Bayesian Reasoning
Bayesian Reasoning · Clinical Module
Clinical Reasoning Series

Bayesian Reasoning
for Clinical Practice

Navigate uncertainty, update beliefs, and make better clinical decisions

Module Overview
Welcome to Bayesian Clinical Reasoning

Your value as a clinician isn't just in immediate pain relief — it's in your expertise, your process, and your ability to navigate uncertainty. Bayesian reasoning provides a framework to manage expectations, make better diagnostic decisions, and update your understanding as new evidence emerges.

"You don't need to know everything. You need to be confident in showing patients you have a PROCESS that will get them where they need to go."

Why Bayesian Reasoning?
  • Navigate diagnostic uncertainty with confidence
  • Avoid anchoring on initial impressions
  • Make better rebooking decisions using natural history data
  • Understand chronic pain and belief change
  • Interpret imaging findings with proper context
The Bayesian Brain
Your brain naturally functions as a Bayesian machine, constantly updating its understanding based on new evidence. Being aware of this process allows you to use it strategically in clinical practice.
For New Graduates

You don't need to know everything. Patients gain confidence from seeing you have a systematic process for navigating complexity. Talk through your reasoning OUT LOUD with them.

Pro Tip

Check the Prognosis & Base Rates section for natural history data on 13 common MSK conditions and imaging base rates. The Communication Strategy section shows you how to talk through Bayesian reasoning with patients to build trust.

Foundations
Understanding Bayes Theorem
The Three Key Components
1. Prior Probability

The initial likelihood of a hypothesis being true — e.g., disease prevalence in the population. This is your starting point before any evidence is collected.

2. New Evidence

Information that updates the prior probability — e.g., test results, clinical observations, mechanism of injury.

3. Posterior Probability

The updated probability of the hypothesis being true after considering new evidence.

The Classic Example
A rare disease affects 0.1% of the population. A test for this disease is 99% accurate. If you test positive, what's the chance you actually have the disease?
Intuitive Answer

99%

Most people think this — it's wrong.

Bayesian Answer

Only 9%!

Out of 1000 people, 1 has the disease and tests positive. But 10 people without the disease also test positive (1% false positive). You're 1 of 11 positives.

With a second positive test, the probability jumps to 91%. Multiple positive tests are far more powerful than one.

Setting Expectations
Mean vs Median
Why This Matters
  • Setting accurate patient expectations
  • Deciding when to rebook patients
  • Identifying patients who deviate from expected trajectories
  • Weighting your prior probabilities appropriately
Mean (Average)

The arithmetic average of all outcomes. Can be pulled higher or lower by outliers.

Acute LBP mean: ~6 weeks

Includes patients who recover in 2 weeks AND those who take 12–24 weeks, pulling the average higher than what most people experience.

Median (Most Common)

The middle value — represents what the majority of patients actually experience.

Acute LBP median: ~3–4 weeks

Most patients report their pain "isn't really bothering me anymore" around this timeframe, notably shorter than the mean.

Clinical Application: Cervical Radiculopathy
Median: 16 weeks

Most common timeframe for significant improvement

Mean: 6–8 months

Average pulled higher — not uncommon for recovery to extend to 12+ months

Decision Framework
  • Good prognostic signs? Anchor expectations closer to median (16 weeks)
  • Poor prognostic indicators? Anchor expectations closer to mean (6–8 months)
  • Update continuously based on how patient tracks in first 2–4 weeks
Weight Toward MEDIAN When:
  • Start Back tool scores LOW
  • Good social support and health literacy
  • Positive early feedback (first 2 weeks showing good progress)
  • Patient demonstrating good self-efficacy
Weight Toward MEAN When:
  • Start Back tool scores HIGH
  • Poor prognostic indicators present
  • Slow or minimal improvement in first 2 weeks
  • Multiple comorbidities or complexity factors
The Bell Curve Reality
  • Most patients cluster around the median (the peak of the bell curve)
  • Some patients take much longer — the tail extends right
  • This tail pulls the mean higher than where most patients sit
  • Your job: use the initial assessment to predict which part of the curve this patient will inhabit
Interactive Tool
Bayesian Calculator
Base rate in the population
Correctly identifies disease (and correctly identifies non-disease)
Independent positive test results
Probability of Having Disease
9.0%
After 1 positive test
Clinical Insight

Notice how low prevalence dramatically affects your posterior probability, even with highly accurate tests. This is why starting with base rate (prevalence) is diagnostically stronger than jumping straight to tests. Your clinical tests are most powerful when they update an already reasonable prior probability.

Clinical Application
Bayesian Clinical Diagnosis
The Optimal Diagnostic Sequence
  1. 1
    Start with Base Rate — What's the prevalence of conditions in this population? This is your prior probability.
  2. 2
    Update with Subjective Information — Mechanism of injury, symptom patterns, timeline. First evidence update.
  3. 3
    Try to Disconfirm Your Leading Hypothesis — Not confirm it! This prevents anchoring bias.
  4. 4
    Clinical Tests — Use specific tests to further update probability before initial diagnosis.
The Golden Question for Every Assessment Tool

Before collecting ANY piece of data, ask: "Why am I collecting this? How will it update my priors?"

Goals vs Current Function
Large gap + negative language = likely slower trajectory, needs more frequent follow-up. Updates prior toward MEAN.
Range of Motion + Behaviour
Reveals if patient is avoider/coper AND if that needs validation/violation. Updates toward standard or prolonged trajectory.
Example
80-Year-Old: Anterior Knee Pain
Base rate: OA highly prevalent in 80+ population

Disconfirming questions:
  • Was there trauma? (No = supports OA)
  • Sudden onset? (No = supports OA)
  • Consistent with degenerative pattern? (Yes = supports OA)
Unable to disconfirm OA → strong posterior probability
Example
30-Year-Old: Anterior Knee Pain
Base rate: Patellofemoral pain syndrome common in young adults

Disconfirming questions:
  • Cutting/pivoting sports? (Yes = consider ACL)
  • Heard a pop? (Yes = update to ACL likely)
  • Immediate swelling? (Yes = ACL more likely)
Successfully disconfirmed PFPS → pivot to ACL testing
Common Diagnostic Errors
  • Starting with tests instead of base rate — jumping to Lachman's without considering prevalence
  • Confirmation bias — only looking for evidence that supports your initial impression
  • Anchoring too early — telling an 80-year-old "you have arthritis" without adequate disconfirmation
  • Over-interpreting imaging — attributing symptoms to common incidental findings
Natural History Data
Prognosis & Base Rates
Why This Matters
  • Natural history data = expected treatment response = your prior for rebooking decisions
  • Base rates of imaging findings = how much weight to give "positive" findings
  • Timeline expectations = when to update your diagnosis if patient isn't following expected trajectory
Low Back Pain
Base Rate

1–2 episodes per year is NORMAL for general population (even without chronic LBP). Chronic LBP patients have higher frequency.

Natural History
  • Acute LBP: Sharp drop in 3–5 days, asymptotes toward zero over 3–6 weeks
  • Chronic LBP: Initial drop, then plateaus around baseline ±1–2 pain points
  • Episodic flares are NORMAL — not "re-injury"
  • Flare duration typically 3–7 days before returning to baseline
  • Median improvement: 3–4 weeks; Mean: 6 weeks (pulled by outliers)
  • Most improvement in first 6 weeks; beyond 6 weeks improvement slows
Expected Timeline

Most improvement in first 6 weeks. Expect episodic flares 1–2×/year minimum.

Rebooking Strategy

Rebook frequently in first 6 weeks if not following expected trajectory. After 6 weeks, longer intervals unless patient deviates. Build a flare management plan — flares don't mean treatment failure.

Imaging Context

84% of asymptomatic people age 48 have disc degeneration on MRI. Spondylolysis (11.5% on CT) has NO significant association with LBP occurrence.

Neck Pain
Base Rate

20–40% of general population experienced neck pain in the previous month.

Natural History
  • 45% show decreased pain/disability in first 6 weeks
  • 60–80% of workers report neck pain 1 year later
  • Course is recurrent/episodic — complete resolution is the exception
  • For acute trauma (WAD): 3 trajectories — mild (45%), moderate (40%), severe (15%)
  • Recovery occurs most rapidly in first 6–12 weeks, then slows
Timeline

Most recovery in first 6–12 weeks; little improvement after 12 months.

Rebooking Strategy

Identify trajectory early. Severe WAD trajectory needs frequent follow-up. After 12 weeks, expect slower gains.

Imaging Context

In those 60–65 years: 95% males and 70% females show degenerative changes.

Ankle Sprain
Natural History
  • Pain decreases rapidly within first 2 weeks
  • 75% of high school athletes return within 3 days, 95% within 10 days
  • However, 5–46% still experience pain at 1–4 years
  • 33–55% report instability long-term
  • 50–85% report full recovery at approximately 3 years
Timeline

Rapid early improvement (2 weeks), but significant minority have long-term issues.

Rebooking Strategy

Most can be managed with 1–2 week follow-ups initially. If pain persists beyond 4 weeks, increase follow-up frequency to identify those at risk for chronic issues.

Rotator Cuff / Subacromial Pain
Base Rate

Up to 25% of adults experienced shoulder pain in last year.

Natural History
  • About 50% recover completely within 6 months
  • 32% continue to have more than minor symptoms at 1 year
  • Pain beyond 3 months associated with poorer recovery
  • Minimum 23 weeks for favourable prognosis; improvement may continue to 24 weeks
  • Rapid improvement first 12 months, then plateau
Timeline

Expect improvement over 23–24 weeks; reassess if not improving by 3 months.

Rebooking Strategy

Weekly follow-ups if pain >3 months at presentation. For acute onset, 2–3 week intervals, extending if following expected trajectory.

Imaging Context

Asymptomatic RC tears: 8–40% partial thickness, 0–46% full thickness. In 60s: 25% have full tears. In 80s: 50% have full tears.

Adhesive Capsulitis
Natural History
  • Average duration 30.1 months (range 12–42 months)
  • At 12–18 months, mild to moderate disability may persist
  • 40–56% report ongoing pain and ROM loss up to 20 years after onset
  • The belief that all cases resolve naturally is challenged by recent evidence
Timeline

12–42 months; NOT always self-limiting. Manage patient expectations accordingly.

Rebooking Strategy

Can distinguish from rotator cuff pain within 7 days based on treatment response. If true frozen shoulder, longer intervals are acceptable. Focus on education and self-management.

ACL Reconstruction
Base Rate

1–2% per year in cutting/pivoting sports teams.

Natural History
  • 80% return to some form of sport after ACLR
  • Only 67% return to pre-injury level
  • 55% return to competitive level
  • 20% sustain reinjury to either knee (90% during high-risk sports)
  • Minimum RTS: 9–12 months (aligns with graft maturation)
Timeline

9–12 months minimum for RTS. Manage expectations about level of return — not all will reach pre-injury performance.

Rebooking Strategy

Frequent early rehab (2–3×/week first 6 weeks), then 1–2×/week. Monthly check-ins from 3–9 months. Intensive RTS testing at 9–12 months.

Meniscus Tear
Base Rate

30% of asymptomatic knees have meniscal tears (18% have degeneration).

Natural History
  • 88–90% follow moderate or high trajectory (gradual or early improvement)
  • 10–12% have severe baseline impairments with minimal improvement over 5 years
  • Majority experience early improvements, within normative data by 12 months
  • Clinically relevant improvements can occur up to 24 months
  • 84% of menisci remain stable over 8 years
Timeline

Most improvement in first 12 months; can continue to 24 months.

Rebooking Strategy

If following favourable trajectory, can extend intervals. If in the 10–12% severe group, needs frequent monitoring and possible surgical consultation.

Patellofemoral Pain
Natural History
  • 40% report unfavourable 12-month recovery
  • Majority who recover favourably do so within first 3 months
  • 71–91% report ongoing pain up to 20 years after initial diagnosis
  • In adolescents: 55% still have pain at 2-year follow-up
  • Improvements after 5–6 weeks intervention may plateau
Timeline

If going to improve, expect it within 3 months. Many have persistent symptoms long-term.

Rebooking Strategy

Intense first 6–8 weeks (weekly sessions). If no improvement by 3 months, reassess diagnosis and consider comorbidities. Longer symptom duration at baseline = poorer prognosis.

Hamstring Strain
Natural History
  • Sprinting-type injuries: median 15–23 days to RTS
  • Stretching-type injuries: median 43–50 days to RTS (much longer)
  • Wide variation in time to RTS
  • MRI has negligible additional predictive value over history and examination
Timeline

2–8 weeks depending on mechanism; stretching-type take 2–3× longer.

Rebooking Strategy

Weekly for first 2–3 weeks, then based on functional progression. Cannot accurately predict RTS from initial assessment — need serial monitoring.

Achilles Tendinopathy
Natural History
  • Patients can expect improvement between 3–12 months after treatment
  • NOT expected to improve beyond 12 months
  • Chronic symptoms persist in ~25% of patients at 10 years
  • Wait-and-see: minimal improvement expected short-term
  • 85% of athletes return to sport (level uncertain)
Timeline

3–12 months; no improvement expected after 12 months. Advise against wait-and-see.

Rebooking Strategy

Every 2–3 weeks initially, monthly after 3 months. If no improvement by 12 months, consider other interventions.

Lateral Epicondylitis (Tennis Elbow)
Natural History
  • 90% achieve resolution at 1 year WITHOUT treatment
  • Normal course: 12–18 months
  • Long symptom duration does NOT affect prognosis
  • Spontaneous improvement is likely even in prolonged cases
Timeline

12–18 months; excellent prognosis even without treatment. Reassurance is powerful here.

Rebooking Strategy

Monthly intervals given high spontaneous resolution rate. Main clinical value is education and reassurance. Duration doesn't worsen prognosis — maintain hope.

Concussion
Natural History
  • 80–90% have symptom resolution by 7–10 days
  • Cognitive deficits, balance, and symptoms improve in first 2 weeks for most
  • RTP often within 10 days
  • Sizeable minority of high school/college athletes take >10 days
  • 5–58% have persistent symptoms affecting daily function
Timeline

Most recover in 7–14 days; minority have persistent issues.

Rebooking Strategy

Review every 2–3 days initially. Greater acute symptom severity = predictor of slower recovery. Low symptoms day 1–2 = favourable sign.

Imaging Context
Imaging Base Rates in Asymptomatic Individuals

Many "abnormal" imaging findings are present in asymptomatic individuals. Always start with clinical presentation and use imaging to update your priors — not as the starting point.

Lumbar Disc Degeneration
84% (age 48), 90% (age 60+)
Part of normal ageing, not necessarily pathological
Cervical Disc Degeneration
95% males, 70% females (age 60–65)
Extremely common in asymptomatic individuals
Meniscal Tears
30% in asymptomatic knees
93% medial, 91% posterior horn. Common incidental finding.
Rotator Cuff Tears
34% overall asymptomatic. Age 60+: 28% full, 26% partial
Increases dramatically with age; presence doesn't equal causation
Hip Labral Tears
41–73% in asymptomatic hips
>50% chance of labral tear in asymptomatic volunteer
CAM Morphology (FAI)
37% athletic, 13–72% in asymptomatic males
Surgery doesn't clearly prevent OA development
Lumbar Spinal Stenosis
77.9% have moderate+ central stenosis
Only 17.5% of severe stenosis is symptomatic
Spondylolysis
11.5% on CT
No significant association with LBP occurrence
Shoulder Labral Tears
55–72% (including superior labrum)
Extremely common even in asymptomatic shoulders
Knee Cartilage Defects
75% of asymptomatic adults age 70
OA features common in older population — presence ≠ pain source
Clinical Skills
Communication Strategy
The Core Principle

You don't need to know everything. You need to show patients you have a PROCESS.

Patients gain confidence not from you claiming to have all the answers, but from seeing you have a systematic approach to navigating uncertainty. Talk through your Bayesian reasoning OUT LOUD with them.

Script 1 — Explaining Your Initial Assessment
"Based on what you've told me today and the examination I've done, here's what I'm thinking. This bit of information [point to Start Back score] makes me think your journey will probably look more like [describe trajectory]. But this other thing [point to another factor] makes me not 100% sure, so I'd like to check in with you in [timeframe] to see how you're tracking."
Why this works: You're being transparent about your reasoning while showing you have a systematic approach to uncertainty.
Script 2 — When Things Are Going Well
"Okay, so when you first came in, I thought you'd probably follow this path [gesture]. And you're actually tracking really well with what I'd expect for someone in your situation. That means I'm confident we can push our next appointment out to [longer timeframe] because you're following the normal course for this condition."
Why this works: You're explaining WHY you're comfortable extending intervals, which builds their confidence in the plan.
Script 3 — When Things Aren't Following Expected Path
"So here's what's interesting. When you first came in, based on [factors], I expected you'd be at about [this point] by now. But you're actually here [different point]. That doesn't mean something's wrong necessarily, but it tells me I need more information. So instead of seeing you in [original timeframe], I'd like to bring you back in [sooner] so we can figure out what's influencing your journey."
Why this works: You're normalising the deviation while showing you're actively managing uncertainty through more frequent data collection.
Script 4 — Natural History Education
"Let me show you what typically happens with this condition. Most people experience [draw or gesture the trajectory]. You're going to feel worst right now, but over the next [timeframe], you should notice [specific changes]. Now, that doesn't mean it's a straight line down — there will be ups and downs. What I'm looking for is that overall, the trend is going this direction [gesture]. Make sense?"
Why this works: Visual explanation of natural history reduces allostatic load by making the journey predictable and controllable.
Script 5 — Normalising Flare-Ups
"One thing that's really important to know about [chronic condition]: flare-ups are NORMAL. Even people without chronic pain get 1–2 episodes a year. For you, you might get more. When you have a flare, it typically lasts [timeframe], feels intense, then settles back down to your baseline. It doesn't mean you've 're-injured' yourself or that treatment isn't working. It's just the nature of the condition. So we're going to build you a flare-up management plan."
Why this works: Pre-framing flares as normal prevents catastrophising and unnecessary re-presentation to healthcare.
For New Graduates: Building Confidence
  • Be explicit about your process: "This is how I work through complex cases..."
  • Acknowledge uncertainty: "I'm not 100% sure yet, which is why..."
  • Show your system: "These are the checkpoints I use to know we're on track..."
  • Explain your reasoning: "This piece of info makes me think... but this other piece makes me think..."

Paradoxically, admitting what you don't know while showing you have a method to find out builds MORE confidence than pretending to have all the answers.

Learning Through Practice
Case Study: Shoulder Pain
Step 1 of 7 Patient Presentation
Patient Presentation
  • Patient: 41-year-old male, car retail business owner
  • Chief Complaint: 3-month gradual onset left shoulder pain
  • History: No acute injury, Type 2 diabetes (poorly controlled), hypertension and obesity
  • Extensive shoulder trauma from semi-professional rugby; multiple previous dislocations
  • Imaging: Ultrasound and X-ray show mild OA
Initial Assessment Findings
  • ROM pattern: AROM = PROM (equal active and passive)
  • Order of loss: Internal rotation > External rotation > Abduction > Flexion
  • Wall reach: 88cm abduction, 110cm flexion
  • Key finding: Equal AROM/PROM with capsular pattern
Bayesian Differential Diagnosis
High Probability

Adhesive Capsulitis — metabolic conditions, age, trauma history, equal AROM/PROM all point here

Moderate Probability

Rotator Cuff Related Shoulder Pain — higher base rate in general population

Low Probability

Osteoarthritis — scan findings mild, wouldn't severely restrict ROM in this pattern

Communication Strategy
What was communicated to the patient
"Based on your presentation, metabolic history, and examination findings, the most likely diagnosis is either frozen shoulder or rotator cuff related pain. I can't be certain yet which is more accurate. We're going to use treatment as both intervention and assessment — your response will help clarify the diagnosis."
Key phrase: "Most likely diagnosis" — not definitive. Maintaining diagnostic humility. Patient task: Home measurements (wall reach, wall slide) to contribute data.
Treatment as Assessment
Interventions Used as Diagnostic Tools
  • Manual therapy (glides, PNF, isometrics)
  • Exercise (broomstick drills, wall slides, theraband)
  • Goal: Can we improve ROM by influencing the nervous system?
  • Reasoning: If ROM significantly improves with these interventions, suggests rotator cuff pain rather than true capsular restriction

Rebooked in 7 days for crucial reassessment — diagnostic uncertainty is high, so short follow-up interval is warranted.

7-Day Follow-Up Results
Significant Improvement!
  • 20 degrees greater abduction
  • 14cm better wall reach
  • External rotators significantly weak on dynamometry
Updated Diagnosis Communicated
"Based on your excellent response to treatment, this is most likely rotator cuff related shoulder pain, not frozen shoulder. True frozen shoulder wouldn't improve this quickly."
Key Lessons
  • Adhesive capsulitis and poorly managed rotator cuff pain can present identically
  • Don't anchor diagnosis too early — maintain openness to updating
  • Communicate uncertainty honestly — explain your reasoning process
  • Book early follow-ups when diagnostic uncertainty is high
  • Treatment response is valuable diagnostic data
Putting It Into Practice
Clinical Applications
Rebooking Strategy
Use natural history data to inform your rebooking intervals. Check the Prognosis section for specific timelines.
Rebook LESS Frequently When:
  • Condition is straightforward and predictable
  • Following textbook recovery pattern
  • Patient is clearly a "normal responder"
  • High confidence in prognosis
  • E.g. Tennis elbow at 6 months (90% spontaneous resolution)
Rebook MORE Frequently When:
  • Diagnostic uncertainty remains high
  • Not responding as expected from natural history
  • Need to determine high/normal/low responder status
  • Gathering data points to update your priors
  • E.g. LBP not improving in first 6 weeks

"Treatment outcomes serve as data points to confirm or contradict initial hypotheses."

Chronic Pain & Belief Change
Key principle: Strong prior beliefs require significant and repeated evidence to change.
Belief Strength Scale (0–10)
  • 0–3 or 7–10: Extremely difficult to change — requires multiple strong interventions over time
  • 4–6: More receptive to new information and belief updates
Patients with chronic pain and anxiety/depression have hypersensitive minds — both body and cognition — making it harder to update their thinking. Expect slower belief change and celebrate small victories.
Programming & Treatment Planning
Understand the mean treatment effect for any program and assess where your patient sits on that curve.
Response Categories
  • High responder: Exceeds typical improvements — may need less input
  • Normal responder: Following expected trajectory — standard protocol
  • Low responder: Underperforming — needs protocol adjustment
Patient Empowerment
Teach patients to monitor leading and lagging indicators to participate in data collection.
Leading Indicators

Early signs of improvement: less morning stiffness, easier movements, better wall reach. These appear before full recovery.

Lagging Indicators

Outcome measures: return to sport, pain-free week, Start Back score. These confirm recovery has occurred.

Neurodivergent Considerations
ASD — Autism Spectrum Disorder

Difficulty updating strong prior beliefs. Unexpected changes cause distress because they disconfirm expectations.

Approach: Graded exposure with predictable small updates. Same environment with incremental changes.

ADHD

Too willing to update priors. Easily distracted by new stimuli, struggling to maintain a coherent worldview.

Approach: Help establish stronger priors through structure, routine, and filtering strategies.

Test Your Knowledge
Self-Assessment