Quantifying Bone Strength Deficits in Young Adults Born Extremely Preterm or Extremely Low Birth Weight Open Access

ABSTRACT

The long‐term bone health of young adults born extremely preterm (EP; <28 weeks' gestation) or extremely low birth weight (ELBW; <1000 g birth weight) in the post‐surfactant era (since the early 1990s) is unclear. This study investigated their bone structure and estimated bone strength using peripheral quantitative computed tomography (pQCT)‐based finite element modeling (pQCT‐FEM). Results using this technique have been associated with bone fragility in several clinical settings. Participants comprised 161 EP/ELBW survivors (46.0% male) and 122 contemporaneous term‐born (44.3% male), normal birth weight controls born in Victoria, Australia, during 1991–1992. At age 25 years, participants underwent pQCT at 4% and 66% of tibia and radius length, which was analyzed using pQCT‐FEM. Groups were compared using linear regression and adjusted for height and weight. An interaction term between group and sex was added to assess group differences between sexes. Parameters measured included compressive stiffness (k_comp), torsional stiffness (k_torsion), and bending stiffness (k_bend). EP/ELBW survivors were shorter than the controls, but their weights were similar. Several unadjusted tibial pQCT‐FEM parameters were lower in the EP/ELBW group. Height‐ and weight‐adjusted k_torsion at 66% tibia remained lower in EP/ELBW (mean difference [95% confidence interval] −180 [−352, −8] Nm/deg). The evidence for group differences in k_torsion and k_bend at 66% tibia was stronger among males than females (p_interactions <0.05). There was little evidence for group differences in adjusted radial models. Lower height‐ and weight‐adjusted pQCT‐FEM measures in EP/ELBW compared with controls suggest a clinically relevant increase in predicted long‐term fracture risk in EP/ELBW survivors, particularly males. Future pQCT‐FEM studies should utilize the tibial pQCT images because of the greater variability in the radius possibly related to lower measurement precision. © 2023 The Authors. Journal of Bone and Mineral Research published by Wiley Periodicals LLC on behalf of American Society for Bone and Mineral Research (ASBMR).

Introduction

With advances in perinatal care, including the introduction of exogenous surfactant use in the early 1990s, more infants born extremely preterm (EP, less than 28 weeks' gestation) or extremely low birth weight (ELBW, birth weight <1000 g) are surviving,^(1,2) necessitating long‐term health studies in this group. The Victorian Infant Collaborative Study (VICS) Group was established to study the EP and/or ELBW population from infancy into adulthood.⁽³⁾ In Australia, approximately 1100 infants annually who are born either EP or ELBW today would be expected to survive into adulthood.⁽¹⁾

EP/ELBW survivors may be at later fracture risk. In the neonatal period, many of them have metabolic bone disease and suboptimal nutrition and receive postnatal corticosteroids to treat or prevent bronchopulmonary dysplasia, the lung disease of prematurity,⁽⁴⁾ which can impair bone formation and remodeling. The long‐term bone health of young adults born EP or ELBW has been unclear. Several studies concluded that EP/ELBW survivors had poorer bone health or higher fracture risk in adulthood,^(5‐14) whereas one study showed no substantial differences between the preterm and term participants,⁽¹⁵⁾ with mean age ranging from 17.2 to 35.5 years.^(9,11) However, there are limitations to the existing studies, including studying participants born before the introduction of surfactant in the early 1990s when survival rates of the tiniest and most immature infants were low,^{(5,7‐11,13,14)} small sample sizes,^(7,9,11,13) and sole use of dual‐energy X‐ray absorptiometry (DXA) to assess bone health.^{(5,6,8,10,11,14,15)} Few studies used peripheral quantitative computed tomography (pQCT), and with mixed results.^(7,9,12,13)

Although DXA is the clinical gold standard to diagnose osteopenia/osteoporosis, it scans the bone only in a single projection; it does not adequately account for variations in bone size and structure. For example, the areal bone mineral density (BMD) measured by DXA can be compromised by different bone sizes, geometry, and microarchitecture (compounded by soft tissue–related inaccuracies) and cannot be separated into cortical and trabecular BMD.^(7,16,17) Peripheral QCT is potentially superior in predicting fracture risk because it also analyzes the volumetric BMD and bone strength, considering the bone structure and rigidity at different sites.^(18‐20) Finite element modeling derived from pQCT images (pQCT‐FEM) goes further by considering the three‐dimensional stresses applied to the bone.⁽²¹⁾ This method has demonstrated the predicted fracture risk in experimental models.⁽²²⁾ It has also yielded better differentiation between fracture‐prone individuals and healthy comparator groups than what is found with standard densitometry techniques in several clinical settings, such as an older low‐trauma fracture cohort,^(21,23) women after surgical menopause,⁽²⁴⁾ and young women with type 1 diabetes.⁽²⁵⁾ However, validation of pQCT‐FEM via an independent analysis has not been available.

The aim of this study was to investigate the bone structure and estimated bone strength of young adults born EP/ELBW in the post‐surfactant era in the VICS study, using pQCT‐FEM. It was hypothesized that young adults born EP/ELBW would have lower bone stiffness and hence lower bone strength than the controls. We also performed an independent validation of pQCT‐FEM, aiming to demonstrate the predictive value of measured bone stiffness for bone strength and failure load.^(21,22)

Materials and Methods

Study design

Details of the VICS study protocol and inclusion criteria have been described elsewhere.⁽³⁾ Briefly, EP/ELBW survivors and contemporaneously term‐born, normal birth weight controls were recruited at birth in Victoria, Australia, over a 24‐month period from January 1, 1991, to December 31, 1992. The bone health of the participants was assessed at age 25 years, and some results of that evaluation have been published before this analysis.⁽¹²⁾ In this study, we report the pQCT‐FEM results; participants with missing pQCT results were excluded from the current analysis.

The Human Research Ethics Committees of the Royal Women's Hospital, Mercy Hospital for Women, Monash Children's Medical Centre, and the Royal Children's Hospital (Melbourne, Australia) approved this study. All participants gave written informed consent to participate in the study.

Data collection and storage

Perinatal and 25‐year data were stored in a password‐protected computer.⁽³⁾ Perinatal characteristics included maternal preeclampsia, use of antenatal corticosteroids, multiple birth, gestation at birth in completed weeks, birth weight, sex, necrotizing enterocolitis, use of postnatal corticosteroids, and use of supplemental oxygen at 36 weeks postmenstrual age.⁽²⁶⁾ Participant characteristics at 25 years included age at assessment, weight, height, and smoking status. At age 25 years, the participants underwent pQCT (slice thickness 2.0 mm, XCT 3000, Stratec Medizintechnik GmbH, Pforzheim, Germany). The axial images were available at four sites, measured proximally from the distal articular surface: distal tibia (4% site of the tibia length), tibia shaft (66% site of the tibia length), distal radius (4% site of the radius length), and radius shaft (66% of the radius length).

The images were analyzed with our established pQCT‐FEM algorithm that uses commercial engineering software (MATLAB; version R2020a Mathworks, Natick, MA, USA) to process images for analysis with commercial FEM software (Abaqus version 6.11, Simulia, Dassault Systemes, Providence, RI, USA).⁽²¹⁾ The algorithm allowed for the export and performance of manual segmentation on the pQCT cross sections, which generated meshes of 0.4 × 0.4 × 2 mm elements. It then re‐sliced these meshes in the z‐direction to produce a uniform mesh of 0.4 × 0.4 × 0.4 mm elements. Using an established equation,^(19,27) the Young's modulus was calculated and assigned to each element. The cortical and trabecular bones were not separated and were converted to modulus based on a single power law.^(28,29) At the 66% site, the intramedullary canal was not explicitly separated; rather its grayscale value was converted to density and the corresponding modulus was low and contributed a negligible amount to the bending or torsional stiffness (0.6% and 0.3%, respectively; Supplemental Fig. S5). The moduli were binned in increments of approximately 250 MPa to help alleviate some of the noise present in the pQCT images. The FEM was finally generated from each voxel mesh in Abaqus.

Because the mechanical behavior of bones may differ as a function of the three‐dimensional load components due to the variation in bone quantity, quality, and geometry impacting on bone strength,^(22,30) the bone structure in each image was originally tested with four separate idealized load cases: axial compression, shear, torsion, and bending (Supplemental Fig. S4).⁽²¹⁾ Axial compression was simulated by a 0.01 mm displacement of the superior surface toward the inferior surface. Shear was simulated by a 0.01 mm displacement of the superior surface in the direction of either the x‐ or y‐axis. Bending was simulated by a 0.0001 radian rotation of the inferior surface about either the x‐ or y‐axis (ie, cross‐sectional neutral axes). Torsion was simulated by a 0.0001 radian rotation of the inferior surface about the z‐axis. The stiffness of the bone was estimated by dividing the reaction forces and moments predicted from the simulations by the respective applied displacement or rotation. The variables measured for each axial image were compressive stiffness (k_comp), shear stiffness (k_shear), torsional stiffness (k_torsion), and bending stiffness (k_bend). In this analysis, we presented the results of k_comp at the 4% site along with k_torsion and k_bend at the 66% site of the bones because they were the variables most highly correlated with the whole‐bone strength for compression, torsion, and bending (see Table 4 and validation results below regarding shear and bending stiffness). Shear stiffness was not presented because this type of load would generate a bending failure mode that was already represented by k_bend. Because pQCT‐FEM is highly dependent on image quality, images with severe motion artifacts and images taken at an inaccurate bone site were excluded.

Statistical analysis

Data were analyzed using Stata Release 17 (StataCorp, College Station, TX, USA). Participant characteristics were compared between EP/ELBW and control participants using linear regression for continuous variables or logistic regression for categorical variables. Bone variables were compared between EP/ELBW and controls using linear regression models fitted with generalized estimating equations and reported with robust (sandwich) estimation of standard errors to account for lack of independence within multiple births from the same family. We reported results both unadjusted and adjusted for height and weight to evaluate the impact of body size on the variables. To determine whether between‐group differences varied between the sexes, we added an interaction term for group‐by‐sex; if there was evidence for an interaction (p < 0.05), we reported differences between EP/ELBW and control groups within each sex separately. We acknowledge the multiple comparisons and have interpreted our findings by focusing on the strength of the evidence for overall patterns and magnitude of differences, rather than on individual p values. With sample sizes of 162 EP/ELBW and 122 controls, we have 80% power to find a difference in means between the two groups as small as one‐third of a SD.

Validity of finite element analysis of 2‐mm slices in predicting full bone strength

To examine the validity of using thin cross sections to predict whole‐bone behavior, the strength predicted by FEM analysis of full tibias was compared with the pQCT‐FEM analysis from the 2‐mm slices of the respective tibias. The full methods are described in further detail in Supplemental Data S1. Briefly, computed tomography (CT) lower‐limb scans (slice thickness: 1.0 mm, in‐plane pixel size: 0.7–1.0 mm) of 12 tibias from 7 females were randomly selected from the New Mexico Decedent Imaging Dataset (Supplemental Table S1).⁽³¹⁾ The Hounsfield unit of each CT voxel was converted to apparent density using a phantomless calibration method,⁽³²⁾ then the elastic modulus was computed using the density‐modulus relationship established for the proximal tibia.⁽³³⁾

Each tibia was segmented using the software Mimics (version 21.0, Materialize, Leuven, Belgium), converted to a 3D surface, meshed with tetrahedral elements of maximum size 1.5 mm, then material properties assigned based on the modulus at each voxel (Supplemental Fig. S1). The meshes were imported into Abaqus and set up to apply displacement or rotation that created bending, shear, compression, and torsion loading (Supplemental Fig. S2). The applied displacement or rotation was increased until full‐bone failure was identified according to 10% of the surface elements exceeding a Coulomb–Mohr failure criterion.⁽²⁸⁾ The respective load at failure was taken as the full‐bone strength for compression, torsion, bending, or shear (S_comp, S_torsion, S_bend, and S_shear, respectively).

Using the whole tibia and patella geometry, the 2‐mm‐thick axial sections at the 4% and 66% sites were located and subsequently segmented using the Mimics software to obtain their respective 3D geometries. Each geometry was meshed with tetrahedral elements with a maximum size of 0.5 mm, then the material properties were assigned based on the derived elastic moduli at each CT voxel (Supplemental Fig. S3). Each mesh was imported into Abaqus, then using the same method outlined for the 2‐mm slices earlier, bending, shear, compression, and torsion simulations were performed from which k_bend, k_shear, k_comp, and k_torsion were, respectively, calculated (Supplemental Fig. S4). To determine the failure load for each loading type, the applied displacement or rotation was then increased until 2% of the volume had exceeded the energy‐equivalent strain criterion.⁽³⁴⁾ The respective load at failure was taken to represent S_comp, S_torsion, S_bend, and S_shear. The coefficient of determination (R²) was computed for linear regressions between each stiffness at the 4% and 66% sites and the full bone strengths. R² was also computed for linear regressions between each strength at the 4% and 66% sites and the full‐bone strengths.

Results

Participant characteristics

The original cohorts comprised 297 EP/ELBW consecutive survivors born in 1991–1992 and 253 controls. From the total of 291 participants assessed at 25 years of age, 3 participants declined to undergo pQCT and 5 participants were excluded because of severe motion artifact in their images or because of imaging taken at an inaccurate bone site. Overall, 283 participants were included in this analysis (Table 1): 161 EP/ELBW (54.2% of original EP/ELBW cohort; 46.0% male) and 122 controls (48.2% of original control cohort; 44.3% male). Perinatal details are shown in Table 1. The perinatal characteristics of participants with bone data at 25 years were similar to those who did not have bone data at 25 years, in both the EP/ELBW and control groups (Supplemental Table S7). At 25 years of age, the EP/ELBW group was shorter than the controls overall (mean difference [95% confidence interval], −6.0 [−8.2, −3.8] cm), and for the males and females separately. The weights were similar between the controls and EP/ELBW group, as was the sex distribution (Table 1).

pQCT‐FEM bone parameters between EP/ELBW and control groups

Descriptive values for each pQCT‐FEM variable are shown in Table 2. On unadjusted analyses, there was evidence that EP/ELBW survivors had lower bone stiffness at both sites of the tibia compared with the control group. In the radius, there was evidence that the EP/ELBW group had lower k_torsion and k_bend at 66% radius. When adjusted for height and weight, the evidence persisted that EP/ELBW survivors had lower k_torsion at 66% tibia.

pQCT‐FEM parameters by sex

At 66% tibia, k_torsion and k_bend showed significant group interactions with sex (Table 2). Separating the adjusted data by sex (Table 3), there was evidence that EP/ELBW males had lower k_torsion and k_bend at 66% tibia compared with the male controls, whereas there was little evidence for between‐group differences among females for these variables. The adjusted reductions among EP/ELBW males averaged 12% for k_torsion and k_bend at 66% tibia.

Validity of finite element analysis of 2‐mm slices in predicting full‐bone strength

The full‐tibia strengths and the stiffness and strengths of the 2‐mm slices are provided in Supplemental Tables S2–S4. The results of the linear regression between the FEM‐derived stiffnesses at the 4% and 66% sites and for the full tibias are provided in Table 4. S_comp for the full bone was most correlated with k_comp at the 4% site (R² = 0.93). S_torsion was most correlated with k_torsion at the 66% site (R² = 0.98), whereas S_bend and S_shear were both most correlated with k_bend at the 66% site (R² = 0.97 and R² = 0.98, respectively). Therefore, in this analysis, we presented the results of k_comp at the 4% site along with k_torsion and k_bend at the 66% site of the bones. Shear stiffness was not presented for the VICS study because this type of load would generate a bending failure mode that was already represented by k_bend.

The linear regression results between the FEM‐derived strengths at the 4% and 66% sites and for the full tibias are provided in Table 5. S_comp and S_torsion for the full bone were most correlated with S_comp and S_torsion at the 4% site (R² = 0.94 and R² = 0.97, respectively). S_bend and S_shear were both most correlated with S_bend at the 66% site (R² = 0.97 and R² = 0.98, respectively).

Discussion

In this study using finite element analysis to investigate bone health, all tibial bone stiffness variables were lower in EP/ELBW survivors than among controls, with k_torsion at 66% tibia remaining low even after adjusting for height and weight. Adjusted k_torsion and k_bend at 66% tibia were particularly lower in EP/ELBW males compared with control males. The differences were less apparent in female participants or in the radius, with only unadjusted k_torsion and k_bend at 66% radius being lower in EP/ELBW compared with their controls.

The current study used stiffness computed for 2‐mm axial slices at the 4% and 66% sites as a proxy for whole‐bone strength. This assumption was supported by FEM results of 12 tibias from The New Mexico Decedent Database which indicated that the whole‐bone compressive strength was most highly correlated with the k_comp at the 4% site (R² = 0.93), whereas the whole‐bone torsional and bending strength most highly correlated with k_torsion and k_bend at the 66% site (R² = 0.98 and R² = 0.97, respectively) (Table 4). It was also possible to compute the local strengths at the 4% and 66% sites assuming bone failure if 2% of the bone tissue volume exceeded an energy equivalent strain limit of 7000 μstrain;⁽³⁴⁾ however, these did not offer any considerable improvement in predicting whole‐bone strength (Tables 4 and 5). Therefore, only stiffnesses at the 4% and 66% sites were used in the current study because they make no assumption regarding a failure criterion and because they are consistent with those used in our previous studies.^(23‐25)

Overall, there is strong evidence to support our contention that localized stiffnesses of bone have predictive value regarding whole‐bone strength. We previously reported from cadaveric experiments on the radius that this approach gave good results where regressions between the whole‐bone strength and k_bend and k_torsion at the 4% site were 0.83 and 0.81, respectively.⁽²²⁾ Anex‐Bustillos and colleagues also computed the rigidity of 3‐mm‐thick axial slices of the femur using composite beam theory and compared these to the whole‐bone strength from mechanical testing.⁽³⁵⁾ The minimum compressive and bending rigidity from all slices of the femur correlated well with the whole‐bone strength (coefficient of determination R² = 0.82 and 0.86, respectively). Indeed, conventional finite element analysis using HR‐pQCT imaging assumes that the compressive stiffness or strength of a 9.02 mm distal section of bone is representative of the whole bone behavior.^(36‐38)

In the current study, stiffnesses for three different loading modes (compression, bending, and torsion) were used because these were considered predictive of bone strength under the respective loading types. This was considered valid because in vivo the tibia is exposed to a combination of loads, such as for walking, where the distal end is exposed mostly to compression, whereas more proximal regions are exposed to combined compression, bending and torsional loads.⁽³⁹⁾ These loading components are associated with their own tibial fracture types such as oblique and transverse fractures caused by bending,⁽⁴⁰⁾ spiral fractures from torsion,⁽⁴⁰⁾ or pilon fractures from compression.⁽⁴¹⁾ It is possible that a bone could be weaker in one type of load than others (Supplemental Fig. S6); hence, we included the aforementioned loading types as we considered they better represent bone strength holistically in its capability to resist all types of fractures. Notably, shear stiffness was not included in the current study because the extent of shear applied to the tibia is likely low in vivo⁽³⁹⁾ and because this type of load would generate a bending moment and the associated bending strength was already represented by k_bend.

Our findings suggest an increased long‐term osteoporotic fracture risk in young adults born EP/ELBW. Although the pQCT‐FEM model does not measure the entire bone, the stiffness estimated when tested with three‐dimensional stresses is a strong predictor for bone strength and failure load.^(21,22) All absolute pQCT‐FEM values of the tibia were lower in young adults born EP/ELBW than in controls, indicating lower strength and failure load and hence suggesting higher fragility fracture risk.^(21,22) The between‐group difference in k_torsion after adjusting for height and weight suggest that bone strength differences are not purely attributable to the size of the individual bone.

An increased predicted fracture risk in the preterm group compared with controls is consistent with other studies in the literature. Studies using DXA found an average deficit of 5% to 8.6% for BMD or bone mineral content of very preterm birth or very low birth weight compared with controls.^(6,13,14) Past studies using pQCT showed mixed results. Backström and colleagues found lower bone strength indices at both the tibia and radius (average deficits 11% to 16%) in preterm born young adults compared with controls,⁽⁷⁾ whereas the other two studies using pQCT did not report group differences.^(9,13) All three studies were in smaller cohorts and before surfactant was introduced. In our previous publication reporting DXA and pQCT results of this cohort, the average difference was 2% to 7% for femoral neck areal BMD measures and 12% for tibial polar stress strain index;⁽¹²⁾ the differences also remained after adjusting for height and weight. In the current study, the magnitude of deficits in unadjusted FEM stiffness variables of the EP/ELBW group compared with controls were 10% for k_comp and 17% to 18% for k_torsion and k_bend, which were largely similar if not greater degrees of deficit compared with the standard DXA and pQCT outcomes.^(7,12) Furthermore, in our earlier fracture cohort study,⁽²³⁾ the deficits for the same FEM variables in the patients with fracture were in the range of 5% to 14%. These comparative data lend further support to the likelihood that predicted long‐term fracture risk is appreciably increased in the EP/ELBW individuals.

There are several potential explanations for the greater deficits in the FEM stiffnesses. These measures use bone modulus calculated by a power law, which increases the weighting of high bone densities, whereas the calculation of polar stress strain index comprises a normalization term for the bone cross‐section size, which masks differences in bone properties attributed to variations in bone size. k_bend also accounts for the distribution of bone in its weakest direction, as compared to polar stress strain index, which accounts for bone distribution in all directions.⁽²¹⁾ Last, k_comp is computed at the 4% site and incorporates trabecular bone, which will lead to differences from the polar stress strain index computed at the 66% site using only cortical bone. Therefore, pQCT‐FEM may be a more robust predictor of fracture risk than standard DXA.

The apparent increased fracture risk is particularly evident in male rather than female EP/ELBW young adults. Sex‐specific differences were also described in previous studies using pQCT.^(7,12) In rats, differential effects of growth restriction due to experimental placental insufficiency on bone health in female versus male offspring were reported.⁽⁴²⁾ An underlying mechanism for this sex‐specific deficit is not ascertained.⁽⁷⁾ In one of our previous cohorts, there was no evidence of pubertal delay at 14 years of age in individuals born very low birth weight.⁽⁴³⁾ Because bone strength is impacted by both bone quantity and quality,⁽³⁰⁾ we hypothesize that these factors should be considered in seeking an explanation. Regarding bone quantity, there is evidence of lower total body bone mineral content and femoral neck and total hip areal BMD in EP/ELBW males compared with controls but not in the female counterparts.⁽¹²⁾ In terms of bone quality, there is also evidence of poorer bone microarchitecture measured by trabecular bone score using DXA in very low birth weight or small for gestational age males than in females.⁽⁸⁾ EP/ELBW males but not EP/ELBW females also demonstrate lower cortical thickness at 66% tibia in pQCT compared with controls.⁽¹²⁾ These reports suggest a complex mechanism involving both bone quantity and quality for sex‐specific deficits and warrant more research into possible explanations.

Among all radius bone variables, k_torsion and k_bend at 66% radius were lower in EP/ELBW survivors in the unadjusted model; however, these differences vanished when adjusted by height and weight and did not show sex interactions. Therefore, the unadjusted differences were presumably explained by the differences in bone size at the radius shaft. Also, we noted that average deficits of radial pQCT‐FEM variables between EP/ELBW and controls were 9% for k_comp at 4% site, and 11% for k_torsion and 15% for k_bend at 66% site, all of which were similar to or lower than the known precision errors of radial pQCT‐FEM variables (8.7% to 21.4%).⁽⁴⁴⁾ We proposed the minimal between‐group pQCT‐FEM differences in the radius, particularly at 4% site, were due to the greater variability in the radius possibly related to lower measurement precision. The radius precision may suffer from greater variability in anatomy, motion artifact, bone curvature, density, and repeatability of locating the articular surface reference point (Supplemental Fig. S7).^(45,46) The radius is also smaller in size, hence is more sensitive to misalignment of the sites. These factors may have reduced the study power of its stiffness,⁽⁴⁶⁾ and the random error due to lower precision may have influenced this subtle difference. The tibia has flatter distal articular surface, does not change shape as rapidly, and has lower precision errors (1.1% to 1.7%).⁽⁴⁴⁾ Our results indicate that differences in radius stiffness appear relatively subtle for EP/ELBW, at least when evaluated using site‐specific pQCT imaging, whereas the deficits at the tibia are pronounced, suggesting the lower limbs should be the preferred target for future pQCT‐FEM studies.

The current study's findings have important clinical and research implications. As young adulthood is the stage of development when peak bone mass is achieved,⁽⁴⁷⁾ individuals born EP/ELBW already with weaker bones may well face an increasing fragility fracture risk with aging. Although the incidences of fractures in EP/ELBW adults remains lacking in the literature, largely because few have survived into adulthood until recent decades, this predicted risk is important since up to 90% of EP/ELBW infants are now expected to survive into adulthood.⁽²⁾ This risk necessitates early awareness of interventions to prevent bone fragility. Conservative management should be advised with adequate calcium, vitamin D, and exercise.⁽⁴⁸⁾ There is space for research because the definition of osteoporosis, intervention threshold, and bone treatment in young adults in general are all unclear, and such evidence in EP/ELBW survivors is even scarcer.⁽⁴⁹⁾ Future research on adults born EP/ELBW should focus on the incidence of fragility fracture and the effectiveness of conservative and perhaps pharmacological bone treatment in this group. It is worth noting that EP/ELBW infants born today now receive regular breast milk fortified with minerals and preterm formula, compared with less optimal unfortified breast milk in the past.^(7,50,51) The current cohort were born in the era when breast milk was fortified and hence their persisting fracture risk despite improved nutrition indicates that further research is needed to evaluate the bone metabolism in these individuals.

This study has several strengths. To the best of our knowledge, no previous study has used finite element analysis to calculate bone stiffness in the EP/ELBW population. Peripheral QCT‐FEM has the advantage of improved sensitivity in detecting deficits in bone strength compared with DXA and standard pQCT.^(20,22,52) It also requires a lower radiation dose than that of QCT and high‐resolution pQCT^(21,53,54) and less time and cost to prepare than high‐resolution pQCT thanks to readily available software.^(21,22,55) Importantly, this study was also one of the few studies to investigate young adults born EP/ELBW in the post‐surfactant era, which provides updated evidence for modern EP/ELBW survivors. Limitations of our study include the attrition rate of the original sample. Also, 5 additional participants were excluded from the original publication because of their images having severe motion artifacts or being taken at inaccurate bone sites. Although the remaining 283 pQCT images underwent FEM analysis, there were no significant changes to their pQCT variables compared with the original 288 participant cohort (data not shown). Further, we had limited data on early bone health, nutrition through infancy and childhood, or physical activity with which to correlate our findings. At the follow‐up of this cohort at 25 years, we do not yet have data on fragility fracture prevalence and await later follow‐up to determine the clinical utility of our findings.

Another potential limitation is the relatively low signal to noise ratio in pQCT images that have been observed to contribute to uncertainty in bone density measurements with standard deviations of around 10%.⁽⁵⁶⁾ For the current study, binning of the bone into modulus increments of 250 MPa may have alleviated this noise slightly, particularly for cortical bone, which has exhibited greater reliability when using simple median filtering.⁽⁵⁶⁾ Furthermore, the mean errors in density measurements by pQCT are much lower,⁽⁵⁶⁾ and since the FEM stiffnesses are a function of the total bone density and its distribution across the entire cross section, it would be expected to be relatively insensitive to the imaging noise.

In conclusion, using finite element analysis of bone pQCT images of a cohort of young adults born EP/ELBW in the post‐surfactant era and of contemporaneous term‐born, normal birth weight controls, this study found between‐group differences in a number of bone structural variables in the EP/ELBW group. These differences suggest lower bone strength and increased predicted long‐term fragility fracture risk in the EP/ELBW group, particularly males. Further studies are needed to evaluate underlying mechanisms of the bone deficits, potentially remediable risk factors for such bone deficits, the incidence of fragility fractures, and prevention/treatment strategies in this group. Future pQCT‐FEM studies should utilize the tibial pQCT images due to the greater variability in the radius possibly related to lower measurement precision.

Acknowledgments

The project is supported by grants from the National Health and Medical Research Council of Australia (Centre of Clinical Research Excellence #546519; Centre of Research Excellence #1060733 and #1153176; Project Grant #108702; Leadership Fellowship #2016390 to JC) and the Victorian Government's Operational Infrastructure Support Program. The funding sources had no role in the study design; in the collection, analysis, and interpretation of data; in the writing of the report; and in the decision to submit the paper for publication. Open access publishing facilitated by The University of Melbourne, as part of the Wiley ‐ The University of Melbourne agreement via the Council of Australian University Librarians.

Author Contributions

Thang Dao: Conceptualization; methodology; writing – review and editing; writing – original draft; data curation; formal analysis. Dale Lee Robinson: Conceptualization; methodology; writing – review and editing; formal analysis; software. Lex W Doyle: Conceptualization; methodology; data curation; formal analysis; writing – review and editing. Peter VS Lee: Writing – review and editing; software. Joy Olsen: Formal analysis; writing – review and editing. Ashwini Kale: Data curation; writing – review and editing. Jeanie LY Cheong: Conceptualization; methodology; data curation; formal analysis; writing – review and editing. John D Wark: Conceptualization; methodology; formal analysis; writing – review and editing.

Peer Review

The peer review history for this article is available at https://www.webofscience.com/api/gateway/wos/peer-review/10.1002/jbmr.4926.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

Disclosures

The authors declare no conflicts of interest.

References

Appendix A

Members of the Victorian Infant Collaborative Study Group.

Convenor: Jeanie LY Cheong.^1,2,3,4 Collaborators (in alphabetical order): Peter Anderson,^2,4,5 Rosemarie Boland,^2,3 Alice Burnett,^2,4,6,7 Margaret Charlton,⁸ Marissa Clark,⁸ Noni Davis,⁴ Lex Doyle,^1,2,3,4,6 Julianne Duff,⁴ Leah Hickey,^2,6,7 Emily Johnston,⁵ Elisha Josev,^2,6,9 Katherine Lee,^6,10 Rheanna Mainzer,¹⁰ Marion McDonald,⁴ Dianne Malcolm,⁹ Bronwyn Novella,⁹ Kerry Ann O‐Connor,⁹ Joy Olsen,^2,4 Gillian Opie,^3,9 Lauren Pigdon,^2,4 Gehan Roberts,^2,4,6,11,12 Alicia Spittle,^1,2,13 Penelope Stevens,⁸ Alice Stewart,⁸ Anne‐Marie Turner,⁹ Tania Woods.^1,4

¹Neonatal Services, Royal Women's Hospital, Melbourne, Australia.

²Clinical Sciences, Murdoch Children's Research Institute, Melbourne, Australia.

³Department of Obstetrics & Gynecology, University of Melbourne, Melbourne, Australia.

⁴Premature Infant Follow‐Up Program, Royal Women's Hospital, Melbourne, Australia.

⁵Monash Institute of Cognitive and Clinical Neurosciences, Monash University, Melbourne, Australia.

⁶Department of Pediatrics, University of Melbourne, Melbourne, Australia.

⁷Department of Neonatal Medicine, Royal Children's Hospital, Melbourne, Australia.

⁸Department of Neonatology, Monash Medical Centre, Melbourne, Australia.

⁹Neonatal Services, Mercy Hospital for Women, Melbourne, Australia.

¹⁰Clinical Epidemiology and Biostatistics, Murdoch Children's Research Institute, Melbourne, Australia.

¹¹Center for Community and Child Health, Royal Children's Hospital, Melbourne, Australia.

¹²Population Health, Murdoch Children's Research Institute, Melbourne, Australia.

¹³Department of Physiotherapy, University of Melbourne, Melbourne, Australia.

The pQCT precision data used for FEM were produced by Dr Emma Callegari¹ and Dr Ashwini Kale.^1,2 The pQCT‐FEM precision data were originally presented by Dr Hongyuan Jiang.⁽⁴⁴⁾

¹Department of Medicine, Royal Melbourne Hospital, University of Melbourne, Melbourne, Australia.

²Bone and Mineral Medicine, Department of Diabetes and Endocrinology, The Royal Melbourne Hospital, Melbourne, Australia.

Author notes