Performance of estimation methods in modelling the kinetics of respiratory virus infection

Background: Viral kinetic (VK) models are progressively being used to support the dose justification of drugs used to treat influenza and Respiratory Syncytial Virus infection 1-2. However, study design considerations and data limitations often present challenges when developing models that describe the VK following placebo and drug treatment. We explore the performance of estimation methods and discuss some of the difficulties associated with characterizing the pharmacodynamic effect.

Objectives: To assess the performance of stochastic approximation expectation–maximization (SAEM), Importance sampling (IMP) and first order conditional estimation with interaction (FOCEI) when modelling respiratory VK data.

Methods: The simulation dataset was based on a typical challenge study population (n = 30) inoculated with 106 viral particles, with sampling over 0 – 7 days 2-4. The VK was described using an established 4-compartment target-cell limited model for virus (V), target epithelial (T), infected non-productive (I1) and infected productive (I2) cells. Parameter estimation of VK in placebo-receiving subjects was performed using SAEM, IMP and FOCEI methods in NONMEM. To reduce parameter identifiability, the ‘lifespan’ of infected cells was assumed using known physiological values 2-4. The performance of estimation methods was assessed by calculating parameter bias and relative standard error (RSE).

Results: The SAEM estimation method (followed by IMP, expectation only) produced the lowest parameter bias (-27% to 11%; RSE 1.2% to 32%). However, FOCEI was unreliable with bias exceeding 50% to 100% for some parameters. When estimating the antiviral effect, a general observation is that the doses administered in human challenge studies often achieve concentrations that far exceed the EC50 value. As a consequence, pharmacodynamic models frequently predict complete inhibition of virus across the entire duration of study at the dose ranges investigated.

Conclusions: Respiratory VK models can provide valuable insights to support antiviral drug development. However, careful consideration of estimation methods and clinical study design is required during modelling to allow for appropriate extrapolation to target populations of interest.


  • Canini L, Perelson AS. Viral kinetic modeling: state of the art. J Pharmacokinet Pharmacodyn 2014 41: 431‐443.
  • Baccam P, Beauchemin C, Macken CA et al. Kinetics of influenza A virus infection in humans. J Virol 2006; 80: 7590-9.
  • Canini L et al. Population Modeling of Influenza A/H1N1 Virus Kinetics and Symptom Dynamics. J Virol 2011; 85:2764-2770.
  • González‐Parra G, Dobrovolny HM. Assessing Uncertainty in A2 Respiratory Syncytial Virus Viral Dynamics. Comput Math Methods Med 2015 doi: 10.1155/2015/567589.