Content uploaded by Petar J Denoble
Author content
All content in this area was uploaded by Petar J Denoble on Apr 07, 2016
Content may be subject to copyright.
Content uploaded by John J Freiberger
Author content
All content in this area was uploaded by John J Freiberger
Content may be subject to copyright.
Delivered by Ingenta to:
Aerospace Medical Association
IP : 24.106.206.82
Mon, 16 Jul 2007 20:07:46
RESEARCH ARTICLE
Consensus Factors Used By Experts in the Diagnosis
of Decompression Illness
John J. Freiberger, Sean J. Lyman, Petar J. Denoble,
Carl F. Pieper, and Richard D. Vann
FREIBERGER JJ, LYMAN SJ, DENOBLE PJ, PIEPER CF, VANN RD. Con-
sensus factors used by experts in the diagnosis of decompression
illness. Aviat Space Environ Med 2004; 75:1023–8.
Introduction: The diagnosis of decompression illness (DCI) is entirely
based on clinical findings and DCI experts are rare. Of all the chambers
reporting to Diver’s Alert Network (DAN), 86% see less than 10 cases
per year. Simulated diving injury cases (vignettes) were used to identify
diagnostic factors important to 11 international experts attending the
2003 Undersea and Hyperbaric Medical Society symposium on DCI
diagnosis. Methods: There were 200 vignettes evaluated for the proba-
bility of DCS and/or arterial gas embolism (AGE). Vignettes were con-
structed from 141 factors that modeled information from DAN’s emer-
gency call system. Factor probability mirrored DAN’s 2001 Report on
Decompression Illness and Diving Fatalities. Factors included: diver
characteristics, exposure characteristics, signs, symptoms, treatment,
and response. Multiple linear regression with stepwise elimination iden-
tified and ordered the significant factors in terms of their importance to
the experts. Results were confirmed with logistic regression. Results: For
DCS, the top five factors in order of importance were: 1) a neurological
symptom as the primary presenting symptom; 2) onset time of symp-
toms; 3) joint pain as a presenting symptom; 4) any relief after recom-
pression treatment; and 5) the maximum depth of the last dive. For AGE,
the top five factors were: 1) onset time of symptoms; 2) altered con-
sciousness; 3) any neurological symptoms as a presenting symptom; 4)
motor weakness; and 5) seizure as the primary presenting symptom.
Age, gender, or physical characteristics were not statistically important.
Conclusions: The vignette concept may be useful in the development of
consensus standards for DCI diagnosis.
Keywords: diving, decompression sickness, arterial gas emboli, hyper-
baric oxygen, vignette simulation.
THE DIAGNOSIS OF decompression illness (DCI) is
unusual in modern medicine in that it relies almost
exclusively on clinical presentation. No laboratory tests
exist that can incontrovertibly confirm or reject the di-
agnosis (5). Although various blood tests have been
proposed (6, 8, 11), they are not specific for DCI, they
are not in general clinical use, and their utility is un-
known. Even the so-called “test of pressure,” a post hoc
response to treatment, is not always reliable. Patients
may not respond if they are severely injured (3,4,12),
treated late (1), or if they are treated after a secondary
decompression stress such as altitude exposure (14).
Some patients without DCI may respond to therapy via
the placebo effect. Because there is no objective “gold
standard,” the diagnosis of DCI relies heavily on clini-
cal experience; however, experience in diagnosing DCI
is extremely difficult for a clinician to obtain. Even
though DCI is comprised of both decompression sick-
ness (DCS) and arterial gas embolization (AGE), cases
of DCI are rare with an estimated incidence of approx-
imately 4 cases for each 10,000 exposures in recreational
divers (13). Very few practicing physicians are able to
acquire sufficient experience to become expert in the
diagnostic process. A small percentage (14%) of the
recompression chambers that report diving injuries to
Diver’s Alert Network (DAN) provide more than 90%
of all cases, meaning that it is difficult for any one
physician to accumulate first hand experience. How-
ever, in April of 2003 the Undersea and Hyperbaric
Medical Society organized an international symposium
to discuss techniques for creating a diagnostic classifi-
cation system in the absence of a gold standard. The
participants included some of the world’s foremost au-
thorities on diving medicine with a collective expertise
far beyond that possible to assemble apart from such a
special occasion. To take advantage of this unique op-
portunity, we asked 11 of these experts to rate a series
of 200 DCI vignettes for the likelihood of DCS and/or
AGE. We then analyzed the responses with linear re-
gression, confirmed by logistic regression, to identify
the diagnostic determinants, or factors, that they used.
METHODS
Factor and Vignette Generation
A Microsoft Excel macro generated 300 virtual cases
or vignettes from 141 separate case factors that were
composed by the authors to model the information
available to a physician or a medic answering calls to
the DAN Emergency Hot Line. Although we recognize
that the DAN injury cases cannot be used as a gold
standard, to provide realism, the macro selected symp-
From the Research Department, Diver’s Alert Network, Durham,
NC (J. J. Freiberger, P. J. Denoble, R. D. Vann); Department of Anes-
thesiology (J. J. Freiberger, R. D. Vann), Center for Hyperbaric Med-
icine and Environmental Physiology (J. J. Freiberger, S. J. Lyman), and
Department of Biometry and Bio-Informatics (C. F. Pieper), Duke
University Medical Center, Durham, NC.
This manuscript was received for review in July 2004. It was
accepted for publication in October 2004.
Address reprint requests to: John J. Freiberger, M.D., M.P.H., who
is Assistant Clinical Professor of Anesthesiology, Center for Hyper-
baric Medicine and Environmental Physiology, DUMC 3823 Room
0584, CRII Building, Durham, NC 27710; jfreiberger@dan.duke.edu.
Reprint & Copyright © by Aerospace Medical Association, Alexan-
dria, VA.
1023Aviation, Space, and Environmental Medicine •Vol. 75, No. 12 •December 2004
Delivered by Ingenta to:
Aerospace Medical Association
IP : 24.106.206.82
Mon, 16 Jul 2007 20:07:46
tom factors based on the probability of their appearance
in Table 2, page 21 of the DAN 2001 Report on Decom-
pression Illness and Diving Fatalities (15). Likewise,
factors for gender, age, medical history, symptom se-
verity, the first symptom’s onset time, number of dives
in the dive series, maximum depth of the last dive,
number of days diving, flying after diving, problems
during the dive, use of surface oxygen, time to admin-
istration of oxygen, time to recompression treatment,
number of recompression treatments and the response
to both oxygen and recompression treatment were also
assigned probabilities based on demographic and other
data from the aforementioned 2001 report. The 300
candidate vignettes were sequentially reviewed for
plausibility by the authors and the first 200 without
unrealistic scenarios, based on the author’s clinical ex-
perience, were selected for the experiment. Because all
of the important factors influencing the development of
DCI are not known, it was recognized that the factors
included in the vignettes were an incomplete list of all
the possible contributing conditions and no claim was
made that any resulting model using solely our factors
could be used to predict the probability of DCS in actual
patients.
Rating Process
The 11 diving experts individually evaluated the
same 200 vignettes. They were instructed to imagine
that they were doing chart reviews on cases treated for
a diving injury and being asked to evaluate each case
post hoc for their expert estimate of the probability of
DCS and/or AGE from the information given. All dives
were assumed to have been within the U.S. Navy no
decompression limits for depth and time unless other-
wise stated and symptom onset times were defined as
the time after surfacing that the diver or buddy re-
ported the first symptom. The experts were instructed
to assume that, as in real life, a finding may have been
present before it was noticed and reported. All possible
explanations for the findings, such as diseases other
than DCS or AGE, including psychosomatic responses,
were to be considered. The raters were also asked to
assume that all concurrent medical conditions de-
scribed in the diver related factors (for example; diabe-
tes, heart disease, hypertension, pneumothorax, loss of
consciousness, seizures, others) were reliably described
and that if a medical event occurred in the vignette it
was diagnosed and/or treated appropriately. This was
done so that the experts would make their decisions
based solely on the experiment’s factors and not be
concerned about the reliability of the medical findings
that were presented. The experts were also told that
some cases would appear more than once.
We theorized that some of the vignettes would con-
tain certain specific factors and their combinations that
would fit well into the rater’s concept of either DCS or
AGE. These vignettes, therefore, would be rated high.
Conversely, some factors or combinations would not fit
either disease in the rater’s minds. In these cases the
rater’s assigned probabilities would be low, indicating
that they felt that the presented scenario represented a
missed diagnosis of another condition. Because the fac-
tors presented in the vignettes were randomly assigned,
some vignettes appeared more like DCI cases than oth-
ers. This was necessary to provide a wide range of
expert assigned probability values against which to
regress the diagnostic factors in the vignettes. However,
because we did not want symptom order to reflect
assumed importance, the factors were always ab-
stracted in the same order relative to each other. After
reading each vignette, the experts were asked to rate the
probability of both DCS and AGE ona0to100% scale
that was graduated in 10% increments. Because the
diagnosis of DCS does not exclude the possibility of
also having AGE, a probability score could be indepen-
dently assigned for both possible diagnoses. The re-
sponses were stored by the macro to allow for easy
retrieval and analysis by the research team without the
need for cumbersome paper forms or duplicate data
entry. A complete list of all factors in the vignettes and
a screen shot of the macro are available by e-mail upon
request.
Statistical Methodology
We assumed that each rater would have an underly-
ing relative propensity to declare a vignette either DCS
or AGE analogous to the probability that an unan-
swered DAN call would have DCS or AGE. We also
assumed that each factor would influence the rater’s
probability score for that vignette. Once the experts had
assigned their subjective probability scores to the 200
vignettes, the average DCS and AGE probability scores
for each was calculated and used as the response or “y”
variable in a multiple linear regression in which the
factors were the predictor or “x” variables. Stepwise
elimination (criteria for probability of F, entry ⫽0.05,
removal ⫽0.10) was used to exclude the factors that did
not significantly improve the model as measured by an
increase in the overall amount of variance explained
(the adjusted R
2
). The statistically significant factors
were then ordered in terms of importance by using the
“p” value and the absolute value of the standardized
partial regression coefficient (beta weight) for each fac-
tor to portray its effect strength. The statistically signif-
icant factors with lowest p values and the largest abso-
lute beta weights were considered to have had the
greatest influence on the DCS or AGE probability as-
signment made by the experts and, therefore, were
considered to have been most important to them in
arriving at a diagnosis (7). In a second analysis, the
factors identified by linear regression were confirmed
by logistic regression using the expert assigned proba-
bility score cut-off point of greater than 0.5 to categorize
each vignette as a DCS or AGE case or not (score ⬎
0.5 ⫽‘case’). After categorization, these binary data
were then regressed against the 141 factors using logis-
tic regression with forward conditional stepwise elimi-
nation (criteria for entry ⫽0.02, removal ⫽0.7, classi-
fication cut-off, 0.5, 20 iterations maximum) to select the
most influential factors. All regressions were performed
with SPSS version 8.0 for Windows.
FACTORS IN DCI DIAGNOSIS—FREIBERGER ET AL.
1024 Aviation, Space, and Environmental Medicine •Vol. 75, No. 12 •December 2004
Delivered by Ingenta to:
Aerospace Medical Association
IP : 24.106.206.82
Mon, 16 Jul 2007 20:07:46
Reliability Measurements
The internal consistency of the raters (inter-rater re-
liability) was tested by repeating three of the 200
vignettes and computing their intra-class correlations
(ICC). Inter-rater agreement, the generalization of the
ICC which allows additional sources of variance, (here,
rater, vignette, and error), was employed to assess
agreement among the experts about the disease proba-
bility for the individual vignettes (10).
RESULTS
Regressions
After linear regression and stepwise elimination, 25
of the 141 possible factors remained as significant
predictors of DCS probability and 10 factors re-
mained as significant predictors of AGE probability.
Some of the factors increased the probability of a
particular diagnosis whereas others decreased it. The
overall ANOVA for both the DCS and AGE regres-
sions was significant to p ⬍0.001. The significant
factors explained 76% and 46% of the variance of the
expert’s assigned probability for DCS and AGE re-
spectively. The adjusted R
2
was 0.76 for DCS and 0.46
for AGE. Analysis of Studentized residuals for DCS
showed them to be approximately normally distrib-
uted. For AGE there was some indication of an in-
crease in variance with predicted values of the de-
pendent variable. Logistic regression confirmation
identified 19 factors for DCS and 7 factors for AGE,
with 9 out of the 19 DCS factors and 4 out of the 7
AGE factors common to both the logistic and linear
methods.
Specific Factors Used by the Experts for the Diagnosis of
DCS and AGE
The factors used by the experts to diagnose DCS were
different from those used to diagnose AGE. Table I
shows the 25 most important factors used by the experts
to predict DCS and Table II shows the 10 most impor-
tant factors used by the experts to diagnose AGE as
revealed by linear regression. Factors identified by both
linear and logistic regression are note by bold text. All
other factors identified by logistic regression are listed
in the table descriptions.
Reliability
Three cases were repeated to test inter-rater reliabil-
ity. Of the three repeated cases, one was very ambigu-
ous, one was moderately ambiguous, and one was ob-
viously either DCS or AGE. Intra-class correlations for
these cases were 0.3, 0.57, and 0.7 respectively, indicat-
ing moderate internal consistency (reliability). The
agreement among the experts about the disease proba-
bility for the individual vignettes or inter-rater agree-
ment was moderate ranging from 0.49 to 0.61.
TABLE I. THE 25 FACTORS MOST IMPORTANT TO THE EXPERTS IN MAKING THE DIAGNOSIS OF DCS. THE FACTORS IN BOLD
WERE IDENTIFIED BY BOTH LOGISTIC AND LINEAR REGRESSION.*
Factor sig Absolute value of beta
1. Neuro symptom (any type) as primary presenting symptom 0 0.447
2. Onset time in min after surfacing of 1st symptom (longer onset time was a negative predictor) 0 0.374
3. Joint pain (may be multiarticular) as primary presenting symptom 0 0.278
4. Any relief after recompression treatment 0 0.266
5. Max depth last dive 0 0.254
6. Seizure as primary presenting symptom (negative predictor) 0 0.252
7. Motor weakness (anywhere) reported as secondary symptom 0 0.222
8. Complete relief after first recompression 0 0.213
9. Unusual fatigue as primary presenting symptom 0 0.173
10. Fine cerebellar signs reported as secondary symptom or finding 0.001 0.132
11. Vertigo (any symptom) (negative predictor) 0.002 0.123
12. Missed decompression stop in recent dive exposure history 0.001 0.122
13. Heart disease history (any) (negative predictor) 0.002 0.114
14. Skin symptoms (any type) as primary presenting symptom 0.003 0.111
15. Unusual fatigue reported as secondary symptom 0.003 0.11
16. Max depth of dive series 0.012 0.11
17. Pulmonary (any type) symptom as primary presenting symptom (negative predictor) 0.004 0.108
18. Previous DCS in past medical history 0.003 0.107
19. Diabetes in past medical history (negative predictor) 0.003 0.105
20. Numbness or paresthesia reported as secondary symptom 0.012 0.089
21. Evolution (worsening) of primary presenting symptom 0.023 0.079
22. Bilateral symmetry of first symptom findings (negative predictor) 0.031 0.079
23. Time to recompression after surfacing 0.046 0.077
24. Thought aberration reported as secondary symptom (negative predictor) 0.036 0.074
25. Symmetrical upper extremity motor weakness as primary presenting symptom or finding (negative predictor)0.037 0.073
* Additional factors identified solely by logistic regression included: back pain as primary presenting symptom, headache as primary presenting
symptom (negative), presence of any symptoms before last dive, rapid ascent in dive exposure history (negative), time to recompression after
surfacing, nystagmus reported as secondary symptom, diffuse abdominal pain as primary presenting symptom, complete relief after all
recompressions, evolution (worsening) of primary presenting symptom, bowel or bladder dysfunction reported as secondary symptom, missed
decompression stop in recent dive exposure history, fine cerebellar signs reported as secondary symptom or finding, number of dives in dive
series, previous DCS in past medical history.
FACTORS IN DCI DIAGNOSIS—FREIBERGER ET AL.
1025Aviation, Space, and Environmental Medicine •Vol. 75, No. 12 •December 2004
Delivered by Ingenta to:
Aerospace Medical Association
IP : 24.106.206.82
Mon, 16 Jul 2007 20:07:46
DISCUSSION
Method Used
Linear and logistic regressions identified the same or
very similar factors as important to this expert panel in
their diagnosis of DCI. Although there was only ap-
proximately 50% homology among the factors identi-
fied by the two methods, when the factor lists are
compared it is reassuring to see that many of the non-
matching factors are very similar (complete relief after
all recompressions versus any relief after recompres-
sion). However, because a continuous probability scale
was used to score the vignettes, linear regression con-
served more information than the forced case-no case
dichotomy required to use logistic regression. This is
the primary argument for using the linear regression
derived factors if conclusions are to be drawn.
Factors for DCS
To make the diagnosis of DCS, the experts considered
the following categories of factors in order of impor-
tance: 1) type of symptoms at presentation; 2) onset
time; 3) the pressure time exposure; and 4) the response
to treatment. The expert’s responses indicated that in
making the diagnosis of DCS they placed great impor-
tance in observing symptoms typically associated with
type II DCS including neurological symptoms such as
numbness, paresthesias, or motor weakness. However,
bilateralism of symptoms was a negative factor for mo-
tor findings as well as for paresthesias. DCS does not
often present symmetrically in the opinion of these
experts. Onset time was critically important. DCS prob-
ability rapidly decreased with symptom onset times
greater than 2-3 h after a dive. The mean, median, and
mode for symptom onset time were 3 h, 2 h, and 1 h,
respectively, for the vignettes in the highest quartile of
the expert assigned probability range. Seizures, isolated
nausea and vertigo, the presence of a pre-existing car-
dio-pulmonary condition or diabetes prompted the ex-
perts to lower their probability assignment for DCS.
Presumably, this is because these symptoms are com-
monly caused by conditions other than DCS and these
classes of coexisting diseases can frequently cause DCS-
like symptoms. Although headache after a dive is often
attributed to DCS, in this study headache did not
achieve statistical significance as a diagnostic factor for
the expert raters. Constitutional symptoms such as ex-
treme fatigue were important as were the presence of
ataxia or cerebellar signs. Also, an unexplained skin
rash was considered noteworthy. Pressure and time
exposure were important to the experts. The experts
indicated that they believed a minimum dive depth and
time is required to contract DCS and a missed a decom-
pression stop was an important positive diagnostic fac-
tor. Finally, the experts placed importance in the diver’s
response to recompression treatment. Relief of symp-
toms after recompression was a positive factor in as-
signing a diagnosis of both DCS and AGE.
Factors for AGE
To make the diagnosis of AGE the experts considered
the following categories of factors in order of impor-
tance: 1) symptom onset time; 2) loss of consciousness
or other neurological symptoms; 3) a seizure as the
presenting symptom; and 4) a plausible history of a
rapid decompression such as a rapid ascent from depth.
For the most important factor, symptom onset time, the
vignettes in the upper half of the expert assigned prob-
ability range for AGE had a mean symptom onset time
of 12 min and a median and mode of 0 min for both.
After symptom onset time, the presence of neurological
symptoms, such as loss of consciousness, motor weak-
ness, and gait ataxia were strong positive factors in
making the diagnosis of AGE. As opposed to DCS, the
occurrence of seizures was a positive factor for the
expert diagnosis of AGE. A history of a rapid ascent
and a positive response to treatment were also impor-
tant. Vignettes with asthma as a pre-existing medical
condition were rated by the experts as having a slightly
higher AGE probability score than those without
asthma. However, dive depth and time were not im-
portant to the experts in making a diagnosis of AGE.
Factors Without Impact on the Experts Decision
When the experts made their diagnostic decisions
about either DCS or AGE, with the exception of asthma,
cardio-pulmonary disease, and diabetes, they did not
use pre-existing medical conditions. The significance
values are greater than 0.05 and the beta weights are
TABLE II. THE 10 FACTORS MOST IMPORTANT TO THE EXPERTS IN MAKING THE DIAGNOSIS OF ARTERIAL GAS EMBOLISM
(AGE). THE FACTORS IN BOLD WERE IDENTIFIED BY BOTH LOGISTIC AS WELL AS LINEAR REGRESSION.*
Factor sig Absolute value of beta
1. Onset time in min after surfacing of 1st symptom (longer onset time was a negative predictor) 0 0.37
2. Altered consciousness reported as secondary symptom 0 0.35
3. Neuro symptom (any type) as primary presenting symptom 0 0.266
4. Motor weakness (anywhere) reported as secondary symptom 0 0.196
5. Seizure as primary presenting symptom 0 0.191
6. Rapid ascent in dive exposure history 0.001 0.182
7. Memory aberration reported as secondary symptom (negative predictor) 0.029 0.151
8. Gait ataxia reported as secondary symptom or finding 0.03 0.117
9. Any relief after recompression treatment 0.033 0.114
10. Asthma in past medical history 0.046 0.105
* Additional factors identified solely by logistic regression included: any relief after recompression treatment, altered consciousness as primary
presenting symptom, pneumothorax reported as secondary finding, seizure reported as secondary symptom, thought aberration reported as
secondary symptom, vertigo (any symptom).
FACTORS IN DCI DIAGNOSIS—FREIBERGER ET AL.
1026 Aviation, Space, and Environmental Medicine •Vol. 75, No. 12 •December 2004
Delivered by Ingenta to:
Aerospace Medical Association
IP : 24.106.206.82
Mon, 16 Jul 2007 20:07:46
low for gender, hypertension, back disease, chronic
joint pain, advanced age, and obesity. The maximum
depth of the last dive was a significant factor in the
diagnosis of DCS but not AGE. A history of rapid ascent
or air exhaustion was an important factor influencing
the diagnosis of AGE, but not DCS. Previous DCS was
important to the experts in making the diagnosis of
DCS but not AGE.
Reliability
Both systematic as well as random error potentially
degraded the model. The systematic sources of error
included raters selecting different factors as important
and giving them different weights or using a different
rating method. Random error sources include rater fa-
tigue and missed information. Repeating 3 of the 200
vignettes tested reliability, the degree with which re-
sults can be reproduced. Although it is recognized that
subsequent ratings may be contaminated by knowledge
of earlier ratings, the intra-class correlations indicated
moderate internal consistency (reliability). The experts
were not always in agreement with each other about the
disease probability for the individual vignettes. We
measured this by calculating inter-rater agreement. In
this study, the inter-rater agreement was moderate and
ranged from 0.49 to 0.61. However, as was mentioned
before, some of the cases were diagnostically ambigu-
ous purposely to help identify individually important
factors in the expert’s decision-making process. To put
these findings in context, Edmonds et al. found a level
of inter-rater agreement similar to ours when two
trained observers measured vital signs in 140 patients
presenting to an emergency department with acute
complaints (2). In their study, the inter-rater agreement
was between 0.40 and 0.60 for detecting systolic hyper-
tension and less than 0.40 in detecting systolic hypoten-
sion or tachypnea. Our level of inter-rater agreement is
consistent with Edmonds’s findings and within accept-
able bounds. A possible criticism of our check of reli-
ability measures is that we did not check for a rater ⫻
factor effect. If one expert always relied on a particular
factor to assign his probability, then by consistently
doing so he would raise the importance of that partic-
ular factor entirely by himself, skewing the results. In
addition, we do not know if we had, within our expert
panel, raters who used very different sets of criteria that
we were unable to measure. We may have had a pos-
sible “Dr. Goldstandard” who always rated the cases
correctly paired with a “Dr. Doesn’t-know” who always
rated incorrectly. This would unavoidably contribute to
a decrease in inter-rater agreement and reliability.
However, even if some raters were better than others in
terms of skill and consistency, overall we were able to
accurately discriminate the large and statistically signif-
icant effects.
Conclusions and Future Directions
This study was an attempt to better define the factors
important in the diagnosis of DCS and AGE to help
decrease ambiguity and improve scientific communica-
tion. Ambiguity in the diagnosis of DCS and AGE is
more of a research problem than a clinical one. When
the indicated treatment is benign, tests of low specificity
yet high sensitivity are not medically harmful. Hyper-
baric oxygen treatment is relatively safe. Other than
delaying definitive treatment, recompression of false
positive cases of DCS or AGE is not harmful. In most
practices, if a diver has even marginal symptoms or
signs after decompression he is usually assumed to
have DCS and treated. The problem arises when an
investigator tries to use clinical records to advance
knowledge in the field. Low specificity and high sensi-
tivity result in “over-diagnosis bias” which limits con-
clusions about treatment effectiveness or prevention
strategies. In addition, an economic burden is imposed
on patients as well as treatment facilities when therapy
is based on non-specific criteria and over-diagnosis re-
sults. At the very least, consensus regarding diagnostic
criteria would lead to more focused scientific inquiry
about mechanisms, prevention and treatment.
A common method used to reach consensus is the
Delphi process, a procedure developed by the Rand
corporation in the 1950s (9). The original Delphi tech-
niques were research methods designed to use experts
in a particular field to standardize decision-making
where published information was inadequate or non-
existent. In the standard Delphi process, a group of
experts use either the literature or their own experience
to select a list of individual factors they believe influ-
ence the outcome in question. The experts make a con-
scious decision about how factors will be ranked and
then strive to reach a consensus through multiple meet-
ings and discussion. Questions have been developed to
assist the experts in prioritizing their responses and
methods are used to assure anonymity so that the un-
checked force of a strong personality does not over-
whelm the process. However, the Dephi process re-
quires a strong commitment from the experts involved.
It is very time consuming and achieving consensus
requires cooperation.
Although our study also was an attempt to achieve
consensus, we approached it in a different manner. The
experts were allowed to choose whatever factors they
liked and individually make their decisions about their
importance. Our process was more efficient in that no
repeated meetings were required. In addition, our pro-
cess possibly had greater power since numerous repli-
cates (200 vignettes x 11 raters) were presented allow-
ing for the disentanglement of confounders and
mediators. Because the probabilities assigned to each
vignette by each expert were averaged, the influence of
each expert’s opinion carried equal weight in the final
outcome. Our model represents the “first round” con-
sensus opinion of the expert group in a Delphi-like
process. However, because it is only a “first round,” the
experts were not able to benefit from an organized
discussion of the topic until after the vignettes were
rated and the actual symposium had taken place.
In conclusion, our study showed that certain specific
factors influence the expert diagnosis of DCS and AGE.
Identification of these factors is a first step in codifying
a standard methodology for diagnosis of a condition
without an objective laboratory test. Collectively decid-
FACTORS IN DCI DIAGNOSIS—FREIBERGER ET AL.
1027
Aviation, Space, and Environmental Medicine •Vol. 75, No. 12 •December 2004
Delivered by Ingenta to:
Aerospace Medical Association
IP : 24.106.206.82
Mon, 16 Jul 2007 20:07:46
ing how these factors are used is the next step in this
important process.
REFERENCES
1. Ball R. Effect of severity, time to recompression with oxygen, and
re-treatment on outcome in forty-nine cases of spinal cord
decompression sickness. Undersea Hyperbaric Med 1993; 20:
133–45.
2. Edmonds ZV, Mower WR, Lovato LM, Lomeli R. The reliability of
vital sign measurements. Ann Emerg Med 2002; 39:233–7.
3. Francis TJ, Dutka AJ, Flynn ET. Experimental determination of
latency, severity, and outcome in CNS decompression sick-
ness. Undersea Biomed Res 1988; 15:419–27.
4. Freiberger J, Denoble PJ, Vann R, et al. The association of present-
ing symptoms of DCI with residual neurological abnormalities
after treatment. [Abstract]. Undersea Hyperbaric Med 2001;
28(2001 Supplement):69.
5. Moon RE, Sheffield PJ. Guidelines for treatment of decompression
illness. Aviat Space Environ Med 1997;68:234–43.
6. Neuman TS, Harris MG, Linaweaver PG, Jr. Blood viscosity in
man following decompression: correlations with hematocrit
and venous gas emboli. Aviat Space Environ Med 1976; 47:
803–7.
7. Norusis MJ. Linear regression and correlation, testing regression
hypotheses, analyzing residuals and building multiple regres-
sion models. In: SPSS 7.5 guide to data analysis. Upper Saddle
River, N.J.: Prentice Hall; 1997: 363–497.
8. Nyquist PA, Dick EJ, Jr., Buttolph TB. Detection of leukocyte
activation in pigs with neurologic decompression sickness.
Aviat Space Environ Med 2004; 75:211–4.
9. Pill J. The Delphi method: Substance, context, a critique and an
annotated bibliography. Socio-Economic Planning Sciences
1970; 5:57–71.
10. Shrout P, Fleiss JL. Intraclass correlations: uses in assessing rater
reliability. New York: Wiley; 1979: 420–8.
11. Smith RM, Neuman TS. Abnormal serum biochemistries in asso-
ciation with arterial gas embolism. J Emerg Med 1997; 15:
285–9.
12. Sykes JJ, Hallenbeck JM, Leitch DR. Spinal cord decompression
sickness: a comparison of recompression therapies in an ani-
mal model. Aviat Space Environ Med 1986; 57:561–8.
13. Vann RD, ed. Report on decompression illness, diving fatalities
and project dive exploration: DAN’s annual review of rec-
reational scuba diving injuries and deaths based on 2000
data. 2002 edition. Durham, NC: Diver’s Alert Network;
2002:88
14. Vann RD, Denoble P, Emmerman MN, Corson KS. Flying after
diving and decompression sickness. Aviat Space Environ Med
1993; 64(9, Pt 1):801–7.
15. Vann RD, Uguccioni D, M, eds. Report on decompression illness,
diving fatalities and project dive exploration: DAN’s annual
review of recreational scuba diving injuries and deaths based
on 1999 data. 2001 edition. Durham, NC: Diver’s Alert Net-
work; 2001:21, 35–59
FACTORS IN DCI DIAGNOSIS—FREIBERGER ET AL.
1028 Aviation, Space, and Environmental Medicine •Vol. 75, No. 12 •December 2004