The Phenomenology of the Diagnostic Process:
A Primary Care–Based Survey
Norbert Donner-Banzhoff, MD, MSc, Judith Seidel, MD, Anna Maria Sikeler, MD,
Stefan Bo
¨
sner, MD, MPH, Maria Vogelmeier, MD, Anja Westram, PhD,
Markus Feufel, PhD, Wolfgang Gaissmaier, PhD, Odette Wegwarth, PhD,
Gerd Gigerenzer, PhD
Background. While dichotomous tasks and related cogni-
tive strategies have been extensively researched in cogni-
tive psychology, little is known about how primary care
practitioners (general practitioners [GPs]) approach ill-
defined or polychotomous tasks and how valid or useful
their strategies are. Objective. To investigate cognitive
strategies used by GPs for making a diagnosis. Methods.
In a cross-sectional study, we videotaped 282 consulta-
tions, irrespective of presenting complaint or final diag-
nosis. Reflective interviews were performed with GPs
after each consultation. Recordings of consultations and
GP interviews were transcribed verbatim and analyzed
using a coding system that was based on published litera-
ture and systematically checked for reliability. Results.In
total, 134 consultations included 163 diagnostic episodes.
Inductive foraging (i.e., the initial, patient-guided search)
could be identified in 91% of consultations. It contribu-
ted an average 31% of cues obtained by the GP in 1
consultation. Triggered routines and descriptive ques-
tions occurred in 38% and 84% of consultations, respec-
tively. GPs resorted to hypothesis testing, the hallmark of
the hypothetico-deductive method, in only 39% of con-
sultations. Limitations. Video recordings and interviews
presumably interfered with GPs’ behavior and accounts.
GPs might have pursued more hypotheses and collected
more information than usual. Conclusions. The testing of
specific disease hypotheses seems to play a lesser role
than previously thought. Our data from real consultations
suggest that GPs organize their search for information in
a skillfully adapted way. Inductive foraging, triggered
routines, descriptive questions, and hypotheses testing
are essential building blocks to make a diagnosis in
the generalist setting. Key words: general practice;
decision making; hypothesis testing; qualitative research;
diagnosis, family medicine; cues. (Med Decis Making
2017;37:27–34)
M
aking a diagnosis is perhaps the most intellec-
tually challenging task of the physician. Often
within minutes or even seconds, experienced physi-
cians manage to narrow the range of possible diseases.
While invasive and costly diagnostic procedures are
widely debated by researchers and health planners, it
is the patient’s history that provides the most impor-
tant material for this task.
1
To a physician making a diagnostic assessment,
plenty of information is available, ranging from the
patient’s utterances, knowledge of the patient’s his-
tory, disease prevalences, visual impression of the
patient, and findings from the physical examination.
Given the limited capacity of the human brain and
that much information is noise,
2
physicians can and
should process only part of it. But how do they collect,
select, and, inevitably, ignore information? When do
they stop and when do they continue to collect more?
In medicine, the most influential model of the diag-
nostic process has perhaps been the hypothetico-
deductive method proposed by Elstein and others.
3
According to this view, early in the encounter with the
patient, physicians form diagnostic hypotheses in
their minds. These guide further data collection,
which aims at confirmation or disconfirmation of pos-
sible explanations for the patient’s problem. This may
result in an iterative process of rejection and reformu-
lation hypotheses until an adequate conclusion has
been found.
4
It remains unclear, however, which processes
precede the formulation of the first hypothesis.
Once they are entertaining one or more hypotheses,
physicians have narrowed the range of possible
Ó The Author(s) 2016
Reprints and permission:
http://www.sagepub.com/journalsPermissions.nav
DOI: 10.1177/0272989X16653401
ORIGINAL ARTICLE
MEDICAL DECISION MAKING/JANUARY 2017 27
Konstanzer Online-Publikations-System (KOPS)
URL: http://nbn-resolving.de/urn:nbn:de:bsz:352-0-354508
Erschienen in: Medical Decision Making ; 37 (2017), 1. - S. 27-34
https://dx.doi.org/10.1177/0272989X16653401
explanations from several hundreds, or even thou-
sands, to perhaps 3 or 4 at most. This reduction of
uncertainty must be regarded as a considerable
achievement, especially in generalist settings, such
as primary care or hospital emergency departm ents.
We have proposed ‘‘inductive foraging’’ to have a
central role in the first stage of diagnostic data col-
lection. Inductive foraging refers to an initial open
search for information guided by the patient. This
stage usually starts with an open prompt by the
physician, giving the patient the opportunity to ela-
borate on his or her problem. It can be terminated
by the patient or by the physician interrupting the
patient.
5
Inductive foraging is more than the patient
stating a presenting complaint. With the diagnostic
task represented as the search through an almost
infinite pro blem space, inductive foraging helps
define and limit the problem space. It is only the
patient who can provide this indispensable init ial
guidance. For the first time, we here report empiri-
cal data about the occurrence of this and related
strategies (for the definition and operationalization
of strategies, see Table 1 and Figure 1).
Although the initial diagnostic assessment
achieved by physicians is important, it is difficult
to investigate, resulting in a plethora of discussion
and educational statements, but only a few empiri-
cal studies of the process itself. We therefore under-
took a survey of real-life primary care consultations.
Our objective was to explore cognitive strategies
used by general practitioners (GPs) for their diagnos-
tic assessment. Based on our data, we propose a phe-
nomenology that has implications for the generalist
setting and any other setting where a large number of
diagnostic possibilities must be considered.
METHODS
Twelve full-time GPs in the Marburg-Biedenkopf
district of Hessen, Germany, were asked to take part
in the study and all agreed to participate. We
required participants to have been in practice for at
least 5 years and to have under- or postgraduate
teaching experience.
We covered 3 ha lf-day sessions with each partici-
pating GP. Patients were included irrespective of their
symptoms or possible diagnoses. Only consultations
exclusively planned for nondiagnostic reasons, such
as chronic disease monitoring or follow-up for a previ-
ously identified problem, were excluded beforehand.
Participating GPs informed each patient about
the study and asked for written consent to partici-
pate and have his or her consultation video-
recorded. GPs were instructed not to address the
patient’s presenting complaint(s) at this stage. After
consent was obtained, consultations proceeded as
usual. Sessions were scheduled so that after each
consultation, GPs had sufficient time for a semi-
structured interview to explain their diagnostic rea-
soning. These interviews were also video-recorded.
GPs were asked to use the initial utterance by the
patient as the starting point of their reflection. GPs
elaborated on their first impression and previous
knowledge of each patient, as well as diagnostic
hypotheses considered.
Recordings of consultations and GP interviews
were transcribed verbatim. Transcripts were coded
with MAXQDA software for qualitative data analy-
sis.
6
We defined a consultation as containing a diag-
nostic episode when the patient brought up a new
complaint, which resulted in some kind of data col-
lection by the GP, such as taking a history or exam-
ining the patient.
Received 23 September 2015 from the Department of Primary Care,
Philipps University of Marburg, Marburg, Germany (ND-B, JS, AMS,
SB, MV); Harding Center for Risk Literacy, Max-Planck-Institute for
Human Development, Berlin, Germany (AW, MF, OW, GG); and
Department of Psychology, University of Konstanz, Konstanz,
Germany (WG). This research was supported by the German
Research Foundation (DFG) with grants DO 513/2-1 and GI 170/8-1.
All authors have completed the Unified Competing Interest form at
www.icmje.org/coi_disclosure.pdf (available on request from the cor-
responding author) and declare that ND-B, JS, AMS, SB, MV, AW,
MF, WG, OW, and GG have no relationships with companies that
might have an interest in the submitted work in the previous 3 years;
their spouses, partners, or children have no financial relationships that
may be relevant to the submitted work; and ND-B, JS, AMS, SB, MV,
AW, MF, WG, OW, and GG have no nonfinancial interests that may
be relevant to the submitted work. The funder of the study had not
influence on study design, collection, analysis, and interpretation of
data; on the writing of the report; or the decision to submit the article
for publication. All authors had full access to all of the data in the
study and can take responsibility for the integrity of the data and the
accuracy of the data analysis. The lead author (ND-B) affirms that this
manuscript is an honest, accurate, and transparent account of the
study being reported; that no important aspects of the study have
been omitted; and that any discrepancies from the study as planned
have been explained. Data sharing: no additional data available.
Revision accepted for publication 13 May 2016.
Supplementary material for this article is available on the Medical
Decision Making Web site at http://mdm.sagepub.com/supplemental.
Address correspondence to Stefan Bo
¨
sner, MD, Department of
Primary Care, Philipps University of Marburg, Karl-von-Frisch-Str. 4,
D-35043 Marburg, Germany; telephone: +49 6421 286 5122; fax:
+49 6421 286 5121; e-mail: boesner@staff.uni-marburg.de.
DONNER-BANZHOFF AND OTHERS
28
MEDICAL DECISION MAKING/JANUARY 2017
Table 1 Cognitive Strategies: Definitions and Operationalization
Definition Operationalization Illustrative Example
Inductive foraging The patient is guiding the
physician to relevant
problem areas. Search of a
wide problem space is thus
possible (see Figure 1a).
This is the only phase
controlled by the patient.
Open question or invitation
by physician regarding the
reason for visit; subsequent
time period with patient
elaborating spontaneously.
Usually occurs at the
beginning of the
consultation but
occasionally marking the
beginning of a second
diagnostic episode within
the same consultation.
Terminated either by the
patient pausing or
suggesting a diagnosis
himself or herself or by the
physician interrupting,
usually with a closed
question.
GP: What brings you here
today?
Patient: What I mentioned
before; I stop breathing at
night.
GP: Yes.
Patient: While I’m sleeping. .
. . My wife says, when it
happens, it happens
several times, and it takes
30 seconds until after she
has given me a kick, until I
start breathing again.
GP: Are you snoring?
[continued below
‘‘hypothesis testing’’]
(0310-P)
Descriptive questioning Physician obtains
descriptions of symptoms
or problems. Phase
controlled by physician.
Closed questions aiming at
descriptions of a symptom
or problem already
mentioned by the patient.
GP: And where, exactly, do
you feel the pain? (0212-P)
Triggered routines Physician obtains
information on problem
area becoming relevant
because of symptom or
problem mentioned by
patient (see Figure 1b).
Phase controlled by
physician.
Closed questions, exploring
a problem area, such as an
organ system. Triggered by
symptoms mentioned by
the patient. Formal clinical
prediction rules also fulfill
this definition.
[Patient complaining of
diarrhea] GP: Blood in your
stool?
Patient: No.
GP: Do you feel sick?
Vomiting?
Patient: Not really, but I feel
a bit funny—not all the
time though. . . . And there
was a loud noise in my
tummy last night, but
that’s gone now.
GP: OK, any temperature?
Patient: No, I didn’t take it.
(0301-P)
Hypothesis testing Physician evaluates defined
hypotheses (diseases)
potentially explaining the
patient’s problem.
Following the space
metaphor, this corresponds
to deep
digging for findings
or abnormalities
specifically associated with
hypothesis (Figure 1c).
Phase controlled by
physician.
Closed questions related to a
specific disease, aiming at
confirmation or
disconfirmation.
Hypothesis either
explicitly mentioned in
interview or questions
asked relate to a specific
disease as assessed by a
medically trained coder.
GP: Are you snoring? Has
your wife mentioned that
you snore?
Patient: [always answers in
the affirmative]
GP: How do you feel when
you get up in the morning?
Tired? Exhausted?
Patient: . . .
GP: You are saying, when
you sit comfortably, you
easily fall asleep. . . . There
is a possibility that you
have sleep apnea. (0310-P)
GP, general practitioner.
PHENOMENOLOGY OF THE DIAGNOSTIC PROCESS
ORIGINAL ARTICLE 29
Drawing from previously published work,
3,5,7
we
developed a coding tree composed of categories
describing GPs’ diagnostic reasoning and data-
collecting behavior (available upon request).
Extensive discussion in our group and several itera-
tive loops of coding and modification of the coding
tree resulted in operational definitions of categories.
After completing this phase, we tested the reliabil-
ity of our coding procedure on 3 consultations and
3 interviews, all randomly selected. Two blinded
independent raters double-coded the material with
high reliability, resulting in a weighted mean of
84% agreement.
8
Remaining discrepancies in
coding were resolved by discussion with senior
members of the research team.
To quantify the yield of cognitive strategies, we
identified all verbal and nonverbal diagnostic cues
presented by patie nts. A cue is any piece of infor-
mation made available to the GP, either sponta-
neously or during questioning. Nonverbal cues
were usually GPs’ visual impressions and findings
from the physical examination. For this kind of
analysis, we selected a random sample of 5 consul-
tations from each practice, including GP interviews
(n = 58). In one practice, where only 3 consultations
were included, we analyzed all consultations in
this manner.
Each item of diagnostic information gathered by
the GP during the first diagnostic episode of a con-
sultation was coded irrespective of whether it was
obtained spontaneously or in response to a closed
question. Information modifying a previous cue in a
relevant way, such as the quality of pain or specific
triggers causing a symptom, was counted as a sepa-
rate cue. We could thus quantify to what degree dif-
ferent strategies contributed to the information
acquired from each patient. For each strategy, we
report frequency of occurrence within the consulta-
tion, including 95% confidence intervals.
Two authors (ND-B and GG) developed the gen-
eral design of the study. Details of the study proto-
col, such as recruitment and data collection and
analysis, were discussed by the entire study team.
Two authors (MV and AW) collected data in partici-
pating practices and lead interviews with GPs, and
2 qualified physicians (JS, AMS) coded and ana-
lyzed the text material. They were directly super-
vised by ND-B and SB. Selected text passages were
discussed by the entire team, which included
researchers with medical (ND-B, JS, AMS, SB, MV)
and cognitive psychology (WG, MAF, OW, AW, GG)
backgrounds. ND-B is the guarantor for this work.
The study obtained ethical ap proval by the Ethics
Committee of the Faculty of Medicine, University of
Marburg (39/2010).
RESULTS
Participating GPs and Patients
Of 12 participating GPs, 5 were female. On aver-
age, they were 53 years old and had been in primary
care practice for 21 years (for details, see Table 2).
After the exclusion of consultations with techni-
cally unsatisfactory recordings and/or without
diagnostic content, 134 consultations with 163 diag-
nostic episodes were available for analysis (see
flowchart in Figure 2; for patients’ characteristics,
see Table 3 and Suppl. Table SA). Practices contrib-
uted between 3 and 16 consultations with diagnos-
tic content. Their average duration was 9 minutes
59 seconds (range, 2 minutes 45 seconds to 28
Figure 1 Illustration of cognitive strategies: (a) inductive fora-
ging, (b) triggered routine, and (c) hypothesis testing. The plane
stands for the problem space of the patient. Symptoms, which
the patient would present if given enough time, are shown above
the surface (geometrical shapes). Findings accessible only by
directed questioning or search are hidden below the surface (c).
An initial search guided by the patient will lead physicians to
problem areas where directed search (‘‘digging’’) will be highly
efficient. Efficiency is lost if physicians start directed search
(hypothesis testing) too early.
DONNER-BANZHOFF AND OTHERS
30
MEDICAL DECISION MAKING/JANUARY 2017
minutes 15 seconds). Reflective interviews lasted
between 2 and 18 minutes (median, 6 minutes 35
seconds).
Inductive Foraging
Inductive foraging (i.e., the initial search guided
by the patient) could be identified in 122 (91%)
consultations (for definition and operationalization
of strategies, see Table 1 and Figure 1). In 5 of the
remaining 12 cases, the beginning of the consulta-
tion was not recorded for technical reasons.
Inductive foraging is thus likely to occur even more
frequently. The median duration of this phase was
34 seconds (range, 6–176 seconds). Inductive fora-
ging took a proportion of 14.6% (median) of diag-
nostic episodes (range, 1.7%–93.1%). See online
supplemental material for the effect of practice and
presenting symptoms (Suppl. Table SD and SE).
In diagnostic episodes beginning without induc-
tive foraging, GPs started with testing a hypothesis
derived from previous encounters or from infor ma-
tion obtained by receptionists.
In 70 (57%) cases, inductive foraging ended by
the GP interrupting the patient, usually by asking a
closed question. In 6 consultations, there was a
return to inductive foraging at a later stage .
GPs employed several tactics to support patients
in their foraging for complaints and symptoms;
paraphrasing, verbal prompts, and nonverbal
encouragements were the most frequently used.
Triggered Routines and Descriptive Questions
We identified an intermediate stage in the diag-
nostic process in which GPs explored a limited
problem area. GPs consistently used these ‘‘trig-
gered routines’’ for identical symptoms (see Table 1
and Figure 1 for definitions of strategies). This strat-
egy was observed in 62 (38%) of 163 diagnostic
episodes.
GPs asked descriptive questions in 137 (84%)
diagnostic episodes. A physical examination was
conducted in 120 (89%) consultations.
Figure 2 Flowchart of participating physicians and patients.
Table 3 Participating Patients (n = 134)
Characteristic Value
Age, mean (standard deviation), y 54.6 (4.8)
Sex: female, n (%) 85 (63)
Family, n (%)
Single/divorced/widow 47 (35)
Married/living with partner 81 (60)
Adolescent living with parents 6 (5)
Level of education,
a
n (%)
Low 31 (23)
Intermediate 73 (55 )
High 30 (22)
a. Levels defined according to German educational system: low, basic
(Hauptschule) or no secondary education; intermediate, equivalent to
O-levels (Realschule, mittlere Reife, etc.); high, graduation from gram-
mar school or equivalent (Abitur).
Table 2 Participating General Practitioners
(n = 12)
Characteristic Value
Sex: proportion female, n 5/12
Age, median (range), y 53 (49–62)
Practice, n
Single handed 2
Group (2 partners and more) 10
Years in primary care practice, median (range) 21 (9–30)
Location of practice, n
Urban—small town 9
Rural 3
PHENOMENOLOGY OF THE DIAGNOSTIC PROCESS
ORIGINAL ARTICLE 31