Are healthcare workers more likely than the general population to consult in primary care for an influenza‐like illness? Results from a case‐control study

Abstract Background Healthcare workers are at increased risk of contracting influenza. However, existing studies do not differentiate professional categories or domains of the healthcare system that are most at risk. Methods This case‐control study compared proportions of patients with professional activity in the healthcare system between cases consulting their primary care physician for an influenza‐like illness (ILI) and controls from the general patient population of the same practices of the Swiss sentinel network. Influenza was confirmed by rRT‐PCR in a subset of practices. Analysis used a mixed logistic regression model, including age and sex as potential confounders. Results During the 2018/2019 influenza surveillance season, out of 4287 ILI cases and 28 561 controls reported in 168 practices, 235 (5.5%), respectively 872 (3.1%), were active in the healthcare system. After adjustment, being active in health care increased the odds of consulting for an ILI (OR 1.66, 95% CI 1.40‐1.97). The association was strongest for physicians and nursing aides. In terms of work setting, odds of consulting for ILI were increased for professionals of almost all healthcare settings except home‐based care. Conclusion Individuals active in the healthcare system were more likely to consult their primary care physician for an influenza‐like illness than for another reason, compared with individuals not active in the healthcare system. These results warrant further efforts to understand influenza transmission in the healthcare system at large.


| INTRODUC TI ON
Healthcare workers are at increased risk of influenza infection compared to non-HCW. [1][2][3] For example, influenza-like illness (ILI) among Italian medical residents peaks earlier compared to the general population. 3 General practitioners (GPs) in particular have been shown to have high levels of basic immunity to influenza, probably resulting from frequent contacts with influenza viruses in the past. 4 Already during the 1918 influenza pandemic, social class based on occupation had an impact on mortality. 5 Occupation of influenza cases has been explored in more details during the 2009 H1N1 pandemic. In a study conducted in four American states, the proportion of healthcare workers was three times higher among laboratory-confirmed influenza cases compared to its proportion in the general workforce. 6 In a Spanish matched case-control study, being a healthcare worker was associated with consulting as an outpatient for influenza. 7 However, existing studies of influenza risk based on occupation do not differentiate between the different settings of the healthcare system, such as hospitals, residential homes, physician practices.
Direct transmission from healthcare workers has been documented, 8 but whether patients acquire influenza mostly from other patients or from healthcare workers is still debated. 9,10 Most of the work on healthcare-associated influenza has been conducted in hospitals 11 or long-term care institutions. In hospitals, a significant proportion of influenza infections is acquired during admission. 12 Patients visiting the emergency department for another reason than influenza during the influenza season have an increased risk of contracting influenza compared with community controls. 13 In an outpatient setting, one retrospective cohort study among children aged two to five years old reported an increased risk of 36% (incidence rate ratio 1.36; 95% CI 1.22-1.52) of presenting for an ILI visit in the 8 days after a non-ILI visit to a pediatric clinic. 14 Our research question was whether being professionally active in the healthcare system (exposure) increases the risk of influenza infection, assessed by consulting a primary care practitioner for influenza-like illness (outcome). We assumed that healthcare workers would mostly consult their primary care practitioner in case of influenza-like illness. Therefore, we estimated the association between seeking consultation for an influenza-like illness or having confirmed influenza, and being professionally active in the healthcare system, differentiating by type of profession and work setting.  For both cases and controls, the following variables were obtained from the routinely collected Sentinella: week, age, sex. In addition, for ILI cases we collected whether the swab was sent to the reference laboratory, and rRT-PCR result. At practice level, we obtained the region and total number patient-physician contacts during influenza surveillance season. The project made full use of the quality assurance system of Sentinella. Declaring GPs received instructions about data collection, with main messages reinforced by regular Newsletters. Predefined checks in electronic data entry diminished the risk of data entry errors. The Sentinella program Commission, consisting of regional representatives of declaring physicians, Swiss family medicine institutes, and the SFOPH, reviewed the study protocol and data collection forms.

| MATERIAL S AND ME THODS
Analysis of this case-control study was based on a mixed logistic regression model, taking into account the clustering by practice by including a random intercept. We considered age and sex as potential confounders, because age was associated with both types of profession and ILI incidence, and sex was associated with types of profession, as well as possibly associated with ILI incidence and health-seeking behavior in case of ILI. Profession and work setting of patients active in the healthcare system were compared to those not active, excluding those with unknown or missing activity information (complete case analysis). If active, other professions with <5% of total and unknown profession were regrouped into a single category. If active, but profession, respectively, work setting, was missing, it was recoded as unknown. For confirmed influenza cases, the dataset was restricted to practices where swabs were performed.
Separate models were used for activity in the healthcare system in general, categories of professional activity if active in the healthcare system, and categories of work settings, because of collinearity between these variables. In a sensitivity analysis, we repeated the model for activity in the healthcare system, setting all missing data to "inactive," To examine possible over-or underrepresentation of some professions among controls, we compared the proportion of individuals active in each professional category among subjects aged 15-64 years old with national occupational statistics. 15 We used the Stata 15 software for all analyses.
The investigators had access only to anonymized data. Neither additional health-related data nor biological material was collected specifically for the study. As such, the project was not under the scope of the Swiss human research law (LRH) and did not require formal ethical review.

| RE SULTS
During the 2018/2019 influenza surveillance season, there were 4287 ILI cases reported from 168 practices, out of which 346 were confirmed for influenza from the 79 practices swabbing ILI cases.
During weeks 11 and 12, 28 561 controls were recorded, reduced to 15 463 after restricting the dataset to practices doing swabs.
Being active in the healthcare system was associated with increased odds of consulting for an ILI (crude OR 1.91, 95% CI 1.65-2.21; Odds were also increased for administrative staff and for other or unknown profession. After adjustment, we found no increased odds for nurses nor for medical assistant and paramedical staff.
In terms of work setting, we found increased odds of consult- confirmed influenza and being an administrative staff or a staff active in another or unknown profession.
In sensitivity analyses, we considered all individuals with unknown or missing activity in the healthcare system as not active instead of excluding them from the logistic regression models (Table   S1). All associations found in the main analysis were confirmed.
Associations were also consistent when restricting the data to cases and controls to individuals aged 15-64 years old (Table S2). Finally, to get a sense of the healthy worker bias present in our data, we compared the proportions of individuals working in different categories or work settings among our control population with available national statistics (Table S3). With the exception of nurses, all professional categories were rather underrepresented among con-

| D ISCUSS I ON
In this study, individuals active in the healthcare sector were more likely to consult their primary care physician for an influenza-like illness, respectively, confirmed influenza, than for another reason. In terms of professional categories, the association was particularly strong for physicians and nursing aides. Surprisingly, being active either as an administrative staff or as any other or unknown profession in the healthcare system was also associated with an increased risk of consulting for an ILI. This could be due both to a higher risk of infection and to more sensitization in healthcare settings to abstain from work in case of ILI symptoms. In terms of work settings, private practices and nursing home particularly stood out, followed by hospitals. modes and evidence on effective interventions should be directly generated in the relevant settings, and not extrapolated from hospitals.
For example, a prospective cohort study among staff of primary care practices should be conducted to estimate infection rates without being confounded by differences in health-seeking behavior.
While sentinel practices do not constitute a representative sample of all primary care practices, we have no reason to believe that Sentinella practices would be more or less likely to have health professionals among their patients than other private practices. Also, the Swiss sentinel network covers all six regions of the country, and the demographic structure of the adult patient population is overall similar to the one of Swiss practices. 19 While these results cannot be used to extrapolate the proportions of professionals working in the healthcare system, we believe that the reported associations are valid. Still, we cannot exclude the possibility that health professionals were more likely to consult their physician for ILI, knowing that their physician was part of Sentinella. Overall, these findings certainly justify further attention to prevention of influenza transmission in the health system, particularly outside hospitals.

ACK N OWLED G EM ENTS
We