An actual framework for clinical reasoning
To prove or not to prove, that is the question... Be all my errors re e er d." J. Van den Ende, IMTA
Aristoteles • Categorico-deductive – All humans are mortal – Socrates is human – Thus: Socrates is mortal
• Hypothetico-inductive – Socrates, Plato, Aristotoles and Toon Hermans are mortal – They are human – Hence: Humans are mortal
Popper: hypothetico-deductive • H potheti o a du ti e – Humans are mortal – Socrates is mortal – Thus : Is Socrates human???
• In human medicine: – TB patients cough – Yakasi coughs. – Does Yakasi has got TB?
Reverend Bayes • Pretest – +
• Sensitivity – +
• False positives – =
• Post test Probability
“et theor : of set theor , li era os do i e
Claude Elwood Shannon: information theory
• Entropy= uncertainty • Entropy= chaos • Chaos=molecule movements • Molecule movements=temperature
Lowering entropy in physics
7
Entropy representation: maximum at 50%
Entropy as function of pretest
pretest 0,001
0,09
1
entropy
0,01 0,5 0,99 0,999
1,2
post test prob 0,01
0,8 pretest
0,6
pure entropy
0,4
0,91 1,00 1,00
0,2 0 1
2
3 steps
4
5 8
Entropy change: not illuminating
Entropy change
pretest
0,001 0,01 0,5 0,99 0,999
post test prob
1 0,8 0,6
pretest
0,4
entropy change
0,2
0,01 0,09 0,91 1,00 1,00
0 1
2
3
4
5
9
Entropy of a set of diseases
10
Entropy of a set of diseases 2
11
Entropy: Conclusions • • • •
Entropy study for a set of hypothesis No value for a single hypothesis Uncertainty maximum at 0.5 Log10odds much more intuitive for clinicians
12
Kahneman: thinking fast and slow
• Heuristics – – – –
Availability Confirmation Anchoring Hindsight
• Thinking fast: pattern recognition • Thi ki g slo : true reaso i g
The concept of threshold: Balance between utility of treating vs not treating
Threshold The threshold
Probability of disease
sure
Enough evidence to treat
DECISION THRESHOLD
Not enough evidence to treat
impossible
Factors factors influencing the thresholds
Probability of disease
sure
impossible
treatment risk and cost
disease seriousness, contagiousness
availability and effectiveness of treatment
Sackett
• Predictive values : useless for the clinician • Use likelihood ratios • Use Fagan nomogram • Apply threshold
The Antwerp framework for clinical reasoning
a history of three rebellions and of serendipity (Rie de Ridder) and a lot of intellectual fun.
I the earl
i eties …
• Prof. Gigase: tour in Western-Africa – Depressed – Average diagnostic spectrum 12 diseases – Organize some "thinking lessons"
• Formal and applied course in clinical epidemiology. TB positives EXAMEN +
TB negatives
0,10
EXAMEN -
12,0 99
12
1
88
0,001
87,9
0,10
99,9
A bridge between clinical epidemiology and clinical reasoning • Clinical practice and clinical epidemiology – same phenomena – no bridge – no common language.
• cross-fertilization • resulting in a new model for clinical logic.
Bayes for 1 argument
TB positives EXAMEN +
TB negatives
0.10
EXAMEN -
33.0 99
33
1
67
0.001
66.9
0.10
99.9
Bayes for n arguments
Rebellion I • Students: – "What is all this athe ati al ru ish for! – Gi e us so e real-life e a ples!
• Course closed.
But: students were right!
• Foreign language – English – Not clinical
• Where to find LR? • Negative LR difficult: 0.03 ?? • Convert probability to odds !!
New language
• li i ia -frie dl la guage • L‘+ = o fir i g po er • LR- = e ludi g po er • pre-test probability = li i al suspi io , roo ha e • post-test pro a ilit = ertai t le el
aiti g
Rebellion II • Nobody less than Paul Janssen – eminent founder of Janssen pharmaceuticals – You are o pletel ro g – Human mind thinks in categories, not in continuous numbers
• Ho to ultipl ategories to at h Ba es’ theore ? • That’s our pro le , ot i e!
Answer: • Work in logarithms of likelihood ratios • Use u likelihood ratio : -10 ~ 0.1 • Round • Add in stead of multiplying
Turing,
• father of the term artifi ial i tellige e • father of the ROC curve • same idea • never published • gone our great discovery!
Add in stead of multiplying
pre-test odds
x likelihood ratio = post-test odds
log10 pre-test odds
+ log10 LR = log10 post-test odds
Work with cathegories, rounded values
clinical class pre-test+ log10 LR
= clinical class post-test
clinical class pre-test+ power class
= clinical class post-test
The Feinman-Tufte principle
• What can be shown in
a graph, • should not be explained in text.
Power of arguments
Diseases Probability
certain
very strong argument
probable
possible
less probable
impossible
weak argument
good argument
strong argument
Advantages • • • • • •
Added value constant Symmetry Multiple tests Extremities of scale visible Multiple diseases OR
Effect of a positive test on disease probability, on a linear scale 100 90 80 70 60 50 40 30 20 10 0
student prostitute
waiting room
HIV test pos
Difference between added value of a positive HIV elisa 99.9
3
99
2
90
1
50
0
student
10
-1
1
-2
0.1
-3
0.01
-4
waiting room
HIV test +
prostitute
Asymmetry 2
1
Log10 odds
0
Pretest
Thick film
confirmer
-1
excluder -2
-3
-4
Consecutive steps
Asymmetry 2
1.5
Log10 odds
1
confirmer
0.5
0
Pretest
Tertian fever
-0.5
-1
-1.5
-2
Consecutive steps
excluder
di m
w
sp in dl e
x
fe ve r
ia
an gu di sc la t io it i sn os te ol ys is
vi ty
ro om pa ra pl eg
ng
se ns it i
ai ti
in is he d
pr ev al en ce
Probability evolution
99.99 4
99.9 3
99 2
90 1
50 0
10 -1
1 -2
0.1 -3
0.01 -4
Extremities of scale 100 90 80 70 60 50 40 30 20 10 0
Odds ratio • Total discriminative value – Positive LR + Negative LR – Pos LR / (inverse of negative LR) – 10/0.1
• Show it on a graph
Odds ratio 2
1.5
1
Odds Ratio
Log10 odds
0.5
confirmer
0
Pretest
Tertian fever
-0.5
-1
-1.5
-2
Consecutive steps
excluder
A LGORI THM E RA M IFI A NT DY SPNEE DE L'A DULTE
inspiratoire
globale expiratoire
fièvre voix rauque
fièvre
asthme
goit re fièvre
toux
orthopnée
eosinophilie massive goit re plongeant
corps etranger pneumonie
fausses membranes
diphtérie
katayama
laryngite
signes de phlébite
râles crépit ants pieds gonflés
bronchite, bronchiolite
insuff isance cardiaque
embolie pulmonaire
autres causes
Rebellion III
• Dr Jansen of the Damian Foundation – algorith i thi ki g is ot hu a • Answer: – propose a other logi al fra e . • Jansen: – this is our pro le , ot i e
Lumbago in Verona • Algorithms – Serial – Flip-flop, dichotomous
• Human brain – Parallel – Fuzzy logic
• City of Romeo and Juliet • Tablecloth: the solution
The idea of the diagnostic landscape or panorama • Avoid the tunnel trap – Go for serious and treatable diseases first – Do not stop at the first plausible pattern
• Look for key findings – Confirmers – Excluders
Typhoid fever Borreliosis Acute glaucoma Malaria, acute
Leptospirosis
acute headache woman
Arteritis temporalis
Bact. meningitis
CO-Intoxication Mastoiditis Pre-eclampsia
Hypertensive crisis
Typhoid fever recurrent exclusion diagnosis, leucopenia, Abdominal symptoms
Borreliosis thick smear, Splenomegaly pallor
Acute glaucoma
Fever
Malaria, acute red eye (blue ring), pain, vision disturbance
acute headache woman
Leptospirosis Exposure, jaundice, Haemorrhages, proteinuria
Arteritis temporalis
Neck stiffness
Mastoiditis
CO-Intoxication proteinuria blood pressure
local pain, ear infection
Unilateral, ESR
No fever
Bact. meningitis
Pre-eclampsia
Exposure, red face, heating in house
Hypertensive crisis
Viral Infection, incl Arbo, Hanta, Lassa
heatstroke
Typhoid fever
recurrent
Sinusitis, acute exclusion diagnosis, leucopenia, Abdominal symptoms
Borreliosis thick smear, Splenomegaly pallor
Fever
Acute glaucoma
Malaria, acute
acute headache woman
Leptospirosis Exposure, jaundice, Haemorrhages, proteinuria
red eye (blue ring), pain, vision disturbance
migraine
Arteritis temporalis
Viral meningitis Unilateral, ESR
No fever
Bact. meningitis Neck stiffness
CO-Intoxication Exposure, red face, heating in house
proteinuria
Mastoiditis
blood pressure local pain, ear infection
Pre-eclampsia
cluster headache tension headache
Hypertensive crisis
trauma
Acute renal failure Hemorrhage, intracerebral, subdural, epidural
Zona
Future I • Train students in clinical reasoning • Train trainers: university professors • Through thorough remodeling of the logic – More patients can be cured – For much lower expenses. – In developing countries of utmost importance !
Future II • Evidence of usefulness of the model • Fundamental research in – Thresholds – Powers – Logic
Thanks Al a s make the audience suffer as much as possi le . Alfred Hitchkock