American Journal of Epidemiology Advance Access originally published online on August 29, 2007
American Journal of Epidemiology 2007 166(8):983; doi:10.1093/aje/kwm240
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
LETTERS TO THE EDITOR |
RE: "MULTIPARAMETER CALIBRATION OF A NATURAL HISTORY MODEL OF CERVICAL CANCER"
Department of Epidemiology and Public Health, Yale University School of Medicine, New Haven, CT 06510
(e-mail: sanjay.basu{at}yale.edu)
In a recent article, Kim et al. (1) addressed the important question of how best to estimate parameters in complex models of human papillomavirus. Unfortunately, in their paper they make a number of incorrect claims about the advantages of their method and the disadvantages of more established methods. First, they wrongly claim that Bayesian techniques for identifying model parameters require informative prior distributions (2). They also claim that their technique of identifying multiple discrete parameter sets to fit curves is superior to the approach of finding a joint probability density of the parameters in a model, partly because the former uses common measures of effect produced in cross-sectional, case-control, and cohort studies as part of the approach to model calibration. In fact, more rigorous Bayesian inference procedures regularly incorporate meta-analysis with these common measures, projecting the results of this analysis directly from prior data onto the inference used to construct posterior parameter distributions (3). The software and techniques for implementing these analyses have been made simple, free, and open-source (4).
Kim et al. claim that their estimation method is novel for providing a range of results to allow a more complete representation of model uncertainty. The Bayesian approach inherently accounts for uncertainty in a probabilistic manner and would be superior to the authors' approach, which would not find a probability distribution but simply produce more discrete parameter estimates with more sampling. The approach described by Kim et al. is not capable of discriminating among "good-fitting" parameter sets. Perhaps most importantly, the Bayesian approach offers specific criteria for determining when estimation has converged onto posterior distributions of parameter estimates (5, 6) and for distinguishing among alternative model structures (7, 8). The authors' output appears to be heavily oriented towards their baseline precalibrated parameter set, and their model was unable to fit the available data, unless they assumed that natural immunity exceeded 50 percent for high-risk types of human papillomavirus. This suggests a basic underlying problem with the model used. The model adopted by Kim et al. violates principles of identifiability (9), given its large negative degrees of freedom, and the authors could benefit from becoming more familiar with such identifiability issues if they intend to make bold claims about parameter estimation.
 |
ACKNOWLEDGMENTS |
|---|
Conflict of interest: none declared.
|
References |
|---|
 TOP References |
|---|
- Kim JJ, Kuntz KM, Stout NK, et al. Multiparameter calibration of a natural history model of cervical cancer. Am J Epidemiol (2007) 166:137–50.
[Abstract/Free Full Text] - Boys RJ, Giles PR. Bayesian inference for stochastic epidemic models with time-inhomogeneous removal rates. J Math Biol (2007) 55:223–47.[CrossRef][Web of Science][Medline]
- Cooper NJ, Sutton AJ, Abrams KR, et al. Comprehensive decision analytical modelling in economic evaluation: a Bayesian approach. Health Econ (2004) 13:203–26.[CrossRef][Web of Science][Medline]
- Spiegelhalter DJ, Best NG. Bayesian approaches to multiple sources of evidence and uncertainty in complex cost-effectiveness modelling. Stat Med (2003) 22:3687–709.[CrossRef][Web of Science][Medline]
- Brooks SP, Gelman A. General methods for monitoring convergence of iterative simulations. J Comput Graph Stat (1997) 7:434–55.[CrossRef]
- Geweke J. Evaluating the accuracy of sampling-based approaches to calculating posterior moments. In: Bayesian statistics 4: proceedings of the Fourth Valencia International Meeting—Bernardo JM, Berger JO, Dawid AP, eds. (1992) New York, NY: Oxford University Press. 169–93.
- Spiegelhalter DJ, Best NG, Carlin BP, et al. Bayesian measures of model complexity and fit (with discussion). J R Stat Soc Ser B (2002) 64:583–639.[CrossRef]
- Akaike H. A new look at the statistical model identification. IEEE Trans Automatic Control (1974) 19:716–23.[CrossRef]
- Jacquez JA. Identifiability and model distinguishability. In: Compartmental analysis in biology and medicine. (1985) 2nd ed. Ann Arbor, MI: University of Michigan Press.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  S. Basu, G. B. Chapman, and A. P. Galvani Integrating epidemiology, psychology, and economics to achieve HPV vaccination targets PNAS, December 2, 2008; 105(48): 19018 - 19023. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. J. Kim, E. L. Franco, N. K. Stout, and S. J. Goldie THE AUTHORS REPLY Am. J. Epidemiol., October 15, 2007; 166(8): 983 - 984. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||