British Journal of Anaesthesia, 122 (1): 131e140 (2019)

doi: 10.1016/j.bja.2018.08.027
Advance Access Publication Date: 27 October 2018
Quality and Patient Safety

Non-opioid analgesic modes of pain management are

associated with reduced postoperative complications
and resource utilisation: a retrospective study of
obstructive sleep apnoea patients undergoing
elective joint arthroplasty
C. Cozowicz1,2, J. Poeran3, N. Zubizarreta3, J. Liu1, S. M. Weinstein1,
L. Pichler1,2, M. Mazumdar3 and S. G. Memtsoudis1,2,*
1
Department of Anesthesiology, Critical Care and Pain Management, Hospital for Special Surgery, Weill
Cornell Medical College, New York, NY, USA, 2Department of Anesthesiology, Perioperative Medicine and
Intensive Care Medicine, Paracelsus Medical University, Salzburg, Austria and 3Institute for Healthcare
Delivery Science, Department of Population Health Science and Policy, Icahn School of Medicine at Mount
Sinai, New York, NY, USA

*Corresponding author. E-mail: memtsoudiss@hss.edu

Abstract
Background: Studies on the effectiveness of multimodal analgesia, particularly in patients at higher perioperative risk
from obstructive sleep apnoea (OSA), are lacking. We aimed to assess the impact of multimodal analgesia on opioid use
and complications in this high-risk cohort.
Methods: We conducted a population-based retrospective cohort study of OSA patients undergoing elective lower ex-
tremity joint arthroplasty (2006e16, Premier Healthcare database). Multimodal analgesia was defined as opioid use with
the addition of one, two, or more non-opioid analgesic modes including, nonsteroidal anti-inflammatory drugs (NSAIDs),
cyclooxygenase-2 inhibitors, paracetamol/acetaminophen, peripheral nerve blocks, steroids, gabapentin/pregabalin, or
ketamine. Multilevel multivariable regression models measured associations between multimodal analgesia and opioid
prescription (primary outcome; oral morphine equivalents). Secondary outcomes included opioid- and OSA-related
complications, and resource utilisation. Odds ratios (OR) or % change and 95% confidence intervals (CI) are reported.
Results: Among 181 182 OSA patients included, 88.5% (n ¼ 160 299) received multimodal analgesia with increasing uti-
lisation trends. Multivariable models showed stepwise beneficial postoperative outcome effects with increasing addi-
tional analgesic modes compared with opioid-only analgesia. In patients who received more than two additional
analgesia modes (n ¼ 64 174), opioid dose prescription decreased by 14.9% (CI 17.0%; 12.7%), while odds were
significantly decreased for gastrointestinal complications (OR 0.65, CI 0.53; 0.78), mechanical ventilation (OR 0.23, CI 0.16;
0.32), and critical care admission (OR 0.60, CI 0.48; 0.75), all P<0.0001.

Editorial decision: 21 August 2018; Accepted: 21 August 2018

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

131
 132 - Cozowicz et al.

Conclusions: In a population at high risk for perioperative complications from OSA, multimodal analgesia was associated
with a stepwise reduction in opioid use and complications, including critical respiratory failure.

Keywords: arthroplasty; hip replacement; knee replacement; sleep apnoea, obstructive; multimodal analgesia; post-
operative complications; postoperative outcome; opioids

exempt from requirements for consent, as data are de-

Editor’s key points identified and compliant with the Health Insurance Porta-
 Obstructive sleep apnoea may increase the risk of bility and Accountability Act.
perioperative complications, but the impact of multi-
modal analgesia on this is unknown. Analysis plan
 In this large retrospective cohort study of patients with
The analysis plan was established at the onset of the study
obstructive sleep apnoea undergoing lower limb joint
(before data access) to define the population and intervention
arthroplasty, most received some form of multimodal
of interest, baseline variables of significance, and healthcare
analgesia.
outcomes of importance in OSA patients.
 Use of multimodal analgesia is associated with reduced
perioperative complications and opioid use, with the
Participants
greatest benefit coming from increasing analgesic
modes. We included patient records with a diagnosis of OSA who
underwent elective, primary total hip arthroplasty (THA) or
total knee arthroplasty (TKA); ICD-9-CM codes 81.51 and 81.54,
Given the prospect of improved postoperative pain control and respectively. ICD-9-CM definitions for the presence of OSA,
reduced opioid consumption, multimodal analgesia seems comorbidities, and complications are presented in Appendix
particularly appealing in patients at higher perioperative risk Tables 3e5.8
because of conditions such as obstructive sleep apnoea (OSA).
Sedative and opioid-based medications account for a major Variables
concern in OSA,1 as these drug categories interfere with res-
piratory drive and alertness.2e4 Thus, current perioperative Baseline characteristics
OSA recommendations include multimodal analgesia.5 A According to the pre-specified analysis plan, the following
recent study demonstrated incremental benefits with the use baseline characteristics were investigated. Patient-related
of additional analgesic modes in joint replacement surgery.6 variables included age, gender, race (white, black, other) and
Nevertheless, evidence on a population level to assess bene- insurance type (commercial, Medicaid, Medicare, uninsured,
fits of this practice in patients at higher perioperative risk other). Hospital-related variables included, location (urban
because of OSA is largely lacking. We therefore investigated rural), size (<300, 300e499, 500 beds), teaching status, and
the effectiveness of multimodal analgesia among patients number of annual hip/knee arthroplasties. Procedure-related
with OSA undergoing lower extremity joint arthroplasty sur- variables included, year of procedure, use of general or neu-
gery. We analysed associations between multimodal analgesia raxial anaesthesia, and use of patient-controlled analgesia
and opioid use, opioid-related side-effects, OSA characteristic (PCA). The comorbidity burden was assessed using the Quan
adverse outcomes, and resource utilisation. We hypothesised adaptation of the Charlson-Deyo Comorbidity Index.9 Vari-
that the addition of non-opioid analgesic modes to opioid ables describing history of substance use/abuse,10e12 chronic
analgesia is associated with a reduction in opioid consump- pain conditions,13 and psychiatric conditions12 were included
tion and complications. as possible drivers of perioperative outcome, particularly
through their correlation with opioid utilisation.
Methods
Exposure
Study design and data source
The exposure of interest was the use of multimodal analgesia,
In this population-based retrospective cohort study, we
while the comparator was the utilisation of opioid-only anal-
extracted data (from January 2006 to December 2016) from the
gesia. Multimodal analgesia was defined as opioid analgesia plus
Premier Healthcare Database7 (Premier Healthcare Solutions,
the addition of non-opioid analgesic modes on the day of surgery
Inc., Charlotte, NC, USA), a national claims-based database
or the day after, and categorised into the addition of one, two, or
with patient-specific inpatient billing information covering
more than two non-opioid analgesic modes. Non-opioid anal-
approximately 25% of US hospitals. Diagnoses and services
gesic modes included, peripheral nerve blocks, paracetamol/
provided were determined through International Classifica-
acetaminophen, steroids, gabapentin/pregabalin, ketamine,
tion of Diseases, Ninth Revision, Clinical Modification codes
NSAIDs, or cyclooxygenase-2 (COX-2) inhibitors, administered
(ICD-9-CM, US health system’s adaption of the international
on the day of surgery or the day thereafter.
ICD-9 standard), Current Procedural Terminology codes, and
standardised billing items. Approval was obtained by the
Outcomes
Institutional Review Boards of the Hospital for Special Surgery
(New York, NY, USA; #2012-050-CR2) and Icahn School of The primary outcome was inpatient opioid prescription dose
Medicine (New York, NY, USA; #14-00647). This study was (extracted from billing); this was analysed separately as opioid
 RRH: Multimodal analgesia in obstructive sleep apnoea - 133

cases in hospitals with <30 primary lower extremity joint re-

placements (to ensure sufficient sample size per cluster18), no
OSA diagnosis, absent billing for perioperative opioids (this
may reflect missing data on opioids more than purposeful
omission of opioid use, which is rare), and opioid prescription
greater than the 95th percentile to exclude outliers. Figure 1
depicts the patient flow chart.

Statistical analysis
Univariable associations between the number of modes used
and study variables, and outcomes were analysed using the
c2 test for categorical and the Kruskal-Wallis test for
continuous variables. Multilevel, multivariable regression
models (logistic and linear regression for binary and contin-
uous outcomes, respectively) measured the association be-
tween multimodal analgesia categories (compared with
opioids only) and outcomes. On the hospital level, models
controlled for the correlation of patients within hospitals and
fit separate regression lines by hospital,19 because patients
within the same hospital may be correlated, because of
similarities in treatment care. Furthermore, within the
multilevel model, we controlled for hospital-related vari-
ables. On the patient level, multivariable regression models
were adjusted for all patient-related variables, insurance
type, procedure-related variables, Charlson-Deyo Comorbid-
ity index, history of substance use/abuse, chronic pain con-
ditions, and psychiatric conditions. As parsimony is less of an
issue when dealing with larger datasets, in practice our
strategy ended up including all available covariates in the
multivariable models. Adjusted odds ratios (OR) and
Bonferroni-adjusted 95% confidence intervals (CI) were re-
ported taking into account the number of hypotheses tested
for in the main analyses (42 hypotheses; 14 outcomes, and
Fig 1. Study flow chart. OSA, obstructive sleep apnoea; THA, three multimodal comparisons). Based on our sample size of
total hip arthroplasty; TKA, total knee arthroplasty. 181 182 patients, our model achieves 100% power to detect a
5% decrease in opioid utilisation using a two-sided test with
an alpha of 0.05 (under the conservative assumption that a
prescription during the entire hospital stay, and by hospital- 5% reduction in opioid utilisation would be the minimal
isation day (day of surgery, postoperative Day 1, and inpatient reduction needed to be translated into differences in out-
days thereafter). Secondary outcomes were comprised of three comes). It must be noted that while this step may reduce the
categories. Adverse effects commonly attributed to opioids as risk of type I errors, the likelihood of type II errors may be
previously established by Kessler and colleagues14 included increased.20 For all models, the PROC GLIMMIX procedure in
respiratory, gastrointestinal, genitourinary, CNS and ‘other SAS version 9.4 statistical software (SAS Institute, Cary, NC,
complications’ (composite of ICD-9 codes for postoperative USA) was used; for opioid prescription, LOS, and cost of
bradycardia, rash or itching, drugs causing adverse effects hospitalisation the gamma distribution with a log link func-
with therapeutic use, and fall from bed). Postoperative out- tion was applied as these variables are skewed.21,22 Effect
comes characteristic for the OSA population included CPAP estimates for these continuous outcomes are reported as %
use, mechanical ventilation, and the utilisation of critical care change and CI. Additionally, we used the CONTRAST state-
services. Lastly, outcomes reflecting the typically increased ment in the SAS PROC GLIMMIX procedure to test whether a
utilisation of resources in OSA, included cost and length of linear trend existed between effect estimates with increasing
stay (LOS).15 numbers of modes used in the multimodal analgesic
Inpatient opioid prescription was expressed in oral approaches.23
morphine equivalents, calculated by converting inpatient
opioid charges using the Lexicomp® ‘opioid agonist conver- Post hoc analyses
sion’16 and the GlobalRPH ‘opioid analgesic converter’17 Cost
of hospitalisation was adjusted for inflation and expressed in Several post hoc analyses were performed: first, to test the
2016 US dollars. robustness of our results and the selected cohort, patients
with opioid prescription above the 95th percentile were added
to a sensitivity analysis (Appendix Tables 1 and 2). Moreover,
we assessed the individual impact of each non-opioid anal-
Study size
gesic mode (peripheral nerve blocks, paracetamol/acetamin-
Exclusion criteria included non-elective procedures, unknown ophen, steroids, gabapentinoids, ketamine, NSAIDs, or COX-2
gender, unknown discharge status, outpatient procedures, inhibitors) on opioid prescription dose.
 134 - Cozowicz et al.

Table 1 Study variables by multimodal categorisation. *Continuous variable median and inter-quartile range (IQR) reported, instead of
n and %, respectively. COX, cyclooxygenase, PAP, positive airway pressure

Opioids only 1 mode 2 modes 2þ modes P-value

n¼20,883 n¼45,382 n¼50,743 n¼64,174

n % n % n % n %

Patient characteristics
Median age (yr)* 65 59e72 65 58e71 64 58e71 64 58e70 <0.0001
Gender <0.0001
Female 9610 46.0 21 344 47.0 24 080 47.5 29 624 46.2
Male 11 273 54.0 24 038 53.0 26 663 52.5 34 550 53.8
Race/ethnicity <0.0001
White 15 690 75.1 36 029 79.4 41 123 81.0 53 172 82.9
Black 1613 7.7 3693 8.1 3944 7.8 5171 8.1
Hispanic 167 0.8 298 0.7 208 0.4 147 0.2
Other 3413 16.3 5362 11.8 5468 10.8 5684 8.9
Healthcare-related
Insurance Type <0.0001
Commercial 7894 37.8 17927 39.5 20 605 40.6 26 784 41.7
Medicaid 528 2.5 1264 2.8 1438 2.8 1796 2.8
Medicare 11 608 55.6 24 285 53.5 26 581 52.4 33 052 51.5
Uninsured 50 0.2 160 0.4 151 0.3 149 0.2
Unknown 803 3.8 1746 3.8 1968 3.9 2393 3.7
Hospital location <0.0001
Rural 2129 10.2 4456 9.8 4910 9.7 5411 8.4
Urban 18 754 89.8 40 926 90.2 45 833 90.3 58 763 91.6
Hospital bed size <0.0001
<300 7654 36.7 16 726 36.9 19 899 39.2 24 801 38.6
300e499 7875 37.7 17 029 37.5 16 930 33.4 22 951 35.8
500 5354 25.6 11 627 25.6 13 914 27.4 16 422 25.6
Hospital teaching status <0.0001
Non-teaching 12 848 61.5 26 097 57.5 27 904 55.0 34 433 53.7
Teaching 8035 38.5 19 285 42.5 22 839 45.0 29 741 46.3
Median no. of annual hip 521 277e778 547 319e927 623 355e1005 709 401e1068 <0.0001
and knee arthroplasties
per hospital*
Procedure-related
Year of procedure <0.0001
2006 1636 7.8 2224 4.9 1153 2.3 296 0.5
2007 1676 8.0 2837 6.3 1728 3.4 674 1.1
2008 1806 8.6 3154 7.0 2214 4.4 1201 1.9
2009 2189 10.5 3718 8.2 3011 5.9 2039 3.2
2010 2380 11.4 4368 9.6 3858 7.6 3035 4.7
2011 2404 11.5 4842 10.7 4667 9.2 4193 6.5
2012 2407 11.5 5234 11.5 5737 11.3 6010 9.4
2013 1975 9.5 5063 11.2 6570 12.9 8040 12.5
2014 1752 8.4 4794 10.6 7022 13.8 10 243 16.0
2015 1546 7.4 4986 11.0 7576 14.9 13 811 21.5
2016 1112 5.3 4162 9.2 7207 14.2 14 632 22.8
General anaesthesia 14 872 71.2 31 916 70.3 34 221 67.4 41 828 65.2 <0.0001
Neuraxial anaesthesia 4265 20.4 9443 20.8 10 427 20.5 14 074 21.9 <0.0001
Patient-controlled analgesia 5563 26.6 8717 19.2 6936 13.7 4963 7.7 <0.0001
NSAIDs e e 18 195 40.1 29 312 57.8 47 170 73.5 <0.0001
COX-2 inhibitors e e 7541 16.6 19 937 39.3 49 513 77.2 <0.0001
Ketamine e e 728 1.6 2020 4.0 7601 11.8 <0.0001
Pregabalin/gabapentin e e 3585 7.9 12 393 24.4 40 768 63.5 <0.0001
Paracetamol/acetaminophen e e 11 649 25.7 29 636 58.4 56 150 87.5 <0.0001
Steroids e e 583 1.3 1206 2.4 2570 4.0 <0.0001
Peripheral nerve block e e 3101 6.8 6982 13.8 17 399 27.1 <0.0001
Comorbidity-related
Charlson-Deyo comorbidity index (categorised) <0.0001
0 12 557 60.1 27 812 61.3 31 039 61.2 40 072 62.4
1 5501 26.3 11 650 25.7 12 993 25.6 16 071 25.0
2 1592 7.6 3337 7.4 3928 7.7 4814 7.5
2þ 1233 5.9 2583 5.7 2783 5.5 3217 5.0
History of substance use/abuse 2080 10.0 4508 10.0 4728 9.0 6150 10.0 <0.0001
Pain conditions 3523 17.0 8366 18.0 10 675 21.0 15 124 24.0 <0.0001
Psychiatric comorbidities 5141 25.0 11 744 26.0 14 112 28.0 18 903 29.0 <0.0001

Continued
 RRH: Multimodal analgesia in obstructive sleep apnoea - 135

Table 1 Continued

Opioids only 1 mode 2 modes 2þ modes P-value

n¼20,883 n¼45,382 n¼50,743 n¼64,174

n % n % n % n %

Opioid-related adverse effects

Respiratory 956 4.6 1789 3.9 1749 3.4 1774 2.8 <0.0001
Gastrointestinal 603 2.9 957 2.1 972 1.9 1174 1.8 <0.0001
Genitourinary 551 2.6 1177 2.6 1241 2.4 1476 2.3 0.00456
CNS 239 1.1 459 1.0 471 0.9 539 0.8 0.0003
‘Other’ 446 2.1 824 1.8 825 1.6 755 1.2 <0.0001
OSA-related adverse effects
PAP ventilation 3072 14.7 6953 15.3 7466 14.7 9601 15.0 0.04411
Mechanical ventilation 310 1.5 555 1.2 377 0.7 259 0.4 <0.0001
Critical care admission 516 2.5 1038 2.3 1052 2.1 1142 1.8 <0.0001
Resource utilisation Median IQR Median IQR Median IQR Median IQR
Total opioid prescription* 350 215e523 323 202e482 299 185e450 270 165e420 <0.0001
Day 0 opioid prescription* 154 60e260 150 80e242 143 75e228 130 75e210 <0.0001
Day 1 opioid prescription* 60 20e120 60 30e110 60 30e105 60 30e100 <0.0001
Day 1þ opioid prescription* 85 38e158 71 30e135 60 23e120 50 15e105 <0.0001
Cost of hospitalisation* $16 555 $13 499e21 $16 249 $13 273e20 $16 023 $13 195e19 $15 783 $13 186e19 <0.0001
031 377 749 202
Length of stay* 3 3e4 3 3e3 3 2e3 3 2e3 <0.0001

Results analgesia (26.6% in opioids only vs 19.2%, 13.7%, and 7.7% with
increasing modes of multimodal analgesia, P<0.0001). The
Univariable analyses
highest opioid dosage, LOS, cost, and rate of complications
Overall, 181 182 OSA patients undergoing joint arthroplasty was observed in the ‘opioids only’ group.
(n¼48 475 THA; n¼132 707 TKA) were included. Multimodal Figure 2 shows trends in the utilisation of multimodal
analgesia was used in 88.5% (n¼160 299). Table 1 presents analgesia by number of analgesic modes where particularly
study variables by multimodal categorisation. Patients the group of patients receiving more than two non-opioid
receiving multimodal analgesia were more likely to be white, analgesic modes appear to be increasing. Figure 3 shows an
on commercial insurance, in teaching hospitals, or in hospi- overall decreasing trend in median annual per-procedure
tals with higher arthroplasty volume. opioid dose prescription in each multimodal analgesia
The most commonly utilised non-opioid analgesics were category.
paracetamol/acetaminophen (53.8%, n¼97 435/N¼181 182),
NSAIDs (52.3%, n¼94 677/N¼181 182), COX-2 inhibitors (42.5%,
n¼76 991/N¼181 182), and gabapentin/pregabalin (31.3%, n¼56
Multivariable analyses
746/N¼181 182), followed by peripheral nerve blocks (15.2%, When analysing separate effects of non-opioid analgesic
n¼27 482/N¼181 182), ketamine (5.7% n¼10 349/N¼181 182), modes for multimodal analgesia, the strongest reduction in
and steroids (2.4%, n¼4359/N¼181 182). opioid prescription was associated with the use of COX-2 in-
One of the most pronounced differences between multi- hibitors [12.6% (CI 14.0%; 11.1%)] and NSAIDs [11.0%
modal categories was the lower use of PCA in multimodal (CI 12.2%; 9.85%)] (Table 2).

Table 2 Results from multilevel regression models providing separate effect estimates for individual multimodal analgesia modalities.
Models were adjusted for: age, gender, race, insurance type, hospital location, hospital size, hospital teaching status, number of
annual hip and knee arthroplasties per hospital, year of procedure, general anaesthesia, neuraxial anaesthesia, patient-controlled
analgesia, NSAIDs, COX-2 inhibitors, ketamine, pregabalin/gabapentin, paracetamol/acetaminophen, steroids, peripheral nerve
block, Charlson-Deyo comorbidity index, history of substance use/abuse, pain conditions, and psychiatric comorbidities. COX,
cyclooxygenase

Total opioid Day 0 opioid Day 1 opioid Day 1þ opioid

prescription (%) prescription (%) prescription (%) prescription (%)

Peripheral nerve block 0.3 (1.4; 1.9) 8.1 (9.8; 6.2)* 8.9 (6.3; 11.4)* 3.4 (0.7; 6.2)*
Steroids 1.3 (1.5; 4.2) 2.1 (1.3; 5.6) 3.2 (7.1; 0.8) 3.6 (1.1; 8.5)
Pregabalin/gabapentin 2.0 (0.9; 3.1)* 1.4 (0.1; 2.7)* 0.4 (1.1; 2.0) 3.4 (1.5; 5.2)*
Ketamine 3.8 (1.8; 5.9)* 0.6 (1.7; 2.9) 3.2 (0.3; 6.1)* 5.6 (2.2; 9.1)*
NSAIDs 4.3 (5.3; 3.4)* 1.5 (2.6; 0.3)* 11.0 (12.2; 9.8)* 4.1 (5.7; 2.6)*
COX-2 inhibitors 6.1 (7.1; 5.2)* 2.6 (3.8; 1.4)* 8.0 (9.3; 6.6)* 12.6 (14.0; 11.1)*
Paracetamol/acetaminophen 2.3 (3.4; 1.3)* 0.2 (1.1; 1.5) 3.8 (5.3; 2.3)* 3.5 (5.1; 1.7)*
 136 - Cozowicz et al.

Discussion
In this retrospective cohort analysis of 181 182 OSA patients
undergoing major orthopaedic surgery, multimodal analgesia
was utilised in 88.5% of the patients, with increasing use over
time. With increasing numbers of non-opioid analgesic
modes, a stepwise decrease in opioid prescription dose,
resource utilisation, and odds for postoperative complications
was observed. Notably, the strongest beneficial stepwise effect
in the context of a dose-response gradient was observed in the
association between multimodal analgesia and the signifi-
cantly reduced need for postoperative mechanical ventilation
and critical care admission. Analysis of individual multimodal
Fig 2. Trends in the utilisation of multimodal analgesia by analgesia components showed that COX-2 inhibitors and
number of modes used. NSAIDs were associated with the strongest individual effect in
opioid dose reduction. Although effect estimates appeared
more modest for other analgesia modes, these may be
augmented when various analgesic modes are utilised
concurrently.
As a perioperative risk factor, OSA has particularly raised
the level of concern for acute respiratory complications,24,25
potentially causing respiratory failure, requirement for crit-
ical care interventions, and possibly even death in the post-
operative period.26e28 Given its phenotype of chronic
respiratory impairment with recurrent apnoea and hypo-
pnoea, OSA by nature predisposes affected individuals to risks
associated with anaesthetic medications and opioids in
particular. The effects of these drugs can cause various de-
grees of airway collapse1 and interact with consciousness,
sleep, and respiratory drive.2e4 Moreover, OSA has been linked
to an increased risk for the presence of a difficult airway.29,30
This is important, as the postoperative period is crucial given
Fig 3. Trends in opioid dose prescription by utilisation of
the exacerbation in sleep disordered breathing31 at a time of
multimodal analgesia. OME, oral morphine equivalents.
highest levels of pain and analgesic requirement. Chung and
colleagues31 demonstrated that cumulative 72-h opioid dose
appears to independently drive a higher postoperative
Adjusted effect estimates for opioid prescription dose, apnoea-hypopnoea index, while others have described
resource utilisation, and postoperative complications are re- possible alterations in pain sensitivity and augmented opioid
ported in Tables 3 and 4. With increasing modes of non-opioid potency in OSA.32e34 Collectively, these factors have raised
analgesia in addition to opioids, a stepwise decrease in opioid concerns among perioperative caregivers and spurred the
prescription dose of up to 14.9% (CI 17.0%; 12.7%) was release of recommendations on the use of risk-mitigating
observed on postoperative Day 1, while LOS decreased by up strategies including opioid dose reduction and safer anal-
to 11.8% (CI 13.0%; 10.7%) and minor decreases were gesic pathways, such as multimodal analgesia.5
observed for cost of hospitalisation, 3.2% (CI 4.2%; 2.2%). Consistent with previous publications,35,36 the current
With regards to associations between multimodal anal- analysis demonstrates that, with additional analgesic modes,
gesia use and the occurrence of postoperative complications, a gradual decrease in opioid prescription dose is observed.
the most pronounced estimates of effect were observed in However, the sheer reduction in utilised opioid dose does not
reduced odds for postoperative mechanical ventilation (OR per se establish a clinically meaningful benefit of multimodal
0.23, CI 0.16; 0.32) and postoperative critical care admission analgesia unless its utilisation also results in reduced adverse
(OR 0.60, CI 0.48; 0.75), which significantly decreased with the events in the absence of adverse effects attributable to sup-
addition of two or more non-opioid analgesic modes. Notably, plemental analgesic techniques.37 To that extent, there is a
this was observed in a stepwise manner, in the context of a lack of population-based studies investigating the impact of
dose-response relationship. Similarly, a stepwise decrease in multimodal pain management on postoperative complica-
odds for gastrointestinal complications of up to 35% (OR 0.65, tions and resource utilisation, and potential harm reduction
CI 0.53; 0.78) was found. with the use of increasing numbers of pain management
modalities.35,38
In this context, our study indeed shows beneficial re-
Sensitivity analysis
ductions in complications. The odds for gastrointestinal
Sensitivity analysis, applying the same multivariable models complications significantly decreased in a stepwise manner by
in a cohort including patients with opioid prescription above up to 35%, which mirrored a stepwise decrease in opioid pre-
the 95th percentile (n¼11 118) corroborated our initial findings. scription with increasing analgesic modalities used. However,
Notably, associations between multimodal analgesia and the most consistent stepwise beneficial effects were observed
opioid prescription dose were more pronounced. Results are for the need of postoperative mechanical ventilation and use
presented in Appendix Tables 1 and 2. of critical care services. These outcomes are of particular
 RRH: Multimodal analgesia in obstructive sleep apnoea - 137

Table 3 Multivariable regression results for opioid prescription dose and resource utilisation for each multimodal category (refer-
ence¼‘opioids only’ group). Odds ratios for opioid-related adverse effects and for continuous outcomes exponentiated coefficients
from the log model depicting % change comparing with reference (¼‘opioids only’). Models were adjusted for: age, gender, race/
ethnicity, insurance type, hospital location, hospital size, hospital teaching status, number of annual hip and knee arthroplasties per
hospital, year of procedure, general anaesthesia, neuraxial anaesthesia, patient-controlled analgesia, NSAIDs, COX-2 inhibitors, ke-
tamine, pregabalin/gabapentin, paracetamol/acetaminophen, steroids, peripheral nerve block, Charlson-Deyo comorbidity index,
history of substance use/abuse, pain conditions, and psychiatric comorbidities. CI, confidence interval; COX, cyclooxygenase; Day 0,
day of surgery; LOS, length of stay, POD, postoperative day

Opioid analgesia þ 1 additional þ 2 additional þ >2 additional Linear trend

analgesic mode analgesic modes analgesic modes

Opioid dose % Change (95% CI) P-value % Change (95% CI) P-value % Change (95% CI) P-value P-value
Day 0 1.5 (3.3; 0.4) 0.4704 2.3 (4.2; 0.4) 0.0042 3.5 (5.4; 1.5) <0.0001 0.0683
POD 1 5.0 (7.2; 2.8) <0.0001 10.4 (12.6; 8.3) <0.0001 14.9 (17.0; 12.7) <0.0001 <0.0001
POD 1þ 5.7 (8.1; 3.2) <0.0001 9.0 (11.4; 6.5) <0.0001 12.5 (14.9; 10.0) <0.0001 <0.0001
Total opioid use 1.7 (3.3; 0.1) 0.0252 4.3 (5.9; 2.6) <0.0001 7.2 (8.9; 5.5) <0.0001 <0.0001
LOS 4.6 (5.8; 3.5) <0.0001 7.8 (8.9; 6.6) <0.0001 11.8 (13.0; 10.7) <0.0001 <0.0001
Cost 1.4 (2.4; 0.5) <0.0001 2.5 (3.5; 1.6) <0.0001 3.2 (4.2; 2.2) <0.0001 <0.0001

importance in OSA, because of the higher anticipated risk for most resource utilisation (and thus costs) may be incurred on
life-threatening respiratory failure24 and resource the day of the surgery. Thus, a longer LOS may not be pro-
utilisation.15,27 portionally linked to total cost of hospitalisation.
The lack of increase in genitourinary complications An increasing body of evidence has suggested a potential
throughout all multimodal categories should also be noted, in impact of regional anaesthesia related procedures on post-
light of the fact that complementary use of drugs such as operative outcomes.42 This concept has increasingly been
NSAIDs, COX-2 inhibitors, and paracetamol/acetaminophen, applied to patients at higher perioperative risk such as those
have been associated with their own risks, including renal with OSA. Consistent with our findings for multimodal anal-
toxicity.29,30 gesia, a recent population-based analysis in OSA patients
Notably, multimodal analgesia was also associated with a similarly showed that the utilisation of regional vs general
stepwise, continuous decrease in LOS by up to 11.8%, while a anaesthesia was associated with reduced odds for mechanical
negligible effect was observed for cost of hospitalisation. The ventilation, critical care requirement, prolonged LOS, and
decrease in LOS, cost, and gastrointestinal complications with increased cost.43
the use of multimodal analgesia could be related, given a
recent population-based analysis in OSA patients demon-
strating reduced odds for these complications at lower vs
Limitations
higher perioperative opioid dose concentrations.39 Opioid A number of limitations, primarily attributable to the retro-
dose-related side-effects, such as postoperative sedation, spective observational nature of this analysis, should be
nausea and vomiting, urinary retention, and ileus may mentioned. Causality cannot be established, thus conclusions
contribute to delayed hospital discharge at increased cost.40,41 require cautious interpretation. Despite adjustments for
However, the modest effect observed for cost may be reflective numerous baseline characteristics (including substance abuse/
of the fact that in major surgeries with relatively short LOS, use, chronic pain, psychiatric comorbidities, and preoperative

Table 4 Multivariable regression results for postoperative complications for each multimodal category (reference¼‘opioids only’
group). Models were adjusted for: age, gender, race/ethnicity, insurance type, hospital location, hospital size, hospital teaching status,
number of annual hip and knee arthroplasties per hospital, year of procedure, general anaesthesia, neuraxial anaesthesia, patient-
controlled analgesia, NSAIDs, cyclooxygenase-2 inhibitors, ketamine, pregabalin/gabapentin, paracetamol/acetaminophen, ste-
roids, peripheral nerve block, Charlson-Deyo comorbidity index, history of substance use/abuse, pain conditions, and psychiatric
comorbidities. PAP, positive airway pressure

Opioid analgesia þ 1 additional analgesic þ 2 additional analgesic þ >2 additional analgesic Linear trend
mode modes modes

Or (95% CI) P-value OR P-value OR P-value P-value

Respiratory 1.00 (0.86e1.15) >0.999 0.96 (0.82e1.11) >0.999 0.86 (0.74e1.01) 0.1092 0.4930
Gastrointestinal 0.75 (0.62e0.89) <0.0001 0.69 (0.57e0.83) <0.0001 0.65 (0.53e0.78) <0.0001 <0.0001
CNS 0.96 (0.74e1.26) >0.999 0.96 (0.73e1.27) >0.999 0.94 (0.71e1.25) >0.999 0.7142
Genitourinary 0.96 (0.80e1.15) >0.999 0.91 (0.75e1.09) >0.999 0.83 (0.68e1.00) 0.0546 0.8584
Other 0.95 (0.78e1.16) >0.999 0.95 (0.77e1.17) >0.999 0.79 (0.63e1.00) 0.0378 0.8566
PAP use 1.03 (0.94e1.12) >0.999 1.01 (0.92e1.11) >0.999 0.99 (0.90e1.10) >0.999 0.2901
Mechanical ventilation 0.60 (0.46e0.79) <0.0001 0.33 (0.25e0.45) <0.0001 0.23 (0.16e0.32) <0.0001 <0.0001
Critical care admission 0.81 (0.66e1.00) 0.0420 0.73 (0.59e0.90) <0.0001 0.60 (0.48e0.75) <0.0001 0.0292
 138 - Cozowicz et al.

opioid use disorder to minimise the potential effect of preop- Acknowledgments

erative opioid use), the lack of detailed clinical information (e.g.
Data were purchased from Premier and use was restricted for
postoperative pain, local anaesthetic infiltration use, and OSA
this project and cannot be shared because of this restriction on
severity) in our dataset bears the risk for residual confounding,
use of data. Syntax is available from the corresponding author.
including indication or selection bias. However, we believe this
risk has been minimised by comprehensive adjustments for
baseline variables, by sensitivity analysis and given the dose- Funding
response gradient observed in outcomes. This analysis relates
to opioid prescription, rather than actual opioid consumption, This work was solely supported by institutional funding. No
because of the lack of such information in most large health- grants or other external support was received for this work.
care databases. Therefore, the observed effect of prescription
dose reduction by multimodal analgesia could be under- Declarations of interest
estimated, as patients may not have consumed all opioid pre-
scriptions. Given the overflow of information on numerous S.G.M. is a director on the boards of the American Society of
combinations of multimodal analgesia, dose information could Regional Anesthesia and Pain Medicine (ASRA) and the Society
not be included. Our variable, however, reflects the intention to of Anesthesia and Sleep Medicine (SASM). He is a one-time
multimodal analgesia, thus is reflective of our main study in- consultant for Sandoz Inc. and the holder of US Patent US-
terest. Although adjustment for enhanced recovery pathways 2017-0361063, Multicatheter Infusion System. He is the
was not possible, multimodal analgesia is a key component of owner of SGM Consulting, LLC and co-owner of FC Monmouth,
these pathways, therefore our main exposure variable likely LLC. None of the above relations influenced the conduct of the
adjusted for this indirectly. ICD-9 coding of OSA bears its own present study.
limitations, given the variability in OSA assessment methods
and disease severity. There has been one validation study,
Appendix A. Supplementary data
however, using Canadian data, which may not be reflective of
the US, especially as coding practices may differ with varying Supplementary data to this article can be found online at
financial incentives.44,45 The overall underestimation of true https://doi.org/10.1016/j.bja.2018.08.027.
OSA burden because of the large proportion of undiagnosed
OSA patients will apply to our dataset as well. Our findings may
References
therefore only apply to those with diagnosed OSA (presumably
those with the most complaints) and further research is needed 1. Ehsan Z, Mahmoud M, Shott SR, Amin RS, Ishman SL. The
on how multimodal strategies affect perioperative risk in the effects of anesthesia and opioids on the upper airway: a
full scope of OSA patients. Finally, despite rigorous efforts by systematic review. Laryngoscope 2016; 126: 270e84
Premier to provide data review and validation, potential coding 2. Taylor JM, Gropper MA. Critical care challenges in ortho-
errors and inconsistencies cannot be fully excluded. Never- pedic surgery patients. Crit Care Med 2006; 34: S191e9
theless, the stepwise nature of changes observed in outcomes, 3. Hillman DR, Loadsman JA, Platt PR, Eastwood PR.
is in line with a dose-response relationship, a factor raising the Obstructive sleep apnoea and anaesthesia. Sleep Med Rev
quality of evidence.46 2004; 8: 459e71
Given the potential for significant improvement of periop- 4. Ankichetty S, Wong J, Chung F. A systematic review of the
erative safety in OSA, future research is needed to further spur effects of sedatives and anesthetics in patients with
the simultaneous utilisation of multiple non-opioid analgesia obstructive sleep apnea. J Anaesthesiol Clin Pharmacol 2011;
pathways. An increased use of multimodal analgesia seems 27: 447e58
particularly desirable in bariatric surgery, given the high OSA 5. Gross JB, Apfelbaum JL, Caplan RA, et al. Practice guide-
prevalence.47 Moreover, besides opioid dose reduction, other lines for the perioperative management of patients with
analgesia related mechanisms, such as nerve blockade, may obstructive sleep apnea: an updated report by the Amer-
also significantly contribute to improved outcome.48 ican Society of Anesthesiologists Task Force on Perioper-
In conclusion, the utilisation of an increasing number of ative management of patients with obstructive sleep
non-opioid analgesic modes was associated with a significant, apnea. Anesthesiology 2014; 120: 268e86
stepwise decrease in opioid prescription dose, resource uti- 6. Memtsoudis SG, Poeran J, Zubizarreta N, et al. Association
lisation, and postoperative complications in a group consid- of multimodal pain management strategies with periop-
ered at high risk for perioperative complications because of erative outcomes and resource utilization: a population-
OSA. The most pronounced effects were observed in the based study. Anesthesiology 2018; 128: 891e902
reduction of the need for mechanical ventilation and critical 7. Makadia R, Ryan PB. Transforming the premier perspec-
care admission. These findings were observed in the context of tive hospital database into the observational medical
a dose-response relationship, a factor established to represent outcomes partnership (OMOP) common data model.
increased quality of evidence. Given the significant concern Egems 2014; 2: 1110
for critical respiratory failure in OSA, multimodal analgesia 8. Lyons PG, Zadravecz FJ, Edelson DP, Mokhlesi B,
may present an effective strategy for perioperative risk Churpek MM. Obstructive sleep apnea and adverse out-
reduction in this patient population. comes in surgical and nonsurgical patients on the wards.
J Hosp Med 2015; 10: 592e8
9. Quan H, Sundararajan V, Halfon P, et al. Coding algo-
Authors’ contributions rithms for defining comorbidities in ICD-9-CM and ICD-10
Study design: C.C., J.P., N.Z., M.M., S.G.M. administrative data. Med Care 2005; 43: 1130e9
Study conduct: C.C., J.P., N.Z., M.M., S.G.M. 10. Vadivelu N, Mitra S, Kaye AD, Urman RD. Perioperative
Data analysis, writing paper, revising paper: all authors. analgesia and challenges in the drug-addicted and drug-
 RRH: Multimodal analgesia in obstructive sleep apnoea - 139

dependent patient. Best Pract Res Clin Anaesthesiol 2014; 28: 29. Kheterpal S, Healy D, Aziz MF, et al. Incidence, predictors,
91e101 and outcome of difficult mask ventilation combined with
11. Bedard NA, Pugely AJ, Westermann RW, Duchman KR, difficult laryngoscopy: a report from the multicenter
Glass NA, Callaghan JJ. Opioid use after total knee perioperative outcomes group. Anesthesiology 2013; 119:
arthroplasty: trends and risk factors for prolonged use. 1360e9
J Arthroplasty 2017; 32: 2390e4 30. Cattano D, Killoran PV, Cai C, Katsiampoura AD,
12. Armaghani SJ, Lee DS, Bible JE, et al. Preoperative opioid Corso RM, Hagberg CA. Difficult mask ventilation in gen-
use and its association with perioperative opioid demand eral surgical population: observation of risk factors and
and postoperative opioid independence in patients un- predictors. F1000Res 2014; 3: 204
dergoing spine surgery. Spine Phila Pa 1976; 39: E1524e30 31. Chung F, Liao P, Elsaid H, Shapiro CM, Kang W. Factors
13. Khan TW, Manion S. Perioperative surgical home for the associated with postoperative exacerbation of sleep-
patient with chronic pain. Anesthesiol Clin 2018; 36: 281e94 disordered breathing. Anesthesiology 2014; 120: 299e311
14. Kessler ER, Shah M, Gruschkus SK, Raju A. Cost and quality 32. Doufas A. Obstructive sleep apnea, pain, and opioid analgesia
implications of opioid-based postsurgical pain control us- in the postoperative patient. Curr Anesthesiol Rep 2014; 4: 1e9
ing administrative claims data from a large health system: 33. Lam KK, Kunder S, Wong J, Doufas AG, Chung F.
opioid-related adverse events and their impact on clinical Obstructive sleep apnea, pain, and opioids: is the riddle
and economic outcomes. Pharmacotherapy 2013; 33: 383e91 solved? Curr Opin Anaesthesiol 2016; 29: 134e40
15. Memtsoudis SG, Stundner O, Rasul R, et al. The impact of 34. Turan A, You J, Egan C, et al. Chronic intermittent hypoxia
sleep apnea on postoperative utilization of resources and is independently associated with reduced postoperative
adverse outcomes. Anesth Analg 2014; 118: 407e18 opioid consumption in bariatric patients suffering from
16. Limberg JK, Dube S, Kuijpers M, et al. Effect of hypoxia on sleep-disordered breathing. PLoS ONE 2015; 10, e0127809
heart rate variability and baroreflex sensitivity during 35. Marret E, Kurdi O, Zufferey P, Bonnet F. Effects of
hypoglycemia in type 1 diabetes mellitus. Clin Auton Res nonsteroidal antiinflammatory drugs on patient-
2015; 25: 243e50 controlled analgesia morphine side effects: meta-
17. McAuley D, Pharm.D. GlobalRPH Opioid Analgesic Con- analysis of randomized controlled trials. Anesthesiology
verter http://globalrph.com/narcoticonv.htm Accessed 19 2005; 102: 1249e60
September 2016. 36. Song K, Melroy MJ, Whipple OC. Optimizing multimodal
18. Moineddin R, Matheson FI, Glazier RH. A simulation study analgesia with intravenous acetaminophen and opioids in
of sample size for multilevel logistic regression models. postoperative bariatric patients. Pharmacotherapy 2014;
BMC Med Res Methodol 2007; 7: 34 34(Suppl 1): 14se21s
19. Witte JS, Greenland S, Kim LL, Arab L. Multilevel modeling in 37. Elia N, Lysakowski C, Tramer MR. Does multimodal
epidemiology with GLIMMIX. Epidemiology 2000; 11: 684e8 analgesia with acetaminophen, nonsteroidal antiin-
20. Perneger TV. What’s wrong with Bonferroni adjustments. flammatory drugs, or selective cyclooxygenase-2 in-
BMJ 1998; 316: 1236e8 hibitors and patient-controlled analgesia morphine offer
21. Moran JL, Solomon PJ. A review of statistical estimators for advantages over morphine alone? Meta-analyses of ran-
risk-adjusted length of stay: analysis of the Australian and domized trials. Anesthesiology 2005; 103: 1296e304
New Zealand Intensive Care Adult Patient Data-Base, 38. Kehlet H, Wilmore DW. Multimodal strategies to improve
2008e2009. BMC Med Res Methodol 2012; 12: 68 surgical outcome. Am J Surg 2002; 183: 630e41
22. Rascati KL, Smith MJ, Neilands T. Dealing with skewed 39. Morwald EE, Olson A, Cozowicz C, Poeran J, Mazumdar M,
data: an example using asthma-related costs of medicaid Memtsoudis SG. Association of opioid prescription and
clients. Clin Ther 2001; 23: 481e98 perioperative complications in obstructive sleep apnea
23. SAS PROC GLIMMIX procedure. Available from https:// patients undergoing total joint arthroplasties. Sleep Breath
support.sas.com/documentation/cdl/en/statug/63033/ 2018; 22: 115e21
HTML/default/viewer.htm#statug_glimmix_a0000001408. 40. White PF. The changing role of non-opioid analgesic
htm Accessed 3 May 2018. techniques in the management of postoperative pain.
24. Kaw R, Chung F, Pasupuleti V, Mehta J, Gay PC, Anesth Analg 2005; 101: S5e22
Hernandez AV. Meta-analysis of the association between 41. Oderda GM, Evans RS, Lloyd J, et al. Cost of opioid-related
obstructive sleep apnoea and postoperative outcome. Br J adverse drug events in surgical patients. J Pain Symptom
Anaesth 2012; 109: 897e906 Manage 2003; 25: 276e83
25. Memtsoudis S, Liu SS, Ma Y, et al. Perioperative pulmo- 42. Ambrosii T, Sandru S, Belii A. The prevalence of periop-
nary outcomes in patients with sleep apnea after erative complications in patients with and without
noncardiac surgery. Anesth Analg 2011; 112: 113e21 obstructive sleep apnoea: a prospective cohort study. Rom
26. Ramachandran SK, Haider N, Saran KA, et al. Life- J Anaesth Intensive Care 2016; 23: 103e10
threatening critical respiratory events: a retrospective 43. Memtsoudis SG, Stundner O, Rasul R, et al. Sleep apnea
study of postoperative patients found unresponsive dur- and total joint arthroplasty under various types of anes-
ing analgesic therapy. J Clin Anesth 2011; 23: 207e13 thesia: a population-based study of perioperative out-
27. Melamed R, Boland LL, Normington JP, et al. Postoperative comes. Reg Anesth Pain Med 2013; 38: 274e81
respiratory failure necessitating transfer to the intensive 44. McIsaac DI, Gershon A, Wijeysundera D, Bryson GL,
care unit in orthopedic surgery patients: risk factors, Badner N, van Walraven C. Identifying obstructive sleep
costs, and outcomes. Perioper Med (Lond) 2016; 5: 19 apnea in administrative data: a study of diagnostic accu-
28. Subramani Y, Nagappa M, Wong J, Patra J, Chung F. Death racy. Anesthesiology 2015; 123: 253e63
or near-death in patients with obstructive sleep apnoea: a 45. Poeran J, Cozowicz C, Chung F, Mokhlesi B,
compendium of case reports of critical complications. Br J Ramachandran SK, Memtsoudis SG. Suboptimal diag-
Anaesth 2017; 119: 885e99 nostic accuracy of obstructive sleep apnea in one
 140 - Cozowicz et al.

database does not invalidate previous observational programs for bariatric surgery. Results from the Franco-
studies. Anesthesiology 2016; 124: 1192e3 phone large-scale database. Surg Obes Relat Dis 2018; 14:
46. Guyatt GH, Oxman AD, Sultan S, et al. GRADE guidelines: 99e105
9. Rating up the quality of evidence. J Clin Epidemiol 2011; 48. Kehlet H. Modification of responses to surgery by neural
64: 1311e6 blockade. In: Neural blockadevols. 129e75. Philadelphia:
47. Deneuvy A, Slim K, Sodji M, Blanc P, Gallet D, Lippincott-Raven; 1998
Blanchet MC. Implementation of enhanced recovery

Handling editor: L. Colvin