Search results
368 studies were extracted from the PubMed, Embase, Cochrane, and Google Scholar databases initially. After review of the abstract and title, 310 studies were excluded for the following reasons: duplicates; unrelated studies; studies of corpses or animals; case reports; and review articles. Then, the full texts of the remaining 58 studies were reviewed. A further 47 studies were excluded for the following reasons: single-arm studies (n = 13); less than four segments fused (n = 3); insufficient data (n = 20); and pooled results without segregating termination site L5 or the sacrum (n = 11). Hence, a total of 11 studies were finally selected into the meta-analysis [10, 14,15,16,17,18, 23, 28,29,30,31]. The flow diagram is shown in Fig. 1.
Study characteristics
All of the 11 studies were non-randomized retrospective studies. The basic characteristics of included studies are shown in Table 1. There were no significant differences on age, gender, diagnosis, and follow-up between the two groups. Because the preoperative scoliosis angles, lordosis angles, SVA, ODI, and VAS scores were significantly different in some studies [14, 16, 17, 29], the change values from pre-operation to the last follow-up were used for analysis to eliminate the influence of different preoperative values.
Table 1 Basic characteristics of included studies in this meta-analysis Quality assessment and publication bias assessment
Two investigators assessed the quality of studies independently. All of the 11 included studies scored more than 7 points; hence, the studies were of relatively high quality. The quality assessment is summarized in Table 2. Evidence of publication bias was not indicated in the symmetrical funnel plot on visual inspection (Online Resource 2).
Table 2 Quality assessment according to the Newcastle–Ottawa scale Final correction of scoliosis
We used the change values from pre-operation to the last follow-up for analysis. The data of main scoliosis angles were extracted from four articles [14, 16, 17, 29] (Fig. 2). There was no significant difference between L group and S group in analysis (MD 0.50; 95% CI − 1.44 to 2.44, P = 0.61) with moderate heterogeneity (P = 0.14, I2 = 45%).
Final correction of lordosis
Data on the preoperative and final lordosis were available from four studies [14, 16, 17, 29] (Fig. 3). There was a significant difference between L group and S group in analysis (MD − 4.12; 95% CI − 7.85 to − 0.38, P = 0.03) with significant heterogeneity (P = 0.03, I2 = 66%).Hence, the random effects model was used. Sensitivity analysis was conducted to explain the source of heterogeneity. When omitting the study Koller 2016 [14], the I2 index fell to 0%. The reason might lie in their adequate correction and little loss of lumbar lordosis with the use of appropriate surgical approach in L group.
Final correction of SVA
The SVA was calculated as the distance from the C7 plumb line to the posterior endplate corner of S1. Data on the preoperative and final SVA were available from five studies [14, 16,17,18, 29] (Fig. 4). There was a significant difference between L group and S group in analysis (MD − 1.15; 95% CI − 1.80 to − 0.50, P = 0.0005) with moderate heterogeneity (P = 0.17, I2 = 38%).
Overall complication rate and revision rate
The records of complications were different in forms—some studies listed all complications, whereas some provided the overall complication rate. Data on overall postoperative complications were retrieved from seven studies totalling 527 patients [16,17,18, 23, 28,29,30] (Fig. 5). There was no significant difference between L group and S group in analysis (OR, 0.89; 95% CI 0.60 to 1.30, P = 0.54) with moderate heterogeneity (P = 0.12, I2 = 41%). As for revision rate, data of revisions were available from seven studies totalling 673 patients [14,15,16,17,18, 28, 30] (Fig. 6). There was also no significant difference between two groups (OR, 1.08; 95% CI 0.59 to 1.99, P = 0.80) with significant heterogeneity (P = 0.06, I2 = 51%). When omitting the study Complex Spine Study Group (CSSG) 2011 [30], the I2 index dropped to 14%. The relatively low incidence of pseudoarthrosis and revision rate in this study’s S group might lead to the clinical heterogeneity.
Rate of proximal/distal adjacent segment disease
The evidence of adjacent segment disease is based on clinical symptoms instead of radiographic assessment, which included symptomatic junctional degeneration, kyphosis, and failure. Data on proximal adjacent segment disease were available from seven studies [15,16,17,18, 23, 29, 31] (Fig. 7). There was a significant difference between L group and S group in analysis (OR, 0.57; 95% CI 0.35 to 0.92, P = 0.02) without heterogeneity (P = 0.81, I2 = 0%). Data on distal adjacent segment disease were available from eight studies [14,15,16,17,18, 23, 28, 29] (Fig. 8). There was a significant difference between L group and S group in analysis (RD, 0.15; 95% CI 0.10 to 0.19, P < 0.00001) with moderate heterogeneity (P = 0.12, I2 = 39%). In other words, since no distal junctional disease would occur in S group, the incidence of advanced degeneration of L5-S1 disc with clinical sequelae in L group was 15% (10 to 19%).
Rate of pseudarthrosis and implant-related complications
Data on pseudarthrosis and implant-related complications occurred during follow-up were available from nine studies totalling 887 patients [10, 14,15,16,17,18, 23, 28, 29] (Fig. 9). There was a significant difference between L group and S group in analysis (OR, 0.43; 95% CI 0.29 to 0.64, P < 0.0001) without heterogeneity (P = 0.74, I2 = 0%).
Final improvement of ODI and VAS scores
We used the change values from pre-operation to the last follow-up for analysis. Data on the preoperative and final ODI were available from three studies [16, 17, 29] (Fig. 10). There was no significant difference between L group and S group in analysis (MD 0.94; 95% CI − 1.19 to 3.07, P = 0.39) with moderate heterogeneity (P = 0.18, I2 = 41%).Data on the preoperative and final VAS scores were available from two studies [17, 29] (Fig. 11). We introduced two subgroups according to the painful place (back and leg). There was no significant difference between L group and S group in subgroup analysis, respectively (MD = − 0.22, 95% CI − 1.29 to 0.86, P = 0.69; MD = − 0.46, 95% CI − 1.63 to 0.72, P = 0.45). Heterogeneity existed within both subgroups (P = 0.07, I2 = 70%; P = 0.11, I2 = 61%); hence, the random effects model was conducted.
Subgroup meta-analysis
Although we required that the fusion length must be greater than four in the inclusion criteria, the fusion segments were significantly different in some studies. We made the subgroup meta-analysis to eliminate the influence of different fusion lengths. There were two groups as subgroups, 4 ≤ fusion segments ≤ 8 group and fusion segments > 8 group. Eight were chosen because fusion of so many segments could just cross the thoracolumbar junction to the thoracic spine.
For the final correction of SVA, there was significant difference in each subgroup (MD = − 0.74, 95% CI − 1.47 to − 0.01, P = 0.05; MD = − 2.70, 95% CI − 5.33 to − 0.07, P = 0.04), which was consistent with the previous meta-analysis. There was no significant difference between subgroups (P = 0.16) (Online Resource 3).
For the overall complication rate, there was no significant difference in each subgroup (OR = 1.16, 95% CI 0.60 to 2.23, P = 0.67; OR = 0.52, 95% CI 0.20 to 1.36, P = 0.19), which was also consistent with the previous meta-analysis. There was no significant difference between subgroups (P = 0.18) (Online Resource 4). For the revision rate, there was no significant difference in each subgroup (OR = 1.08, 95% CI 0.69 to 1.70, P = 0.74; OR = 0.40, 95% CI 0.12 to 1.39, P = 0.15), which identified with the previous meta-analysis. There was no significant difference between subgroups (P = 0.14) (Online Resource 5).
For the rate of proximal adjacent segment disease, there was no significant difference in each subgroup (OR = 0.68, 95% CI 0.26 to 1.81, P = 0.44; OR = 0.65, 95% CI 0.35 to 1.21, P = 0.17), which was still consistent with the previous meta-analysis. There was no significant difference between subgroups (P = 0.94) (Online Resource 6). Similarly, for the rate of distal adjacent segment disease, there was significant difference in each subgroup (RD = 0.17, 95% CI 0.11 to 0.24, P < 0.00001; RD = 0.15, 95% CI 0.04 to 0.26, P = 0.007), which was consistent with the previous meta-analysis. There was no significant difference between subgroups (P = 0.70) (Online Resource 7).
For the rate of pseudarthrosis and implant-related complications, there was significant difference in each subgroup (OR = 0.53, 95% CI 0.29 to 0.97, P = 0.04; OR = 0.33, 95% CI 0.19 to 0.60, P = 0.0002), which was still in line with the previous meta-analysis. There was no significant difference between subgroups (P = 0.27) (Online Resource 8). Other variables could not be divided into two subgroups.