Appropriate methods to meet project objectives...

Methods

Decision Trees are employed when accurate costing and prediction of disease progression are independent of incremental time advances called cycles. In Decision Trees, all members of a cohort start at a single, starting node and are segregated into a discrete number of final branches.

Markov Models are employed when multiple cycles iteratively executed are necessary to properly evaluate a disease state. True Markov Models do not require any memory of a previous state, thus all members of a cohort are segregated using information as they exist in a single moment in time rather than what path they took to get there.

Semi-Markov Models work in much the same way as pure Markov Modeling but with memory that allows for small differences between cycles.

Monte Carlo Microsimulation is employed when many individual attributes of cohort members can affect outcomes (Age, gender, race, etc.). Monte Carlo Microsimulations contain time advances in a fixed manner or using time-to-event approaches and are designed to track many attributes throughout the simulation. Attriubutes are assigned by random sampling from distributions representative of the population or disease state.

Time-to-Event (TTE) is used when TTEs can be calculated to move ahead within the model timeframe without rigidly defined cycle lengths. TTE can be achieved if credible estimates of the relationship between the expected event and time values can be derived. These estimates may take the form of linear or higher order functions or may simply be determined from discrete tables.

Discrete Event Simulation in situations where the competition for resources among entities can have an impact in continuous time requiring the placement of entities in a queue and listing events on a calendar. The foundations of DES are reliable distributions representing the spectrum of times at which entities make request and the times at which resources are made available. These distributions take various forms (e.g., Normal, Exponential, Beta, Poisson) depending on the characteristics of the event or resource.

Platforms

Decision Tree Software - Strength lies in pure Decision Trees but may be adapted to perform Markov, Semi-Markov, and Monte Carlo Microsimulation modeling. User interface functionality can be limited, which necessarily limits the end-user to those familiar with the Decision Tree development platform.

Excel Spreadsheets - Most functional Decision Trees and Markov models. User interfaces can be programmed in Visual Basic. Semi-Markovs are feasible with additional Visual Basic programming but limited as complex algorithms test the limits of code maintenance and integrity.

Custom Software - MDM has developed a software development toolkit that leverages the flexibility and platform independence of the Java programming language to develop software customized to the client's requirements as determined by the disease state and treatment comparators.