Translate this page into:
Using core clinical epidemiology principles for bedside evidence-based diagnosis
Correspondence to RAVIKUMAR CHODAVARAPU; drravikch@gmail.com
[To cite: Chodavarapu R, Raza PC. Using core clinical epidemiology principles for bedside evidence-based diagnosis. Natl Med J India 2025;38:240-3. DOI: 10.25259/NMJI_843_2023]
INTRODUCTION
Research by the Johns Hopkins Armstrong Institute Centre for Diagnostic Excellence and partners from the Risk Management Foundation of the Harvard Medical Institutions shows that in the USA, the annual estimate of mortality or serious morbidity due to diagnostic error is approximately 795 000.1
Diagnosis of an illness is a vital step in proper clinical management. It often involves a certain amount of uncertainty that cannot be clearly defined by qualitative terms such as possible, probable, likely, and unlikely. They give variable impressions to different clinicians and patients, and are differently interpretable. Clinical epidemiology provides quantitative methods to minimize subjectivity and enhance objectivity in the diagnostic process. In developing countries, medical schools often lack departments of clinical epidemiology. Consequently, training in clinical diagnostic skills is not integrated with clinical epidemiology, and the clinical management of patients does not explicitly consider the principles of clinical epidemiology. Medical students, and some clinicians, may think that medical subject knowledge and clinical skills, by themselves, without competency in the application of principles of clinical epidemiology, will always result in correct diagnoses. Continuing medical education programmes do not deal with the utility of such skills. There is a need to develop medical education in bedside objective diagnosis, utilizing principles of clinical epidemiology.
We present essential concepts of clinical epidemiology in the form of practical tips for making bedside diagnoses. Practising these 10 tips makes a learner competent in bedside diagnosis, reduces diagnostic errors, and thus, lowers therapeutic errors. Understanding the clinical utility of the principles is more important than knowing the underlying mathematical information. This article is intended to serve as a student’s learning tool and a teacher’s training module for bedside diagnosis. We formulated these learning–teaching tips based on scientifically established diagnostic clinical epidemiology principles and the corresponding author’s 30 years of experience and observations in training undergraduate and postgraduate students of paediatrics diagnostic clinical skills.
APPLYING PRINCIPLES OF CLINICAL EPIDEMIOLOGY DURING BEDSIDE DIAGNOSIS
Tip 1: Make accurate and complete measurements of symptoms and signs of the patient
Appropriate and specific diagnostic questions can be framed only when the patient’s clinical characteristics are measured with as much accuracy and precision as possible. Inaccurate clinical measurements render even the best quality systematic reviews pertaining to the pre-test probability of diseases and the diagnostic power of various diagnostic tests futile in making the right diagnosis. The importance of accurate clinical examination and the power of clinical observations in making useful diagnostic decisions is well demonstrated through relevant clinical epidemiology principles by various pioneers.2
Tip 2: Make a list of differential diagnoses with a focus on presenting complaints or symptoms
Each diagnosis is a synthesis of three components: anatomical diagnosis, pathophysiological diagnosis, and aetiological diagnosis.3,4 Better diagnostic decision is based on the three components instead of depending on only one component. Diagnosis based solely on aetiology (the cause of disease) without considering anatomical (where the disease is located) and pathological diagnosis (how the disease is caused) is often uncertain. For example, the identification of lobar pneumonia (anatomical–pathophysiological diagnosis) is a better predictor of pneumococcal pneumonia (aetiological diagnosis).
Diagnosis by synthesizing the three components is more accurate and precise when the diagnosis is made considering the diagnostic power (specificity, sensitivity and diagnostic likelihood ratios) of a diagnostic test (symptoms, signs or laboratory values) that is relevant to each component, e.g. tubular breathing is specific for lobar consolidation and sputum culture for pneumococcal aetiology.
Differential diagnosis making starts from the stage of complaints and other symptoms.5
Tip 3: Prioritise each diagnosis in the differential diagnosis according to its probability and prognosis
In a probabilistic approach, diseases with the highest probability are considered initially in the diagnostic list from high probability to low probability.6,7 In the prognostic approach, diseases with even low probability are considered in the differential diagnosis when they have serious outcomes if left untreated.5,7 The diagnostic list is modified based on high probability and prognosis, weighing the benefits and risks and considering the harms of both a missed diagnosis (not treating) and a misdiagnosis resulting in unnecessary treatment. The benefits versus risks of diagnostic tests are also considered.
In a pragmatic approach, diseases that respond well to therapy when treated early are considered in the differential diagnosis. A novice in training approaches diagnosis typically by a possibilistic approach, considering differential diagnosis for each symptom and sign. A broad list of diagnoses is compiled, incorporating textbook-described causes as well as recent, rare case reports from continuing medical education (CME) programmes or dramatic single personal case experiences. A possibilistic diagnostic approach often involves simultaneously testing for all considered diagnoses or treating many of them. This results in over-testing and over-treatment, leading to unnecessary harm from testing and treatment.
Tip 4: Diagnostic probability of each diagnosis is to be on the principle of Bayes probability estimation
In this process, the diagnostic probability of a disease is estimated based on its pre-test probability and the diagnostic power of a diagnostic test, such as symptoms, signs, or laboratory results.8,9 The diagnostic power or disease predictive power of a diagnostic test revises the pre-test probability of the diagnosis, upwards or downwards, based on whether the test result is positive or negative.10
The mathematical process of modifying or refining the diagnostic probability of a disease from its pre-test probability to post-test probability is the same for all diagnostic tests: symptoms, signs and laboratory tests such as blood tests, radiological tests and biopsies, as per Bayes principle.11
Example of steps of Bayes diagnostic probability estimation. If a 7-year-old child presents with fever, sore throat, and anterior cervical adenopathy, the diagnostic question is: What is the probability of Group A Streptococcal pharyngitis (GAS)?
The prior or pre-test probability of GAS pharyngitis in the child is about 30%.12
If a rapid antigen diagnostic test (RADT) is used to diagnose GAS pharyngitis, the RADT has a diagnostic power of 85.6% sensitivity and specificity of 95.4%.13
The post-test probability of GAS pharyngitis (post-RADT probability of GAS pharyngitis) if the test is positive is 88.9% and if the test is negative is 6.1%.
Tip 5: Pre-test probability is the likelihood of disease before undertaking a diagnostic test and is considered at each stage of the diagnostic process and changes dynamically with each diagnostic test from symptoms to signs and laboratory values
It is the probability of a diagnosis (disease) before doing a diagnostic test. It depends on the information available about a patient before doing a diagnostic test. The prior probability of a disease in a person, even before any symptoms have occurred, is usually referred to as prevalence or incidence.4 It is the likelihood of having the disease in a population to which the patient belongs. Using prevalence or incidence and symptoms as diagnostic tests, the diagnostic probability of a disease can be estimated. The post-symptom based diagnostic probability of a disease becomes the prior probability of the disease at the stage of using signs as diagnostic tests.7 The diagnostic probability of the disease after signs is the pre-test probability of the disease for laboratory tests.
Example. What is the probability of pneumonia in a population of children?
The incidence of pneumonia in children in South Asia is 2500 cases per 100 000 children.14
It is the pre-test probability of pneumonia in a child in South Asia before symptoms, signs, or diagnostic tests, such as a chest X-ray, are considered.
Prior probability, also known as pre-test probability, can be expressed as a percentage for convenience.
The incidence of pneumonia in children in South Asia is expected to be 2.5%.
Tip 6: Diagnostic power of a diagnostic test: Diagnostic power of any symptom, sign or laboratory value is its ability to distinguish diseased from non-diseased and is measured and expressed as diagnostic sensitivity (%), diagnostic specificity (%) and diagnostic likelihood ratios: Positive likelihood ratio and negative likelihood ratio
To diagnose tricuspid regurgitation, the physical sign of holo- systolic murmur at the left lower sternal border has a diagnostic power of 52% sensitivity and 95% specificity. If the pre-test probability of tricuspid regurgitation in a patient is 42%, the murmur as a diagnostic test changes the diagnostic probability of tricuspid regurgitation. Post-test probability of tricuspid regurgitation increases to 88% with the presence of a murmur and decreases to 27% with the absence of a murmur.15
Compared with chest X-ray, positive lung ultrasound findings (consolidation measuring >1 cm or a focal/asymmetrical B-lines pattern) showed a sensitivity of 87.8%, a specificity of 58.5%, a positive likelihood ratio of 2.12 and a negative likelihood ratio of 0.21.16
Tip 7: Order only diagnostic tests with appropriate diagnostic power
Order investigations that have the highest diagnostic power to increase or decrease the likelihood of a disease suitably. Diagnostic tests with the highest specificity or positive likelihood ratio, when they are positive, are useful for increasing the positive predictive value, thereby increasing the probability of a disease. Conversely, diagnostic tests with the highest sensitivity or lowest negative likelihood ratio, when they are negative, are useful for increasing the negative predictive value, thereby decreasing the probability of disease.2,8
Hypothetical example. A disease has a pre-test probability of 20%. A diagnostic test is used with a positive likelihood ratio (LR+) of 2.0. If it is positive, it increases the diagnostic probability of the disease by 15% from its pre-test probability level (20%+15%=35%). If a diagnostic test with a LR+ of 5.0 is used, it increases the diagnostic probability of the disease by 30% from its pre-test probability (20%+30%=50%). A diagnostic test with an LR+ of 10.0 increases diagnostic probability by 45% (20%+45%=65%).
Tip 8: Positive predictive value (PPV) (Post-test probability of disease when test is positive) and negative predictive value (NPV) (Post-test probability of no disease when test is negative) of diagnostic tests depend on both pre-test probability of each diagnosis and diagnostic power of each diagnostic test
The relation between pre-test probability, diagnostic power of a test, PPV and NPV with underlying mathematical concepts can be understood by observing Tables 1 and 2.
| Post-test probability | Pre-test probability % | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| 1 | 5 | 10 | 30 | 50 | 70 | 90 | 95 | 99 | |
| Post-test probability of disease with positive test (PPV) % | 50.0 | 83.9 | 91.7 | 97.7 | 99.0 | 99.6 | 99.9 | 99.9 | 100 |
| Post-test probability of no disease with negative test (NPV) % | 100 | 99.9 | 99.9 | 99.6 | 99.0 | 97.7 | 91.7 | 83.9 | 50.0 |
| Post-test probability of disease with negative test (1-NPV) % | 0.0 | 0.1 | 0.1 | 0.4 | 1.0 | 2.3 | 8.3 | 16.1 | 50.0 |
PPV positive predictive value NPV negative predictive value
| Post-test probability | Pre-test probability % | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| 1 | 5 | 10 | 30 | 50 | 70 | 90 | 95 | 99 | |
| Post-test probability of disease with positive test (PPV) % | 16.1 | 50.0 | 67.9 | 89.1 | 95.0 | 97.8 | 99.4 | 99.7 | 99.9 |
| Post-test probability of no disease with negative test (NPV) % | 99.9 | 99.7 | 99.4 | 97.8 | 95.0 | 89.1 | 67.9 | 50.0 | 16.1 |
| Post-test probability of disease with a negative test. (1-NPV) % | 0.1 | 0.3 | 0.6 | 2.2 | 5.0 | 10.9 | 32.1 | 50.0 | 83.9 |
PPV positive predictive value NPV negative predictive value
A student can comprehend from Tables 1 and 2 the following mathematical concepts that help proper estimation of diagnostic probabilities:
PPV and NPV values change with changing pre-test probabilities even when the diagnostic power of a test remains constant (within Table 1 and Table 2).
PPV and NPV can change at the same pre-test probability level when the diagnostic power changes (Table 1, diagnostic power is 99% versus Table 2, diagnostic power is 95%).
PPV and NPV indicate the degree of diagnostic certainty and uncertainty associated with a disease in a patient and are more important for clinical decisions because the presence or absence of a disease cannot be determined with certainty when a patient presents with symptoms. The prediction of a disease’s probability often requires both the pre-test probability and the diagnostic power of a test.17,18 Most often, diagnostic tests by themselves cannot accurately diagnose a disease without considering the pre-test probability of the disease.
In two rare situations, a diagnostic test on its own gives a more reliable estimate of the diagnostic probability of a disease:
It is rare that a diagnostic test has a sensitivity of 100% or specificity of 100%. In such a situation, a negative diagnostic test rules out the disease, and a positive result confirms the diagnosis. When the sensitivity and specificity of a test are <100%, absolute certainty in the diagnosis is not possible based on the test result alone. Tables 1 and 2 show the influence of pre-test probability on the positive and negative predictive values of diagnostic tests even when the diagnostic powers are high (99% sensitivity and specificity – Table 1 and 95% sensitivity and specificity – Table 2).
When the pre-test probability of a disease is 50%, the PPV and NPV will be closer to the specificity and sensitivity of the diagnostic test, respectively. However, the pre-test probability of a disease may be different from 50%. Hence, it is essential to have an estimate of the pre-test probability of a disease in a given patient and then consider the diagnostic power of the test to estimate the diagnostic probabilities (PPV and NPV) of the disease.
Tip 9: Order investigations after measuring symptoms and signs and interpret the investigation results in a valid way to get the estimate of the diagnostic probability that is relevant to a patient
Initial clinical measurements indicate the appropriate differential diagnosis.
Diagnosis based on clinical measurements will improve the estimate of the pre-test probability of disease (patient-specific) and aid in interpreting the results of investigations, considering the pre-test probability for each diagnosis in a patient.
The pre-test probability of a disease is practice-specific, varying from one clinical setting to another (primary, secondary, or tertiary care).
If investigation results are interpreted without considering the pre-test probability of disease relevant to the patient, it leads to greater diagnostic error.17 Consider the diagnostic power of the test ordered and derive the best possible estimate of the post-test diagnostic probability.
Positive results of investigations in the context of a higher pre-test probability will have a lower false positivity rate. If the pre-test probability of a disease is high and the diagnostic test result is negative, one must consider the possibility of a false-negative test result before ruling out the disease from the diagnostic list. Similarly, if the pre-test probability of disease is low and a diagnostic test result is positive, one must consider the possibility of a false-positive test result before deciding to treat.
Tip 10: Generate a comprehensive list of potential diagnoses with their diagnostic probability that could explain the patient’s symptoms
Refine the differential diagnosis based on the updated diagnostic probabilities, taking into account the sequential pre-test probabilities and the diagnostic power of the investigations.
Continue this process until a final diagnosis is reached that satisfactorily explains the patient’s symptoms.
CLINICAL DECISIONS
Considering the amount of uncertainty in each diagnosis, along with the patient’s condition, a clinician can estimate two clinical decision thresholds.5,17
Treatment threshold: A level of probability estimate of a diagnosis that is high enough to require treatment without further testing. Identifying treatment thresholds can be done clinically based on the clinical features that have adequate diagnostic power or the patient’s clinical condition.
Test threshold: A level of probability estimate of a diagnosis that is low enough not to require treatment, and further testing is not needed.
When the diagnostic probability of a disease is above the test threshold but below the treatment threshold, further testing is typically required before initiating treatment. Diagnostic tests will have greater utility when the diagnostic probability of a disease is above the test threshold and below the treatment threshold.2,19
A clinician must order a diagnostic test only when its result can significantly alter the pre-test probability of disease, such that the post-test probability of disease falls above or below one of the thresholds.
A diagnostic test should be obtained only when its outcome could alter the patient’s management. If the probability of disease after a diagnostic test is nearly equal to the probability of disease before the test, it is not worth performing the test.20
DIAGNOSTIC UTILITY OF PROBABILITY ESTIMATION PROCESS
Probability is a useful representation of diagnostic uncertainty
Diagnostic uncertainty can be better estimated using probability theory through the Bayes probability estimation process.
Bayes probability estimate of diagnostic certainty and uncertainty is useful:
To identify diagnostic probability thresholds to take critical clinical decisions such as to test or not test, to treat or not treat, and to observe.
As a preceding step to categorize numerical diagnostic probability, when needed into clinically useful categories: Low or medium or high probability disease.
PRACTICE POINTS
Clinical measurements must be complete and accurate
Focus differential diagnosis on complaints and symptoms
Estimate the probability of each disease in the list of differential diagnoses by considering the pre-test probability of the disease and the diagnostic power of the test
Prioritise diagnoses according to the probability of each diagnosis and prognosis for each disease according to benefit versus risk of treatment or testing
Decide whether to test or treat according to test or treatment thresholds.
Conflicts of interest
None declared
References
- Burden of serious harms from diagnostic error in the USA. BMJ Qual Saf. 2024;33:109-20.
- [CrossRef] [PubMed] [Google Scholar]
- Clinical epidemiology: A basic science for clinical medicine (2nd ed). Philadelphia, PA, USA: Lippincott Williams and Wilkins; 1991.
- [Google Scholar]
- Clinical examination: A systematic guide to physical diagnosis (7th ed). Australia: Elsevier; 2014.
- [Google Scholar]
- Symptom to diagnosis: An evidence-based guide (4th ed). New York, USA: McGraw Hill; 2020.
- [Google Scholar]
- Symptom to diagnosis: An evidence-based guide (2nd ed). New York, USA: McGraw Hill; 2010.
- [Google Scholar]
- A model for reasoning In: Stern SD, Cifu AS, Altkorn D, eds. Symptom to diagnosis: An evidence-based guide (4th ed). New York, USA: McGraw-Hill; 2020.
- [Google Scholar]
- Improved diagnostic accuracy through probability-based diagnosis Rockville MD, USA: Agency for Healthcare Research and Quality; 2022.
- [Google Scholar]
- Clinical epidemiology: The essentials (5th ed). Philadelphia, USA: Wolters Kluwer, Lippincott Williams and Wilkins; 2014.
- [Google Scholar]
- Basic methods of medical research (1st ed). New Delhi, India: AITBS Publishers; 2006.
- [Google Scholar]
- Evidence-based diagnosis: An introduction to clinical epidemiology. (2nd ed). London, UK: Cambridge University Press; 2020.
- [CrossRef] [Google Scholar]
- Acute pharyngitis In: Kliegman RM, Stanton BF, St Geme JW 3rd, Schor NF, Behrman RE, Emeritus, eds. Nelson textbook of pediatrics (20th ed). Philadelphia, PA, USA: Elsevier; 2016.
- [Google Scholar]
- Rapid antigen detection test for group A Streptococcus in children with pharyngitis. Cochrane Database Syst Rev. 2016;7:CD010502.
- [CrossRef] [Google Scholar]
- Child pneumonia data visualizer, based on UNICEF global databases. Available at https://data.unicef.org/resources/child/pneumonia/data-visualization-toolkit (accessed on 13 Jan 2020)
- [Google Scholar]
- Evidence-based physical diagnosis. (4th ed). Philadelphia, PA, USA: Elsevier; 2018.
- [CrossRef] [Google Scholar]
- Lung ultrasound performed by primary care physicians for clinically suspected community-acquired pneumonia: A multicentre prospective study. Ann Fam Med. 2022;20:227-36.
- [CrossRef] [PubMed] [Google Scholar]
- Diagnosis in general practice. Using probabilistic reasoning. BMJ. 2009;339:b3823.
- [CrossRef] [PubMed] [Google Scholar]
- Oxford handbook of epidemiology for clinicians. (1st ed). London, UK: Oxford University Press; 2012.
- [CrossRef] [Google Scholar]
- The challenge of medical diagnosis: A primer on principles, probability, process and pitfalls. Natl Med J India. 2015;28:24-8.
- [Google Scholar]
- Assessment of diagnostic technology in health care: Rationale, methods, problems, and directions: Monograph of the council on health care technology Washington, USA: National Academic Press; 1989.
- [Google Scholar]