To the Editor:--Manley et al. found the true incidence of previously unrecognized pregnancy in menstruating women presenting for elective ambulatory surgery to be 7 of 2,056 (0.3%), as determined by urinary or serum human chorionic gonadotrophin immunoassay. Although this information is valuable, alone it is insufficient for the clinician to evaluate the significance of a positive or negative pregnancy test result in a single patient.
The probability of pregnancy in a patient with a positive test result (positive predictive value) equals the number of true positives divided by the total number of positives (true positives + false positives); the probability of nonpregnancy in a patient with a negative test result (negative predictive value) equals the number of true negatives divided by the total number of negatives (true negatives + false negatives). The clinical significance of the pregnancy test is therefore given by the positive predictive and the negative predictive values. These values can be calculated from intrinsic properties of the test (sensitivity and specificity) and the estimated prevalence of pregnancy in the population in question. [2,3].
Manley et al. quote one false-positive pregnancy test among 179 adolescent women (0.6%) during preoperative testing from data by Malviya et al. The specificity of the pregnancy test is therefore 178 of 179, 99.4%. Manley et al. also state that the possibility of a false-negative result (pregnancy despite a negative test result) is "extremely remote." Let us assume, therefore, that the sensitivity of the test is 100% (a sensitivity of 99% would give virtually the same result in the following discussion).
Consider a sample of 100,000 menstruating women presenting for elective ambulatory surgery. Given the original findings by Manley et al. of a 0.3% pregnancy rate, there would be 300 pregnant and 99,700 nonpregnant women in the sample. Pregnancy testing of the 300 pregnant women would result in 300 true positives and no false negatives (100% sensitivity); pregnancy testing of the 99,700 nonpregnant women would result in 99,102 true negatives and 598 false positives (99.4% specificity). The positive predictive value is 300/300 + 598 = 33%, and the negative predictive value is 99,102/99,102 + 0 = 100%.
Given that the aforementioned assumptions are valid, the clinician can be virtually 100% confident that a negative preoperative pregnancy test means that the patient is not pregnant. On the other hand, only 33% of women who test positive will be pregnant; the remaining 67% will have their surgery delayed but will not be pregnant. The latter statement is based on the specificity derived from data by Malviya et al. Even if this assumption is not valid, when a population has a low overall prevalence of pregnancy (0.3% in this example), it is essential for the pregnancy test to be highly specific (few false positives) if a low positive predictive value is to be avoided. Improving the specificity to 99.95% (a false-positive rate of 1 in 2,000) would only result in an 86% positive predictive value given the same prevalence of pregnancy (0.3%).
Although Manley et al. and Malviya et al. found what appear to be minor differences in specificity for preoperative pregnancy testing, the low prevalence of pregnancy in the population to be tested will result in large differences in positive predictive values. More data are required, preferably from large samples of women, before we can judge the clinical usefulness of preoperative pregnancy testing.
Ian Lewis, M.B., B.S., M.R.C.P., F.R.C.A., Consultant in Anesthesiology.
Jonathan Cooper, M.B., B.S., F.R.C.A., Senior Registrar in Anesthesiology, Shackleton Department of Anaesthetics, Level E, Center Block, Southampton University Hospital, Tremona Road, Southampton, SO 16 6YD, United Kingdom.