about

Tool : Measures of diagnostic accuracies of a 2x2 table

Thomas Keller, ACOMED statistik, www.acomed-statistik.de.

Disease
Test D+ D- Total
T+ 191 19 210 width of CI
T- 90 112 202 case-control study y/n
Total 281 131 412
logit
Sens 0.6797 0.6217 0.7339 0.7525 Sens:
Spec 0.8550 0.7828 0.9104 1.7741 Spec:
PPV 0.9095 0.8623 0.9446 2.3078 PPV:
NPV 0.5545 0.4831 0.6242 0.2187 NPV:
Prev 0.6820 0.6347 0.7268 0.7632 Prev:
DLR+ 4.6865 3.0687 7.1572 DLR+
DLR- 0.3746 0.3115 0.4505 DLR-
Calculation of exact CI for Sens, Spec Calculation of appr. CI for predictive val
Sens Spec Prev PPV NPV (PPV and NPV for case control study)
n1 191 112 281 191 112
n2 281 131 412 210 202
alpha/2 0.0250 0.0250 0.0250 0.0250 0.0250 ln(DLR+)
u 0.6217 0.7828 0.6347 0.8623 0.4831 ln(DLR-)
x 0.6797 0.8550 0.6820 0.9095 0.5545 logit(prev)+ln(DLR+)
o 0.7339 0.9104 0.7268 0.9446 0.6242 -logit(prev)-ln(DLR-)
Tool to calculate measures of diagnostic accuracy with confidence intervals
If disease prevalence is correctly reflected by study, then case-control-study = n
If disease prevalence is correctly reflected by study, then case-control-study = n
If disease prevalence is NOT correctly reflected by study, then case-control-study = y. Please input assumed prevale
Sens=P(T+|D+)=n(D+,T+)/n(D+)=TP/(TP+FN) probability of pos. test result given disease
Spec=P(T-|D-)=n(D-,T-)/n(D-)=TN/(TN+FP) probability of neg. test result given non-disease
PPV=P(D+|T+)=n(T+,D+)/n(T+)=TP/(TP+FP) probability of disease given positive test result
NPV=P(D-|T-)=n(T-,D-)/n(T-)=TN/(TN+FN) probability of non-disease given neg. test-result
DLR+=P(T+|D+)/P(T+|D-) = Sens/(1-Spec) post odds/pre odds for positive test result, large value indicates good
DLR-=P(T-|D+)/P(T-|D-) = (1-Sens)/Spec post odds/pre odds for negative test result, low value indicates good
Don't use measures like efficiacy, sum of sens and spec, … to describe diagnostic accuracy of a test!!!
Abbreviations:
Sens Sensitivity D+ Disease
Spec Specificity D- Non disease
PPV positive predictive value T+ positive test results
NPV negative predictive value T- negative test result
DLR+ positive diagnostic likelihood ratio P(a|b) conditional probability: probability of a given b
DLR- negative diagnostic likelihood ratio
prev prevalence
odds probability/(1-probability)

Seite 1
 about

logit ln(x/(1-x))

Protection: no password is neccecary.

28.7.2011, Vs. 2
12.1.2012, Vs. 3: Handling of cells=0 (sens or spec =0 or 1)
18.2.2016, Vs. 4: , and . Can be used as decimal separators.
15.7.2016, Vs. 5: check whether input of prevalence ist between 0 and 100%

Seite 2
 about

0.001

95% Ignore this cell

n 0.1

67.97% (95%-width of CI: 62.17% .. 73.39%)

85.5% (95%-width of CI: 78.28% .. 91.04%)
90.95% (95%-width of CI: 86.23% .. 94.46%)
55.45% (95%-width of CI: 48.31% .. 62.42%)
68.2% (95%-width of CI: 63.5% .. 72.7%)
4.686 (95%-width of CI: 3.069 .. 7.157)
0.375 (95%-width of CI: 0.312 .. 0.45)
Calculation of appr. CI for predictive values and DLR
PPV and NPV for case control study)

1.545 1.121 1.968

-0.982 -0.797 -1.166
2.308 1.884 2.731
0.219 0.034 0.403

y = y. Please input assumed prevalence.

given disease
given non-disease
ositive test result
en neg. test-result
e test result, large value indicates good test
ve test result, low value indicates good test
ic accuracy of a test!!!

bability: probability of a given b

Seite 3
 tool

Tool : Measures of diagnostic accuracies of a 2x2 table

Thomas Keller, ACOMED statistik, www.acomed-statistik.de.

Disease
Test D+ D- Total
T+ 191 19 210 width of CI
T- 90 112 202 case-control study y/n
Total 281 131 412
logit
Sens 0.6797 0.6217 0.7339 0.7525 Sens:
Spec 0.8550 0.7828 0.9104 1.7741 Spec:
PPV 0.9095 0.8623 0.9446 2.3078 PPV:
NPV 0.5545 0.4831 0.6242 0.2187 NPV:
Prev 0.6820 0.6347 0.7268 0.7632 Prev:
DLR+ 4.6865 3.0687 7.1572 DLR+
DLR- 0.3746 0.3115 0.4505 DLR-
Calculation of exact CI for Sens, Spec Calculation of appr. CI for predictive value
Sens Spec Prev PPV NPV (PPV and NPV for case control study)
n1 191 112 281 191 112
n2 281 131 412 210 202
alpha/2 0.0250 0.0250 0.0250 0.0250 0.0250 ln(DLR+)
u 0.6217 0.7828 0.6347 0.8623 0.4831 ln(DLR-)
x 0.6797 0.8550 0.6820 0.9095 0.5545 logit(prev)+ln(DLR+)
o 0.7339 0.9104 0.7268 0.9446 0.6242 -logit(prev)-ln(DLR-)
Tool to calculate measures of diagnostic accuracy with confidence intervals
If disease prevalence is correctly reflected by study, then case-control-study = n
If disease prevalence is correctly reflected by the study, then case-control-study = n
If disease prevalence is NOT correctly reflected by study, then case-control-study = y.
In that case, input assumed prevalence (>0 and <100%).
Sens=P(T+|D+)=n(D+,T+)/n(D+)=TP/(TP+FN) probability of pos. test result given disease
Spec=P(T-|D-)=n(D-,T-)/n(D-)=TN/(TN+FP) probability of neg. test result given non-disease
PPV=P(D+|T+)=n(T+,D+)/n(T+)=TP/(TP+FP) probability of disease given positive test result
NPV=P(D-|T-)=n(T-,D-)/n(T-)=TN/(TN+FN) probability of non-disease given neg. test-result
DLR+=P(T+|D+)/P(T+|D-) = Sens/(1-Spec) post odds/pre odds for positive test result, large value indicates good t
DLR-=P(T-|D+)/P(T-|D-) = (1-Sens)/Spec post odds/pre odds for negative test result, low value indicates good te
Don't use measures like efficiacy, sum of sens and spec, … to describe diagnostic accuracy of a test!!!
Abbreviations:
Sens Sensitivity D+ Disease
Spec Specificity D- Non disease
PPV positive predictive value T+ positive test results
NPV negative predictive value T- negative test result
DLR+ positive diagnostic likelihood ratio P(a|b) conditional probability: probability of a given b
DLR- negative diagnostic likelihood ratio
prev prevalence

Seite 4
 tool

odds probability/(1-probability)
logit ln(x/(1-x))

Protection: no password is neccecary.

28.7.2011, Vs. 2
12.1.2012, Vs. 3: Handling of cells=0 (sens or spec =0 or 1)
18.2.2016, Vs. 4: , and . Can be used as decimal separators.

Seite 5
 tool

0.001

Ignore this cell

95% 200%
n please input a number
>0% and <100%!

67.97% (95%-width of CI: 62.17% .. 73.39%)

85.5% (95%-width of CI: 78.28% .. 91.04%)
90.95% (95%-width of CI: 86.23% .. 94.46%)
55.45% (95%-width of CI: 48.31% .. 62.42%)
68.2% (95%-width of CI: 63.5% .. 72.7%)
4.686 (95%-width of CI: 3.069 .. 7.157)
0.375 (95%-width of CI: 0.312 .. 0.45)
Calculation of appr. CI for predictive values and DLR
(PPV and NPV for case control study)

1.545 1.121 1.968

-0.982 -0.797 -1.166
2.308 1.884 2.731
0.219 0.034 0.403

given disease
t given non-disease
positive test result
ven neg. test-result
ive test result, large value indicates good test
tive test result, low value indicates good test
ostic accuracy of a test!!!

obability: probability of a given b

Seite 6