A Meta-Analysis and Meta-Regression of Frequency and Risk Factors for Poststroke Complex Regional Pain Syndrome

Background and Objectives: This article aimed to investigate the risk factors for poststroke complex regional pain syndrome (CRPS). Materials and Methods: We searched electronic databases including PubMed, Medline, Web of Science, Cochrane Library, and Embase up to 27 October 2021. We enrolled analytical epidemiological studies comprising cohort, case-control, and cross-sectional studies. A quality assessment was performed using the Newcastle–Ottawa Quality Assessment Scale for cohort and case-control studies and the Joanna Briggs Institute critical appraisal checklist for analytical cross-sectional studies. Binary outcomes were reported as odds ratios (ORs), and continuous outcomes were described as standardized mean differences (SMDs) with 95% confidence intervals. For the meta-regression, beta coefficient and p value were adopted. Results: We included 21 articles comprising 2225 participants. Individuals with shoulder subluxation and spasticity were found to have higher risks for poststroke CRPS. Spasticity with higher modified Ashworth scale score, lower Brunnstrom hand stage, and inferior Barthel index scores were observed in patients with poststroke CRPS. The pooled incidence proportion in nine articles was 31.7%, and a correlation was found between effect sizes and the ratio of women and the proportion of left hemiparesis. The summarized prevalence in nine cross-sectional studies was 33.1%, and a correlation was observed between prevalence and the subluxation ratio and Brunnstrom stage. Conclusions: Based on our meta-analysis, being female, left hemiparesis, shoulder subluxation, spasticity, a lower Brunnstrom stage of distal upper limb, and an inferior Barthel index are all features for poststroke CRPS. Larger studies with greater statistical power may confirm our findings and clarify some other unknown risk factors for poststroke CRPS.


Introduction
Complex regional pain syndrome (CRPS) is a clinical syndrome characterized by pain, sensory, motor, vasomotor, and trophic changes [1]. Different diagnostic criteria have been proposed, including the Budapest criteria, which is currently the criteria most commonly adopted [2]. Most poststroke CRPS is considered CRPS type one, also known as reflex sympathetic dystrophy. The pathogenesis of poststroke CRPS is still unclear [3], and the prevalence varies from 2% to 50% based on previous studies [4]. Poststroke CRPS usually occurs one to six months after a cerebrovascular accident (CVA), which happens to be the period with the highest potential for rehabilitation [5]. Hence, prevention, early diagnosis, and treatment of poststroke CRPS are important following a stroke.
Currently, no guidelines for the prevention of poststroke CRPS have been established. Some investigations have revealed early rehabilitation may decrease the likelihood of

Quality Assessment
We used the Newcastle-Ottawa Quality Assessment Scale for the cohort and casecontrol studies and the Joanna Briggs Institute (JBI) critical appraisal checklist for analytical cross-sectional studies [17]. Two authors (Y.-C.S. and Y.-H.G.) independently evaluated the articles, and disagreements were resolved by discussion with the senior author (Y.-C.L.). Reviewer Manager version 5.3 was utilized to summarize the results in the form of a graph of risk of bias and summary table.

Statistical Analysis
The primary outcomes were factors considered to contribute to the frequency of poststroke CRPS. We conducted a meta-analysis if a determinant was appropriately mentioned at least three times in similar populations. The odds ratios (ORs) with 95% confidence intervals (CI) were reported for binary outcomes, and standardized mean differences (SMDs) with a 95% CI were used for the continuous outcomes. The secondary outcome was the frequency of CRPS following a CVA, presented with a 95% CI. We used a random effects model for the pooling of effect sizes. Between-study heterogeneity was assessed using I 2 . Moderate heterogeneity was defined as an I 2 > 50% and high heterogeneity as an I 2 > 75% [18]. Post hoc analyses were conducted for outcomes with I 2 > 50%. This included random effects meta-regression that explored the correlations between effect sizes and the different characteristics of the study populations. Continuous variables comprised age, proportion of females, rate of hemorrhagic stroke, ratio of left hemiparesis, duration of the CVA, proportion of subluxation, and the Brunnstrom stage. If an article reported the Brunnstrom stage of both arm and hand, the mean of the two was adopted for post hoc analysis. The categorical variable was the diagnostic criteria for CRPS. The results of the meta-regression were considered to be statistically significant when p < 0.05. Funnel plots and Egger's test were used to assess publication bias, and a two tailed p < 0.1 was regarded as statistically significant [19]. Sensitivity analyses were also carried out by removing cross-sectional studies to determine their contribution to the overall effect size in the meta-analyses of primary outcomes. The statistical analyses were conducted using Comprehensive Meta-analysis Software version 3 (Biostat, Englewood, NJ, USA).

Identification of studies via databases and registers
Studies included in quantitative synthesis (n = 20) Reports excluded: Quantitative data unavailable (n = 1)  Results are given as mean (standard deviation), unless otherwise noted; CRPS: complex regional pain syndrome; SEM: standard error of mean.

Risk of Bias Assessment
Two of the three cohort studies did not demonstrate whether the outcome of interest was presented at the start of the research. One of the eight nested case-control studies did not mention recruitment of all eligible cases within a defined period of time. None of the 11 articles referenced above reported the methods used to increase comparability (Table 3). Two cross-sectional studies did not describe inclusion and exclusion criteria clearly, and another two did not disclose the study setting in detail. Furthermore, none of the 10 cross-sectional studies identified or dealt with confounding factors (Figure 2). Studies were assessed for risk of bias by the Newcastle-Ottawa Scale for cohort or case control studies. *: The criteria were met; -: The criteria were not met. a 1: Representativeness of the exposed cohort; 2: Selection of the non-exposed cohort; 3: Ascertainment of exposure; 4: Demonstration that outcome of interest was not present at start of study; 5: Comparability of cohorts on the basis of the design or analysis; 6: Comparability of cohorts on the basis of the design or analysis, controls for any additional factor; 7: Assessment of outcome; 8: Was follow-up long enough for outcomes to occur; 9: Adequate of follow-up of cohorts. b 1: Is the case definition adequate; 2: Representativeness of the cases; 3: Selection of controls; 4: Definition of controls; 5: Comparability of cases and controls on the basis of the design or analysis; 6: Comparability, controls for any additional factor; 7: Ascertainment of exposure; 8: Same method of ascertainment for cases and controls, 9: Non response rate.
3). Two cross-sectional studies did not describe inclusion and exclusion criteria clearly, and another two did not disclose the study setting in detail. Furthermore, none of the 10 cross-sectional studies identified or dealt with confounding factors (Figure 2). Studies were assessed for risk of bias by the Newcastle-Ottawa Scale for cohort or case control studies. *: The criteria were met; -: The criteria were not met. a 1: Representativeness of the exposed cohort; 2: Selection of the non-exposed cohort; 3: Ascertainment of exposure; 4: Demonstration that outcome of interest was not present at start of study; 5: Comparability of cohorts on the basis of the design or analysis; 6: Comparability of cohorts on the basis of the design or analysis, controls for any additional factor; 7: Assessment of outcome; 8: Was follow-up long enough for outcomes to occur; 9: Adequate of follow-up of cohorts. b 1: Is the case definition adequate; 2: Representativeness of the cases; 3: Selection of controls; 4: Definition of controls; 5: Comparability of cases and controls on the basis of the design or analysis; 6: Comparability, controls for any additional factor; 7: Ascertainment of exposure; 8: Same method of ascertainment for cases and controls, 9: Non response rate.

(a)
Were the criteria for inclusion in the sample clearly defined?
Were the study subjects and the setting described in detail?
Was the exposure measured in a valid and reliable way?
Were objective, standard criteria used for measurement of the condition?
Were confounding factors identified?
Were strategies to deal with confounding factors stated?
Were the outcomes measured in a valid and reliable way?
Was appropriate statistical analysis used? 0% 25% 50% 75% 100% Low risk of bias Unclear risk of bias High risk of bias
Was the exposure measured in a valid and reliable way?
Were objective, standard criteria used for measurement of the condition?
Were confounding factors identified?
Were strategies to deal with confounding factors stated?
Were the outcomes measured in a valid and reliable way?
Was the exposure measured in a valid and reliable way?
Were objective, standard criteria used for measurement of the condition?
Were confounding factors identified?
Were strategies to deal with confounding factors stated?
Were the outcomes measured in a valid and reliable way?

Spasticity
Eight investigations assessed spasticity. Four studies [4,21,26,31] measured the severity of spasticity in patients with and without poststroke CRPS, while the other four [8,20,35,36] compared the proportion of participants with signs of spasticity between groups. Five of the eight articles [4,20,26,31,36] reached statistical significance. Three studies [4,21,26] reporting the severity of spasticity were eligible for the meta-analysis, where the results indicated more severe spasticity in patients with poststroke CRPS with low heterogeneity (SMD, 0.488, 95% CI, 0.111 to 0.866, I 2 = 33.1%, Figure 4). The funnel plot and Egger's test did not reveal publication bias (p = 0.18). Sensitivity analysis was not performed, since only one article was left after excluding cross-sectional studies [26]. Another three investigations [8,20,35] reporting the proportion of patients with spasticity were deemed suitable for the meta-analysis. The results demonstrated a higher risk of poststroke CRPS with low heterogeneity in patients with spasticity (OR, 1.505, 95% CI, 1.073 to 2.111, I 2 = 0.0%, Figure 5). No publication bias was observed based on the funnel plot and Egger's test (p = 0.94). None of the papers included were cross-sectional studies, so sensitivity analysis was not conducted.

Spasticity
Eight investigations assessed spasticity. Four studies [4,21,26,31] measured the severity of spasticity in patients with and without poststroke CRPS, while the other four [8,20,35,36] compared the proportion of participants with signs of spasticity between groups. Five of the eight articles [4,20,26,31,36] reached statistical significance. Three studies [4,21,26] reporting the severity of spasticity were eligible for the meta-analysis, where the results indicated more severe spasticity in patients with poststroke CRPS with low heterogeneity (SMD, 0.488, 95% CI, 0.111 to 0.866, I 2 = 33.1%, Figure 4). The funnel plot and Egger's test did not reveal publication bias (p = 0.18). Sensitivity analysis was not performed, since only one article was left after excluding cross-sectional studies [26]. Another three investigations [8,20,35] reporting the proportion of patients with spasticity were deemed suitable for the meta-analysis. The results demonstrated a higher risk of poststroke CRPS with low heterogeneity in patients with spasticity (OR, 1.505, 95% CI, 1.073 to 2.111, I2 = 0.0%, Figure 5). No publication bias was observed based on the funnel plot and Egger's test (p = 0.94). None of the papers included were cross-sectional studies, so sensitivity analysis was not conducted.

Brunnstrom Stage
Eight articles mentioned the Brunnstrom stage. Three of the eight [20,22,30] reported the Brunnstrom stage for the arm and hand separately, and one analyzed hand only [21].

Spasticity
Eight investigations assessed spasticity. Four studies [4,21,26,31] measured the severity of spasticity in patients with and without poststroke CRPS, while the other four [8,20,35,36] compared the proportion of participants with signs of spasticity between groups. Five of the eight articles [4,20,26,31,36] reached statistical significance. Three studies [4,21,26] reporting the severity of spasticity were eligible for the meta-analysis, where the results indicated more severe spasticity in patients with poststroke CRPS with low heterogeneity (SMD, 0.488, 95% CI, 0.111 to 0.866, I 2 = 33.1%, Figure 4). The funnel plot and Egger's test did not reveal publication bias (p = 0.18). Sensitivity analysis was not performed, since only one article was left after excluding cross-sectional studies [26]. Another three investigations [8,20,35] reporting the proportion of patients with spasticity were deemed suitable for the meta-analysis. The results demonstrated a higher risk of poststroke CRPS with low heterogeneity in patients with spasticity (OR, 1.505, 95% CI, 1.073 to 2.111, I2 = 0.0%, Figure 5). No publication bias was observed based on the funnel plot and Egger's test (p = 0.94). None of the papers included were cross-sectional studies, so sensitivity analysis was not conducted.

Brunnstrom Stage
Eight articles mentioned the Brunnstrom stage. Three of the eight [20,22,30] reported the Brunnstrom stage for the arm and hand separately, and one analyzed hand only [21]. No spasticity Spasticity Figure 5. Forest plot of odds ratio in participants with spasticity.

Shoulder Pain
Three investigations [4,22,38] reported the proportion of individuals with shoulder pain, and one study described more shoulder pain in patients with poststroke CRPS. The pooled effect sizes were statistically nonsignificant with low heterogeneity (OR, 3.466, 95% CI, 0.978 to 12.277, I 2 = 47.5%, Figure 9). The funnel plot and Egger's test did not indicate publication bias (p = 0.29). Sensitivity analysis for shoulder pain was not tested because all three articles were cross-sectional studies.

Shoulder Pain
Three investigations [4,22,38] reported the proportion of individuals with shoulder pain, and one study described more shoulder pain in patients with poststroke CRPS. The pooled effect sizes were statistically nonsignificant with low heterogeneity (OR, 3.466, 95% CI, 0.978 to 12.277, I 2 = 47.5%, Figure 9). The funnel plot and Egger's test did not indicate publication bias (p = 0.29). Sensitivity analysis for shoulder pain was not tested because all three articles were cross-sectional studies.

Shoulder Pain
Three investigations [4,22,38] reported the proportion of individuals with shoulder pain, and one study described more shoulder pain in patients with poststroke CRPS. The pooled effect sizes were statistically nonsignificant with low heterogeneity (OR, 3.466, 95% CI, 0.978 to 12.277, I 2 = 47.5%, Figure 9). The funnel plot and Egger's test did not indicate publication bias (p = 0.29). Sensitivity analysis for shoulder pain was not tested because all three articles were cross-sectional studies.

Duration of Stroke
Six articles [4,[20][21][22]28,33] compared the duration of the CVA between individuals with and without CRPS. One article [20] exhibited a shorter stroke duration in patients with poststroke CRPS. The meta-analysis included all six studies, and no between-group differences were found with low heterogeneity (SMD, −0.065, 95% CI, −0.267 to 0.137, I 2 = 26.3%, Figure 15). Significant publication bias was found based on the funnel plot and Egger's test (p = 0.05, File S2). Sensitivity analysis for duration of stroke was not conducted because only one [20] article was left after removing the cross-sectional studies.

Duration of Stroke
Six articles [4,[20][21][22]28,33] compared the duration of the CVA between individuals with and without CRPS. One article [20] exhibited a shorter stroke duration in patients with poststroke CRPS. The meta-analysis included all six studies, and no between-group differences were found with low heterogeneity (SMD, −0.065, 95% CI, −0.267 to 0.137, I 2 = 26.3%, Figure 15). Significant publication bias was found based on the funnel plot and Egger's test (p = 0.05, File S2). Sensitivity analysis for duration of stroke was not conducted because only one [20] article was left after removing the cross-sectional studies.
Six articles [4,[20][21][22]28,33] compared the duration of the CVA between individuals with and without CRPS. One article [20] exhibited a shorter stroke duration in patients with poststroke CRPS. The meta-analysis included all six studies, and no between-group differences were found with low heterogeneity (SMD, −0.065, 95% CI, −0.267 to 0.137, I 2 = 26.3%, Figure 15). Significant publication bias was found based on the funnel plot and Egger's test (p = 0.05, File S2). Sensitivity analysis for duration of stroke was not conducted because only one [20] article was left after removing the cross-sectional studies.

Discussion
Our systematic review and meta-analysis identified 21 analytical epidemiological studies investigating risk factors for poststroke CRPS. The results of the meta-analysis demonstrated risk factors, including shoulder subluxation, spasticity, lower Brunnstrom hand stage, and an inferior Barthel index. The post hoc analysis revealed a positive correlation between the incidence proportion with women and left hemiparesis. In addition, a positive correlation between the prevalence with shoulder subluxation, and a negative correlation between prevalence with the Brunnstrom stage were observed in the post hoc analysis. Age, side of lesion, etiology of the stroke, the Brunnstrom arm stage, the duration of a CVA, and shoulder pain were not found to be associated with poststroke CRPS.
Although the pathophysiology of poststroke CRPS remained unclear, some scholars suggested that repeated microtraumas in the shoulder joint caused chronic pain and the initiation of an abnormal sensory-sympathetic reflex arch [39]. In stroke patients, the stability of the glenohumeral joint may be affected due to paresis or palsy of the shoulder girdle muscles. Such instability may further cause injury in the shoulder joint [20]. In addition, spasticity of shoulder muscles may cause glenohumeral capsulitis and pain, and some researchers suggested that it contributes to CRPS [40]. These hypotheses correlated with the conclusions of our review. Furthermore, the occurrence of shoulder subluxation has been reported to be negatively correlated with the Brunnstrom stage [41], which may explain the findings in our work. As for Barthel index, the effect observed in our review may have been derived from the positive correlation between the index and the Brunnstrom stage [42].
Daviet et al. [31] concluded that shoulder subluxation may only play a secondary role in the development of poststroke CRPS, being a reflection of the severity of paresis. However, a later published article by Kocabas et al. [26] disagreed such statement and further concluded that subluxation can also be a factor for the development of type 1 CRPS. We believe that the relationship between subluxation and poststroke CRPS needs further investigation given that there was significance between study heterogeneity detected in our review. Moreover, our summarized odds ratio was derived from studies that were not matched for severity of paresis, meaning that we could not support nor oppose the conclusion of Daviet et al. Future studies that are matched for severity of paresis are warranted to delineate the relationship between shoulder subluxation and poststroke CRPS.
In the post hoc analysis, women and paralysis of left limbs were found to be more likely to be associated with poststroke CRPS. In a population-based study [43], female patients had a higher incidence of CRPS compared with male patients, which may explain the finding in our review, although the mechanisms are not clear so far. Furthermore, one article showed that patients with hemianopia or hemineglect were more subject to CRPS [20]. Since both hemineglect and weakness left side limbs occur more often in a right hemispheric stroke, we hypothesized that patients with right hemispheric lesions have a higher risk for traumatizing the contralateral shoulder due to neglect syndromes [36,44].
There were several limitations in this article. First, various diagnostic criteria were adopted. Variations in criteria may derive distinct risk factors [21] and may also decrease the generalizability of our results. A full assessment of contributing elements using distinct diagnostic criteria was not possible here due to low numbers of studies. In recent years, the Budapest criteria has become the mainstream diagnostic tool for CRPS, and larger cohorts based on such criteria are warranted to confirm the findings of our review. Second, we could not conduct a meta-analysis for neglect because of too few researchers, and the summarized effect of side of brain lesions in radiological findings were non-significant probably due to insufficient statistical power. To form a stronger link between left hemiparesis and poststroke CRPS, delineating these two factors of neglect and side of brain lesions is necessary in further research. Third, research that failed to achieve statistical significance may have gone unpublished, which may have caused false positives. Brunnstrom stage of the hand may possess such publication bias in our meta-analysis, and future research was necessary to delineate the influence of such bias. Fourth, only a few of the enrolled articles adopted adjustments for confounding factors. Hence, we could not conduct a sensitivity analysis to estimate the effects of confounding factors on our results. Fifth, most of the study population in our review was recruited from among hospitalized patients. Investigations targeting individuals from outpatient departments are needed to clarify the disease frequency and contributing factors in such groups. Finally, non-English publications were not enrolled in our review; nonetheless, the authors believe that it was unlikely to cause the exclusion of any major articles.
This review highlighted several risk factors of poststroke CRPS, which aids in the identification of patients who are at high risk. A previous study found a 2.17 fold of increase in the healthcare utilization cost after diagnosis of CRPS in the general population, and such increase persisted at least 8 years after diagnosis [45]. Furthermore, oral corticosteroids are currently the only anti-inflammatory drugs with level 1 evidence [2]. However, this treatment remains problematic for its adverse events, including hyperglycemia and hypertension [46], which were common comorbid diseases in stroke patients [47]. Hence, prevention methods should be applied after stroke, especially in those with high risk of developing poststroke CRPS. Calcitonin, early rehabilitation, and restriction of passive movement of the affected limb have been examined through clinical trials for prevention of poststroke CRPS [6,7,48]. Starting early rehabilitation not only for the prevention of CRPS but also shoulder subluxation in the acute phase with the aid of adequate nursing care is essential [49]. Rehabilitation centers may utilize the findings in our study to increase the outcomes of poststroke patients.

Conclusions
This article of meta-analysis and meta-regression revealed that being female, left hemiparesis, shoulder subluxation, spasticity, a lower Brunnstrom hand stage, and inferior Barthel index are risk factors for development of poststroke CRPS. Larger studies with greater power may confirm our findings and clarify some other unknown risk factors for poststroke CRPS.
Supplementary Materials: The following are available online at https://www.mdpi.com/article/10 .3390/medicina57111232/s1, File S1: Search strategies for meta-analysis, File S2. Funnel plots for publication bias and bubble plots of meta-regression, File S3. Details of articles excluded after full text retrieved.