|Year : 2015 | Volume
| Issue : 2 | Page : 88-90
Basics of diagnostic evaluation part 2
Areej Abdul Ghani Al Fattani
Department of Pediatrics, King Faisal Specialist Hospital and Research Centre, Riyadh 11211, Saudi Arabia
|Date of Web Publication||7-Jul-2015|
Areej Abdul Ghani Al Fattani
Department of Pediatrics, King Faisal Specialist Hospital and Research Centre, MBC 58, PO Box 3354, Riyadh 11211
Source of Support: None, Conflict of Interest: None
Despite a long tradition of reporting diagnostic accuracy in terms of sensitivity and specificity, only a minority of physicians correctly interprets and applies this information. Likelihood ratios (LRs) are more intelligible and more meaningful because it connects the preprobability of a person to postprobability that guide the physician to make the right decision about further procedures. Using Fagan Nomogram, makes the application of the LRs more customized to each case. In case of testing a continuous diagnostic test, receiver operating curve is used to set a cut-off value that will be the line between disease and nondisease persons. When determining the cut-off, the trade-offs between increasing the sensitivity with lowering specificity and vise-versa depends on the nature of the disease, as well as whether this test will be used as screening or as diagnostic.
Keywords: Diagnosis, likelihood ratio, preprobability, receiver operating curve
|How to cite this article:|
Al Fattani AA. Basics of diagnostic evaluation part 2. J Appl Hematol 2015;6:88-90
| Likelihood ratio|| |
Sensitivity and specificity used to assess the accuracy of a given test, compared to a gold standard. Whenever sensitivity rises, the specificity goes down. Deciding on which side to go depends on the purpose behind using that test. High sensitivity test is useful for ruling out a disease (i.e., for screening of HIV, your concern about ruling out because you don't want to tell a person that he is infected with HIV while he isn't). On the other hand, high specificity test is useful for confirming the presence of the disease (i.e., for a confirmatory test of HIV, you need a very specific test, which does not miss any infected cases).
In daily clinical practice people commonly focus on sensitivity and specificity of the test, remember sensitivity give you the probability of having positive result giving a person with the disease. However in ER, for example, you do not know whether the person you face has the disease or not. In fact, the physician wants to know if a patient got a positive result, what is his probability for actually having the disease. That's what exactly the positive predictive value (PPV) tells us. The physician also might want to know if the person got a negative result, what is his probability to be not diseased. That's what exactly the negative predictive value (NPV) tells us. One limitation we should be aware of, that PPV and NPV are affected by the prevalence of the disease and the prevalence in our clinic might not be similar to the previously published one. 
Another way to describe the performance of a test is the likelihood ratios (LRs). A positive LR (LR+) tells us, how many times a positive test will be found in persons with the disease compared to persons without the disease. Negative LR (LR−) tells us, how many times a negative test will be found in persons with the disease compared to those without the disease. Both LRs give a summarization to the values of sensitivity and specificity but are more stable when the prevalence changes. This feature makes LR more relevant measure for the use of health care professionals. That will be illustrated using our previous example, comparing the diagnostic value of the sickling test in clinically suspected patients (prevalence 40%) versus unsuspected patients (prevalence 10%). ,
In the example shown in [Table 1] and [Table 2], the sensitivity and specificity for the sickling test were 65% and 95%, respectively. Although PPV and NPV have changed between samples, LR+ and LR− were constant 13 and 0.4%, respectively. A diagnostic test with a higher LR+ (>10) and lower LR− (<0.1) generates large change from pretest probability to postprobability. In our example, we conclude that sickling test will be useful in case of positive result, but not in case of negative result. Fortunately, an intelligent tool is available to put that theory in a practice called Fagan Nomogram.
We want to evaluate the benefit of the sickling test in the diagnosis of sickle cell anemia (SCA) with three patients with different levels of risk and different pretest probabilities. Using the LRs and Fagan Nomogram, we will get the answer. [Figure 1], shows the Fagan Nomogram of three patients who have different preprobabilities. In Saudi Arabia, the crude prevalence of sickle-cell trait ranges from 2% to 27% according the regions.  Patient 1 comes to your clinic from the central region and has low expected possibility (2%) to have SCA. By plotting values of LR+ and LR− (13 and 0.4), we find that if the test was positive he is likely to have the disease by 25%, which is not informative as a screening. If the test was negative, his postprobability would be around 0.5%, which is not that far from the pretest probability. We note here, this test is not that helpful for patient 1 and maybe we need to obtain more history or consider other diagnosis.
Patient 2 comes from the eastern region and his preprobability for having the disease around 10%. If he test positive he is likely to have SCA by 60%, while a negative result will decrease the probability for not having the sickle cell trait to 2%. Again, this would still need more information to raise or to lower the preprobability in order to have a significant difference. Patient 3 comes from the eastern region, and his parents were both traits, so his preprobability will be raised to around 25%. In case of positive result, the postprobability will be more than 82%. While if he test negative the postprobability will be around 6%. This patient benefits more from that test, because there is a notable difference between his pre- and post-probability. 
We conclude from that, LRs are very informative in the daily practice of physicians, but the patient characteristics and medical history enhance its meaning and importance in diagnosis.
| Receiver operating characteristic curve|| |
Sometimes the measurement used for the diagnostic test is on the continuous scale such as most important biochemical tests and biomarkers for diagnosis of cancers. In this case, there are many options about where to set the cut-off point between positive and negative test result. (Receiver operating curve) ROC uses sensitivity in the horizontal axis and 1-specificity-or false-positive rate-on the vertical axis to construct a graph that shows the possible cut-off values like in [Figure 2]. The diagonal line reflects the reference of the test with no diagnostic value where LR+ always equals 1, means the positive result is equally found in persons with and without the disease. ROC is used to define the best cut-off value for a continuous diagnostic test that could distinguish diseased from nondiseased persons. By lowering the cut-off level, we increase the sensitivity but decrease the specificity. While by raising the cut-off level, we decrease the sensitivity but increase the specificity.
The choice of which cut-off to use, depends on the nature of the disease and the purpose of the test. The perfect test (or the most optimal) is the closest to the upper-left corner. For a diagnostic conformation test, you want to be more concerned about increasing the specificity or decreasing the false positive rate. The cut-off point for such test would be on the left hand of the curve. On the other hand, for a screening test, where you will be concerned about the sensitivity, the cut-off point would be on the right side of the curve. ROC can be used-in addition to determine the best cut-off for a scale diagnostic test to assess accuracy quantitatively or to compare accuracy between more than one predictive models. 
The area under the curve (AUC) used to interpret the summary of the accuracy. When the AUC gets bigger, the performance of the test gets better. The value of AUC ranges from 1 to 0.5, where 1 indicates perfect diagnostic value and 0.5 indicates that test does not distinguish between diseased and nondiseased persons. Statistical software's report the ROC with a table shows all the possibilities of the cut-off and the corresponding sensitivity and 1-specificity values. It also reports the ROC with a table with its AUC, its significance level (P value) and the confidence interval. 
Financial Support and Sponsorship
Conflicts of Interest
There are no conflicts of interest.
| References|| |
Weinstein S, Obuchowski NA, Lieber ML. Clinical evaluation of diagnostic tests. AJR Am J Roentgenol 2005;184:14-9.
Knottnerus JA, van Weel C, Muris JW. Evaluation of diagnostic procedures. BMJ 2002;324:477-80.
Jastaniah W. Epidemiology of sickle cell disease in Saudi Arabia. Ann Saudi Med 2011;31:289-93.
Hajeer AH, AL Knawy BA. Doctor′s Guide to Evidence-Based Medicine. Riyadh: Alasr Printing Company; 2006.
Cook NR. Statistical evaluation of prognostic versus diagnostic models: Beyond the ROC curve. Clin Chem 2008;54:17-23.
Greenberg RS, Daniels SR, Flanders WD, Eley JW, Boring JR. Medical Epidemiology. 4 th
ed. United States of America: McGraw-Hill Companies, Inc.; 2005.
[Figure 1], [Figure 2]
[Table 1], [Table 2]