--- module: 006-02 language: en chapter: 6 title: "Clinical Reasoning, History, Examination, and Diagnostic Probability" module_title: "Bayesian updating, decision thresholds, and diagnostic safety" source_sha256: dba09072c4d938db627ddcbc2607b770711d3ff15ae1b730fbf7f9762875162d --- # Bayesian updating and diagnostic safety ## Evidence changes odds ### Prior odds times likelihood ratio gives posterior odds ### Keep the direction and scale of the update ### Predictive values answer the inverse question ### Likelihood ratios are not fixed #### Disease spectrum and test threshold #### Operator skill and timing #### Definition of the reference standard ### Correlated tests are not independent evidence ## Decision thresholds ### Below the testing threshold, investigation harms ### Between thresholds, information can change management ### Above the treatment threshold, act without waiting ### Thresholds move with stakes #### Dangerous and rapidly treatable justifies low-probability testing #### Toxic or invasive treatment demands stronger evidence #### Reliable follow-up makes watchful waiting safer ### Empirical treatment degrades later evidence #### Antimicrobials sterilise cultures #### Glucocorticoids suppress inflammatory findings ### Value of information: will the result change an action ## Causal problem representation ### Semantic qualifiers separate mechanisms #### Acute against chronic #### Focal against diffuse #### Exertional against random ### List what the representation does not explain ### Multimorbidity is common, parsimony is not mandatory ### Mechanism-based differentials #### Syncope: reflex, orthostatic, cardiac, metabolic, drug #### Hypoxaemia: hypoventilation, diffusion, mismatch, shunt ### Time course carries causal information ## Examination quality ### Technical and biological variance #### Cuff size, position, noise, effort, expectation ### Repeat a surprising measurement before explaining it ### Interobserver agreement is not diagnostic accuracy ### Absence of a sign can be uninformative ### Coherent signs from one mechanism beat any single sign ### Point-of-care devices need the same discipline ## Cognitive forcing and team reasoning ### Pattern recognition is fast when the case is familiar ### Analysis is for unfamiliar, high stakes, inconsistent cases ### Switch modes deliberately ### Diagnostic timeout triggers #### Transfer of care #### Unexpected deterioration or treatment failure #### A result that does not fit ### Team diversity uncovers blind spots if challenge is allowed ### Safe handover names escalation criteria and ownership ## Diagnostic harm and safety-netting ### Overdiagnosis is not a false positive ### Incidental findings carry surveillance and anxiety ### Underdiagnosis from access, atypical presentation, fragmented records ### Diagnostic overshadowing #### Symptoms wrongly attributed to a known diagnosis #### Recheck physical and medication causes ### Safety-netting #### Working diagnosis and what remains uncertain #### Specific warning symptoms #### Where and how urgently to seek help #### Who reviews pending tests ### Follow-up gives evidence only when designed ### The final question is which action is safest at this probability