This talk was presented at the Advances in Precision and Personalized Medicine Workshop.

1. Impact of measurement error on positive controls
Rosa Gini
Agenzia regionale di sanit`a della Toscana, Florence, Italy
Advances in Precision and Personalized Medicine, SAMSI
March 14-15, 2019

2. Disclosure
I am employee of Agenzia regionale di sanit`a della Toscana, a public
agency which conducts pharmacoepi studies compliant with the
Code of Conduct of the European Network of Centres for
Pharmacoepidemiology and Pharmacovigilance
Measurement error is one of the main topics of my research

3. Motivation

4. You take the blue pill, the story ends, you wake
up in your bed and believe whatever you want
to believe. You take the red pill, you stay in
Wonderland, and I show you how deep the
rabbit hole goes. Remember, all I’m offering is
the truth, nothing more.
Morpheus to Neo
Matrix, directed by The Wachowskis

5. Building positive controls
from negative controls
Treated with A Treated with B

6. Building positive controls
from negative controls
Treated with A Treated with B

7. Building positive controls
from negative controls
Treated with A Treated with B
From assumptions we know the observed
difference is an artefact: the true RR is 1

8. Building positive controls
from negative controls
Treated with A Treated with B

9. Building positive controls
from negative controls
Treated with A Treated with B

10. Building positive controls
from negative controls
Treated with A Treated with B

11. Building positive controls
from negative controls
Treated with A Treated with B

12. Building positive controls
from negative controls
Treated with A Treated with B
“If we now add an additional n simulated occurrences
[. . . ] we have doubled the risk”
(Schuemie et al, 2018, PNAS)

13. Building positive controls
from negative controls
Treated with A Treated with B

14. Building positive controls
from negative controls
Treated with A Treated with B
If phenotype in A has low sensitivity, RR of
positive control is lower than 2

15. Building positive controls
from negative controls
Treated with A Treated with B
If phenotype in A has low PPV, RR of
positive control is higher than 2

16. Formula for expected RR
when simulated outcomes are injected at random
1
1.5
2
3
20 30 40 50 60 70 80 90 100 20 30 40 50 60 70 80 90 100 20 30 40 50 60 70 80 90 100
p=1% p=5% p=10%
50%
75%
100%
PPV
RR
Sensitivity (%)
RR = 1 +
1
kv
1 − p
1 − p
kv
p true frequency
po observed frequency
v positive predictive value
k reciprocal of sensitivity
in those treated with A

17. Formula for expected RR
when simulated outcomes are injected at random
1
1.5
2
3
20 30 40 50 60 70 80 90 100 20 30 40 50 60 70 80 90 100 20 30 40 50 60 70 80 90 100
p=1% p=5% p=10%
50%
75%
100%
PPV
RR
Sensitivity (%)
With random injection, RR of positive control
is only 2 if the phenotype of the negative
control has PPV=sensitivity in those treated
with A

18. Questions
Is this RR significantly different from 2 in real
life?

19. Questions
Is this RR significantly different from 2 in real
life?
How can this formula be applied at all, as we
normally do not have p, v, k?

20. Reminder
RR = 1 +
1
kv
1 − p
1 − p
kv
p true frequency
po observed frequency
v positive predictive value
k reciprocal of sensitivity
in those treated with A

21. Reminder
RR = 1 +
1
kv
1 − p
1 − p
kv
p true frequency
po observed frequency
v positive predictive value
k reciprocal of sensitivity
in those treated with A
p = pokv

22. Estimates of validity without a gold standard in
multidatabase studies
OHDSI Europe Symposium. Rotterdam, The Netherlands. March 23rd
-24th
, 2018.Disclosure This research received support from the Innovative Medicines Joint Undertaking under ADVANCE grant agreement Nr: 115557
Interpreting the effect of different concept sets, data domains and data provenances in cohorts from heterogeneous European data
sources: examples of component strategy application from the EMIF and the ADVANCE projects
Rosa Gini1
, Giuseppe Roberto1
, Caitlin Dodd2
, Kaatje Bollaerts3
, Alessandro Pasqua4
, Lars Pedersen5
, Miguel Angel Mayer6
, Ron Herings7
, David Ansell8
, Sulev Reisberg9
, Lara Tramontan10
, Gino Picelli11
, Consuelo Huerta12
, Elisa Martin-Merino12
, Talita Duarte-Salles13
,
Gianfranco Spiteri14
, Emmanouela Sdona14
, Paul Avillach15
, Peter Rijnbeek2
, Miriam Sturkenboom3,16
(1) Osservatorio di Epidemiologia, Agenzia regionale di sanit`a della Toscana, Florence, Italy; (2) Department of Medical Informatics, Erasmus Medical Center, Rotterdam, The Netherlands; (3) P-95, Leuven, Belgium; (4) Health Search, Italian College of General Practitioners and Primary Care, Florence, Italy; (5) Department of Clinical Epidemiology, Aarhus University Hospital, Aarhus, Denmark; (6) Hospital del Mar Medical Research Institute, Barcelona, Spain; (7) PHARMO Institute for Drug Outcomes Research, Utrecht, Netherlands; (8) The Health
Improvement Network, Cegedim Strategic Data Medical Research Ltd, London, UK; (9) Quretec, Software Technology and Applications Competence Center, University of Tartu, Tartu, Estonia; (10) Arsen`al.IT Consortium, Veneto’s Research Centre for eHealth Innovation, Treviso, Italy; (11) Pedianet, Padua, Italy; (12) Agencia Espaola de Medicamentos y Productos Sanitarios, Madrid, Spain; Institut Universitari d’Investigaci´o en Atenci´o Primria (13) SIDIAP database, Primary Care Research Institute Jordi Gol (IDIAP Jordi Gol), Barcelona, Spain; (14)
European Centre for Disease Prevention and Control, Sweden; (15) Department of Biomedical Informatics, Harvard Medical School & Children’s Hospital Informatics Program, Boston Children’s Hospital, Boston, MA, USA; (16) University Utrecht Medical Center, Utrecht, The Netherlands.
Identifying conditions in Europe
In a typical European data source based on
electronic records, data may be only collected
when a patient visits a primary care practice,
or only when patients visit a hospital for
inpatient care. When conducting a
multi-national, multi-database study in Europe,
medical conditions may be identified by
different case identification algorithms. A
process, called component strategy, was
developed and tested in two European
projects: EMIF and ADVANCE
Define component algorithms
Case-finding algorithms are split in simpler algorithms,
each defined by a triple. The Unified Medical Language
System (UMLS) was used to project concept sets to
local terminologies and free text keywords.
Data domain
involved, among
diagnosis, drug,
result from test, . . .
Concept set what
is the meaning of
the information that
is searched?
Data provenance
where was the
information
collected?
Extract, compose and compare
All the data sources extract all the available components and compute the occurrence of
components and of meaningful compositions in the study population. Occurence is
represented in the graphs below which enable comparisons. As an example of comparison:
when a composite algorithm A OR B is represented, the share of the pink bar accounts for the
share of subjects that would not be captured if B was not available: under suitable
assumptions, this provides an estimate of sensitivity of A.
Interpret and decide
Decide which data-source tailored composite
algorithm should be used. Obtain information
on validity. Design sensitivity analyses using
other algorithms.
An acute non-communicable disease:
acute myocardial infarction
AMI is a cardiac emergency, which requires
immediate medical attention and may lead to
death before access to a medical facility. Six data
sources from five European countries participating
in the EMIF project were used. The one year
cumulative incidence of new AMI cases in 2012
among subjects aged 45+, with at least two year of
look-back and no previous record of an AMI
diagnosis, was computed.
Component Concept set Data domain Data provenance
PC diagnosis (AMI) Diagnosis Primary care practice
Specialist diagnosis (AMI) Diagnosis Specialist practice
Inpatient diagnosis (AMI) Diagnosis Hospital
ER diagnosis (AMI) Diagnosis Emergency room
Cause of death (AMI) Diagnosis Death registry Inpatient/ER OR death
PC OR inpatient diagnosis
Cause of death
ER diagnosis
Inpatient diagnosis
Specialist diagnosis
PC diagnosis
0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60
Primary care DB, Spain Hospital DB, Spain Primary care DB, Italy Record linkage DB, Italy Primary care DB, The Netherlands Primary care DB, UK Record linkage DB, Denmark
Left-hand side component only Both components Right-hand side component only
Cumulative incidence of new cases (per 10,000)
Record linkage DBs could extract diagnoses from
inpatient, emergency care and death registry: they
probably captured most of the cases in the
underlying populations, although PPV of
diagnoses may vary according to data
provenance. Variability in cumulative incidence in
primary care DBs was possibly due to different
local recording habits, such as use of free text. In
one of the four primary care DBs, it was possible
to extract diagnoses from inpatient care as well:
assuming consistent PPV of primary and inpatient
care diagnoses, sensitivity of primary care records
was less than 60%.
An infectious disease:
bordetella pertussis
Pertussis is an infectious disease of the respiratory
tract, caused by the bacterium Bordetella
pertussis. Only a specific laboratory tests can
ascertain the specific microorganism responsible
for the infection. National notifications of cases to
the public health authority are reported annually by
EU/EEA countries to the European Centre for
Disease Prevention and Control (ECDC). In five
databases participating in the ADVANCE project
incidence rates per 100,000 person-years during
the year 2012 and during the year 2014 were
computed in the population aged 0-14 years. The
incidence rates of the notified cases in the three
countries were computed from the ECDC
database.
Component Concept set Data domain Data provenance
PC diagnosis, specific (Bordetella pertussis) Diagnosis Primary care practice
PC diagnosis, unspecified(Pertussis) Diagnosis Primary care practice
Laboratory results (Positive result from a
bordetella pertussis test)
Result from test Primary care practice
Symptoms (Symptoms of pertussis) Symptoms Primary care practice
PC diagnosis OR symptoms
Lab OR PC diagnosis
PC specific OR unspecified diagnosis
Laboratory results
PC diagnosis, unspecified
PC diagnosis, specific
0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50
Primary care DB, Spain (A) Primary care DB, Spain (B) Primary care DB, Italy Primary care DB, UK (A) Primary care DB, UK (B)
Left-hand side component only Both components Right-hand side component only
Notification
Incidence rate (per 100,000 person-years)
In three data sources the IRs obtained by
combining cohorts of specific and unspecified
diagnoses were compatible with the notification,
taking into account some misclassification in the
unspecified diagnosis. In one data source the
estimate was much higher, which might indicate
under -notification of suspected cases not finally
confirmed to the relevant public health authority, or
a higher rate of misclassification in the unspecified
diagnosis in this data source. In another database,
a high number of cases was recorded in the
symptoms component, which in this data source
was identified by a very specific string of free text
(‘tos perusoide’) and the pooled estimate was
compatible with the notification. In all the
databases the percentage of cases recorded with
a specific diagnosis was very low. The percentage
of cases confirmed by a record of a positive
diagnostic test was negligible in all the three
databases where this data domain was available.
A chronic disease:
type 2 diabetes mellitus
T2DM is a chronic metabolic disease which needs
diagnostic follow-up and, at a more advanced
stage, regular pharmaceutical treatment with
hypoglycemic drugs. T2DM care is typically
provided by primary care physicians. In six
population-based data sources from four
European countries participating in the EMIF
project prevalence of the cohorts at 1st January
2012 was computed in the adult population (16+).
Component Concept set Data domain Data provenance
PC diagnosis (T2DM) Diagnosis Primary care practice
Inpatient diagnosis (T2DM) Diagnosis Hospital
Laboratory results (Positive result from a
T2DM test)
Result from test Primary care practice
NIAD (Non-insulin antidiabetic
drugs)
Drug Primary care practice or
pharmacy NIAD OR validated algorithm
Validated algorithm
NIAD
Laboratory values
Inpatient diagnosis
PC diagnosis
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9
Primary care DB, Italy Record linkage DB, Italy Primary care DB, The Netherlands Record linkage DB, The Netherlands Primary care DB, UK Record linkage DB, Denmark
Left-hand side component only Both components Right-hand side component only
Prevalence (%)
T2DM diagnoses in primary care could be
extracted from 3 data sources, where a validated
algorithm was also available. Between 54% and
76% of the cases detected by the validated
algorithms were users of NIAD. The record linkage
DBs could only extract diagnoses from inpatient
care, which yielded non-realistically low
prevalences. In those data sources, the
component of NIAD may be used to detect T2DM
cases: the sensitivity may be estimated between
54% and 76% from primary care DBs.
Application
In order to implement the component analysis in the OHDSI ecosystem, information on data provenance should be standardized. In
a first attempt, a simple classification in primary care practice, specialist practice, hospital, emergency room, and death registry, may
lead to interpretable information. The impact of using SNOMED CT instead of UMLS should be assessed. The possibility of
incorporating free text keywords in OHDSI concept sets should be explored in European data sources.
Conclusion
The nature of the disease under study is an important factor in the sensitivity and/or positive predictive value of a component. The
systematic creation and comparison of component-based algorithms could be useful in the OHDSI ecosystem to empower the
validity and efficiency of the data extraction. A systematic approach is needed in OHDSI to address the impact of the phenotypic
definitions in a multi-database setting on study results.

23. An acute disease: acute myocardial infarction
Inpatient/ER OR death
PC OR inpatient diagnosis
Cause of death
ER diagnosis
Inpatient diagnosis
Specialist diagnosis
PC diagnosis
0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60
Primary care DB, Spain Hospital DB, Spain Primary care DB, Italy Record linkage DB, Italy Primary care DB, The Netherlands Primary care DB, UK Record linkage DB, Denmark
Left-hand side component only Both components Right-hand side component only
Cumulative incidence of new cases (per 10,000)

24. An acute disease: acute myocardial infarction
Inpatient/ER OR death
PC OR inpatient diagnosis
Cause of death
ER diagnosis
Inpatient diagnosis
Specialist diagnosis
PC diagnosis
0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60
Primary care DB, Spain Hospital DB, Spain Primary care DB, Italy Record linkage DB, Italy Primary care DB, The Netherlands Primary care DB, UK Record linkage DB, Denmark
Left-hand side component only Both components Right-hand side component only
Cumulative incidence of new cases (per 10,000)

25. An acute disease: acute myocardial infarction
Inpatient/ER OR death
PC OR inpatient diagnosis
Cause of death
ER diagnosis
Inpatient diagnosis
Specialist diagnosis
PC diagnosis
0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60
Primary care DB, Spain Hospital DB, Spain Primary care DB, Italy Record linkage DB, Italy Primary care DB, The Netherlands Primary care DB, UK Record linkage DB, Denmark
Left-hand side component only Both components Right-hand side component only
Cumulative incidence of new cases (per 10,000)

26. An acute disease: acute myocardial infarction
Inpatient/ER OR death
PC OR inpatient diagnosis
Cause of death
ER diagnosis
Inpatient diagnosis
Specialist diagnosis
PC diagnosis
0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60
Primary care DB, Spain Hospital DB, Spain Primary care DB, Italy Record linkage DB, Italy Primary care DB, The Netherlands Primary care DB, UK Record linkage DB, Denmark
Left-hand side component only Both components Right-hand side component only
Cumulative incidence of new cases (per 10,000)
observed: po = .0040
po = .0054
% = 89%
po = .0043
% = 85%

27. An acute disease: acute myocardial infarction
Inpatient/ER OR death
PC OR inpatient diagnosis
Cause of death
ER diagnosis
Inpatient diagnosis
Specialist diagnosis
PC diagnosis
0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60
Primary care DB, Spain Hospital DB, Spain Primary care DB, Italy Record linkage DB, Italy Primary care DB, The Netherlands Primary care DB, UK Record linkage DB, Denmark
Left-hand side component only Both components Right-hand side component only
Cumulative incidence of new cases (per 10,000)
observed: po = .0040
po = .0054
% = 89%
po = .0043
% = 85%
assumed:
v = 1
k = 1
v = 1
k = 1
v = 1
k = 1
.87

28. An acute disease: acute myocardial infarction
Inpatient/ER OR death
PC OR inpatient diagnosis
Cause of death
ER diagnosis
Inpatient diagnosis
Specialist diagnosis
PC diagnosis
0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60
Primary care DB, Spain Hospital DB, Spain Primary care DB, Italy Record linkage DB, Italy Primary care DB, The Netherlands Primary care DB, UK Record linkage DB, Denmark
Left-hand side component only Both components Right-hand side component only
Cumulative incidence of new cases (per 10,000)
observed: po = .0040
po = .0054
% = 89%
po = .0043
% = 85%
assumed:
v = 1
k = 1
v = 1
k = 1
v = 1
k = 1
.87
computed: p = .0046

29. An acute disease: acute myocardial infarction
Inpatient/ER OR death
PC OR inpatient diagnosis
Cause of death
ER diagnosis
Inpatient diagnosis
Specialist diagnosis
PC diagnosis
0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60 0 10 20 30 40 50 60
Primary care DB, Spain Hospital DB, Spain Primary care DB, Italy Record linkage DB, Italy Primary care DB, The Netherlands Primary care DB, UK Record linkage DB, Denmark
Left-hand side component only Both components Right-hand side component only
Cumulative incidence of new cases (per 10,000)
observed: po = .0040
po = .0054
% = 89%
po = .0043
% = 85%
assumed:
v = 1
k = 1
v = 1
k = 1
v = 1
k = 1
.87
computed: p = .0046
obtained RR: 1.9

30. A chronic disease: type 2 diabetes mellitus
NIAD OR validated algorithm
Validated algorithm
NIAD
Laboratory values
Inpatient diagnosis
PC diagnosis
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9
Primary care DB, Italy Record linkage DB, Italy Primary care DB, The Netherlands Record linkage DB, The Netherlands Primary care DB, UK Record linkage DB, Denmark
Left-hand side component only Both components Right-hand side component only
Prevalence (%)

31. A chronic disease: type 2 diabetes mellitus
NIAD OR validated algorithm
Validated algorithm
NIAD
Laboratory values
Inpatient diagnosis
PC diagnosis
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9
Primary care DB, Italy Record linkage DB, Italy Primary care DB, The Netherlands Record linkage DB, The Netherlands Primary care DB, UK Record linkage DB, Denmark
Left-hand side component only Both components Right-hand side component only
Prevalence (%)

32. A chronic disease: type 2 diabetes mellitus
NIAD OR validated algorithm
Validated algorithm
NIAD
Laboratory values
Inpatient diagnosis
PC diagnosis
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9
Primary care DB, Italy Record linkage DB, Italy Primary care DB, The Netherlands Record linkage DB, The Netherlands Primary care DB, UK Record linkage DB, Denmark
Left-hand side component only Both components Right-hand side component only
Prevalence (%)

33. A chronic disease: type 2 diabetes mellitus
NIAD OR validated algorithm
Validated algorithm
NIAD
Laboratory values
Inpatient diagnosis
PC diagnosis
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9
Primary care DB, Italy Record linkage DB, Italy Primary care DB, The Netherlands Record linkage DB, The Netherlands Primary care DB, UK Record linkage DB, Denmark
Left-hand side component only Both components Right-hand side component only
Prevalence (%)
observed:
po = .088
% = 55%
po = .064
po = .067
% = 77%
po = .041
po = .078
% = 53%
po = .042

34. A chronic disease: type 2 diabetes mellitus
NIAD OR validated algorithm
Validated algorithm
NIAD
Laboratory values
Inpatient diagnosis
PC diagnosis
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9
Primary care DB, Italy Record linkage DB, Italy Primary care DB, The Netherlands Record linkage DB, The Netherlands Primary care DB, UK Record linkage DB, Denmark
Left-hand side component only Both components Right-hand side component only
Prevalence (%)
observed:
po = .088
% = 55%
po = .064
po = .067
% = 77%
po = .041
po = .078
% = 53%
po = .042
assumed:
v = 1
k = 1
v = 1
k = 1
v = 1
k = 1
v = 1
k = 1
.62
v = 1
k = 1
.62
v = 1
k = 1
.62

35. A chronic disease: type 2 diabetes mellitus
NIAD OR validated algorithm
Validated algorithm
NIAD
Laboratory values
Inpatient diagnosis
PC diagnosis
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9
Primary care DB, Italy Record linkage DB, Italy Primary care DB, The Netherlands Record linkage DB, The Netherlands Primary care DB, UK Record linkage DB, Denmark
Left-hand side component only Both components Right-hand side component only
Prevalence (%)
observed:
po = .088
% = 55%
po = .064
po = .067
% = 77%
po = .041
po = .078
% = 53%
po = .042
assumed:
v = 1
k = 1
v = 1
k = 1
v = 1
k = 1
v = 1
k = 1
.62
v = 1
k = 1
.62
v = 1
k = 1
.62
computed: p = .103 p = .067 p = .067

36. A chronic disease: type 2 diabetes mellitus
NIAD OR validated algorithm
Validated algorithm
NIAD
Laboratory values
Inpatient diagnosis
PC diagnosis
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9
Primary care DB, Italy Record linkage DB, Italy Primary care DB, The Netherlands Record linkage DB, The Netherlands Primary care DB, UK Record linkage DB, Denmark
Left-hand side component only Both components Right-hand side component only
Prevalence (%)
observed:
po = .088
% = 55%
po = .064
po = .067
% = 77%
po = .041
po = .078
% = 53%
po = .042
assumed:
v = 1
k = 1
v = 1
k = 1
v = 1
k = 1
v = 1
k = 1
.62
v = 1
k = 1
.62
v = 1
k = 1
.62
computed: p = .103 p = .067 p = .067
obtained RR: 2.0 1.6 2.0 1.6 2.0 1.6

37. An infectious disease: pertussis
PC diagnosis OR symptoms
Lab OR PC diagnosis
PC specific OR unspecified diagnosis
Laboratory results
PC diagnosis, unspecified
PC diagnosis, specific
0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50
Primary care DB, Spain (A) Primary care DB, Spain (B) Primary care DB, Italy Primary care DB, UK (A) Primary care DB, UK (B)
Left-hand side component only Both components Right-hand side component only
Notification
Incidence rate (per 100,000 person-years)

38. An infectious disease: pertussis
PC diagnosis OR symptoms
Lab OR PC diagnosis
PC specific OR unspecified diagnosis
Laboratory results
PC diagnosis, unspecified
PC diagnosis, specific
0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50
Primary care DB, Spain (A) Primary care DB, Spain (B) Primary care DB, Italy Primary care DB, UK (A) Primary care DB, UK (B)
Left-hand side component only Both components Right-hand side component only
Notification
Incidence rate (per 100,000 person-years)

39. An infectious disease: pertussis
PC diagnosis OR symptoms
Lab OR PC diagnosis
PC specific OR unspecified diagnosis
Laboratory results
PC diagnosis, unspecified
PC diagnosis, specific
0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50
Primary care DB, Spain (A) Primary care DB, Spain (B) Primary care DB, Italy Primary care DB, UK (A) Primary care DB, UK (B)
Left-hand side component only Both components Right-hand side component only
Notification
Incidence rate (per 100,000 person-years)

40. An infectious disease: pertussis
PC diagnosis OR symptoms
Lab OR PC diagnosis
PC specific OR unspecified diagnosis
Laboratory results
PC diagnosis, unspecified
PC diagnosis, specific
0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50
Primary care DB, Spain (A) Primary care DB, Spain (B) Primary care DB, Italy Primary care DB, UK (A) Primary care DB, UK (B)
Left-hand side component only Both components Right-hand side component only
Notification
Incidence rate (per 100,000 person-years)
observed:
po = .00040
No = .00021
po = .00017
No = .00021
po = .00044
No = .00005
po = .00018
No = .00013
po = .00022
No = .00013

41. An infectious disease: pertussis
PC diagnosis OR symptoms
Lab OR PC diagnosis
PC specific OR unspecified diagnosis
Laboratory results
PC diagnosis, unspecified
PC diagnosis, specific
0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50
Primary care DB, Spain (A) Primary care DB, Spain (B) Primary care DB, Italy Primary care DB, UK (A) Primary care DB, UK (B)
Left-hand side component only Both components Right-hand side component only
Notification
Incidence rate (per 100,000 person-years)
observed:
po = .00040
No = .00021
po = .00017
No = .00021
po = .00044
No = .00005
po = .00018
No = .00013
po = .00022
No = .00013
assumed:
k = 1
p = No
k = 1
p = No
k = 1
p = No
k = 1
p = No
k = 1
p = No

42. An infectious disease: pertussis
PC diagnosis OR symptoms
Lab OR PC diagnosis
PC specific OR unspecified diagnosis
Laboratory results
PC diagnosis, unspecified
PC diagnosis, specific
0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50
Primary care DB, Spain (A) Primary care DB, Spain (B) Primary care DB, Italy Primary care DB, UK (A) Primary care DB, UK (B)
Left-hand side component only Both components Right-hand side component only
Notification
Incidence rate (per 100,000 person-years)
observed:
po = .00040
No = .00021
po = .00017
No = .00021
po = .00044
No = .00005
po = .00018
No = .00013
po = .00022
No = .00013
assumed:
k = 1
p = No
k = 1
p = No
k = 1
p = No
k = 1
p = No
k = 1
p = No
× ×
× ×

43. An infectious disease: pertussis
PC diagnosis OR symptoms
Lab OR PC diagnosis
PC specific OR unspecified diagnosis
Laboratory results
PC diagnosis, unspecified
PC diagnosis, specific
0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50
Primary care DB, Spain (A) Primary care DB, Spain (B) Primary care DB, Italy Primary care DB, UK (A) Primary care DB, UK (B)
Left-hand side component only Both components Right-hand side component only
Notification
Incidence rate (per 100,000 person-years)
observed:
po = .00040
No = .00021
po = .00017
No = .00021
po = .00044
No = .00005
po = .00018
No = .00013
po = .00022
No = .00013
assumed:
k = 1
p = No
k = 1
p = No
k = 1
p = No
k = 1
p = No
k = 1
p = No
× ×
× ×
computed: v = .53 v = .76 v = .60

44. An infectious disease: pertussis
PC diagnosis OR symptoms
Lab OR PC diagnosis
PC specific OR unspecified diagnosis
Laboratory results
PC diagnosis, unspecified
PC diagnosis, specific
0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50
Primary care DB, Spain (A) Primary care DB, Spain (B) Primary care DB, Italy Primary care DB, UK (A) Primary care DB, UK (B)
Left-hand side component only Both components Right-hand side component only
Notification
Incidence rate (per 100,000 person-years)
observed:
po = .00040
No = .00021
po = .00017
No = .00021
po = .00044
No = .00005
po = .00018
No = .00013
po = .00022
No = .00013
assumed:
k = 1
p = No
k = 1
p = No
k = 1
p = No
k = 1
p = No
k = 1
p = No
× ×
× ×
computed: v = .53 v = .76 v = .60
obtained RR: 2.9 2.3 2.7

45. Wrap up

46. Wrap up

47. Wrap up

48. Wrap up

49. Wrap up
Thanks