Estimation Techniques

CPM (Confidence Profile Method)


1. Description

CPM (Confidence Profile Method) uses conditional probabilities as arbitrarily specified in a "chain of evidence", similar to DAGs.[1][2][3] A benefit-risk metric is calculated from single link chains for direct evidence and from multiple link chains for linking together indirect evidence.

2. Evaluation

2.1 Principle
  • The principles of CPM are mathematically sound but exhaustive.
  • Evidence data are structured in "chains",that link the pieces of information together.
  • The equations for various formulations and circumstances are available in the original literature as guidance.[1]
  • Transparency is increased through five basic application steps.
  • Statistical uncertainties are propagated from the likelihood functions and prior distributions (in Bayesian context).
  • Other sources of uncertainties including biases can be specified as stochastic parameters.
  • CPM targets intuitively accessible elements where empirical evidence or practical experience is often available.
  • It allows uncertainty parameters to be expressed as probability distributions.
  • It requires explicit assumptions and judgements and allows these collective judgements to be combined in a probability distributions.
  • It can estimate the value of additional information about a parameter.

2.2 Features
  • Numerical representation of benefit-risk profiles are dependent on the choice of model fitted.
  • CPM can deal with multiple benefit-risk criteria and multiple sources of evidence through model parameterisation.
  • CPM naturally allows sensitivity analysis through changes in model assumptions and specifications.
  • The central idea of CPM is Bayesian, and so it can readily update new data and changes in assumptions formally through Bayesian inference.

2.3 Visualisation
  • Visualisation of CPM results may vary depends on the requirements and the choice of metric indices used.
  • Posterior distributions plots have been used to show the difference in probabilities depicting benefit-risk trade-offs.

2.4 Assessability and accessibility
  • The parameters in CPM should be carefully specified using probability distributions.
  • Although CPM is conceptually acceptable and interpretable, in practice they might not be so straightforward.
  • The results themselves, if specified correctly, are acceptable and are easy to interpret but are dependent on the complexity of the model and the choice of benefit-risk metric.
  • CPM has some similarities to SMAA and is a special case of CDS, and therefore shares some of their strengths and weaknesses as well as practicality of application in real-life decision making.

3. References

[1] Eddy DM. The confidence profile method: a Bayesian method for assessing health technologies. Oper Res 1989 Mar;37(2):210-28.
[2] Ades AE, Sutton AJ. Multiparameter evidence synthesis in epidemiology and medical decision-making: current approaches. Journal of the Royal Statistical Society: Series A (Statistics in Society) 2006;169(1):5-35.
[3] Eddy DM, Hasselblad V, McGivney W, Hendee W. The Value of Mammography Screening in Women Under Age 50 Years. JAMA: The Journal of the American Medical Association 1988 Mar 11;259(10):1512-9.