We recruited subjects for this case–control study from the patient population who underwent radiography of the forearm at Cincinnati Children’s Hospital Medical Center for a suspected forearm fracture. Both fractured and non-fractured subjects were recruited from this patient pool: all had sustained an injury to the forearm. Subjects were 5 to 16 years of age; there was no restriction regarding race/ethnicity. We excluded children with chronic diseases that affect bone, with injuries due to a motor vehicle accident, with bilateral forearm fractures or with a concurrent fracture at another skeletal site. Recruitment was stratified by sex and presence of a fracture. The goal was to have similar age distributions of forearm fracture cases and injured controls within each sex. Midway through the study we noted that fracture cases were younger than controls and had sustained injuries due to a fall rather than a collision. Thus, recruitment of controls was restricted to better match cases in terms of age and mechanism of injury (fall versus collision). The Institutional Review Board approved the study protocol, and the parent or guardian provided informed consent.
We recruited subjects as a forearm fracture case or injured control based on the initial radiography report. Subsequently, the study radiologist (TL) reread all radiographs to confirm group assignment. Criteria for classification as a forearm fracture were radiographically evident bony contour deformity of the radius or ulna on the initial imaging examination, or callus formation on a follow-up radiograph. Criteria for controls were no bony contour deformity on the initial radiograph, no callus formation on follow-up radiographs, clinical management consistent with no fracture (i.e., no cast), and no report of persistent pain at screening. All bony fractures were characterized according to bone involved (radius and/or ulna) and site (physis, metaphysis or metadiaphysis [end of the metaphysis abutting the diaphysis], or diaphysis).
PQCT scans of the non-injured forearm were obtained at enrolment with a Stratec XCT2000 6-detector scanner at sites 4% (metaphysis) and 20% (diaphysis) from the distal end of the radius. We performed a scout scan for scan positioning according to the landmarks described by Neu et al. [16
]. For subjects with an open physis, the reference line was placed at the proximal edge of the physis in the center of the radius. The reference line was placed at the ulnar edge of the articular surface if the physis was closed or closing (13% of girls, 8% of boys). All scans were obtained with a voxel size of 0.2 mm and speed of 25 m/s. Scans were analyzed using the manufacturer’s software version 5.50. Scans of the 4% site were analyzed with an iterative algorithm that identifies the steepest gradient on the cortical shell to identify the periosteal bone edge (Contour mode 2). This analysis provided measures of total or integral vBMD, bone mineral content (BMC), and cross-sectional area. The interior trabecular bone region was identified using a threshold of 400 mg/cm3
(Peel mode 1) allowing measurement of trabecular vBMD.
We identified the periosteal and endosteal bone edges of scans at the 20% site using two different analysis thresholds, 710 and 280 mg/cm3
(Cort mode 1), as there has been no consensus as to a single best threshold [17
]. The differences between cases and controls were similar regardless of threshold used (data not shown). We presented results using thresholds that theoretically best reflect the characteristic being measured. A threshold of 710 mg/cm3
was used to identify bone edges for measurement of cortical vBMD, cortical area, and periosteal and endosteal circumferences. A threshold of 280 mg/cm3
was used to identify bone edges for BMC and the strength–strain index (SSI) in order to capture the maximum amount of bone mineral. We used the manufacturer’s software “circular ring model” to calculate periosteal and endosteal circumferences, and cortical thickness. SSI (section modulus accounting for cortical density) also was calculated by the manufacture’s software as follows: SSI = Σvoxels
(A × d2
, where A
is the area of a voxel, d
is the distance of the voxel from the center, dmax
is the maximum distance of any of the voxels from the center, and vBMDvox
is the vBMD for each voxel.
All subjects had DXA scans (Hologic, QDR 4500, Bedford, MA, USA) of the total body, lumbar spine and proximal femur (hip). A DXA scan of the non-injured forearm was obtained on the latter half of subjects to enable comparison with forearm pQCT measures. We performed DXA scans according to Hologic guidelines, and analyzed DXA scans by software version 12.1 using the automatic low-density option. To prevent measurement artifacts due to the cast material, we excluded the scan region containing the injured arm and replicated the information from the non-injured arm to generate all total body results. Information on BMC, aBMD, and bone area were obtained for following regions of interest (ROI) lumbar spine, total hip, femoral neck, ultradistal radius (metaphysis), and 1/3 radius (diaphysis). The total body less head (TBLH; i.e., subcranial skeleton) ROI was used instead of the total body ROI because the head contains an inconsistent proportion of total body BMC relative to body size in children [19
]. Information on lean mass and fat mass was obtained from the total body scans.
Information about arm dominance and mechanism of injury was ascertained by questionnaire at enrolment. Subjects reported the type of activity in which they were engaged when injured and whether the injury was a result of a fall or a collision. We further classified mechanism of injury as low, moderate, or severe trauma according to criteria described by Landin [20
] and modified by Clark et al. [21
]. Low trauma: falls during informal play, ball sports, and gymnastics, from standing heights or less (<0.5 m), to a trampoline from ≤3 m; wrestling; being hit by a soccer or soft ball; and collisions with a stationary object while moving slowly. Moderate trauma: falls onto a non-resilient surface from 0.5 to 3 m, down stairs, from a moving bicycle, swing, slide, skateboard, rollerblades, or skis moving at moderate or fast speed; being hit by a bicycle; collisions between two moving objects; collision with a stationary object while moving at moderate or fast speed; and falls from standing height with another person landing on top of them. Severe trauma: falls from height >3 m, and being hit by a moving heavy object (>body weight).
Height and weight were measured in duplicate with children dressed in light clothing and without shoes. Height was measured with a stadiometer, and weight was measured with an electronic balance. Body mass index (BMI) was calculated as weight/height2
), and BMI Z-scores were calculated using the CDC 2000 growth reference [22
]. Subjects ≥8 years were given pictures illustrating the five stages of pubertal development, breast (girls) and gonadal (boys), according to criteria of Tanner [23
]. Subjects, with help of their parents, were asked to select the Tanner stage most similar to them. We assumed that subjects <8 years of age were Tanner stage 1 (pre-pubertal).
Statistical analyses were performed using JMP statistical software (version 5.1, SAS Institute, Cary, NC, USA). We conducted multiple regression analyses to determine whether bone measures differed between cases and controls. The dependent variables were the bone measures, and the primary independent variable was group (case vs. control). All bone measures, except cortical vBMD and cortical thickness, were log-transformed to remove heteroscedasticity and improve the fit of the regression models. Results were reverse transformed for presentation. To adjust for imbalances between groups and to reduce the residual error, we included the following covariates in all regression models: sex, age, black race, Tanner stage, height, and injury type (fall vs. collision) or severity. Regression models for forearm bone measures also included arm dominance and forearm length if the latter was a stronger predictor than height (based on model R2
) for that bone measure. Additional variables (height2
, Tanner stage-by-sex, and sex-by-fracture group interactions) were tested and kept in the model if p
<0.05. Given that cortical vBMD measurements are subject to “partial volume effect” especially when the cortical shell is thin, cortical thickness was included in the regression models for cortical vBMD to statistically adjust for this potential bias [24
]. We calculated the percentage difference between groups and the 95% confidence interval (95% CI) around the percentage difference [25
We calculated Z-scores for each bone measure to enable expression of that bone measure relative to the average value given the child’s age, sex, race, height, and Tanner stage in our study sample. Z-scores were calculated as the Studentized residual (mean=0, standard deviation=1) from the multiple regression models. We then performed logistic regression analyses, where fracture group was the dependent variable, to determine the odds ratio (OR) and 95% CI associated with a change in the Z-score for each bone measure. This method enabled us to compare the relative importance of each bone measure for predicting fracture risk in a standardized way. Covariates considered in the logistic regression models were injury type or severity and arm dominance. Receiver operating characteristic (ROC) curves were generated for each bone Z-score, and the area under the curve (AUC) was calculated to reflect the potential of that bone measure for identifying individuals at risk of fracture. ROC curves are commonly used to evaluate the performance of a diagnostic test. An AUC of 1 indicates a perfect test; an AUC of 0.5 indicates that the test is equivalent to chance.