Prevalence and risk factors of postoperative delirium after spinal surgery: a meta-analysis

Objective Postoperative delirium (POD) was common after spinal surgery, but the main findings in previous studies remained conflicting. This current meta-analysis was aimed at exploring the prevalence and risk factors of POD after spinal surgery. Methods PubMed and Embase were searched from inception to June 2019. Studies which reported the prevalence and risk factors of POD after spinal surgery were included. STATA version 12.0 was employed to analyze the pooled data. Statistical heterogeneity across included studies was identified using the I2 statistics. Results A total of 28 studies with 588,732 patients were included in the meta-analysis. The pooled prevalence of POD after spinal surgery was 0.85% (95%CI, 0.83–0.88%) with substantial heterogeneity (I2 = 97.3%). The central nervous system disorder (OR 4.73; 95%CI, 4.30–5.19) was a strong predictor for POD, whereas age (OR 1.16; 95%CI, 1.05–2.47; I2 = 99.2%) and blood loss (OR 1.10; 95%CI, 1.01–1.20; I2 = 93.3%) were weaker predictors. The funnel plot and statistical tests suggested that there existed potential publication bias, but the trim and fill method indicated that the pooled prevalence basically kept stable after adding two “missing” studies. Conclusions The pooled POD after spinal surgery ranges from 0.83 to 0.88%. The central nervous system disorder, age, and blood loss were potential risk factors for POD.


Introduction
Delirium, an acute state of confusion, is characterized as the distortion in consciousness and perception, decreased capacity of focusing one's attention, deteriorated cognitive functions, and disturbed sleep-wake cycles [1,2]. Postoperative delirium (POD) is a common complication after any major surgical procedure, which predominantly occurs in elderly [3,4]. POD is associated with loss of independence, longer hospital stay, aggravated cognitive capacity, increased morbidity and mortality risk, and greater medical economic burden [5][6][7]. Unfortunately, the treatments for POD are full of challenges currently [3]. In general, identifying POD-associated risk factors is a potential useful way to understand the characteristics of POD, so it is essential to identify the potential perioperative risk factors which may help to establish effective strategies for prevention and treatment. In 2015, Shi et al. performed a meta-analysis to identify the POD-associated risk factors after spinal surgery [8]. However, the previous meta-analysis only included six studies, and thus, the reliability of its conclusion may be limited by the small sample size. Moreover, following the meta-analysis by Shi and coworkers, a large body of studies was performed to make a further exploration on the POD-associated risk factors after spinal surgery. Additionally, an accurate estimation of the POD incidence is also of much significance. On the one hand, in the intervention studies without a placebo group, a precise estimate of incidence is needed for comparison to determine whether the intervention could effectively prevent POD. The researchers could calculate the appropriate number of subjects needed for intervention studies based on the incidence [9]. On the other hand, an accurate estimate of incidence may help to identify some certain subgroups of patients, which would attract more attention of doctors to adjust interventions for the specific patient populations, and it meanwhile may guide researchers to establish the scientific inclusion of clinical trials, matching the interventional and control groups well in terms of POD-associated baseline risk variables. Nevertheless, currently, the reported prevalences of POD vary too widely following spinal surgery with inconsistency.
Therefore, in this meta-analysis, we combined the currently available studies to systematically examine the prevalence and risk factors of POD following spinal surgery.

Methods
This meta-analysis was performed according to the guideline of the Meta-analysis of Observational Studies in Epidemiology checklist and the Cochrane Handbook [10]. Two reviewers separately performed selection criteria, data extraction, quality assessment, and statistical analysis, with inconsistence resolves by a third reviewer.

Search strategy and selection criteria
A systematic literature search was performed in PubMed and Embase for studies published before June 10, 2019. The search terms included delirium, risk factor, spinal surgery, and their variants. Additionally, the reference lists of the eligible studies and relevant reviews were carefully screened to identify any potential inclusion. All eligible observational studies, which reported the incidence of delirium after spinal surgery or provided relevant information to calculate the incidence of POD, were assessed for inclusion in this meta-analysis. Studies enrolling fewer than 25 subjects, with overlapping patients, or without available data were excluded. When two or more studies included the overlapping populations, the one with the largest sample size and the longest duration was chosen for the current meta-analysis.

Data extraction
Two co-authors independently extracted relevant data by using a pre-determined Excel sheet. The items of data extraction included the first author, year of publication, country, years of survey, type of operation, study design, mean age, sample size, risk factor, the number of delirium, and delirium assessment methods. The primary outcome is the prevalence of POD after spinal surgery. Additionally, the odds ratios (ORs) with corresponding 95% confidence intervals (CIs), which described the perioperative risk factors for POD after spinal surgery, were also extracted. Specially, we merely extracted individual risk factors which were assessed on multivariate or adjusted analysis in at least two studies.

Quality assessment
The quality of eligible studies was evaluated using the Newcastle-Ottawa Scale (NOS) score [11,12]. This score system is established specifically for assessing the quality of observational studies, in which scores are assigned for three dimensions, including selection criteria of participants, comparability, exposure, and outcome. A maximum score of NOS is up to 9, suggesting the highest quality.

Data synthesis and analysis
Stata SE12.0 (Stata Corp., College Station, TX, USA) was used to estimate the pooled prevalence of POD after spinal surgery. According to the Cochrane Handbook (9.5.4), the random-effects estimate and its confidence interval address the question "what is the average effect?" while the fixed-effect estimate and its confidence interval address the question "what is the best estimate of the effect?" When the results of pooled analysis based on random-effects model and fixed-effect model are very different from each other owing to substantial heterogeneity across included studies, the fixed-effect estimate and its confidence interval, but not the random-effects estimate may more truthfully reflect the pooled prevalence of POD in spine surgery. Besides, sensitivity analyses were performed by excluding one study at each step to further assess the influence of individual included studies on the overall synthesis analyses. If individual risk factors of interest were reported in two or more studies, the pooled OR estimates with 95% CIs were calculated. Statistical heterogeneity across studies was evaluated using I 2 statistic (I 2 > 50% was regarded as substantial heterogeneity) [13,14]. Subgroup analysis and meta-regression analysis for the primary outcome were used to detect the potential source of heterogeneity. The following categorical variables were analyzed in subgroup analyses: (1) region: Asia vs. Europe vs. North America; (2) year of the survey: before 2010 vs. after 2010 by the median splitting method; (3) sample size: ≤ 500 vs. > 500 by the median splitting method; (4) type of operation: oncological spine surgery vs. non-oncological spinal surgery; (5) mean age of patients: ≤ 60 vs. > 60; (6) study design: database analysis vs. non-database related observational study; and (7) preoperative disease status: with preoperative cerebrovascular disorders vs. without preoperative cerebrovascular disorders. Continuous variables including age, year of publication, and NOS score were analyzed by meta-regression analysis. Publication bias was assessed using Begg and Egger's test and funnel plot [15,16]. If there was significant publication bias, the "trim and fill method" was used to determine whether it obviously affected the robustness of the synthesis analysis [17]. P ≤ 0.05 was deemed to be statistically significant.

Study selection and characteristics
A total of 576 articles were identified initially. After removing duplicate records and irrelative titles and abstracts, the full texts of 66 articles were further screened for eligibility. Finally, a total of 28 studies with 588,732 patients were included in this meta-analysis . The flow diagram for the study selection was presented in Fig. 1. The included studies were published from 2006 to 2019. Among all eligible studies, two were performed in Europe, nine in North America, and the others in Asia. The detailed characteristics of the included studies were shown in Table 1. The NOS score of all the included studies ranged from 6 to 9, suggesting the quality of included studies was relatively high for the current meta-analysis. The detailed NOS score of the included studies are shown in Table 2.

Postoperative delirium after spine surgery
All included studies reported POD after spinal surgery. The pooled prevalence of POD was 0.85% (95%CI, 0.0083-0.0088; I 2 = 97.3%; Fig. 2a) using fixed-effect model, but the pooled prevalence based on randomeffects model was 12% (95%CI, 0.09-0.14; I 2 = 97.3%; Fig. 2b). According to the Cochrane Handbook, the fixed-effect estimate may more truthfully reflect the authentic pooled prevalence of POD in spine surgery when there existed a significant difference between the fixedeffect and random-effects estimates with substantial heterogeneity. Furthermore, we performed sensitivity analyses to explore the influence of individual included studies on the overall pooled effect. The results indicated that the pooled prevalence of POD basically remained   (Table 3). Subgroup analyses indicated that there were significant differences in the incidences of POD among Asia (7.7%), Europe (5.3%), and North America (0.8%), as well as between oncological spinal surgery (5.3%) and nononcological spinal surgery (0.9%). In subgroup analyses stratified by year of survey, the incidence of POD after 2010 (4.1%) was higher than that before 2010 (0.9%).
Similarly, in stratified analyses by mean age, we observed that the incidence of POD in patients older than 60 years (8.2%) was also higher than that in patients younger or equal to 60 years (0.9%). In subgroup analysis by study design, the incidence of POD in the subgroup of nondatabase related observational study (8.4%) was higher than that in the database analysis subgroup (0.8%). Subgroup analysis based on preoperative disease status showed that the incidence of POD in patients with preoperative cerebrovascular disorders (8.7%) was significantly higher than that in patients without preoperative cerebrovascular disorders (0.8%). When stratified by sample size, the incidence of POD in ≤ 500 group (8.8%) was higher than that in > 500 group (0.8%). The subgroup analyses were detailed in Table 4. Furthermore, the meta-regression analyses showed that publication time (p = 0.041), but not sample size (p = 0.183) and NOS score (p = 0.975), was significantly associated with higher POD after spine surgery. The funnel plot seemed to be asymmetric and statistical tests (Egger's test, p = 0.797 and Begg's test, p = 0.008; Fig. 3a) also suggested the significant evidence of publication bias. However, the pooled prevalence for POD (0.9%; 95%CI, 0.008-0.009) did not change significantly after adding two "missing" studies from the "trim and fill" analysis ( Fig. 3b).

Perioperative risk factors for postoperative delirium
A total of nine risk factors associated with POD after spine surgery were reported on multivariate or adjusted  analysis in two or more included studies (

Discussion
The current study shows that the POD was a serious complication after spine surgery with pooled prevalence ranging from 0.83 to 0.88%. The central nervous system disorder was a strong predictor, whereas age and blood loss were the weaker predictors for POD after spine surgery.
In the present meta-analysis, a total of 28 studies were identified to evaluate the incidence of POD following spinal surgery. The overall pooled prevalence of POD was 0.85 % with substantial heterogeneity. We found that the incidence of POD in elder patients (> 60) was higher than that in younger patients (≤ 60), suggesting that older age might be a risk factor for POD after spinal surgery. In accordance with these, our data synthesis analysis revealed that older age was a significant risk factor for POD after spinal surgery (OR 1.16, 95%CI 1.05-2.47). In general, the elderly patients usually have poor general health status, more physical and psychological problems, and decreased functioning; all of which might contribute to the occurrence of POD after spinal surgery [46,47]. Meta-regression analyses also found that publication time was significantly associated with higher POD prevalence. A possible interpretation was that the global aging population trend in these years may be an important contributor to the increased prevalence of POD accompanied with years [48]. Subgroup analysis also revealed that the pooled prevalence of POD in patients with preoperative cerebrovascular disorders was approximately two times higher than those without preoperative cerebrovascular disorders. Additionally, we found that the central nervous system disorder was identified as a strong predictor for POD after spine surgery as well. Actually, numerous previous studies have demonstrated that some cerebrovascular disorders including Alzheimer's disease and dementia were associated with a high risk of delirium, which may help to explain our findings [49][50][51][52]. Our study also found that the male gender may be a significant risk factor for POD after spine surgery. A possible reason was that cerebrovascular disorders have a high prevalence among men versus women [53][54][55], which may lead to a high risk of POD after spine surgery. Furthermore, we identified that hemoglobin < 100 g/L, blood loss, and blood transfusion were potential predictors for POD. An important reason for this was that perioperative oxygen insufficiency of the central nervous system facilitated the development of POD after spine surgery. Other risk factors, such as operative time, MMSE score, and ASA score were possible risk factors for POD, although the pooled results were no statistical significant. Taken together, perioperative management focused on these aforementioned risk factors may reduce the risk of POD after spine surgery. The current study also existed many limitations. Firstly, in the current study, the results of pooled analysis based on random-effects model and fixed-effect model are very different from each other owing to substantial heterogeneity across included studies. According to the Cochrane Handbook (9.5.4), the random-effects estimate and its confidence interval address the question "what is the average effect?" while the fixed-effect estimate and its confidence interval address the question "what is the best estimate of the effect?" Therefore, we just chose the fixed-effect estimate and its confidence interval, but not the random-effects estimate, which may more truthfully reflect the pooled prevalence of POD in spine surgery. The choice may be inappropriate, but can show the authentic pooled prevalence of POD in spine surgery.
Our study indicated that the overall pooled prevalence of POD was approximately 0.85% with substantial heterogeneity. Meta-regression analyses indicated that the year of publication was a significant contributor to statistical heterogeneity. Moreover, we performed subgroup analyses based on the year of survey and found that the incidence of POD after 2010 was lower than those before 2010. However, the statistical heterogeneity within subgroups was still substantial, so there may exist other  Additionally, we also performed the sensitivity analyses by sequentially excluding single study and subgroup analyses to explore the robustness and creditability of our overall pooled effect. We found that there is a high possibility that our pooled result was skewed in favor of the statistics reported by Fineberg. Of course, we cannot totally exclude the bias risk since all the included studies were retrospective studies and unavoidable heterogeneity. Thus, further homogeneous and prospective studies should be warranted to elucidate the prevalence of POD following spine surgery. Thirdly, some risk factors were reported in limited included studies, but the pooled estimate based on limited studies may bias the authenticity. The pooled analysis based on two studies found that the ASA score was associated with more than four-time risk of POD after spine surgery, but with no statistical significance. Actually, many studies revealed ASA score was a significant predictor for POD [56][57][58][59]. Moreover, there are other factors like electrolyte imbalance and general condition of the patient before surgery which were not identified in the current meta-analysis, but may also be potential risk factors for POD. However, these factors were reported in the limited studies, so we did not include them in this meta-analysis for the pooled estimate, considering that the limited studies may bias the authenticity of our pooled analysis. Therefore, our results in this meta-analysis may be too conservative and should be interpreted cautiously. Meanwhile, many other potential risk factors for POD should be further assessed in future studies. Finally, the funnel plot and statistical tests suggested that the current meta-analysis may exist publication bias, regardless of the fact that we have performed a systematic literature search. However, the pooled prevalence for POD basically remained stable after adding two "missing" studies, which further supported the reliability of the pooled effect.
To sum up, our study indicated that the pooled POD after spinal surgery was approximately 0.85%. The central nervous system disorder, age, and blood loss were potential risk factors for POD. Further studies with more homogeneous clinical parameters should be warranted to illuminate the prevalence and risk factors of POD after spine surgery.