Diagnostic Odds Ratio Calculator
Diagnostic Odds Ratio Calculator
Collapse a test’s whole performance into one number for comparing tests and pooling studies — then see why that number is the wrong tool at the bedside.
Diagnostic Odds Ratio
Sens, spec → DORA test with 90% sensitivity and 90% specificity
Formula
= [sensitivity ÷ (1 − specificity)] ÷ [(1 − sensitivity) ÷ specificity]
= (sensitivity × specificity) ÷ [(1 − sensitivity) × (1 − specificity)]
- LR+
- sensitivity ÷ (1 − specificity): how many times more often a positive result occurs in disease than in health
- LR−
- (1 − sensitivity) ÷ specificity: how many times more often a negative result occurs in disease than in health. A good rule-out test has a very small LR−
- DOR
- the ratio of the two. Equivalently, the odds of a positive result in someone with the disease divided by the odds of a positive result in someone without it. One number for the whole 2×2 table
- DOR = 1
- no discrimination whatsoever. Below 1 the test is pointing the wrong way, which in practice means an input or a coding error
- undefined at 100%
- if sensitivity is 100%, LR− is zero and the ratio is infinite; if specificity is 100%, LR+ is infinite. Meta-analyses handle this by adding 0.5 to each cell of the 2×2 table, a continuity correction that is a convenience rather than a truth
- what it cannot tell you
- which direction the test moved the probability. The DOR is symmetric: swap sensitivity and specificity and it does not change, even though the clinical use of the test reverses completely. For one patient you need LR+ and LR− separately
- what it is good for
- comparing tests, summarising a body of literature, and meta-analysis — where the summary receiver operating characteristic curve is built on the log of this ratio precisely because a single number can be pooled across studies with different cut-offs
Worked example
A test with 90% sensitivity and 90% specificity
LR+ = 0.90 ÷ (1 − 0.90) = 0.90 ÷ 0.10 = 9.0
LR− = (1 − 0.90) ÷ 0.90 = 0.10 ÷ 0.90 = 0.111
DOR = 9.0 ÷ 0.111 = 81.0
The same result the short way: (0.90 × 0.90) ÷ (0.10 × 0.10) = 0.81 ÷ 0.01 = 81
Now take a test that is 98% sensitive and 40% specific — a classic rule-out test. LR+ = 0.98 ÷ 0.60 = 1.63, LR− = 0.02 ÷ 0.40 = 0.05, and the DOR is 32.7
And a test that is 40% sensitive and 98% specific — a rule-in test. LR+ = 0.40 ÷ 0.02 = 20, LR− = 0.60 ÷ 0.98 = 0.61, and the DOR is 32.7 again. Two tests you would use in opposite clinical situations, and one identical summary number. That is the limitation of the DOR in a single line
Two tests, one diagnostic odds ratio
| Sensitivity | Specificity | LR+ | LR− | DOR | What it is for |
|---|---|---|---|---|---|
| 98% | 40% | 1.63 | 0.05 | 32.7 | Ruling out — a negative result is decisive, a positive one means little |
| 40% | 98% | 20.0 | 0.61 | 32.7 | Ruling in — a positive result is decisive, a negative one means nothing |
| 90% | 90% | 9.0 | 0.11 | 81.0 | Both directions, moderately well |
| 95% | 95% | 19.0 | 0.05 | 361.0 | Both directions, strongly |
| 80% | 80% | 4.0 | 0.25 | 16.0 | Both directions, weakly |
| 50% | 50% | 1.0 | 1.0 | 1.0 | Nothing at all |
Rough orientation for the size of a diagnostic odds ratio
| DOR | Equivalent symmetric test | Reading |
|---|---|---|
| 1 | 50% / 50% | No discrimination whatsoever |
| 4 | 67% / 67% | Almost none |
| 16 | 80% / 80% | Weak to moderate |
| 50 | 88% / 88% | Moderate |
| 81 | 90% / 90% | Good |
| 361 | 95% / 95% | Strong |
One number for a whole test, and what it hides
The diagnostic odds ratio is the ratio of the two likelihood ratios: LR+ divided by LR−. Written out in terms of sensitivity and specificity it is (sensitivity × specificity) divided by ((1 − sensitivity) × (1 − specificity)), and read as odds it is the odds of a positive result in someone with the disease divided by the odds of a positive result in someone without it. For a test with 90% sensitivity and 90% specificity that is 9.0 ÷ 0.111, or 81. A value of 1 means the test discriminates nothing at all; below 1 it is pointing the wrong way, which in practice means an input error.
Its virtue is that it is a single number, and it is a single number that does not change with prevalence. That makes it genuinely useful for two jobs. The first is comparing tests: saying that one assay has a diagnostic odds ratio of 80 and another 20 is a fair summary of which discriminates better overall. The second is meta-analysis, where studies report different cut-offs and therefore different pairs of sensitivity and specificity; the summary receiver operating characteristic approach is built on the logarithm of the diagnostic odds ratio precisely because one number per study can be pooled when two correlated numbers cannot.
What it cannot do is help with a patient, and the reason is worth being precise about. The formula is symmetric in sensitivity and specificity: swap them and the answer is unchanged. A test that is 98% sensitive and 40% specific has a diagnostic odds ratio of 32.7, and so does a test that is 40% sensitive and 98% specific. The first is a rule-out test whose negative result is decisive and whose positive result means very little; the second is a rule-in test with the opposite profile. They would be ordered in different clinical situations, interpreted in opposite ways, and they share a summary statistic. No amount of care in reading that statistic recovers the distinction, because the information was discarded when the two likelihood ratios were divided.
So the diagnostic odds ratio belongs in a comparison table and in a meta-analysis, and it does not belong in a decision about one person. There are two further practical cautions. It is undefined when sensitivity or specificity reaches 100% — one of the likelihood ratios becomes zero or infinite — and the usual remedy of adding half a count to each cell of the 2×2 table is a convenience rather than a truth. And because it is a ratio of ratios it grows very fast: 90%/90% gives 81 and 95%/95% gives 361, so an impressive-looking value can rest on a small study and a wide confidence interval. Look for the interval, and then look at LR+ and LR− separately before you use the test on anybody.
Frequently asked questions
How do you calculate the diagnostic odds ratio?
Divide LR+ by LR−, or equivalently compute (sensitivity × specificity) ÷ ((1 − sensitivity) × (1 − specificity)). A test with 90% sensitivity and 90% specificity has LR+ 9.0 and LR− 0.111, giving a diagnostic odds ratio of 81.
What is a good diagnostic odds ratio?
There is no agreed classification, which is itself informative. The most honest way to read one is to ask what symmetric performance would produce it: 16 corresponds to 80% sensitivity with 80% specificity, 81 to 90%/90%, and 361 to 95%/95%. A value of 1 means the test discriminates nothing.
Why can’t I use the diagnostic odds ratio for an individual patient?
Because it is symmetric in sensitivity and specificity and therefore says nothing about which direction the test moved the probability. A test that is 98% sensitive and 40% specific and one that is 40% sensitive and 98% specific both have a diagnostic odds ratio of 32.7, and they are used in opposite clinical situations. For a patient you need LR+ and LR− separately, applied to a pre-test probability.
What happens when sensitivity or specificity is 100%?
The diagnostic odds ratio is undefined. A sensitivity of 100% makes LR− zero and the ratio infinite; a specificity of 100% makes LR+ infinite. Meta-analyses usually add 0.5 to each cell of the 2×2 table as a continuity correction, which lets the arithmetic proceed but does not create information that the study did not contain.
Does the diagnostic odds ratio change with prevalence?
No, and that is one of its genuine strengths. Like sensitivity, specificity and the likelihood ratios it is calculated within the columns of the 2×2 table, so it is a property of the test rather than of the population tested — unlike the predictive values, which move substantially with prevalence.
Related calculators
References
- Glas AS, Lijmer JG, Prins MH, Bonsel GJ, Bossuyt PMM. The diagnostic odds ratio: a single indicator of test performance. J Clin Epidemiol. 2003;56(11):1129–1135.
- Deeks JJ, Altman DG. Diagnostic tests 4: likelihood ratios. BMJ. 2004;329(7458):168–169.
- Moses LE, Shapiro D, Littenberg B. Combining independent studies of a diagnostic test into a summary ROC curve: data-analytic approaches and some additional considerations. Stat Med. 1993;12(14):1293–1316.
Medical Disclaimer: The tools and content provided here are for educational and reference purposes only. They are not intended to substitute for professional medical advice, diagnosis, or treatment. Clinical decisions should always be based on the comprehensive assessment of a qualified healthcare professional.