Sensitivity and Specificity for NEET PG: A Worked 2x2 Table, PPV, NPV and the Prevalence Effect

Sensitivity and specificity look easy until a question hands you a table and asks for a predictive value. Then the denominators blur, and a familiar formula gives the wrong answer. This post builds one 2x2 table slowly, reads every statistic off it, and then shows what happens to those numbers when the same test is used in a different population.

Every number in the worked examples below is hypothetical. They are invented to make the arithmetic clean, and they do not describe any real test or disease. The formulas themselves come from a standard review of the topic (Florkowski, Clinical Biochemist Reviews, 2008).

Key Takeaways

• Sensitivity, TP/(TP+FN), is the share of diseased people who test positive. Specificity, TN/(TN+FP), is the share of healthy people who test negative. Both are calculated within the disease columns of the 2x2 table.

• PPV = TP/(TP+FP) and NPV = TN/(TN+FN). Both use the test-result rows, so both change when prevalence changes.

• In the hypothetical 90% sensitive, 80% specific test, PPV rises from 4.3% to 33.3% to 81.8% as prevalence goes from 1% to 10% to 50%, while NPV falls from 99.9% to 88.9%.

• When a very sensitive test comes back negative, disease becomes unlikely (SnNout). When a very specific test comes back positive, disease becomes likely (SpPin). Treat these as shortcuts, not laws.

• Requiring two tests to both be positive raises specificity and lowers sensitivity. Accepting either test as positive does the reverse.

What Sensitivity and Specificity Measure

Sensitivity and specificity describe how a test behaves in people whose true disease status is already known. The truth comes from a reference standard, and the test is judged against it. Sensitivity looks only at the diseased group and counts how many of them the test picks up. Specificity looks only at the healthy group and counts how many the test correctly clears.

Those two questions give the formulas:

• Sensitivity = TP / (TP + FN), the positivity in disease.
• Specificity = TN / (TN + FP), the negativity in health.

Few false negatives means a highly sensitive test seldom lets a case slip through. Few false positives means a highly specific test seldom flags someone who is well. Because moving a test's cut-off trades one for the other, a lower cut-off (higher sensitivity) suits screening, and a higher cut-off (higher specificity) suits confirmation (Florkowski, 2008). The wider public-health logic of screening is covered in the levels of prevention post.

The 2x2 Table Layout

The 2x2 table is the single tool that makes all of this mechanical. Put the test result in the rows (positive or negative) and the true disease status in the columns (present or absent). Crossing the two gives four cells: true positives (TP), false positives (FP), false negatives (FN) and true negatives (TN).

Here is the layout to draw before every question:

	Disease present	Disease absent
Test positive	TP	FP
Test negative	FN	TN

One rule goes with this layout. Move down a column for sensitivity and specificity, and along a row for PPV and NPV. Picking the wrong direction causes most of the errors students make here.

A Worked 2x2 Table at 10% Prevalence

A worked table is the fastest way to see every statistic at once. Imagine a hypothetical test with a sensitivity of 90% and a specificity of 80%, applied to 1,000 people, of whom 10% have the disease. Again, these figures are illustrative only.

Start with the people, then fill the cells. Ten per cent of 1,000 is 100 diseased people, which leaves 900 healthy people. A sensitivity of 90% means 90 of the 100 diseased people test positive (TP = 90) and 10 test negative (FN = 10). A specificity of 80% means 720 of the 900 healthy people test negative (TN = 720) and 180 test positive (FP = 180).

	Disease present	Disease absent	Total
Test positive	TP = 90	FP = 180	270
Test negative	FN = 10	TN = 720	730
Total	100	900	1,000

Now read the statistics off the table:

• Sensitivity = 90/100 = 90%
• Specificity = 720/900 = 80%
• PPV = 90/270 = 33.3%
• NPV = 720/730 = 98.6%

The surprise is the PPV. A test that is 90% sensitive and 80% specific still gives a positive result that is right only one time in three here. Of the 270 people who test positive, 180 are healthy.

Predictive Values: PPV and NPV

A clinician with a result in hand wants to know what it means for the patient in front of them, and predictive values give that answer. Given a positive test, the chance that disease is really present is the PPV, TP/(TP+FP). Given a negative test, the chance that the patient is truly disease-free is the NPV, TN/(TN+FN).

Notice that the denominators are the row totals (270 and 730), not the column totals. That single switch is the entire difference between the two pairs of statistics. A review of the topic stresses that predictive values are not intrinsic to the test and depend on the clinical context in which it is used (Florkowski, 2008).

Other Statistics Derived From the Table

Three more figures can be derived directly from the same four cells. They are not named in the core references used here, so treat them as derived from the 2x2 table rather than as separately sourced definitions.

• False-positive rate = FP/(FP+TN) = 180/900 = 20%, which equals 1 minus specificity.
• False-negative rate = FN/(FN+TP) = 10/100 = 10%, which equals 1 minus sensitivity.
• Accuracy = (TP+TN)/total = 810/1,000 = 81%.

A frequent slip is to write the false-positive rate as 1 minus PPV. It is not. The false-positive rate uses the healthy column, so it is always 1 minus specificity.

Likelihood Ratios From the Same Numbers

Likelihood ratios package sensitivity and specificity into a single multiplier. LR+ = sensitivity/(1 − specificity) = 0.90/0.20 = 4.5. LR− = (1 − sensitivity)/specificity = 0.10/0.80 = 0.125.

They connect to the table through odds. Pre-test odds are prevalence/(1 − prevalence), here 0.10/0.90 = 0.111. Post-test odds equal pre-test odds times the likelihood ratio, so 0.111 × 4.5 = 0.5. Converting odds back to probability gives 0.5/1.5 = 33.3%, which matches the PPV already calculated.

One trap is worth memorising: LR− is not 1/LR+. It has its own formula. Large changes in probability are expected only from fairly extreme values: an LR+ greater than 10, or an LR− smaller than 0.1 (Florkowski, 2008).

How Prevalence Changes PPV and NPV

Hold the test fixed and change only the population, and the predictive values move sharply. Under the usual exam convention, sensitivity and specificity stay constant while prevalence changes. PPV then rises with prevalence and NPV falls.

The table below applies the same hypothetical test (90% sensitive, 80% specific) to 1,000 people at three prevalences. The 10% row repeats the table above.

Prevalence	Diseased / healthy	TP	FN	TN	FP	PPV	NPV	Accuracy
1%	10 / 990	9	1	792	198	9/207 = 4.3%	792/793 = 99.9%	801/1,000 = 80.1%
10%	100 / 900	90	10	720	180	90/270 = 33.3%	720/730 = 98.6%	81.0%
50%	500 / 500	450	50	400	100	450/550 = 81.8%	400/450 = 88.9%	850/1,000 = 85.0%

The pattern is the point. At 1% prevalence, 198 healthy people are flagged for every 9 true cases, so a positive result is wrong about 19 times out of 20. At 50% prevalence the same test is right more than four times in five when it is positive.

Why a Good Test Can Still Have a Poor PPV

The reason is simple arithmetic about who is in the pool. In a rare disease, the healthy group is far larger than the diseased group. A small false-positive rate multiplied by that large group can still yield more false positives than a high sensitivity can yield true positives from the few people who are ill.

This is why a good sensitivity and specificity do not guarantee a useful positive result. It also means the same biomarker can show a lower PPV in a low-prevalence general-practice setting than in an emergency department (Florkowski, 2008).

Accuracy Is Not Immune

Accuracy is often assumed to be a stable property of a test. The table shows otherwise: accuracy drifts from 80.1% to 81.0% to 85.0% as prevalence rises, even though sensitivity and specificity never moved. Accuracy is a weighted average of the two, and the weights are the proportions of diseased and healthy people. Change the prevalence and you change the weights.

A Note on Real Populations

The convention that sensitivity and specificity are fixed is a teaching simplification. In real studies they can vary with case mix, a phenomenon called the spectrum effect. A BMJ analysis reports a simulation in which a fall in prevalence changed likelihood ratios by up to ten-fold, and a hypothetical in which a test looked like 97% sensitive and 97% specific in a case-control study but 72% and 92% in a general-population study (Usher-Smith et al., BMJ 2016).

For the exam, use the convention. In practice, remember that a test's performance in a hospital clinic may not carry over to a community screening camp.

SnNout and SpPin: Useful Shortcuts With a Caveat

Two mnemonics compress the rule-in and rule-out logic. SnNout: with a highly sensitive test, a negative result helps rule disease out. SpPin: with a highly specific test, a positive result helps rule disease in.

The reasoning follows the table. With very few false negatives, a negative result from a highly sensitive test seldom conceals disease. With very few false positives, a positive result from a highly specific test seldom turns out to be wrong.

Do not over-apply them. A 2004 BMJ note of caution (SpPIn and SnNOut) warns that these mnemonics should not be applied blindly. For MCQs they are reliable. In real decisions the prevalence and the pre-test probability still matter.

A related pitfall is mixing the two mnemonics. Sensitivity pairs with ruling out (the N in SnNout), and specificity pairs with ruling in (the P in SpPin).

Cut-offs and the Sensitivity-Specificity Trade-off

Most real tests give a number, not a yes or no, so someone must choose a cut-off. Moving the cut-off trades one statistic against the other. An ROC curve shows this trade visually: every possible cut-off becomes one point, with 1 − specificity (the false-positive rate) on the x-axis and sensitivity (the true-positive rate) on the y-axis.

Lowering the cut-off labels more people as positive. Sensitivity rises, specificity falls, and NPV rises. This suits screening and rule-out. Raising the cut-off does the opposite: specificity and PPV rise at the cost of sensitivity, which suits confirmation or rule-in (Florkowski, 2008).

A single number, the area under the ROC curve (AUC), sums up how well the test discriminates overall. One definition describes the area under the receiver operating characteristic curve as the probability that a randomly chosen diseased person is ranked higher than a randomly chosen non-diseased person (Usher-Smith et al., 2016).

Change	Sensitivity	Specificity	PPV	NPV	Better suited to
Lower the cut-off	Rises	Falls	Tends to fall	Rises	Screening, ruling out
Raise the cut-off	Falls	Rises	Rises	Tends to fall	Confirming, ruling in

Combining Tests: Series and Parallel Testing

Using two tests together changes the sensitivity and specificity of the combination. The effect depends on the decision rule, not on the order in which tests are done. Radiopaedia's article on multiple tests gives formulas that assume the tests are conditionally independent.

Take two hypothetical tests. Test A has sensitivity 0.90 and specificity 0.80. Test B has sensitivity 0.95 and specificity 0.85.

Decision rule	Combined sensitivity	Combined specificity	Net effect
Both must be positive ("A and B")	0.90 × 0.95 = 0.855	0.80 + 0.85 − (0.80 × 0.85) = 0.97	Sensitivity falls, specificity rises
Either positive ("A or B")	0.90 + 0.95 − (0.90 × 0.95) = 0.995	0.80 × 0.85 = 0.68	Sensitivity rises, specificity falls

The intuition is straightforward. Demanding two positives makes a positive call harder to earn, so fewer healthy people qualify and fewer diseased people do too. Accepting either test as positive makes the call easier, so almost no diseased person is missed but more healthy people are flagged.

When to Use Which Rule

The combined-sensitivity and specificity table above points to a practical reading. The "either positive" rule gives the higher sensitivity (0.995), so it fits situations where missing a case is costly and extra false positives are tolerable. The "both positive" rule gives the higher specificity (0.97), so it fits confirming a diagnosis when false positives are costly, accepting that some true cases will be missed.

Watch the Terminology

Terminology varies between textbooks. Radiopaedia keeps two ideas apart: timing (parallel means done at the same time, serial means one after the other) and the decision rule (both positive, or either positive). Some sources use the series and parallel labels the other way round. When an MCQ uses either word, check how the question defines it before answering.

Common Mistakes With Sensitivity and Specificity

Most lost marks on this topic come from a handful of repeatable errors. Check each of these points before you lock in an answer.

1. Wrong denominator. Sensitivity and specificity divide by disease columns. PPV and NPV divide by test rows.
2. Assuming good sensitivity and specificity mean a good PPV. At 1% prevalence, a 90% sensitive and 80% specific test gives a PPV of 4.3%.
3. Forgetting that prevalence leaves sensitivity and specificity unchanged by convention. Real populations can differ because of the spectrum effect.
4. Swapping SnNout and SpPin. Sensitive tests rule out; specific tests rule in.
5. Writing false-positive rate as 1 − PPV. It is 1 − specificity.
6. Mixing up series and parallel. Confirm which definition the question uses.
7. Moving the cut-off the wrong way. A lower cut-off raises sensitivity and lowers specificity.
8. Treating accuracy as fixed. In the table above it changed from 80.1% to 85.0%.
9. Taking LR− as 1/LR+. LR− = (1 − sensitivity)/specificity.

A Quick Method for Any Question

When numbers are given, draw the empty 2x2 table first. Fill in the people totals from prevalence, then the cells from sensitivity and specificity, then read off whatever the question asks. Working with 1,000 people usually keeps the arithmetic whole-number.

When only percentages are given and no counts, still choose a convenient population such as 1,000. The answer does not depend on the number you pick, because every statistic is a ratio.

Putting It Into Practice

Test yourself by changing one input at a time. Keep the 90% and 80% figures, pick a new prevalence such as 20%, rebuild the table and recompute PPV and NPV. Then keep the prevalence and change the specificity to see which cell moves. A few of these by hand will do more than rereading the formulas.

Here is the worked answer for 20% prevalence with the 90% sensitive, 80% specific test, again using hypothetical numbers. Of 1,000 people, 200 are diseased and 800 are healthy. Among the 200 diseased, TP = 180 and FN = 20. Among the 800 healthy, TN = 640 and FP = 160.

• PPV = 180/(180 + 160) = 180/340 = 52.9%
• NPV = 640/(640 + 20) = 640/660 = 97.0%

Compare this with the 10% table: as prevalence doubled, PPV rose from 33.3% to 52.9% and NPV fell from 98.6% to 97.0%.

When you want exam-style questions on this and related biostatistics topics, work through the community medicine biostatistics practice questions or the wider NEET PG question bank, and sort the ones you miss by the mistake they represent.

Conclusion

Sensitivity and specificity belong to the test and the reference standard, and they are read down the columns of the 2x2 table. PPV and NPV belong to the test in a particular population, and they are read across the rows.

Prevalence is the bridge between the two, which is why the same hypothetical test can have a PPV of 4.3%, 33.3% or 81.8% without changing at all. If you draw the table, fill it from a population of 1,000 and name which denominator each statistic needs, most questions on this topic become routine arithmetic.