Send to

Choose Destination
See comment in PubMed Commons below
Stud Health Technol Inform. 1997;43 Pt B:666-70.

Use of statistical classifiers as support tools for the diagnosis of iron-deficiency anemia in patients on chronic hemodialysis.

Author information

  • 1Medical Informatics Unit, Salvatore Maugeri Foundation, IRCCS, Medical Center of Pavia, Italy.


Discriminant analysis, logistic regression and neural network models were applied to the diagnosis of iron-deficiency anemia in hemodialyzed patients. The ability of the three quantitative approaches to distinguish between subjects suffering or not from iron-deficiency anemia was compared by re-substitution and cross-validation testing. Methods performance was evaluated by means of sensitivity, specificity and accuracy. All the methods performed globally well (sensitivity and specificity > 0.85), revealing that the problem is classifiable. Neural networks showed the highest accuracy, both in the re-substitution (models developed and tested on the complete data set) and 3-way cross-validation (data set randomly splitted into 3 developmental and validation data sets) testing. These preliminary results suggest that the correct classification of iron status in the hemodialytic population can be treated as a pattern classification problem, for which neural networks and traditional statistical modelling can be a valuable aid to the clinical diagnosis of iron-deficiency anemia. A better performance of the neural network model must be confirmed through prospective testing on a larger data set.

[PubMed - indexed for MEDLINE]
PubMed Commons home

PubMed Commons

How to join PubMed Commons

    Supplemental Content

    Full text links

    Icon for IOS Press
    Loading ...
    Support Center