This is the title of a useful article by Karnan and Haji Ali Afzali (Pharmacoeconomics 2014) A review and critique of the costs and benefits of DES, with a subtitle.
The paper begins by highlighting the 3 most common types of models: decision trees, Markovian group models, and individual-level models. decision tree Are useful for simple, short-term models, but not for more complex modeling tasks or longer term horizons. group model are the most popular, but “…[b]Because such models follow a group rather than individuals, they are subject to the Markovian assumption, which states that an object’s next (or future) path depends only on the current health state, not on the sequence of states preceding it. personal model include
Discrete time and continuous time state transition models, and discrete event simulation (DES). “All of these models overcome the Markovian assumption by providing individuals with properties that can influence their progress through the model, but which can also be updated as individuals experience events within the model.” What is the difference between discrete/continuous time transition models and DES? The former focus on health conditions; DES is organized around events.
What are the advantages and disadvantages of DES? I summarize some of these below.
ProsThe primary advantage of DES is that they can more easily cover the following 4 scenarios:
- basic diversity“Most cohort-based models define a single, homogeneous population (for example, women aged 60 years with node-positive early breast cancer [10]), for which a set of input parameter values is specified. By facilitating the representation of heterogeneity in the eligible population, DES can improve model accuracy if baseline characteristics are not jointly and individually normally distributed, and baseline characteristics are not linear in effect (e.g., for each unit increase in age, the probability of an event increases by 1%). The value of representing baseline heterogeneity is a function of the magnitude of the potential improvement in model validity and the quality of the data available to populate the model.
- persistent disease markersIn the real world, diseases often change slowly over time, DES allows this type of modeling, For example, if a clinical trial has only surrogate markers, it is very helpful to be able to link these surrogate measures to outcomes, How is it done? “First, a model is constructed to predict changes in disease marker variables over time, Second, given current disease marker levels (and other patient characteristics), models are applied at each time point to predict the probability of relevant clinical events, Such models require consideration of disease marker progression, and the relationship between disease markers and the incidence of relevant clinical events (for example, diabetic retinopathy, cardiovascular events, and kidney disease) Strong descriptive data are needed,”
- time varying incidence ratesTime-varying event rates can be implemented in group models but are much more complex than DES, “Representation of time-varying probabilities [cohort models requires] States that are entered after the start time of the model require the use of tunnel states: a health state is divided into a number of sub-states, each of which represents the health state at some specified time beyond the start time of that state. The implementation of time-varying event rates in DES models is more straightforward – the model simply samples event times from a specified survival function. When does it matter? When an event occurs on a large scale, it has a ‘knock-on effect’. “The improvement in model validity from representing time-varying event rates to nonobservable health conditions is likely to be large. In these cases, variation in the distribution of timing of the next event has an impact. In the stroke prevention model, new times to subsequent secondary events were sampled after each secondary event was experienced.
- Effect of prior events on subsequent event ratesThen, the effect of prior events on subsequent infection states can be modeled in cohort models but the complexity increases exponentially, “The models that best illustrate this factor are diabetes models, which use mostly UKPDS outcome model risk equations to describe the incidence of seven diabetes-related complications, To represent all possible combinations of the seven complications would require 127 different health conditions,,, Other examples include models in which the same event can be experienced multiple times and the number of episodes is a significant predictor; for example, a model of schizophrenia has used risk equations as a patient characteristic, The number of relapse episodes has been recorded,”
Shortcoming, The main drawback is complexity and the need for additional validation.
- complexity“The cost of DES is related to the time and expertise required to implement and review complex models, when perhaps a simpler model would suffice, The costs are borne not only by the analyst, but also by the reviewers,”
- ValidationAll models require validation, Because DESs are more complex and may appear less transparent to reviewers (e,g,, academic reviewers, HTA bodies, other researchers), it is important to conduct validation exercises,
So which model should you use? Decision stress, group models, or individual-level models such as DES? The authors turn to Albert Einstein to provide some guidance:
‘Everything should be made as simple as possible, but no simpler than that.’
Albert Einstein
The article also conducts a literature review of all DES models that were used as part of economic evaluation of health technologies. You can read the full article Here,
