Medical Decision-Making in Clinical Practice

Medical decision-making (MDM) is a central aspect of clinical practice.
The presentation explores clinical decisions, pretest probability, and diagnostic test characteristics.

Clinical decision-making has inherent uncertainty due to imperfect data and unpredictable treatment outcomes.
Understanding probability and using new information are crucial for decision-support systems.
Probabilistic medical reasoning helps handle uncertainty but isn't suitable for all cases.

Good medicine involves understanding probabilities and possibilities for tests to give valuable information, rather than indiscriminate testing

Example of the accuracy of the polymerase chain reaction (PCR) test for detecting HIV in blood donors.
The challenge involves determining the likelihood of a donor having HIV based on test results.
The true positive rate is 98%, when an HIV antibody is present
The true negative rate is 99% when an HIV antibody is absent.

Ms. Kamala is a 66-year-old woman with coronary artery disease.
Ms. Kamala had two coronary artery bypass graft surgeries.
Her chest pain has returned and worsened despite medication.
Ms. Kamala faces the choice of a third operation with surgical risk, or avoiding surgery at the risk of a heart attack.

Many medical decisions rely on imperfect associations between symptoms and diseases, leading to uncertainty.
"Medical decisions based on probabilities are necessary but also perilous. Even the most astute physician will occasionally be wrong." - Smith L. (1985)
Probability and decision analysis provide clearer guidance in complex medical situations.

Ms. Kirk is a 33-year-old woman with a history of a blood clot with pain and swelling in her left leg.
A suspicion of deep vein thrombosis (DVT) is raised.
An ultrasound shows abnormal blood flow, but a new clot’s presence is unclear.
Weighing the risks of an untreated clot leading to pulmonary embolism against bleeding from anticoagulant treatment.

Clinicians use vague terms like "probable" or "highly likely", which may cause misunderstanding when making decisions.
Probability-based expressions in medicine can reduce ambiguity and improve communication on the likelihood of outcomes.

First, data is collected using observation and history.
The clinician gathers patient history, symptoms, and physical examination findings.
For example, the patient reports chest pain and shortness of breath.
List of possible diagnoses (differential diagnosis).
Possible causes of chest pain include heart attack, pneumonia, and acid reflux.
Diagnostic tests are ordered to confirm or rule out diagnoses with ECG, blood tests, or imaging.
An ECG and blood test might help confirm a heart attack.
The most likely diagnosis is determined combined with an appropriate treatment plan.

Probability theory is used to update the likelihood of a disease based on tests (Bayes' theorem).
A doctor considers test accuracy (sensitivity/specificity) and prevalence before confirming a diagnosis.
The doctor considers a positive HIV test when the disease is rare in that population.
This approach is best in situations where false positives/negatives are common, like a cancer screeing.
This approch requires statistical thinking, not always practical in real-time clinical settings.

Pretest Probability Estimation:
Clinician evaluation of likelihood of heart disease based on symptoms, history, and experience.
Example, a 60-year-old man with chest pain with estimated pretest probability: 50% (1:1 odds).
Diagnostic Testing
An exercise stress test is performed to lessen uncertainty.
Abnormal ECG results support the presence of heart disease.
Post-Test Probability Calculation
Bayes' theorem the clinician updates the probability of disease based on the test results.
Then determines next steps in diagnosis and treatment.

Subjective Probability Estimation:
Clinicians judge probability based on how closely a patient resembles their mental image of a disease.
Probability estimation is influenced by how easily similar cases come to mind.
Objective Probability Estimates:
Prevalence (frequency of a disease in a population) serves as an anchor for probability estimation.
Refine probability estimates by placing patients into subgroups with known disease probabilities.

Biological measurements typically follow a symmetrical distribution.
Laboratories define the upper limit of normal as two standard deviations above the mean.
In reality, overlap occurs leading to misclassification.
True Positive (TP)
Diseased
True Negative (TN)
Healthy
False Positive (FP)
False Negative (FN)

Spectrum Bias:
Studies include severely ill patients and healthy volunteers, which makes detection easier.
Inflating sensitivity and specificity.
Test-Referral Bias:
Only patient positive index test results undergo the gold standard test.
Overestimated sensitivity and underestimated specificity.
Test-Interpretation Bias:
Interpretation of the index test influences the gold standard test.
Both sensitivity and specificity can be artificially increased.