Relationship of parameters to the real world

Abstract

Realistic models for a specific disease must be based on agreement between statisticians and epidemiologists, with regard to overall structure and specific flow-rates between compartments. Relevant parameters must be estimated from appropriate data. Models must be properly checked out and validated, e.g. by generating verifiable predictions.

Satisfactory models must incorporate the major epidemiological states considered of greatest relevance to specific situations under examination. Typical items are listed. Similarly, all relevant parameters must be incorporated, dealing with a variety of clinical, epidemiological, demographic, economic, and other factors. Again, typical items are listed.

Proper statistical analysis should be undertaken whenever possible, including parameter estimation, calculation of standard errors, testing of goodness-of-fit, and comparison of predictions with observations.

Many of these ideal requirements cannot be achieved in practice, at least to begin with. Therefore a priori assumptions must be made about both model structure and parametric values. Such problems can be handled by the judicious use of sensitivity analysis, and more attention to this approach is needed.

In addition, special efforts should be made to achieve the minimum model complexity required for practical efficiency. An appropriate balance must be found between the conflicting requirements of realism and tractability.