Since 2002, 4 million visitors plus:
hit counters
search engine optimization service

  Appletcollection Vertical Menu java applet, Copyright 2003 GD

SENSITIVITY, SPECIFICITY, AND POSITIVE AND NEGATIVE PREDICTIVE POWER FOR CONTINUOUS PERFORMANCE TESTS

Ron P. Dumont, Ed.D., John Willis, Ed.D., Casey Stevens, M.A., C.A.S.

The Gordon Diagnostic System is another in the array of computerized CPTs (Continuous Performance Tests) that have been used to help diagnose attentional difficulties. It, like many of the CPTs, lacks strong validity. I do not deny that CPTs can be useful in a comprehensive evaluation of children suspected of attentional problems, but they are not 'proof' in and of themselves. There are a number of important confounds that could impact upon the score obtained on this and other computerized measures.

Reviews of research on computerized continuous performance tasks have generally been favorable, and they are seen as playing a role, albeit limited, in the evaluation of attention disorders. Barkley and Grodzinksi (1994), for instance, evaluated the utility of neuropsychological measures, including continuous performance tests (CPTs) for distinguishing children with ADHD from normal controls and children with learning disabilities. They found CPT measures among the most useful of the assessment procedures investigated. Nonetheless, they noted that positive but not negative CPT findings can have diagnostic utility. Thus, while poor performance on a CPT measure was indicative of an attention disorder, good performance did not necessarily rule out attention disorders. These results have also been replicated by Matier-Sharma and colleagues (Matier-Sharma et. al., 1995).

One recent study, Wherry et al., 1993, Psychology in the Schools, investigated the validity of the GDS and the results were fairly poor. These authors stated that "The results failed to demonstrate the discriminant validity of any GDS score regardless of the behavior rating used." As Barkley and others (1994) have noted, in order for a test to be diagnostically useful, it must be able to not only identify the children with ADHD, but it must also accurately identify children without ADHD. One very important issue regarding the typical validity studies is their use of already identified clients. This does provide some aspect of validity but it is also necessary to investigate the sensitivity and specificity of the measures (something typically lacking).

Ellwood (1993) discusses parameters that can be used to examine a test's diagnostic usefulness. Test specific parameters include sensitivity, or the proportion of individuals with a disorder that exhibit the sign (i.e., the proportion of children with ADHD who receive scores within the abnormal range) and specificity, or the proportion of individuals without a disorder that do not exhibit the sign (i.e., the proportion of controls who receive scores within the normal range). These two parameters are calculated in the research setting by first knowing the diagnosis of the children (through test-independent criteria) and noting how they perform on the test of interest. However, as Ellwood (1993) points out, this is the opposite of the way an evaluator uses a test. The evaluator starts with the test score and attempts to determine the child's diagnosis. In order to judge the usefulness of a test for this purpose, the evaluator will need to look at a test's sensitivity and specificity in light of the disorder's base rate in their referral population.

For example, if a test was used as a screening measure on a population of 1000 children in which 4% (40) of the children have ADHD, and that test gives an abnormal score for 90% of the children with ADHD (i.e., sensitivity) and gives a normal score for 90% of the children without ADHD (specificity), the following diagnostic properties result.

Calculation of Sensitivity, Specificity, PPP, and NPP

ADHD
Control

a

b

Abnormal Score

36
96

132

c

d

Normal Score

4
864

868

40
960

1000

Sensitivity = a/a+c = .90

Specificity = d/b+d = .90

PPP = a/a+b = .27

NPP = d/d+c = .99


Using this table, one can calculate Positive Predictive Power (PPP), or the chances that a child who receives an abnormal test score actually has ADHD. PPP = a/a+b = 36/132 = 0.27. A test with 90% sensitivity and specificity has restricted usefulness as a diagnostic tool if it is used on a population with a 4% base rate of the disorder because if the child receives an abnormal score, (s)he is still much more likely to be a control than a child with ADHD.

The issue of PPP and NPP was from an article we wrote on the use of the Mesulam CPT (a paper and pencil CPT) that takes about 3 minutes and can be used in an entire class. We found the PPP and NPP to be similar or better than the Computerized tests.


Barkley, R. A. and Grodzinksi, G. M. (1994). Are tests of frontal lobe functions useful in the diagnosis of Attention Deficit Disorders? The Clinical Neurologist, 8, 121-139.

Ellwood, R.W. (1993). Clinical discriminations and neuropsychological tests: An appeal to Bayes' theorem. The Clinical Neuropsychologist, 7, 224-233.

Matier-Sharma, K., Perachio, N., Newcorn, J.H., Sharma, V., & Halperin, J. M. (1995). Differential diagnosis of ADHD: Are objective measures of attention, impulsivity, and activity level helpful? Child Neuropsychology, 1, 118-127.

Wherry, J. N., Paal, N., Jolly, J. B., Balkozar, A., Holloway, C., Everett, B., & Vaught, L. (1993). Concurrent and discriminant validity of the Gordon Diagnostic System: A preliminary study. Psychology in the Schools, 1, 29-36.