You are on page 1of 2

British Journal of Anaesthesia, ▪ (▪): 1e2 (2018)

CORRESPONDENCE

Hyperoxia is a modifiable anaesthetic risk factor that varies in the


practice of individual anaesthetists
A. K. Staehr-Rye1,2,*, T. Kurth3, F. T. Scheffenbichler1, L. S. Rasmussen4 and
M. Eikermann1,5
1
Department of Anaesthesia, Critical Care, Pain Medicine, Massachusetts General Hospital, 55 Fruit Street,
Boston, MA 02114, USA, 2Department of Anaesthesiology, Herlev and Gentofte Hospital, University of

Copenhagen, Herlev Ringvej 75, 2730 Herlev, Denmark, 3Institute of Public Health, Charite
Universitatzmedizin Berlin, Germany, 4Department of Anaesthesia, Centre of Head and Orthopaedics,
Rigshospitalet, University of Copenhagen, Denmark and 5Klinik fur Anaesthesie und Intensivmedizin,
Universitaetsklinikum Essen, Essen, Germany
*Corresponding author. E-mail: akstaehr1@hotmail.com

We thank Acka1 and de Jonge and colleagues2 for their in- the preference of the anaesthetist. Additional data show that
terest in our study analysing the association between intra- this assumption does not hold true. To quantify the distribu-
operative inspiratory oxygen fraction and risk of respiratory tion of inter-provider variability in utilization of high FIO2, we
complications.3 We agree with both that for cause and effect plotted the proportion of cases where the individual anaes-
relationships, more targeted research is needed. We do not thesia provider administered high FIO2 defined as median FIO2
follow the arguments of Acka that the results are probably >0.63 (upper quintile), a statistical approach we have reported
caused by an epiphenomenon triggering high FIO2 rather than before.4e6 The data show that the median FIO2 varied widely
an effect of high FIO2. Acka mentions that patients receiving across practitioners within our cohort. A small number of
the highest FIO2 were different from the patients receiving anaesthesia providers consistently used lower FIO2, whilst
lower FIO2 in terms of several characteristics such as comor- another group regularly used higher FIO2. Similar distributions
bidities, volatile anaesthetic dose, duration of surgery, pro- were observed before and after propensity-matched adjust-
cedural severity, etc. That is correct. We anticipated that a ment using covariates that can influence FIO2 (including pro-
high FIO2 would be used in the sickest patients, and in the cedure duration, procedural severity score, diagnosis of
primary analysis we therefore adjusted for several covariates chronic obstructive pulmonary disorder before surgery,
including procedural risks, comorbidities, and anaesthesia- admission status, PEEP, tidal volume, surgical service, body
related risk factors. In a subgroup analysis, we included mass index, weight, and minimum alveolar concentration of
adjustment for minutes with hypotension and oxygenation inhaled anaesthetics) suggesting that this variability is
(ratio of PaO2 and FIO2) as additional potential confounding driven more by individual preference than by patient or
factors, and this did not change the results. These analyses procedure characteristics. That is concerning because
show that the association between FIO2 and respiratory harmful effects of high FIO2 have been detected in a number
complications is robust and not an epiphenomenon of a high of studies.6e8
FIO2 requirement. When analysing blood gas samples in De Jonge and colleagues are concerned about suboptimal
approximately 6000 patients, we saw that >30% of the pa- confounder control. Our statistical analyses have been a priori
tients in the high FIO2 group had in fact a PaO2 above 40 defined and all sensitivity analyses have been clearly indi-
kPadthis hyperoxia indicates that these patients did not need cated. Our rationale for confounding control is strictly based
this high FIO2. on clinical knowledge, and we have not used data mining for
Acka assumes that the concentration of FIO2 is mainly any aspect of the analysis. We are not concerned about
driven by patient and procedural characteristics and not by collinearity; we exclude any collinearity related bias by

© 2018 British Journal of Anaesthesia. Published by Elsevier Ltd. All rights reserved.
For Permissions, please email: permissions@elsevier.com

1
2 - Correspondence

analysing the condition index and the variance decomposition References


matrix. Our methods follow modern causal inference princi-
ples, and we have conducted several sensitivity analyses to 1. Acka O. High intraoperative oxygen utilizationdcontext
evaluate the robustness of our findings. With the availability of matters. Br J Anaesth [in January issue].
large datasets, powerful statistical analysis tools are currently 2. de Jonge SW, Posthuma LM, Boldingh QJ, Boermeester MA,
applied in many fields of medicine.9 We have taken a careful Hollmann MW. Reverse causation. Br J Anaesth 2018
look at our data and have conducted additional analyses as [in this issue]
suggested by the reviewers of our manuscript. For instance, 3. Staehr-Rye AK, Meyhoff CS, Scheffenbichler FT, et al. High
we modified the categorization of the outcome and our results intraoperative inspiratory oxygen fraction and risk of
did not change. Overall, our results are very robust and we major respiratory complications. Br J Anaesth 2017; 119:
have no reason to believe that confounding or misclassifica- 140e9
tion is driving our results. 4. Ladha KS, Bateman BT, Houle T, et al. Variability in the
We have carefully tried to isolate the impact of intra- use of protective mechanical ventilation during general
operative FIO2 on postoperative outcomes in the context of anaesthesia. Anesth Analg July 2017. https://doi.org/
many other factors that might also be related to that risk.10,11 10.1213/ANE.0000000000002343 [epub ahead of print]
The study was conducted on a large sample of diverse pa- 5. Shin CH, Long DR, McLean D, et al. Effects of intraoperative
tients undergoing surgery and represents the typical patient fluid management on postoperative outcomes. Ann Surg
undergoing surgery at our institutions (i.e. there is no re- Mar 2017. https://doi.org/10.1097/SLA.0000000000002220
striction of range of predictor classes). We observed a sub- [epub ahead of print]
stantial variability in treatment, which is needed to make 6. Fonnes S, Go € genur I, Søndergaard ES, et al. Perioperative
meaningful comparisons in observational research. It is not hyperoxiadlong-term impact on cardiovascular compli-
possible to explain why a high FIO2 was used and this natu- cations after abdominal surgery, a post hoc analysis of the
rally leads to some speculation and interesting hypotheses PROXI trial. Int J Cardiol 2016; 215: 238e43
that should be tested in future research. Finally, the charac- 7. Girardis M, Busani S, Damiani E, et al. Effect of conser-
teristics of the groups being compared have a reasonable vative vs conventional oxygen therapy on mortality
amount of overlap. Thus, we are convinced that our study has among patients in an intensive care unit the oxygen-ICU
generated important new information for researchers and randomized clinical trial. JAMA 2016; 316: 1583e9
clinicians alike. 8. Kilgannon JH, Jones AE, Shapiro NI, et al. Association be-
In summary, our data demonstrate that hyperoxia is an tween arterial hyperoxia following resuscitation from
anaesthetic risk factor that varies in the practice of individual cardiac arrest and in-hospital mortality. JAMA 2010; 303:
anaesthetists. As this factor is potentially modifiable, we 2165e71
encourage other groups to focus research on this factor. We 9. Herna  n MA, Robins JM. Using big data to emulate a target
believe that our data provide evidence for avoiding the trial when a randomized trial is not available. Am J Epi-
routine use of high FIO2dwe should rather titrate the con- demiol 2016; 183: 758e64
centration of this drug that has desirable and unwarranted 10. Brueckmann B, Villa-Uribe JL, Bateman BT, et al. Devel-
effects. opment and validation of a score for prediction of post-
operative respiratory complications. Anesthesiology 2013;
118: 1276e85
11. Canet J, Gallart L. Postoperative respiratory failure: path-
Declaration of interest ogenesis, prediction, and prevention. Curr Opin Crit Care
None declared. 2014; 20: 56e62

doi: 10.1016/j.bja.2017.12.028

You might also like