NOTES NOTES TESTING SENSITIVITY (SN) & SPECIFICITY (SF) osms.it/sensitivity-specificity ▪ Validity measure; concerned with how close test’s result is to truth (i.e. did test/ instrument measure what it is intended to measure?) ▫ No perfect test → some miscalculation degree inevitable (i.e. healthy individual tests positive for disease → false positive; sick individual tests negative → false negative) ▫ Sn, Sp: complementary test characteristic measures must be used together SENSITIVITY ▪ Population proportion who test positive for disease, have disease ▪ AKA true positive rate ▪ Highly sensitive test with positive result identiﬁes people who are truly diseased (true positives), some healthy people (false positives) ▪ Sensitivity: proportion containing all truly positive, false positives ▪ Can assume two things ▫ Test with high sensitivity is negative, individual must be healthy → rule out disease ▫ Test with high sensitivity is positive, individual may/may not have disease (ensure lack of false positive; further testing required) ▫ High sensitivity negative test → useful for ruling-out disease SPECIFICITY ▪ Population proportion tests negative for disease, free of disease ▫ AKA true negative rate ▪ Highly speciﬁc test with negative result ▪ Identiﬁes all people who are truly free of disease (true negatives), some sick people (false negatives) ▪ Speciﬁcity: proportion containing all truly negative, false negatives; two things assumed ▫ Test with high speciﬁcity positive → conﬁrm disease ▫ Test with high speciﬁcity negative → individual may/may not have disease (ensure not false negative; further testing required) ▪ Positive test with high speciﬁcity → useful disease conﬁrmation CUTOFF POINT ▪ For continuous variables: sensitivity, speciﬁcity may overlap → midpoint usually sought (avoids misclassiﬁcation) ▪ Cutoff point needed to distinguish between normal/healthy, abnormal/unhealthy results High cutoff point ▪ Highly speciﬁc: low false positives ▫ Everyone categorized as abnormal has disease ▪ Poorly sensible: high false negatives ▫ Not everyone categorized as normal is free of disease OSMOSIS.ORG 79
▪ I.e. previous hypertension deﬁnition stated 140/90mmHg as cutoff point ▫ Highly speciﬁc: everyone categorized as abnormal has disease ▫ Poorly sensitive: not everyone categorized as normal is free of disease Low cutoff point ▪ Poorly speciﬁc: high false positives ▫ Not everyone categorized as abnormal has disease ▪ Highly sensitive: low false negatives ▫ Everyone categorized as normal is free of disease ▪ I.e. new hypertension deﬁnition states 120/80mmHg as cutoff point ▫ Poorly speciﬁc: not everyone categorized as abnormal has diseases ▫ Highly sensitive: everyone categorized as normal is free of disease Cutoff point determined by test’s purpose ▪ Screening test ▫ Needs to detect all possible diseased → low cutoff point → highly sensitive → low false negatives ▪ Conﬁrmatory test ▫ Need to be sure of disease presence → high cutoff point → highly speciﬁc → low false positives ▫ First test sensitivity x second test sensitivity ▪ Net speciﬁcity: proportion of healthy people that test negative on either ﬁrst, second test ▫ (First test speciﬁcity + second test speciﬁcity) - (ﬁrst test speciﬁcity * second test speciﬁcity) Simultaneous testing ▪ Two tests with different characteristics performed at same time → more sensitive results ▫ Simultaneous testing: three groups of people ▫ People detected only by Test A ▫ People detected only by Test B ▫ People detected by both Test A and Test B ▫ Pools all possibly relevant information → more sensitive results ▪ Sensitivity, speciﬁcity calculations must include both tests’ characteristics ▪ Net sensitivity: proportion of true cases that test positive on either test A or B ▫ (Test A sensitivity + Test B sensitivity) (Test A sensitivity x Test B sensitivity) ▪ Net speciﬁcity: proportion of healthy people that test negative on both tests A and B ▫ Test A speciﬁcity x Test B speciﬁcity SEQUENTIAL & SIMULTANEOUS TESTING Sequential testing ▪ AKA two-stage testing ▪ Consecutive tests performed with different characteristics → obtain more speciﬁc results ▫ Perform ﬁrst test → positive → perform second test → positive → disease likely present ▫ Perform ﬁrst test → negative → disease not likely present ▪ Similar to “double checking” results ▪ First test often easier/cheaper/less invasive than second test ▪ Sensitivity, speciﬁcity calculations must include both tests’ characteristics ▪ Net sensitivity: proportion of true cases that test positive on both ﬁrst, second test 80 OSMOSIS.ORG Figure 13.1 Illustration showing how sensitivity and speciﬁcity are affected by moving the cut-off point.
Chapter 13 Biostatistics & Epidemiology: Testing POSITIVE & NEGATIVE PREDICTIVE VALUE osms.it/positive-negative-predictive-value ▪ PPV: probability that if test is positive, person has disease ▫ Divide true positives, total positive test number ▪ NPV: probability that if test is negative, person is free of disease ▫ Divide true negatives, total negative test number ▪ Both measures directly inﬂuenced by prevalence, test speciﬁcity ▫ High prevalence: more likely that person has disease → ↑ PPV ▫ Low prevalence: less likely that person has disease → ↑ NPV ▫ Low prevalence: need a good test in conﬁrming disease (high speciﬁcity) → ↑ PPV OSMOSIS.ORG 81
TEST PRECISION & ACCURACY osms.it/test-precision-accuracy ▪ Both concerned with how likely test to be reproduced → return results close to truth ▫ Neither measuring devices nor people perfect → affects test precision, accuracy ▪ Test precision: how repeatable test results are over time, regardless of result accuracy ▫ High precision test: consistently deliver similar results, regardless of whether true/not ▪ Test accuracy: how true test results are, regardless of test repeatability ▫ High accuracy test: gives correct results; cannot always be reproduced 82 OSMOSIS.ORG ▪ Comparing test precision, accuracy ▫ Oximeter consistently (precisely) reports true pO2 (accurately) ▫ Oximeter consistently (precisely) reports pO2 20% lower than truth (not accurate) ▫ Oximeter inconsistently (not precise) reports true pO2 (accurate) ▫ Oximeter inconsistently (not precise) reports pO2 20% lower than truth (not accurate)