A meta-analysis indicating extra-short implants (≤ 6 mm) as an alternative to longer implants (≥ 8 mm) with bone augmentation

Extra-short implants, of which clinical outcomes remain controversial, are becoming a potential option rather than long implants with bone augmentation in atrophic partially or totally edentulous jaws. The aim of this study was to compare the clinical outcomes and complications between extra-short implants (≤ 6 mm) and longer implants (≥ 8 mm), with and without bone augmentation procedures. Electronic (via PubMed, Web of Science, EMBASE, Cochrane Library) and manual searches were performed for articles published prior to November 2020. Only randomized controlled trials (RCTs) comparing extra-short implants and longer implants in the same study reporting survival rate with an observation period at least 1 year were selected. Data extraction and methodological quality (AMSTAR-2) was assessed by 2 authors independently. A quantitative meta-analysis was performed to compare the survival rate, marginal bone loss (MBL), biological and prosthesis complication rate. Risk of bias was assessed with the Cochrane risk of bias tool 2 and the quality of evidence was determined with the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) approach. 21 RCTs were included, among which two were prior registered and 14 adhered to the CONSORT statement. No significant difference was found in the survival rate between extra-short and longer implant at 1- and 3-years follow-up (RR: 1.002, CI 0.981 to 1.024, P = 0.856 at 1 year; RR: 0.996, CI 0.968 to 1.025, P  = 0.772 at 3 years, moderate quality), while longer implants had significantly higher survival rate than extra-short implants (RR: 0.970, CI 0.944 to 0.997, P < 0.05) at 5 years. Interestingly, no significant difference was observed when bone augmentations were performed at 5 years (RR: 0.977, CI 0.945 to 1.010, P = 0.171 for reconstructed bone; RR: 0.955, CI 0.912 to 0.999, P < 0.05 for native bone). Both the MBL (from implant placement) (WMD: − 0.22, CI − 0.277 to − 0.164, P < 0.01, low quality) and biological complications rate (RR: 0.321, CI 0.243 to 0.422, P < 0.01, moderate quality) preferred extra-short implants. However, there was no significant difference in terms of MBL (from prosthesis restoration) (WMD: 0.016, CI − 0.036 to 0.068, P = 0.555, moderate quality) or prosthesis complications rate (RR: 1.308, CI 0.893 to 1.915, P = 0.168, moderate quality). The placement of extra-short implants could be an acceptable alternative to longer implants in atrophic posterior arch. Further high-quality RCTs with a long follow-up period are required to corroborate the present outcomes. Registration number The review protocol was registered with PROSPERO (CRD42020155342).


Search strategy. Two reviewers conducted electronic systematic literature searches independently through
PubMed, Web of Science, EMBASE and the Cochrane Library databases (until November 2020) using the following search terms: (a) Pubmed: ((short) OR (extra-short) OR (ultra-short)) AND ((implant) OR (implants) OR (dental implant) OR (dental implants)) AND (clinicaltrial [Filter]); (b) EMBASE: 'short implants':ti,ab,kw OR 'short dental implants':ti,ab,kw OR 'short implant':ti,ab,kw OR 'short dental implant':ti,ab,kw; (c) Web of Science: TOPIC: (short implant) OR TOPIC: (short implants)OR TOPIC: (short dental implants) ORTOPIC: (short dental implant) Refined by: Databases: (WOS) ANDDOCUMENT TYPES: (CLINICAL TRIAL); (d) the Cochrane Library databases: short implants in Title Abstract Keyword OR short dental implants in Title Abstract Keyword-in Trials (Word variations have been searched)-Source: CT.gov. Moreover, a thorough hand-searching incorporated the related journals and grey literature (from January 2016 to November 2020) supplemented by references within the retrieved articles.

PICOS (patient, intervention, comparison, outcome, study design).
According to the PICOS format, a specific answerable question was illustrated as follows: (P) Patients: Patients who received at least one extra-short dental implant (≤ 6 mm) or longer implant (≥ 8 mm) with or without bone augmentation followed for ≥ 12 months. Gender, nationality and race of patients are not restricted. (I) Intervention: One or more extra-short (≤ 6 mm) implants placed in the maxilla and/or mandible. www.nature.com/scientificreports/ (control group); (c) fixed prostheses was used as final restorations. (d) The survival rate of extra-short implant (≤ 6 mm) compared with longer implant (≥ 8 mm) were considered as the primary outcome which should be available in all the included studies. In addition, secondary outcomes comprised difference in marginal bone loss (MBL), supplemented with biological and prosthesis complication rates in this review.
Study selection and data extraction. Initially, titles and abstracts of all studies were scanned and excluded by two reviewer authors independently and in duplicate. Full-text reading were required for further information to confirm the eligibility and fulfill the predetermined data extraction form in the final stage of screening. The data from each included study, such as number of implants, patient characteristics, implant characteristics, surgical procedure, was presented in Table 1. All disagreements were resolved by discussion or consulting a third author.

Statistical analysis.
Only the studies made similar comparisons reporting the same outcomes, could a meta-analysis be conducted by software Stata version 15 (StataCorp. 2017. Stata Statistical Software: Release 15. College Station, TX: StataCorp LLC). The main effect size measure for quantitative continuous data (MBLs) was considered weighted mean difference (WMD). The quantitative binary data (implant survival rate, biological and prosthesis complication rate) were evaluated using risk ratio (RR). Inverse Variance methods and Mantel-Haenszel were used as the Weighting Methods for WMD and RR, respectively. By definition, RR < 1 indicated a lower event rate of test group, and WMD < 0 indicated a lower MBL was observed in test group. Q Cochrane test, the related P values, I 2 and the 95% confidence intervals for I 2 were used to evaluate the heterogeneity. Summary estimates of RR were calculated by random-effects models if heterogeneity was proved to be high (P < 0.05, I 2 > 50%) 32 . According to the recommendations of Higgins 33 , subgroup analysis, metaregression, sensitivity testing and exploration of publication bias were conducted to investigate the heterogeneity, and the significance was set at P < 0.05. Subgroup analyses were performed to test the effect of bone augmentation procedure. Meta-regression analyses were performed to test categorical variables such as loading method (immediate/early/conventional), and exclusion or inclusion of heavy smokers. The effect of smoking habits on the clinical outcomes was investigated through the smoking percentage ratio between short and long implant groups (S/L) and the total percentage of smokers. Additionally, one-out-removed method was performed for the sensitivity analysis. Funnel plots and Egger tests were implemented to assess the probability of publication bias.

Risk of bias and quality of evidence.
The methodological quality assessment of included articles was undertaken by two investigators based on the published full-text articles independently using the Cochrane risk of bias 2 (ROB 2) assessment tool for RCTs 34 . In case of disagreement, it was solved by a discussion with the third author. ROB 2, focusing on different aspects of trial design, conduct and reporting, is structured into a fixed set of domains of bias, which include a series of questions ('signalling questions') for elicit information about features of the trial that are relevant to risk of bias. A judgement arising from each domain, which could be 'Low' , 'High' risk of bias, or 'Some concerns' about the risk of bias, is proposed based on answers to the signalling questions.
The Grades of Recommendation, Assessment, Development and Evaluation (GRADE) tool has been used to summarise the overall quality of the evidence 35 . Issues with bias, inconsistency, imprecision, indirectness and publication bias can decrease certainty, whereas large effect, plausible confounding and dose response can increase.

Results
Study selection. Electronic searches identified a total of 4010 publications including 1757 from PubMed, 1387 from Web of Science, 570 from EMBASE, 296 from the Cochrane Central Register of Controlled Trials. Furthermore, an additional 31 articles were collected through manual screening. After removal of duplicates and screening the title and abstract, 69 publications were selected. After full-text screening, 38 articles were excluded for reasons (Fig. 1, Supplementary file 2), leaving a total of 31 articles with different follow up times for inclusion in the following statistical analysis and interpretation. While, articles which reported the outcomes of the same RCT at different follow-up would be counted as the same study. Therefore, the 31 included articles were categorized into 21 series of studies and each series reported the outcomes of one independent RCT.
Risk of bias and quality of evidence. The methodological quality assessment of 21 series of studies was undertaken by ROB 2 tool and shown in Table 2 and Supplementary Figure 3. Five studies were considered as having a high risk of bias, and judgments expressed "some concern" in seven studies, while the remaining were characterized by a low risk of bias. According to the GRADE system, pooling of studies on implants survival rate, MBL and complications rate provided low-to moderate-quality evidence in Table 3.
Characteristics of the studies. The characteristics of the 31 included articles, which were categorized into 21 series of studies, are listed in Table 1. Overall, 2576 implants have been placed compromising 1243 extra-short implants and 1333 longer implants in 1387 patients. Subgroups were performed when maxillary and mandibular implants were reported as separate analysis units in the RCTs 5,36-56 , while 4 studies combined the outcomes of implants installed in both maxilla and mandible 12,57-59 . 7 out of 21 RCTs restored the implants with single-crown prostheses 36,38,40,41,[50][51][52]54,56,60 , while 5 with splinted prostheses exclusively 43,45,47,57,59,61 and the others with either a single-crown or splinted prostheses. Eight studies used only screw-retained prostheses 5,39,41,45,47,49,54,[57][58][59][60]62 , while four studies applied only cement-retained prostheses 51 19 selected articles employed the patient as the unit of analysis, while the outcomes of implant-level were reported in the remaining studies. Different approaches to address the within-patient correlation, which is very common in oral research since multiple sites within a single patient may be inappropriately considered as independent analysis units, were adopted in few studies 53,65 . Only one implant was installed in each patient in some studies resulting in the avoidance of the problem of within-patient correlation 12,41,52,56 . Nevertheless, survival rate without adjustment for within-patient correlation were reported in a few studies, which utilized all implants while ignored the dependence among implants from the same subject 59,66,67 .

Implant survival rate.
There were 31 studies included with different follow-up years, which revealed that the individual survival rate for the reported extra-short and longer implants throughout the studies was 95.98% and 96.77%, and the overall survival rate of the implants was 96.39%. The meta-analysis revealed that the survival rate of extra-short and longer implants failed to prove a significantly statistical difference in both jaws at 1-and 3-year follow up (RR: 1.002, CI 0.981 to 1.024, P = 0.856 at 1 year; RR: 0.996, CI 0.968 to 1.025, P = 0.772 at 3 years, Fig. 2a,b). However, statistical significance was demonstrated in the survival difference between two groups at 5-years follow up (RR: 0.970, CI 0.944 to 0.997, P < 0.05) (Fig. 2c), which proved that longer implants have a higher survival rate than extra-short implants in longer follow up periods. While, no significant difference was found between two groups in the maxilla (RR: 0.987, CI 0.956 to 1.018, P = 0.399 for 1-year; RR: 0.978, CI 0.936 to 1.022, P = 0.327 for 3-year; RR:0.892, CI 0.783 to 1.015, P = 0.084 for 5-year) or mandible (RR: 1.039, CI 0.998 to 1.083, P = 0.063 for 1-year; RR: 1.026, CI 0.966 to 1.091, P = 0.404 for 3-year; RR: 0.918, CI 0.824 to 1.023, P = 0.122 for 5-year) at different defined follow up, respectively ( Fig. 3a-c). Furthermore, the arch had a significant impact on the risk ratio difference throughout different follow up periods (P < 0.05), while no influence was found when the risk ratio difference of defined follow up (1-, 3-, 5-year independently) was evaluated (P = 0.062, P = 0.310, P = 0.897, respectively). Finally, subgroup analysis (of only the maxilla/mandible independently) in the eight articles 12,40,57-59,61-63,68 was impractical to conduct since the combination data of both jaws.
For further analysis, the influence of augmentation procedure was evaluated. The RRs for reconstructed bone up to 1-year, 3-years and 5-years follow-up were 1.010 (CI 0.978 to 1.044, P = 0.542), 0.997 (CI 0.964 to 1.031, P = 0.861), 0.977 (CI 0.945 to 1.010, P = 0.171), respectively ( Fig. 4a-c). And the RRs for native bone up to 1-year, 3-years and 5-years follow-up were 0.989 (CI 0.969 to 1.009, P = 0.270), 0.992 (CI 0.938 to 1.049, P = 0.786), 0.955 (CI 0.912 to 0.999, P < 0.05), respectively ( Fig. 4a-c). The subgroup analysis displayed that the survival differences between two groups did not vary significantly when an augmentation procedure was performed or not, while the survival rate of longer implants was higher than that of extra-short implants in native bone after 5-years measurement. Moreover, augment procedure did not impact the results in different defined follow up periods (P = 0.228 for 1-year, P = 0.933 for 3-year, P = 0.436 for 5-year, respectively).
In investigations with mandibular implants in reconstructed bone 46,48,49,54 , the longer implants showed higher survival rate (RR: 1.058, CI 1.002 to 1.117, P < 0.05) than extra-short implants at 1-year follow-up. However, the survival rate of extra-short and longer implants was not significantly different when bone augmentation was  www.nature.com/scientificreports/ augmentation procedure at 3-years follow up was not attainable in either maxilla or mandible, since all the studies which reported the survival rate of upper or lower jaw independently underwent augment procedures. In addition, the RRs of implant survival rate at patient level were calculated by fixed-effects model since the heterogeneity was proved to be low (P > 0.05, I 2 = 0%). The RR for overall survival rate between two groups was 0.975 (CI 0.946 to 1.005, P = 0.101). Similar with the results of implant level, no significant difference was found at different follow up (RR = 0.979, CI 0.945 to 1.015, P = 0.25 at 1-year; RR = 0.995, CI 0.941 to 1.053, P = 0.872 at 3-year; RR = 0.956, CI 0.899 to 1.016, P = 0.145 at 5-year) (Supplementary Figure 1).
Meta-regression analyses of the RRs for survival rate were performed and results showed that categorical moderators, such as short/long smoking ratio, total smoking percentage, loading time, test of initial stability, were not in significant association with the survival rate differences between two group (P > 0.05) at 1-, 3-or 5-year recalls.
The sensitivity analyses for all studies in implant and patient level were performed respectively, as well as the studies with correct statistical analyses (Fig. 6a-c). Figure 6a showed that the exclusion of Bernardi et al. 54 Table 3. Grades of recommendation, assessment, development and evaluation approach summarizing the evidence. Question: Extra-short implants (≤ 6 mm) compared to longer implants (≥ 8 mm) for partially or totally edentulous patients. GRADE Working Group grades of evidence. High certainty: We are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect. Very low certainty: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect. Generated by GRADEpro GDT web application, http:// grade pro. org. CI: confidence interval; RR: risk ratio; MD: mean difference. a Small simple size (less than OIS), CI of RR included 1. b Heterogeneity across the studies, I 2 = 58.10%, CI for I 2 = 30.3% to 74.8%. c Heterogeneity across the studies, I 2 = 48.8%, CI for I 2 = 5.1% to 72.39%. d Small simple size (less than OIS), CI of RR included 1.   www.nature.com/scientificreports/ seemed to result in a relatively different meta-analytic estimate. Nevertheless, this difference was insignificant. Accordingly, the combination of investigations was not influenced by a particular one. Similarly, the combination was not influenced by a particular one (Fig. 6b,c).

Marginal bone loss.
Peri-implant marginal bone loss were measured from different baseline between studies. Only studies reporting the outcomes of MBL in patient-level would be analyzed in the present review. 16 articles utilized the baseline measured at the time of implant placement 12,[37][38][39][40][42][43][44][46][47][48][49]52,61,62,69 , two of which also reported the bone loss from the baseline at prosthetic loading 36,38 . In contrast, nine articles 41,45,53,56,[58][59][60]66,70 considered the time of prosthetic restoration as baseline exclusively. The data of marginal bone loss was absent in one included study 54 . Therefore, the between-group comparison of marginal bone loss was performed depending on these two different baseline criteria.  (Fig. 8b). A significant effect of augmentation procedure on the mean difference of maxillary MBL at 1 year was found (coefficient: 0.324, P < 0.05). The influence of bone augmentation at 3 and 5 years in both jaws was impossible to be analyzed since all the included studies performed augmentation.
The comparison between extra-short and standard length implants following immediate loading protocol was reported in two included studies at 1-year follow up 47,52 . The loading method (convention/immediate) was demonstrated to significantly relate to the mean difference of maxillary MBL at 1 year through meta-regression analysis (coefficient: 0.324, P < 0.01); however, non-significant correlation was found in its mandibular counterpart (P = 0.437). Subgroup analyses for the effects of loading method (immediate/conventional) on MBL from IP at 1-year follow up showed significant difference between two groups on maxilla (WMD: − 0.470, CI − 0.534 to − 4.06, P < 0.001 for immediate loading; WMD: − 0.145, CI − 0.202 to − 0.088, P < 0.001 for conventional loading). Other categorical moderators, such as short/long group smoking ratio, total smoking percentage, test of initial stability, inclusion of heavy smokers, had non-significant association with the MBL differences between groups (P > 0.05) at different defined follow up recalls. Interestingly, the total smoking percentage had a positive correlation to MBL difference between two groups at 5-year follow up (coefficient: 1.646, P = 0.16) (Fig. 13a), despite the P value with no statistical significance. It indicates that higher the number of smokers the higher the tendency of greater MBL occurring in extra-short compared to longer implants. Random-effect model and subgroup analysis were performed as above since the heterogeneity in WMD of MBLs at 1-year was high (I 2 = 63.5%, CI 30.16% to 80.93%, P < 0.05). In addition, the one-out remove method with metaninf module was conducted for the sensitivity analysis, and the exclusion of Cannizzaro-2015 47 study seemed to result in a relatively different meta-analytic estimate. Nevertheless, this difference was insignificant. Accordingly, the combination of investigations was not influenced by a particular one (Fig. 6d).
Meanwhile, in spite of no statistical significance, the RR > 1 in native bone, while RR < 1 in reconstructed bone. It implies that the prosthesis complication rate tends to be higher in test group in native bone, and reconstructed bone subgroup had inverse pattern, preferring test group. Meta-regression was performed and no covariate was found statistically associated with RR for prothesis complication rate between two groups throughout different follow up periods (P > 0.5). Nevertheless, despite no statistically significant (P = 0.192), a positive correlation between total smoking percentage and prothesis complications in all studies was found (coefficient = 3.628, Fig. 13b). Hence, it was indicated that the more smokers were included the higher the trend of prothesis complications appearing in test group.

Discussion
To the best of our knowledge, this is the first systematic review comparing the survival rate, marginal bone loss, biological and prosthesis complication rate between extra-short implants (≤ 6 mm) and longer implants (≥ 8 mm) at both jaws, maxilla or mandible independently, with and without bone augmentation procedures.

Survival rate.
The survival rate of extra-short implants was found comparable to the longer implants at 1and 3-years follow-up in the present review, while significantly higher survival rate was found in longer group at 5-year. The present outcomes resembled the reports of previous meta-analyses, in which the survival rate of short implants in long-term follow-up was lower than long implants 14,29 , and more updated RCTs were included in the present meta-analysis compared to preceding. Moreover, moderate consistent findings across studies were corroborated by the relative deficiency of heterogeneity (I 2 = 17.8%, CI 0 to 53.0%, P = 0.241) and the limited dispersion of the funnel plot, suggesting relatively low between-study heterogeneity. According to the GRADE system, pooling of studies on implants survival rate provided moderate-quality evidence.
For subgroup analyses, no significant difference between test and control groups was found when considering the influence of implant position (maxilla/mandible). In spite of this, the survival rates of the extra-short implants displayed a more serious downward trend over time than longer implants both in upper and lower jaws, which may imply the not optimistic long-term (more than 5 years) clinical outcomes. Interestingly, in favor of the longer implants was found in the native bone group at 5 year, while its reconstructed counterpart failed to attain significance, indicating the extra-short implants could be an acceptable alternative to longer implants in atrophic posterior arch. Meanwhile, a better result for extra-short implants could be anticipated as the development of implant surface modification and shape design 71,72 . Interestingly, mandibular longer implants with vertical bone augmentation displayed a slightly less survival rate than the extra-short implants at 1 year, while the survival rate of extra-short implants was no better than that of mandibular long implants without augmentation. Although the rapid development and wide application of vertical bone augmentation, the high technique-sensitivity and possible complications could remain the essential stimulus of implant failure 73 , which may contribute to the above outcomes.
Information of multiple sites within a single subject are frequently collected in oral health study designs, introducing generally positive correlation among responses within subject 74 . Much less attention has been devoted to this essential issue. In fact, however, increased risk of bias, leading to overestimation of performance of implants, or inappropriate conclusions drawing from the outcomes may be attributed to this substantial problem 75 . Among the 31 included RCTs, 4 of them assigned one implant per patient, in which the within-patient correlation did not need to take into consideration. When it comes to the rest of these RCTs with multiple implants each patient, only one study 53 randomly selected one implant per patient, in which the within-patient correlation could be avoided to some extent. Nevertheless, none of the remaining 25 RCTs adjusted for within-patient correlation. Almost all of these RCTs just reported the number of survived/total implants in contingency www.nature.com/scientificreports/ www.nature.com/scientificreports/ the result of survival rate, not only in the present systematic review but also probably all the previous review on this topic, should be interpreted with caution, since it is particularly hard for subsequent meta-analysis based on these RCTs to calculate the adjusted RR estimate. Furthermore, even though the only RCT attempted to deal with the within-patient correlation by randomly selecting one implant per participant, it is an inefficient design due to the loss of potentially highly valuable information and inferior statistically valid standard error estimate 76 . The simplest strategy to avoid this problem, which is also the majority of the included RCTs employed, is to generate a summary statistic over all implants in the same patient, and then the standard statistical methods for independent observations could be applied. However, as the result, the sensitivity to effects, power and precision may be decreased due to the averaging over many sites and reduced sample size 74 . The RR estimation (at patient level) of the studies in which patient was considered as analysis unit was calculated so as to avoid the influence of within-patient correlation on pooled RR as much as possible. And the results seemed to be similar to those at implant level.
Marginal bone loss. MBL calculated and compared at the patient level, instead of the implant level, were analyzed in this systematic review. The MBL from IP showed that the bone resorption for test group was less than control group with the statistical significance. And as the time goes on, there was greater mean difference of the MBL between two groups, which indicated that, in terms of MBL, extra-short implants would prior to longer implants especially in long-term prognosis. Nevertheless, when the length of implant was taken into consideration, the greater absolute value of MBL may not exactly equal to the greater bone loss relative to the length of implant. In other words, greater MBL of longer implants compared to extra-short implants may not absolutely leads to the greater C/I ratio of longer implants. Moreover, bone augmentation procedure was performed in all the included 3-and 5-years RCTs, which may contribute to the greater MBL in control group. In addition, when it comes to the subgroup analyses for the effect of augmentation, the MBL from IP at both jaws of extra-short implants was statistically significantly less than longer implants in reconstructed bone at 1 year, while insignificant difference was found in native bone, which corroborated the effect of bone augmentation on MBL.
Similar results were found in subgroup analysis of implant position (maxilla/mandible) at 1-and 3-years follow-up. And significant difference was found in maxilla with and without augmentation at 1 year. Surprisingly, mean differences of MBL with augmentation was less than that without augmentation, which was contrary to the above outcomes. Nevertheless, this result, which was speculated from the analyses of only one investigation 47 installing long implants without bone augmentation and seven articles 38,[42][43][44]46,48,55 utilizing augmentation procedure (sinus lift) when indicated. Moreover, that investigation 47 was the only study that employed the immediately loaded implants among all the included 1-year studies on maxilla. In present analysis, the loading method (immediate/conventional) had a significant impact on the mean difference of MBL at maxilla after 1-year measurement, and greater mean difference was found in the immediately loaded implants. In addition, significant resorption was prone to occur after vertical bone augmentation, particularly in the mandible 77 , and higher MBL after augmentation was validated in the previous meta-analyses 17,25,78 . Therefore, the immediate loading was likely to bear the mainly responsibility for the greater mean difference of MBL between two groups in this study. The heterogeneity in the MBL measured from IP at 1 year was relatively severe, and the potential sources of heterogeneity would be augmentation procedure, implants location and loading method. While, based on the sensitivity analysis and publication bias assessment, this result was credible. In addition, according to the GRADE system, pooling of studies on MBL measured from IP provided low-quality evidence and from PR were rated as moderate-quality. www.nature.com/scientificreports/ There was no significant difference of MBL from PR between test and control groups in all defined follow up periods. Besides, the subgroup analyses for the effect of implants position and augmentation procedure showed the similar results. Thus, a higher MBL from IP was observed in control group in this article as above. The bone remodeling process, which was known to proceed along with an adaptive biological width after the second stage of implant surgery and prior to prosthetic loading, could account for the different results of MBL measured from IP and PR 79 . Compared to implant placement, the bone should be more stable when prosthesis loaded and the impact of the initial bone remodeling could be primarily avoided if baseline was measured at this time. In this way, prosthetic factors which may affect the marginal bone resorption could be better analyzed and comprehended. In addition, implant placement without bone augmentation procedure were employed in some of the selected RCTs, as well as relatively small number of analyzed studies, may account for the insignificant difference between two groups.
Complications. Complications in this systematic review were calculated and compared at the patient level, instead of the implant level. The extra-short implants displayed a significantly lower biological complication rate than longer implants in both maxilla and mandible, especially in reconstructed bone, which was corroborated in the previous meta-analyses 2,24,25,29 . The higher complication rate of longer implants in reconstructed bone, such as paresthesia, graft infection, graft resorptions, perforation of the sinus membrane, could be caused by the augmentation procedures. Moreover, compared to short implant placement, the augmentation procedure is characterized by relatively time-consuming and suffering, which might be slightly related to the high complication rate of long implants with augmentation, as some biological complications were self-reported by patients.
In present meta-analysis, no statistically significant difference of prosthesis complication rate was found between two groups, which was similar with previous studies 15 . Meanwhile, despite no statistical significance, the extra-short implants trended to have higher prosthesis complication rate in native bone, and inverse pattern was observed in reconstructed bone at 1 year. It was also reported that the rate of prosthesis complication in extra-short implants (≤ 6 mm) was higher than long implants (≥ 10 mm) 29 . The difference may be caused by the different definition of control group that the implants of length no less than 8 mm were all considered as control group in this review. The higher C/I ratio is often considered as a risk of extra-implants for prosthesis restoration, but several investigations have failed to demonstrate a detrimental effect of this parameter on the rates of prosthesis complications 10,[80][81][82] . In addition, meta-regression of C/I ratio was hard to accomplish since only four series of studies in included articles reported C/I ratio 40,41,50,57 . Thus, together with our results, the authors suggest that the possible prosthesis complication rate of extra-short implants would be approximately higher than longer implants in long term prognosis.
The Consolidated Standards of Reporting Trials (CONSORT) statement, which was developed in 1996, updated in 2001 and 2010 [83][84][85][86] and endorsed by many biomedical journals, aims to improve clarity and consistency of transparency of reporting in RCTs. There was more possibility for RCTs published after 2010 to have superior performance when evaluated with Cochrane Risk of Bias tool in total 87 . However, the use of the CONSORT statement varies among RCTs since only 20 out of 31 included articles adhere to the CONSORT statement, according to the present systematic review. Moreover, only eight out of thirty included RCTs had been registered in a public database and only two of them were prior registered 53,66 , which has been required by many medical journals for years. Therefore, a further encouragement to dental researchers to register in public database and adhere to the CONSORT statement should be established so as to improve the quality of RCTs and consequently better patient care.
Compared to previous meta-analyses 2,29 , more articles reporting a 5-years follow-up were included, together with low publication bias and heterogeneity, leading to a more reliable outcome of long-term follow up. The quality of evidence was assessed using the GRADE approach and the majority were found to be moderate-quality, while the MBL measured from IP showed low-quality due to the imprecision and inconsistency of included RCTs 88,89 . The result of MBL from IP should be interpreted with caution since the low-quality and high heterogeneity. Some subgroup analyses in long-term follow up studies, such as maxilla/mandible for 5-years MBL and complication rate, were impossible to perform since the lack of maxillary implants studies at 5 years. Moreover, the meta-regression analyses of potential associating clinical variants in long-term follow up, suggesting the direction for future research, should be interpreted with caution due to the limited number of 5-years studies reporting analyzable clinical details. And more notably, nearly all of the previous RCTs and systematic reviews on this topic ignored the importance of the statistical issues, especially the problem of within-patient correlation during the analysis of implant survival rate. The comparisons of survival rate at implant level between short and long implants in systematic reviews should be interpreted with caution since the imperfect statistical method of design and analysis in included RCTs.

Conclusion
The above outcomes indicate that the placement of extra-short implants (≤ 6 mm) is an acceptable alternative to longer implants (≥ 8 mm) with bone augmentation in atrophic posterior arch, due to the comparable survival rate, less bone resorption as well as lower biological complication rate. Further high-quality and prior registered RCTs with a longer follow-up period (at least 5 years), appropriate statistics approaches, satisfactory adherence to CONSORT statement are required to corroborate the present outcomes.