Perioperative Sleep Disturbances and Postoperative Delirium in Adult Patients: A Systematic Review and Meta-Analysis of Clinical Trials

Background: The aim of this systematic review and meta-analysis of clinical trials was to investigate the effects of perioperative sleep disturbances on postoperative delirium (POD). Methods: Authors searched for studies (until May 12, 2020) reporting POD in patients with sleep disturbances following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. Results: We identified 29 relevant trials including 55,907 patients. We divided these trials into three groups according to study design: Seven retrospective observational trials, 12 prospective observational trials, and 10 randomized controlled trials. The results demonstrated that perioperative sleep disturbances were significantly associated with POD occurrence in observational groups [retrospective: OR = 0.56, 95% CI: [0.33, 0.93], I2 = 91%, p for effect = 0.03; prospective: OR = 0.27, 95% CI: [0.20, 0.36], I2 = 25%, p for effect < 0.001], but not in the randomized controlled trial group [OR = 0.58, 95% CI: [0.34, 1.01], I2 = 68%, p for effect = 0.05]. Publication bias was assessed using Egger's test. We used a one-by-one literature exclusion method to address high heterogeneity. Conclusions: Perioperative sleep disturbances were potential risk factors for POD in observational trials, but not in randomized controlled trials.


INTRODUCTION
Postoperative delirium (POD) is a state of brain dysfunction following surgery, and it features acute onset and fluctuating occurrence (1). The typical clinical manifestations of POD include alterations of consciousness, attention, and cognition. According to reports, POD affects 11-51% of patients after major surgery, and it is independently associated with prolonged intensive care, long-term postoperative cognitive dysfunction, and increased mortality (2)(3)(4)(5). However, the pathogenesis of POD remains unclear, so it is particularly important to identify risk factors to prevent its occurrence.
Perioperative sleep disturbances are common among surgery patients. The disturbances include obstructive sleep apnea (OSA), reduced total sleep time, sleep fragmentation, circadian rhythm disruption, and so on (6)(7)(8). Over 40% of patients complained about poor sleep quality during the first night following surgery, and the sleep problems continued several days postoperation (9). Some observational studies reported that patients with poor sleep quality were predisposed to mental disorders including delirium and cognitive dysfunction (10)(11)(12). In addition, several randomized controlled trials (RCTs) found that improving sleep quality, be it through medication or other interventions, strikingly decreased the incidence of delirium (13)(14)(15). Although multiple studies have supported the viewpoint that sleep problems are significantly associated with delirium, some studies obtained negative results (16)(17)(18). Thus, we designed this systematic review and meta-analysis to clarify the effect of sleep disturbance on the incidence of delirium in adult surgery patients.

METHODS
This systematic review and meta-analysis was performed according to the guidelines of the 2009 Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) (Supplementary Table 1) (19).

Search Strategy
Hongbai Wang and Liang Zhang were responsible for document retrieval. We searched the databases of Pubmed, Embase, Cochrane Library, and Web of Science using the PICOS (Population, Intervention, Comparison, Outcome, Study design) method. Our last search was completed on May 12, 2020. The search terms included "sleep" OR "insomnia" OR "sleep disturbance" OR "night" OR "circadian" AND "surgery" OR "operation" OR "postoperative" OR "anesthesia" OR "anesthesia" AND "delirium" OR "confusion" OR "agitation" OR "acute confusional state" OR "acute confusional syndrome, " and the search scope was "title and abstract." Because we sought to examine all studies about the effect of sleep disturbances on POD incidence in adult patients undergoing surgery, we did not constrain the search terms for study designs.

Study Selection
Zhe Zhang and Yinan Li performed the screening process for titles and abstracts, while Hongbai Wang and Su Yuan performed the screening process for full texts. The inclusion criteria were (1) participants aged 18 years or older; (2) patients undergoing surgery; and (3) articles reporting the effect of sleep on delirium. The exclusion criteria were: (1) duplicate articles; (2) participants younger than 18 years old; (3) review or meta-analysis; (4) articles published as an abstract, letter, case report, basic research, editorial, note, method, or protocol; (5) articles presented in a non-English language; (6) studies without statistical differences in sleep quality between intervention and control groups; (7) studies without a specific number of patients with sleep problems (observational studies) and/or delirium; and (8) studies including some patients not undergoing surgery.

Quality Assessment of Included Studies
Qipeng Luo and Su Yuan independently assessed the quality of included studies. For retrospective and prospective observational trials, risk of bias was assessed using the Newcastle-Ottawa Quality Assessment Scale (NOS), which comprises the following three domains: selection, comparability, and outcome for cohort studies (20). There were four stars in the selection domain, two stars in the comparability domain, and three stars in the exposure domain. Trials with seven or more cumulative stars were considered to be of high quality, those with six stars of moderate quality, and those with <6 stars of low quality (20). For RCTs, risk of bias was assessed using the Cochrane Collaboration Risk of Bias Assessment tool, which included the following seven items: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting, and others (bias due to vested financial interest and academic bias). If a trial was found to have one or more of the items associated with high or unclear risk of bias, this trial was classified as high risk (21). If the two authors disagreed on their assessment, they consulted the third or fourth author. Eventually, we reached a consensus.

Data Extraction
Yinan Li and Qipeng Luo were responsible for extracting the following information: (1) authors; (2) publication year; (3) total number of participants in each study; (4) age range of all the participants; (5) country of publication; (6) percentage of males; (7) procedures that the participants underwent; (8) methods of sleep disturbance assessment; (9) methods of POD assessment; (10) number of patients with and without POD; (11) number of patients with good and poor sleep quality; and (12) the followup time. Hongbai Wang and Liang Zhang were responsible for adjusting data discrepancies.

Outcome Measures
The sole aim of this meta-analysis was to determine whether perioperative sleep disturbances were associated with increased POD in adult surgery patients.

Data Synthesis
We divided all the included trials into three groups according to their study design to facilitate data synthesis. These were retrospective observational trials (ROTs), prospective observational trials (POTs), and RCTs.

Data Analysis
RevMan Review Manager version 5.3 (Cochrane collaboration, Oxford, UK) and Stata version 12.0 (Stata Corp, College Station, TX, USA) were used to perform statistical analyses. We assessed the heterogeneity of included studies using the values of I 2 and the Mantel-Haenszel chi-square test (p-value for heterogeneity). The values of I 2 < 40%, I 2 = 40-60%, and I 2 > 60% indicated low, moderate, and high heterogeneity, respectively (22). If we identified I 2 > 50% or a p-value for heterogeneity < 0.1, we used a random-effect model to analyze the data. Conversely, if we identified I 2 < 50% or a p-value for heterogeneity ≥ 0.1, we used a fixed-effect model to analyze the data (23). The dichotomous outcomes were reported as odds ratios (OR) with 95% confidence intervals (CI). Publication bias was assessed using Begg's test, and studies with a p < 0.05 were adjusted using trim-and-fill analysis (24). The statistical tests were two-sided, and overall effects with a p < 0.05 were considered to exhibit significant differences.
We conducted sensitivity analysis to address high heterogeneity (I 2 > 40%) through the methods of subgroup analysis or one-by-one article removal. We used meta-regression to identify the sources of high heterogeneity according to possible risk factors (25). Meta-regression analyses that produced a risk factor of p < 0.05 were followed by subgroup analysis, while those that produced a risk factor of p ≥ 0.05 were followed by one-by-one article removal (26). Figure 1 presents the PRISMA flow chart for our screening process. We obtained 257 trials from Pubmed, 437 from Embase, 292 from Cochrane Library, and 342 from Web of Science. We removed 389 duplicate trials and excluded 857 trials at the title-and-abstract review stage based on our exclusion criteria.

Study Selection
We excluded 53 trials at the full-text review stage, including 35 without statistical differences in sleep quality between intervention and control groups, 10 without a specific number of patients with sleep problems and/or delirium, and eight that enrolled some patients without surgery. Eventually, our search strategy yielded 29 relevant trials with a total of 55,907 patients ( Figure 1) (16)(17)(18).

Publication Bias
We assessed publication bias using Egger's test by Stata 12.0 software. We did not find publication bias in the ROT (p = 0.085), POT (p = 0.764), or RCT (p = 0.933) groups (Figure 2 and Supplementary Table 3).

Sensitivity Analysis
We performed meta-regression to identify the sources of heterogeneity in the ROT and RCT groups, assessing possible risk factors including publication year, average age (≥65 years and <65 years), male proportion (≥50% and <50%), surgery types (non-cardiac surgery, cardiac surgery, and cardiac and non-cardiac surgeries), onset time for POD (>3 days and ≤3 days), and study quality (low quality and high quality). Unexpectedly, all p-values for these risk factors were over 0.05 (Supplementary Tables 4, 5). Afterwards, we used the method of one-by-one literature removal and found that 4 trials were the main sources of heterogeneity in the ROT group (I 2 dropped  from 91 to 12%) and four trials were in the RCT group (I 2 dropped from 68% to 19%).

DISCUSSION
This meta-analysis investigated the effect of perioperative sleep disturbance on the incidence of POD. The results from observational trials (retrospective and prospective) demonstrated that perioperative sleep disturbances were significantly associated with elevated POD incidence, while those from RCTs did not confirm this positive association. POD, as a kind of mental disorder, is a knotty problem that patients may face following major surgery. It arises as the combined effect of multiple factors, which include advanced age, low education level, preoperative impaired cognition, alcohol abuse, smoking, cardiac or macrovascular surgery, major non-cardiac surgeries, perioperative administration of sedative and analgesic drugs, and postoperative imperfect analgesia, among others (53)(54)(55). Sleep problems are a hot topic in current clinical research due to their prevalence and potential negative impact on cognitive functions, including learning, memory, spatial orientation, behavioral capacity, and so on (9,(56)(57)(58). Furthermore, long-term sleep disturbances are intimately associated with major depression and dementia in adult populations, especially the aged (59,60). The mechanism underlying sleep-disturbance-related cognitive dysfunction is still unclear. Some studies reported that poor sleep quality (OSA, disordered circadian rhythms, and psychologically-based sleep deprivation) could lead to neuronal apoptosis in those areas of the brain related to cognition, by means of neuroinflammation,  changes in neurotransmitter activity (e.g., adenosine), and cerebral hypoxic and hypoperfusion injury (61)(62)(63)(64). Other studies confirmed that sleep disturbance was rather common in the perioperative period, and that it seriously affected postoperative cognitive function (65,66). A great number of studies have reported the effects of perioperative sleep disturbance on POD, and most of them concluded that poor perioperative sleep quality was an important risk factor of POD (67)(68)(69). We performed this meta-analysis to clarify and substantiate these findings, collecting as many articles as possible, listing their trial characteristics, and synthesizing their results.
All enrolled observational trials reported the POD incidence in patients exposed to sleep disturbances and non-sleep disturbances. Because the aim of this meta-analysis was to investigate the effect of perioperative sleep quality on POD, we selected studies with significant differences in sleep quality between intervention and control groups in RCTs. Different from the observational trials, in RCT group, some patients in the intervention group suffered sleep disturbances and some in the control group exhibited good sleep quality; thus, the negative pooled result of RCTs may be unreliable. We detected high heterogeneity in the ROT and RCT groups, which could affect the reliability of our meta-analysis results. We used sensitivity analysis-namely subgroup analysis and one-by-one literature exclusion-to address high heterogeneity (70,71). In addition, though meta-analysis using a random-effect model did not solve heterogeneity, it did decrease the impact of significant heterogeneity on the pooled results (72). As a result, we conjectured that high heterogeneity may be the result of a combination of factors, and we used the method of one-byone literature exclusion to solve this problem. Consequently, eight trials were excluded from the ROT and RCT groups (that is, 4 trials each). We then used a fixed-effects model with OR to conduct meta-analyses for the remaining literature in these groups, and the pooled results were consistent with those prior to sensitivity analysis.
Publication bias was another problem affecting the reliability of meta-analysis results (73). Detecting and adjusting publication bias is indispensable for meta-analysis. Currently, the primary methods for detecting publication bias include the rank correlation test (Begg's test, Schwarzer's test, and arcsine Begg's test), the regression test (Egger's test, Macaskill's regression, Harbord's test, Peters' test, and arcsine regression methods), and funnel plots. Funnel plots are not suitable for meta-analyses with few traits and high heterogeneity due to their tendency to produce asymmetric graphs (74,75). In this meta-analysis, there were only seven trials in the ROT group, so we did not draw a funnel plot. Of the other methods, Egger's test has the highest power and the most accurate p-value, and it is easy to understand. As such, it is the most popular method for detecting publication bias (24). If the p-value for a given group was <0.05 following Egger's test, we determined that this group exhibited significant publication bias. We used trim-and-fill analysis to adjust publication bias (24). Unexpectedly, there was no evidence of significant publication bias in any of the three groups in this meta-analysis.
The advantages of this meta-analysis were as follows. First, this meta-analysis included as many trials as possible, and it did not exclude trials based on study methods. Second, we selected observational trials with exposure and non-exposure factors (sleep disturbances) and RCTs with significant differences in risk factors (sleep quality) between intervention and control groups; thus, the results were more convincing. Third, grouping the studies according to their different study methods helped us to synthesize data. Fourth, different sleep disturbances (OSA, disordered circadian rhythms, and psychologically-based sleep deprivation) were included in this meta-analysis, allowing us to analyze the effects of perioperative sleep quality on POD more comprehensively.
Several limitations should also be taken into consideration in our meta-analysis. First of all, although we observed a positive correlation between sleep disturbance and POD in the observational study groups (retrospective and prospective), this finding may be less reliable than it could be because of the inevitable selection bias (76). Meanwhile, the most enrolled observational studies in this meta-analysis presented small sample size, which may attenuate the reliability of synthesized results as well (77). Furthermore, the study from Gupta et al. (28) only provided the number of matched patients in the control group, therefore we were not sure whether real-world research would affect the pooled results. In addition, although there was a striking difference in sleep quality between the RCTs' intervention and control groups, some patients in the intervention group suffered sleep disturbances and some in the control group exhibited good sleep quality; thus, the negative pooled result of RCTs may be unreliable. Therefore, a realworld prospective observational study with large sample size can significantly elevate the validity and reliability of results in spite of selection bias (78,79). Different onset time and POD assessment methods may also have affected the reliability of our pooled results. Lastly, the low-quality literature in each of the three groups likely compromised the reliability of our pooled results as well.

CONCLUSION
This systematic review and meta-analysis demonstrated that perioperative sleep disturbances were significantly associated with elevated POD incidence in observational trials (retrospective and prospective), but not in RCTs. Despite this inconsistency, we suggest that perioperative sleep disturbance could be a potential risk factor for POD, and that clinicians should pay careful attention to this phenomenon. In the future, the high-quality real-world prospective observational trial with large sample size will be required to further prove the effect of perioperative sleep disturbances on POD.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author/s.

AUTHOR CONTRIBUTIONS
HW and LZ were responsible for document retrieval and were responsible for adjusting data discrepancies. ZZ and YL performed the screening process for titles and abstracts. HW and SY performed the screening process for full texts. YL and QL were responsible for extracting the data. QL and SY independently assessed the quality of included studies. QL conducted the statistical analysis and made the figures and tables. HW prepared the manuscript. FY supervised the whole process and ensured the effectiveness of the meta-analysis. All authors read and approved the submission of the final manuscript.

ACKNOWLEDGMENTS
We would like to thank Dr. Yang Wang (Department of Biostatistics, the Chinese Academy of Medical Sciences, Fuwai Hospital, China) for his help with statistical data management.