of confounding and imputation of missing values, but also provides new quantitative information about associations of a large number of occupational and non-occupational factors (e.g. air temperature, dust, hand washing, long sleep, exercise) with risks of common viral infections.

Limitations

Limitations worth highlighting include uncertain external validity (I) and possible threats to internal validity (II).

1. (I). First, the study was confined to one medium-sized Swedish industrial town. Second, no more than 16% of the invited individuals participated. Selecting subjects from a restricted source population, however, might enhance feasibility of the study and increase completeness of follow-up [35]. In a small community with high participation density it is also easier to keep participants alerted to their reporting responsibility since information about the study tends to propagate by word-of-mouth. Given the significant commitment required, the low participation rate was expected. In addition, to preserve internal validity, the invitees were discouraged from signing up if they doubted that they were able to completely fulfil the commitments [21]. In the present study, the expected under-representation of men and of the age stratum 25–44 years was compensated for in the recruitment; despite that, the proportion of men and individuals in the age stratum 25–44 is still lower compared to the working age Swedish population in 2011. An older age, a higher degree of education and a predominance of females in respondents compared to the total population have been previously reported also in other epidemiological studies [36, 37] and it might cast doubt on the generalizability. It has been argued, however, that a skillful control for confounding and the understanding of mechanisms of effect [38] might be even more important than representativeness of the sample. Additionally, unintentional non-representativeness might cause greater bias when the probability of participation in the study is affected by the outcome [39], but we have no reason to believe that a common outcome as viral infections is linked to decision to take part in the study. Lastly, the potential lack of representativeness due to self-selection has to be balanced against the potential bias from poor response to follow-up in a more representative sample [39].
2. (II). Since assessment of exposures was only made once, there may be non-differential misclassification of the more relevant exposures that occurred in the few days that preceded the onset of disease episodes. From a purely biological point of view, this may lead to underestimations of the strength of causal associations, but since aspects of the work environment that are potentially modifiable tend to be more stable over time (e.g. working with young children, working in hospitals, working indoors/outdoors, being in contact with several people…), associations that are of interest from a public health perspective may be less liable to underestimation. Given the prospective nature of the study, exposure misclassification that is differential in regard to the outcome is unlikely. However, since participants did not answer the risk factor questionnaires before the start of follow-up, but rather spread them out over the first months, there is some scope for reverse causation. As all responses are recorded with a time stamp, there are ample possibilities to do sensitivity analyses, where each individual’s start of follow-up is shifted to the point when all the relevant exposure information had been registered. Misclassification of the outcome is another internal validity issue. The total number of reported infections might be inaccurate, in particular we observed that individuals with more missing answers in the exposure questionnaires had a lower number of infections reported. The significant commitment required may have caused in some individuals reporting fatigue which, as a matter of fact, may result in fewer or zero infections being reported. In addition, it is not possible to distinguish between a zero occurrence and a nonresponse, but methods to account for the presence of excess zeros are available [40]. Extensive evaluations of the validity of self-initiated, event-driven infectious disease reports have been done in two previous population-based surveillance cohorts, each followed during one infectious disease season [25]. Briefly, while the specificity was very high (essentially no false positive reporting), there was substantial under-reporting. The estimate was 60%, but since the reference standard measurements (retrospective reporting at the end of the surveillance period) were likely over-estimating the number of disease events due to so called telescoping bias, the true extent of under-reporting is likely to be of the order of 50% [25]. The constancy across sociodemographic strata in the validation studies allays the obvious concern that the under-reporting might be differential in regard to the studied exposures. Some further comfort inheres in the rule of thumb that independent non-differential disease misclassification with perfect specificity will not bias the risk-ratio estimate [41].