|
OBUMRM
A
FORTRAN Program to Analyze Multi-Reader, Multi-Modality ROC Data
Statistical Method described in:
Obuchowski NA, Rockette HE. Hypothesis testing of diagnostic accuracy
for multiple readers and multiple tests: an ANOVA approach with
dependent observations. Communications in Statistics - Simulations
1995; 24: 285-308.
Obuchowski NA. Multireader, multimodality receiver operating characteristic
curve studies: hypothesis testing and sample size estimation using
an analysis of variance approach with dependent observations. Academic
Radiology 1995; 2: S22-S29.
Software
written by Nancy A. Obuchowski, Ph.D., The Cleveland Clinic Foundation
Purpose:
To perform the computations of the Obuchowski-Rockette statistical
method for multireader, multimodality ROC datasets. The inputted
dataset should have the following specifications: N subjects without
the condition ("normals") and M subjects with the condition
("diseased") each underwent I diagnostic tests. The results
of the I diagnostic tests were interpreted by the same sample of
J readers. The objective of the study is to compare the mean accuracies
of readers in the I tests to determine if the accuracy of the I
tests differ.
Notes:
- We
assume there is a gold standard that provides each patients' true
diagnosis, i.e. distinguishes normal from abnormal patients without
error.
- We
assume that the readers interpreted the test results blinded to
other test results and the gold standard result.
- We
assume there is only one test result per patient, per reader,
per diagnostic test. The following scenarios are not appropriate
for this program: analysis of the test results of the two eyes
of a patient, analysis of the test results of the two kidneys
of a patient.
- We
assume there is no missing datapoints.
- The
program can accomodate either ordinal or continuous-scale test
results.
-
Under FORMAT A, the program estimates the area under the empirical
ROC curve (so called "nonparametric estimate") and performs
the necessary calculations using these estimates of accuracy.
***To use other estimates of the ROC area (e.g. binormal MLEs)
or to use other measures of accuracy (e.g. partial area under
the ROC curve), the data must be inputted using FORMAT B.
- The
program treats the diagnostic tests as fixed effects, and the
patient sample and readers as random effects.
- The
program is currently dimensioned for either 2 or 3 modalities,
up to 15 readers, and up to 500 patients with and 500 without
the condition. These dimensions can be changed.
- The
current name of the input file is 'obumrm_in.dat'. The current
name of the output file is 'obumrm_out.dat'.
- There
are example input and output files that can be used as templates
for data entry and as verification that the program is performing
properly.
Files
for Downloading:
The following plain text files may be downloaded by clicking on
the filenames.
ReadMe file on OBUMRM.FOR readme.txt
Instructions for using FORMAT A instructionsA.txt
formAordin.dat: An example input dataset in format A (ordinal-scale
data)
formAcont.dat: An example input dataset in format A (continuous-scale
data)
Instructions for using FORMAT B InstructionsB.txt
formB.dat: An example input dataset in format B
OBUMRM.FOR: The actual FORTRAN program obumrm.for
formAordin_out.dat: The output file corresponding to formAordin.dat
formAcont_out.dat: The output file corresponding to formAcont.dat
formB_out.dat: The output file corresponding to formB.dat
|