(1)
Department of Neurology Neurosciences Centre, and Clinical Epidemiology Unit, All India Institute of Medical Sciences, New Delhi Delhi, India
Abstract
Making a diagnosis is moving from possibilities to high or low probabilities. Based on history and physical examination, we think of one or more possibilities. Some, we think, are more likely than others, while some may have to be ruled out. We do diagnostic tests to increase the probability of more likely ones to nearly 95 % or above and decrease the probability of the ‘rule out’ diagnoses to near ‘zero’ or less than 5 %. Thus, the function of a diagnostic test is to increase or decrease (in one word, revise) the probability of the diseases under consideration. The probability of a disease we consider before ordering a diagnostic test (called ‘pretest probability’) should substantially change after the test. The probability we get after the test result is called ‘posttest probability’. Pretest probability is usually based on history and physical examination. In fact, it develops through several revisions, starting with some probability with the first symptom and changing the probability as newer findings emerge on history and physical examination. For example, as soon as a patient complains of chest pain of 2 h of duration (without history of trauma), we think of certain possibilities – like acute myocardial infarction (MI), pericarditis, pneumonia, pleurisy and dissection of aorta. Looking at his age of, say, 60 years, and considering the frequency, we think MI more likely than others. We ask about the characteristics of pain (onset, character, radiation, etc.) and risk factors (like diabetes, hypertension, smoking, hyperlipidemia) and accordingly revise the probability of MI to a high level (say, 60 %). Then we do certain tests like ECG and serum CPK or TropT. Each revises the probability further. If CK and ECG changes are borderline, the probability may not change much, but if ECG shows ST–T changes, the probability goes up, and if CK is also raised 2× normal, then the probability is nearly 99–100 % and the diagnosis is confirmed. The tests in this case are revising the pretest probability of 60 % to a posttest probability of 99–100 %. The example illustrates that the function of a diagnostic test is to revise the pretest probability of the diseases which are being considered in the differential diagnosis (7.1).
The Clinical Diagnostic Process
Clinical diagnosis is sometimes a spot diagnosis, but more often a matter of pattern recognition and probabilistic thinking. Usually, making a diagnosis is moving from possibilities to high or low probabilities. Based on history and physical examination, we think of one or more possibilities. Some, we think, are more likely than others, while some may have to be ruled out. We do diagnostic tests to increase the probability of more likely ones to nearly 95 % or above and decrease the probability of the ‘rule out’ diagnoses to near ‘zero’ or less than 5 %. Thus, the function of a diagnostic test is to increase or decrease (in one word, revise) the probability of the diseases under consideration. The probability of a disease we consider before ordering a diagnostic test (called ‘pretest probability’) should substantially change after the test. The probability we get after the test result is called ‘posttest probability’. Pretest probability is usually based on history and physical examination. In fact, it develops through several revisions, starting with some probability with the first symptom and changing the probability as newer findings emerge on history and physical examination. For example, as soon as a patient complains of chest pain of 2 h of duration (without history of trauma), we think of certain possibilities – like acute myocardial infarction (MI), pericarditis, pneumonia, pleurisy and dissection of aorta. Looking at his age of, say, 60 years, and considering the frequency, we think MI more likely than others. We ask about the characteristics of pain (onset, character, radiation, etc.) and risk factors (like diabetes, hypertension, smoking, hyperlipidemia) and accordingly revise the probability of MI to a high level (say, 60 %). Then we do certain tests like ECG and serum CPK or TropT. Each revises the probability further. If CK and ECG changes are borderline, the probability may not change much, but if ECG shows ST–T changes, the probability goes up, and if CK is also raised 2× normal, then the probability is nearly 99–100 % and the diagnosis is confirmed. The tests in this case are revising the pretest probability of 60 % to a posttest probability of 99–100 %. The example illustrates that the function of a diagnostic test is to revise the pretest probability of the diseases which are being considered in the differential diagnosis (Table 7.1).
Table 7.1
Steps in making a diagnosis (probabilistic method)
Steps

Process of diagnosis


1.

Collect initial information (from history or examination)

2.

Generate possibilities

3.

Attach probability to each possibility

4.

Collect more information

5.

Revise probabilities

6.

Perform diagnostic tests

7.

Revise probabilities further to a level that helps decisionmaking

The function of a diagnostic test is to revise pretest probabilities of the diseases under consideration to posttest probabilities.
Dichotomous and Multilevel Results
What revises the pretest probability? The test results. Test results may take a number of forms: Sometimes they are numbers like erythrocyte sedimentation rate (ESR); sometimes they are multiple categories – strongly positive, moderately positive, mild or borderline positive and negative or strongly negative. Even the results expressed in numbers are most of the time converted in our mind into multiple categories: very high, high, borderline high, normal, borderline low, very low, etc. Both the forms may be termed ‘multilevel results’. At other times, results are expressed as positive and negative, termed dichotomous results. No matter how the results are expressed, they are useful only if they revise pretest probability to such a level of posttest probability that helps us to make a decision about the next step, for example, in treatment (i.e. helps to cross the treatment threshold). The usefulness of a test depends on whether its results help in treatment decisionmaking. A test is only as good as its results. Every type of test result needs to be assessed – whether a strongly positive, moderately positive or negative. Even if one of the results is helpful, the test may be considered useful.
More on Pretest Probabilities, Posttest Probabilities and Predictive Values
Let us start with a scenario. You want to develop a facility to determine foetal sex using ultrasound. You come across two ultrasonographers who claim to be correct nearly always in sex determination of foetus at 14 weeks’ gestation. You discuss and decide to conduct a study to establish the facility. You buy a stateoftheart ultrasound machine, which gives 3D picture of the foetus. Ultrasound is an operatordependent technology. First you decide to establish the consistency of the ultrasonographers (US) – both interobserver and intraobserver. Same patients are tested twice by the same observer (without his knowledge) and some by both the observers independently. Interobserver and intraobserver consistency are examined and you find them acceptable.
The next question is how to check their correctness. It was clear that the US report had to be compared with the final diagnosis at birth. You verify the US findings with diagnosis at birth in 400 consecutive births. Let us look at the results. Well, before you look at the results, think for a moment. Suppose US prediction of male was found correct in 60 % and of females 65 %. Do you think it is good? Surely not. As such without any test, what do you guess is the chance (probability) of male babies, 50 %, and of female babies, 50 %. Therefore, after the test if the chance becomes 60 or 65 %, this cannot be called acceptable, because any decision on the basis of such prediction or even communication to the patient is far too often wrong. You want the probabilities after the test to be much more (say, 90–99 %) than the probabilities before the test (which is 50 %). Probabilities before the test are called ‘pretest probabilities’ and those after the test results, ‘posttest probabilities’. A good test should give good results. Good results are those which revise pretest probabilities to such an extent that correct decisions can be taken or correct information can be given to patients.