Accuracy and Precision of Laboratory Tests

Technical accuracy, or trueness, is an important consideration in interpreting the results of a laboratory test. Because of improvements in methods of analysis and elimination of analytic interference, the accuracy of most tests is limited primarily by their precision. Accuracy is a measure of the nearness of a test result to the actual value, whereas precision is a measure of the reproducibility of a result. No test can be more accurate than it is precise. Analysis of precision by repetitive measurements of a single sample gives rise to a gaussian distribution with a mean and an SD. The estimate of precision is the coefficient of variation (CV):

The CV is not likely to be constant over the full range of values obtained in clinical testing, but it is approximately 5% in the normal range. The CV is generally not reported, but is always known by the laboratory. It is particularly important in assessing the significance of changes in laboratory results. For example, a common situation is the need to assess hepatotoxicity incurred as a result of the administration of a therapeutic drug and reflected in the serum alanine aminotransferase (ALT) value. If serum ALT increases from 25 U/L to 40 U/L, is the change significant? The CV for ALT is 7%. Using the value obtained ±2 × CV to express the extremes of imprecision, it can be seen that a value of 25 U/L is unlikely to reflect an actual concentration of >29 U/L, and a value of 40 U/L is unlikely to reflect an actual concentration of <34 U/L. Therefore, the change in the value as obtained by testing is likely to reflect a real change in circulating ALT levels. Continued monitoring of serum ALT is indicated, even though both values for ALT are within normal limits. Likely in this case is only a probability. Inherent biologic variability is such that the results of 2 successive tests may suggest a trend that will disappear on further testing.

The precision of a test may also be indicated by providing confidence limits for a given result. Ordinarily, 95% confidence limits are used, indicating that it is 95% certain that the value obtained lies between the 2 limits reported. Confidence limits are calculated using the mean and SD of replicate determinations:

where t is a constant derived from the number of replications. In most cases, t = 2.

Accuracy is expressed by determining the difference, or bias, between results from a comparative method and a definitive or reference method. A definitive or reference method provides results with increased precision and accuracy compared to the clinical laboratory. When these methods are used, along with highly purified materials (i.e., Standard Reference Materials from the National Institute of Standards and Technology) to establish values for assay calibrators used in the clinical laboratory, the accuracy of patient results is improved. Creatinine, hemoglobin A1c, and neonatal bilirubin are examples in which the accuracy of these tests has been improved.

Sensitivity, Accuracy, and Analytic Testing

In some circumstances, the sensitivity and accuracy of an analysis are reduced or increased as functions of clinical purpose. For example, ion exchange chromatography of plasma amino acids for the diagnosis of inborn errors of metabolism is usually performed at an analytic sensitivity that allows measurement of all of the amino acids with a single set of standards. The range of values is approximately 20-800 µmol/L, and accuracy is poor at values of ≤20 µmol/L. The detection of homocysteine in this type of analysis suggests an inborn error of methionine metabolism. If the analysis is adjusted to achieve greater analytic sensitivity, it is possible to measure homocysteine accurately in normal plasma (3-12 µmol/L). This more sensitive test is used to assess cobalamin status and analyze risk factors for atherosclerotic cardiovascular disease.

Predictive Value of Laboratory Tests

Predictive value (PV) theory deals with the usefulness of tests as defined by their clinical sensitivity (ability to detect a disease) and specificity (ability to define the absence of a disease).

The problems addressed by PV theory are false-negative and false-positive test results. Both are major considerations in interpreting the results of screening tests in general and neonatal screening tests in particular.

Testing for HIV seroreactivity illustrates some of these considerations. If it is assumed that approximately 1,100,000 of 284,000,000 residents of the USA are infected with HIV (prevalence = 0.39%) and that 90% of those infected demonstrate antibodies to HIV, then we can consider the usefulness of a simple test with 99% sensitivity and 99.5% specificity. If the entire population of the USA were screened, it would be possible to identify most of those infected with HIV.

However, there will be 119,900 false-negative test results. Even with 99.5% specificity, the number of false-positive test results would be larger than the number of true-positive results:

In addition, there will be 281,480,000 true-negative results.

Given the high cost associated with follow-up and the anguish produced by a false-positive result, it is easy to see why universal screening for HIV seropositivity received a low priority immediately after the introduction of testing for HIV infection.

By contrast, we can consider the screening of 100,000 individuals from groups at increased risk for HIV in whom the overall prevalence of disease is 10%, with all other considerations being unchanged.

These 2 hypothetical testing strategies show that the diagnostic efficiency of testing depends heavily on the prevalence of the disease being tested for, even with a superior test, such as the test for HIV antibodies. Because the treatment of pregnant women infected with HIV is effective in preventing vertical transmission of the infection, screening has now been expanded to all pregnant women. The proven effectiveness of current therapy in preventing neonatal infection has intensified screening for HIV early in pregnancy.

However, because of the long time needed to test for HIV antibodies, it was difficult to screen women during labor and provide the necessary therapy. Recently, rapid HIV antibody testing procedures using a fingerstick or venipuncture to obtain whole blood, plasma, or serum, and tests using oral fluid were approved (Table 707-2). The HIV test results are usually obtained in <20 min. The collection of oral fluid samples provides an alternative for individuals who avoid HIV testing because of their dislike of needlesticks. HIV testing using whole blood or oral fluid is classified as a waived test under the Clinical Laboratory Improvement Amendments of 1988 (CLIA), and these tests are allowed in a point-of-care setting. Waived tests are simple laboratory procedures that use methodologies that are so simple and accurate as to render the likelihood of an erroneous result by the user negligible. A positive rapid HIV test result is then confirmed by Western blot analysis or immunofluorescence assay.

According to the U.S. Centers for Disease Control and Prevention, in the USA, between 280 and 370 infants were born with HIV in 2000. Rapid HIV testing during labor allows for implementation of antiretroviral therapy for HIV-infected women who have not been tested or are unaware of their HIV status. The initiation of therapy at the time of labor or within the 1st 12 hr of an infant’s birth significantly reduces the risk of mother-to-child transmission. In the mother-infant rapid intervention at delivery study, it was shown that the sensitivity and specificity of a rapid whole blood test for HIV during labor were 100% and 99.9%, respectively, with a positive PV of 90%. The median turnaround time for obtaining results from blood collection to patient notification was only 66 min. The performance of the rapid blood test was better than that of the standard HIV enzyme immunoassay, which had sensitivity and specificity of 100% and 99.8%, respectively, with a positive PV of 76%. In addition, the median turnaround time from blood collection to patient notification was 28 hr. As a result, rapid whole blood HIV testing is now the standard of care for women in labor with undocumented HIV status.

Rapid HIV testing can also be used in developing countries. In resource-poor settings, because of the lack of properly equipped laboratories, skilled technologists, and basic resources, such as electricity and water, these self-contained, point-of-care HIV tests are very attractive. In areas of Asia and Africa in which HIV is epidemic, screening pregnant women with rapid HIV tests and offering antiretroviral therapy can significantly reduce the transmission of HIV to hundreds of thousands of infants.

Neonatal Screening Tests

Almost all of the diseases detected in neonatal screening programs have a very low prevalence, and for the most part, the tests are quantitative rather than qualitative. In general, the strategy is to use the initial screening test to separate a highly suspect group of patients from normal infants (i.e., to increase the prevalence) and then to follow this suspect group aggressively. There are 2 common strategies used to detect congenital hypothyroidism, 1 uses thyroid-stimulating hormone for the initial screen and the other uses thyroxine. In the thyroxine strategy for congenital hypothyroidism, which has a prevalence of 25/100,000 liveborn infants, the initial test performed is for thyroxine in whole blood. Infants with the lowest 10% of test results are considered suspect. If all infants with hypothyroidism were included in the suspect group, the prevalence of disease in this group would be 250/100,000 infants. The original samples obtained from the suspect group are retested for thyroxine and are tested for thyroid-stimulating hormone. This 2nd round of testing results in an even more highly suspect group composed of 0.1% of the infants screened and having a prevalence of hypothyroidism of 25,000/100,000 subjects. This final group is aggressively pursued for further testing and treatment. Even with a 1,000-fold increase in prevalence, 75% of the aggressively tested population is euthyroid. The justifications advanced for the program are that treatment is easy and effective and that the alternative, if congenital hypothyroidism is undetected and untreated—long-term custodial care—is both unsatisfactory and expensive.

At its inception, newborn screening was driven by the selection of genetic diseases whose clinical manifestations developed postnatally, such as phenylketonuria, galactosemia, and hypothyroidism. Diseases selected for screening typically had to meet certain criteria. The prevalence of disease had to meet a minimum, typically 1 in 100,000. Disease selection required demonstrated reduction in morbidity and mortality in the neonatal period. Effective therapies needed to be available, and the cost of screening and the feasibility of laboratory testing were also considerations in this selection process.

More common diseases have also become targets for neonatal screening programs. Sickle cell disease, easily detected using liquid chromatography or isoelectric focusing, can be treated more effectively if it is diagnosed before clinical signs appear. In addition, the results of neonatal screening for cystic fibrosis (CF) show that there are clear benefits associated with preclinical diagnosis, but also that there are some inherent difficulties associated with genetic screening for complex autosomal recessive diseases that are common and are caused by a rather large number of mutations (>1,500) of a single gene. The definitive diagnostic test for CF is the measurement of concentrations of chloride in sweat, a test that is not practical during the 1st wk of life. Neonates with CF generally have elevations in whole blood trypsinogen. This test allows the identification of a group of neonates at risk for CF. Unfortunately trypsinogen as an initial screening test has a high false-positive rate, a unfavorable characteristic that creates unnecessary anxiety among newborn parents and families, and is costly due to the time and expense for medical follow up. Performing DNA analysis for common mutations that cause CF reduces the size of the suspect group and identifies neonates with a higher likelihood of disease. This 2-tiered strategy identifies a manageable number of infants on whom to perform sweat tests. Problems include the following: (1) uncommon mutations are not included in the screening panel (thus, cases of CF caused by these mutations can be missed); (2) common mutations that cause clinically innocent elevations of whole blood trypsinogen in heterozygous neonates cause potentially alarming false-positive findings; and (3) CF in patients with normal sweat test results is rare, but is likely to be missed.

Tandem mass spectrometry (MS/MS) is a technically advanced method in which many compounds are initially fragmented and separated by molecular weight. Each compound is then fragmented again. Identification of compounds is based upon characteristic fragments. The process requires roughly 2 min/sample and can detect 20 or more inborn errors of metabolism. The effects of prematurity, neonatal illness, and intensive neonatal management on metabolites in blood complicate the interpretation of results. The PV of a positive screening result is likely to be <10%; that is, 90% of positive results are not indicative of a genetic disorder of metabolism. Nonetheless, MS/MS permits a diagnosis to be made before clinical illness develops, and has revolutionized the purpose and ability of newborn screening. MS/MS is not directed toward diseases defined as treatable, but toward all of the diseases, each of which is rare, that the technique can identify.

Electrospray tandem mass spectrometry permits the detection of rare inborn errors of metabolism and has been introduced as a newborn screening tool all around the world. In the 4 yr since mass spectrometry was implemented in Australia, the rate of detection per 100,000 births was 15.7, significantly higher than the rate of 8.6-9.5 in the 6 preceding 4-yr periods. Disorders of fatty acid oxidation, particularly medium-chain acyl coenzyme A dehydrogenase deficiency, accounted for the majority of increased diagnoses.

Expanded newborn screening programs using MS/MS increase the detection of inherited metabolic disorders. As of 2009, 49 U.S. states use MS/MS in their neonatal screening programs. The remaining state has required expanded newborn screening, but has not yet implemented the testing. However, the metabolic conditions screened for by states using MS/MS vary, ranging from <10 to >29.

In an attempt to standardize newborn screening programs, the American College of Medical Genetics recommends that every baby born in the USA be screened for a core panel of 29 disorders (Table 707-3). An additional 25 conditions were recommended as secondary targets because they may be identified while screening for the core panel disorders. The March of Dimes and the American Academy of Pediatrics also endorse the recommendation by the American College of Medical Genetics. However, expansion of the screening test menu raises several issues. For example, the cost of implementation can be significant because many states will need multiple MS/MS systems. In addition, staffing the laboratory with qualified technical personnel to run the MS/MS system and qualified clinical scientists to interpret the profiles can be a challenge. A number of false-positive results will also be obtained with these newborn screening programs. Many of these findings are due to parenteral nutrition, biologic variation, or treatment, and are not the result of an inborn error of metabolism. Therefore, qualified staff will be needed to ensure that patients with abnormal results are contacted and receive follow-up testing and counseling, if needed. Even with these concerns, the American College of Medical Genetics report is a step in the right direction toward standardizing guidelines for state newborn screening programs.

Testing in Refining a Differential Diagnosis

The use of laboratory tests in refining a differential diagnosis satisfies PV theory because a correct differential diagnosis should result in a relatively high prevalence of the disease under consideration. An example of testing in refining a differential diagnosis is the measurement of urinary vanillylmandelic acid (VMA) for the diagnosis of neuroblastoma. A simple spot test for VMA is not useful in general screening programs because of the low prevalence of neuroblastoma (3 cases/100,000) and the low sensitivity of the test (69%). Even though the specificity of urinary VMA is 99.6%, testing of 100,000 children would produce 2 true-positive test results, 400 false-positive results, and 1 false-negative result. The PV of a positive result in this setting is 0.5%, and the PV of a negative result is 99.99%, not much different from the assumption that neuroblastoma is not present. Testing for urinary VMA in a 3 yr old child with an abdominal mass, however, gives a useful result because the prevalence of neuroblastoma is at least 50% in 3 yr old children with abdominal masses. If 100 such children are tested and the prevalence of neuroblastoma in the group is assumed to be 50%, then a satisfactory PV is obtained.

Thus, in this situation, a test with low sensitivity is powerful in refining the differential diagnosis because the PV of a positive result is almost 100% in the setting of high prevalence.

Acknowledgments

The author gratefully acknowledges the original contributions by Michael A. Pesce, on which portions of this chapter are based.