• Advanced differential diagnosis for the practicing rehabilitation clinician
• Includes advanced knowledge for life long learning to improve differential diagnosis skills
  • Underlying causal connections
  • Reasoning patterns
• Focuses on screening for referral
  • Standard to complex clinical cases across practice settings
• Goal is to develop advanced diagnostic reasoning when faced with a myriad of presenting signs and symptoms, and the ability to develop novel differential diagnosis approaches for your practice setting

• Explain the use of causation, reasoning patterns and inference in differential diagnosis.

• Develop a causal network and consider the features of its structure in reasoning and making diagnostic inferences.
• Interpret a set of signs and symptoms in light of possible causes.
• Identify red flags that warrant a deeper consideration of alternative explanations for signs and symptoms.
• Generate a differential diagnosis and subsequent plan of action for complex patient cases across practice settings.

Includes:

• Consideration of the connection between research and practice
• Critique of a strictly empirical framework for decision making
• Development of the curriculum at Plymouth State DPT program
• Investigation of approaches of clinical reasoning
  • Cause, Models & Inference: Building a Knowledge Based Practice

Hope to encourage development of:

• Teaching and developing inferential reasoning based on causal models
• Development of specific Bayesian networks with practical relevance
• Continued discussion.....

• Differential diagnosis requires abductive inference (infer cause from observed effects) and conditional probability

• Abductive inference requires knowing cause - effect relationships
• Knowing cause - effect relationships requires inductive inference
• What we know from induction is an estimate and is based on probability
• Causal conditional probability allows us to infer effects from cause
• Bayes theorem allows us to invert conditional probability to infer cause from effect
• This is till too simple to be useful in clinical practice since the true set of cause and effect relationships are part of a large causal network
• We can learn, communicate and utilize the structure of these networks

Causation, reasoning patterns and inference

Explain the use of causation, reasoning patterns and inference in differential diagnosis.

Epistemology

As a critical realist I do not fully subscribe to pure or hyper empirical approaches to knowledge; nor do I believe that rationalism or idealism individually provide the necessary framework for knowledge in practice

But all we need to start is a foundation - "common sense" realism:

• Law of causation
• Law of non contradiction (X cannot be & not be in the same way at the same time)
• Analogical use of language
• Basic reliability in sensory perception

• Most of what we discuss we will make "discrete" (not as in be quiet about this, but as in a category, not continuous)
• Consistent with making continuous phenomena discrete (i.e. blood pressure becomes normal or hypertensive; blood values become anemia) we will assume normal is normal (not pathological), and that abnormal is abnormal (possibly pathological or at least a positive sign (+S))
• We will use
  • +D to indicate "Disease is present" as a fact
  • -D to indicate "Disease is not present" as a fact
  • +S to indicate that a "symptom or sign is present" as a fact
  • -S to indicate that a "symptom or sign is not present" as a fact
  • All test results and 'observable' clinical manifestations will be considered a sign
    • By observable we include all sensory modalities not just vision

• We will make use of 2x2 tables a lot

Relevant to our above discussion:

Relevant to disease and symptoms / signs:

For use in probability equations:

\(P(D|S)\)

\(P(S|D)\)

\(P(D|S)=\dfrac{P(S|D)\cdot P(D)}{P(S)}\)

But, conditional probability help establish causation

Causation based on probability

• Definite cause (\(+D \rightarrow +S\))
  • \(P(+S|+D) = 1\)
• Probable cause (\(+D \rightarrow +S\))
  • \(P(+S|+D) > P(+S|-D)\)
• Exacerbating factor (\(+D \rightarrow +S\))
  • \(P(+S|+D \wedge E) > P(+S|+D)\)

You have heard, and it is true:

• Correlation does not imply causation

But it is also true that:

• Causation does imply correlation

What is Differential Diagnosis?

Do you have any examples to share?

At your table - briefly define and discuss:

1. Probability
2. Conditional Probability
3. Inference
4. Causation
5. Rule in / Rule out
6. Diagnosis / Disease / Conditions / Injury
7. Sensitivity / Specificity
8. Likelihood ratios

• Definite cause (\(+D \rightarrow +S\))
  • \(P(+S|+D) = 1\)
• Probable cause (\(+D \rightarrow +S\))
  • \(P(+S|+D) > P(+S|-D)\)
• Exacerbating factor (\(+D \rightarrow +S\))
  • \(P(+S|+D \wedge E) > P(+S|+D)\)

• Etiology (a cause of disease)
  • \(+E \rightarrow +D\)
  • \(P(+D|+E) > P(+D|-E)\) (assuming a probable cause)
• Risk factor (can be an etiology, can cause an etiology, or both (confounding))
  • \(+R \rightarrow +D \vee +E\)
  • \(P(+D \vee +E|+R) > P(+D \vee +E|-R)\) (assuming a probable cause)
• Clinical manifestations
  • Symptoms (what people feel)
  • Signs -Observable (all sensory modalities) manifestations 1. Anatomical changes 2. Physiological changes 3. Behavioral changes (assumed but not able to identify either 1 or 2) 4. Tests and measures are typically of signs

• Causation flows from cause TO effect:
  • \(+D \rightarrow +S\)
• Diagnostic inferences are cause FROM effect (abduction):
  • \(+D \leftarrow +S\)
• So while we start with gathering conditional probability on:
  • \(P(+S|+D)\)
• We really need the conditional probability of:
  • \(P(+D|+S)\)
• And this is only part of the picture, we also need:
  • \(P(-D|-S)\)
• And there are bound to be errors:
  • \(P(-D|+S)\), \(P(+D|-S)\)

From: http://getthediagnosis.org/finding/Ottawa_Knee_Rule.htm

From: http://getthediagnosis.org/diagnosis/Congestive_Heart_Failure.htm#1311

Positive Likelihood Ratio

\(+LR=\dfrac{P(+S|+D)}{1-P(-S|-D)}\)

\(+LR= \dfrac{a(b+d)}{b(a+c)}\) = 13

Negative Likelihood Ratio

\(-LR=\dfrac{1-P(+S|+D)}{P(-S|-D)}\)

\(-LR= \dfrac{c(b+d)}{d(a+c)}\) = 0.8787879

Computes the probability of disease given the sign/symptom (\(P(D|S)\))

\(P(+D|+S)=\dfrac{P(+S|+D)\cdot P(D)}{P(S)}\)

\(P(+D|+S)\)= 0.9285714

Based on:

• Sensitivity (Rate of True Positives): \(P(+S|+D)=\dfrac{a}{a+c}\) = 0.13
• \(P(D)=\dfrac{a+c}{a+c+b+d}\) = 0.5
• \(P(S)=\dfrac{a+b}{a+b+c+d}\) = 0.07

But this based on one sample; and this sample may have an elevated \(P(D)\) and \(P(S)\)

The \(P(D)\) and \(P(S)\) are the "priors" - or "baseline" probabilities of the disease and the sign

\(P(+D|+S)=\dfrac{P(+S|+D)\cdot P(D)}{P(S)}\)

For example, for Small Pox, with S = spots:

• $P(+S|+D)= 0.76
• $P(D)= 0.0011
• $P(S)= 0.08

\(P(+D|+S)\)= 0.01045

But for chicken pox, with S = spots:

• \(P(+S|+D)\)= 0.9
• \(P(D)\)= 0.075
• \(P(S)\)= 0.08

\(P(+D|+S)\)= 0.84375

• Difference between knowledge (justified true belief) when there is a probabilistic justification and when there is a direct observation (the example of knowing someone in China is making a doll right now, or knowing that someone is taking a breath right now even though you do not see it happening)

• Relate to inductive inference and the process of developing general theories about how the world works, of what is true about the world

Inversion of causation -

24% of people having a stroke have difficulty with speech; is not the same and does not mean mean that 24% of people that are having difficulty with speech are having a stroke

expectations about the world - our "baseline" or "apriori" probability

top diseases in terms of prevalence and then the distribution of these conditions among groups of people - looks a lot like "stereotyping"

Sanity - your chance to break away (perhaps unnoticed)

Causal network, structure and reasoning

Develop a causal network and consider the features of its structure in reasoning and making diagnostic inferences.

New Hampshire Department of Health and Human Services

http://www.dhhs.nh.gov/data/

Hungry, possibly Hangry at this point.....

Interpret a set of signs and symptoms; Generate a differential diagnosis

Interpret a set of signs and symptoms in light of possible causes.

Generate a differential diagnosis and subsequent plan of action for complex patient cases across practice settings.

By this point probably not this long - sort of a buffer in the schedule

Red flags that warrant a deeper consideration; Generate a differential diagnosis

Identify red flags that warrant a deeper consideration of alternative explanations for signs and symptoms.

Generate a differential diagnosis and subsequent plan of action for complex patient cases across practice settings.