Craniofacial growth predictors for class II and III malocclusions: A systematic review

Abstract Objective To evaluate the validity of craniofacial growth predictors in class II and III malocclusion. Material and methods An electronic search was conducted until August 2020 in PubMed, Cochrane Library, Embase, EBSCOhost, ScienceDirect, Scopus, Bireme, Lilacs and Scielo including all languages. The articles were selected and analyzed by two authors independently and the selected studies was assessed using the 14‐item Quality Assessment Tool for Diagnostic Accuracy Studies (QUADAS‐2). The quality of evidence and strength of recommendation was assessed by the GRADE tool. Results In a selection process of two phases, 10 articles were included. The studies were grouped according to malocclusion growth predictor in (1) class II (n = 4); (2) class III (n = 5) and (3) class II and III (n = 1). The predictors were mainly based on data extracted from cephalometries and characterized by: equations, structural analysis, techniques and computer programs among others. The analyzed studies were methodologically heterogeneous and had low to moderate quality. For class II malocclusion, the predictors proposed in the studies with the best methodological quality were based on mathematical models and the Fishman system of maturation assessment. For class III malocclusion, the Fishman system could provide adequate growth prediction for short‐ and long‐term. Conclusions Because of the heterogeneity of the design, methodology and the quality of the articles reviewed, it is not possible to establish only a growth prediction system for class II and III malocclusion. High‐quality cohort studies are needed, well defined data extraction from cephalometries, radiographies and clinical characteristics are required to design a reliable predictor.


| INTRODUCTION
The precision in the diagnosis and evaluation of growing patients is relevant in the field of orthodontics, since it allows the prediction and assessment of the amount of growth for planning orthopedic, orthodontic or surgical treatment (Alexander et al., 2009) with the aim of a successful outcome.
The great variability in the direction and quantity of craniofacial growth implies great importance for success in orthodontic treatment, which has generated great interest in the search for methods of predicting individual facial growth in terms of direction and magnitude (Solow & Siersbaek-Nielsen, 1992), since it would allow to estimate future changes in the vertical or horizontal relationship (Turchetta et al., 2007).
In the past, the theory popularized by Brodie (1941Brodie ( , 1946 and Brodie et al. (1938) indicated that growth patterns were established at an early age; however, evidence would show that there are changes in the growth pattern over time in both direction and quantity, which would support the search for some system to predict craniofacial growth in the future (Rudolph et al., 1998). The interaction between all components of the craniofacial system, such as genetic and environmental factors (Auconi et al., 2014), increases the complexity of their growth prediction. Therefore, the integration of the components should be established to obtain predictive models developed in recent times and that have allowed us to infer the progression of the dentoalveolar imbalance congruent with the biological principles of growth and development (Araya-Díaz et al., 2013;Auconi et al., 2014;Janes & Yaffe, 2006;Ruz & Araya-Díaz, 2018).
Among the different prediction methods available for craniofacial growth, there are systems based on statistical information according to averages of growth increments (Solow & Siersbaek-Nielsen, 1992); Another approach uses facial structure characteristics: Facial types, structural features of lower face (Lavergne, 1982), n-tgo gn angle, proportion of anterior to posterior facial height (Solow & Siersbaek-Nielsen, 1992), regression equations to predict mandibular rotation (Skieller et al., 1984), graphic projection techniques (Ricketts, 1972), cervical and craniofacial posture (Solow & Siersbaek-Nielsen, 1992) and development of mathematical models from computational techniques extracted from cephalometric data (Auconi et al., 2014), cephalometric criteria and procedures such as meshing criteria, grids among others (Johnston, 1975;Moorrees & Lebret, 1962;Popovich & Thompson, 1977;Ricketts, 1972). Despite the existence of these predictors, there would not be methods with relevant clinical acceptability to predict growth (Hirschfeld & Moyers, 1971), which makes it difficult to generate a proposal for use in orthodontic practice.
Pre-adolescent subjects with class II malocclusion have favorable and unfavorable growth patterns and their predictability could determine the planning and result of orthodontic treatment (Rudolph et al., 1998).
Despite the characterization of these patients, there is no precise method to predict the amount, direction and magnitude of their growth, as it would be difficult to determine the contribution of the predictors when craniofacial changes occur due to treatment or growth.
In subjects with class III malocclusions, evidence based on longitudinal studies would indicate differences in mandibular growth compared to class I subjects, where skeletal and dental components tend to manifest early in class III children (Guyer et al., 1986;Tollaro et al., 1994) and they would worsen with growth (Alexander et al., 2009). Reyes et al., 2006, indicate that there would be no tendency for sagittal self-limitation in class III malocclusions (Reyes et al., 2006). In addition, there would be multiple environmental, behavioral and genetic factors contributing to the determination of mandibular morphology and where genetic factors would play a significant role (Bayram et al., 2014;Huh et al., 2013). This multifactorial characteristic would make it more difficult to establish a prediction system.
The purpose of this study was to identify and analyze prediction methods to determine growth in subjects with class II and III malocclusions to estimate skeletal, sagittal and vertical dentoalveolar changes.

This systematic review was conducted according to the Preferred
Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) statement (Moher et al., 2009).
The aim of this systematic review was to answer the PICO question (Population, Intervention, Control groups and Outcome): "What are the prediction methods (I) to accurately determine the growth in the short and long term (O) in patients with class II and III malocclusion (P) when comparing craniofacial growth over time (C)?" an electronic search was conducted in April 2019, updated on 23 August 2020. The electronic databases used were PubMed, Cochrane Library, Embase, Scopus, EbscoHost, ScienceDirect, Bireme, Lilacs y Scielo.

| STUDY SELECTION
3.1 | Inclusion criteria for this review were as follows:

| Types of studies
Cohort studies with the objective of designing or proposing some method to predict growth in patients with skeletal class II and III malocclusion.

| Language of the studies
Search of studies without limitation of language, but the studies included for analysis were in Spanish, English and Portuguese. This is based on the fact that these are the languages used by researchers.

| Types of participants
Selected studies included growing subjects of both genders, with the clinical/imaging diagnosis of skeletal class I, II, and III malocclusions.
The participants included were not subjected to a surgical procedure in the facial skull region, were not subjected to any previous orthopedic or orthodontic treatment and nor did they present any syndrome or alteration of facial skull growth.

| Intervention types
Studies without intervention, with the aim of designing and proposing predictors of growth in the short and long term in subjects with class II and III malocclusion.

| Types of results
Primary outcomes: Analyze studies that design and propose predictors of vertical and/or sagittal growth in growing subjects with class II and III malocclusions from clinical or imaging data, analyze the available evidence when determining the cephalometric or clinical predictors constructed using computational modeling, mathematical equation and other methods based on statistical analysis. In addition, establish the risk of biases of these studies to determine their methodological quality.

| Data collection
For class II and III diagnostic: Data obtained from cephalometric methods (Steiner, Ricketts, Delaire analysis among others), radiographs for orthodontic planning. Clinical methods (occlusal, intraoral and extraoral examination), laboratory (biological samples analysis) and methods based on mathematical models with data obtained from clinical, imaging and/or cephalometric data.
Predictor construction: Multivariate or univariate analysis, computational methods (based on discriminant analysis, machine learning), mathematical modeling among others.

| Search strategy
For the identification and selection of the number of potentially eligible studies for this systematic review (N), a specific and individualized search strategy was developed for each database. A semantic field was determined for the term "Class II and III malocclusion" and another semantic field related to the term "Growing Predictors." The search strategy is found in Table A1 in Appendix of this review.

| Study selection
In a first screening, the title and abstract of all potentially eligible articles were listed and evaluated by two researchers independently (J.A., C.R.). In a second stage, the full text of articles that potentially met eligibility criteria based on the first screening was assessed independently by the same two researchers (J.A., C.R.) according to inclusion criteria (study design: clinical trial, diagnostic studies; objective: to propose predictors based on clinical, imaging, cephalometric methods, mathematical models among others, that allow to predict growth for class II and III patients; type of participants: patients in the growth stage). When no agreement was found, the inclusion of the article within the sample was discussed with a third researcher (A.P.) who acted as an arbiter. Articles that met inclusion criteria were included in the review for the final analysis. The reasons why some studies were excluded were recorded in an adjacent column (Table A2 in Appendix). The quality of assessment according to GRADE, was performed by two independent reviewers (V.S. and T.J.). To determine the quality and methodological validity in relation to the diagnostic methods of the selected studies, Quality assessment of studies of diagnostic accuracy included in Systematic Reviews -QUADAS-2 was used (V.S. and T.J.).

| Extracting data from studies and data synthesis
The PICO format (Population, Intervention, Control groups and Outcome) was used to make the tables of analyzed articles: Population (sample size, distribution by gender, age range and SD); Intervention: (Instrument for malocclusion diagnostic, image acquisition protocol and type of predictor); Comparison criteria or control: (comparison of craniofacial growth over time) and Outcomes (including the answer to the hypothesis, statistical analysis. Finding overall).

| Risk of bias in individual studies
The Grading of Recommendations Assessment, Development and Evaluation (GRADE) system (GRADE, 2014), was used to evaluate the quality of evidence. Two authors independently assessed the quality of the evidence and the strength of the recommendations according to the risk of bias. The methodological quality of the selected studies was evaluated with the QUADAS-2 tool (Whiting et al., 2011), used to assess the quality of diagnostic accuracy studies. Two authors independently rated each item as "yes," "no," "unclear," "low" or "high."

| RESULTS
2445 articles were identified from the 9 electronic databases. The studies were exported to an-Excel table, and of these articles, 196 were eliminated because they were duplicates. The remaining 2249 studies were evaluated by the authors in a first screening and 2221 of these were eliminated because they were not relevant for this study. Of the remaining 28 studies, 18 were eliminated in a second screening when the full text of the articles was analyzed, and the reasons for exclusion are shown in Table A2 in Appendix. Finally, ented in Table A1 in Appendix and the flowchart of the literature search is presented in Figure 1.

| Characteristics of participants
In the articles analyzed (Tables 1-3), a total of 1313 participants were investigated, with an age range between 6 to 20 years, both genders were included, although three studies included only female subjects (Auconi et al., 2014;Chen et al., 2005;Scala et al., 2012) and 1 only male (Buschang et al., 1986). According to the type of malocclusion, the studies analyzed included class II skeletal malocclusion (n = 4); class III (n = 5) and class II/III (n = 1) ( Table 4).

| Characteristics of predictors
All studies included predictors designed from cephalometric data obtained from growing patients and considered only cohort study designs (Table 4). For class II malocclusions, 4 articles were analyzed and the proposed predictors consisted of mathematical equation (Arias et al., 2006;Rossouw et al., 1991;Rudolph et al., 1998) and computerized structural superimposition (Solow & Siersbaek-Nielsen, 1992). For class III malocclusions, six studies were identified.

| Risk of bias of included studies
The studies in general were methodologically heterogeneous, because the types of analysis differed among the included studies, although they all proposed growth predictors from cephalometric and/or clinical data. The methodological quality of the predictors according to QUADAS-2 was low to moderate and none of the articles met all its Growth prediction formulas: It has been shown that individualizing prediction by assessing maturational development rather than chronologic age can greatly increase the accuracy of prediction.
Abbreviations: Ar-Po, linear distance from Ar to Po; ArGoMe, Gonial angle; ANB, anteroposterior relation of the maxilla and mandible; Ba-Na, cranial base length from Ba to Na; Co-A, midfacial length as distance from Co to A; Co-Gn, mandibular length as distance from Co to Gn; GP, mandibular growth potential; GPM, mandibular growth potential method;  Buschang et al., 1986;Scala et al., 2012;Solow & Siersbaek-Nielsen, 1992;Turchetta et al., 2007). The analysis of the quality of the evidence, according to the GRADE tool (Table 6, Figure 3) indicated that the available evidence regarding growth predictors in patients with class II and III malocclusions was low.

| Synthesis of results
The results collected from the included studies were based on levels of prediction of measurements made on clinical and cephalometric ANB angle and its capacity of improvement through the years. appropriate interval between index test(s) and reference standard?  (Turchetta et al., 2007) on the basis of maxillary and mandibular angular estimates for classes I and II and for Class III mandibular group estimates. In addition, the method proposed by Buschang et al. (1986), presents an approach based on a polynomial model, which would provide estimates to describe the average size, speed and acceleration, reducing the required longitudinal cephalometric data.

| Quality of the evidence
Ten studies were included for qualitative analysis in this systematic review. Based on their design, all the articles were cohort studies, which indicates that there was a follow-up in the growth of the subjects to adequately design the prediction systems. All articles presented a high risk of bias when analyzed with GRADE, although the studies conducted by Buschang et al. (1986), Turchetta et al. (2007), Abu Alhaija and Richardson (2003), presented a better methodological quality, particularly in the random sequence generation, allocation concealment and other bias domains compared to the rest of the studies.
When the QUADAS-2 tool was considered to determine the pre- growing patients in all the studies analyzed. Of these, three studies consisted only of female subjects (Auconi et al., 2014;Chen et al., 2005;Scala et al., 2012) and one only of male subjects (Buschang et al., 1986), which could limit the interpretation of the results to growing patients in the general population .

| Predictors of growth for class II malocclusions
Five studies proposed predictors of growth for subjects with class II malocclusions. Three articles designed predictors based on mathematical models (Arias et al., 2006;Buschang et al., 1986;Rudolph et al., 1998) and a study in computerized structural superimposition (Solow & Siersbaek-Nielsen, 1992)

| Predictors of growth for class III malocclusions
Six articles designed growth predictors for class III malocclusions.
Some studies included the design of predictors using resources such as: Computational Modeling (Auconi et al., 2014), Network Modeling (Scala et al., 2012), Cluster analysis (Abu Alhaija & Richardson, 2003) and Software methods (Schulhof et al., 1977), all constructed from cephalometric data. Meanwhile, other prediction systems were designed based on cephalometric analysis (Ricketts analysis, the Johnston grid system, and the Fishman) (Turchetta et al., 2007) and the use of a linear equation based on a mathematical model to predict mandibular growth (Chen et al., 2005). Of these, the predictors designed in the Turchetta and Abu Alhaija studies (Abu Alhaija & Richardson, 2003;Turchetta et al., 2007) presented the lowest risk of bias according to the GRADE and QUADAS-2 tools.
These predictors could have clinical relevance in subjects who will undergo orthodontic and/or orthopedic treatment with the objective of defining the beginning of the camouflage treatment during growth or waiting until the growth is complete to plan an orthodontic-surgical treatment (Ghiz et al., 2005). Turchetta et al. (2007), concluded that Fishman's method could be the best in the short and long term. This method is based on maturational age determined by hand-wrist radiograph. The percentages of total growth completed are considered instead of linear growth in absolute terms, several facial linear measurements are applied to construct a prediction. According to the authors, when evaluating maturational development instead of chronological age, physiological variability among children at the same chronological age is reduced (Turchetta et al., 2007), which would increase the accuracy. In the study performed by Abu Alhaija and Richardson (2003), 3 clusters were formed: long facial types (cluster I), short (severe class III discrepancy or cluster II) and intermediate (moderate intermaxillary discrepancy or cluster III). The percentage of discrimination was 80% when DFA was performed (discriminant function analysis), which was satisfactory, but when the analyses were performed in the groups separately, the results varied for cluster I with a good or bad result in 92%, 85% cluster II and 100% cluster III. The The other four studies (Auconi et al., 2014;Chen et al., 2005;Scala et al., 2012;Schulhof et al., 1977), presented a high risk of Schulhof et al. (1977), designed a predictor (=V CN/SD × V), and con- Although all the predictors analyzed in this review were constructed from the follow-up of growing patients and the data were obtained from the clinic, cephalometry and/or radiographs, the genetic factor should be considered for future studies. The new findings could explain the genetic susceptibility to the class III phenotype with mandibular prognathism when there is presence of GHR and FGF polymorphisms, and could also explain the CA genotype of P561T with greater mandibular length (Co-Gn) (Bayram et al., 2014) .
The natural progression of class III has not been accurately tested yet, F I G U R E 3 Criteria met, according to the GRADE tool since most of the evidence is based on case-control studies that cannot yet establish an association between genetic variation and class III malocclusion (Cruz et al., 2017), which has not been evaluated in the studies analyzed in this review.

| Predictors of growth in treated patients
There are proposals of predictors to determine the success of treatment in growing subjects, however, these studies were not included in this review because they considered the intervention of subjects under orthopedic and/or orthodontic treatment. Despite the evidence in this topic, there would be no consensus predictor, since the differences given in the sample collection, the characterization of the subjects (excessive mandibular growth, lack of maxillary growth, combination of both and hypo or hyperdivergent growth pattern), long-term follow-up and different classification criteria make this difficult. In the study by Ghiz et al. (2005), a logistic regression model was  Rossouw et al. (1991) suggested that in class I and III malocclusions, the frontal sinus surface (in mm 2 ) would be an indicator to predict increased mandibular growth in subjects with a larger frontal sinus. Arntsen and Sonnesen (2011), showed that fusion abnormalities in the cervical spine would be associated with a greater mandibular sagittal relationship, mandibular retrognathia, greater mandibular inclination and an extended head posture; for class III malocclusions, Yang and Kim (1995) presented the sum of Björk, the gonial angle and the occlusal plane to the angle of the AB plane; Ko et al. (2004), the incisor inferior to the angle of the mandibular plane and Baccetti et al. (2004), the mandibular ramus, angle of the skull base and angle of the mandibular plane.

| Limitations
The limited evidence and risk of bias found in most articles constitutes a limitation of this SR. Although all the studies designed predictors based on cephalometric data, these were not similar and most authors proposed different types of predictors. In spite of the similarity of the design in most of the articles, (all were cohort studies), the heterogeneity of the methodology to propose prediction models does not allow comparisons between them, and neither does the difference in the system of prediction. In addition, the risk of bias present in most of the studies analyzed using the GRADE and QUADAS-2 tools would be mainly due to the absence of randomization of the sample, shielding and interpretation of the index test, which suggests improving these items in future research.
It was not possible to propose a single method, since most of the predictors designed in the studies were established from multiple cephalometric variables, and there is no standardization of the points, angular and/or linear measurements, also considering the heterogeneity of the designs (prospective or retrospective cohort) and characterization of malocclusions, which makes it even more difficult to establish any comparison. Despite this, and based on the findings made in this review, it is possible to suggest that the predictors for the growth of classes II and III proposed by Buschang et al. (1986) and Turchetta et al. (2007) could be useful in orthodontic practice as their methodological quality is better.
Given the heterogeneity of the methodology used in the studies, in the designs of the predictors, number of patients and distribution by gender, it was not possible to perform a meta-analysis.

| CONCLUSIONS
Predicting growth is one of the most relevant challenges in the field of craniofacial growth and development, as it would allow the planning and prediction of timing and prognosis of first and second phase treatments in orthodontics.
From the findings made in this systematic review, it is possible to conclude the following: • The available evidence from studies that design class II and III predictors is scarce and their methodological quality in general is low to moderate.
• There is no consensus to establish a single predictor, since the designs of the studies are heterogeneous, the extraction of data from the studies was not standardized and in general they do not characterize the patients.
• More cohort studies with a higher level of evidence are suggested, with more homogeneous designs and standardized methods to extract the data from the clinic, radiographs and cephalometrics methods.