sensitivity versus predictive value: implications in low-risk populations

1
CORRESPONDENCE Division of Neonatology Children's Hospital of Philadelphia Philadelphia, Pennsylvania 19104 ]. Sanford Schwartz, M.D. Section of General Medicine Department of Medicine University of Pennsylvania Philadelphia, Pennsylvania 19104 REFERENCE 1. Greiner PF, Mayewski RJ, Mushlin AI, Greenlan P. Selec- tion and interpretation of diagnostic tests and procedures. Principles and applications. Ann Intern Med (Supp!.) 1981;94:553. To the Editors: In regard to the recent article by Johnson et al. I was confused by their data concerning the sensitivity and specificity of both the LIS ratio and A 650 amniotic fluid studies with respect to the respiratory distress syn- Disease absent False positive (FP) True negative (TN) Disease present True positive (TP) False negative (FN) Test + Test- Table I. Classification of test results specificity is not. Results reported from high-risk cen- ters will show, therefore, a higher positive predictive value for RDS since it occurs more frequently than in low-risk centers, where most of the positive tests will be in normal subjects with false positive results. The (PV -) test will be even higher since most patients are normal. Recalculation of the data shows that both the LiS ratio and A 650 are, indeed, equally sensitive in identify- ing fetuses destined to develop RDS, but at a lower level than reported. Although the PV - of both tests is 97%, results from other centers may differ since pre- dictive value is dependent upon the frequency of RDS in the population studied. Thus, when tests are stan- dardized and compared, sensitivity and specificity are more appropriate measures. Paul]. Weinbaum, M.D. Fellow in Maternal-Fetal Medicine Division of Maternal-Fetal Medicine Department of Obstetrics and Gynecology Hospital of the University of Pennsylvania Philadelphia, Pennsylvania 19104 Douglas Richardson, M.D., M.B.A. Fellow in Neonatology Sensitivity versus predictive value: Implications in low-risk populations To the Editors: We read with interest the report of Khouzami et aI., entitled "Amniotic fluid absorbance at 650 nm: Its re- lationship to the lecithin/sphingomyelin ratio and neo- natal pulmonary sufficiency" (AM.]. OBSTET. GYNECOL. 147:552, 1983) evaluating the relationship of the am- niotic fluid absorbance at 650 nm (A 650 ) to the lecithin/ sphingomyelin (LIS) ratio and fetal pulmonary matu- ration. We are soon to report our experience with the A 650 , LIS ratio, and other tests of fetal lung maturity, and agree that the simpler methodology and rapidity with which the A 650 can be performed certainly make it an attractive alternative to the fetal lung profile. However, we would like to point out an error in the statistical analysis of the data in this report. Table II in the report illustrates the relationship between LIS ratio, A 650 , and neonatal respiratory distress syndrome (RDS). The sensitivity and specificity of the LIS ratio are incorrectly described as 97.9% and 37.5%, respec- tively, whereas the A 650 is noted to be 97.5% sensitive and 15.8% specific. The terms sensitivity and specificity have apparently been confused in this report with positive and negative predictive value. All test results can be classified by means of Table 1. 1 Sensitivity (se) properly defined is the likelihood that a given test will be positive when a disease is present [se = TP/(TP + TN)], whereas spec- ificity (sp) is the probability that a test will be negative when the disease is absent [sp = TN/(FP + TN)]. On the other hand, positive predictive value (PV +) is the chance that the disease is present when the test is posi- tive [TP/(TP + FP)], and negative predictive value (PV -) is the probability that a disease is not present when a test is negative [TN/(TN + FN)]. Using these techniques, we have recalculated the au- thor's data. Three of four patients with RDS were de- tected with the use of an LIS ratio <2.0 (TP), whereas one displayed an LIS ratio >2:0 (FP). The sensitivity of the LiS ratio then is not 97.9% as stated, but 75%. The sensitivity of the A 650 is also 75%. The specificities of the LiS ratio and A 650 are 90% and 69%, respectively. It is the PV - of both the A 650 and LiS ratio which were the 97.9% and 97.3% figures reported. Another impor- tant point is the evident effect of small numbers. If a single infant with RDS is misclassified, the calculated sensitivity can vary from 50% to 100%. Thus, estimates of sensitivity must be based on a larger series. The distinction between predictive value and sensi- tivity/specificity of a test is very important. The predic- tive value of a diagnostic test is affected by the preva- lence of disease in the population, whereas sensitivity/ 470

Upload: jsanford

Post on 26-Jan-2017

213 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Sensitivity versus predictive value: Implications in low-risk populations

CORRESPONDENCE

Division of NeonatologyChildren's Hospital of PhiladelphiaPhiladelphia, Pennsylvania 19104

]. Sanford Schwartz, M.D.Section of General MedicineDepartment of MedicineUniversity of PennsylvaniaPhiladelphia, Pennsylvania 19104

REFERENCE

1. Greiner PF, Mayewski RJ, Mushlin AI, Greenlan P. Selec­tion and interpretation of diagnostic tests and procedures.Principles and applications. Ann Intern Med (Supp!.)1981;94:553.

To the Editors:In regard to the recent article by Johnson et al. I was

confused by their data concerning the sensitivity andspecificity of both the LIS ratio and A650 amniotic fluidstudies with respect to the respiratory distress syn-

Disease absent

False positive (FP)True negative (TN)

Disease present

True positive (TP)False negative (FN)

Test +Test-

Table I. Classification of test results

specificity is not. Results reported from high-risk cen­ters will show, therefore, a higher positive predictivevalue for RDS since it occurs more frequently than inlow-risk centers, where most of the positive tests will bein normal subjects with false positive results. The(PV -) test will be even higher since most patients arenormal.

Recalculation of the data shows that both the LiSratio and A650 are, indeed, equally sensitive in identify­ing fetuses destined to develop RDS, but at a lowerlevel than reported. Although the PV - of both tests is97%, results from other centers may differ since pre­dictive value is dependent upon the frequency of RDSin the population studied. Thus, when tests are stan­dardized and compared, sensitivity and specificity aremore appropriate measures.

Paul]. Weinbaum, M.D.Fellow in Maternal-Fetal Medicine

Division of Maternal-Fetal MedicineDepartment of Obstetrics and GynecologyHospital of the University of PennsylvaniaPhiladelphia, Pennsylvania 19104

Douglas Richardson, M.D., M.B.A.Fellow in Neonatology

Sensitivity versus predictive value: Implicationsin low-risk populations

To the Editors:We read with interest the report of Khouzami et aI.,

entitled "Amniotic fluid absorbance at 650 nm: Its re­lationship to the lecithin/sphingomyelin ratio and neo­natal pulmonary sufficiency" (AM.]. OBSTET. GYNECOL.147:552, 1983) evaluating the relationship of the am­niotic fluid absorbance at 650 nm (A650) to the lecithin/sphingomyelin (LIS) ratio and fetal pulmonary matu­ration. We are soon to report our experience with theA650, LIS ratio, and other tests of fetal lung maturity,and agree that the simpler methodology and rapiditywith which the A650 can be performed certainly make itan attractive alternative to the fetal lung profile.

However, we would like to point out an error in thestatistical analysis of the data in this report. Table II inthe report illustrates the relationship between LISratio, A650, and neonatal respiratory distress syndrome(RDS). The sensitivity and specificity of the LIS ratioare incorrectly described as 97.9% and 37.5%, respec­tively, whereas the A650 is noted to be 97.5% sensitiveand 15.8% specific.

The terms sensitivity and specificity have apparentlybeen confused in this report with positive and negativepredictive value. All test results can be classified bymeans of Table 1.1 Sensitivity (se) properly defined isthe likelihood that a given test will be positive when adisease is present [se = TP/(TP + TN)], whereas spec­ificity (sp) is the probability that a test will be negativewhen the disease is absent [sp = TN/(FP + TN)]. Onthe other hand, positive predictive value (PV +) is thechance that the disease is present when the test is posi­tive [TP/(TP + FP)], and negative predictive value(PV -) is the probability that a disease is not presentwhen a test is negative [TN/(TN + FN)].

Using these techniques, we have recalculated the au­thor's data. Three of four patients with RDS were de­tected with the use of an LIS ratio <2.0 (TP), whereasone displayed an LIS ratio >2:0 (FP). The sensitivity ofthe LiS ratio then is not 97.9% as stated, but 75%. Thesensitivity of the A650 is also 75%. The specificities ofthe LiS ratio and A650 are 90% and 69%, respectively. Itis the PV - of both the A650 and LiS ratio which werethe 97.9% and 97.3% figures reported. Another impor­tant point is the evident effect of small numbers. If asingle infant with RDS is misclassified, the calculatedsensitivity can vary from 50% to 100%. Thus, estimatesof sensitivity must be based on a larger series.

The distinction between predictive value and sensi­tivity/specificity of a test is very important. The predic­tive value of a diagnostic test is affected by the preva­lence of disease in the population, whereas sensitivity/

470