We simulated the occurrence of influenza virus outbreaks during an 80-day period in a typical long-term care nursing home department (30-bed unit with 15 two-bed rooms and a team of 30 HCWs) in the Netherlands. HCWs worked 8-hour shifts; according to a weekly schedule 5, 3, and 1 HCW(s) worked during the day, evening, and night shifts, respectively, which has been observed in some nursing homes in the Netherlands. The average length of stay for a patient was 14 months (16,17). Because we simulated a small population where chance events can have major effects, we used a stochastic transmission model. The model is described in the online supporting information (Technical Appendix) and has been described in detail in a previous study (18). Here, we describe the essential elements of the model’s structure for the baseline scenario (parameters for the baseline scenario are shown in Table 1).

According to a standard model for infectious disease transmission, persons could be in 1 of several stages of influenza virus infection: susceptible, infected but not yet infectious (exposed), infectious, or recovered/immune (Figure 1) (19). The durations of the exposed and infectious periods were exponentially distributed with means of 1.4 days; the resulting generation time equaled 2.8 days, which agrees with observations of generation times during influenza epidemics (20,21). At the start of the influenza season, 30% of the adult nursing home population was assumed to be immune to infection because of cross protection from earlier infections (22). Since the elderly have weakened immune systems (23,24), but exact estimates are absent, we made the most conservative assumption that their immune systems had no memory of previous infections.

According to our model, both patients and HCWs could receive influenza vaccine before the influenza season. The average vaccination rate was 75% for nursing home patients (25) and 40% for HCWs (2). We assumed that for each person vaccination either led to perfect immunity against infection or had no effect (18). In a previous study, we showed that this all-or-nothing assumption for vaccine-induced immunity yielded similar results to those of an alternative assumption of incomplete immunity in which vaccinated persons had a lower probability of acquiring infection upon contact with an infectious person (18). The assumption of all-or-nothing immunity due to prophylaxis has also been made in other modeling studies (26). We assumed the vaccine efficacy against influenza virus infection in healthy adults, and thus HCWs, was 73% (27). For elderly nursing home patients, no statistically significant vaccine efficacy against infection has been observed (28). However, because other evidence showed that the vaccine protected against influenza disease and complications, we assumed patient efficacy to be 25% (28,29).

We compared 2 strategies of prophylaxis with oseltamivir to a control situation in which no neuraminidase inhibitors were used: continuous (seasonal) prophylaxis was given to all patients during 8 weeks (the longest period of prophylaxis described in effectiveness studies) (30) around the peak of the influenza season; or postexposure prophylaxis was started for all patients as soon as 1 patient had a laboratory-confirmed influenza virus infection. Because recognition of a possible influenza infection is required before doing a laboratory test, we assumed that only the fraction of infected patients in whom influenza disease developed (the symptomatic patients) could trigger the start of postexposure prophylaxis. We assumed that, for every first symptomatically infected person, the delay between the start of infectiousness and the start of prophylaxis followed a distribution with a mean of 3.5 days. This interval was determined by the time to onset of symptoms, the time to recognition of symptoms, the time to a positive laboratory test, and the delay to start of prophylaxis (Technical Appendix). Postexposure prophylaxis was given to all patients in the department for at least 2 weeks and was continued until no new cases occurred during a period of 8 days (2). Because we did not have data on the efficacy of oseltamivir in elderly persons, we used estimates from household studies (4) as the best available evidence. We assumed oseltamivir induced immunity to infection by wild-type strains in 55% of the susceptible patients as soon as it was administered and for the duration of prophylaxis. Immunity did not develop in the other patients, but when they were infected they were considered to become less infectious than persons who did not take oseltamivir (26). Based on estimates of the total reduction in infectiousness in persons treated with oseltamivir (4), we assumed the probability that the virus was transmitted during contact with a susceptible person was reduced by 20%. In the Technical Appendix, we describe some uncertainty analyses that we performed for the parameters describing oseltamivir efficacy.

On the basis of household studies, we assumed that influenza disease would develop in 50% of patients infected with influenza virus (4). For those receiving oseltamivir prophylaxis, this probability was only 20% (4).

A person’s risk of being infected depended on the number and type of contacts with infectious persons. We distinguished between casual and close contacts; casual contact was considered as conversation and close contact occurred with physical contact. We parameterized the contact model; the expected numbers of contacts, specified by type of persons and kind of contact, matched the number of contacts that we observed in 2 nursing home departments in the Netherlands (18). The probability of contact between 2 persons, given their type (HCW or patient), as well as the probability that this contact was close (physical contact), is given in Table 2. During the night shift, patients did not have contact with other patients, except for their roommates, who were assumed to be casual contacts. During the day and evening shifts, patients could also have contact with visitors. All contacts with visitors were considered close. The expected number of visitors was based on a study in the Netherlands on nursing home patients and visitors and was estimated to be 0.7 visitors per patient per day (31).

For every pair of persons with a casual or close contact, a probability existed that the virus was transmitted if the persons involved in the contact were infectious and susceptible. This probability was determined by sampling from a Bernoulli distribution with mean set equal to the transmission probability. For a casual contact, the transmission probability was 0.13; we chose this probability because the expected infection attack rate among patients in the absence of HCW vaccination was similar to observed attack rates for influenza-like-illness (18,25,32). For close contacts, the probability of transmission was assumed to be 2× as high as that of casual contacts.

The rate at which influenza virus was introduced into the nursing home by HCWs, visitors, and patients depended on the prevalence of the virus in the community; we used a simulation of an influenza epidemic in a large population (Technical Appendix). In each simulation, a constant proportion of infections in the community was assumed to be caused by resistant strains.

Resistant viruses were assumed to be completely insensitive to oseltamivir, and therefore prophylaxis had no effect on the susceptibility of a person who was exposed to a resistant strain. We also assumed that use of oseltamivir neither affected the infectiousness nor the development of symptoms in a person infected with a resistant strain. Apart from oseltamivir sensitivity, resistant and nonresistant strains were assumed to be similar. Infection with 1 of the strains conferred cross-protection against infection with other strains during the season.

We defined the infection attack rate and the disease attack rate as the total number of infections or influenza diseases among patients, respectively, divided by the total number of patients in the nursing home department during the study period. We distinguished between infections caused by oseltamivir-sensitive and -resistant strains and compared scenarios with increasing prevalence of oseltamivir resistance. Based on the distribution of infection attack rates in a nursing home in the absence of preventive measures (18), we used the proportion of infection attack rates of >0.3 as a proxy for the probability of a large outbreak. We calculated the absolute and relative risk reductions for both strategies of prophylaxis (efficacy) and determined the fraction of infections caused by resistant strains. We also computed the number of daily doses of prophylaxis needed to prevent 1 infection or disease (DNP) as the total number of doses administered divided by the number of influenza infections or diseases prevented (the absolute risk difference) (efficiency). Information on the statistical precision of the effect estimates can be found in the Technical Appendix.

In addition to the baseline scenario previously described, we considered an alternative scenario in which both patients and HCWs received continuous or postexposure prophylaxis according to the same rules. Postexposure prophylaxis was started after detection of infection in a patient and was given to all patients and all HCWs. We also studied a scenario in which the HCW vaccination rate was only 10%, as was observed in the Netherlands (33). Here we considered prophylaxis to patients only and to patients and HCWs.

In the Technical Appendix, additional scenarios are described for the following circumstances: 1) different delays between the start of infectiousness of the first symptomatic patient and the start of postexposure prophylaxis, 2) different levels of influenza virus activity in the community, 3) higher percentage of HCWs vaccinated, 4) lower patient vaccine uptake, 5) greater percentage of patients with prior immunity, and 6) a 60-bed nursing home department.

In the absence of resistance, the prophylactic use of oseltamivir reduced the number of influenza virus infections among patients during the influenza season. The infection attack rate among patients decreased from 0.19 in the control setting without prophylaxis to 0.13 (relative risk [RR] 0.67) when postexposure prophylaxis was given to all patients (first 2 bars, Figure 2, panel A). The fraction of large outbreaks with an infection attack rate of >0.3 decreased from 0.31 to 0.17 (RR 0.55), and outbreaks with attack rates >0.4 rarely occurred (Figure 3). If continuous prophylaxis was given for 8 weeks, the infection attack rate decreased to 0.05 (RR 0.23) (Figure 2, panel B), and the percentage of large outbreaks decreased to 0.03 (RR 0.09). Because of continuous prophylaxis, not only did large outbreaks disappear, but also the percentage of departments without any patient infection increased (Figure 3). Rates of influenza disease decreased from 0.10 to 0.06 (RR 0.60) and 0.01 (RR 0.13), respectively, for the 2 different strategies of prophylaxis (Figure 1, panels C, D). Although the number of infections that could be prevented was higher for continuous prophylaxis, the DNP was ≈3× higher with this strategy than with postexposure strategy (Figure 4). Without resistance, the DNP was 118 for postexposure prophylaxis and 323 for continuous prophylaxis.

An increase in the proportion of oseltamivir-resistant influenza virus strains in the community reduced the efficacy of prophylaxis with oseltamivir against infection and disease (Figure 2). In addition, both prophylaxis strategies became less efficient and the DNP increased rapidly, in particular for the continuous prophylaxis strategy (Figure 4). Prophylaxis caused a selection pressure for resistant strains; the percentage of infections caused by resistant strains in the nursing home was higher than in the community (Figure 5). The selection of resistant strains was most pronounced for continuous prophylaxis strategy.

Extension of prophylaxis strategies to include both HCWs and patients offered little additional protection to patients (Figure 6). In the absence of resistance, postexposure and continuous prophylaxis reduced the infection attack rate in HCWs from 0.14 to 0.10 and 0.05, respectively. The attack rate among patients decreased from 0.19 to 0.12 (RR 0.65) and 0.03 (RR 0.15), respectively. Taken together, the DNP for infection (of either patient or HCW) was 140 for postexposure prophylaxis and 366 for continuous prophylaxis; the total number of doses administered was 2 × as high as in the scenario in which only patients received prophylaxis.

When the HCW vaccination rate was 0.1, the infection attack rate among patients without prophylaxis was 0.23. This could be reduced to 0.15 (RR 0.67) when postexposure prophylaxis was given to patients alone and to 0.14 (RR 0.63) when it was given to HCWs as well (Figure 6). Continuous prophylaxis given to patients only or to both patients and HCWs could reduce the infection attack rate to 0.06 (RR 0.26) and 0.04 (RR 0.16), respectively. However, the number of doses required per department was approximately 6 × higher for continuous prophylaxis than for postexposure prophylaxis. Results of other alternative scenarios and the uncertainty analyses are described in the supporting information (Technical Appendix).