Skip to main content

Frontiers of Engineering Reports on Leading-Edge Engineering from the 2009 Symposium (2010) / Chapter Skim
Currently Skimming:

Calibration in Computer Models for Medical Diagnosis and Prognostication--Lucila Ohno-Machado, Frederic Resnic, and Michael Matheny
Pages 91-98

The Chapter Skim interface presents what we've algorithmically identified as the most significant single chunk of text within every page in the chapter.
Select key terms on the right to highlight them within pages of the chapter.


From page 91...
... . Statistical and machine-learning techniques applied to large clinical data sets are used to develop the models, which are used by both health care professionals and patients.
Read the entire page →
From page 92...
... A fundamental problem in evaluating the calibration of a model in a health care setting is the lack of a gold standard against which individual risk estimates can be compared. A gold standard would be based on a sufficient number of exact replicas of the individual, accurately diagnosed or followed without censoring, so that the proportion of observed events would be equal to the "true estimate" for that individual.
Read the entire page →
From page 93...
... , the overall errors were smaller. In this example, the neural network provided better approximations of the true underlying probability of the event in clusters 2 and 3, as can be seen in the ranges of estimates in these clusters, as well as in their maximum residuals.
Read the entire page →
From page 94...
... Bottom panel: Projection of the points into an "axis" for neural network estimates. The neural network model comes closer to the true probabilities for clusters 2 and 3 than the logistic-regression model.
Read the entire page →
From page 95...
... Proportion LR NN LR NN LR NN Cluster of Events LR Mean NN Mean Std Dev Std Dev Minimum Minimum Maximum Maximum 1 0.01 0.0338 0.0219 0.0172 0.0069 0.0066 0.015 0.0949 0.059 2 0.42 0.4129 0.4819 0.1080 0.0207 0.1955 0.431 0.8013 0.584 3 0.64 0.6291 0.6852 0.1149 0.0146 0.2873 0.647 0.8507 0.732 4 0.98 0.9740 0.9908 0.0127 0.0011 0.9301 0.985 0.9954 0.992 
Read the entire page →
From page 96...
... Ohno Fig 2 IMPLICATIONS FOR MEDICAL DECISIONS In clinical practice, incorrect estimates have significant implications. For example, the widely used clinical practice guideline from the report by the Adult Treatment Panel III (NCEP, 2002)
Read the entire page →
From page 97...
... , but much remains to be done to inform health care workers and the public about the potential shortcom ings of this aspect of personalized medicine. As new molecular-based biomarkers for a variety of health conditions are developed and used in multidimensional models to diagnose or prognosticate these conditions, it will become even more important to develop accurate methods of assessing the quality of estimates derived from predictive models.
Read the entire page →
From page 98...
... 2008. Optimizing logistic regression coefficients for discrimination and calibration using estimation of distribution algo rithms.
Read the entire page →

Key Terms

  • logistic regression
  • predictive models
  • underlying probability
  • average estimate
  • prognostic models

  • incorrect estimates
  • panel iii
  • adult treatment
  • calibration plot
  • std dev

  • gold standard
  • risk estimates
  • regression models
  • true underlying
  • clinical practice

  • regression model
  • treatment panel
  • network model
  • exact replicas
  • estimates according

  • significantly according
  • national cholesterol
  • network provided
  • traumatic brain
  • resulting logistic

  • amia annuual
  • estimate individualized
  • predicted events
  • network estimates
  • predictive model


    • This material may be derived from roughly machine-read images, and so is provided only to facilitate research.
    More information on Chapter Skim is available.