Ó The Author 2008. Published by Oxford University Press. All rights reserved. For permissions, please e-mail: journals.permissions@oxfordjournals.org.

doi:10.1093/fampra/cmn028 Family Practice Advance Access published on 17 June 2008

The coming of age of ICPC: celebrating the 21st

birthday of the International Classification of
Primary Care
Jean-Karl Solera, Inge Okkesb, Maurice Woodc and Henk Lambertsb

Soler J-K, Okkes I, Wood M and Lamberts H. The coming of age of ICPC: celebrating the 21st
birthday of the International Classification of Primary Care. Family Practice 2008; 25: 312–317.
The International Classification of Primary Care (ICPC) has, since its introduction in 1987, been
quite successful. Now in its second revised version, it has been translated in 22 languages,
accepted by the World Health Organization (WHO) as a member of the Family of International
Classifications, and is being widely used both in routine daily practice and in research.
In this contribution, it is explained that ICPC was designed as a theoretical classification, and that
it has especially great potential when used (1) supported by the ICPC2/ICD10 Thesaurus, (2) in
sufficiently large studies to allow all classes to be observed often enough to provide reliable
data, and (3) in studies based on data on episodes of care, rather than encounter data only. Under
these conditions, the likelihood ratios of symptoms given a diagnosis, and of co-morbidity be-
come available, which define the clinical content of family practice.
Keywords. ICPC, episode of care, reason for encounter, prior and posterior probabilities.

Introduction Malta and Serbia) and also in encounter studies

In 1987, the International Classification of Primary
Care (ICPC) was published as a tool to order the do-
main of family practice. It was empirically designed,
from family medicine data, to appropriately classify
and define relationships between events across the
whole breadth of the discipline by using the concept
of episode of care. An episode of care, as distinct from
an episode of illness or disease, is a health problem or
disease from its first presentation to the health care
provider to the last presentation for the same prob-
lem.1,2 As a theoretical classification, several of its
characteristics (Box 1) were quite distinct from the
dominant perspectives on the content of family prac-
tice at that time and constituted a true paradigm shift
in many ways.2–5 Notwithstanding initial misunder-
standings, it grew up handsomely and became quite
successful.3 Translated in 22 languages, accepted by
the World Health Organization (WHO) as a member
of the Family of International Classifications (WHO/
FIC), widely used for the routine collection of data
on episodes of care (The Netherlands, Japan, Poland,
(Australia, Norway, Denmark and The Netherlands),
it is now supported by a large empirical database.6 In
the index of family practice, the search term ‘ICPC’
refers to over 100 articles (since 1984); in contrast,
‘Read Codes’ as a search term refers to 39 articles.
This trend is also seen in PubMed (108 against 26).
ICPC reached ‘adulthood’ with ICPC-2-Revised,
published in 2005 by Oxford University Press, present-
ing the latest revision of ICPC with inclusion and ex-
clusion criteria and its mapping to ICD-10.7 This
publication also included a companion CD8 containing
this version of ICPC in electronic form, an applied ep-
idemiological retrieval programme with data from the
Dutch Transition Project ‘Episodes of Care in Family
Practice (EFP)’,9 the ICPC2-ICD-10 Thesaurus,10
a Glossary of terms and an ICPC Tutorial.
Although this is reason for satisfaction, it appears
that on occasions two of the characteristics of ICPC
(Box 1) have been misunderstood: namely the frequency
requirement for classes in ICPC and the significant ad-
ditional utility of the use of episodes of care to structure
data versus that of encounter-based data.6,11–13

Received 9 October 2007; Revised 29 March 2008; Accepted 29 April 2008.

a
Visiting Professor, Institute of Postgraduate Medicine and Primary Care, Faculty of Life and Health Sciences, University of
Ulster, Coleraine BT52 1SA, UK, bformerly Department of Family Medicine, Academic Medical Center, University of
Amsterdam, Amsterdam, The Netherlands and cformerly Department of Family Medicine, Virginia Commonwealth University,
Richmond, VA, USA. Correspondence to Jean-Karl Soler, The Family Practice, Bay Street (Triq ir-Rand), Attard ATD 1300,
Malta; Email: info@thefamilypractice.com.mt

312
 The coming of age of ICPC
BOX 1 The main characteristics of the ICPC1–5
Its purpose is to order the domain of family practice in the format
of episodes of care.
It provides a single terminology for the patient’s RFE and the
family physicians diagnosis, thus representing both sides of the
same coin.
It captures the changes (transitions) in the content of episodes of
care over time.
It follows strict taxonomic rules, and so its classes are mutually
exclusive.
It offers–if possible–one class for common (occurring >1 per 1000
patient years) reasons for encounter and diagnoses. Less common
classes are included in ‘ragbags’.
Its biaxial structure (chapters for body systems/problem areas and
components identical throughout all chapters) results in three-
digit mnemonic, alphanumeric codes.
Its reliability and validity are supported by its coding rules and
a growing comparative international database.
In the coding process, localization takes precedence over aetiology.
Symptom diagnoses take precedence over disease diagnoses that
are uncertain (i.e. do not fulfil the inclusion criteria).
It does not cater for mind-body metaphors: ‘psychosomatic’ and
‘somatoform’ disorders are not included.
Frequency of observations
In comparison with mortality and hospital data with
a large denominator, ICPC data from family practice
are often derived from a relatively small number
of patients.5,6 ICPC produces an effective reduction
of primary care data, as it has only 684 diagnostic
classes (a code or rubric in ICPC which defines one
concept, a symptom or sign, intervention or disease),
mainly selected on the basis of real frequencies of oc-
currence in daily family practice. Still, problems may
arise in the interpretation of such data in comparative
studies.
Let us consider a practical example. In countries
with well-defined practice populations, a family doctor
(FD) might care for around 2000 listed patients. In 1
year, she would thus collect data on approximately
1400 complete patient years.14–17 A group of FDs may
deal with ten-fold (14 000 patient years), while a large
multi-site study might provide data on 140 000 patient
years (from, say, 100 practices).18–20
In Table 1, the distribution of prevalences (a table
of the annual prevalences of each component 7 (dis-
ease label) ICPC class, calculated on the basis of the
three populations listed in the previous paragraph) of
diagnostic ICPC classes is given for each of these three
denominators, together with the width of their 95%
confidence intervals (CIs). This is the preferred pre-
sentation of rates under the assumption that the data
are unbiased.21,22 In reliability studies in family prac-
tice, the error rate (including missing data) is reported
313
to vary between 10% and 30%.23–27 In EFP,9 the error
rate, including missing data, has repeatedly been es-
tablished as approximately 5%.28–30 Infrequent diag-
noses included in ragbags (a class which includes
symptoms or diseases with a low prevalence, not else-
where classified), however, appear to have a relatively
low reliability.29 A prevalence rate which is not larger
than the width of its own CI cannot be interpreted,
since the observation is lost in the ‘noise’ of its own
observation.31,32 Considering that ICPC covers the do-
main of family medicine, it would be desirable to have
prevalences of most classes which can be usefully ana-
lysed, i.e. which are larger than their CIs.
Consequently, it appears from Table 1 that data
from a single practice are insufficient to characterize
the wide variation in morbidity seen in family practice.
A group practice does significantly better, but still ap-
proximately half of the available classes are not suffi-
ciently represented. In a large study, however,
practically all the available classes allow for detailed
analysis since the diagnoses are observed commonly
enough to provide reliable data.33
However, the situation is worse when only encoun-
ter data with no structuring based on episodes of care
are available. In those data sets, one can only estimate
whether encounters for a certain diagnosis represent
recurrence of disease, multiple encounters for the
same problem or else various combinations of both.
In such cases, the rates are calculated with the number
of encounters as the denominator, in contrast with
data structured using episode of care with a denomina-
tor expressed as the number of patient years.6,11,13 A
multi-site episode of care study with approximately
140 000 patient years will contain almost 1 000 000
subencounters, that part of an encounter which deals
with one single episode of care, since an average pa-
tient year approximately includes three episodes of
care with two encounters each (Table 1). A similar
study based on encounter data thus requires a much
larger denominator to estimate incidence rates, while
it is practically impossible to calculate reliable preva-
lences. However, a rough estimate may be obtained
by using all patients visiting only once in 1 year as a de-
nominator, under the assumption that differences in
the utilization per episode of care do not substantially
skew this estimate.
Table 2 contains primary care encounter data de-
rived from the US National Ambulatory Medical Care
Survey (NAMCS 1995–1999, which includes data from
paediatricians, gynaecologists and internists working
in the community)18,19 compared with the episode of
care data from the Transition Project (1985–2002,
FDs in The Netherlands).8,15 This also allows a com-
parison between ICPC and ICD-9 data. It is evident
that the use of ICD-9-CM as a classification (with
2463 diagnostic classes) in ambulatory care will not
produce adequately useful clinical or epidemiological
314 Family Practice—an international journal

TABLE 1 Ranking of the prevalences (described as rates per 1000 patient years) of ICPC classes for diagnoses presenting in family practice, with the
width of the 95% CI for the observation within 1 practice (1356 patient years), 10 practices (13 560 patient years) and 100 practices (135 600 patient
years)

ICPC class Rank in Prevalence per 95% CI width for 95% CI width for 95% CI width for
prevalence 1000 patient years 1356 patient years 13 560 patient years 135 600 patient years

K86/87 (hypertension) 3 80 31 9 3
R74 (URTI) 8 54 25 5 2
N89 (migraine) 51 15 14 4 1
B80 (iron deficiency anaemia) 75 11 12 4 1
B81 (pernicious anaemia) 241 3 6 2 0.6
A20 (euthanasia discussion/request) 316 2 6 2 0.4
P77 (suicide attempt) 397 1 4 1 0.3
N86 (multiple sclerosis) 434 1 3 1 0.3
D76 (pancreatic malignancy) 587 0.2 2 0.5 0.2
A73/(malaria) 609 0.1 1 0.5 0.1

Source: EFP-extended.17

TABLE 2 Frequency of use of classifications in primary care. ICPC (for diagnoses and RFE) is compared to ICD-9-CM (for diagnoses) and the RVC
(for RFV). Data from The Netherlands (FDs) and the US (FDs, internists and paediatricians) expressed as rates per 1000 observations (subencounters)

Number of classes Available classes Classes unused, N (%)

occurring once or more
(>1) per 1000 observations

The Netherlands, FD, diagnosis (ICPC) 205 686 9 (13)

USA, FD, diagnosis (ICD-9-CM) 168 2463 374 (15)
USA, IM, diagnosis (ICD-9-CM) 166 2463 1113 (45)
USA, Ped, diagnosis (ICD-9-CM) 135 2463 1532 (62)
Netherlands, RFE (ICPC) 179 686 14 (2)
USA, FD, RFV (RVC) 134 1898 1242 (65)
USA, IM, RFV (RVC) 168 1898 1424 (75)
USA, Ped, RFV (RVC) 108 1898 1544 (81)

Sources: EFP8,32 and NAMCS 1995–1999.15,16

RVC, reason for visit classification.

data. This is because the large majority of available
classes will occur less than once per 1000 observations
(Table 2) and will consequently have large CIs. It is
surprising that, for both common diagnoses and RFE/
RFV, a need for similar levels of detail and differenti-
ation exists in family practice, internal medicine and
paediatrics. In ambulatory care, the main workload
for doctors consists of approximately 170 diagnoses
and 150 RFE. These form the core of this group’s pro-
fessional frame of reference. It is also apparent that
FDs do use a large number of available diagnostic la-
bels infrequently (only once or twice over the period
of observation).
Advantages of using episodes of care
A core element of the professional identity of FDs
is their ability to reliably estimate the probabilities
of diagnoses, and to assess the clinical utility of in-
terventions, in their practice populations. The
estimation of the predictive value of symptoms and
complaints and of the co-morbidity of any combina-
tion of two diseases form the basis of this applied
knowledge.1,21,34
Episode of care data are particularly useful to illus-
trate this. For example, the probability of the diagno-
sis ‘asthma’ in an episode of care that starts with the
RFE ‘shortness of breath’ is very high. The odds ratio
is 23.35 with a narrow CI (Table 3), representing a high
post-test (posterior) probability. The positive likeli-
hood ratio (LR+) is substantially greater than unity
(14.52; 95% CI 13.51–15.60), but the negative likeli-
hood ratio (LR–) (0.62; 95% CI 0.60 to 0.65) really
makes the difference. The low LR– supports the FD
in considering asthma to be a less likely diagnosis in
those patients who do not present with shortness of
breath. The familiar ‘abcd’ matrix of test result against
disease present/absent, shown in Table 3, results in
a straightforward formula for the odds ratio: (ad/bc).
Clearly, it is the very substantial (d) of 194 388 patient
years—very characteristic for family practice
 The coming of age of ICPC 315
TABLE 3 Prior (pre-test) and posterior (post-test) probabilities of the diagnosis asthma (R96) for the RFE shortness of breath (R02) in 201 037 patient
years

Patient years with Patient years with Total patient

episode of care R96 other episodes of care years
(asthma)

Presented with RFE shortness of breath (R02) 518 (a) 5.438 (b) 5.956
Presented with other RFE 793 (c) 194.388 (d) 195.181

Source: EFP.8 Sensitivity of RFE shortness of breath for the diagnosis asthma: 0.40. Specificity: 0.97. LR+ (of RFE present in episodes with asthma
against episodes with another diagnosis): 14.52 (95% CI 13.51–15.60). LR– (of RFE absent in episodes with asthma against episodes with other
diagnoses): 0.62 (95% CI 0.60–0.65). Positive predictive value (PV+): 0.09 (prior probability). Negative predictive value (PV–): 0.99 (prior prob-
ability). Diagnostic odds ratio: 23.35 (95% CI 20.84–26.17) (posterior probability).

TABLE 4 Co-morbidity of ‘acute bronchitis’ (R78) and asthma (R96) with LR+ and LR– and 95% CI

Co-morbidity of: % of R78 % of R96 LR+ LR– Odds ratio

with R96 with R78 (95% CI) (95% CI) (95% CI)
(PV+) (PV+)

Incident R78 (n = 7622) with incident R96 (n = 1311) 4.8 7.67 (7.01–8.39) 0.75 (0.72–0.77) 10.24 (9.05–11.58)
Incident R96 (n = 7622) with incident R78 (n = 1311) 27.8 9.80 (8.70–11.03) 0.96 (0.95–0.96) 10.24 (9.05–11.58)
Incident R78 (n = 7622) with prevalent R96 (n = 4267) 11.8 5.91 (5.54–6.30) 0.83 (0.81–0.84) 7.15 (6.61–7.74)
Prevalent R96 (n = 4267) with incident R78 (n = 7622) 20.3 6.46 (6.01–6.93) 0.90 (0.90–0.91) 7.15 (6.61–7.74)

PV+, positive predictive value.

research—that gives power to the calculations and al-
lows such precise and reliable estimates!
With regard to co-morbidity, it is essential to pres-
ent odds ratios in addition to the LR+ and LR–. This
is because the mutual overlap between two episodes
of care (shown as percentages in Table 4) is not identi-
cal (e.g. 4.8 versus 27.8%; 11.8 versus 20.3%). Thus
the LRs, especially the LR–, differ substantially for
the same odds ratio (Table 4). Moreover, the calcula-
tion of co-morbidity between a chronic disease, such
as asthma, and an acute disease, such as acute bronchi-
tis, requires distinguishing between incident and prev-
alent episodes of care. Tables 3 and 4 show that data
from an episode of care database can provide an epi-
demiological profile of the occurrence of respiratory
problems in a family practice population. The inci-
dence of either condition makes the incidence of the
other more likely (LR+ is 7.7 and 9.8), but although
incident acute bronchitis is made less likely without
incident asthma (LR– is 0.75), the converse is not
true (LR– is 0.96). Such an insight into these relation-
ships cannot be derived from an encounter-based
study.
Final remarks
ICPC has certainly grown into early adulthood: it is
full of promise and well prepared for a long and pro-
ductive life. The potential for the routine use of ICPC,
supported by the ICPC2/ICD-10 Thesaurus included
with ICPC-2-R which facilitates automatic double cod-
ing in electronic patients records, is now increasingly
appreciated and understood.35 Several of the theoreti-
cal principles underlying ICPC have proven to be
sound and resilient, and deserve to remain guiding
principles for an international family of classifications
in the 21st century.
Two elements require further attention: inter-doctor/
practice variation and its impact of data precision and
the implementation of computer-assisted data entry.
Inter-practice variation (nesting) in coding RFE,
diagnoses and prescriptions appears to be substantial
and often follows a characteristic and stable pat-
tern.36,37 Part of this variation is a welcome proof of
personalized doctoring and patient choice, another
part, however, may be less desirable. As a conse-
quence, empirical data are needed to develop models
for computer-assisted data entry by FDs.38
It will be a challenge to incorporate new ordering
principles of morbidity into the future versions of
ICPC (ICPC-3) based on data drawn from interna-
tional family practice settings.
See the website www.transitieproject.nl for more in-
formation, including two public domain databases of
ICPC data (used for this study) and ICPC tools and in-
formation/references. An ICD-10 ICPC-2 English
Dutch thesaurus is included. All these are also avail-
able in a CD ROM accompanying the ICPC-2-R book,
available from Oxford University Press or from
316 Family Practice—an international journal
Wonca (e-mail Alfred Loh, Wonca CEO, ceo@wonca.
com.sg).
Acknowledgements
We are greatly indebted to Dr Kerr White for his
never-ending support for ICPC. We thank Prof. GE
Fryer for the calculation of the NAMCS data in
Table 2.
Declaration
Funding: None.
Ethical approval: None.
Conflicts of interest: None.
References
1 encounters: the Primary Care Network Survey and the
Lamberts H, Wood M. ICPC. International Classification of Pri-
mary Care. Oxford: Oxford University Press, 1987.
2 2007; 5: 39–47.
Okkes IM, Lamberts H. Classification and the domain of family
practice. In Jones R (ed.). The Oxford Textbook of Primary
Medical Care, Volume 1. Oxford: Oxford University Press,
2003: 139–152.
3 Okkes IM, Oskam SK, Van Boven K, Lamberts H. EFP-Extended.
Lamberts H, Wood M. The birth of The International Classifica-
tion of Primary Care (ICPC). Serendipity at the border of
Lac Léman. Fam Pract 2002; 19: 433–435.
4 Primary Care (ICPC) in the Transition Project of the Aca-
Lamberts H, Meads S, Wood M. Classification of reasons why per-
sons seek primary care: pilot study of a new system. Public
Health Rep 1984; 99: 597–605.
5
Lamberts H, Wood M, Hofmans-Okkes IM. The International
Classification of Primary Care in the European Community
(with a sixteen language disk of ICPC). Oxford: Oxford Univer-
sity Press, 1993.
6
De Lusignan S, Van Weel C. The use of routinely collected com-
puter data for research in primary care: opportunities and chal-
lenges. Fam Pract 2006; 23: 253–263.
7
International Classification Committee of Wonca. ICPC-2-R. In-
ternational Classification of Primary Care. Revised second edi-
tion. With a CD-Rom: ICPC in the Amsterdam Transition
project. Oxford: Oxford University Press, 2005.
8
Okkes IM, Oskam SK, Lamberts H. ICPC in the Amsterdam Tran-
sition Project. CD-Rom. Amsterdam: Academic Medical Cen-
ter/University of Amsterdam, Department of Family
Medicine, 2005. http://www.transitieproject.nl/ (last accessed
15 May 2008).
9
Okkes IM, Oskam SK, Van Boven K, Lamberts H. EFP. Episodes
of Care in Family Practice. Epidemiological data based on the
routine use of the International Classification of Primary Care
(ICPC) in the Transition Project of the Academic Medical
Center/University of Amsterdam. (1985–2003). In Okkes IM,
Oskam SK, Lamberts H (eds). ICPC in the Amsterdam Transi-
tion Project. CD-Rom, Amsterdam: Academic Medical Center/
University of Amsterdam, Department of Family Medicine,
2005.
10
Becker HW, Oskam SK, Okkes IM, Van Boven K, Lamberts H.
ICPC2-ICD10 Thesaurus. A diagnostic terminology for semi-
automatic double coding in Electronic Patient Records. In
Okkes IM, Oskam SK, Lamberts H (eds). ICPC in the Amster-
dam Transition Project. CD-Rom, Amsterdam: Academic
Medical Center/University of Amsterdam, Department of
Family Medicine, 2005.
11
Britt H. A new coding tool for computerised clinical systems in pri-
mary care—ICPC plus. Aust Fam Physician 1997; 26 (suppl 2):
S79–S82.
12
Lamberts H, Okkes I. Sense and specificity in computer based pa-
tient records in general practice—the ICPC-ICD-10 conversion
structure as the Holy Grail [editorial]. Aust Fam Physician
1997; 26 (suppl 2): S57–S9.
13
Soler JK, Okkes IM. Outlining errors and inaccuracies in this re-
view. Available at http://fampra.oxfordjournals.org/cgi/elet-
ters/23/2/253 (last accessed on 15 May 2008).
14
Deveugele M, Derese A, Van den Brink-Muinen A, Bensing J, De
Maeseneer J. Consultation length in general practice: cross sec-
tional study in six European countries. Br Med J 2002; 325:
472–477.
15
Okkes IM, Polderman GO, Fryer GE et al. The role of family prac-
tice in different health care systems. A comparison of reasons
for encounter, diagnoses, and interventions in primary care
populations in the Netherlands, Japan, Poland, and the United
States. J Fam Pract 2002; 51: 72.
16
Jones R, Latinovic R, Charlton J, Gulliford MC. Alarm symptoms
in early diagnosis of cancer in primary care: cohort study using
General Practice Research Database. Br Med J 2007; 334
(7602): 1040.
17
Fryer GE, Green LA, Dovey SM, Yawn BP, Phillips RL, Lanier D.
Variation in the Ecology of Primary Care. Ann Fam Med 2003;
1: 81–9.
18
Binns HJ, Lanier D, Pace WD et al. Describing primary care
National Ambulatory Medical Care Survey. Ann Fam Med
19
Tierney WM, Oppenheimer CC, Hudson BL et al. A National
Survey of Primary Care Practice Based Research Networks.
Ann Fam Med 2007; 5: 242–250.
20
Episodes of Care in Family Practice. Epidemiological data
based on the routine use of the International Classification of
demic Medical Center/University of Amsterdam, extended
version: a retrieval program for the Amsterdam Transition
Project’s data with TransHis as an Electronic Patient Record.
1995–2004. 2005. Academic Medical Centre, University of Am-
sterdam; Amsterdam, The Netherlands.
21
Eldridge S. Good practice in statistical reporting for Family Prac-
tice. Fam Pract 2007; 24: 93–94.
22
Altman DG, Machin D, Bryant T, Gardner MJ. Statistics with
Confidence. Confidence Intervals and Statistical Guidelines.
London: Blackwell BMJ Books, 2000.
23
Jordan K, Porcheret M, Croft P. Quality of morbidity coding in
general practice computerized medical records: a systematic
review. Fam Pract 2004; 21: 396–412.
24
Porcheret M, Hughes R, Litt M et al. Data quality of general prac-
tice electronic health records: the impact of a program of as-
sessments, feedback, and training. J Am Med Inform Assoc
2004; 11: 78–86.
25
Peabody JW, Luck J, Jain S, Berthenthal D, Glassman P. Assessing
the accuracy of administrative data in health information sys-
tems. Med Care 2004; 42: 1066–1072.
26
Thiru K, Hassey A, Sullivan F. Systematic review of scope and
quality of electronic patient record data in primary care. Br
Med J 2003; 326: 1070–1072.
27
Brown PJB, Warmington V, Laurence M, Prevost AT. Rando-
mised crossover trial comparing the performance of Clinical
Terms Version 3 and Read Codes 5 byte set coding schemes
in general practice. Br Med J 2003; 326: 1127–1131.
28
Lamberts H, Brouwer HJ, Marinus AMF, Hofmans-Okkes IM.
The use of ICPC in the Transition project. Episode oriented
epidemiology in general practice. In Lamberts H, Wood M,
Hofmans-Okkes IM (eds). The International Classification of
Primary Care in the European Community, Oxford: Oxford
University Press, 1993: pp. 45–93.
29
Meijer JS. [A perspective on disease. The classification of disease
and health problems in general practice] [Dutch, with a sum-
mary in English]. PhD Thesis, University of Amsterdam,
Amsterdam, 1998.
 The coming of age of ICPC 317
30
Veltman MTM. [Episodes of care in general practice. An analysis of symptoms to Dutch family physicians. J Fam Pract 2002; 51:
seven years’ data] [Dutch]. PhD Thesis, University of 31–36.
35
Amsterdam, Lelystad, Meditekst, 1995. Okkes IM. Banff declaration calls for ICPC in the United States.
31
Burnand B, Kernan WN, Feinstein AR. Indexes and boundaries Wonca News 2004.
36
for ‘quantitative significance’ in statistical decisions. J Clin Dickinson LM, Basu A. Multilevel modelling and practice-based
Epidemiol 1990; 43: 1273–1284. research. Ann Fam Med 2005; 3 (suppl 1): S52–S60.
32 37
Feinstein AR. P-values and confidence intervals: two sides of Marinus AMF. [Inter-doctor variation in general practice] [Dutch,
the same unsatisfactory coin. J Clin Epidemiol 1998; 51: 355–360. with a summary in English]. PhD Thesis, University of
33
Eimerl TS. Organized curiosity: a practical approach to the Amsterdam, Meppel, Krips, 1993.
38
problem of keeping records for research purposes in general George (Spenceley) SE, Warren JR. Statistical modelling of gen-
practice. J R Coll Gen Pract. 3: 246–252. eral practice medicine for computer assisted data entry in elec-
34
Okkes IM, Oskam SK, Lamberts H. The probability of specific tronic medical record systems. Int J Med Inform 2000; 57:
diagnoses for patients presenting with common 77–89.