Applicability of two commonly used bone age assessment methods to twenty-first century UK children

Bone age assessment plays an important role in clinical practice, permitting investigation of whether bone maturity is occurring at a rate consistent with chronological age (CA). In this context, bone age (BA) assessment is useful for managing children with skeletal dysplasias and endocrine disorders, as well as planning for orthopaedic procedures [1]. Approximately 160,000 unaccompanied children entered European countries during 2015 and 2016 [2]. Although there is no precise figure, numbers are significant and authorities have faced challenges in estimating some of their ages [3]. In these situations, CA has occasionally been deduced by comparing BA of the individual in question with the existing BA standards [4]. This practice is particularly common at geographical borders where conflicts or crises are occurring. Whether to aid clinical management of paediatric patients or to determine chronological age when this is unknown, it is crucial to have a reliable and appropriate method of determining bone age [5]. However, the European Society of Paediatric Radiology musculoskeletal task force has recently advised against the practice of estimating chronological age based on an assessment of bone age [6].

Numerous approaches have been developed to determine BA. Among these, two methods are widely utilised based on left hand and wrist radiographs, namely the Greulich and Pyle (G&P) and Tanner and Whitehouse (TW) methods [7, 8]. The G&P method is based on matching the child’s hand radiograph to standard plates provided by the G&P atlas; thus, this method compares the hand’s general maturational status. The population providing the G&P standard atlas were originally North American Caucasians of “good” socioeconomic status in 1938. The “good” socioeconomic status was designated because recruited children were above average both economically and educationally (they were also free of physical, mental, nutritional and environmental factors detrimental to growth) [9]. In contrast to the G&P atlas, the TW method undertakes an assessment and scoring of skeletal maturity for each individual hand and wrist bone. Data provided by the Harpenden Longitudinal Growth Study enabled the TW method’s development. In 2001, the TW3 method replaced the TW1 and TW2 methods as a result of documented secular change (as stated by the authors). The data that formed the TW3 method was collected from European and American Caucasian children of average socioeconomic status during the 1980s and 1990s [10]. Following the introduction of G&P and TW3 standards, numerous investigations have been undertaken internationally, in order to identify the extent to which these standards are relevant to various populations. This issue is significant, especially in light of the growing volume of studies concluding that certain methods are inappropriate for particular ethnic groups and as a result of improvements in socioeconomic status [11‐14].

BoneXpert software was developed in 2009, enabling automatic calculation of bone age, according to the G&P and TW3 standards [15]. The software provides standard deviation scores for each hand radiograph, thus assisting the comparison of a child’s bone age with healthy children of the same sex and age. There are several advantages in utilising this software tool, including eliminating observer variability and saving rating times.

This study aims to use BoneXpert to test the applicability of the G&P and TW3 methods to United Kingdom (UK) children born in the twenty-first century, whose standard of living (across all socioeconomic categories) is likely to be higher than that of the children used to develop the G&P and TW3 methods. Our hypothesis was that improved living standards and therefore improved nutrition would render their bone age advanced when compared with their chronological age [16].

In total, we identified 401 potentially eligible hand and wrist radiographs of which 9 were omitted due to BoneXpert failing to provide a reading for the following reasons as provided by the software: (1) “radiograph too sharp” in six images (this terminology is provided by the software for images with excessive edge enhancement or other post-processing), (2) poor image quality in two and (3) inconsistent lengths in one. Therefore, results are from 392 radiographs, comprising 206 males, 296 Caucasians, 71 Asians, 20 Africans and 5 mixed (Caucasian/Asian). Figure 1 illustrates the number of children per age and sex. In regard to socioeconomic status, 216 (55%), 74 (19%) and 102 (26%) children were of low, average and high socioeconomic status, respectively.

Concerning G&P, mean difference between BA and CA ranged from 33-month underestimation to 36-month overestimation in both females and males. Although differences were not statistically significant, G&P underestimated females’ ages by 1 month and overestimated males’ ages by 1.6 months (Table 1). BA was lower than CA in 51% of females and 44% of males, while being equal in 1% of males. With the cohort divided into yearly intervals, G&P overestimated females aged from 2 to 7 years by between 0.8 and 6 months, apart from at 4 years of age. This overestimation was statistically significant (p < 0.05) at age 6, in females (Table 2). After 7 years of age, G&P consistently underestimated females until 12 years of age by between 0.1 and 11 months, with underestimation being statistically significant (p < 0.05) at 12 years of age (Table 2). Subsequently, G&P overestimated females’ ages. Concerning males, G&P overestimated in all age groups apart from at 3, 6 and 12 years of age, with no statistical difference between BA and CA. ANOVA test showed no statistical difference between low, average and high socioeconomic status groups when using the G&P atlas for either females (p = 0.171) or males (p = 0.204). However, in females, the mean difference between BA and CA tended to be larger in low and average socioeconomic status groups, while in males, the difference tended to be larger within the higher socioeconomic status group.

Concerning TW3, overall mean difference between BA and CA showed a statistically significant difference in females but not in males. The mean difference between BA and CA ranged from 37-month underestimation to 32-month overestimation in both females and males. BA was lower than CA in 64.5% of females and 49.5% of males, while being equal in 0.5% of males. TW3 underestimated females’ ages by between 2 and 15 months (mean 5.2 months, p < 0.01) for all chronological age groups above 3 years (Table 3). TW3 significantly underestimated females at 8, 11, 12 and 15 years of age (p < 0.05). There was a statistically significant difference between the three socioeconomic groups as determined by one-way ANOVA (p = 0.019). Post hoc ANOVA showed that 17.8% of the variation between CA and TW3. BA as assessed by BoneXpert was accounted for by socioeconomic status. Observed differences were larger and significant (p < 0.001) in females of low and average socioeconomic status (Table 4). In males, TW3 underestimated age for those 10 years or above; this was statistically significant in Caucasians at ages 9, 12 and 13 years. There was no statistically significant difference between socioeconomic groups as determined by one-way ANOVA (p = 0.91). Distribution of the mean difference between CA and BA estimated via both G&P and TW3 methods for each sex is illustrated in Figs. 2 and 3.

Analysis of the Caucasian data showed no statistically significant difference when compared with the results from overall analysis, which included all ethnicities (Tables 1 and 2). In particular, the mean difference between CA and BA estimated by TW3 was statistically significant both in females of all ethnicities and in Caucasian females alone. An independent t test showed no significant difference between the mean difference of BA and CA when acquired from either the left hand or the right hand for both G&P (p = 0.58 females, p = 0.07 males) and TW3 (p = 0.08 females, p = 0.30 males) methods. Mean differences between BA and CA according to body side are illustrated in Table 5.

The G&P and TW3 methods showed comparable accuracy in females with the standard error of the estimate (SEE) of ± 1.05 and ± 1.06 years, respectively. Similar accuracy for the two methods was also observed in males with SEE of ± 1.10 and ± 1.00 years for G&P and TW3 respectively.

Several variables may affect the applicability of BA methods. One is socioeconomic status, which refers to a combination of environmental factors such as nutritional status, state of health and economical and social class of an individual. Being of “high” socioeconomic status infers improved access to healthcare, sufficient food, exercise and housing, allowing full growth potential to be achieved [20]. Studies have shown that high socioeconomic status is more likely to accelerate skeletal maturation rate [12]. This might be related to nutritional factors with over-nutrition leading to overweight/obesity, which in children has been linked to BA advancement [21, 22]. In contrast, individuals from low socioeconomic groups are more likely to have poor diets and lower weight and are more likely to experience growth retardation [23]. Bearing in mind that the TW2 method was updated because of perceived effects of secular change [8], whereas G&P has never been updated, we questioned the reliability of bone age assessment methods. We sought to analyse the reliability of the G&P and TW3 methods within the modern-day UK context.

Breaking the cohort into yearly intervals showed statistical significance for varying age groups in females and males, when using the G&P atlas. These differences (overestimation at age of 6 and underestimation at age of 12, in females) were still significant when only data from Caucasian children was analysed. In spite of these sub-group differences, there was no statistical difference between overall mean BA and overall mean CA in either males or females. To convey a comprehensive picture, we contrasted our findings—especially mean difference between BA and CA—with previous studies that focused on the Caucasian population (Supplementary Table 1). Some of these studies have concluded that Caucasian children mature skeletally at approximately the same rate as the G&P standard in males across all age groups [14, 24‐28]. However, other authors recommend that the G&P atlas be used with reservation due to mean BA being retarded in some age groups compared to the reference population [29‐32]. Common findings among these studies of the G&P atlas include underestimation of males aged below 13 years and overestimation during adolescence [30‐36]. G&P was applicable to females during adolescence while overestimation was reported before the age of 12 years [31, 32]. Others have recommended that a new standard altogether is required for precise bone age assessment, given the significant advancement of BA due to secular changes in skeletal maturation, which is thought to be due to improved standard of living [28, 30, 35, 36]. For example, Calfee et al reported that G&P overestimated males and females between 12 and 15 years old, for whom BA exceeded CA by at least 2 years [35]. All of these studies used the subjective assessment of experienced raters; our results using an objective software program indicate that overall, G&P currently remains applicable.

In contrast to the G&P atlas, we found that TW3 significantly underestimated females’ ages after 3 years of age. The mean difference between BA and CA was statistically significant in females, especially at the ages of 8, 11, 12 (Fig. 4) and 15 years, for all ethnicities and for Caucasians alone. In Caucasian males, the mean BA was significantly lower than CA at age of 9, 12 and 13 years.

A large number of children included in this study (55%) were of low socioeconomic status according to IMD and socioeconomic status explained 17.8% of the difference between bone age (TW3 method) and chronological age. Although there have been improvements in standard of living over the past decade [16] (expected to advance bone age), our results show delayed BA in girls when using the TW3 method. In line with our results, other studies have shown delayed BA compared with CA in females after the age of 10 years [14, 29, 37]. These results potentially support recent views of some researchers, who argue that the improved secular trend has eased or stopped [38, 39]. As a result of an improving secular trend in standard of living, the TW3 method was established in 2001 such that the TW3 BA is about a year ahead of the previous (TW2) method, especially after the age of 10 or 11 years [8]. Our results suggest that a return to TW2 may be necessary.

Several authors argue that socioeconomic status is the predominant reason behind the difference in skeletal maturational rates among populations [12, 14, 31]. Schmeling et al found that bone age was retarded among 27 studies that reported the socioeconomic status of their participants [12]. This retardation was due to the high socioeconomic status of the children recruited to develop the G&P atlas compared with the children within these studies, such that even the secular trend of increasing standard of living was not sufficient to eliminate any differences in socioeconomic status of the various cohorts [29].

In spite of the likely effects of socioeconomic status, the impact of ethnicity cannot be neglected. Studies on two different ethnic groups residing in the same region have shown that bone age assessment methods may reveal different results [24, 34]. Ontell et al showed that the G&P atlas is applicable to Caucasian girls at all ages but not to boys before the age of 13, while in Asians in the same region, the G&P atlas is applicable to girls at all ages but only to boys between 7 and 13.3 years. Zhang et al concluded that Asian children mature sooner than do Caucasian children, especially between 10 and 13 years of age in girls and between 11 and 15 years of age in boys. In a recent meta-analysis, bone age was significantly delayed in African females, while advanced in Asian males when compared with the G&P standard [40]. Furthermore, it has been shown that young Asian adults reach the end of maturity prior to the age observed through the TW3 method (25–27). Research focusing on South African individuals found that TW3 underestimated CA for boys but not for girls [41]. These variations within populations must be considered when assessing bone age [42]. In this current study, we demonstrated no significant difference between all ethnic groups compared with Caucasians alone; it should be noted that Asians and Africans made up only 20% and 5% of the study population respectively.

Measuring BA according to a subjective technique has a greater likelihood of introducing rating variations across analysts, due to varying degrees of expertise. However, this disadvantage was overcome in the current study through the use of BoneXpert, which is an automated bone age analysis software tool that eliminates observer variability and has the advantage of saving significant time. Our observed 5-month persistent discrepancy between chronological age and TW3 bone age as determined by BoneXpert in females appears to be a disadvantage not of the software per se, but of the reference standard (TW3) on which the software depends. Despite this, the software showed acceptable accuracy when using the G&P and TW3 methods for both sexes with the SEE being approximately ± 1 year.

The limitations of this study include the following:

Progress in medicine, education, industry and economic growth have all contributed to higher socioeconomic status which in turn is expected to have had a positive impact on children’s skeletal maturation [8, 24]. Our results show retardation of BA appears counterintuitive, but may not be if the socioeconomic status of the TW3 reference children was on average higher than that of the children we recruited and suggest that perhaps we should revert to the TW2 method.

Our results indicate that (1) secular change does not appear to have advanced skeletal maturity of UK children; (2) no significant difference exists between BoneXpert-derived BA and CA when using the G&P atlas; therefore, this method can be utilised for the modern UK population; and (3) BoneXpert-derived TW3 BA in current UK children is consistently below the CA of females by an average of 5 months; the clinical significance of this will have to be determined by the requesting clinician and will be greater in younger children who have a lower standard deviation. Developers of BoneXpert may wish to consider this in future upgrades of the software.