In medicine, diagnosis of a disease is rarely definitive. Clinicians often use the terms ‘most likely’ or ‘not likely’ to rule in or rule out the disease after clinical evaluation. Diagnostic tests are often needed to help clinician to make the diagnosis. This article will discuss the methods which can be used for evaluating the accuracy of the diagnostic tests and for interpreting the results of a positive or negative test result in a particular patient. Some of the most difficult situations in the neonatal medicine relate to evaluation and management of suspect neonatal sepsis. Therefore most of the examples taken in this article to explain the principles of evidence based medicine will be from the field of neonatal sepsis. For example, whether to order a panel of diagnostic tests for making the diagnosis of neonatal sepsis, what benefit it gives if we add another sophisticated marker like IL-6 to the usual sepsis screen panel and whether to do a lumbar puncture in a sepsis screen positive but clinically asymptomatic newborn.Every additional diagnostic test causes an additional cost and sometimes an additional risk to the patient. Causing cost and risk to the patient can be justified only if it changes the management or provides clinically important information about the prognosis. Whether test result changes therapy depends on: <list list-type="bullet"><list-item>The action threshold i.e. the probability of the disease at which the clinician starts treatment. Action threshold of a disease is inversely proportional to the perceived risk of untreated diseases.</list-item><list-item>Prior probability which depends upon disease frequency and clinical evaluation</list-item><list-item>Diagnostic test performance</list-item></list></sec><sec id="sec002"><title>Prior probability and action threshold

Clinicians often use the term ‘index of suspicion’ for the probability of a given disease prior to performing the diagnostic tests. One must know or make a calculated guess about the prevalence of a particular disease in the targeted population and then use the information derived from history and clinical examination to upgrade or downgrade the risk to reach the value of ‘prior probability’. ‘Test:No test threshold’ is the probability level below which probability of a disease is so low that there is no use of performing a test. ‘Test:Treat threshold’ is the level above which prior probability of presence of the disease is so high that treatment is warranted even without performing the test. But most of the times prior probability of the disease lies between ‘Test:No test’ and ‘Test:Treat’ threshold. In such a scenario applying a good diagnostic test will help by either increasing the posterior probability to above level of action threshold or will rule out the disease.

Test performance

Test performance is measured by comparing the test against a gold standard for the diagnosis and then measuring following variables using a 2X2 table.

Sensitivity: Sensitivity is the proportion of diseased patients who have a positive test. So it’s the ability to correctly diagnose the disease i.e. true positive of all diseased (true positive and false negative).

Specificity: Specificity is the proportion of nondiseased patients who have a negative test. Therefore it’s ability to correctly identify the absence of disease i.e. true negative of all nondiseased (true negative and false positive).

Positive Predictive Value (PPV): PPV is the proportion of test positive patients who have the disease i.e. true positive of all positives.

Negative Predictive Value (NPV): NPV is the proportion of test negative patients who don’t have the disease i.e. true negative of all negatives.

Likelihood Ratio (LR): LR is the probability of a particular test result among those with the diseased compared with the probability of the same test result among those without the disease. Therefore

Choosing among tests with different properties or choosing the cutoff value for a single test depends on the purpose of getting the test result and may seem counterintuitive. When the objective is to rule out a disease, the test with the fewest falsenegative results should be chosen. Because sensitivity is the proportion of people with disease who have a positive test result, a test with high sensitivity has the fewest false-negatives. Because nearly all diseased patients have a positive test result, few diseased patients have a negative test result (few false-negatives). On the other hand, when the objective is to confirm a diagnosis, the physician should choose the test with the fewest false positives; that is, a test with high specificity. A simple way of remembering the effect of different properties of tests is to use the mnemonics “SpPIn” and “SnNOut.” If a highly Specific test is used, a Positive result rules In the diagnosis; hence, the mnemonic SpPIn. On the other hand, if a highly Sensitive test is used, a Negative result rules Out the diagnosis: SnNOut.

Clinical suspicion or presence of risk factor(s) is often the starting point of evaluation and management of neonatal sepsis. But the clinical signs in a newborn are often non-specific and unreliable. In addition blood culture the gold standard for diagnosis of neonatal sepsis is positive in only 20-60% (Indian data) of clinically suspected sepsis. Here comes the role of performing diagnostic tests. An ideal diagnostic marker for evaluation of neonatal sepsis should have an optimal comparable cutoff, sensitivity approaching 100%, specificity of >85%, PPV of >85%, NPV approaching 100%, should detect infection at an early stage, differentiate b/w viral vs. bacterial, guides antibiotic use, monitors progress of treatment and help in prognostication. Singh M et al (1) studied the performance of various diagnostic markers in evaluation of sepsis. Used individually these markers had unsatisfactorily low sensitivity but reasonable specificity (Table 1). Table 1

Sensitivity and specificity of each individual tests

Test	Sensitivity (%)	Specificity (%)

Micro ESR	55	81
GA for polymorphs	38	81
ANC	65.5	48
Band/Neutrophil count	62	81
CRP	80	91

This means that if we use a marker like micro ESR alone for diagnosis of infection we will be missing 45% cases, which is unacceptable. Similarly Chandana et al (2) found positive predictive value of CRP as low as 57% which means that if we treat all CRP positive cases as sepsis we will be unnecessarily treating almost equal number of uninfected babies, which also is undesirable.

Many combinations of diagnostic tests have been tried to increase the diagnostic yield (Table 2). Singh M et al (1) found that combining any two or more of the tests mentioned in Table 1 increases the sensitivity to 86%. Table 2

Sepsis screen panels

Study	Year	Criteria	Sn	Sp	PPV	NPV	PlR	NLR

Koenig³	1988	TLC>15000&	60	86	32	95	4.15	0.47
		I/T > 0.2
Krediet⁴	1992	CRP&	60	74	71	63	2.3	0.54
		I/T
Hiew⁵	1992	CRP & ANC	47	64	36	74	1.31	0.83
Hiew⁵	1992	CRP & TLC	27	89	50	74	2.33	0.83
Hiew⁵	1992	TLC & ANC	23	83	37	72	1.36	0.93
DaSilva⁶	1994	ICC/TLC/I:T	61	76	37	89	2.49	0.52
Philip⁷	1994	I:T+CRP	22	98	54	92	10.17	0.80

A look at Table 2 shows that PLR of a positive panel ranges from 1.31-10.17 but in most of the studies NLR is near 0.8. This means that probability of a negative test result among those with the disease (sepsis) is almost same as compared with the probability of negative test result among those without the disease (no sepsis).

Odds and Probabilities

Now let’s look at the use of the evidence based medicine while performing diagnostic test in the perspective of an individual patient. Suppose incidence of neonatal sepsis in term babies is 0.1%. A 3 day old newborn presents to a clinician with fever. From his experience and literature knowledge the clinician feels that probability of the newborn having sepsis is 30% (baseline incidence revised according to clinical presentation) which is a pretest probability of 0.3. Of course the clinician is in dilemma and needs a diagnostic test which will either decrease or increase this probability substantially. What if he orders the sepsis screen panel studied by Philips et al (7) and gets a positive result?

This means that a positive test has significantly increased the possibility of sepsis from 30% to 81%. Suppose the clinician gets the negative result from same sepsis screen panel. NLR of the positive test from Table 2 is 0.8. From calculations as shown above the posttest probability with the negative test result will be 0.25 (25% from 30%). This means a negative result will still leave the clinician stranded. One can avoid using these lengthy calculations by using the probability nomogram (Figure 3) designed by Fagan (8). Figure 1Threshold Model

Figure 2Test performance variables

Figure 3Probability nomogram (Adapted from Fagan TJ) (8)

It has been suggested that a likelihood ratio between 0.1 and 10 is of limited use for predicting the presence or absence of disease, since it will not substantially alter the pretest probability. Apart from a few exceptions, the likelihood ratios calculated from the studies included Table 2 lie within this indeterminate range and so they appear to be of limited value, either as individual tests or in combination.

Receiver Operating Characteristic Curve

When test result is expected as a continuous variable there can be varying sensitivity or specificity rates for differing values of the test result. Lower values tend to be more sensitive while higher values tend to be more specific. The choice of cut-off depends upon the relative merits of sensitivity and specificity for the diagnosis of disease in question. In order to evaluate the performance of test result over a range of possible cut-off values and to compare different diagnostic tests employing different continuous variables, the Receiver Operating Characteristic (ROC) curve is a very useful mode of analysis. These plots were developed in the 1950s for evaluating radar signal detection. Assume a radar system, which is highly sensitive for picking enemy fighter planes. The alarm will get activated more often without there being actual fighter plane in addition to actual threat. This means the system is highly sensitive but less specific. And assume the same system, you make the system less sensitive then the alarm will get activated when there is actual threat of fighter plane. Meaning it is less sensitive but more specific. So there is a traid-off between sensitivity and specificity. Increasing one compromises the other. A receiver operating characteristic plot is obtained by calculating the sensitivity and specificity of every observed data value and plotting sensitivity (Y axis) against 1 – specificity (X axis), as in Figure 4. A test that perfectly discriminates between the two groups would yield a “curve” that coincided with the left and top sides of the plot (curve A). A test that is completely useless would give a straight line from the bottom left corner to the top right corner (curve C). In practice there is virtually always some overlap of the values in the two groups, so the curve will lie somewhere between these extremes (curve B). A global assessment of the performance of the test is given by the area under the receiver operating characteristic curve. Figure 4ROC Curve

Use of Diagnostic Tests in Evidence Based Medicine