We conducted a prospective contact survey of nurses in charge of inpatient care for 5 hospitals in the German federal state of Bavaria by using a paper diary approach. We compared the data with data from a matched sample of the German general population obtained in a previously conducted survey of contacts (25). In both surveys, a contact was defined as a 2-way conversation of >2 words (not by telephone) or skin-to-skin contact (17,18) as a surrogate for exposure to disease. Actual rates of disease transmission were not measured.

Our findings correspond with the nature of contact between HCWs and patients. Nurses have a high frequency of close contact with patients, but the time they can dedicate to each patient is limited. This pattern differed fundamentally from that of their work contacts with staff members or other persons, the characteristics of which agreed with those described by Mossong et al. (25), who found that contact intensity (i.e., whether physical contact was involved) correlated positively with contact duration in a general population sample.

Contact data are important for modeling the spread of infectious diseases. Our results show that the use of general population data might lead to inaccurate modeling results for the healthcare sector because HCWs’ patterns of contact with patients differ fundamentally from the day-to-day contacts of the general population. However, although contact data form the basis of simulation models, other parameters may modify the influence of contact patterns. Our data might change current modeling predictions with respect to 1) spread of disease within the nurse population that results in staff shortage and 2) spread of disease from nurses to the general population.

Our findings suggest that the risk for infectious diseases by airborne transmission might be greater for nurses than for the general population because of nurses’ more frequent and more intensive professional contacts with potentially infected patients. In addition, nurses’ high proportion of work-related skin-to-skin contact highlights the potential for fecal–oral transmission of disease (e.g., norovirus infection).

Whether increased risk for infection in nurses during an influenza pandemic would lead to an earlier peak in the number of infections in HCWs than in the general population needs to be investigated in modeling studies. During a pandemic, maintaining sufficient staff during peak hospitalizations of the general population will be crucial. Our data may help guide public health interventions to prevent the infections in hospital staff and in the general population from peaking simultaneously.

Nurses most likely would take preventive measures, such as wearing personal protective equipment, when handling patients with clinically manifested influenza. However, their risk for infection from patients hospitalized for other reasons and already infected with influenza, but not yet clinically ill or with influenza diagnosis, might be more important for initial spread of disease in the healthcare sector. We collected our data during the interpandemic period, so we cannot predict how HCWs’ contact behavior might change during a pandemic. Models using our data will need to account for this uncertainty.

Nurses can potentially spread infection from infected patients to the general population. In Germany, nurses represent 0.5% of the population (28). After infection is introduced into the nurse population, further contact with patients, other staff members, and relatives might result in faster spread to the general population during the early phase of a pandemic. The combination of contact with persons from older age groups at work and from younger age groups at home might facilitate spread among different age groups that would not otherwise occur.

These points support the need to add a separate healthcare component to models, i.e., modeling both HCWs and general population by using self-reported contact data instead of assuming similar contact rates for nurses and the general population. Additional data are required for modeling studies to determine the extent to which nurses and patients would respectively benefit from vaccination of HCWs against influenza.

Our contact definition captures only the amount of social contact with other persons. Because we did not measure rates of transmission or disease, use of our data for modeling relies on the assumption that the number of contacts correlates with the amount of exposure to disease.

Regarding the ongoing discussion about the major route of transmission of influenza (29,30), we might overestimate exposure to disease by counting conversational contacts if influenza is transmitted predominantly by droplet spread. On the other hand, accounting only for skin-to-skin contact might considerably underestimate exposure because, among other reasons, transmission by contaminated surfaces or fomites would not be included. To provide data for modeling the spread of influenza and other infectious diseases with various routes of transmission, we dichotomized the data.

Furthermore, for nurses, the distribution of conversational and skin-to-skin contacts differs among contacts with patients and with others. Consequently, different contact matrices should be used for modeling when different transmission probabilities are assumed for conversational and skin-to-skin contacts.

Our study has some limitations. First, our nurse sample might not be entirely random because potential participants could refuse participation. If nurses willing to participate in the study had more contacts than nurses not willing to participate, our results could overestimate nurses’ contact numbers. Second, nurses’ motivation to participate might have been higher than that of controls, resulting in more reported contacts; a reason for nurses’ increased motivation might be our visits to study hospitals to inform nurses about the aims of the study and the potential benefits of participating for their occupation. The method used to recruit controls from the POLYMOD study required conduct of the survey in study participants’ homes. Even though home visits were timed to minimize this possible selection bias, the sampling still might have resulted in a higher proportion of housewives and unemployed and retired persons, all of whom may have fewer contacts than the general population. However, because patients’ needs largely determine the nurses’ number of work contacts and because the numbers of non–work-related contacts were comparable for nurses and controls, we do not expect that a large selection bias accounts for the difference.

A third potential limitation is the difference in work patterns of nurses and controls. Nurses worked on days they were surveyed, even on weekends, and controls might not have had work contacts during some days of diary completion. This difference is regarded, not as bias, but as potentially meaningful information that needs to be considered when modeling nurses’ contacts. Still, the separate analysis by day of the week persistently differed between the groups. The difference in work contacts reported by nurses and controls also might be affected by controls’ reporting of aggregated work contacts instead of single diary entries if they usually had >10 contacts at work. Not considering these aggregated work contacts for weekend days might have underestimated controls’ contact numbers. However, again, the separate comparison by day points to more contacts of nurses than controls. Evidence conflicts about whether disaggregated contact reporting is more complete than aggregated reporting. One study showed higher median contact numbers comparing prospective to retrospective reporting among university students (22). By contrast, another group that compared a diary approach with a 1 time-point estimation of contacts decided in favor of aggregated contacts because most diaries were completed retrospectively (31). However, most respondents in the POLYMOD sample from Germanyof the POLYMOD study (25) stated that they completed the diary during the day and not retrospectively. The large difference in work contacts between nurses and controls is difficult to explain solely by different methods of contact reporting.

Finally, we perhaps missed types of contacts that do not include conversation or physical contact. Missed contacts might include crowds (e.g., during public transportation), which have been discussed in the context of airborne spread of infections (29,30). However, the average risk associated with a conversation or with physical contact can be reasonably assumed to be substantially higher than just presence in the same room.

Hospital structure and tasks of nurses in charge of inpatient care are identical in the different German federal states. Therefore, we expect our results from hospitals in the German federal state of Bavaria to be representative of all of Germany. However, because structures and standards of care might differ in other countries, our data require country-specific validation. Furthermore, patterns of contact for hospital workers other than nurses in charge of inpatient care (e.g., physicians, technical assistants, cleaning personnel, or nurses in emergency departments) might differ substantially and influence the patterns of infectious disease spread in the hospital. However, because most routine daily patient contacts are with inpatient nurses, we are confident our data reflect the pattern of a large proportion of contacts in German hospitals.

Our survey did not account for length of hospital stay. Repeated contacts with long-term patients might not bear the same risk for infection as contacts with newly admitted patients because long-term patients are less likely to import an infection to the hospital. Future modeling studies might investigate the impact of length of hospital stay on disease spread.

In our study, nurses’ patterns of contact differed from those of the general population. Our findings support the need for explicit modeling of the healthcare sector to assess the spread of epidemics. To this aim, our study provides quantitative estimates of contact patterns. On the basis of results of modeling approaches for the healthcare sector, public health policies should be reassessed and revised as necessary.