Ranges Used in the Interpretation of Test Results
In clinical practice, the laboratory test result is typically placed alongside a range of values for that test. In most cases, this is the reference range, which is often considered to be the normal range. It is important to understand that individuals with values inside the reference range may have subclinical disease, despite the presence of an apparently normal value. The reference range is dependent on the instrument and reagent used to perform the test. The reference ranges are ideally established inside the laboratory where the test is being performed. Reference ranges supplied by instrument and reagent manufacturers are not likely to correspond perfectly to ranges generated within an individual laboratory. This is because the population used to establish the range by the manufacturer and/or the instruments and reagents used by the manufacturer are likely to be different from those in an individual clinical laboratory.
To obtain a reference range, individuals without disease and on no medications donate samples for testing. A distribution of these values, which should be numerous enough to be statistically reliable, is plotted. The data are not always distributed in a Gaussian pattern. Therefore, statistical methods that are nonparametric are used to identify the central 95% of values. This range, representing the middle 95% of results, is the reference range. As an indication that being outside the reference range does not always reflect the disease, 5% of normal healthy, nonmedicated individuals who donated samples for the reference range determination now fall outside of what has become the reference range for the test.
To obtain a reference range, individuals without disease and on no medications donate samples for testing. The middle 95% of results is the reference range.
Several decades ago, the results for cholesterol testing demonstrated that individuals eating a high-fat diet showed high cholesterol levels that were associated with atherosclerotic vascular disease. When these apparently healthy, nonmedicated individuals provided samples for reference range determinations, the central 95% of values from this population provided an inappropriately high reference range. Therefore, the use of the classical reference range for selected laboratory tests in certain populations was not recommended. For that reason, desirable or prognosis-related ranges were developed. These are commonly established by groups of experts associating laboratory test results with clinical outcome.
For certain medications, a therapeutic window exists to provide a target for a blood, plasma, or serum level for the medication. Values below the therapeutic range typically reflect an inadequate amount of medication, and values above the therapeutic range may be associated with a particular toxic effect. In some cases, the therapeutic range does not reflect the amount of medication in the blood, but instead reflects a therapeutic effect produced by the drug. For example, patients taking the drug warfarin are not monitored with warfarin levels in the blood. Instead, the warfarin decreases the level of coagulation factors, which results in a prolonged prothrombin time (PT), and a calculated value known as the international normalized ratio (INR). The therapeutic range of warfarin, therefore, is determined by its effect rather than its concentration in the blood.
For certain laboratory tests, the presence of disease is associated with a value that is above a threshold.
Interpretations of Clinical Laboratory Test Results that Do Not Involve the Use of a Range
For certain laboratory tests, the presence of disease is associated with a value that is above a threshold. The use of troponin as a marker for myocardial infarction involves a threshold value, such that a level above the threshold is consistent with cardiac ischemia. Another prominent example is related to the detection of drugs of abuse. Any level above zero, as a threshold value, provides evidence for the ingestion of an illicit drug.
For laboratory tests that show too much variability to permit the use of a range or a threshold, an individual laboratory result for a specific patient can be compared with a result for that same patient that was generated previously. The longitudinal analysis of results over time can indicate the progression or regression of the disease.
The Need for a Diagnostic Cutoff
Figure 1–1 shows 2 populations of individuals and their results for a particular test. All of the individuals who do not have disease have a low value for the test, and all of the individuals with disease have a high value for the test. There is no overlap between groups in Figure 1–1. In Figure 1–2, a more commonly encountered situation is shown. There is overlap in laboratory values between individuals with disease and those without disease. This means that the diagnostic threshold will necessarily misclassify some patients to create false-positives, false-negatives, or both.
A clinical situation in which the diagnostic threshold completely separates those with disease from those without disease.
A clinical situation in which a diagnostic threshold is selected to maximize sensitivity.
The Definition of Sensitivity of a Laboratory Test
The population of individuals who have disease is the focus of sensitivity. The sensitivity of a laboratory test is its capacity to identify all individuals with disease. The threshold used in Figure 1–2 maximizes sensitivity by placing all those with disease above the line. This placement of the diagnostic threshold would decrease the number of false-negatives (those with disease who fall below the line), because everybody with the disease would have a positive test result. However, there is a significant misclassification of individuals without disease. As the diagnostic threshold is lowered, an increasing number of patients without disease would be told they have a positive test result, and by implication, the disease in question. The formula for sensitivity is:
True-positives and false-negatives are groups with disease; as noted above, sensitivity focuses on those with disease.
The sensitivity of a laboratory test is its capacity to identify all individuals with disease. Specificity is a statistical term that indicates the effectiveness of a test to correctly identify those without disease.
The Definition of Specificity of a Laboratory Test
The population of individuals without disease is the focus of specificity. Specificity is a statistical term that indicates the effectiveness of a test to correctly identify those without disease. When used to describe a laboratory test, it does not refer to its ability to diagnose a “specific” disease among a group of related disorders. One could maximize specificity by raising the threshold shown in Figure 1–3 to place all those without disease below the line. This would decrease the number of false-positives because everyone without disease would have a negative test result. However, there would be a significant misclassification of the individuals with disease. As the diagnostic threshold is raised, an increasing number of patients with disease would be told they have a negative test result and, by implication, no disease. The formula for specificity is:
A clinical situation in which a diagnostic threshold is selected to maximize specificity.
True-negatives and false-positives are the groups without disease; as noted above, specificity focuses on those without disease.
The Identification of the Appropriate Value for the Diagnostic Threshold
For diseases that are serious and treatable, and for which a second confirmatory laboratory test exists, it is important to maximize sensitivity as in Figure 1–2. For example, for diagnosis of AIDS, it is better to have a few false-positives that can be subsequently correctly identified with a confirmatory test than to fail to identify individuals with HIV infection who might unknowingly infect others. However, for diseases that are serious and not curable, a false-positive result is catastrophic for the patient. For such diseases, such as pancreatic cancer, it is better to use the threshold shown in Figure 1–3 for diagnosis because if individuals with disease are missed, it will have no effect on the treatment or outcome. When there are no compelling reasons to maximize either sensitivity or specificity, the threshold value should be established to minimize the total number of false-positives and false-negatives, as shown in Figure 1–4.
A clinical situation in which a diagnostic threshold is selected to minimize the number of false-positives and false-negatives.
The Definition of Predictive Value of a Positive Test
The population of individuals with a positive test result is the focus of positive predictive value. The positive predictive value for a laboratory test indicates the likelihood that a positive test result identifies someone with disease. It should be noted that the predictive value of a positive test is greatly influenced by the prevalence of the disease in the area where testing is performed. As an example, a screening test for HIV infection is more likely to be confirmed as positive in an area where many individuals are infected with HIV, as opposed to a location where there is only a rare case of HIV infection. In the latter situation, most of the positive HIV tests in the initial evaluation of a patient are found to be false-positives by confirmatory tests. A high percentage of false-positives from a low prevalence disease, as shown in the following formula, decreases the predictive value of the positive test:
The positive predictive value for a laboratory test indicates the likelihood that a positive test result identifies someone with disease. The negative predictive value for a laboratory test indicates the likelihood that a negative test result identifies someone without disease.
True-positives and false-positives are the groups with a positive test result; as noted above, positive predictive value focuses on those with a positive test.
The Definition of Predictive Value of a Negative Test
The population of individuals with a negative test result is the focus of the negative predictive value. The negative predictive value for a laboratory test indicates the likelihood that a negative test result identifies someone without disease. It is not greatly influenced by the prevalence of disease because false-positives are not included in the formula for negative predictive value. The formula for predictive value of a negative test result is:
True-negatives and false-negatives are the groups with a negative test result; as noted above, negative predictive value focuses on those with a negative test result.
The Difference Between Prevalence and Incidence
The prevalence of a disease reflects the number of existing cases in a population. It is usually expressed as a percentage of a certain population. Incidence refers to the number of new cases occurring within a period of time, usually 1 year. For example, in the United States, sore throat has a low prevalence because considering the size of the population there is a low percentage of individuals at a given time afflicted with sore throat. However, it has a high incidence because many new cases of sore throat appear each year.
Precision versus Accuracy
Precision refers to the ability to test 1 sample and repeatedly obtain results that are close to each other. This does not infer that the mean of these very similar numbers is the correct number (see Figure 1–5). Some analyses, which have great precision, are very inaccurate. The accuracy reflects the relationship between the number obtained and the true result. Thus, a sample could have high accuracy but low precision if it provides the correct answer but has substantial variability as the sample is repeatedly tested.
Precision refers to the ability to test 1 sample and repeatedly obtain results that are close to each other. This does not infer that the mean of these very similar numbers is the correct number.
A series of “bulls-eye” illustrations that display excellent or poor precision and accuracy.
Analyzing Errors in Laboratory Performance
There are 3 phases of laboratory analysis. The first of these is the preanalytical phase. This time frame is from patient preparation for the laboratory test, through the time of sample collection, until the sample arrives in the laboratory. Most of the errors in laboratory test performance occur in this phase. Examples of preanalytical errors are: inappropriate preparation of the patient, such as not fasting for a particular test in which fasting is required; ingesting drugs that will interfere with the laboratory tests; collection of the specimen in the wrong tube; delayed transport of the specimen to the laboratory; storage of the sample at an incorrect temperature; and collection of an inadequate amount of blood in vacuum tubes containing a fixed amount of anticoagulants. All these errors occur before the sample arrives for analysis and make it impossible, no matter how great the analytical precision within the laboratory, to provide a test result that truly reflects the patient's condition. The second phase is the analytical phase, which is the time that the sample is being analyzed in the laboratory. Errors can occur during this process, but they are much less common now because of the high level of automation of many laboratory instruments. Examples of analytical errors are: incorrect use of the instrumentation and the use of expired reagents. The third phase of laboratory test performance is the postanalytical phase, which begins when the result is generated and ends when the result is reported to the physician. Example of errors in this phase, which are more common than analytical errors but less common than preanalytical errors, are: delay in time to enter a completed result into the laboratory information system and reporting results for the wrong patient.
Minimizing Errors in the Selection of Laboratory Tests
As the number of laboratory tests has increased in size, complexity, and cost, health care providers are highly challenged to select the correct tests, and only the correct tests, in pursuit of a diagnosis. One approach commonly implemented to assist physicians in correct test selection is the use of a reflex test algorithm. Tests are ordered by algorithm selection, such as selection of an algorithm to determine the cause of a prolonged PTT result. Using the algorithm, the clinical laboratory notes the results of the first test in the algorithm, and that result determines which test is performed next. For example, if the prolonged PTT is further evaluated with a PTT mixing study, a normal result would direct testing toward assays for factors VIII, IX, XI, and XII. An elevated result in the PTT mixing study, on the other hand, would direct testing toward an inhibitor in the PTT reaction, such as a lupus anticoagulant. Testing is continued within the algorithm until a diagnosis, which explains the prolonged PTT in this case, is identified.
Laboratory test selection is also made more difficult because many laboratory tests have synonyms, and many compounds have related forms. For example, the test most commonly known as the lupus anticoagulant is also called the lupus inhibitor, and the general term that includes the lupus anticoagulant and related entities is antiphospholipid antibody. Vitamin D has several isoforms that include 25-hydroxyvitamin D and 1,25-dihydroxyvitamin D. Incorrect laboratory test selection is a major source of medical error.
Minimizing Errors in the Interpretation of Laboratory Test Results
With thousands of tests on the clinical laboratory test menu, it is impossible for a health care provider to understand the clinical significance of an abnormality for each test. This has become particularly noteworthy with the introduction of tests for genetic alterations, because there are so many and the clinical significance of the alterations may not yet be well established. In some institutions, narrative interpretations of complex clinical laboratory evaluations are prepared by experts in the field. In most institutions, such narratives require a special request for completion, but an emerging concept is to provide narrative interpretations for all complex clinical laboratory evaluations automatically, as they are provided in radiology and in anatomic pathology. Misinterpretation of laboratory test results has been increasingly noted as a source of poor patient outcome.