How good is a test?

Start here

How do you find out whether a test gives the right answer?

Do the new test and the best available test on the same patients, then count how often they agree. The counts go into a 2 × 2 table.

The test being studied is the index test. The best available method of deciding who has the disease is the reference standard, often called the gold standard. Every patient falls into one of four cells.

Two properties of the test

Of the patients who have the disease, how many does the test pick up?

Test results in people with and without a disease usually overlap. Wherever you put the cut-off, some people land on the wrong side.

Sensitivity is the proportion of people with the disease whose test is positive. Specificity is the proportion of people without the disease whose test is negative. Both are calculated inside one disease group, so in principle they do not depend on how common the disease is.

A very sensitive test misses few cases, so a negative result makes the disease unlikely. A very specific test gives few false positives, so a positive result makes the disease likely. Pewsner and colleagues (BMJ, 2004) added two cautions. The ability to rule out also falls when specificity is low, and the ability to rule in also depends on sensitivity. Studies quoted as showing SnNout or SpPin properties may be affected by spectrum bias, partial verification bias or incorporation bias, or may be too small.

The formula

Working through the 2 × 2 table

The numbers from the sepsis example (made-up teaching numbers).

TP

true positive: disease present, test positive (90)

FP

false positive: disease absent, test positive (180)

FN

false negative: disease present, test negative (10)

TN

true negative: disease absent, test negative (720)

Sensitivity = TP ÷ (TP + FN) = 90 ÷ 100 = 90%

Specificity = TN ÷ (TN + FP) = 720 ÷ 900 = 80%

PPV = TP ÷ (TP + FP) = 90 ÷ 270 = 33.3%

NPV = TN ÷ (TN + FN) = 720 ÷ 730 = 98.6%

Prevalence = (TP + FN) ÷ total = 100 ÷ 1,000 = 10%

Accuracy = (TP + TN) ÷ total = 810 ÷ 1,000 = 81%

Youden index J = sensitivity + specificity − 1 = 0.90 + 0.80 − 1 = 0.70

Diagnostic odds ratio = (TP × TN) ÷ (FP × FN) = 64,800 ÷ 1,800 = 36

Accuracy mixes the two kinds of error and changes with prevalence, so it is a poor single summary. The answer to the parents is 33%, not 90%.

The patient's question

My patient tested positive. How likely is the disease?

That is the positive predictive value. It depends on how common the disease is in the people you test.

Positive predictive value (PPV) is the proportion of people with a positive test who have the disease. Negative predictive value (NPV) is the proportion of people with a negative test who are free of it.

PPV = (sensitivity × prevalence) ÷ [sensitivity × prevalence + (1 − specificity) × (1 − prevalence)]

This is why a test that performs well in a referral hospital gives many more false positives when it is used to screen the general population.

When the disease is rare, PPV depends much more on specificity than on sensitivity. The number of false positives is set by the specificity and by the size of the group without the disease, and that group is large. For the sepsis test at 1% prevalence, raising sensitivity from 90% to 99% raises the PPV only from 4.3% to 4.8%. Raising specificity from 80% to 99% raises it to 47.6%. NPV depends more on sensitivity. At 50% prevalence, raising sensitivity to 99% raises the NPV from 88.9% to 98.8%, while raising specificity to 99% raises it only to 90.8%.

Interactive

Change the prevalence and watch PPV and NPV change

Try this: move the prevalence from 10% down to 1% and watch the PPV fall from 33.3% to 4.3% while the NPV rises.

Probability

Why is the chance of a positive test in disease different from the chance of disease after a positive test?

Sensitivity and PPV are both conditional probabilities. The condition is on opposite sides.

A conditional probability is a probability calculated within one group only. P(A | B) is read as “the probability of A given B”. Sensitivity is P(positive | disease), calculated among people who have the disease. PPV is P(disease | positive), calculated among people whose test is positive. In the sepsis example the first is 90% and the second is 33.3%.

Hospital records and published case series are usually organised by diagnosis, so they report how often patients with a disease had a symptom or a positive test. At the bedside you need the reverse. Bayes' rule converts one into the other.

P(disease | positive) = P(positive | disease) × P(disease) ÷ P(positive)

P(positive) = P(positive | disease) × P(disease) + P(positive | no disease) × P(no disease)

Sepsis example: 0.90 × 0.10 ÷ (0.90 × 0.10 + 0.20 × 0.90) = 0.09 ÷ 0.27 = 33.3%

P(disease)

the prior (pre-test) probability: prevalence in patients like yours, or your estimate from the history and examination (0.10)

P(positive | no disease)

1 − specificity, the false-positive rate (0.20)

P(positive)

the proportion of all patients tested who test positive (270 of 1,000 = 0.27)

P(disease | positive)

the posterior (post-test) probability, which is the PPV

The PPV formula in the section above is Bayes' rule written for a test result. The likelihood ratio method in the next section is the same rule written in odds. When several diagnoses compete for the same presentation, the rule extends to all of them at once, and when tests are combined their errors may not be independent. Both are covered in the Practical Statistics chapter on Bayes' rule and combined tests.

Likelihood ratios

How much does this result change the probability of disease?

A likelihood ratio tells you how many times more likely a result is in a person with the disease than in a person without it.

LR+ = sensitivity ÷ (1 − specificity) = 0.90 ÷ 0.20 = 4.5

LR− = (1 − sensitivity) ÷ specificity = 0.10 ÷ 0.80 = 0.125

Likelihood ratios work on odds, not probabilities. Start with the pre-test probability: your estimate before the test, from prevalence and clinical findings. The result gives the post-test probability.

Pre-test odds = 0.10 ÷ 0.90 = 0.111

Positive: 0.111 × 4.5 = 0.50 odds → 0.50 ÷ 1.50 = 33.3%

Negative: 0.111 × 0.125 = 0.0139 odds → 0.0139 ÷ 1.0139 = 1.4%

odds

probability ÷ (1 − probability)

probability

odds ÷ (1 + odds)

LR above 1 raises the probability of disease, LR below 1 lowers it, and LR = 1 does not change it. McGee (2002) gave a bedside approximation that avoids odds. It is accurate to within 10 percentage points for pre-test probabilities between 10% and 90%.

Likelihood ratios come from sensitivity and specificity, so they do not change with prevalence in the way PPV does. You can apply them to your own patient's pre-test probability. Tests with several result levels can have a likelihood ratio for each level.

Interactive

Change the pre-test probability and the likelihood ratio, and watch the post-test probability

ROC curve

The test gives a number. Which cut-off should you use?

Every cut-off gives a different pair of sensitivity and specificity. The receiver operating characteristic (ROC) curve plots all of them.

• Axes: sensitivity (true-positive rate) on the y-axis, 1 − specificity (false-positive rate) on the x-axis.
• Corners: a cut-off above every result makes everyone negative, the point (0, 0). A cut-off below every result makes everyone positive, the point (1, 1). So every ROC curve runs from the lower left corner to the upper right corner. Tests differ in the path between them.
• Diagonal: a test no better than chance lies on the line from (0, 0) to (1, 1). The closer the curve comes to the top left corner (0, 1), the better the test separates the two groups.
• Area under the curve (AUC): 0.5 for chance, 1.0 for perfect separation. Hanley and McNeil (1982) showed that the AUC is the probability that a randomly chosen person with the disease is rated as more abnormal than a randomly chosen person without it, the same quantity estimated by the Wilcoxon statistic.
• Choosing a cut-off: the Youden index (sensitivity + specificity − 1, the largest vertical distance from the diagonal) or the point closest to the top left corner give equal weight to both errors. If a missed case is worse, as in neonatal sepsis, choose a lower cut-off with higher sensitivity. If a false positive leads to harmful treatment, choose a higher cut-off.
• Uses: choosing a cut-off for a laboratory test or score, comparing two tests over all cut-offs, judging how well a prediction model separates patients with and without an outcome, and reader studies in radiology.
• Slope: the slope of the curve at a point is the likelihood ratio for a result at that cut-off.

Interactive

Move the cut-off and watch sensitivity, specificity and the ROC point change

The formula

How do you build a ROC curve and calculate its area from real counts?

Teaching example (made-up numbers): an ultrasound grade from 1 (normal) to 5 (clearly abnormal) in 100 children with appendicitis and 150 without.

Step 1. Take each grade in turn as the cut-off, and call that grade and every higher grade positive. Step 2. At each cut-off, calculate sensitivity and 1 − specificity from the counts. Step 3. Plot the points, add (0, 0) and (1, 1), and join them with straight lines.

Distance to (0, 1) = √[(1 − sensitivity)² + (1 − specificity)²]

Grade 4 or more has both the highest Youden index (0.617) and the shortest distance to the top left corner (0.283). The two rules do not always choose the same cut-off.

AUC (trapezoidal rule) = sum over the strips of width × average height

AUC = 0.0075 + 0.0600 + 0.1375 + 0.2790 + 0.3920 = 0.876

The same number comes from comparing children in pairs. Of the 15,000 possible pairs of one child with and one child without appendicitis, the child with appendicitis has the higher grade in 12,360 pairs and the same grade in 1,560. Counting each tie as half, (12,360 + 780) ÷ 15,000 = 0.876. This is the Mann–Whitney (Wilcoxon) statistic divided by the number of pairs, which is why Hanley and McNeil could describe the AUC as a probability.

• Empirical or smooth curve: the trapezoidal method uses only the observed points and makes no assumption about the shape of the distributions. A binormal (parametric) method fits a smooth curve, like the one in the ROC figure above, by assuming that the results in each group follow normal distributions after a suitable transformation.
• Precision: report the AUC with its 95% confidence interval. When two tests are done on the same patients, their AUCs are correlated. DeLong, DeLong and Clarke-Pearson (Biometrics, 1988) gave a nonparametric method that estimates this correlation so that the two areas can be compared.
• Curves that cross: two tests can have the same AUC and still differ. One is better where high sensitivity is needed, the other where high specificity is needed. The partial AUC, the area over a chosen range of false-positive rates, compares them only over the range that matters clinically.
• AUC below 0.5: the test ranks people the wrong way round, with lower values in people with the disease. Reverse the direction of positivity, and the AUC becomes 1 minus the original value.

Interactive

Change the counts at each grade and watch the ROC curve and its area change

Pre-test and post-test probability

When will the test result change what you do for the patient?

A test changes management only if its result can move the probability of disease across the level at which you would treat.

The vertical distance between a curve and the diagonal is the change the result makes. For cut-off B, a positive result adds the most, 30 percentage points, when the pre-test probability is about 35%. At a pre-test probability of 90% it adds about 7 points, to 96.9%. When the pre-test probability is very high or very low, the result changes it little.

The cut-off changes the curves. Cut-off C (specificity 96%) can raise the probability by up to 52 percentage points after a positive result. Cut-off A (sensitivity 96%) can lower it by up to 52 points after a negative result. A highly specific test is most useful for ruling in, and a highly sensitive test for ruling out, as SpPin and SnNout state.

The threshold approach

Pauker and Kassirer (N Engl J Med, 1980) defined two threshold probabilities. If the probability of disease is below the testing threshold, withhold treatment and do not test. If it is above the test-treatment threshold, treat without testing. Between the two, do the test and let the result decide. In their method the thresholds also depend on the risks of the test and on the benefits and risks of the treatment.

Teaching example, ignoring any risk from the test itself. You would treat at a probability of 30% or more, and the test is cut-off B (LR+ 3.413, LR− 0.293). The testing threshold is the pre-test probability at which a positive result just reaches 30%. The test-treatment threshold is the pre-test probability at which a negative result just falls to 30%.

Odds at the treatment threshold = 0.30 ÷ 0.70 = 0.4286

Testing threshold: odds 0.4286 ÷ 3.413 = 0.126, probability 0.126 ÷ 1.126 = 11.2%

Test-treatment threshold: odds 0.4286 ÷ 0.293 = 1.463, probability 1.463 ÷ 2.463 = 59.4%

The test is worth doing when the pre-test probability lies between about 11% and 59%. Below 11%, even a positive result leaves the probability under 30%. Above 59%, even a negative result leaves it at 30% or more. The two vertical red lines in the figure mark the same points.

The ROC curve shows sensitivity and specificity but not predictive values. Shiu and Gatsonis (2008) described the predictive ROC (PROC) curve, which plots every pair of PPV and NPV as the cut-off for a positive result is varied.

Decide

Which measure answers your question?

Start from the clinical question, not from the formula.

Comparison

Why does a screening test need high sensitivity, and a diagnostic test high specificity?

A screening test only decides who needs further tests. A diagnostic test decides who is treated.

Using two tests together

Parallel testing (positive if either test is positive) raises sensitivity. Serial testing (positive only if both are positive, the usual screening-then-confirmation pathway) raises specificity. The formulas assume the two tests make errors independently.

Parallel: sensitivity = 1 − (1 − Se_A)(1 − Se_B), specificity = Sp_A × Sp_B

Serial: sensitivity = Se_A × Se_B, specificity = 1 − (1 − Sp_A)(1 − Sp_B)

Teaching example: test A has sensitivity 90% and specificity 80%, test B 80% and 95%. In parallel: sensitivity 98%, specificity 76%. In series: sensitivity 72%, specificity 99%.

These formulas use the product rule of probability. For independent events, the probability that both occur is the product of their probabilities. If the two tests tend to make the same mistakes, for example two imaging tests that both miss small lesions, the formulas overstate the gain in sensitivity from parallel testing and the gain in specificity from serial testing. Combining tests is treated in more depth in the Practical Statistics chapter on Bayes' rule and combined tests.

At 10% prevalence, test A alone has PPV 33.3% and NPV 98.6%. In series, the PPV rises to 88.9% and the NPV falls to 97.0%. In parallel, the NPV rises to 99.7% and the PPV falls to 31.2%. Serial testing is used to confirm a diagnosis and parallel testing to exclude one.

Serial testing is usually done in sequence, with test B given only to people who are positive on test A. The classification is the same as giving both tests to everyone and calling positive only those positive on both, but far fewer B tests are needed. Among 1,000 people tested at 10% prevalence, 270 are positive on test A (90 true and 180 false positives), so 270 people receive test B instead of 1,000.

Screening programmes

When is a disease worth screening for?

In 1968 WHO published Wilson and Jungner's report, Principles and practice of screening for disease. Its ten criteria are still the starting point.

1. The condition sought should be an important health problem.
2. There should be an accepted treatment for patients with recognized disease.
3. Facilities for diagnosis and treatment should be available.
4. There should be a recognizable latent or early symptomatic stage.
5. There should be a suitable test or examination.
6. The test should be acceptable to the population.
7. The natural history of the condition, including development from latent to declared disease, should be adequately understood.
8. There should be an agreed policy on whom to treat as patients.
9. The cost of case-finding (including diagnosis and treatment of patients diagnosed) should be economically balanced in relation to possible expenditure on medical care as a whole.
10. Case-finding should be a continuing process and not a “once and for all” project.

Why screen-detected patients seem to do better

• Length-time bias: slow-growing disease stays longer in the stage that screening can detect, so screening finds a larger share of the slow, less dangerous cases.
• Overdiagnosis: screening finds disease that would never have caused symptoms or death in that person's lifetime. These patients are counted as cures.
• Volunteer (healthy screenee) bias: people who come for screening are often healthier and more health-conscious than people who do not.

Because of these biases, longer survival after diagnosis does not prove that screening works. The evidence needed is a randomised trial showing fewer deaths from the disease in the group invited for screening.

Characteristics of a test

Apart from accuracy, what makes a test good?

A test must give the right answer, give the same answer when repeated, and be usable in the setting where it is needed.

A reliable test is not always valid: two observers can agree and both be wrong. A valid test must be reasonably reliable.

Analytical and diagnostic sensitivity are different things

Definitions from Saah and Hoover (1997), who advise always adding the word “analytical” or “diagnostic”.

Biases that make a test look better than it is

• Spectrum bias: comparing clear-cut severe cases with healthy volunteers. Ransohoff and Feinstein (1978) showed that without a broad spectrum of patients, sensitivity and specificity can be falsely high.
• Review bias: reading the index test while knowing the reference result, or the reverse. The two must be interpreted independently.
• Partial verification (work-up) bias: only patients with a positive index test receive the reference standard.
• Incorporation bias: the index test is part of the reference standard.
• Differential verification: patients receive different reference standards depending on the index test result, for example surgery for positives and follow-up for negatives. Lijmer and colleagues (JAMA, 1999) found that this overestimated diagnostic performance (relative diagnostic odds ratio 2.2, 95% CI 1.5 to 3.3).
• Imperfect reference standard: the reference standard itself makes some errors. Every disagreement is then counted as an error of the index test, so the accuracy estimates are biased. If the index test and the reference standard tend to make the same errors, accuracy is overestimated (Reitsma and colleagues, 2009).

Comparison

Which study design gives which measures of accuracy?

Sensitivity and specificity can come from either design below. PPV and NPV come directly only from a study that enrols patients as they present.

Lijmer and colleagues examined 184 original studies of 218 diagnostic tests, taken from published meta-analyses, and compared the diagnostic odds ratios of studies with and without each design fault. A study that enrols patients according to their test result, for example all test-positives and a sample of test-negatives, gives PPV and NPV but not sensitivity and specificity, unless the sampling fractions are known and used to reweight the counts.

By specialty

Diagnostic test problems by specialty

Common myths

Common mistaken beliefs

• ✗Myth: A test with 90% sensitivity means a positive patient has a 90% chance of disease.

  ✓That chance is the PPV, which depends on prevalence. In the sepsis example it is 33%.

• ✗Myth: Sensitivity and specificity are fixed properties of a test.

  ✓STARD 2015: "sensitivity and specificity are not fixed test properties." They vary with the patients studied and the setting.

• ✗Myth: A PPV from a hospital study applies to screening in the community.

  ✓Prevalence is lower in the community, so PPV is lower and false positives are more common.

• ✗Myth: A negative result on any sensitive test rules the disease out.

  ✓The ability to rule out also depends on specificity and on the pre-test probability. Check the negative likelihood ratio and its confidence interval.

• ✗Myth: Screen-detected patients live longer after diagnosis, so screening works.

  ✓Lead-time bias, length-time bias and overdiagnosis produce longer survival from diagnosis without fewer deaths.

• ✗Myth: A high AUC means the test performs well at the cut-off I use.

  ✓AUC summarises all cut-offs. In practice you use one cut-off, so look at sensitivity and specificity at that point.

Why this matters to you

Five consequences of misreading a diagnostic test

1. Protocol sent back

   A diagnostic accuracy study with no named reference standard, or with the reference test only for positives, is returned by the committee.

2. Inflated results in your thesis

   Comparing confirmed severe cases with healthy volunteers gives sensitivity and specificity that fall when the test is used on real patients.

3. A common viva error

   Quoting sensitivity when the examiner asks how likely a patient with a positive test is to have the disease.

4. Misreading papers

   Applying a PPV from a tertiary referral centre to screening in a district population.

5. Harm to patients

   When prevalence is low, most positive results are false. Each can mean anxiety, further invasive tests or treatment that was not needed.

History

History of diagnostic test evaluation

1. 1947

   J. Yerushalmy

   Published "Statistical problems in assessing methods of medical diagnosis, with special reference to X-ray techniques". It introduced the terms sensitivity and specificity for the proportions of diseased and non-diseased people correctly classified.

2. 1950

   W. J. Youden

   Published "Index for rating diagnostic tests" in Cancer. The Youden index, sensitivity + specificity − 1, is still used to choose cut-offs.

3. 1950s

   Signal detection theory

   ROC analysis originated in electronic signal detection theory in the early 1950s. One of its first applications was radar.

4. 1960

   Lee B. Lusted

   Calculated the first ROC curve in diagnostic radiology, re-analysing published data on reading chest films for pulmonary tuberculosis. He introduced ROC analysis to medical imaging.

5. 1968

   J. M. G. Wilson and G. Jungner, WHO

   Published Principles and practice of screening for disease, with ten criteria for deciding whether a condition is suitable for screening.

6. 1975

   T. J. Fagan

   Published a nomogram for Bayes’s theorem in a letter to the New England Journal of Medicine, converting pre-test to post-test probability with a likelihood ratio.

7. 1978

   David Ransohoff and Alvan Feinstein

   Showed that a narrow spectrum of patients and non-independent interpretation of tests can falsely raise sensitivity and specificity, using the carcinoembryonic antigen and nitro-blue tetrazolium tests as examples.

8. 1980

   Stephen Pauker and Jerome Kassirer

   Described the threshold approach to clinical decision making, with a testing threshold and a test-treatment threshold for the probability of disease.

9. 1982

   James Hanley and Barbara McNeil

   Showed that the area under the ROC curve is the probability that a randomly chosen diseased subject is ranked above a randomly chosen non-diseased subject.

10. 1988

    Elizabeth DeLong, David DeLong and Daniel Clarke-Pearson

    Published a nonparametric method for comparing the areas under correlated ROC curves from tests done on the same patients.

11. 1999

    Jeroen Lijmer and colleagues

    Showed in 184 studies of 218 tests that design faults inflate diagnostic accuracy. Studies of cases and separate controls overestimated the diagnostic odds ratio about threefold.

12. 2003

    STARD group

    Published the Standards for Reporting of Diagnostic Accuracy: a 25-item checklist and a flow diagram.

13. 2011

    Whiting and colleagues

    Published QUADAS-2 for judging risk of bias in diagnostic accuracy studies, with four domains.

14. 2015

    STARD group

    Published STARD 2015 with 30 items. A flow diagram of participants is now expected in every report.

Guidelines

Diagnostic tests in guidelines

Steps

Designing a diagnostic accuracy study, step by step

1. State the target condition, the intended use of the index test (screening, diagnosis, triage) and its role beside existing tests.
2. Choose the reference standard, and write why it was chosen.
3. Enrol a consecutive or random series of patients in whom the test would be used in practice.
4. Fix the cut-off for a positive result before the data are collected.
5. Do both tests on every patient, with a short interval and no treatment in between.
6. Read each test without knowing the result of the other.
7. Calculate the sample size for the precision you need around sensitivity and specificity (see the sample size chapter).
8. Report the flow diagram, the 2 × 2 table, how indeterminate results were handled, and estimates with 95% confidence intervals.

Reading a paper

Five checks on any diagnostic accuracy study

• Were the patients a consecutive or random series in whom the disease was suspected, rather than severe cases and healthy volunteers?
• Did every patient receive the same reference standard, whatever the index test result?
• Were the index test and the reference standard read without knowledge of each other’s result?
• Was the cut-off fixed before the data were seen?
• Is the 2 × 2 table given, with 95% confidence intervals for sensitivity and specificity?

Why is it called that?

Word surgery

Exam answers

Model answer outlines

Past papers

Every TNMGRMU question on diagnostic tests

Each one is answered in this chapter. Tap a sitting to open the university's own question paper.

1. ×2Receiver operating characteristic curve (ROC)DM Neonatology · Aug 2025DM Neonatology · Feb 2014
2. ×1Statistical methods of validation of diagnostic testsMD Biochemistry · Oct 2025
3. ×1Likelihood ratios & ROC curvePh.D. RM · Jan 2025
4. ×1A screening test requires high sensitivity but a diagnostic test requires high specificity. ExplainMD Community Medicine · May 2023
5. ×1Predictive valuesPh.D. RM · Jan 2023
6. ×1Regarding randomised clinical trials, explain the following terms: a. External validity. b. Allocation concealment. c. Stratification. d. Sensitivity analysis. e. Fragility IndexDM Critical Care Medicine · Aug 2022
7. ×1Discuss on validation of diagnostic tests. Write briefly on the Receiver operating characteristic curves and likelihood in clinical practiceDM Neonatology · Aug 2022
8. ×1Interpretation of likelihood ratiosDM Endocrinology · Aug 2022
9. ×1A receiver operating characteristic curve must inevitably pass through the upper right cornerMD Biochemistry · Jul 2021
10. ×1An Receiver Operating Charecteristic curve must inevitably pass through the lower left cornerMD Biochemistry · Nov 2021
11. ×1Receiver Operating Characteristic (ROC) CurveMD Community Medicine · Nov 2020
12. ×1Define sensitivity, specificity and predictive value of a cancer screening test. Discuss screening programme for cancer breastMD Radio Therapy · May 2019
13. ×1Draw a 2 X 2 table and describe how to calculate sensitivity, specificity, positive predictive value and negative predictive value for a diagnostic testDM Neonatology · Aug 2019
14. ×1Analytical sensitivity and specificityMD Biochemistry · Oct 2018
15. ×1Receiver operating characteristic curve and its applicationsMD Biochemistry · Oct 2018
16. ×1Describe the various characteristics of a Diagnostic test. Discuss the utility of Receiver Operating curves and Likelihood ratios in clinical practiceDM Neonatology · Aug 2018
17. ×1Positive predictive value, Negative predictive valueDM Neonatology · Feb 2017
18. ×1Likelihood RatioDM Neonatology · Aug 2017
19. ×1Receiver operating curveMD Biochemistry · Oct 2016
20. ×1Sensitivity and specificity of a diagnostic testDM Cardiology · Aug 2016
21. ×1ROC Curve and its significanceMD Biochemistry · Apr 2015
22. ×1Significance of ROC CurveMD Community Medicine · Apr 2013
23. ×1Describe the role of sensitivity and specificity of screening in clinical medicine with examples. Add a note on receiver operative charecterestic curvesMD Community Medicine · Oct 2012
24. ×1Receiver Operating Characteristics CurveMD Biochemistry · May 2011
25. ×1Discuss various characteristics of a diagnostic test in research methodologyDM Neonatology · Aug 2011
26. ×1How is a screening test for cancer evaluated? Discuss the biases caused as a result of a screening programmeMCh Surgical Oncology · Apr 2004
27. ×1Describe the imaging techniques and procedures avallable for the diagnosis of suspected calculi In the common bile duct. Highlight the sensitivity and specificity of the various techniques. Describe the roentgen features and write short notesMD Radio Diagnosis · Mar 1990
28. ×1Describe the imaging procedures available with their specific features, for the diagnosis of a suspected case of pancreatitis versus pancreatic carcinoma High- light the sensitivity and specificity of the procedures usedMD Radio Diagnosis · Oct 1990