GP 4001 Lecture Series
2006-2007
3. Dealing with undifferentiated
problems in primary care II
Learning Outcomes for this
course - I
• Develop a rapport with patients such that
patients are at ease in discussing their health
problem(s) (comm)
• Gather appropriate information on the
patient’s health problem(s) including
information on the patient’s own perspective
on the problem(s). (udp, comm)
• Generate a reasonable range of diagnostic
possibilities for undifferentiated medical
problems presented by patients (udp)
• Investigate these diagnostic possibilities
using appropriately focused history taking
and selective physical examination (udp,
comm)
Learning Outcomes for this
course - II
• Construct a general model for the safe and
effective management of patients with
multiple and long term health problems
(cdm)
• Adapt this model to the long term health
problems commonly encountered by doctors
(cdm)
• Construct an appropriate and feasible
management plan to deal with the physical,
psychological and social aspects of patient’s
problem(s) (udp, cdm)
• Negotiate this plan with the patient. (comm)
Characteristics of tests –
Sensitivity and Specificity
• Sensitivity of a test is the proportion of
patients who test positive for the
disease who actually have the disease
• Specificity of a test is the proportion of
the patients who test negative for the
disease who actually do not have the
disease
Sensitivity and Specificity
TARGET DISORDER
DIAGNOSTIC
TEST
RESULT
PRESENT
ABSENT
+
a
b
a+b
-
c
d
c+d
a+c
b+d
a+b+c+d
Sensitivity = a/(a+c) Specificity = (d/b+d)
Bayes’ Theorem and the
characteristics of tests
• Let us suppose we have a ‘test’ for the
presence or absence of URTI in a population
35 year old women with 3 day history of
cough, temperature and green sputum
• Let us suppose in this population the
prevalence of URTI is 80% (4 in 5)
• Let us suppose for this test the sensitivity is
90% and the specificity is 90%
• Let us say ‘having a runny nose’ is the test
True and false positives
True and false negatives
• How many people will test positive who have
the disease (true positives)
• How many people will test positive who do
not have the disease (false positives)?
• How many people will test negative who do
not have the disease (true negatives)?
• How many people will test negative who do
have the disease (false negatives)?
Answers
Out of 100
Total number of cases
80
Total number of non-cases
20
Total positives
74
Total negatives
26
True positives
72
False positives
2
True negatives
18
False negatives
8
Conclusions
• ‘Having a runny nose’ is a pretty good
predictor of having an URTI (in a patients
with other relevant features)
• Not having a runny nose, however, is not
such a reliable indicator of not having an
URTI
• In the jargon of clinical epidemiology
• This ‘test’ has a good positive predictive value
• But has poor negative predictive value
Another example to show the impact
of prevalence on predictive value
• Let us suppose we have a ‘test’ for the presence
or absence of pneumonia in a population 35 year
old women with 3 week history of cough,
temperature and green sputum
• Let us suppose in this population the prevalence
of pneumonia is 20% (1 in 5)
• Let us suppose for this test the sensitivity is 90%
and the specificity is 90%
• Let us suppose the presence or absence of basal
crepitations on lung auscultation is the ‘test’
True and false positives
True and false negatives
• How many people will test positive who have
the disease (true positives)
• How many people will test positive who do
not have the disease (false positives)?
• How many people will test negative who do
not have the disease (true negatives)?
• How many people will test negative who do
have the disease (false negatives)?
Answers
Out of 100
Total number of cases
20
Total number of non-cases
80
Total positives
26
Total negatives
74
True positives
18
False positives
8
True negatives
72
False negatives
2
Conclusions
• For a test with a fairly typical sensitivity and
specificity (i.e. 90% and 90%) we can say:• When the disease has a high prevalence we tend
to have more false negatives and fewer false
positives
• When the disease is less prevalent we tend to have
more false positives and fewer false negatives
• The ‘performance’ of a test (in terms of its
ability to tell cases from non-cases) is
critically dependent on the prevalence of the
condition in the population (even if it has
good sensitivity and specificity)
Predictive values of tests
• The positive predictive value (PPV) is
the proportion of patients with positive
test results who are correctly
diagnosed.
• The negative predictive value (NPV) is
the proportion of patients with negative
test results who are correctly diagnosed
Positive and negative
predictive values
TARGET DISORDER
DIAGNOSTIC
TEST
RESULT
PRESENT
ABSENT
+
a
b
a+b
-
c
d
c+d
a+c
b+d
a+b+c+d
PPV = a/(a+b)
NPV = d/(c+d)
URTI Example
URTI
Runny nose
PRESENT
ABSENT
+
72
2
74
-
8
18
26
80
20
100
PPV = 72/74 =.97 NPV = 18/26=.69
Pneumonia Example
Pneumonia
Basal
crepitations
PRESENT
ABSENT
+
18
8
26
-
2
72
74
20
80
100
PPV = 18/26=.69 NPV = 72/74 =.97
The positive predictive value for some values
of prevalence, sensitivity and specificity
Prevalence (%) Sensitivity and specificity (%)
99
95
90
80
20
96.1
82.1
69.2
50.0
10
91.7
67.9
50.0
30.8
5
83.9
50.0
32.1
17.4
1
50.0
16.1
8.3
3.9
0.1
9.0
8.7
4.3
2.0
What does this all mean for
making diagnoses in general
• Knowing the prevalence of disease in a
population is a very important
consideration in making good diagnoses
• The prevalence is essentially the same as
‘prior probability’ in the Bayesian model
• Every bit of additional information we get
about a patient functions as if it were a
‘test’ and has:• Sensitivity
• Specificity
• Positive and negative predictive values
Implications for general practice
• Where prevalence (the prior probability) is high
‘tests’ have good positive predictive value but
poor negative predictive value
• Where prevalence (the prior probability) is low
(the typical situation in general practice) tests
generally have poor positive predictive value
but better negative predictive value
• In general practice we tend to be better able to
rule diseases ‘out’
• In hospital we tend to be better able to rule
diseases ‘in’
Ruling in and ruling out
• Hospital doctors focus on ruling disease
in – i.e. establishing the presence of
disease/ confirming a diagnosis
• GPs focus on ruling disease out – i.e.
establishing the absence of disease/
refuting a diagnosis
Types of error
• Type 1 – accepting the null hypothesis
when it ought to have been rejected
i.e. missing a disease in a patient who has
one
• Type 2 – rejecting the null hypothesis
when it ought to have been accepted
i.e. diagnosing disease in a patient who
does not have one
Who makes what kind of error?
• Type 1 – accepting the null hypothesis when
it ought to have been rejected
i.e. missing a disease in a patient who has one
GPs
• Type 2 – rejecting the null hypothesis when it
ought to have been accepted
i.e. diagnosing a disease in a patient who
does not have one
Hospital doctors
Mary had a little cough
Mary had a little cough
There was a lot of it about
Does she have pneumonia?
That’s rare, so we can usually rule it out
But what is the diagnosis?
Whatever could it be?
To say for sure, to rule it in
That’s not the task of her GP
“It is very difficult to make
an accurate prediction,
especially about the future."