Stats
Actions
Tags
From medsyniq-lite
Pre-test probability estimation, likelihood ratios, post-test probability calculation, Fagan nomogram, and threshold-based decision-making
How this skill is triggered — by the user, by Claude, or both
Slash command
/medsyniq-lite:bayesian-reasoningThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Bayesian reasoning provides the mathematical foundation for diagnostic thinking. It formalizes how new information (test results, clinical findings) should update diagnostic probability, enabling rational decisions about testing and treatment.
Bayesian reasoning provides the mathematical foundation for diagnostic thinking. It formalizes how new information (test results, clinical findings) should update diagnostic probability, enabling rational decisions about testing and treatment.
The estimated probability of disease before a test is performed. Sources for estimation:
Pre-test probability is the most important determinant of post-test probability. A test is only as useful as the clinical context in which it is applied.
Limitations: Sensitivity and specificity describe test performance but do not directly tell you the probability of disease given a test result. That requires Bayesian updating.
Likelihood ratios (LRs) express how much a test result changes the odds of disease. They are the preferred metric for Bayesian updating because they are independent of prevalence.
Positive Likelihood Ratio (LR+): LR+ = Sensitivity / (1 - Specificity)
Interpretation: How many times more likely is a positive result in a patient WITH disease compared to one WITHOUT disease?
Negative Likelihood Ratio (LR-): LR- = (1 - Sensitivity) / Specificity
Interpretation: How likely is a negative result in a patient WITH disease compared to one WITHOUT disease?
| LR+ | LR- | Shift in Probability |
|---|---|---|
| >10 | <0.1 | Large, often conclusive |
| 5-10 | 0.1-0.2 | Moderate |
| 2-5 | 0.2-0.5 | Small but sometimes important |
| 1-2 | 0.5-1.0 | Rarely important |
An LR of 1.0 provides no diagnostic information whatsoever.
Pre-test odds = Pre-test probability / (1 - Pre-test probability)
Example: 30% pre-test probability → 0.30 / 0.70 = 0.43
Post-test odds = Pre-test odds × LR
Example: Pre-test odds 0.43 × LR+ of 6.0 = 2.57
Post-test probability = Post-test odds / (1 + Post-test odds)
Example: 2.57 / 3.57 = 0.72 = 72%
For quick bedside estimates, approximate probability shifts:
These approximations work best in the mid-range of probability (20-80%).
A graphical tool for Bayesian updating:
Clinically useful for bedside estimation when a calculator is not available.
When multiple independent tests are performed sequentially:
Example of valid sequential testing: D-dimer followed by CTPA for PE — these test different aspects (fibrinolysis vs anatomy).
Example of invalid sequential testing: CRP followed by ESR — both reflect acute-phase response and are highly correlated.
The probability below which disease is sufficiently unlikely that no further testing is warranted and the diagnosis is excluded. Determined by:
The probability above which disease is sufficiently likely that treatment should begin without further testing. Determined by:
When probability falls between the test threshold and treatment threshold, testing is indicated. Testing is only useful when it can move probability across one of these thresholds.
Key insight: If a test cannot move probability across either threshold regardless of the result, the test should not be ordered. This is common when pre-test probability is either very low or very high.
Ordering a test without considering pre-test probability. A positive D-dimer in a low-risk patient (pre-test probability 5%) yields a post-test probability of only ~20-30%, not a confirmed PE.
Many tests produce continuous results. Likelihood ratios vary across the result spectrum. A highly elevated troponin has a much higher LR+ than a marginally elevated one. Use interval LRs when available.
Combining LRs from correlated tests overestimates diagnostic certainty. CRP + procalcitonin are partially redundant — their combined LR is less than the product of individual LRs.
Clinicians tend to overestimate low probabilities and underestimate high probabilities. Anchor estimates to validated prediction rules when available.
When assisting with diagnostic reasoning:
npx claudepluginhub proflow-labs-ai/medsyniq-lite --plugin medsyniq-liteGenerates and systematically ranks differential diagnoses using anatomical, pathophysiological, and probabilistic frameworks to reduce diagnostic error.
Searches 12+ authoritative clinical guideline sources (NICE, WHO, NCCN, AHA, ADA, SIGN, USPSTF, IDSA, ESMO, ESC, EASL) for evidence-graded treatment recommendations, dosing protocols, and screening guidance with source prioritization.
Applies Bayes' Theorem to update beliefs given a specific prior and new evidence. Use when interpreting test results, metrics, or diagnostic signals to avoid overreacting.