Novel modelling approaches to predict the role of antivirals in reducing influenza transmission

Abstract

To aid understanding of the effect of antiviral treatment on population-level influenza transmission, we used a novel pharmacokinetic-viral kinetic transmission model to test the correlation between nasal viral load and infectiousness, and to evaluate the impact that timing of treatment with the antivirals oseltamivir or baloxavir has on influenza transmission. The model was run under three candidate profiles whereby infectiousness was assumed to be proportional to viral titer on a natural-scale, log-scale, or dose-response model. Viral kinetic profiles in the presence and absence of antiviral treatment were compared for each individual (N = 1000 simulated individuals); subsequently, viral transmission mitigation was calculated. The predicted transmission mitigation was greater with earlier administration of antiviral treatment, and with baloxavir versus oseltamivir. When treatment was initiated 12-24 hours post symptom onset, the predicted transmission mitigation was 39.9-56.4% for baloxavir and 26.6-38.3% for oseltamivir depending on the infectiousness profile. When treatment was initiated 36-48 hours post symptom onset, the predicted transmission mitigation decreased to 0.8-28.3% for baloxavir and 0.8-19.9% for oseltamivir. Model estimates were compared with clinical data from the BLOCKSTONE post-exposure prophylaxis study, which indicated the log-scale model for infectiousness best fit the observed data and that baloxavir affords greater reductions in secondary case rates compared with neuraminidase inhibitors. These findings suggest a role for baloxavir and oseltamivir in reducing influenza transmission when treatment is initiated within 48 hours of symptom onset in the index patient.

Fig 1. Schematic representation of viral kinetic model and study overview.

The viral kinetic model assumes that the initial viral load increases (A) by infection (rate β) of and replication within target cells (T), which subsequently shed progeny virions (free virus, V; virus production rate, p). Antiviral treatment prevents the increase of viral load via inhibition of virus production; for baloxavir, this was informed by PK estimates from clinical trial data, whereas an inhibition factor was used for oseltamivir (in lieu of PK model data). The viral load decreases (B) with the death (rate δ) of infected cells (I) and by clearance of free virus (rate c). The PK–VK model was used to generate viral shedding curves (i.e. viral titer time courses) for 1000 simulated individuals treated with either baloxavir, oseltamivir, or placebo. These shedding curves were transformed to generate infectiousness profiles using one of three epidemiological models, whereby infectiousness was assumed to be proportional to viral load on natural-scale, a logarithmic-scale, or a dose–response transform. Transmission mitigation (i.e. the reduction in secondary transmission) was computed by comparing the area under the respective model-transformed infectiousness profiles for individuals treated with an antiviral or placebo. Figure adapted from Kamal et al. 2015 [23]. PK, pharmacokinetics; VK, viral kinetics.

Fig 2. Influenza virus titer time course in response to antiviral treatment, simulated using the pharmacokinetic–viral kinetic model.

The line represents the median of 1000 simulated individuals with the shaded areas corresponding to the 95% confidence interval around this median. Symptoms were assumed to start 36 hours after infection, and treatment was administered 24 hours after symptom onset (sampled from the observed distribution).

Fig 3. Estimated population-level influenza infectivity profiles over time from symptom onset according to the three epidemiological models studied.

The probability densities are shown for the timing of secondary infections expected under each of the models considered. The VK model was used to simulate individual viral shedding trajectories and the epidemiological model was used to translate these into infectiousness profiles over time. These individual profiles were then normalized so that the area under the curve was equal to 1 (once averaged), with each individual contributing to the average proportional to their time-integrated total infectiousness.

Fig 4. Estimated mitigation of influenza transmission by baloxavir and oseltamivir.

Estimated mitigation of influenza transmission by baloxavir and oseltamivir according to treatment initiation time from symptom onset using natural-scale, log-scale, and dose–response models, with time presented on a continuous scale (A) or as discrete time intervals (B). Both figures show interindividual variability. (A) The line represents the median; lighter shaded areas correspond to the 95% confidence interval around this median, and darker shaded areas show the interquartile range. (B) The boxplots show median, interquartile range, and 95% confidence intervals; the dots correspond to the mean.

Fig 5. Estimated reduction in secondary cases of influenza with baloxavir and oseltamivir.

The boxplots show median, interquartile range, and 95% confidence intervals. (A) Estimated reduction in secondary case rate among household contacts with prophylactic baloxavir and oseltamivir according to natural-scale, log-scale, and dose–response models, assuming treatment within 12–48 hours of symptom onset in the index case. Variability across individuals is shown. (B) Relative reduction in secondary cases with baloxavir treatment, comparing simulations with observed clinical trial data. Data are shown for baloxavir treatment at 0–24 and 24–48 hours after symptom onset in the index case, as well as overall (irrespective of treatment initiation time). The dashed line indicates data from the Phase 3 BLOCKSTONE study [14]. Variability in the observed reduction in secondary cases across simulations of the clinical trial is shown, which includes sampling and individual variation effects.