Introduction

Stroke is the fourth leading cause of death, accounting for 5.2% of all deaths in the United States1. Intravenous thrombolysis within 3 to 4.5 hours of onset of stroke improves outcomes in acute ischemic stroke, however, space-occupying, large hemispheric infarction, usually resulting from acute occlusion of the internal carotid artery (ICA) or the middle cerebral artery (MCA), represents a devastating sub-group of severe ischemic stroke2,3,4. Patients with large hemispheric infarction typically present with hemiparalysis, severe sensory deficits and aphasia when the dominant hemisphere is affected5,6. Despite optimal medical treatment such as ICP lowering therapy, controlled hyperventilation and hypothermia, malignant MCA infarction still leads to death in 70–80% of cases7,8. Moreover, survivors were always associated with long-term disability. The poor prognosis is, at least partially, attributed to space-occupying edema, which can induce secondary mechanical and ischemic neuronal injury by compression of normal brain and blood vessels and lead to herniation and death9.

Decompressive craniectomy (DC) has been shown to be effective in lowering mortality in patients with malignant MCA infarction. DC is aimed to remove part of the ipsilateral cranium to allow outward herniation of the infarcted brain tissue before compression of formerly healthy brain tissue occurs, decrease the ICP and improve cerebral perfusion pressure, thereby aiding the blood flow to the ischemic penumbra10,11,12. Recently, three randomized controlled trials (RCTs) and their pooled analyses suggested that early DC significantly reduced mortality and improved favorable outcome defined as modified Rankin scale (mRS) ≤ 413,14,15,16. However, a more recent meta-analysis following completion of HAMLET demonstrated a non-significant benefit of DC on the favourable outcome defined as mRS ≤ 313. Moreover, there are some important questions unsolved regarding DC for treatment of malignant MCA infarction, such as whether age of patients and timing of surgery influence the effects of surgery. Thus, we present a meta-analysis of the all available studies to with the goal of identifying an optimum time-point and a potential cut-off age for the surgery.

Results

Search Results and Study Characteristics

A diagram summarizing the process of study selection is shown in Figure 1. The combined search yielded 1998 citations, of which 1978 were excluded by review of titles and abstracts, because they were reviews, case reports, letters to editor, comments and duplicate studies. Further, full texts of the remaining 20 articles were reviewed and analyzed in detail. Eventually, 14 studies met our inclusion criteria, comprising of 349 patients in the surgical treatment group and 398 patients in the conservative group. Of which, eight studies (341 patients) were RCTs13,14,16,17,18,19,20,21, 4 (286 patients) were retrospective studies and 2 (120 patients) were prospective studies22,23,24,25,26,27, The main characteristics of the studies included in this meta-analysis are summarized in Table 1.

Table 1 Summary of results included in the meta-analysis
Full size table
Figure 1
figure 1

Flow chart of the literature search.

Full size image

Quality assessments for RCTs and cohort studies were summarized in the Supplementary Table S1–S2 and Figure 2. Briefly, for RCTs, randomization methods were described in 3 studies13,17,19 and allocation concealments were adequate in 2 studies14,21. For blinding, 6 studies used blind observers to assess outcome, while blinded for carers or patients were unlikely in all RCTs. In addition, 2 studies reported the follow-up rates as 10013,21%, which were not mentioned in other studies. Finally, none study had selective outcome reporting. (Figure 2) For cohort studies, the results of quality assessment showed that all six studies had a moderate risk of bias and reached 6–7 out of 9 points. (Supplementary Table S1 and 2)

Figure 2
figure 2

Risk of bias assessment for randomized controlled trials.

‘+’: low risk of bias, ‘−’: high risk of bias and ‘?’: Indicates unclear risk of bias.

Full size image

Early Decompressive Craniectomy

Effect on Neurological outcome

The proportion of patients with poor neurological outcome (defined as mRS>3) was reported for 3-month follow-up in 3 studies, 6 months follow-up in 8 studies, 12 months follow-up in 6 studies and 36 months follow up in 1 study. The results indicated decompressive surgery significantly decreased number of patients with a poor outcome for 3 months follow up (OR = 0.1, 95%CI = 0.02, 0.48, p = 0.004), 6 months follow-up (OR = 0.34, 95%CI = 0.20, 0.59, p = 0.0001) and 12 months follow-up (OR = 0.38, 95%CI = 0.20, 0.73, p = 0.004), but not for 36 months follow-up (OR = 0.91, 95%CI = 0.20, 4.09). (Figure 3 and Table 2)

Table 2 ARR and OR estimates with the corresponding 95% CI for unfavourable outcome in early and later DC
Full size table
Figure 3
figure 3

Forest plot with OR estimating with the corresponding 95% CI for unfavourable outcome (defined as mRS>3) associated with early DC versus medical treatment for individual trials and the pooled population at 3 months, 6 months, 12 months and 36 months follow-up (patients in all ages) CI, confidence interval; DC: decompressive craniectomy; OR, odds ratio; mRS: modified Rankin scale.

Full size image

In the subgroup analysis stratified by age of patients, the number of patients with poor functional outcome was significantly decreased in early DC group for 6 months follow-up (OR = 0.28, 95%CI = 0.10, 0.80, p = 0.02 in younger patients and OR = 0.28, 95%CI = 0.10, 0.78, p = 0.02 in older patients). (Figure 4A and Table 2) Moreover, no significant between-study heterogeneity was detected in either subgroup or overall analysis (ph>0.2 for all comparisons).

Figure 4
figure 4

Forest plot with OR estimating with the corresponding 95% CI for (A) unfavourable outcome (defined as mRS>3) or (B) the proportion of survivors with moderately severe or severe disability (defined as mRS = 4 or 5) associated with early DC versus medical treatment for individual trials and the subgroup population stratified by age at 6 months follow-up.

CI, confidence interval; DC: decompressive craniectomy; OR, odds ratio; mRS: modified Rankin scale.

Full size image

Effect on Mortality

The results of pooled analysis demonstrated that early DC significantly reduced mortality (OR = 0.14, 95%CI = 0.08, 0.25, p<0.0001). (Supplementary Figure S1) Similar results were observed in younger patients (OR = 0.14, 95%CI = 0.06, 0.35, p<0.0001) and in older patients (OR = 0.20, 95%CI = 0.11, 0.33, p<0.0001). (Supplementary Figure S2) Heterogeneity was not observed in all comparisons except for in one comparison (ph = 0.07), in which a random-effects model was used.

Survivors with moderately severe or severe disability

Meta-analysis of all 13 studies suggested there was no significant difference between early DC and best medical treatment in terms of proportion of survivors with moderately severe or severe disability for 3 months (OR = 0.20, 95%CI = 0.04, 1.03, p = 0.05), 6 months (OR = 0.65, 95%CI = 0.34, 1.24, p = 0.19) and 12 months (OR = 1.67, 95%CI = 0.74, 3.79, p = 0.22) follow-up. (Figure 5 and Table 2) Similarly, in the subgroup analysis, early DC did not increase survivors with moderately severe or severe disability in younger patients (OR = 0.68, 95%CI = 0.21, 2.23, p = 0.53) or in older patients (OR = 0.53, 95%CI = 0.17, 1.65). (Figure 4B and Table 2)

Figure 5
figure 5

Forest plot with OR estimating with the corresponding 95% CI for the proportion of survivors with moderately severe or severe disability (defined as mRS = 4 or 5) associated with early DC versus medical treatment for individual trials and the pooled population at 3 months, 6 months, 12 months and 36 months follow-up (patients in all ages) CI, confidence interval; DC: decompressive craniectomy; OR, odds ratio; mRS: modified Rankin scale.

Full size image

Later Decompressive Craniectomy

A total of 4 studies involved 97 patients were eligible for the analysis of the efficacy of later DC on functional outcome in patients with malignant MCA infarction, which indicated DC performed after 48 h of stroke onset was not associated with the improvement of functional outcome or reduction of mortality ((OR = 0.16, 95%CI = 0.02, 1.11, p = 0.06 and OR = 0.43, 95%CI = 0.09, 1.94, p = 0.27, respectively).

Assessment of Publication Bias

Publication bias was assessed by funnel plots and Egger's test. The shape of funnel plot did not reveal evidence of obvious asymmetry. (Figure 6) Then, the Egger's test was used to provide statistical evidence of funnel plot symmetry, which did not show any evidence of publication bias (p>0.12 for all comparisons), indicating that our results are statistically robust.

Figure 6
figure 6

Funnel plot to detect publication bias.

No significant funnel asymmetry was observed which could indicate publication bias. (P value for Egger test was 0.38) logor Natural logarithm of the OR, s.e. of logor standard error of the logOR.

Full size image

Discussion

The present meta-analysis demonstrated that early decompressive surgery (within 48 hours after the onset of stroke) significantly increased the chance of a favorable functional outcome in patients with malignant MCA infarction, whereas later decompressive surgery might not reduce mortality or poor functional outcome. Moreover, early DC did not increase the rate of patient survival with moderately severe or severe disability.

This study included a larger number of RCTs than two previous meta-analyses, which indicated early DC saved lives and increased good functional outcome (defined as mRS ≤ 4) in patients with malignant MCA infarction, but had a non-significant improvement in the percentage of survivors with good outcomes defined as mRS ≤ 313,15. A mRS of 4 implied that the patient was unable to walk or attend to their own body need without assistance, thus, which was usually identified as an unfavourable outcome13,14,28. Moreover, in a recent study, a mRS of 4 was considered acceptable by less than half of the physicians (38.0%), whereas a mRS of 3 was considered acceptable by the majority (79.3%)29. Thus, we defined poor neurological outcome as mRS>3 and favourable outcome as mRS ≤ 3 and the results showed that DC performed within 48 h after stroke onset significantly reduced mortality and poor neurological outcome at 3 months, 6 months and 12 months follow-up.

We further investigated the association between age and functional outcome in patients underwent decompressive surgery, which was inconclusive in previous studies30,31,32. In the subgroup analysis stratified by age, the results showed reduced mortality with early DC compared with medical management (OR = 0.20, 95%CI = 0.11, 0.33, p<0.0001) in patients >60 years of age. Moreover, there was a significant reduction of unfavorable clinical outcome (mRS>3) in older patients (OR = 0.20, 95%CI = 0.10, 0.78, p = 0.02), suggesting early DC had a benefit effect on functional outcome in older patients with malignant infarction.

Recent observational studies have strongly suggested that age is a main predictor of poor functional outcome after decompressive surgery33. In our meta-analysis, the proportion of older patients who had poor functional outcome (88.3% ± 6.1%) was significant higher than that of younger patients (66.8% ± 11.3%, p = 0.016). These results were consistent with a recent review, in which just 6 (8%) of 72 older patients (>60 years of age) had a favorable outcome after surgery34. Together, these results suggested that early DC saved lives and improved neurological outcome in patients >60 years of age. In addition, higher age was a powerful predictor of poor outcome in patients with DC.

The time-point for decompressive surgery is another clinical factor that is associated with the efficacy of decompressive surgery on patients with malignant MCA infarction5,29,35. The present pooled-analysis, including 97 patients who underwent DC after 48 h of stroke onset, showed that later DC did not decrease unfavourable outcome and mortality in patients with malignant infarction. (OR = 0.16, 95%CI = 0.22, 1.11, p = 0.06 and OR = 0.43, 95%CI = 0.09, 1.94, p = 0.27, respectively) Generally, patients with malignant infarction deteriorate after 48 h from edema formation and death usually occurs in most patients within 72 h to 96 h36. By that time, cerebral edema would be getting close to its normal peak, thus, DC seems to lose its superiority over medical treatment. However, the lack of perceived benefit in the later DC group might not be very conclusive owning to relatively small number of patients in previous studies.

In addition, recent studies indicated DC for malignant MCA infarction thus resulted in a significant reduction in mortality, but nearly all survivors suffer moderately severe or severe disability (defined as mRS = 4 or 5). Whether there was a trend towards an increase in the number of patients with moderately severe or severe disability undergoing DC is unclear15,18. Our meta-analysis indicated that, the proportion of patients surviving with moderately severe or severe disability was not increased in DC group compared with conservative treatment group (OR = 0.68, 95%CI = 0.21, 2.23, p = 0.53 in younger patients and OR = 0.53, 95%CI = 0.17, 1.65, p = 0.17 in older patients).

Several limitations of our study should be considered. First, the mRS score mainly reflects motor abilities and dependency and may have a neglect for neuropsychological functions and quality of life. Second, for the analysis in older patients, 2 RCTs and 1 retrospective study including patients >60 years and the other retrospective study used ≥70 years as their age. Thus, this might introduce some bias due to the possibility that some of patients between 60 and 70 years were not included in this meta-analysis. Third, in the subgroup analysis by age, 40% percents older patients were from non-RCTs, which might have a clear selection bias toward healthy or active older patients for performing the surgical procedure. Also, findings of our study could be limited by the inclusion of published data only, such as, DEMITUR, a RCT involving 151 patients with malignant infarction of MCA carried from January 2003 to December 2007, has not published yet37. Finally, a language bias might have been introduced because only studies in English were included.

In conclusion, despite these limitations, our study demonstrated that early DC was effective in lowering mortality and improving functional outcome in patients with malignant MCA infarction. Although higher age was an important predictor of unfavorable outcome in patients underwent decompressive surgery, early DC did increase the probability of survival and reduce poor functional outcome in patients >60 years of age. Finally, large, multicenter RCTs comparing the efficacy of decompressive surgery and conservative treatment are required, especially in older patients.

Methods

Search Strategy

A comprehensive electronic search in PubMed, Medline, Embase, Cochrane library, Web of Science database was carried out using the following search terms: “hemicraniectomy” or “craniectomy” or “decompressive surgery” or “decompressive craniectomy” or “decompression”, “middle cerebral artery” or “internal carotid artery”, AND “stroke” or “infarction” or “infarct” (the last search update was 5 June 2014). In addition, the references of all retrieved articles were checked for additional potential studies.

Study Eligibility

The inclusion criteria were the following: 1) studies comparing effects of DC and medical treatment alone as control on patients with MCA infarction; 2) assessing outcome as death and functional outcome defined by mRS or Glasgow outcome scale (GOS) score (if mRS score was unavailable) at 3 months, 6 months, 12 months or 36 months follow-up. Exclusion criteria for our primary analysis were as follows: 1) unavailability of a medical treatment comparison group 2) unavailability of numbers of patients survival or with functional outcome at 3 months, 6 months, 12 months or 36 months follow-up 3) review articles, meta-analysis and guidelines

Data Extraction

The following data was extracted independently from each study by two authors using a standardized data extraction form: study design, patient eligibility criteria, duration of follow-up, sex and age of patients, National Institutes of Health stroke scale (NIHSS), vascular territories and side of the infarction, presence of preoperative clinical signs of herniation, time to surgery, mRS scores, GOS scores and mortality. Disagreements were resolved by consulting with a third author.

Quality Assessment

Quality assessment for included studies was assessed by 2 independent authors. Briefly, Cochrane Collaboration's tool were used for assessing quality according to the following domains: selection bias (random sequence generation and allocation concealment), attrition bias (incomplete outcome data), performance and detection bias (blinding of participants, personnel and outcome assessment), reporting bias (selective reporting) and other bias (other sources of bias)38. In addition, we used Newcastle–Ottawa Scale (NOS) to assess the quality in the non-randomized cohort studies39.

Outcome

We assessed the following outcomes to explore the effect of DC on malignant MCA infarction: 1) death at 3 months, 6 months, 12 months and 36 months follow-up, 2) a poor functional outcome (defined as mRS = 4 to 6 or GOS = 1 to 3) at 3 months, 6 months, 12 months and 36 months follow-up 3) survival with moderately severe or severe disability defined as mRS 4 and 5 at 3 months, 6 months, 12 months and 36 months follow-up. Pooled analyses of the association between DC and outcome adjusted by other variables were not carried out due to the unavailable data for individual subjects.

Statistical Analysis

Absolute risk reductions (ARRs), odds ratios (ORs) and 95% confidence intervals (CIs) were calculated for the specified outcome15. The significance of the pooled OR and ARR was determined by the Z-test, a P-value less than 0.05 was considered significant. The heterogeneity between studies was assessed by Chisquare based Q test and I2 test40. Heterogeneity was considered significant when P<0.10 and pooled estimates were calculated using the random-effects (DerSimonian-Laird) model, otherwise, a fixed-effects (Mantel-Haenszel) model was used41. Publication bias was investigated using visual evaluation of funnel plots and Egger regression asymmetry test42. All statistical analyses were performed by Review Manager (RevMan) (Version 5.2) or STATA software (version 12).