1Principles of ScreeningJeffrey Murawsky, MDAssistant Professor, MedicinePerfect Screening Test Always correct Repeatable Safe, painless, quick, inexpensive Makes a clinical differenceReality is Quite a Different Prospect!Basic Two by TwoTrue Positives False PositivesFalse Negatives True NegativesGold StandardDisease Positive Disease NegativeTestPositiveNegativeSensitivity Proportion of those with disease defined by gold standard testing who are labeled by the test in question as positive True positives/ all subjects with gold standard proven diseaseSpecificity Proportion of those without disease defined by gold standard testing who are labeled as negative by the test in question True Negatives/ all subjects disease free by gold standard testingPrevalence In the 2x2 table: the number of those with disease by gold standard ie 5/100,000 Clinically Pretest Probability of the patient is very similar (ie 30% chance of a disease based on risk factors)2Defining a Positive Test Tests are usually yield a continuous variable An artificial cut off is needed to define the positive or abnormal values from the normal or negative values. The Receiver operating characteristic curve demonstrates the trade offROC Curve As specificity is increased sensitivity is lost. The closer to the upper left corner your values yield the better balance in the test characteristics.ExampleND Disease: “ an uncontrollable urge to watch a football team without any hope of winning (especially bowl games)”We know from research at LUMC that in our medical school this disease occurs in 1 in 10 medical studentsJeff’s Test is questionnaire available and has defined sensitivities andspecificities by population testingGiven 2000 medical students Have ND Don’t Have ND200 with ND 1800 without NDTrue Positives False PositivesFalse Negatives True NegativesJeff’s TestPositiveNegativeJeff’s TestSensitivity: 80%Specificity: 90%That means: 80% of 200 or 160 will be True Positives and90% of 1800 or 1620 will be True NegativesThru the magic of Mathematics:Given 2000 medical students Have ND Don’t Have NDPrevalence of 1 in 10 = 200 with ND 1800 without NDTrue Positives False Positives160 180False Negatives True Negatives20 1620Jeff’s TestPositiveNegativeJeff’s TestSensitivity: 80%Specificity: 90%Now let’s assume that the prevalence of ND disease changes to 1 in 200 in another population.Given 2,000 persons in Boston Have ND Don’t Have ND10 with ND 1990 without NDTrue Positives False Positives4199False Negatives True Negatives6 1791Jeff’s TestPositiveNegativeWhat’s the Most Important Clinical Question?If the test is positive, what is the chance that the patient really has diseaseorIf the test is negative, what is the chance that the patient does not have the disease?3Predictive Value The proportion of patients testing positive who actually have the disease (by gold standard) True Positives / All positivesPositive Predictive ValuePredictive Value The Proportion of patients testing negative who are truly free of the disease (by gold standard) True Negatives / All NegativesNegative Predictive ValueJeff’s Test Sensitivity: 80% Specificity: 90%Positive Predictive Value: 160 True Positives / 160 + 180 (all testing positive)Negative Predictive Value: 1620 True Negative/ 1620 + 20 (all testing negative)Given 2000 medical students Have ND Don’t Have NDPrevalence of 1 in 10 = 200 with ND 1800 without NDTrue Positives False Positives160 180False Negatives True Negatives20 1620Jeff’s TestPositiveNegative340 Tested Positive1640 Tested NegativeJeff’s Test Sensitivity: 80% Specificity: 90%The test is only right about a positive result 47 % of the timeThe test is right about a negative result 99% of the timeGiven 2000 medical students Have ND Don’t Have NDPrevalence of 1 in 10 = 200 with ND 1800 without NDTrue Positives False Positives160 180False Negatives True Negatives20 1620Jeff’s TestPositiveNegative340 Tested Positive1640 Tested NegativeJeff’s Test Sensitivity: 80% Specificity: 90%PPV: 4/203 =.02 or 2% NPV: 1791/1797 =.99 or 99%Given 2,000 persons in Boston Have ND Don’t Have ND10 with ND 1990 without NDTrue Positives False Positives4199False Negatives True Negatives6 1791Jeff’s TestPositiveNegativeA quick short cut: As prevalence increases: PPV increases and NPV decreases As prevalence decreases: PPV decreases and NPV increasesGiven a Test with a sensitivity of 80% and a specificity of 90%:Pretest Probability1% 10% 50% 90%Positive Predictive Value 7.5% 47.1% 88.9% 98.6%Negative Predictive Value 99.8% 97.6% 81.8% 33.3%4The point is . . .  Sensitivity and Specificity are functions of the operating curves of the test Predictive Values are related to prevalence or pre test probabilities Statistical difference doesn’t necessarily relate to clinical relevance Clinically Rule In/ Rule Out may not be accomplished by one testNow let’s assume that two tests are now available to screen for a new disease (Barrister’s Syndrome) and if detected in the asymptomatic phase can be cured with minimal therapy. The rate of Barrister’s Syndrome is 5% in the screened population.Test #1 Test #2Sensitivity= 80% Sensitivity= 85%Specificity= 70% Specificity= 50%How can one compare these tests for clinical utility?Assume 10,000 as a population sample and a prevalence of 5%: Test #1 Test #2Sensitivity= 80% Sensitivity= 85%Specificity= 70% Specificity= 50%Disease + Disease - Disease + Disease -Test + 400 2850 Test + 425 4750Test - 100 6650 Test - 75 4750500 9500 500 9500Predictive ValuesPPV: 400/3250 = 12.3% PPV: 425/5175 = 8.2%NPV: 6650/6750 = 98.5% NPV: 4750/4825 = 98.4%Which test is better? Likelihood ratio of a positive test is probability of a true positive (given disease) to false positives (without disease) Likelihood ratio of a negative test is probability of a false negative (with disease) to a true negative (without disease)Likelihood ratios compare probabilities of true results to false resultsMore simply . . .Likelihood ratio of a positive test isSensitivity100% - SpecificityThe larger the likelihood ratio the better the ability of the test to Rule In diseaseAnd . . . Likelihood