A Model-based Assessment of Oseltamivir Prophylaxis Strategies to Prevent Influenza in Nursing Homes

Postexposure prophylaxis can prevent more influenza virus infections among nursing home patients per dose than continuous prophylaxis.

oseltamivir resistant virus strain (E R ), infectious with an oseltamivir resistant virus strain (I R ), immune by prophylaxis (R p )}. At each moment t, the state of the system was completely characterized by the state vectors x and y. For convenience, we also used aggregate variables whose values were completely determined by the state variables: • the number of patients that were infectious with a non-resistant virus strain at time t, I P (t); • the number of HCWs at work that were infectious with a non-resistant virus strain at time t, I H (t); • the number of patients that were infectious with an oseltamivir resistant virus strain at time t, I RP (t); the number of HCWs that were infectious with an oseltamivir resistant virus strain at time t, I RH (t).

Update Rules
At each time step Δt the values of the state variables were updated to account for transitions. The probability of each of these transitions to occur was specified according to the rules in Table S2.   Day ë 1 (t) = ë 11 +ë 12 +ë vis ë 11 = (ð 11 ñ +(1-ð 11 )) p c c 11 I P (t) Evening ë 1 (t) = ë 11 +ë 12 +ë vis ë 12 = (ð 12 ñ +(1-ð 12 )) p c c 12 I H (t) Night ë 1 (t) =Ind(x(t,j+Neighbor(j))=I) p c +ë 12 * ë vis = g ñ p c i (1-ö) For HCWs # see table S1 for the meaning on the symbols used ¶ we use ¬vacant to denote any possible state except vacant ‡ we use {at work, ·} to denote any possible state where the first state variable is equal to the state at work * we use Ind(x(t,j)=J) to mean an indicator function that returns the value 1 if its argument is a correct expression, and 0 if its argument is false; we use Neighbor(j)=j-1+2 Mod[j,2] as a function that returns the index of the roommate of the patient, such that Neighbor(1)=2, Neigbor(2)=1 $ P start and P stop determine the moments of start and end of prophylaxis with oseltamivir; P o (t)= 1 if prophylaxis is being administered. Figure S1. Incidence and prevalence of influenza virus infections in the community.

Prophylaxis with Oseltamivir
Continuous prophylaxis was given during 8 weeks around the peak of the community influenza epidemic, starting from t=15 to t=71. Post-exposure prophylaxis was started for all patients as soon as one patient had a laboratory-confirmed influenza virus infection. Since recognition of a possible influenza virus infection is required before doing a laboratory test, we assumed only the fraction of infected patients that develop influenza disease (the symptomatic patients) could trigger the start of post-exposure prophylaxis. We assumed that for every first symptomatically infected individual the time between becoming infectious and the start of prophylaxis followed a distribution that was determined by  Table S3.  days, respectively. Table S3. Mean duration and range of the steps leading to the delay between the start of infectiousness of the first symptomatically infected individual and the start of post-exposure prophylaxis for three scenarios.
Step Duration (day ) s Mean

Precision of the Effect Estimates
In the main text we did not give confidence intervals around the effect estimates because these depended on the number of simulations we performed. The 95% bootstrap confidence intervals around the effect estimates (Table S4), show us that the 4000 simulations for the baseline scenario and the 2000 simulations for the alternative scenarios sufficed to obtain reliable effect estimates (Table S4).

Uncertainty analyses
We used Latin hypercube sampling (15,16) to do uncertainty analyses for the four parameters describing oseltamivir effectiveness. Therefore we chose a likely range for the parameter values in question and drew actual values from a uniform distribution over this range.   Table S5 shows the results of some additional scenarios:

Alternative Scenarios
1) Two scenarios with delays between the start of infectiousness of the first symptomatic patient and the start of post-exposure prophylaxis of 1.75 and 6 days instead of 3.5 days.
Changes in the delay did not have a large influence on the results.
2) Two scenarios with higher and lower influenza virus activity in the community (with community attack rates of 15 and 5 percent as compared to 10 percent in the baseline scenario).
Apparently the number of doses needed to prevent one infection was not very sensitive to the annual influenza activity when prophylaxis was given post-exposure. With continuous prophylaxis, the number of prevented cases increased with higher influenza prevalence and the strategy became more efficient, although it did not approximate the efficiency of post-exposure prophylaxis.
3) A scenario in which the HCW vaccination rate was 0.75 instead of 0.4. In this scenario the influenza virus attack rate among patients was already decreased in the absence of prophylaxis. Although the relative risk reductions for both strategies of prophylaxis were similar to those in the baseline scenario, the DNP increased due the lower actual number of infections prevented.

4)
A scenario in which the patient vaccine uptake was 0.40 instead of 0.75. In this scenario, the infection attack rate was slightly increased in the absence of prophylaxis, which increased the efficiency of prophylaxis.

5)
A scenario in which 30% instead of 0% of the patients was immune at the start of the season. In this scenario the effectiveness and efficiency of prophylaxis decreased since more individuals were protected prior to prophylaxis and a lower number of infections was prevented.
6) A scenario for a larger department with 60 beds. In this simulation we assumed the ratio of patients and HCWs and the average number of contacts per person per day to be the same as we had observed in the 30-bed departments. The effect of prophylaxis became slightly higher in this scenario due to the higher attack rate in the absence of prophylaxis.