Discretely Integrated Condition Event simulation
Discretely Integrated Condition Event (DICE) simulation, created by J. Jaime Caro[1] and first implemented by Jörgen Möller, is an approach to modeling the trajectory of a disease in a given population and in the context of particular interventions, with applications spanning decision analysis through modeling of randomized clinical trials to formal health economic assessments.
Introduction
The techniques commonly used to model disease trajectories have been borrowed from other fields and can lead to oversimplification of the problem and unnecessary complexity in implementation. DICE simulation was specifically designed to overcome these limitations. With DICE, any aspect of a disease and its management can be simulated, including interactions among components, changes over time, sequences of treatments, and health-care resource use.[2][3] DICE can handle both Markov modeling and time-to-event simulation (unconstrained discrete event simulation[4]), as well as combined approaches.[2] Multiple types of analyses can be run using the same core DICE structure, which can be implemented in a spreadsheet without requiring additional software.
The DICE simulation approach has been validated against the sample models produced in 2014 by the Decision Support Unit of the National Institute for Health and Care Excellence (NICE).[5][6] Other agencies, such as the French Haute Autorité de Santé, have also implemented or reviewed models using this method.[7]
Conditions and Events
All information in a DICE model is specified in conditions. Each condition has a unique name. Conditions can characterize aspects of the disease (e.g., viral load, cancer stage, osteoporosis, severity score) or its consequences (e.g., renal impairment, pain, disability); features of the treatment (e.g., dose, price, compliance) or its unintended effects (e.g., neutropenia, weight change); timings of events; outputs; or any other aspects that are relevant (e.g., time horizon, discount rate, utilities[8]).
The information in conditions changes at uniquely-named events (hence “discretely-integrated”). Events have no duration, but their time of occurrence may be of interest. Many events reflect what happens in the real world during the disease (e.g., relapse) or as its consequence (e.g., fracture, hypoglycemia, progression); they can also represent aspects of the treatment (e.g., switch to another treatment, start dialysis) or of its unintended effects (e.g., anaphylaxis[9]). Other events are included purely for modeling purposes (e.g., update values, Markov transition) and to facilitate the functioning of the model (e.g., Start and End).
The time spent at a given level of a condition can have value, measured in whatever units are of interest (utilities, quality of life score, budget impact, willingness-to-pay, weights for a multicriteria decision analysis, etc.). Events can also have value, measured in whatever units are relevant, and this value may depend on when the event occurs.
| Conditions | Events |
| All information in the model:
Patient profiles Treatment features Values Event timings Outputs Equations and parameters Structure |
Points in time where one or more conditions change value:
Real world occurrences like Death Hospitalization Treatment change Model occurrences like Start/end Transition |
| Interested in time spent at a given level (value) | Interested in number that happen (and when) |
Expressions
The consequences of each Event are listed sequentially as expressions that specify how each affected Condition changes when the event occurs. Each expression is an instruction specified using text that follows the syntax of the selected spreadsheet software. The symbol that signals that the expression is a formula (e.g., the “=” in MS Excel) is omitted to ensure that the instruction is not immediately carried out. Expressions are read and acted on one by one in sequence by the discrete-integrator macro.
Profiles
A profile is a set of conditions that hold the determinant values that sufficiently characterize a population of interest. Profiles may contain demographics (e.g., age, sex), behavioral aspects (e.g., smoking), medical history (e.g., comorbidities, prior treatments), disease characteristics (e.g., severity, duration, stage), biomarkers, prognostic indices, and any other relevant information that changes from patient to patient. These Profiles are stored in arrays of specialized conditions. The profiles are read in and processed one at a time during execution.
Specific profiles are defined to represent the population adequately for the purpose at hand. Each profile denotes a “subgroup” of interest. Defining a single profile is tantamount to specifying a Markov cohort. There can be as many profiles as the analyst wants, including all the profiles manifested in a particular population of patients (e.g., the participants in a clinical trial).
Outputs
The results of a DICE simulation are stored in special output conditions. Those outputs are accrued during the simulation following the instructions specified in expressions at appropriate events. Outputs may continuously accumulate a quantity (e.g., time alive), may accrue amounts at discrete points in time (e.g., cost of a clinical visit); count occurrences (e.g., number of hospitalizations), or record a quantity as it accrues at particular points in time (e.g., changes in quality of life).
Specification
A DICE simulation is specified using a set of tables that itemize all of the model’s components: Conditions, Events, Outputs, Profiles. In each table, the applicable information is listed. The first column in each table displays the unique name of the item. Additional columns provide for aspects specific to the type of component. In the Conditions table, there is a column that lists the initial value of each condition (if known at the start of execution); in the All Events table, the initial time of occurrence of each event is specified and another column lists the names of the corresponding consequences tables. In the Outputs table, there is a column specifying the type of output. In addition, each event has a corresponding table that lists all its consequences in appropriate Expressions, row by row in the desired order of execution. Other helpful tabulations can list conditions that control a run (e.g., number of profiles, number of replications); the features of the setting (e.g., discount rate, perspective) in a Context table; the particulars of a given analysis (e.g., time horizon) in a Scenario table; the core formulas that control the simulation in an Equations table, and other information needed for an analysis (e.g., conversion factors) in a Constants table. This set of tables is the full specification of a DICE simulation and can be implemented on any spreadsheet or other software that supports tabulation.
Execution
A DICE simulation is executed by a macro that reads all the tables into memory, selects the first event (always “Start”) and processes its consequences table, one instruction at a time in the order they are listed. When the macro reaches the end of the table, it selects the next event — the one with the most proximal time — finds its consequences table and processes the instructions listed. This continues until the macro reaches an event called End, at which point it will stop simulating that intervention for that profile. For convenience, the execution of other interventions for that profile, and of additional profiles can be automated using additional loops in the macro. This macro can be written using the macro language often included with spreadsheet software or in any software language that provides the required functionality.
Advantages
The two concepts (conditions, events) that define a DICE are straightforward and correspond directly to the disease and its management. Since DICE uses a standard framework, terminology and a generic macro, users and reviewers need not “relearn” each new model nor re-validate the macro (e.g., the VBA code in an MS Excel implementation). The disease-specific terms will change but the structuring and implementation remain consistent across models. The ability to fully implement a DICE in familiar commonly-used software (e.g., MS Excel) removes the need to purchase and learn new software.[10]
DICE simulation is very flexible. It can accommodate anything from very simple models to vast complex structures, all the while remaining very transparent.[11] As the model is entirely specified by the Tables, DICE is readily communicated to clinicians, modelers, reviewers, decision makers and other stakeholders and the method does not itself introduce complexity. Even a person completely unexposed to modeling should be able to quickly understand the concept and review a model (the equations may require specialized statistical knowledge, but that is not specific to DICE. Modifying a DICE model is uncomplicated because changes to the tables — insertions, deletions, edits — are simple alterations of the table rows or text. There are no linkages that need to be updated or redone.
Limitations
DICE was developed to meet the needs of health technology assessment[12][13][14]. It is not meant for models that require explicit resources with capacities and queues (as in constrained discrete event simulation; nor for simulations that entail interactions with the environment or other people (agent-based simulation should be used in that case).[15][16]
See also
Markov model
References
- ↑ Caro JJ. Discretely Integrated Condition Event (DICE) Simulation for Pharmacoeconomics. Pharmacoeconomics. 2016 Jul;34(7):665-72. doi: 10.1007/s40273-016-0394-z.
- ↑ 2.0 2.1 Caro JJ, Moller J. Adding Events to a Markov Model Using DICE Simulation. Med Decis Making. 2018 Feb;38(2):235-245. doi: 10.1177/0272989X17715636. Epub 2017 Jul 5.
- ↑ Treatment sequences for advanced renal cell carcinoma: A health economic assessment Deniz B, Ambavane A, Yang S, et al. Treatment sequences for advanced renal cell carcinoma: A health economic assessment. PLoS One. 2019;14(8):e0215761. Published 2019 Aug 29. doi:10.1371/journal.pone.0215761
- ↑ Arlegui, Hugo; Nachbaur, Gaëlle; Praet, Nicolas; Bégaud, Bernard; Caro, J. Jaime (October 2020). "Using Discretely Integrated Condition Event Simulation To Construct Quantitative Benefit–Risk Models: The Example of Rotavirus Vaccination in France". Clinical Therapeutics. 42 (10): 1983–1991.e2. doi:10.1016/j.clinthera.2020.08.013. ISSN 0149-2918. PMID 32988633 Check
|pmid=value (help). Unknown parameter|s2cid=ignored (help) - ↑ Möller J, Davis S, Stevenson M, Caro JJ. Validation of a DICE Simulation Against a Discrete Event Simulation Implemented Entirely in Code. PharmacoEconomics. 2017;35(10):1103-9.
- ↑ http://nicedsu.org.uk/technical-support-documents/patient-level-simulation-tsd/
- ↑ Ghabri, S., Binard, A., Pers, Y. M., Maunoury, F., & Caro, J. J. (2020). Economic Evaluation of Sequences of Biological Treatments for Patients With Moderate-to-Severe Rheumatoid Arthritis and Inadequate Response or Intolerance to Methotrexate in France. Value in Health. https://doi.org/10.1016/j.jval.2019.12.003
- ↑ Ghabri, Salah; Binard, Aymeric; Pers, Yves-Marie; Maunoury, Franck; Caro, J. Jaime (April 2020). "Economic Evaluation of Sequences of Biological Treatments for Patients With Moderate-to-Severe Rheumatoid Arthritis and Inadequate Response or Intolerance to Methotrexate in France". Value in Health. 23 (4): 461–470. doi:10.1016/j.jval.2019.12.003. ISSN 1098-3015. PMID 32327163 Check
|pmid=value (help). Unknown parameter|s2cid=ignored (help) - ↑ Graves, J.; Garbett, S.; Zhou, Z.; Schildcrout, J.; Peterson, J. (May 2020). "Pbi20 Comparison of Decision Modeling Approaches for Health Technology and Policy Evaluation". Value in Health. 23: S17. doi:10.1016/j.jval.2020.04.087. ISSN 1098-3015. Unknown parameter
|s2cid=ignored (help) - ↑ Deniz, Baris; Ambavane, Apoorva; Yang, Shuo; Altincatal, Arman; Doan, Justin; Rao, Sumati; Michaelson, M. Dror (2019). "Treatment sequences for advanced renal cell carcinoma: A health economic assessment". PLOS ONE. 14 (8): e0215761. Bibcode:2019PLoSO..1415761D. doi:10.1371/journal.pone.0215761. PMC 6715231 Check
|pmc=value (help). PMID 31465470. - ↑ Smoking Cessation: A Comparison of Two Model Structures Pennington, B., Filby, A., Owen, L. et al. Smoking Cessation: A Comparison of Two Model Structures. PharmacoEconomics 36, 1101–1112 (2018). https://doi.org/10.1007/s40273-018-0657-y
- ↑ Hummel, James P.; Leipold, Robert J.; Amorosi, Stacey L.; Bao, Haikun; Deger, Kristen A.; Jones, Paul W.; Kansal, Anuraag R.; Ott, Lesli S.; Stern, Sean; Stein, Kenneth; Curtis, Jeptha P. (2019). "Outcomes and costs of remote patient monitoring among patients with implanted cardiac defibrillators: An economic model based on the PREDICT RM database". Journal of Cardiovascular Electrophysiology. 30 (7): 1066–1077. doi:10.1111/jce.13934. ISSN 1540-8167. PMC 6850124 Check
|pmc=value (help). PMID 30938894. - ↑ Reifsnider, Odette; Kansal, Anuraag; Pimple, Pratik; Aponte‐Ribero, Valerie; Brand, Sarah; Shetty, Sharash (2020). "Cost-effectiveness analysis of empagliflozin versus sitagliptin as second-line therapy for treatment in patients with type 2 diabetes in the United States". Diabetes, Obesity and Metabolism. 23 (3): 791–799. doi:10.1111/dom.14268. ISSN 1463-1326. PMC 7898389 Check
|pmc=value (help). PMID 33236481 Check|pmid=value (help). Unknown parameter|s2cid=ignored (help) - ↑ Guillemin, Francis; Wit, Maarten de; Fautrel, Bruno; Grimm, Sabine; Joore, Manuela; Boonen, Annelies (2020-11-01). "Steps in implementing a health economic evaluation". RMD Open. 6 (3): e001288. doi:10.1136/rmdopen-2020-001288. ISSN 2056-5933. PMC 7856109 Check
|pmc=value (help). PMID 33148785 Check|pmid=value (help). - ↑ Sullivan, W.; Bullement, A.; Lee, D. (2018-10-01). "Prm135 - Dice for Nice? Lessons from a Single Technology Appraisal". Value in Health. 21: S379. doi:10.1016/j.jval.2018.09.2255. ISSN 1098-3015.
- ↑ "DICE for NICE? Lessons from a Single Technology Appraisal" (PDF). Unknown parameter
|url-status=ignored (help)
Discretely Integrated Condition Event (DICE) Simulation
This article "Discretely Integrated Condition Event simulation" is from Wikipedia. The list of its authors can be seen in its historical and/or the page Edithistory:Discretely Integrated Condition Event simulation. Articles copied from Draft Namespace on Wikipedia could be seen on the Draft Namespace of Wikipedia and not main one.
