Screening and diagnostic testing

Editor's Notes

• #3: To understand how a disease is transmitted and develops and to provide appropriate and effective health care, it is necessary to distinguish between people in the population who have the disease and those who do not. This is an important challenge, both in the clinical arena, where patient care is the issue, and in the public health arena, where secondary prevention programs that involve early disease detection and intervention are being considered and where etiologic studies are being conducted to provide a basis for primary prevention. Thus, the quality of screening and diagnostic tests is a critical issue. Regardless of whether the test is a physical examination, a chest X-ray, an electrocardiogram, or a blood or urine assay, the same issue arises: How good is the test in separating populations of people with and without the disease in question? This chapter addresses the question of how we assess the quality of newly available screening and diagnostic tests to make reasonable decisions about their use and interpretation.
• #4: A large group centers on the value of 0 mm—no induration—and another group centers near 20 mm of induration. This type of distribution, in which there are two peaks, is called a bimodal curve. The bimodal distribution permits the separation of individuals who had no prior experience with tuberculosis (people with no induration, seen on the left) from those who had prior experience with tuberculosis (those with about 20 mm of induration, seen on the right). Although some individuals fall into the “gray zone” in the center, and may belong to either curve, most of the population can be easily distinguished using the two curves. Thus, when a characteristic has a bimodal distribution, it is relatively easy to separate most of the population into two groups (e.g., ill and not ill, having a certain condition or abnormality and not having that condition or abnormality).
• #5: In general, however, most human characteristics are not distributed bimodally. Figure shows the distribution of systolic blood pressures in a group of men. In this figure there is no bimodal curve; what we see is a unimodal curve—a single peak. Therefore, if we want to separate those in the group who are hypertensive from those who are not hypertensive, a cutoff level of blood pressure must be set above which people are designated hypertensive and below which they are designated normotensive. No obvious level of blood pressure distinguishes normotensive from hypertensive individuals. Although we could choose a cutoff for hypertension based on statistical considerations, we would ideally like to choose a cutoff on the basis of biologic information; that is, we would want to know that a pressure above the chosen cutoff level is associated with increased risk of subsequent disease, such as stroke, myocardial infarction, or subsequent mortality. Unfortunately, for many human characteristics, we do not have such information to serve as a guide in setting this level.  In either distribution—unimodal or bimodal—it is relatively easy to distinguish between the extreme values of abnormal and normal. With either type of curve, however, uncertainty remains about cases that fall into the gray zone.
• #6: According to National cholesterol education program S Ch > 240 mg/dl is an indication for drug tt, 200 – 239 is boderline where diet and lifestyle corrections are considered
• #8: To minimize individual variability – repeat measure and take average, measure 24 hr, for measurement variability – standardize instrument-caliberation, repeat measure and take average, technique (fasting or not); same laboratory type of analyser
• #14: Berkson’s bias – hospital patients usually has more than one disease therefore false associations can be registered
• #15: Unacceptability Bias e.g. – in a case control study regardless of disease status participants may under report eating high fat diet thinking its not good making it difficult for the researchers to identify an association- bias towards the null hypothesis
• #17: There are two accepted methods for dealing with potential confounder Consider them in design by matching on potential confounder or by restricting the sample to limited levels of potential confounders Evaluate confounder in analysis by stratification or by using multivariate analysis (multiple logistic regression)
• #20: Clinical decision making is weighing of probabilities… The purpose of a diagnostic test is to move the estimated probability of the presence of a disease toward either end of the probability scale based on new meaningful information that will alter subsequent treatment / diagnostic plans If the patient’s sister or mother had been previously diagnosed with CA Br patient’s likelihood of having breast cancer prior to any test could have been as high as 1%, if palpable lump would have been there probability would raise to 20-40% Different mammographic and FNA characteristics eg appearance of the nucleus or nuclear/ cytoplasmic ratio change the the estimate of br ca probability …raised/ lower. Further more different pathologists may have different opinion – definite cancer cell/ suspicious
• #23: Sensitivity and specificity are descriptors of the accuracy of a test. Sensitivity – defined as the percentage of persons with the disease of interest having positive test results. Tests with great sensitivity are useful clinically to rule out the presence of a disease. Specificity – defined as the percentage of the persons without the disease of interest having negative test results Positive Predictive Value (PV+) :– defined as percentage of the persons with the positive test result actually having the disease Negative Predictive Value (PV –) :– defined as percentage of the persons with the positive test result actually having the disease
• #24: Before the FNA was performed, the average likelihood of not having breast cancer among the sample women was 87% (99 out of 114). After a negative FNA test the probability of not having cancer is raised to 99% - (91/92=99%) Usefulness of the FNA test For a patient without a palpable lump is considered: - a +ve result increased the probability of Br. Ca from 13% to 64% (but further workup is still required since we cannot take chance of missing a case) a negative test result, however, would reduce the probability that breast cancer is present to 1% (100-99=1%) - now decision could be made to defer surgical biopsy and repeat mamographic and physical examination in several months for women with abnormal mamogram but normal FNA accepting a 1 in 100 risk of mistakenly delaying treatment of an existing cancer. An aggressive approach could be to perform FNA in all cases considering 1% of the total population is a lot of women Although the FNA test has identical sensitivity and specificity in pts with and without lump but PV+ increased from 64% in women without lump to 88% in women with the palpable lump (high prevalence)– therefore its easier to confirm the presence of breast cancer Av. Likelihood of not having Ca Br. = 99/114 =87% Likelihood of having Ca Br. after a -ve FNA test =(100-99= 1%)
• #33: Moving the cut off point changes the sensitivity, specificity and +ve and –ve predictive values and hence the way the test is used! With a cutoff point set btw the categories of benign and suspicious, a –ve FNA test would reduce the probability of Br. Ca by 96% but with 4% chance of Br. Ca. biopsy may still be warranted. A +ve FNA result indicate 88% likelihood of having CA Br. But still would not confirm the diagnosis absolutely. Alternatively, by setting the cutoff point btw suspicious and malignant (a more stringent requirement to consider test +ve) the PV + = 100%. This could be useful clinically as now women with +ve FNA would require no further testing prior to definitive treatment.
• #34: The smallest possible value of LR+ve occurs when numerator is minimized (sensitivity =0) and maximum when denominator is minimized i.e. specificity is 100% (so 1-1=0) resulting in LR of positive infinite. An LR+ve of 1 indicates a test with no value in sorting out persons with and without the disease of interest, since the probability of a +ve test result is equally likely for affected and unaffected persons The larger the value of LR+VE the stronger the association btw having a positive test result and having the disease.
• #35: The sizes of the two likelihood ratios indicate the strength of association btw a test result and likelihood of the disease A diagnostic test with a large LR+ value increases the suspicion of disease for patients with positive results – larger the size better is the diagnostic value of the test…arbitrarily a value of 10 is perceived as an indication of a test of high value for LR+ and  0.08 for LR-
• #48: The ROC plot of a given test is obtained by calculating the sensitivity and specificity of every observed value, and then plotting sensitivity (on the Y axis) against 1 - specificity (on the X axis). A test that does not discriminate between normal and abnormal would give a diagonal straight line from the bottom left corner to the top right corner. All points on such a line represent a 1:1 ratio of true to false positives. An ideal test would give a rectangular plot passing from the origin at the bottom left hand corner towards top left hand corner at first and thence to the top right hand corner. In reality the ROC curves of many of the tests in common use fall in between these extremes. The cut-off point for deciding between normal and abnormal is selected arbitrarily where the ROC curve changes direction from being vertical to horizontal. The more the ROC curve arches into the upper left hand corner away from the diagonal, the better the test.
• #50: Breast cancer is an important public health problem with sufficiently high mortality and morbidity. Early detection allows less extensive surgical treatment and reduces mortality and morbidity. Since the incidence increases steadily with advancing age a high risk group can be constructed - > 50 yrs or so; recommending screening above 50 routinely.
• #51: The overall odds ratio is then calculated by pooling the data of all the studies. This is also called "typical" odds ratio. It is calculated by the difference between number of deaths in the treatment group (i.e. the number observed) and the number of deaths in this group if the treatment were ineffective (i.e. number expected). This gives the Observed minus the Expected statistic. The confidence interval of O  E is also calculated The outcome in the case of each study can be estimated separately by calculating the O  E value. If the observed number (O) differs systematically from the expected number (E), there is clear evidence of effect. The totaled O  E gives a measure of the overall statistical significance and the effect size