Sensitivity, specificity and predictive values of psychiatric measures

Eric Fombonne, R. Fuhrer

Research output: Contribution to journalArticle

3 Citations (Scopus)

Abstract

Clinical assessments, psychological examinations, biological parameters and other indices are commonly used in psychiatry to measure disease status. These evaluation tools are not perfect and, in general, they will fail to perfectly discriminate 'cases' from normal controls. When a gold standard is available, their performances can be evaluated. If the results of a measurement are tabulated with true disease status in a contingency table, the proportions of cases and normals correctly identified (sensitivity and specificity) are useful coefficients of validity of the instrument studied. It is shown that unbiased estimates of the prevalence rate cannot however be derived directly from the proportion of sample respondents identified by a test. Practical examples with a psychiatric rating scale are provided which illustrate the importance and magnitude of this bias. Formulas are subsequently given which permit the calculation of the 'true' prevalence rate from the proportion of screen positives when the sensitivity and the specificity of the test are known. The proportion of true cases among those individuals identified by a test is another important property of psychiatric measures, particularly for screening purposes. It is also called the positive predictive value and, like sensitivity and specificity predictive values are probabilities for which high values are desirable. However, this critical characteristic of a test depends on the prevalence rate, the sensitivity and the specificity of the test used. Using an example, major changes in the probability of a screen positive individual to be truly diseased are shown to depend upon varying levels of the base rate of the disease. Different values of the positive predictive value are also listed in a table according to different values of prevalence, sensitivity and specificity. The practical implications of these notions are finally highlighted, both for clinicians and researchers. In particular, the performance of a measure should not be thought of as an intrinsic or unalterable feature; rather, they will vary with the context and purpose of measurement. If inefficient studies and false inferences are to be avoided, one should critically reassess the psychometric properties when transposing an instrument from one cultural setting to another, or from clinical to community samples.

Original languageEnglish (US)
Pages (from-to)73-77
Number of pages5
JournalEncephale
Volume17
Issue number2
StatePublished - 1991
Externally publishedYes

Fingerprint

Psychiatry
Sensitivity and Specificity
Psychometrics
Research Personnel
Specificity
Psychology
Proportion

ASJC Scopus subject areas

  • Psychiatry and Mental health
  • Neuroscience(all)

Cite this

Sensitivity, specificity and predictive values of psychiatric measures. / Fombonne, Eric; Fuhrer, R.

In: Encephale, Vol. 17, No. 2, 1991, p. 73-77.

Research output: Contribution to journalArticle

@article{1bc0d24b61754c20a3529f573fb8b3ef,
title = "Sensitivity, specificity and predictive values of psychiatric measures",
abstract = "Clinical assessments, psychological examinations, biological parameters and other indices are commonly used in psychiatry to measure disease status. These evaluation tools are not perfect and, in general, they will fail to perfectly discriminate 'cases' from normal controls. When a gold standard is available, their performances can be evaluated. If the results of a measurement are tabulated with true disease status in a contingency table, the proportions of cases and normals correctly identified (sensitivity and specificity) are useful coefficients of validity of the instrument studied. It is shown that unbiased estimates of the prevalence rate cannot however be derived directly from the proportion of sample respondents identified by a test. Practical examples with a psychiatric rating scale are provided which illustrate the importance and magnitude of this bias. Formulas are subsequently given which permit the calculation of the 'true' prevalence rate from the proportion of screen positives when the sensitivity and the specificity of the test are known. The proportion of true cases among those individuals identified by a test is another important property of psychiatric measures, particularly for screening purposes. It is also called the positive predictive value and, like sensitivity and specificity predictive values are probabilities for which high values are desirable. However, this critical characteristic of a test depends on the prevalence rate, the sensitivity and the specificity of the test used. Using an example, major changes in the probability of a screen positive individual to be truly diseased are shown to depend upon varying levels of the base rate of the disease. Different values of the positive predictive value are also listed in a table according to different values of prevalence, sensitivity and specificity. The practical implications of these notions are finally highlighted, both for clinicians and researchers. In particular, the performance of a measure should not be thought of as an intrinsic or unalterable feature; rather, they will vary with the context and purpose of measurement. If inefficient studies and false inferences are to be avoided, one should critically reassess the psychometric properties when transposing an instrument from one cultural setting to another, or from clinical to community samples.",
author = "Eric Fombonne and R. Fuhrer",
year = "1991",
language = "English (US)",
volume = "17",
pages = "73--77",
journal = "L'Encephale",
issn = "0013-7006",
publisher = "Elsevier Masson",
number = "2",

}

TY - JOUR

T1 - Sensitivity, specificity and predictive values of psychiatric measures

AU - Fombonne, Eric

AU - Fuhrer, R.

PY - 1991

Y1 - 1991

N2 - Clinical assessments, psychological examinations, biological parameters and other indices are commonly used in psychiatry to measure disease status. These evaluation tools are not perfect and, in general, they will fail to perfectly discriminate 'cases' from normal controls. When a gold standard is available, their performances can be evaluated. If the results of a measurement are tabulated with true disease status in a contingency table, the proportions of cases and normals correctly identified (sensitivity and specificity) are useful coefficients of validity of the instrument studied. It is shown that unbiased estimates of the prevalence rate cannot however be derived directly from the proportion of sample respondents identified by a test. Practical examples with a psychiatric rating scale are provided which illustrate the importance and magnitude of this bias. Formulas are subsequently given which permit the calculation of the 'true' prevalence rate from the proportion of screen positives when the sensitivity and the specificity of the test are known. The proportion of true cases among those individuals identified by a test is another important property of psychiatric measures, particularly for screening purposes. It is also called the positive predictive value and, like sensitivity and specificity predictive values are probabilities for which high values are desirable. However, this critical characteristic of a test depends on the prevalence rate, the sensitivity and the specificity of the test used. Using an example, major changes in the probability of a screen positive individual to be truly diseased are shown to depend upon varying levels of the base rate of the disease. Different values of the positive predictive value are also listed in a table according to different values of prevalence, sensitivity and specificity. The practical implications of these notions are finally highlighted, both for clinicians and researchers. In particular, the performance of a measure should not be thought of as an intrinsic or unalterable feature; rather, they will vary with the context and purpose of measurement. If inefficient studies and false inferences are to be avoided, one should critically reassess the psychometric properties when transposing an instrument from one cultural setting to another, or from clinical to community samples.

AB - Clinical assessments, psychological examinations, biological parameters and other indices are commonly used in psychiatry to measure disease status. These evaluation tools are not perfect and, in general, they will fail to perfectly discriminate 'cases' from normal controls. When a gold standard is available, their performances can be evaluated. If the results of a measurement are tabulated with true disease status in a contingency table, the proportions of cases and normals correctly identified (sensitivity and specificity) are useful coefficients of validity of the instrument studied. It is shown that unbiased estimates of the prevalence rate cannot however be derived directly from the proportion of sample respondents identified by a test. Practical examples with a psychiatric rating scale are provided which illustrate the importance and magnitude of this bias. Formulas are subsequently given which permit the calculation of the 'true' prevalence rate from the proportion of screen positives when the sensitivity and the specificity of the test are known. The proportion of true cases among those individuals identified by a test is another important property of psychiatric measures, particularly for screening purposes. It is also called the positive predictive value and, like sensitivity and specificity predictive values are probabilities for which high values are desirable. However, this critical characteristic of a test depends on the prevalence rate, the sensitivity and the specificity of the test used. Using an example, major changes in the probability of a screen positive individual to be truly diseased are shown to depend upon varying levels of the base rate of the disease. Different values of the positive predictive value are also listed in a table according to different values of prevalence, sensitivity and specificity. The practical implications of these notions are finally highlighted, both for clinicians and researchers. In particular, the performance of a measure should not be thought of as an intrinsic or unalterable feature; rather, they will vary with the context and purpose of measurement. If inefficient studies and false inferences are to be avoided, one should critically reassess the psychometric properties when transposing an instrument from one cultural setting to another, or from clinical to community samples.

UR - http://www.scopus.com/inward/record.url?scp=0025727497&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0025727497&partnerID=8YFLogxK

M3 - Article

C2 - 2049997

AN - SCOPUS:0025727497

VL - 17

SP - 73

EP - 77

JO - L'Encephale

JF - L'Encephale

SN - 0013-7006

IS - 2

ER -