OBJECTIVES: Recent ISPOR Good practice guidelines as well as literature encourage to use a single distribution rather than the latent failure approach to model time to event for patient level simulation models with multiple competing outcomes. Aim was to apply the preferred method of a single distribution on time to event in combination with a multinomial distribution on type of event for parameterizing the primary tumor component of a patient level head and neck cancer model. METHODS: Data on patients treated with radiation therapy as first line therapy for head and neck tumor at two university hospitals in the Netherlands  between 25-02-1980 – 13-12-2010 was used (nUMCG=277 & nVUMC=736). Several distributions were tested for model fit, using QQ-plots, AIC, and simulated versus actual data plots to judge best fit. Covariates tested for inclusion were age, gender, tumor location dummies, nstage and tstage. The final model was applied in the patient simulation model. Multinomial regression with the same covariates and time of event added as a covariate was applied on type of event, distinguishing death, loco regional recurrence and metastasis as events. All analyses were performed in R. RESULTS: The LogNormal distribution showed best fit. The final model had the following coefficients for the location parameter (se in brackets):  Intercept, 8.2 (0.38),   Age -0.026 (0.0053),  Tstage -0.38 ( 0.062), Nstage  -0.21 (0.076), Locd1 -0.91 (0.22) , Locd2 -0.28 (0.15), Locd3 -0.72 (0.22). Locd refers to location dummies. The estimated value for Log(sd) was  -0.44 (0.038).  The multinomial model had age, tstage and time of event as significant covariates CONCLUSIONS: A disadvantage of this method is that a single distribution has to be fit to a time of event which is the result of different interacting stochastic processes. The resulting distributions showed acceptable fit and could be implemented straightforwardly in the patient level simulation model.