Extraction vs nonextraction orthodontic treatment: a systematic review and meta-analysis

The longest running debate in orthodontics, spanning more than a century, has been the effects of extraction and nonextraction treatment.¹  The main concern with extraction treatment has been the possible deleterious effect on facial profile and the main concern with nonextraction treatment being post-treatment stability.²  Edward Angle’s philosophy of preserving the full complement of teeth argued that extraction of teeth would cause an imbalance in facial harmony and abnormal function due to the change in arch width and form.³  Unlike many of Angle’s disciples, Calvin Case opposed this philosophy and defended the extraction of teeth in treating malocclusion to avoid later relapse.⁴  However, it was not until the 1940s, when more members of the orthodontic community (including Charles H. Tweed and Raymond Begg) also supported an extraction treatment approach, that it became a generally accepted option.^5,6 

Since then, the pendulum has swung between extraction and non-extraction treatment, reporting a peak extraction rate of 76% in 1968⁷  declining to 17.6% in 2005⁸  among University of North Carolina patients. At the University of São Paulo, nonextraction treatment continued with an upward trend from 14.29% (1973–1977) to 54.55% (2003–2007).⁹ 

The orthodontic literature has discussed this conundrum, with conflicting results. Bowman and Johnston¹⁰  examined the effects on facial profile and concluded from a sample of 120 patients that extraction treatment had positive results for patients who had initial protrusion relative to the E plane, but it was detrimental for those who had retrusive lips before starting treatment. Boley et al.¹¹  studied profiles of 50 patients and concluded that no difference was found between the two groups as facial profile measurements (Holdaway H-line) were within normal limits. Konstantonis¹²  attributed change in the soft tissue profile of extraction patients to greater incisor retraction, which could be controlled during treatment planning with less retraction mechanics and more mesial movement of posterior segments. These effects were more pronounced in patients with thin lips or high lip strain.

Little et al.¹³  concluded that extraction did not guarantee long term stability and Rossouw et al.¹⁴  reported no significant difference in stability between extraction and nonextraction groups, with similar amounts of relapse.

The literature has previously reported premolar extraction compared to nonextraction treatment focused on limited outcomes.^12,15–23  A recently published scoping review²⁴  outlined the weaknesses of published evidence across the breadth of the current literature but did not include any quantitative evaluation of the available data. This systematic review was, therefore, focused on four first premolar extraction, a broad range of outcomes and quantitative analysis, providing the orthodontist with the evidence required to inform clinical decisions.

The aim of this systematic review was to compare four first premolar extraction and nonextraction treatment effects on arch form, maxillary and mandibular intercanine width and first molar width, profile changes (upper and lower lip prominence to E plane), treatment duration, occlusal outcomes (end treatment UK and US weighted peer assessment rating [PAR] scores, American Board of Orthodontics Objective Grading System [ABO-OGS] score), posttreatment smile aesthetics (aesthetic score, maxillary intercanine width/smile width, visible dentition width/smile width, maxillary intercanine width/visible dentition width) and posttreatment changes of maxillary and mandibular anterior alignment (Little’s irregularity index) to provide orthodontists with the best data available.

This review was prepared in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement and the Cochrane Handbook for Systematic Reviews of Interventions (PROSPERO:CRD42021254523).

Eligibility criteria are in Table 1. Electronic databases were searched until June 2, 2023 without restrictions regarding publication year, study design, or language, with additional searching of gray literature, unpublished literature, and hand-searching of reference lists of included and excluded studies comparing premolar extraction to nonextraction treatment for outcomes of interest. Search strategies and publication date range of the search are in Table 2.

The articles resulting from the search were added to Zotero (version 6.0.26). Duplicates were identified and removed. Articles were manually checked during screening and further duplicates found and removed. Articles were first checked and excluded by title, with the resultant articles screened by their abstract and then full text articles checked for eligibility.

If there were any difficulties in getting the full text of an article, soft copies were obtained from the University of Dundee Library or The British Library. No contact was made with the authors.

Non-English studies without an English version were translated using Google Translate. Sample size and all reported data were checked and values were revised and recalculated whenever raw data were provided to ensure quality of included data.

Two reviewers undertook study selection and data extraction in duplicate using a customized data extraction form (Appendix 1 and 2) prepared by the third reviewer. When there was disagreement, discussion with the third reviewer reached the final decision.

Due to the ethical challenges in undertaking comparative prospective and randomized clinical trials in this subject area, retrospective studies were included. ROBINS-I tool was used to assess the quality of observational (prospective and retrospective) studies, with each outcome being individually judged. Cochrane RoB 2 tool was used for randomized trials.

Where appropriate, continuous data, with sample size, mean value, and standard deviation were available, RevMan (version 5.4.1) was used for quantitative synthesis and narrative synthesis reported when meta-analysis was not possible. Confidence interval (95%) with mean difference was used with significance level P < .05. I² statistic for random effects model meta-analysis was calculated using Comprehensive Meta-Analysis software (version 3.0). Prediction interval (95%) with mean difference, used to describe the distribution of true effect sizes, was calculated using an Excel spreadsheet (Microsoft Corp., Redmond, Washington) based on formulas by Borenstein et al.^25,26 

Outcome measures and time points of assessment are presented in Table 1. Studies reporting female and male subgroups were combined into a single group using RevMan. Random-effects model meta-analysis was used because of the amount of heterogeneity due to the difference in populations and study design. Heterogeneity was assessed by assessing overlap of the confidence intervals on Forest plots and I² statistic with threshold for interpretation as described in the Cochrane Handbook.

Subgroup and sensitivity analyses were carried out to deal with possible sources of heterogeneity of including different malocclusion classes together and differences in outcome measures to isolate their influence.

Publication bias was addressed by including unpublished literature. When more than 10 studies pooled together for an outcome in the meta-analysis, publication bias was identified through a funnel plot. GRADE was used to assess certainty of evidence for each outcome.

The search of databases (including gray literature and hand-searching) identified 2652 articles. Removal of duplicates, exclusion by title (Appendix 3), and screening by abstracts (Appendix 4), resulted in a total of 383 articles (Appendix 5).

Thirty (29 retrospective studies^27–55 , 1 randomized controlled trial [RCT]⁵⁶ ) studies were included with some studies including multiple outcomes. Twenty-four^{27,29–33,35,36,38–44,46–51,54–56}  were included in meta-analysis, of which one thesis⁴⁷  and two non-English articles in Chinese⁵⁴  and Korean⁴²  were included (Table 3). Narrative synthesis included three studies^31,45,53  for UL-E plane (upper lip), five studies^{31,34,37,45,53}  for LL-E plane (lower lip), two studies^28,52  for ABO-OGS, and two studies^32,33  for maxillary and mandibular anterior alignment (Little’s irregularity index) (Table 4). Figure 1 shows the process of study identification and selection.

The included RCT⁵⁶  was judged as being of high risk of bias (Figure 2). For retrospective studies, 23 were of serious risk^{28–33,35–41,43,45–50,53–55}  and six of moderate risk of bias^{27,34,42,44,51,52}  (Table 5) (Appendix 6).

Results of meta-analyses, prediction interval, subgroup, and sensitivity analyses are presented in Table 6 with values rounded to two decimal places except for P value.

Nine retrospective studies^{27,31,39,42,44,47,49,51,54}  and one RCT⁵⁶  reported no statistically significant difference between four first premolar extraction and nonextraction treatment in maxillary intercanine width (MD: 0.02 mm; total 95% CI [−0.38, 0.43]; I² = 0%; P = .91) with significant increase in mandibular intercanine width (MD: 0.68 mm; 95% CI [0.36, 0.99]; I² = 0%; P < .0001) in the nonextraction group.

Eight retrospective studies^{27,31,39,42,44,47,49,51}  and one RCT⁵⁶  reported significant decrease in maxillary (MD: −2.03 mm; total 95% CI [−2.97, −1.09]; I² = 0%; P < .0001) and mandibular interfirst molar width (MD: −2.00 mm; total 95% CI [−2.71, −1.30]; I² = 5.32%; P < .00001) with four first premolar extraction.

Three studies^31,45,53  were included for UL- E plane and five studies^{31,34,37,45,53}  for LL- E plane with vote counting indicating retraction of upper and lower lips with four first premolar extraction.

Choi et al.³¹  compared 15 four first premolar extraction (UL- E plane: −1.61 ± 1.62, LL- E plane: −3.13 ± 1.97) with 17 nonextraction (UL- E plane: −0.07 ± 0.89, LL- E plane: −0.15 ± 0.70) Class I and II female patients and found significant retraction of upper and lower lips in the extraction group.

In an equally divided sample of 20 Class I patients, Freitas et al.³⁴  reported no significant difference between four first premolar extraction (LL- E plane: −0.2 ± 3.7) and nonextraction (LL- E plane: −0.05 ± 1.9) treatment.

Hassan et al.³⁷  reported no significant difference between four first premolar extraction (LL- E plane: −2.15 ± 3.38) and nonextraction (LL- E plane: −0.83 ± 2.75) treatment in a sample of 60 Class I and II Pakistani females.

Konstantonis⁴⁵  compared 30 four first premolar extraction (UL- E plane: −2.75 ± 1.5, LL- E plane: −3.34 ± 1.75) with 32 nonextraction (UL- E plane: −0.68 ± 1.89, LL- E plane: 0.67 ± 2.24) borderline Class I patients and found significant retraction of upper and lower lips in the extraction group.

Xu et al.⁵³  compared 13 four first premolar extraction (UL- E plane: –1.0 ± 1.9, LL- E plane: –2.6 ± 1.9) with 12 nonextraction (UL- E plane: –0.9 ± 2.4, LL- E plane: –0.4 ± 3.4) borderline Chinese patients of different malocclusion and found significant retraction of lower lip in the extraction group with no difference regarding upper lip retraction.

Eight retrospective studies^{29,30,32,33,36,46,47,55}  and one RCT⁵⁶  reported shorter treatment duration in the nonextraction group (MD: 0.36 years; total 95% CI [0.10, 0.62]; I² = 3.18%; P = .007) compared to the four first premolar extraction group.

No eligible studies were found for UK-weighted PAR score. Three retrospective studies^33,38,41  reported no statistically significant difference between four first premolar extraction and nonextraction treatment with US weighted PAR score. (MD: 0.33; total 95% CI [−0.21, 0.87]; I² = 0%; P = .23) (Figure 6).

Two retrospective studies were included with inconclusive evidence. Anthopoulou et al.²⁸  compared 25 four first premolar extraction (total score: 27.04 ± 6.30) with 30 nonextraction (total score: 29.07 ± 7.11) Class I borderline patients and found no statistically significant difference between the two groups.

In 40 Class I borderline patients, Vaidya et al.⁵²  reported lower scores for four first premolar extraction group (total score: 22.0 ± 2.29), compared to nonextraction (total score: 26.8 ± 5.18).

Four retrospective studies^40,43,48,50  for aesthetic score (MD: −0.09; total 95% CI [−0.24, 0.05]; I² = 0%; P = .21) and four retrospective studies^35,40,43,50  for maxillary intercanine width/smile width (MD: 0.01; total 95% CI [−0.00, 0.02]; I² = 0%; P = .12), visible dentition width/smile width (MD: −0.00; total 95% CI [−0.01, 0.01]; I² = 0%; P = .81) and maxillary intercanine width/visible dentition width (MD: 0.00; total 95% CI [−0.02, 0.02]; I² = 0%; P = .94) reported no statistically significant difference between four first premolar extraction and nonextraction treatment.

Two retrospective studies were included with inconclusive evidence. Francisconi et al.³²  compared 40 four first premolar extraction (maxillary Little index: 0.89 ± 1.48, mandibular Little index: 1.64 ± 1.75) with 44 nonextraction (maxillary Little index: 1.64 ± 1.37, mandibular Little index: 1.36 ± 1.33) patients of different malocclusions and found greater maxillary crowding relapse in the nonextraction group and no significant difference between the two treatment groups for mandibular crowding relapse.

Freitas et al.³³  compared 97 four first premolar extraction (maxillary Little index: 1.30 ± 1.75, mandibular Little index: 1.93 ± 2.06) with 58 nonextraction (maxillary Little index: 1.66 ± 1.42, mandibular Little index: 1.40 ± 1.18) patients with no significant difference between the two groups regarding maxillary crowding relapse, and more mandibular crowding relapse in the extraction group.

Prediction interval, the range that in 95% of all populations the true effect size will fall within, was wider than 95% confidence intervals, but with no significant difference, for all outcomes.

Class I subgroup analyses of maxillary and mandibular intercanine width (Figure 3), maxillary and mandibular interfirst molar width (Figure 4) and treatment duration (Figure 5), favored the same results as the main analyses.

For arch width, sensitivity analysis excluded three studies^27,47,56  in which measurement of intercanine (Figure 8) and intermolar (Figure 9) width was from the most labial aspect of buccal surfaces of teeth instead of canine tips and mesiobuccal cusp tips. Sensitivity analysis excluded two studies^30,32  for treatment duration reporting the use of extraoral appliances related to patient compliance (Figure 5), and two studies^35,48  for smile aesthetics (Figure 10), where it was not clear whether included data were end of treatment or posttreatment.

No funnel plot generated, as no more than 10 studies were included in meta-analyses.

GRADE evidence profile was completed for all outcomes (Table 7). No separate grading was undertaken for subgroup/sensitivity analyses.

This review only included studies with four first premolar extraction compared to nonextraction treatment-to-control heterogeneity, with the inclusion of gray literature, non-English studies and published theses to reduce reporting bias. However, eligible studies were of high level of bias except six studies of moderate risk of bias.

Eligibility criteria were set to reduce confounding variables related to effects due to different treatment approaches. For all outcomes, studies included were limited to nonsurgical patients with fixed appliances with no adjunctive procedures. Studies reporting the use of functional appliances or extraoral appliances were excluded for profile.

Confounding variables were further controlled by exclusion of studies with incomplete data reporting to avoid imputations and studies with more than one error in sample size or treatment data, for greater consistency. However, this may have resulted in smaller sample sizes with a potential source of bias,⁵⁷  as not all studies provided raw data for recalculation. No study included in quantitative synthesis reported any error in outcomes of interest, and excluded studies are included in supplementary material with reason for their exclusions if needed for future analysis.

Ideally, age of subjects included would have been limited to 13 years of age or older for arch width⁵⁸  and 15 years of age or younger for profile changes⁵⁹  to exclude growth effects. Significant increase in maxillary and mandibular intercanine and intermolar width occurs due to growth between 3 and 13 years of age, and there is significant upper and lower lip retraction in relation to the E plane between 15 and 25 years of age. However, it was not possible to include an age threshold criterion based on these limits as raw age data were not provided and only mean age reported, a confounding variable of broad age range, increasing indirectness of the results.

There is conflicting evidence in the literature regarding arch width changes, with meta-analysis showing significant increase in mandibular intercanine width in the nonextraction group and no difference regarding maxillary intercanine width. A possible explanation might be related to greater variability in maxillary arch form, whereas mandibular arch form is more influenced by the soft tissue environment, meaning that the need to generate space for alignment in nonextraction treatment leads to mandibular arch width changes while the mandibular arch form is maintained in extraction treatment. The movement of posterior teeth mesially into narrower areas of the dental arch is the cause for maxillary and mandibular intermolar width decrease in the extraction group.⁴⁴ 

The reported results of profile changes, ABO-OGS and stability should be carefully interpreted because of the imprecision of the results due to small sample sizes. Where two studies were included, no meta-analysis was undertaken, as with two studies and in the presence of heterogeneity, confidence intervals based on normal quantiles are not recommended.⁶⁰ 

For profile changes, the current narrative synthesis indicated retraction of upper and lower lips relative to the E plane with four first premolar extraction, matching the findings of the meta-analysis by Konstantonis et al.¹²  However, as ethnic differences were reported, indicating potential additional confounding factors for this outcome, no meta-analysis was undertaken.

Four first premolar extraction took an average of 0.36 years longer to complete in comparison to nonextraction treatment. This might be reasonable to assume as it allows for the time to complete space closure in extraction treatment, although it could also be related to more complex cases being treated with extractions. This difference is one of clinical significance for clinicians and patients. The evidence for this finding was graded as low certainty compared with most other outcomes, which were very low certainty.

Conflicting results on occlusal outcomes were reported in this review. No eligible studies were found for UK-weighted PAR score and no significant difference was reported with US-weighted PAR score. However, it should be noted that end treatment data were pooled rather than percentage improvement for ease of comparison with ABO-OGS.

The results of this review on smile aesthetics indicated no difference between the two treatment approaches in four different smile parameters. Furthermore, Işiksal et al.⁴⁰  stated that inadequate torque expression can affect smile aesthetics regardless of treatment modality.

There was no clear consensus whether four first premolar extraction or nonextraction treatment would provide greater posttreatment stability, as only two studies were included with conflicting results.

In summary, this review found low certainty evidence for a clinically significant difference in treatment duration. However, there are debatable clinical implications of differences found in arch width changes and no differences in occlusal outcomes and smile parameters. The decision whether to extract or not is very situational. Ruellas et al.⁶¹  stated that clinicians should be aware of factors such as compliance, tooth-arch discrepancy, cephalometric discrepancy, facial profile, growth, anteroposterior relationships, dental asymmetry, and pathology in decision making.

In the context of existing data, more robust evidence for changes in outcome between extraction and nonextraction treatment approaches is still needed. However, this is an almost impossible aim, as one of the reasons for the lack of RCTs on this topic is ethical, with patient recruitment dilemmas of randomizing these treatments. Alternatively, high-quality observational studies may be most appropriate and a suggested protocol has been made available recently.²⁴ 

• Studies included in quantitative synthesis were retrospective in nature except one prospective randomized trial with high level of bias. The limitation due to the decision to include observational studies has been discussed in a Cochrane review⁶²  with little evidence for significant effect estimate differences between observational studies and RCTs. However, it is important to consider the level of heterogeneity in meta-analyses of RCTs or observational studies with control for confounding in observational studies.

• There is a possible source of bias due to exclusion of studies with incomplete dataset reporting, but exclusion provides greater confidence in results rather than imputation, with the same for studies with more than one error in sample size or treatment data.

• As malocclusion classes were not studied individually, this baseline characteristic caused an increase in heterogeneity when studies were included for the meta-analysis. Clinical heterogeneity was controlled by limiting the intervention group to four first premolar extraction only and subgroup analyses were carried out where possible.

• As different ages were pooled on the same estimative, without a subgroup analysis, this baseline characteristic caused an increase in the degree of indirectness of the results.

• The subjective nature of the aesthetic score, especially since this was assessed by various raters in the different studies, the authors considered this as a possible source of heterogeneity due to the associated observational bias in the meta analysis.¹⁷ 

• Retention regimen might be a confounding factor related to stability while poor oral hygiene, poor patient compliance, and experience of the operator are factors related to treatment duration.⁶³ 

• There is an imprecision up to ±0.1 in recalculated data of included studies, which may affect the results.

• Non-English studies were translated using Google Translate, which might present a possible source of inaccuracy.

• GRADE does not allow inclusion of multiple study designs per outcome, a recognized limitation of GRADEpro. For outcomes reporting RCT and observational studies, changing study design in certainty assessment did not affect overall certainty.

• Low level evidence indicates clinically significant, shorter treatment duration in the nonextraction group compared to four first premolar extraction.

• Low level evidence indicates mandibular intercanine width increase with nonextraction treatment and mandibular interfirst molar width decrease in the four first premolar extraction group.

• Very low evidence indicates no significant difference in maxillary intercanine width between the extraction and nonextraction groups and decrease of maxillary interfirst molar width with four first premolar extraction.

• Very low level evidence indicates retraction of upper and lower lips-E plane in the four first premolar extraction group.

• Very low level evidence indicates no significant difference regarding US PAR score and posttreatment smile aesthetics.

We would like to gratefully acknowledge Dr. Gavin Revie, Research Methodologist and Research Integrity Lead, School of Dentistry, University of Dundee, for his assistance with the meta-analysis.

This paper was based on a thesis submitted in partial fulfillment of the requirements for the degree of Master of Science (Orthodontics) at University of Dundee.

No external funding was received to conduct this research. No conflicts of interest were declared.

Appendices 1–6 are available online.