Facial morphometrics of children with non-syndromic orofacial clefts in Tanzania

All subjects were African, from northwestern Tanzania, medically assessed and treated at Bugando Medical Centre. The total sample was 126 children, 42 had unrepaired non-syndromic CL/P and 84 were unaffected controls. The CL/P were aged 2 to 120 months, while control group were aged 4 to 120 months. The mean age of the CL/P group was 29.4 ± 31.8 months and the mean age of the control group was 54.1 ± 34.5 months, the difference largely resulting from a greater number of infants in the CL/P group. Information on the structure of the sample in terms of sex, cleft status, and cleft type is shown in Table 1. We used opportunistic sampling for this study, given the location and population under study. We used all possible cases and for the controls, we doubled the sample size to better control for the effects of normal variation in facial shape across ages. Children with unrepaired non-syndromic CL/P were recruited consecutively as they attended surgical wards and clinics for the duration of the study. Children in the control group were recruited conveniently from Bugando Medical Centre and schools in the Mwanza region as long as they met the criteria below. None of the unaffected control children had known craniofacial anomalies and none had received orthodontic treatment. Further, none of the children with unrepaired orofacial clefts had undergone alveolar bone grafting or nasoalveolar molding (NAM) at the time of data collection. Ethical approval was granted from the Tanzania National Institute for Medical Research and the University of Calgary Conjoint Health Research Ethics Board (CHREB). Informed consent was obtained from the parents/guardians of all study subject children. All the facial figures are based on average shape (after morphing) of actual individuals in the unrepaired CL/P and Control groups.

Data collection was done from 2008 to 2012. 3D facial surfaces were captured using the InSpeck 3D MegaCapturor white light digital stereophotogrammetric camera (Creaform Inc., Quebec, Canada), which acquires a 3D surface map in ~0.4 seconds using a 640 × 480 mm field of view with photo-realistic color and texture rendering. Facial scanning was performed with standardized head positions directly in front of the 3D camera while the subject was seated by themselves or on their mother’s lap, approximately 1 meter from the camera, with the lips at rest and eyes closed to minimize involuntary movement. Six standard views were obtained for each subject (frontal, left three-quarter profile, left, right three-quarter profile, right and inferior chin), which were processed and merged to obtain the full 3D facial image using InSpeck Facial Acquisition Processing Software (FAPS) and Editing and Merging software (EM) with a least possible error (approximately 0.300 mm) (Creaform Inc., Quebec, Canada).

From the 3D reconstruction of each subject, 26 soft tissue facial landmarks (Figure 1 and Table 2) were used to characterize facial and nasolabial forms, chosen to provide coverage of soft-tissue facial structures [25, 26]. The stomion landmark was taken as a midpoint of the labial fissure. However, when the lips were not closed in the rest position, it was taken as the midpoint of the lower border of the upper lip. The landmarks were recorded by one individual using MeshLab (V 1.3.0) [27], and the corresponding x,y,z coordinate locations were used for subsequent shape analysis. To ascertain landmark repeatability, a set of five control individuals was landmarked five times each, by one individual observer on five separate days. The average observation and variance was calculated for each landmark’s x,y and z dimensions. Across all landmarks and individual trials, observer precision was calculated to be 0.338 mm. This is well within the precision cited in other similar studies [28].

MorphoJ v1.04a software was used to carry out geometric morphometric analysis of landmark configurations [29]. 3D landmark data for each individual were imported without object symmetry to preserve facial asymmetry resulting from unilateral clefts. The data were subjected to Procrustes superimposition to rescale to standard size, translate to standard position and rotate to standard orientation [30]. To compare unilateral versus bilateral clefts as a group, all individuals with right-side clefts were mirrored to have left-side clefts, by inverting the sign of the x-coordinate for all landmarks. Comparisons of left versus right, however, were carried out on the data prior to mirroring.

Standardization of the dataset for age and size was carried out in MorphoJ [29], using multivariate regression to create a line of best fit for age against the Procrustes coordinates and pooling the data by groups (control and cleft status). The 24 month difference in mean age between the controls and CL/P group is small enough that a linear approximation of shape change with age removes the vast majority of age-related shape variation. The only deviation from this occurs at the extreme lower end of CL/P group, where facial adiposity may introduce some uncontrolled variation into the sample; however, the slope of the age regressions look similar in both the control and CL/P groups. We then adjusted the age-standardized regression residuals for allometry by controlling for centroid size and these residuals, which are an adjusted set of 3D coordinates that describe facial shape after removing the age and size variation within each sample, were used in all subsequent shape analyses.

Using MorphoJ, facial shape variation was analyzed in the combined dataset of cleft and control children and also separately in the CL/P group (unilateral and bilateral) using Principal Component Analyses (PCA). Differences in shape between non-cleft children and unrepaired CL/P children and between unilateral and bilateral clefts in the unrepaired CL/P group were analyzed using Canonical Variate Analyses (CVA). Permutation analysis (10,000 permutations) for the mean Procrustes distance between groups as well as Procrustes ANOVA was used to assess significance of differences between groups. Since geometric morphometric techniques retain the intrinsic geometry in landmark coordinate data, shape variation along a given PC or CV axis can be displayed and visualized graphically. We compared groups using the set of age and centroid size regression residuals due to cleft status (cleft or control), cleft type (isolated CL or combined CL/P), bilateral versus unilateral, and cleft side (right versus left).

Pooled regression found age to explain 4.76% of the variation in the combined control and cleft sample and centroid size explains an additional 1.70% of the residual variation, indicating that facial shape does change with ontogeny and allometry and reinforcing the importance of controlling for age- and size- related facial variation within the two groups before further analysis (Figure 2). Procrustes analysis and other multivariate analyses were carried out after pooling the two groups together and regressing out the effects of age and centroid size on facial shape. Among the control group, the Procrustes distance permutation test showed that facial shape did not differ significantly between males and females (Procrustes distance = 0.0200, p = 0.1856). Among the CL/P group, the Procrustes distance permutation test showed that facial shape did not differ significantly between males and females (Procrustes distance = 0.0336, p = 0.5861). There is no evidence of sexual dimorphism in either group.

First we performed a test for asymmetry due to cleft side (right versus left cleft) for individuals using Procrustes ANOVA on the pre-mirrored coordinates. We found a significant effect of directional asymmetry in the dataset (p < 0.0001), likely resulting from a larger number of individuals with left-sided clefts than right in the sample. A CVA plot showed clear separation due to cleft side pre-mirroring (Procrustes Distance = 0.0866, permutation p < 0.0001). Post-mirroring, the variance is reduced along both CV1 and CV2, with no significant mean shape difference between left- and right-sided clefts (Procrustes Distance = 0.0458, permutation p = 0.1787) (Figure 3). All further analysis were therefore based on the mirrored coordinates.On a PC plot, there is separation between the control group and CL/P individuals (Figure 4A). Within the CL/P group, there is significant variation within and among the different types of clefts. In this combined analysis, PC1 explains 19.41% of the variance and is associated with facial prognathism, nasal and mouth width and facial height (Figure 4C). PC2, which explains 11.98% of facial shape variance, is associated with interorbital distance, mandibular prognathism and flattening of the nose. PCs 3 and 4 explain 8.01% and 5.82% of the variance, respectively. Facial shape changes associated with PC3 included mainly maxillary and mandibular prognathism and, interorbital distance. PC4 is associated with nasal and interorbital distance.Separation of the isolated CL and combined CL/P groups can be seen in the CVA results (Figure 4B). CV1 showed that, compared to the control group, the CL/P group exhibits increased nasal and mouth width, increased interorbital distance, slightly prognathic premaxillary region and decreased facial height (Figure 4D). Permutation tests showed that Procrustes distances comparing the consensus facial shapes of the control versus CL groups (Procrustes distance = 0.0794) and control versus CL/P groups (Procrustes distance = 0.0878) were significantly different (p < 0.0001 for both), indicating that there is a difference in mean facial shape between these groups. The isolated CL and combined CL/P groups (Procrustes distance = 0.0416) were not significantly different (p = 0.50). Procrustes ANOVA in the combined sample (CL, CL/P and control individuals) supported the CVA results, in that cleft status contributed significantly (p < 0.0001) to overall shape variation in the sample.

The first ten PCs accounted for 76% of total shape variance, with PC1 accounting for 17%. Variation along PC1 is associated primarily with lower nasal cavity width, lower facial height and mandibular and maxillary prognathism (Figure 5A). PC2 and PC3 account for approximately 12% and 10% of additional shape variance, respectively. PC2 is associated with interorbital distance, nasal width and maxillary prognathism. Shape changes associated with PC3 included mouth and nasal width and interorbital distance. PC4, which explained 8% of shape variance, described shape changes in nasal width, and positioning of the angle of the mouth.CVA showed some separation between cleft types, laterality and side (Figure 5B), although Procrustes distance permutation tests showed that facial shape did not vary between cleft types (isolated CL or combined CL/P) (Procrustes distance = 0.0416, p = 0.50), indicating there are no significant facial shape differences between individuals with isolated CL versus combined CL/P. Procrustes distance permutation tests within the CL/P group revealed a significant shape difference between unilateral clefts and bilateral clefts (Procrustes distance = 0.0728, p = 0.0001). Procrustes ANOVA of the CL/P individuals supported the permutation test results, showing that unilateral versus bilateral clefts were significantly (p = 0.0001) different in terms of shape variation in this sample. Comparison by cleft type (isolated CL versus combined CL/P) was again not significant (p = 0.79), indicating that combined cleft lip with cleft palate does not significantly differ in facial shape from isolated cleft lip in terms of surface facial morphology.

Based on the significant contribution to facial shape variation by unilateral or bilateral clefts, we performed a separate analysis of the unilateral cleft group, in which we mirrored the left sides so that the clefts always appear on the “right” (N = 36). Permutation tests found no significant differences were seen between the cleft types (Procrustes distance = 0.0566, p = 0.25). In this analysis the first ten PCs explained 79.53% of facial shape variance in the group, with PC1 and PC2 accounting for approximately 17% and 14% of shape variance respectively. Variation along PC1 is associated primarily with nasal and mouth width, lower facial height and interorbital distance. PC2 is associated with nasal width, and maxillary prognathism.