Clinical tests: sensitivity and

specificity
Abdul Ghaaliq Lalkhen MB ChB FRCA
Anthony McCluskey BSc MB ChB FRCA

Key points
Sensitivity and specificity are
Many clinical tests are used to confirm or the disease (true positives) but 20% with the terms used to evaluate a
refute the presence of a disease or further the disease go undetected (false negatives). A high clinical test. They are
diagnostic process. Ideally such tests correctly sensitivity is clearly important where the test is independent of the
identify all patients with the disease, and simi- used to identify a serious but treatable disease population of interest
larly correctly identify all patients who are (e.g. cervical cancer). Screening the female subjected to the test.

Downloaded from http://ceaccp.oxfordjournals.org/ by guest on March 28, 2016

disease free. In other words, a perfect test is population by cervical smear testing is a sensi- Positive and negative
never positive in a patient who is disease free tive test. However, it is not very specific and a predictive values are useful
and is never negative in a patient who is in fact high proportion of women with a positive cer- when considering the value
diseased. Most clinical tests fall short of this vical smear who go on to have a colposcopy of a test to a clinician. They
are dependent on the
ideal. are ultimately found to have no underlying
prevalence of the disease in
pathology.
the population of interest.
Sensitivity, specificity, and The sensitivity and specificity
other terms Specificity of a quantitative test are
The specificity of a clinical test refers to the dependent on the cut-off
The following terms are fundamental to under- value above or below which
ability of the test to correctly identify those
standing the utility of clinical tests: the test is positive. In
patients without the disease.
general, the higher the
1. True positive: the patient has the disease and
True negatives sensitivity, the lower the
the test is positive. Specificity ¼ specificity, and vice versa.
True negatives þ False positives
2. False positive: the patient does not have the
disease but the test is positive. Therefore, a test with 100% specificity cor- Receiver operator
characteristic curves are a
3. True negative: the patient does not have the rectly identifies all patients without the disease.
plot of false positives against
disease and the test is negative A test with 80% specificity correctly reports
true positives for all cut-off
4. False negative: the patient has the disease 80% of patients without the disease as test values. The area under the
but the test is negative. negative (true negatives) but 20% patients curve of a perfect test is 1.0
without the disease are incorrectly identified as and that of a useless test, no
When evaluating a clinical test, the terms sen-
test positive (false positives). better than tossing a coin, is
sitivity and specificity are used. They are inde-
As discussed above, a test with a high sensi- 0.5.
pendent of the population of interest subjected to
tivity but low specificity results in many
the test. The terms positive predictive value
patients who are disease free being told of the
(PPV) and negative predictive value (NPV) are
possibility that they have the disease and are Abdul Ghaaliq Lalkhen MB ChB FRCA
used when considering the value of a test to a
then subject to further investigation. Although Specialist Registrar
clinician and are dependent on the prevalence of
the ideal (but unrealistic) situation is for a Salford Royal Hospitals NHS Trust
the disease in the population of interest. Hope Hospital
100% accurate test, a good alternative is to
Salford M6 8HD
Sensitivity subject patients who are initially positive to a UK
test with high sensitivity/low specificity, to a
The sensitivity of a clinical test refers to the second test with low sensitivity/high specificity. Anthony McCluskey BSc MB ChB FRCA
ability of the test to correctly identify those Consultant
In this way, nearly all of the false positives
Department of Anaesthesia
patients with the disease. may be correctly identified as disease negative. Stockport NHS Foundation Trust
Stepping Hill Hospital
True positives
Sensitivity ¼ Stockport SK2 7JE
True positives þ False negatives Positive predictive value UK
Tel: +44 (0161) 419 5869
A test with 100% sensitivity correctly ident- The PPV of a test is a proportion that is useful Fax: +44 (0161) 419 5045
ifies all patients with the disease. A test with to clinicians since it answers the question: E-mail: [email protected]
(for correspondence)
80% sensitivity detects 80% of patients with ‘How likely is it that this patient has the
doi:10.1093/bjaceaccp/mkn041
Continuing Education in Anaesthesia, Critical Care & Pain | Volume 8 Number 6 2008 221
& The Board of Management and Trustees of the British Journal of Anaesthesia [2008].
All rights reserved. For Permissions, please email: [email protected]
Clinical tests: sensitivity and specificity

disease given that the test result is positive?’ testing negative when they are ill. Therefore, the PPV of this test is
99%. However, if the number of ill people in the population is
True positives
Positive predictive value ¼ only 200 and the number of well people is 3800, the number of
True positives þ False positives
false positives increases from 20 to 38 and the PPV falls to 84%.
This discussion highlights the fact that the ability to make a
diagnosis or screen for a condition depends both on the discrimina-
Negative predictive value tory value of the test and on the prevalence of the disease in the
population of interest. If the data relating to a test are inserted into
The NPV of a test answers the question: ‘How likely is it that this a 22 contingency table, the Fisher’s exact test of many statistical
patient does not have the disease given that the test result is nega- software packages may be used to calculate sensitivity, specificity,
tive?’ PPV, NPPV, and likelihood ratio.
True negatives
Negative predictive value ¼
True negatives þ False negatives Receiver operator characteristic curves
Consider the following hypothetical example: measurement of

Downloaded from http://ceaccp.oxfordjournals.org/ by guest on March 28, 2016

high endorphin levels in SpRs in Anaesthesia has been found to be
Likelihood ratio associated with success in the final FRCA examination. A sample
A final term sometimes used with reference to the utility of tests is of SpRs is tested before the examination resulting in a range of
the likelihood ratio. This is defined as how much more likely is it endorphin values. The data are examined and an arbitrary cut-off
that a patient who tests positive has the disease compared with one point for endorphin levels is chosen above which most of the can-
who tests negative. didates passed with few failures. Despite choosing the cut-off
value in such a way that the maximum possible number of SpRs is
Sensitivity correctly classified, we may find that 10% of the cohort with
Likelihood ratio ¼
1  Specificity endorphin levels above the cut-off level failed the exam (false
positives) and 15% of the cohort with endorphin levels below the
cut-off level passed the exam (false negatives).
The relatively crude measures of sensitivity and specificity dis-
Dependence of PPV and NPV
cussed previously fail to take into account the cut-off point for a
on disease prevalence particular test. If the cut-off point is raised, there are fewer false
Unlike sensitivity and specificity, the PPV and NPV are dependent positives but more false negatives—the test is highly specific but
on the population being tested and are influenced by the prevalence not very sensitive. Similarly, if the cut-off point is low, there are
of the disease. Consider the following example: screening for sys- fewer false negatives but more false positives—the test is highly
temic lupus erythematosis (SLE) in a general population using the sensitive but not very specific.
antinuclear antibody has a low PPV because of the high number of Receiver operator characteristic curves (so called because they
false positives it yields. However, if a patient has signs of SLE were originally devised by radio receiver operators after the attack
(e.g. malar flush and joint pain), the PPV of the test increases on Pearl Harbour to determine how the US radar had failed to detect
because the population from which the patient is drawn is different
(from a general population with a low prevalence of SLE to a
clinically suspicious population with a much higher prevalence).
We may also consider a woman who presents with breathless-
ness post-partum and where one of the differential diagnoses is
pulmonary embolism. A D-dimer test would almost certainly be
elevated in this patient population; therefore, the test has a low
PPV for pulmonary embolism. However, it has a high NPV for
pulmonary embolism since a low D-dimer is unlikely to be associ-
ated with pulmonary embolism.
The dependence of PPV and NPV on the prevalence of a
disease can be illustrated numerically: consider a population of
4000 people who are divided equally into the ill and the well.
A screening test to detect the condition has a sensitivity of 99%
and a specificity of 99%. Screening this population would therefore
yield 1980 true positives and 1980 true negatives with 20 patients Fig 1 Receiver operator curves: (A) line of zero discrimination (AUC=0.5);
being tested positive when they in fact are well and 20 patients (B) typical clinical test (AUC=0.5–1.0); perfect test (AUC=1.0).

222 Continuing Education in Anaesthesia, Critical Care & Pain j Volume 8 Number 6 2008
 Clinical tests: sensitivity and specificity

the Japanese aircraft) are a plot of (12specificity) of a test on the contribution in providing helpful comments and advice on this
x-axis against its sensitivity on the y-axis for all possible cut-off manuscript.
points. An identical plot is produced when the false positive rate of
a test is shown on the x-axis against the true positive rate on the Bibliography
y-axis (Fig. 1). An ideal test is represented by the upper curve in the
1. Bland M. An Introduction to Medical Statistics, 3rd Edn. Oxford: Oxford
figure. The middle curve represents the characteristics of a test more University Press, 2000
typically seen in routine clinical use. The area under this curve 2. Altman DG. Practical Statistics for Medical Research. London: Chapman &
(AUC) represents the overall accuracy of a test, with a value Hall/CRC, 1991
approaching 1.0 indicating a high sensitivity and specificity. The 3. Rumsey D. Statistics for Dummies. New Jersey: Wiley Publishing Inc., 2003
dotted line on the graph represents the line of zero discrimination 4. Swinscow TDV. Statistics at Square One. Available from http://www.bmj.
with an AUC of 0.5 (the test is no better than tossing a coin). com/statsbk/ (accessed 20 October 2008)
5. Greenhalgh T. How to Read a Paper. London: BMJ Publishing, 1997
Acknowledgements 6. Elwood M. Critical Appraisal of Epidemiological Studies and Clinical Trials.
2nd Edn. Oxford: Oxford University Press, 1998
The authors are grateful to Professor Rose Baker,

Downloaded from http://ceaccp.oxfordjournals.org/ by guest on March 28, 2016

Department of Statistics, Salford University for her valuable Please see multiple choice questions 19 –22

Continuing Education in Anaesthesia, Critical Care & Pain j Volume 8 Number 6 2008 223