Transcript - LWW Journals

Issue Highlights
Journal of Trauma and Acute Care Surgery
Volume 78, Issue 4
April 2015
Jo Fields: Welcome to Trauma Loupes, the Journal of Trauma and Acute
Care Surgery’s monthly podcast. This is Jo Fields. For this April 2015 issue,
we hear from our editor in chief, Dr. Ernest Moore, and our biostatistician,
Dr. Angela Sauaia, with their must reads. First, Dr. Moore:
Dr. Gene Moore:
Welcome to the April issue of the Journal. The lead articles are the AAST
Presidential Address entitled “Responsibility” by Dr. Bill Cioffi and the Fitts
Lecture “Genomics of Injury: The Glue Grant Experience” by Dr. Ron
Tompkins. Both are must-reads as they contain invaluable material that all
acute care surgeons should know, but too much to effectively summarize
in a podcast.
The first paper to discuss is by Dr. Marty Schreiber and colleagues of the
Resuscitation Outcome Consortium who completed a randomized trial
comparing prehospital limited crystalloid versus standard crystalloid
resuscitation. The limited cohort received 250 ml boluses of normal saline
to maintain a SBP > versus a target of 110 mm of mercury until in the
control group. Interestingly, the difference in crystalloid administered was
relatively small, one versus two liters. In the blunt injured cohort, the 24
hours mortality was 3% in the limited versus 18% in the standard with an
adjusted odds ratio of 0.17; a 95% confidence interval of 0.03 to 0.92. For
penetrating wounds, there was no difference in mortality. Unfortunately,
this trial was terminated at 192 patients because of a new ROC study. One
of the other limitations was that the median ISS was only nine in these
study patients. Nonetheless, as the authors conclude, this trial indicates
that limited resuscitation is feasible and warrants a large-scale phase III
trial.
The next paper that is particularly timely is by Dr. Tatsuyo Norii and
colleagues who examine the impact of resuscitation endovascular balloon
occlusion of the aorta on survival following blunt trauma using the
Japanese Trauma Data Bank from 2004-2011. Of more than 45,000
patients in the registry, 1% underwent REBOA. Propensity scoring was
employed, matching REBOA to non-REBOA patients in a ratio of 1 : 5. The
provocative finding was that the odds ratio of survival in the REBOA group
was 0.30 with a 95% confidence interval of 0.23 to 0.40. Now it is
important to emphasize that in Japan postinjury REBOA is typically placed
by emergency physicians, with trauma surgeons on call from out of the
hospital.
© 2015 American Association for the Surgery of Trauma
The Journal of Trauma and
Acute Care Surgery
Issue Highlights
Volume 78, Issue 4
April 2015
While the response to trauma is clearly different in Japan, this study has
implications as we continue to integrate REBOA into the resuscitation of
critically injured patients in the United States
Another timely report is by Dr. Peter Rhee and colleagues from the
University of Arizona and the University of Southern California who review
their experience with transfusion of auto transfused whole blood from
hemothoraces, collected with chest tubes and anticoagulated with CPD.
Patients with whole blood auto transfusion numbered 136 were compared
to those who did not, employing a propensity score matching at a 1 : 1
ratio. Patients with auto transfusion required less units of red cells (7
versus 9) and platelets (5 versus 7), although there were no differences in
other outcome measures. This is a particularly interesting report in light of
the ongoing discussion of the potential benefits of fresh whole blood
transfusion to mitigate trauma-induced coagulopathy. But it is important
to recognize that blood collected from a hemothorax is quantitatively
different from whole blood obtained from a venous catheter, and there
has been concern of adverse effects on the coagulation system when large
volumes of intracavitary-collected blood are transfused.
The final paper to highlight is authored by Dr. Ben Howard and colleagues
from the San Francisco General who fundamentally demonstrate that
patients with isolated hypoxemia (Pa02/Fi02 < 300) are different from
those who are hypoxemia and meet Berlin criteria for ARDS, that is,
bilateral pulmonary opacities on chest x-ray. Of 621 injured patients
requiring mechanical ventilation, 65% developed hypoxemia and 30% ARDS
with hypoxemia. Examining the three groups (no hypoxemia, isolated
hypoxemia, and ARDS), there was a progressive increase in severity of
shock, severity of overall injury, and severity of chest trauma as well as
incidences of pneumonia, multiple organ failure, ventilator days, ICU stay,
and mortality – as expected. However, in multiple logistics regression,
independent predictors of hypoxemia were early plasma from (zero to six
hours), delayed crystalloid from (7 to 24 hours), and head and chest injury;
whereas, predicators of ARDS were early crystalloid (zero to six hours),
delayed platelets (7 to 24 hours) and head and chest injury. In sum, the
authors conclude that hypoxemia and ARDS represent distinct clinical
states, with unique predictors that should be recognized in designing
therapeutic strategies. As usual, there is abundant other important
information in this issue. Happy reading!
Jo Fields:
Thank you Dr. Moore. And now, Dr. Sauaia:
© 2015 American Association for the Surgery of Trauma
The Journal of Trauma and
Acute Care Surgery
Issue Highlights
Volume 78, Issue 4
April 2015
Dr. Angela Sauaia: Hello, everyone. Today I would like to acknowledge two
important topics represented in this April issue of our Journal. The first is
the increasing number of articles addressing the elderly population. As I
age myself, a nice alternative to pushing daises, I too became more
interested in this fast growing segment of our population. The research
design and analysis of studies addressing older populations present
interesting challenges.
In fact, a recent article in JAGS, the Journal of the American Geriatric
Association, goes as far as proposing a new sub-discipline named
“gerontologic biostatistics”. This may be going too far, but I certainly
advocate for a special attention to important characteristics of the elderly
as it pertains to research design and statistical analysis. For example,
among the elderly, we have a larger representation of patients who are
receiving anti-coagulants and get injured by a fall from standing. In this
group, two risk factors, i.e., the mechanism of injury and the comorbidity
interact and modify each other.
Another challenge is the multiple risk factors, which represent a challenge
in research design, as they require a large sample size to allow stratified
analysis for risk adjustment. Interventions designed for the older
populations often need to be multicomponent to address the multiple risk
factors. In traditional trials, all participants are eligible for all intervention
components, by contrast, in intervention trials addressing geriatric
participants, the number and type of risk factors and contraindications
vary. Thus, some older patients may be eligible for one component of the
intervention but not for the other components. A practice that is becoming
more popular is called “standardly-tailored”, in which patients randomized
to the intervention arm receive only the intervention components that is
appropriate at the time of enrollment and not the other components.
Another interesting topic in geriatric trauma research is the need to use
multiple outcomes, including cognitive and physical outcomes as well as
quality of life, frailty, activities of daily living, and others. The use of
multiple outcomes can present statistical challenges, similar to the
challenges of multiple comparisons.
The second topic I want to address today is propensity score matching. I
have talked about this issue before, but we can talk again, as we have two
popular studies using this technique in our April issue. Propensity score
matching has been proposed as a good approximation of a randomized
clinical trial. It entails forming matched sets of treated and untreated
subjects who share a similar value of the propensity score. The propensity
score is the probability of receiving the treatment in question based on
observed baseline covariates
© 2015 American Association for the Surgery of Trauma
The Journal of Trauma and
Acute Care Surgery
Issue Highlights
Volume 78, Issue 4
April 2015
. For example, in the study by Dr. Norii and colleagues from Japan on
REBOA, patients who were submitted to REBOA were matched to patients
who had similar propensity score but who, for some reason, were not
submitted to REBOA. Their propensity score was obtained by a regression
model that included vital signs, age, sex, anatomic injuries and physiology
derangement. Traditionally one would match based on these baseline
characteristics. For example, a 29-year-old male with pelvic fracture,
asystolic who underwent REBOA placement would be paired with a 29 year
old male with a pelvic fracture also asystolic who was not treated with
REBOA.
It is pretty intuitive that finding these pairs is not going to be easy,
therefore, we can use a propensity score. The propensity score comes from
a regression equation that enters all the variables and produces a
probability, much like the TRISS methodology that we now name
propensity score. This score can then be used to pair similar, but not
identical patients. There are several techniques for propensity score
matching. First, one must choose between matching with replacement and
matching without replacement. When using matching without
replacement, once an untreated subject has been selected to be matched
to a given treated subject, that untreated subject is no longer available for
matching. As a result, each untreated subject is included in at most one
matched set. In contrast, matching with replacement allows a given
untreated subject to be included in more than one matched set. When
matching with replacement is used, the statistical analysis must account for
the fact that the same untreated subject may be in multiple pairs. A second
choice is between greedy and optimal matching. In optimal matching,
matches are formed to minimize the within-pair difference of the
propensity score. While in greedy matching, the nearest untreated subject
is selected for matching to the given treated subject.
Now, what do I mean by nearest matching score? There are two primary
methods for this: nearest neighbor matching and nearest neighbor
matching within a specified caliper distance. Nearest neighbor matching
selects the untreated subject whose propensity score is closest to that of
the treated subject. If multiple untreated subjects have propensity scores
that are equally close, one of them is chosen at random. Nearest neighbor
matching within a specified caliper distance is similar with the further
restriction that the absolute difference of the propensity scores need to be
within some pre-specified distance, namely the caliper distance. Therefore,
for a given treated subject, one would identify all the untreated subjects
whose propensity score lay within that specified distance of the treated
subject, and from this set of untreated subjects, the untreated subject
whose propensity score was closest to that of the treated subject would be
selected for matching. Ugh….so many things to think about
© 2015 American Association for the Surgery of Trauma
The Journal of Trauma and
Acute Care Surgery
Issue Highlights
Volume 78, Issue 4
April 2015
! But now, the most important question? Is propensity score matching
really a substitute for randomized trials? The answer is, as usual, it
depends. For situations where a RCT is impossible or inappropriate, the
propensity score matching offers a good alternative. Classic cases where
RCTs are not possible are heavy drug use or gun possession, where one
cannot randomize people to be a heavy drug user or to become a gun
owner. However, we must keep in mind that a propensity score is limited
to observed variables that are included in the model used to derive the
score. A randomized trial accounts for variables we have not thought of
beforehand by randomly distributing patients to study groups, satisfying
what is called the ignorability of assignment assumption.
Propensity score matching satisfies this necessary assumption only if we
accounted for all confounders. How often do we account for all
confounders? Well, a recent systematic comparison for propensity score
matching observational and randomized trials in acute coronary syndromes
published in the European Heart Journal, found that observational studies
using propensity-scoring methods produced larger treatment effect
estimates compared with those from randomized clinical trials, however
the differences were rarely statistically significant. Therefore, to our
readers, the guidance I give is: look at the variables included in the
propensity score model and see if you are satisfied with them. Well, I think
that is more than enough statistics for one podcast. See you next month!
Jo Fields:
Thank you Dr. Sauaia. And thank ya’ll for listening. We will be back in a
month for the May issue. If you have any questions or requests, please
send them to [email protected].
© 2015 American Association for the Surgery of Trauma