Strategies for prevention of postoperative delirium: a systematic review and meta-analysis of randomized trials

Introduction The ideal measures to prevent postoperative delirium remain unestablished. We conducted this systematic review and meta-analysis to clarify the significance of potential interventions. Methods The PRISMA statement guidelines were followed. Two researchers searched MEDLINE, EMBASE, CINAHL and the Cochrane Library for articles published in English before August 2012. Additional sources included reference lists from reviews and related articles from 'Google Scholar'. Randomized clinical trials (RCTs) on interventions seeking to prevent postoperative delirium in adult patients were included. Data extraction and methodological quality assessment were performed using predefined data fields and scoring system. Meta-analysis was accomplished for studies that used similar strategies. The primary outcome measure was the incidence of postoperative delirium. We further tested whether interventions effective in preventing postoperative delirium shortened the length of hospital stay. Results We identified 38 RCTs with interventions ranging from perioperative managements to pharmacological, psychological or multicomponent interventions. Meta-analysis showed dexmedetomidine sedation was associated with less delirium compared to sedation produced by other drugs (two RCTs with 415 patients, pooled risk ratio (RR) = 0.39; 95% confidence interval (CI) = 0.16 to 0.95). Both typical (three RCTs with 965 patients, RR = 0.71; 95% CI = 0.54 to 0.93) and atypical antipsychotics (three RCTs with 627 patients, RR = 0.36; 95% CI = 0.26 to 0.50) decreased delirium occurrence when compared to placebos. Multicomponent interventions (two RCTs with 325 patients, RR = 0.71; 95% CI = 0.58 to 0.86) were effective in preventing delirium. No difference in the incidences of delirium was found between: neuraxial and general anesthesia (four RCTs with 511 patients, RR = 0.99; 95% CI = 0.65 to 1.50); epidural and intravenous analgesia (three RCTs with 167 patients, RR = 0.93; 95% CI = 0.61 to 1.43) or acetylcholinesterase inhibitors and placebo (four RCTs with 242 patients, RR = 0.95; 95% CI = 0.63 to 1.44). Effective prevention of postoperative delirium did not shorten the length of hospital stay (10 RCTs with 1,636 patients, pooled SMD (standard mean difference) = -0.06; 95% CI = -0.16 to 0.04). Conclusions The included studies showed great inconsistencies in definition, incidence, severity and duration of postoperative delirium. Meta-analysis supported dexmedetomidine sedation, multicomponent interventions and antipsychotics were useful in preventing postoperative delirium.


Introduction
An estimated 36.8% of surgical patients suffer from postoperative delirium [1]. The incidence is much higher in patients 70 years of age and older [2]. Delirium is associated with increased morbidity and mortality [3], prolonged hospital stay and persistent functional and cognitive decline [4]. Postoperative delirium is also a major burden to medical services with costs in US dollars ranging from $38 to $152 billion per year [5].
Prevention may be the most effective strategy for minimizing the occurrence of postoperative delirium and its adverse outcomes but it is untested or unproven. In hospitalized patients, 30 to 40% cases of delirium are thought to be preventable [6,7]. Multimodal strategies have been used in an effort to counter delirium resulting from diverse causes such as neurotransmitter imbalance, neuroinflammation, pain, infection, metabolic abnormalities and sleep disorders [8,9]. Widely applicable therapeutic countermeasures for delirium have not yet been discovered. It is not presently clear whether a single intervention for patients with different risk factors is a realistic goal, or whether there is an optimal treatment for specific groups of patients.
The purposes of this study were 1) to critically review available randomized clinical trials (RCTs) that assessed the effects of multiple kinds of interventions to prevent postoperative delirium in adult patients, 2) to determine the efficacy of interventions, and 3) to explore whether interventions successful in preventing postoperative delirium also shortened the length of hospital stay.

Materials and methods
This systematic review and meta-analysis was conducted following the guidelines of the PRISMA statement (Additional file 1) [10,11].

Study selection
The initial search returned 2,813 articles. After title and abstract review, 198 potential articles with full texts were further independently reviewed by two coauthors (HZ and YL) to determine the eligibility according to the predefined selection and exclusion criteria. Disagreements between reviewers were resolved by including another coauthor (XS). Completed studies that met all the following criteria were considered eligible for inclusion in the systematic review and meta-analysis: 1) RCTs assessing interventions to prevent postoperative delirium; 2) delirium identified by validated methods including the Diagnostic and Statistical Manual of Mental Disorders, 1987 (DSM-III), DSM-III-R (1994), DSM-IV (1999), the 10th revision of the International Statistical Classification of Diseases and Related Health Problems, 1992 (ICD-10), and clinical diagnostic tools based on these such as the Confusion Assessment Method (CAM), Delirium Rating Scale (DRS) and NEECHAM Confusion Scale [12]; 3) incidence, severity and duration of delirium analyzed independently of other neurologic events such as emergence delirium and dementia. Research articles were excluded if they recruited 1) patients with delirium before surgery; 2) patients with alcohol withdrawal syndrome; 3) groups that also included nonsurgical patients (for example patients in the intensive care unit or ward without surgery); 4) homogeneous populations of patients with certain central nervous system diseases or mental disorders (for example stroke, dementia, schizophrenia and depression).

Data extraction
Data extraction was completed by two coauthors (ML and ZZ) using a predesigned piloted data extraction form. Disagreements were resolved by the third coauthor (XS) consultation. The following study characteristics were collected: primary author, publication year, country of origin, PubMed identifier (if possible), types of surgery, participant characteristics (gender, age, number, existing illness, inclusion and exclusion criteria), intervention (type, dosage, duration and frequency), criteria for delirium, incidence, severity and duration of delirium, P value, duration and frequency of follow-up and the length of hospital stay. Dichotomous data were converted to incidences for data synthesis and continuous data were recorded using mean and standard deviation (SD).

Quality scoring of included trials
The validity and quality of included trials was evaluated independently by two coauthors (FX and LW) using a scoring system ( Table 1) that combined the modified Jadad scale [13] and the delirium-specific score we developed for the current study. The quality review system included eight items with a maximal score of 12. Studies with a score ≤ 5 were arbitrarily defined as lowquality studies with high risk of within-study bias. We designed this delirium-specific scoring system because postoperative delirium is defined subjectively with validated methods such as DSM-IV and ICD-10, has certain risk factors (for example age, sex, comorbidities and medications) and mostly occurs 24 to 72 hours after surgery [1,12,[14][15][16]. Disagreements were resolved by including a third author (XS) for discussion. Studies were not excluded or weighted based on quality scores in the meta-analysis.

Data analysis
The analyses focused on the incidence of postoperative delirium as the primary outcome measure. We further tested the hypothesis that interventions reducing postoperative delirium would shorten the length of hospital stay. Only studies reporting significant differences in the incidences of postoperative delirium between two interventions (P < 0.05, two-tailed) were included. Placebo and control procedures were also considered as interventions when a study aimed to compare the effects between interventions and placebos. Interventions were divided into two groups (interventions with less delirium and interventions with more delirium) and the length of hospital stay was synthesized for comparison.
Meta-analysis was performed when two or more than two studies using similar interventions were identified. Statistical analysis was performed using STATA 11 (Stata-Corp, College Station, TX, USA). A test for heterogeneity was performed using a standard chi-square (χ 2 ) and I-square (I 2 ) statistic. Significant heterogeneity was considered present at χ 2 P < 0.10 or I 2 > 50%. Where no heterogeneity was found, a fixed-effects parametric approach (weighted with inverse variance) was taken. Otherwise a random-effects model was used. For the incidence of postoperative delirium, both pooled relative risk (RR) and incidence with 95% confidence intervals (CI) were calculated. Sample size calculations of different interventions (n1 = n2, α = 0.05 and β = 0.1, two-tailed) were performed based on reported or pooled incidences. For the length of hospital stay, SMD (standard mean difference) was used due to that there was a big intertrial difference. We intended to conduct a subgroup analysis, where possible, to explore 1) the effects of different interventions, or 2) the effects of single intervention in patients with different surgeries. Publication bias was assessed by visually inspecting funnel plot and Begg's test. Meta-regression was performed to help investigate the origin of heterogeneity. For all the analyses, a P value of less than 0.05 (two-tailed) was considered statistically significant.

Study selection
The process of literature identification, screening and selection is summarized by Figure 1. Our primary search yielded 2,813 articles. After screening, 198 studies potentially met the inclusion criteria. After examining the full texts, 160 articles were excluded: 25 studies were not clinical trials; three studies had no control group; 40 studies did not include postoperative delirium as a study variable; three studies tested the diagnosis methods of delirium; 47 studies did not screen postoperative delirium using validated tools; seven studies recruited both surgical and nonsurgical patients; three studies did not provide the delirium data; 11 studies included patients with delirium prior to surgery; 12 studies included homogeneous patients with brain diseases, mental disorders or alcohol withdrawal syndrome; seven studies described ongoing trials; one study was retracted, and one study was identified as a duplicated publication. We ultimately included 38 RCTs  in our systematic review and meta-analysis.

Quality scores of included studies
The scores of included studies were shown in Table 4. The scores ranged from 3 [18] to 12 [36,52,54]. The average score was 8.3 with a standard deviation of 2.2. A score lower than 6 was found in four studies [17,18,21,22]. Three studies got the full score 12 [36,52,54].
Quantitative review and meta-analysis Category 1. Perioperative procedures and drugs (

Neuraxial anesthesia versus general anesthesia
We identified four studies with 511 patients [17,20,27,31] that compared the effects of different anesthesia methods on postoperative delirium. Meta-analysis using a fixed-effects model (χ 2 (3) = 4, P = 0.261, I 2 = 25%) revealed no difference between neuraxial and general anesthesia (pooled RR = 0.99, 95% CI = 0.65 to 1.50, P = 0.962, Figure 2A). The pooled incidences based on a random-effects model were 17.1% (95% CI = 7.8% to 37.8%) for neuraxial anesthesia and 17.1% (95% CI = 9.3% to 31.4%) for general anesthesia. (DRS) scores compared to patients receiving sevoflurane (6 ± 3 vs. 2 ± 1, P = 0.002) in the study by Nishikawa et al. [28]. There was no difference in the incidences of postoperative delirium (16% vs. 0, propofol vs. sevoflurane, P = 0.110). Power calculations suggested that 58 patients per group would be needed to achieve a significant difference in the incidences of delirium but the current study only included 25 patients per group.      Royse et al. [51] tested the influence of either propofol or desflurane on the incidence of postoperative delirium in patients undergoing coronary artery bypass surgery. Seven of 89 patients receiving propofol and 12 of 92 patients receiving desflurane developed delirium. No difference was found in incidence between the two groups (7.9% vs. 13.2%, propofol vs. desflurane, P = 0.245). Power calculations suggested that 732 patients per group would be needed to observe a significant difference in delirium occurrence based on the reported incidences but a total of 171 patients were enrolled in the study.
Less postoperative delirium was found in patients receiving additional ketamine (0.5 mg/kg intravenously, single bolus) for anesthesia induction compared to standard methods (3.45% vs. 31.03%; ketamine vs. standard, P = 0.012). Hudetz et al. [40] recruited 29 patients undergoing cardiac surgery per group for this study.
Leung et al. [34] found no additional effect of N 2 O on the development of postoperative delirium compared to standard anesthesia in older patients undergoing noncardiac surgery. Forty-four of 105 patients (41.9%) exposed to additional N 2 O and 46 of 105 patients

Antipsychotics
We identified six trials with 1,592 patients which tested the role of antipsychotics on preventing postoperative delirium. Three trials used the typical antipsychotic haloperidol [22,29,54]. The doses varied from 1.5 mg/day to 5 mg/day with a duration of one to five days. The other three trials used atypical antipsychotics risperidone and olanzapine [36,45,52].

Anticonvulsants
Leung et al. [33] tested whether oral gabapentin (900 mg/day, for four days) was helpful in preventing postoperative delirium in older patients undergoing spine surgery. Delirium was identified in none of the nine patients receiving gabapentin and in five of the twelve patients receiving placebo (P = 0.045). Pesonen et al. [50] randomly assigned oral pregabalin (150 mg/day, for six days, n = 35) or placebo (n = 35) to patients accepting cardiac surgery. Their study found that pregabalin was able to reduce the severity of delirium (CAM-ICU, 24 ± 8 vs. 21 ± 19, P = 0.04). 2.4 Sleep restoration using diazepam, flunitrazepam and pethidine Aizawa et al. [26] tested whether restoring sleep-wake cycle with medications after surgery was useful to prevent postoperative delirium. The researchers randomly divided 40 patients accepting major abdominal surgeries into two groups. The experimental group (n = 20) received standard treatment plus diazepam/flunitrazepam/pethidine (DFP) for three days to improve sleep disorders and the control group underwent standard treatment (n = 20). Less delirium was developed in the DFP group (5% vs. 35%, DFP vs. standard, P = 0.023).

Sleep restoration using bright light
We identified two studies [38,49] with 33 patients that tested the hypothesis that improving the sleep-wake cycle using bright light (two hours per day in the morning, 2500 to 5000 lx) would be useful to prevent delirium. Meta-analysis using a fixed-effects model (χ 2 (1) = 0.15, P = 0.703, I 2 = 0) found no difference between bright light and control (pooled RR = 0.30, 95% CI = 0.07 to 1.26, P = 0.099). The pooled incidence utilizing a fixed-effects model were 13.0% (95% CI = 2.1% to 78.8%) for bright light and 41.3% (95% CI = 20.9% to 81.5%) for control. Power calculations suggested that 50 patients per group would be needed to get a significant difference in delirium occurrence based on the pooled incidences but only a total of 33 patients were included in the two trials. Besides the effects on the incidences of delirium, Taguchi et al. [38] found bright light therapy reduced the severity of delirium (NEECHAM, 6.7 ± 0.7 vs. 21.1 ± 7, bright light vs. standard, P = 0.014). 2.6 Psychological intervention Schindler et al. [18] detected the role of active daily psychological intervention on postoperative delirium in patients undergoing cardiac surgery. No difference was found on the incidences of delirium between the two groups (0 vs. 12.5%; psychological intervention vs. standard, P = 0.227). Power calculations suggested that a total of 154 patients were needed to observe a difference in delirium occurrence based on the reported incidences but the study included only 33 patients.

Music
McCaffrey et al. [42] recruited 22 patients (11 patients per group) and evaluated the effects of music on delirium prevention. The patients in the music group received standard hospital care plus listening to soothing lullaby music at least four times a day for one hour. Delirium severity was detected using NEECHAM confusion scale on each of the first three postoperative days. They found that listening to music decreased the severity of delirium (NEECHAM, 24 ± 0.97 vs. 22.5 ± 1.22, music vs. standard, P = 0.000).

Multicomponent interventions
Multicomponent interventions that combined both pharmacological and non-pharmacological strategies were performed to prevent postoperative delirium in two RCTs with 325 patients [25,35]. Meta-analysis using a fixed-effects model (χ 2 (1) = 0.55, P = 0.608, I 2 = 0) found that multicomponent interventions decreased the incidence of postoperative delirium (pooled RR = 0.71, 95% CI = 0.58 to 0.86, P = 0.000). The pooled incidences were 43.3% for multicomponent interventions and 62.4% for standard treatment. In addition, Marcantonio et al. [25] found multicomponent interventions reduced the number of patients with severe delirium (7/62 (11.3%) vs. 18/64 (28.1%); multicomponent interventions vs. standard, P = 0.02) and had no effect on the duration of delirium in patients suffering from delirium (2.9 ± 2 vs. 3.1 ± 2.3 days; multicomponent interventions vs. standard, P = 0.73). Lundstrom et al. [35] reported that multicomponent interventions shortened the duration of delirium (5.0 ± 7.1 vs. 10.2 ± 13.3 days; multicomponent interventions vs. standard, P = 0.009). 2.9 Histamine H 2 receptor blockers Kim et al. [21] found no difference in the incidences of postoperative delirium between cimetidine and ranitidine in postoperative cardiac surgical patients. The incidences were close (24.5% vs. 25.9%, cimetidine vs. ranitidine, P = 0.872). More than 20,000 patients for each group would be needed to observe a significant difference in the incidences of delirium and a total of 111 patients were included in the study.
Interventions effective in preventing postoperative delirium did not shorten the length of hospital stay We identified 10 studies with 1,636 patients reporting both different incidences of postoperative delirium between the two interventions and inpatient time [25,26,35,36,40,41,44,47,52,54]. Meta-analysis using a fixed-effects model (χ 2 (9) = 12.1, P = 0.208, I 2 = 25.6%) found no significant difference in the length of hospital stay between interventions with lower or higher incidences of postoperative delirium (pooled SMD = -0.06, 95% CI = -0.16 to 0.04, P = 0.159, Figure 4A). The pooled incidences based on the fixed-effects model were 16.1% for interventions with less delirium and 35.4% for interventions with more delirium. No significant publication bias was found by Begg's test (z = 0.54, P > |z| = 0.592) and by visual inspection of the funnel plot ( Figure 4B).

Summary of evidence
The main findings to emerge from this systematic review and meta-analysis include: 1) There was a huge heterogeneity of interventions among trials. Most of the interventions suffered from only a handful of studies and small sample sizes. Some studies suffered from methodological defects and were at high risk of bias [17,18,21,22]. These shortcomings disallowed sufficient interpretation of the effectiveness of interventions.
2) The meta-analysis showed that dexmedetomidine sedation, multicomponent interventions and antipsychotics were useful in preventing postoperative delirium. 3) Based on the result and quality of individual study, it appeared that light sedation during spinal anesthesia, additional ketamine during anesthesia induction, additional fascia iliaca compartment block, anticonvulsants and sleep restoration using diazepam/flunitrazepam/pethidine were useful in preventing postoperative delirium. 4) The meta-analysis found that interventions useful for preventing postoperative delirium did not shorten the patients' time spent in the hospital. Table 5 provides a summary of the efficacy of interventions for diminishing delirium.
Consistent with our results, Lin et al. [58,59] found that dexmedetomidine sedation was inversely related with the incidence of delirium in patients with cardiac surgery. Contrarily, Tan et al. [60] concluded that the use of dexmedetomidine in critically ill adult patients had no effect on delirium in their meta-analysis. Besides the different doses and durations of dexmedetomidine use, patients with different illnesses might develop delirium due to different reasons [61], which might help explain the discrepancy.
Lonergan et al. [62] found that both typical antipsychotics (haloperidol) and atypical antipsychotics (olanzapine, risperidone and quetiapine) were effective in treating delirium in their meta-analysis. These results were consistent with the current meta-analysis which tested the role of antipsychotics on delirium prevention. Studies by Devlin et al. [63,64] and Skrobik et al. [65] also found a positive role of quetiapine and olanzapine in treating delirium in critically ill patients. Campbell et al. [66] found no superiority for second-generation antipsychotics over haloperidol in managing delirium. Devlin et al. [67] had critically reviewed six studies [54,65,[68][69][70][71] which used haloperidol to prevent or treat delirium in noncritically or critically ill patients. Only the study by Wang et al. [54] showed that low-dose haloperidol reduced the incidence of delirium compared to placebo. The inconsistent results of haloperidol might be due to the following reasons: 1) There was a great heterogeneity of the patient populations among the six studies [54,65,[68][69][70][71]. These studies included patients with severe illnesses (AIDS and cancer) and patients accepting different surgeries or critically ill patients. 2) The comparator of haloperidol was different among studies. Girard el al. [71] used both atypical antipsychotics and placebo as the control. Wang et al. [54] used placebo as the control. Chlorpromazine and lorazepam were used as the control for haloperidol in the study by Breitbart et al. [68]. Another three studies [65,69,70] used atypical antipsychotics (olanzapine or risperidone or ziprasidone) as the control.
3) The dosage and duration of haloperidol differed greatly among studies.
In support of the National Institute for Health and Clinical Excellence guidelines recommending an individual multicomponent intervention package aiming to prevent delirium [72,73], two studies included in our meta-analysis supported multicomponent interventions as a useful way to prevent postoperative delirium [25,35]. A recent study also showed a 30% reduction of delirium by multimodal geriatric consultation versus usual care in older adults with recent hip fracture [74]. Additional similar studies are being performed, which might add evidence to the finding [75][76][77]. Our meta-analysis data found no difference in the length of patient hospital stay between interventions with higher incidences of delirium (pooled incidence, 35.4%) and interventions with lower incidences (pooled incidence,16.1%). This finding was contrary to results from previous observational studies which showed that patients with postoperative delirium stayed longer in the hospital [4,[78][79][80]. Resolving these differences is important as prolonged hospital stay is a heavy burden on the health care system [5] and should also be included as an important clinical outcome during delirium prevention [67]. However, as only 21 of the 38 included trials reported the inpatient time, there was a potential publication bias. Furthermore, our meta-analysis included heterogeneous studies with huge differences in both the incidences of postoperative delirium and the time of hospital stay. Further clinical trials with homogenous patients receiving similar interventions might help to clarify this issue.

Limitations
This systematic review and meta-analysis had several limitations. 1) We included different types of surgeries for a single intervention and this possibly affected the heterogeneity. 2) Multiple methods and different frequencies of postoperative delirium screening across the studies were another source of heterogeneity [1]. 3) For the same intervention, the dose and duration varied greatly among studies and this might account for different effects [70]. 4) Application of the scoring system developed for this study revealed methodological defects in some studies. These defects added a degree of uncertainty to the present results. For example, given that delirium is a multifactorial disorder, similar baseline data were essential when comparing the effects of two interventions. However, 18 [17][18][19]21,25,27,28,30,32,35,37,38,42,45,46,48,49,53] of the 38 studies did not adjust the risk factors before grouping. 5) Publication bias might account for some of the effects reported here. Most of the included studies were small-sampled single-centered studies with less methodological rigor than large-sampled studies. This factor might contribute to an overestimation of effect sizes in small trials. 6) We excluded homogeneous populations of patients with dementia in our study [81,82]. In addition, only two studies [25,35] stated that they included a small subpopulation of patients with dementia. Considering the high morbidity of dementia in the adults and the overlap of dementia with, and contribution to, delirium [3,[83][84][85], we have excluded a large group of patients who were susceptible to postoperative delirium. This exclusion should be seen as a source of potential selection bias and could limit the interpretation of our findings.

Future directions
Our review raised several questions that need to be addressed in future studies: 1) There were three types of postoperative delirium: hyperactive (25%), hypoactive (50%) and mixed (25%) delirium which had different causes and consequences [1,16,86]. However, none of the existing studies tried to distinguish them or tested the specific effects of interventions. Future studies Table 5 Summary of the efficacy of the interventions.

Conclusions based on
Perioperative procedures and drugs (>, superior to; =, equally effective to; UC*, uncertain) Pharmacological, psychological or multicomponent interventions (>, superior to; =, equally effective to; UC, uncertain) should include screening tools such as the Richmond Agitation-Sedation Scale (RASS) to classify the subtypes of delirium [87] and test their reactions to various interventions. 2) The severity and duration of delirium needed more attention. Moreover, the severity and duration of delirium should be averaged for all patients but not only for patients with delirium. 3) High-risk and low-risk patients might show different sensitivity to precipitating factors and interventions. Thus, there is a need for future studies that stratify high-risk patients and low-risk patients in delirium assessment. We identified only two studies stratifying the risk of delirium [29,43]. One study only included patients with subsyndromal delirium [52]. Further studies using valid riskstratifying tools for delirium [12] can make a contribution to this important clinical problem.

Conclusions
Heterogeneity and small sample sizes precluded conclusions regarding the interventions that are likely to prevent postoperative delirium. The limited data suggested that the efficacious way to prevent postoperative delirium included dexmedetomidine sedation, multicomponent interventions and antipsychotics comprising haloperidol, olanzapine and risperidone. Anesthesia types and analgesia methods had no bearing on delirium. Acetylcholinesterase inhibitors were ineffective in preventing delirium. Interventions effective in preventing postoperative delirium did not shorten the length of hospital stay. Considered together, these findings suggested an urgent need for high-quality large-scale RCTs.

Key messages
• Multiple strategies including perioperative management procedures, pharmacological and nonpharmacological interventions have been used in an effort to prevent postoperative delirium.
• There is a consensus in the data that dexmedetomidine sedation, multicomponent interventions and antipsychotics are useful in preventing postoperative delirium.
• Anesthesia types and analgesia methods have no bearing on postoperative delirium.
• Acetylcholinesterase inhibitors are ineffective in preventing postoperative delirium.
• Reduced postoperative delirium is not related with shortened hospital stay. Authors' contributions HZ, YL and XS conceived and designed the study. HZ and YL carried out the literature search. ML, ZZ and XS carried out the data extraction. FX, LW and XS carried out the quality assessment. HZ and YL analyzed and interpreted the data. HZ, YL and XS prepared and revised the manuscript. All authors have read and approved the manuscript.