OBJECTIVES:  Competing risks (CRs) are endemic in HEMs. Techniques for estimation, using cause specific hazard models (CSHMs) are well understood. However, nearly all teaching uses the same parametric family to model each separate risk. This should never be assumed, particularly when the competing events are disparate such as progression, remission or side effect induced treatment switching. Visualisation tools can be crucial in choosing between parametric forms. Combining different parametric families together to make predictions requires flexible analytics and programming. This paper examines these visualisation and analytical/programming practices. METHODS: Monte Carlo simulation techniques generated artificial cause specific competing risk data from different parametric distributions which were then combined appropriately. Subsequent actions treated these parametric distributions as unknown. Graphs against time and log(time) of transformations of cumulative hazard, survival functions and kernel hazard estimates assisted fit statistics in choosing between parametric forms for each risk. Two different modified Newton Raphson techniques plus Brent’s method were compared for speed and convergence in predicting time to events in discrete event simulation models. Vectorised trapezium rule was used for numerical integration to create cumulative incidence functions for cohort models. Tasks were broken down into independent functions and validated against known output values for given inputs. RESULTS: Visualisation aids were crucial in determining parametric form. Negative sigmoidal function shapes typical in survival curves can prevent MNRTs from converging based on survival function inputs – switching to cumulative hazards can rectify this (e.g. worked for Weibull and Log-Logistic joint models): Brent’s method will work if it does not. Benchmarking and profiling tools proved vital to identify efficient code (300+ times increase) capable of performing PSA. CONCLUSIONS: Analysing and programming competing risk models using different parametric families is not easy but should be done if visualisation plots and fit statistics require it - models should fit the data (not vice-versa).