Obstructive sleep apnoea relationship with postoperative delirium and pain: an observational study of a surgical cohort

Summary

Patients with obstructive sleep apnoea are at increased risk of adverse postoperative outcomes, such as cardiac and respiratory complications. It has been hypothesised that obstructive sleep apnoea also increases the risk for postoperative delirium and acute postoperative pain. We conducted a retrospective, observational study investigating the relationship of obstructive sleep apnoea to postoperative delirium and acute postoperative pain severity. Patients were classified as being at high risk for obstructive sleep apnoea if they had been diagnosed with this condition, or if they were positive for more than four factors using the ‘STOP-BANG’ screening tool. Adjusted logistic regression was used to investigate the association between obstructive sleep apnoea and postoperative delirium, and multivariable linear regression to study the relationship between obstructive sleep apnoea and postoperative pain severity. The incidence of postoperative delirium was 307 in 1,441 patients (21.3%; 95%CI 19.2% - 23.5%). In unadjusted analysis, high risk for obstructive sleep apnoea was associated with delirium, with an odds ratio (95%CI) of 1.77 (1.22 – 2.57; p=0.003). After adjustment for pre-specified variables, the association was not statistically significant with odds ratio 1.34 (0.80 – 2.23; p = 0.27). The mean (SD) maximum pain (resting or provoked), reported for the entire cohort was 63.8 (27.9) mm on a 0 – 100 mm visual analogue scale. High risk for obstructive sleep apnoea was not statistically significantly associated with postoperative pain severity (β-coefficient 2.82; 95%CI, −2.34 – 7.97; p = 0.28). These findings suggest that obstructive sleep apnoea is unlikely to be a strong risk factor for postoperative delirium or acute postoperative pain severity.

Introduction

An estimated 20% of the general population suffers from obstructive sleep apnoea [1, 2], many without a clinical diagnosis [3, 4]. This is cause for concern, given the increasing evidence linking obstructive sleep apnoea with adverse postoperative outcomes including neurological, respiratory, cardiac and infective complications [5–7]. Surgical patients with obstructive sleep apnoea therefore have increased overall length of hospital stay, as well as higher admission rates and increased length of stay in intensive care units (ICU) [8–10]. It has been found that sleep disruption in patients with significant burns increases inflammatory markers and reported pain [11–13]; by analogy, patients with obstructive sleep apnoea, who have chronic sleep disruption, may be at higher risk of postoperative delirium and increased postoperative pain. A causal link between obstructive sleep apnoea and delirium would be clinically important, as postoperative delirium is associated with extended ICU stay, persistent cognitive decline, accidental falls, decreased quality of life and death [14–16].

Pain has a negative impact on quality of life and functioning, and thus adequate postoperative analgesia is a crucial component of recovery [17, 18]. Mechanistic evidence in various populations implies that sleep deprivation, which is the hallmark of obstructive sleep apnoea, amplifies up-regulation of cytokines, which may provoke excessive sensitivity to pain [19, 20]. Clinical evidence suggests that interrupted and inadequate sleep promotes hyperalgesia [11, 13, 21]. As patients with obstructive sleep apnoea may have increased susceptibility to the respiratory depressant effects of opioid medications [22, 23], providing safe and sufficient analgesia can be particularly challenging. This complicates investigations of postoperative pain as it is difficult to distinguish whether patients with obstructive sleep apnoea have altered pain perception, or are not adequately treated for pain, or whether both of these apply.

The objectives of this study were to investigate the relationships between obstructive sleep apnoea and postoperative pain and delirium. We hypothesised that patients with obstructive sleep apnoea experience more severe postoperative pain and have a higher incidence of postoperative delirium.

Methods

The primary aim of this retrospective, observational study was to investigate whether obstructive sleep apnoea is independently associated with postoperative delirium and acute postoperative pain severity. The secondary aim was to assess if compliance with prescribed treatment for obstructive sleep apnoea, typically continuous positive airway pressure (CPAP), might decrease the risk of postoperative delirium and the severity of acute postoperative pain.

Patients ≥ 18 years of age who underwent general anaesthesia for a non-neurosurgical inpatient operation at Barnes Jewish Hospital in St. Louis, Missouri, and had a postoperative stay of ≥ 1 day, were included in the analyses. Data were obtained from two randomised controlled trials and one ongoing prospective registry: Systematic Assessment and Targeted Improvement of Services Following Yearlong Surgical Outcomes Surveys Study (SATISFY-SOS) [24, 25]; Electroencephalography Guidance of Anesthesia to Alleviate Geriatric Syndromes study (ENGAGES) [26]; and Prevention of Delirium and Complications Associated with Surgical Treatments study (PODCAST) [27]). These were approved by the Human Research Protection Office at Washington University, who also approved this retrospective study. Patients signed written informed consent for enrolment in the parent studies. We prespecified our analyses and published a detailed protocol [28].

Baseline characteristics and other prespecified independent variables were extracted from electronic charts. Patients were grouped into three categories based upon previous diagnosis of obstructive sleep apnoea or obstructive sleep apnoea risk as determined by the STOP-BANG (Snoring; Tiredness; Observed apnoea; high blood Pressure; Body Mass Index (BMI) > 35kg.m⁻²; Age > 50; Neck circumference > 43cm for men and 41cm for women; male Gender) screening tool [29, 30]. The probability of moderate to severe obstructive sleep apnoea increases in direct proportion to the STOP-BANG score, with patients scoring 0–2 being classified as low risk for obstructive sleep apnoea, and patients scoring 5–8 being classified as high risk for obstructive sleep apnoea [31–33]. For the primary aim, patients were grouped into one of three risk categories: low (STOP-BANG 0–2); intermediate (STOP-BANG 3–4); and high (STOP-BANG 5–8 or previous diagnosis of obstructive sleep apnoea).

Delirium was assessed primarily using the Confusion Assessment Method (CAM) [34]. For patients who were unable to speak, delirium was assessed using the 3-minute Diagnostic CAM (3D-CAM) [35] or the CAM-ICU [36]. The following three criteria had to be present to qualify for a diagnosis: acute onset or a fluctuating course; inattention; and disorganised thinking or altered level of consciousness. Additionally, evidence of delirium was assessed with structured chart review by a researcher who was not aware of the results of the CAM assessment [37]. The presence of delirium was defined as any positive CAM assessment, or positive chart review, up to postoperative day 3.

Postoperative pain was assessed using a 100 mm Visual Analogue Scale (VAS) [38] at rest, taking a deep breath or coughing, and on movement; the maximum pain score recorded on any postoperative assessment was used for analysis. Delirious patients might be unable to report pain, and thus we conducted a sensitivity analysis excluding patients with postoperative delirium. Additionally, since our primary analysis did not distinguish between rest and provoked pain, or consider the duration of severe pain, we conducted a second sensitivity analysis using provoked pain (average of pain scores for deep breath / coughing and movement) during postoperative days 1–3. A small group of expertly trained researchers performed the daily assessments.

Binomial logistic regression was used to test for an association of obstructive sleep apnoea with postoperative delirium. We assessed the overall discriminatory ability of the binomial logistic regression model using area under the receiver operator curve (ROC). Multivariable linear regression was used to test an association of obstructive sleep apnoea with acute postoperative pain severity. The pre-specified regression models adjusted for: obstructive sleep apnoea risk category; age; sex; body mass index (BMI); type of surgery (cardiac vs. non-cardiac); Charlson comorbidity index; procedural cardiac risk; ASA physical status; moderate / severe alcohol use (women > 10 drinks.week⁻¹, men > 15 drinks.week⁻¹; 1 drink unit contains ≈ 14 grams alcohol); intra-operative midazolam dose; volatile anaesthetic concentration in minimum alveolar concentration [MAC]-hours; intra-operative ketamine dose; peri-operative opioid dose (total opioids administered during surgery and in the recovery ward converted to morphine equivalents in mg). We also adjusted for history of any of the following comorbidities: chronic obstructive pulmonary disease or asthma; stroke; dementia or mild cognitive impairment; visual or hearing impairment; depression or anxiety; chronic pain; and diabetes mellitus. Additionally, we included the variable ‘tiredness’ in the logistic regression model, since this particular symptom may be independently associated with delirium. Collinearity impaired our ability to include all of the pre-specified independent variables in our analyses.

For our secondary aim, patients previously diagnosed with obstructive sleep apnoea were asked if they were compliant with prescribed treatment for at least 4 h.night^-1. A binomial logistic regression was used to compare postoperative delirium between five groups: 1. low risk for obstructive sleep apnoea (STOP-BANG < 3) 2. intermediate risk for obstructive sleep apnoea (STOP-BANG 3–4) 3. high risk for obstructive sleep apnoea (STOP-BANG 5–8) 4. confirmed obstructive sleep apnoea + reported compliance with treatment and 5. confirmed obstructive sleep apnoea + reported non-compliance with treatment; with a focus on groups 4 and 5. A linear regression model was used to compare acute postoperative pain severity between the same groups, again focusing on those previously diagnosed with obstructive sleep apnoea. Patients with a low risk for obstructive sleep apnoea were the reference group for both analyses.

Results

Patients were recruited between February 2013 - May 2018. A total of 1441 patients were assessed for postoperative delirium (ENGAGES, 1208; PODCAST, 86; SATISFY-SOS, 147). Of these, 1297 (90%) were assessed for acute postoperative pain on at least one day up to postoperative day 3. Patient refusal to mark the visual analogue pain scale was the most common reason for lack of pain assessment. The variable, ‘history of dementia or mild cognitive impairment’, could not be included in either primary analysis because of only eight cases. Table 1 shows the patient characteristics and primary outcomes grouped by obstructive sleep apnoea risk.

Table 1.

Physical characteristics of patients included in the study, according to risk for obstructive sleep apnoea. Values are mean (SD) or number (proportion).

	Total n = 1441	Low risk n = 268	Intermediate risk n = 580	High risk n = 593
Proportion of total		18.6%	40.3%	41.1%
Age; years	68.8 (9.5)	65.5 (13.5)	70.0 (9.0)	69.0 (7.0)
BMI; kg.m−2	29.7 (7.2)	25.9 (4.5)	28.3 (6.0)	32.8 (8.0)
Sex; male	774 (53.7%)	44 (16.4%)	309 (53.3%)	421 (71.0%)
Black / other minority	145 (10.1%)	25 (9.3%)	72 (12.4%)	48 (8.1%)
ASA physical status
1	21 (1.5%)	16 (6.0%)	4 (0.7%)	1 (0.2%)
2	251 (17.4%)	76 (28.4%)	110 (19.0%)	65 (11.0%)
3	698 (48.4%)	129 (48.1%)	273 (47.1%)	296 (49.9%)
4	471 (32.7%)	47 (17.5%)	193 (33.2%)	231 (38.9%)
History of smoking	826 (57.3%)	121 (45.1%)	329 (56.7%)	376 (63.4%)
Moderate / severe alcohol use	41 (2.8%)	7 (2.6%)	17 (2.9%)	17 (2.9%)
Cardiac surgery	520 (36.1%)	50 (18.7%)	206 (35.5%)	264 (44.5%)
Postoperative delirium	307 (21.3%)	44 (16.4%)	110 (18.9%)	153 (25.8%)
Maximum pain reported	63.8 (27.9)	63.3 (28.0)	61.1 (27.5)	66.8 (27.9)

There were 307 (21.3%; 95%CI 19.2 – 23.5%) cases of delirium. The incidence was 44/268 (16.4%) in the group with low risk for obstructive sleep apnoea, 110/580 (18.9%) in the intermediate risk group, and 153/593 (25.8%) in the high risk group. When compared with low risk for obstructive sleep apnoea, intermediate risk was associated with an unadjusted odds ratio (OR) of 1.19 (95%CI 0.81 – 1.75) and high risk of 1.77 (95%CI 1.22 – 2.57).

Binomial logistic regression was performed to assess the impact of obstructive sleep apnoea risk on the likelihood that participants developed postoperative delirium. Linearity of continuous variables with respect to the logit of Delirium was assessed via the Box-Tidwell procedure. A Bonferroni correction was applied using all 23 terms in the model, resulting in statistical significance being accepted if p < 0.002 [39]. Based on this assessment, all continuous independent variables were found to be linearly related to the logit of delirium incidence. Outliers were assessed using deviance residuals (> 2), leverage (> 0.033), and Cook’s distance (> 1). Outliers were considered for exclusion if they had a deviance absolute value > 2, or a Cook’s value > 1 and leverage value > 0.033. No cases qualified for exclusion, and thus all cases were included in analysis. Cardiac surgery was significantly collinear with many variables, including ASA physical status and procedural cardiac risk. Removing cardiac surgery from the model resolved the collinearity issues, and thus cardiac surgery was excluded from analysis. The logistic regression model was statistically significant, χ²(23) = 120.403, p < 0.0001. Additionally, the model was a good fit as demonstrated by the Hosmer and Lemeshow test, χ²(8)=8.114, p = 0.423. The model explained 12.4% (Nagelkerke R²) of the variance in postoperative delirium. The area under the ROC curve was 0.70 (95%CI: 0.67 – 0.74), which is an acceptable level of discrimination [40]. Of the 23 independent variables, only four were statistically significantly associated with postoperative delirium: race (black and other minority races); functional capacity < 4 METs; age; and peri-operative opioid dose. Obstructive sleep apnoea risk was not significantly associated with postoperative delirium (Table 2).

Table 2.

Odds ratios (OR) for logistic regression analysis predicting postoperative delirium.

	Unadjusted OR (95%CI)	p value	Adjusted OR (95%CI)	p value
Age	1.028 (1.013 – 1.043)	0.0002	1.03 (1.007 – 1.045)	0.006
BMI; kg.m−2	1.002 (0.985 – 1.020)	0.83	0.984 (0.962 – 1.005)	0.13
Sex; male	0.922 (0.716 – 1.187)	0.54	0.794 (0.581 – 1.085)	0.15
Black / other minority	1.587 (1.082 – 2.329)	0.02	1.912 1.267 – 2.884)	0.002
Obstructive sleep apnoea risk
Intermediate vs. low	1.191 (0.811 – 1.749)	0.38	0.85 (0.543 – 1.329)	0.48
High vs. low	1.77 (1.220 – 2.567)	0.003	1.336 (0.798 – 2.234)	0.27
ASA physical status
3 vs. 1 / 2	1.279 (0.858 – 1.907)	0.23	0.873 (0.565 – 1.349)	0.56
4 vs. 1 / 2	3.056 (2.055 – 4.544)	< 0.0001	1.586 (0.985 – 2.552)	0.06
Functional capacity < 4 METs	2.238 (1.723 – 2.908)	< 0.0001	1.673 (1.23 – 2.279)	0.001
High procedural cardiac risk	1.939 (1.503 – 2.501)	< 0.0001	1.002 (0.700 – 1.433)	0.99
Charlson Comorbidity Index	1.001 (0.939 – 1.067)	0.98	1.014 (0.939 – 1.098)	0.74
Tiredness	1.399 (1.085 – 1.803)	0.009	1.13 (0.841 – 1.518)	0.42
Moderate / severe alcohol use	0.755 (0.331 – 1.720)	0.51	0.832 (0.357 – 1.94)	0.67
Diabetes mellitus	1.312 (0.996 – 1.728)	0.05	0.973 (0.703 – 1.347)	0.94
History of stroke	1.783 (1.276 – 2.490)	0.0007	1.297 (0.90 – 1.866)	0.15
Chronic obstructive pulmonary disease / asthma	1.457 (1.091 – 1.946)	0.01	1.26 (0.9 – 1.718)	0.15
Chronic pain	1.177 (0.898 – 1.543)	0.24	1.08 (0.788 – 1.48)	0.60
Depression / anxiety disorder	1.243 (0.912 – 1.694)	0.17	1.156 (0.824 – 1.621)	0.40
Hearing / visual impairment	0.962 (0.747 – 1.239)	0.78	1.006 (0.752 – 1.346)	0.97
Intra-operative ketamine	1.715 (0.983 – 2.995)	0.06	1.296 (0.72 – 2.335)	0.39
Intra-operative midazolam	0.762 (0.592 – 0.981)	0.04	1.124 (0.841 – 1.503)	0.43
Peri-operative opioid dose	1.008 (1.005 – 1.011)	< 0.0001	1.006 (1.003 – 1.009)	< 0.0001
MAC-Hours	1.001 (0.993 – 1.010)	0.83	1.003 (0.993 – 1.013)	0.61

The maximum pain VAS recorded for the entire cohort was mean (SD) 63.8 (27.9) mm. Patients with a high risk for obstructive sleep apnoea had a score of 66.8 (27.9) mm; patients with intermediate risk had a score of 61.1 (27.5) mm; and patients with low risk had a score of 63.3 (28.0) mm. There was a statistically significant difference in means between groups (ANOVA F(2,1292) = 5.72; p = 0.003). Post-hoc analysis with Bonferroni correction showed a statistical difference in pain scores between the intermediate risk group and the high risk group (95% CI 1.6 – 9.8; p = 0.003).

A multivariable regression was performed to assess the relationship between obstructive sleep apnoea risk and maximum postoperative pain, with adjustment for known confounders. Independence of observations was assumed given the study design, and multicollinearity was assessed with the correlation table (values > 0.7 excluded) and Variance Inflation Factor (VIF; values > 10 excluded). Similar to the logistic regression, cardiac surgery demonstrated significant collinearity with multiple variables and was subsequently removed from the regression. There was homoscedasticity, as assessed by visual inspection of a plot of studentised residuals versus unstandardised predicted values. Outliers were assessed with studentised deleted residuals. Two cases were > 3 SD (3.6; 3.3) and were excluded from analysis. One case had high leverage (leverage point = 0.5022) and was excluded. Influential points were assessed with Cook’s distance values, and there were no values above 1. The assumption of normality was met, as assessed by Q-Q plot. The multivariable regression model was statistically significantly associated with maximum postoperative pain, F(23,1274) = 3.83, p < 0.0001, R² = 6.5%, adj. R² = 4.8%. However, only three variables were statistically significantly associated with this outcome. Intra-operative midazolam use, peri-operative opioid dose, and volatile anaesthetic agent MAC-hours were all significantly associated with increasing maximum postoperative pain. Obstructive sleep apnoea risk was not significantly associated with maximum postoperative pain (Table 3).

Table 3.

Linear regression analysis predicting acute postoperative pain.

Variable	B (95%CI)	SEB	β	p value	VIF
Intercept	57.98 (39.57 – 76.39)	9.38		< 0.0001
Age	−0.154 (−0.342 – 0.033)	0.095	−0.053	0.106	1.441
BMI; kg.m−2	−0.85 (−0.321 – 0.151)	0.120	−0.022	0.481	1.30
Sex; male	−1.92 (−5.388 – 1.548)	1.77	−0.034	0.278	1.361
Black / other minority	−0.188 (−5.154 – 4.778)	2.53	−0.002	0.941	1.028
Obstructive sleep apnoea risk
Intermediate vs. low	−2.226 (−6.763 – 2.311)	2.31	−0.039	0.336	2.266
High vs. low	2.816 (−2.341 – 7.974)	2.629	0.049	0.284	2.899
ASA physical status
3 vs. 1 / 2	1.566 (−2.688 – 5.820)	2.168	0.028	0.470	2.060
4 vs. 1 / 2	−2.602 (−8.196 – 2.991)	2.851	−0.043	0.362	2.962
Functional capacity < 4 METs	2.99 (−0.395 – 6.377)	1.73	0.054	0.083	1.299
High procedural cardiac risk	−1.34 (−5.471 – 2.787)	2.105	−0.024	0.524	1.864
Charlson Comorbidity Index	−0.041 (−0.929 – 0.847)	0.453	−0.003	0.928	1.429
Moderate / severe alcohol use	2.416 ( −6.713 – 11.546)	4.654	0.014	0.604	1.026
Diabetes mellitus	−0.873 (−4.656 – 2.91)	1.928	−0.014	0.651	1.273
History of stroke	1.70 (−2.935 – 6.329)	2.36	0.021	0.472	1.108
Chronic obstructive pulmonary disease / asthma	1.76 (−2.107 – 5.633)	1.97	0.026	0.372	1.138
Chronic pain	1.433 (−2.11 – 4.98)	1.806	0.023	0.793	1.191
Depression / anxiety disorder	2.834 (−1.094 – 6.763)	2.003	0.040	0.157	1.068
Hearing / visual impairment	0.965 (−2.27 – 4.20)	1.649	0.017	0.558	1.19
Intra-operative ketamine	−1.387 (−9.001 – 6.227)	3.881	−0.010	0.721	1.036
Intra-operative midazolam	3.462 (0.191 – 6.733)	1.667	0.062	0.038	1.20
Peri-operative opioid dose	0.123 (0.081 – 0.166)	0.021	0.185	<0.0001	1.405
MAC-hours	0.158 (0.010 – 0.307)	0.076	0.060	0.036	1.10

SE_B= Standard error of B coefficient.β=standardised Beta. VIF=variance inflation factor

To reduce any limitations regarding our primary outcome, two pre-specified sensitivity analyses were performed (Table 4). Since our primary analysis did not examine severe pain duration or discriminate between rest and provoked pain, we performed a linear regression with provoked pain as the outcome. Although the overall model showed a significant association with provoked postoperative pain, the model had a small estimated size effect (adjusted R² of 7.5%). Our primary pain analysis did not consider delirium. This is important because delirious patients may be unable to accurately report their pain. Thus, we excluded patients who tested positive for delirium and then performed a binomial logistic regression to assess the relationship between obstructive sleep apnoea risk in non-delirious patients and maximum postoperative pain. Although β-coefficients for high risk for obstructive sleep apnoea improved in these sensitivity analyses, obstructive sleep apnoea risk still did not statistically significantly contribute to maximum or provoked postoperative pain.

Table 4.

Summary of sensitivity analyses for risk of obstructive sleep apnoea, including multiple regression models

	Intermediate risk			High risk			Overall model statistics
	B(SEB)	95%CI	p	B(SEB)	95% CI	p		p	R2(Adj. R2)
Provoked postoperative pain	−0.544 (2.001)	−4.470 – 3.382	0.786	3.528 (2.276)	−0.937 – 7.993	0.121	F(23,1271) = 5.578	< 0.0001	9.2% (7.5%)
Maximum postoperative pain excluding delirious patients	−2.591 (2.508)	−7.513 – 2.331	0.302	4.013 (2.904)	−1.684 – 9.711	0.167	F(23,1045) = 4.025	< 0.0001	8.1% (6.1%)
Primary analysis of maximum postoperative pain	−2.226 (2.31)	−6.763 – 2.311	0.336	2.816 (2.629)	−2.341 – 7.974	0.284	F(23,1274) = 3.832	< 0.0001	6.5% (4.8%)

SE_B= Standard error of B coefficient.

Regarding our secondary aim, of 352 patients previously diagnosed with obstructive sleep apnoea, 262 were prescribed treatment with positive airway pressure (usually CPAP). For those prescribed treatment, 179 reported compliance (68%), with 54 experiencing at least one episode of postoperative delirium. Eighty-three reported noncompliance, with 16 experiencing at least one episode of postoperative delirium.

A binomial logistic regression was performed to investigate if compliance with treatment for obstructive sleep apnoea decreased the incidence of postoperative delirium. Patients reporting compliance with treatment did not have a decreased risk for postoperative delirium (OR 1.77; 95%CI, 0.97 – 3.22; p = 0.06) compared with patients reporting non-compliance (OR 0.90; 95%CI 0.43 – 1.88; p = 0.80). Patients with a high or intermediate risk for obstructive sleep apnoea did not have an increased risk for postoperative delirium, OR 1.30 (95%CI 0.76 – 2.25; p = 0.35) and OR 0.86 (95%CI 0.55 – 1.34; p = 0.51) respectively.

Multivariable linear regression was performed to determine if compliance with treatment for obstructive sleep apnoea decreased acute postoperative pain. Of 352 patients previously diagnosed with obstructive sleep apnoea, 224 were assessed for acute postoperative pain on at least one day up to postoperative day 3. Although patients compliant with obstructive sleep apnoea treatment reported a lower postoperative pain score of 65.7 (27.7) mm compared with 70.1 (26.2) mm for non-compliant patients (t (222) = 1.12; p = 0.26; 95%CI −3.3 – 12.0), neither grouping was significantly associated with acute postoperative pain (compliant with treatment B (SE_B) 4.82 (3.78; 95%CI −2.60 – 12.24); non-compliant with treatment 5.09 (4.27; 95%CI −3.29 – 13.48). Furthermore, classification as high risk of obstructive sleep apnoea (2.81 (3.05; 95%CI −3.17 – 8.80) or intermediate risk of obstructive sleep apnoea (−1.25 (2.52; 95%CI −6.20 – 3.70) was not significantly associated with acute postoperative pain.

Discussion

Contrary to our hypothesis [28], obstructive sleep apnoea risk was not independently associated with the incidence of postoperative delirium or the severity of acute postoperative pain, once adjusted for the pre-specified variables. Race (black and other minority races), functional capacity < 4 METs, age, and peri-operative opioid dose (expressed in morphine equivalents) had the strongest associations with postoperative delirium. Intra-operative midazolam dose, peri-operative opioid dose (expressed as morphine equivalents), and MAC-hours were associated with increasing maximum postoperative pain reported on postoperative days 1–3.

The findings of our study contrast with previous reports that have found an independent association between obstructive sleep apnoea and postoperative delirium [41–43]. These studies were much smaller than ours, and there were only 44, 27 and 21 patients with delirium in each study [41–43]. Their regression models would thus have been constrained regarding the inclusion of potential confounders. In contrast, we included 307 patients with delirium. In common with these studies, we found a strong association between obstructive sleep apnoea and delirium. However, when we adjusted for pre-specified potential confounders, the association was no longer significant. It is also important to note that hypothesis-generating findings are often not reproduced in larger, more rigorous studies [44, 45].

Our study has important implications for patients with obstructive sleep apnoea in the peri-operative period. It supports the finding that delirium is a common postoperative complication [46], including in patients with obstructive sleep apnoea. However, although it is important to diagnose and treat obstructive sleep apnoea in the peri-operative period in order to decrease some complications, our findings suggest that it is unlikely to reduce the incidence of postoperative delirium. This inference is consistent with a finding that peri-operative treatment of obstructive sleep apnoea did not decrease the incidence or severity of postoperative delirium [43]. In relation to pain, we were unable to find evidence that obstructive sleep apnoea is associated with either increased or decreased postoperative pain severity. In view of the risk of airway obstruction and respiratory arrest [47], the treatment of pain in patients with obstructive sleep apnoea remains a major challenge in the peri-operative period.

This study had important limitations. Most notably, since this is a retrospective, observational study, we were limited by the data collected in the parent studies. Although a high-risk grouping using the STOP-BANG screening tool reliably predicts the diagnosis of obstructive sleep apnoea, this does not hold for the intermediate-risk group [31, 32]. Obstructive sleep apnoea can be considered to present a continuum of risk rather than to be a binary outcome [48, 49]. Our analysis treated postoperative delirium as a binary outcome, whereas it may be more important to focus on duration or severity. Regarding pain, the large, significant intercept in our model suggests that a critical predictor of postoperative pain is missing. We were unable to include detailed data on inpatient analgesic medication in our analyses, which is likely to have had an impact on our models.

This study had important strengths. The database included patients who underwent structured pre-operative screening for obstructive sleep apnoea. In additional, rigorous and reliable delirium and pain assessments were performed by a small group of expertly trained researchers. To promote scientific rigor and reproducibility, we prespecified our analyses and published a detailed protocol with peer review [28], which allowed us to improve our study design. We also attempted to mitigate certain limitations by conducting sensitivity analyses. In particular, we excluded patients with postoperative delirium when evaluating risk factors for pain, which other studies have generally not done.

With regard to our secondary aim, and contrary to our initial hypothesis, patients previously diagnosed with obstructive sleep apnoea who reported compliance with prescribed treatment did not experience less postoperative delirium. Moreover, obstructive sleep apnoea risk or compliance status were not significantly associated with postoperative delirium. There was no statistically significant change in acute postoperative pain between patients who were compliant with treatment for obstructive sleep apnoea and those who were non-compliant. Neither compliance status, nor risk of obstructive sleep apnoea, were significantly associated with acute postoperative pain. Even though these secondary analyses were adjusted for pre-specified known and candidate risk factors for postoperative delirium and pain, there are important limitations. Secondary analyses were conducted with data self-reported by patients, without a means to verify true compliance. We found a 68% compliance with obstructive sleep apnoea treatment, higher than a finding of < 50% in previous studies [50, 51]. Furthermore, our secondary analyses did not adjust for suggested factors that influence adherence to obstructive sleep apnoea treatment, such as apnoea-hypopnoea index, Epworth sleepiness score and smoking status [52]. Given the findings and limitations of these analyses, we cannot make a definitive statement regarding the impact of treatment compliance on the incidence of postoperative delirium and acute postoperative pain severity.

In conclusion, this study found that obstructive sleep apnoea risk was not independently associated with the incidence of postoperative delirium or the severity of acute postoperative pain.