DXA-derived hip shape is associated with hip fracture: a longitudinal study of 38,123 UK Biobank participants

22 Despite advancements in fracture prediction tools and osteoporosis management, hip fractures 23 remain a significant consequence of bone fragility, with a 22% one-year mortality. Hip 24 geometric measures (GMs) have been associated with fracture risk; however, their strong 25 correlation hinders the identification of independent influences, leaving their relative predictive 26 value unclear. Statistical shape modelling (SSM) provides a more holistic assessment of hip 27 shape compared to using pre-determined GMs. This study aimed to evaluate whether SSM-28 derived hip shape from dual -energy X-ray absorptiometry (DXA) scans can predict hip 29 fracture, independently of individual GMs. Previously, we applied SSM to left hip DXA images 30 in UK Biobank, a large prospective cohort with link ed hospital records , generating ten 31 orthogonal hip shape modes (HSMs) , that explained 86% of shape variance. Additionally, 32 femoral neck width (FNW), femoral head diameter (FHD), and hip axis length (HAL) were 33 derived from these DXAs . In the current analysis , Cox proportional hazard models , adjusted 34 for age, sex, height, weight, bone mineral density (BMD), and GMs (FNW, HAL, FHD), were 35 used to examine the longitudinal associations between each HSM and first incident hospital 36 diagnosed hip fracture. A Bonferroni adjusted p-value threshold (p<0.004) was used to account 37 for the 13 exposures. Among the 38,123 participants (mean age 63.7 years; 52% female; mean 38 follow-up 5 years), 133 (0.35%) experienced subsequent hip fracture. HSM2, characterised by 39 a narrower FNW, a higher neck shaft angle, and reduced acetabular coverage, showed a strong 40 association with hip fracture risk (HR 1.32, 95% CI 1.11-1.58, P 1.47×10-3), which persisted 41 after full adjustment (1.30, 1.09-1.55, 3.27×10 -3). There was no evidence for an association 42 with other HSMs. These findings suggest that DXA-derived hip shape is associated with hip 43 fracture risk independently of BMD and GMs. Incorporating global hip shape into fracture risk 44 assessment tools could enhance prediction accuracy and inform targeted interventions. 45 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 3 Lay summary 46 Despite improvements in hip fracture prevention , they remain a major problem, with 22% of 47 people dying within a year of sustaining one. This study looked at medical images from 38,123 48 individuals in UK Biobank to assess the shape of their hip using computer -aided statistical 49 techniques. The results indicate that a hip shape variation describing a narrower femoral neck 50 and a larger angle linking the neck and the femoral shaft is linked to fracture. This association 51 persisted after accounting for other known hip shape measures related to fracture risk . 52 Therefore, hip shape could help improve prediction and prevention of hip fractures. 53 Key words: epidemiology, hip morphology, hip fracture, statistical shape modelling, DXA 54 55 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 4

56 The annual number of hip fractures in the UK is projected to rise by 32% over the next 4 years1, 57 highlighting the need for accurate prediction of hip fracture risk. These fractures represent a 58 significant consequence of osteoporosis-related bone fragility and carry a one-year mortality 59 rate of 22%2. However, not all individuals who sustain a hip fracture meet the diagnostic criteria 60 for osteoporosis3, which is primarily based on bone mineral density (BMD) . Clinical risk 61 assessment tools such as FRAX®4 – widely used in over 100 international guidelines – and the 62 UK-specific Qfracture5, have been developed to better predict the risk of incident fractures, but 63 still lack optimal sensitivity 6,7. Consequently, incorporating additional factors not currently 64 considered in existing tools could help improve the accuracy of fracture risk prediction8. 65 Variation in hip shape is increasingly recognised as a contributor to hip fracture9,10, having also 66 been linked to osteoarthritis 11. Hip shape can be assessed through measuring individual 67 geometric measures (GMs), or by evaluating the overall shape. Common examples of GMs 68 include hip axis length (HAL), neck shaft angle (NSA), femoral neck width (FNW), and 69 femoral head diameter (FHD), which can all be derived from DXA scans , either manually or 70 using software such as Hip Structural Analysis12. Although evidence linking G Ms to fracture 71 risk is inconsistent, a recent meta -analysis found that increased HAL, NSA, and FNW are 72 associated with higher fracture risk, with pooled odds ratios (OR) of 1.53 , 1.47, and 2.68 73 respectively10. This did not account for factors such as age and sex. Nonetheless, the 74 International Society of Clinical Densitometry recommends using only HAL for assessing hip 75 fracture risk in females, and advises against us ing GMs to guide treatment decisions 12. 76 Moreover, the high correlation between GMs such as FNW and HAL 11, as well as the 77 correlation between GMs and body size 13, complicates the evaluation of their individual 78 contributions to hip fracture risk . In contrast , assessing hip shape as a whole , rather than 79 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 5 focusing on individual GMs, may provide a more comprehensive understanding of hip health 80 and fracture risk by accounting for overall morphology and the relationships between different 81 features14. 82 Statistical shape modelling (SSM), a computer-aided technique designed to capture the 83 statistical variability of shapes within a dataset15, can be used to provide a more holistic 84 measure of hip shape. SSM uses outline points derived from hip images and employs principal 85 component analysis (PCA) to produce orthogonal modes of shape variation, termed hip shape 86 modes (HSM)16, which each capture a different aspect of hip morphology. Although research 87 linking HSMs to hip fracture risk is limited, one study that applied SSM to radiographs found 88 that a HSM characterised by a longer femoral neck, smaller femoral head and a narrower FNW 89 was associated with a higher fracture risk (OR 2.4 8)8. Studies comparing SSM -derived 90 measures of hip shape to GMs in the context of hip fractures have been limited to small studies9, 91 which have been unable to show that SSM -derived hip fracture risk is independent of GMs. 92 This underscores the need for a comparative analysis to identify the most effective predictors 93 of hip fracture risk. In our recent work using UK Biobank (UKB) we developed a machine-94 learning algorithm that automatically plac es outline points o n high resolution hip DXA 95 images17, facilitating the generation of hip shape measures in large numbers. 96 In the present study, we aimed to establish whether SSM-derived hip shape, obtained using our 97 automated point placement method, is associated with hip fracture risk independently of 98 established risk factors and hip GMs, while also analysing potential sex differences within these 99 associations in the UKB cohort. 100 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 6

and materials 101 Population 102 UKB is a prospective cohort study that recruited ~500,000 males and females, aged 40-69 103 years, from 22 assessment centres across the UK between 2006 -201018. Baseline genetic and 104 phenotypic information was obtained through questionnaires, physical measurements and 105 biological samples. In 2014, UKB launche d the Image Enhancement study , which aims to 106 gather imaging data, including hip DXA scans, from 100,000 participants 19. For this study, 107 ~40,000 left-hip DXA images with outline points delineating the bone contour were available 108 (October 2023). This study is overseen by the UKB Ethics Advisory Committee, and ethical 109 approval was given by the National Information Governing Board for Health and Social Care 110 and North -West Multi -centre Research Ethic s committee (11/NW/0382). All participants 111 provided informed consent for their data to be used in the study. 112 Acquisition of DXA-derived hip shape 113 Hip DXA images were acquired following a standardised protocol using an iDXA scanner (GE-114 Lunar, Madison, WI, USA), with participants ’ legs positioned at an internal rotation of 15-115 25°19. A Random Forest-based machine learning algorithm20 (BoneFinder®, The University of 116 Manchester) had been previously used to automatically place the hip outline points 17. This 117 algorithm was initially trained on a subset of ~7,000 manually marked-up images before being 118 applied to the remaining ~33,000 images . A total of 85 outline points were placed around the 119 femoral head and acetabulum , including the greater and lesser trochanters (Figure 1) . The 120 placement of the outline points was manually verified, with only 10% requiring adjustment and 121 an average correction distance of 1.9 mm. 122 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 7 Once outlined, principal component analysis ( PCA) was performed to generate a set of 123 orthogonal hip shape modes ( HSMs), which collectively explain 100% of the variance 17. To 124 minimise the burden of multiple testing, this analysis focused on the first ten HSMs, which 125 accounted for 86% of the shape variance within the data set. Subsequent HSM s explained 126 minimal additional shape variance and are unlikely to hold clinical significance. Additionally, 127 FNW, FHD, and HAL were previously derived from the DXA scans using an openly available 128 custom Python script, as described elsewhere11,21. 129 Ascertainment of hip fracture 130 Hip fracture data w ere obtained through linkage to hospital episode statistics (HES), which 131 uses the International Classification of Diseases (ICD) 10th revision codes. Hip fractures were 132 identified based on the following codes: fractured neck of femur (S72.0), pertrochanteric 133 fracture (S72.1), subtrochanteric fracture (S72.2), stress fracture, not elsewhere classified 134 (Pelvic region and thigh) (M84.359), or pathological fracture, not elsewhere classified (Pelvic 135 region and thigh) (M84.459). Recording of HES data began on the 1st April 1997. Hip fracture 136 data were downloaded in August 2023, capturing information up until the end of October 2022. 137 Statistical analysis 138 Descriptive statistics, including means, standard deviations (SDs), and ranges, were used to 139 summarise population characteristics and the distribution s of HSMs and G Ms. Histograms 140 were plotted for each HSM to confirm normal distribution. The correlation between each HSM, 141 GM, BMD and demographic factors ( height, weight, age ), were assessed using Pearson 142 correlation co -efficient (r). Cox proportional hazard models were used to examine the 143 longitudinal associations between each HSM and hip fracture risk, as well as between each GM 144 (FNW, FHD, HAL), and hip fracture risk. The follow-up period concluded at the earliest event, 145 which was either the first incident hip fracture during follow-up, withdrawal, censoring due to 146 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 8 death or until the end of the study (31/10/2022). Individuals who had a hip fracture before 147 attending the imaging assessment, i.e. before the DXA scan, were excluded from the analysis. 148 The Cox proportional hazards assumption was tested using the Schoenfeld residuals approach. 149 A Bonferroni adjusted p -value threshold ( P<0.004) was used to account for the thirteen 150 exposures tested (ten HSMs and three G Ms). Results are shown as hazard ratios (HR), which 151 represent the relative risk of experiencing a hip fracture over time, with 95% CI and p-values. 152

are presented across four models: Model 1 is unadjusted; Model 2 adjusts for 153 demographic characteristics ( age, sex, height, and weight ); Model 3 additionally adjusts for 154 left hip femoral BMD; and Model 4 further adjusts for GMs (FNW, FHD, and HAL). When a 155 GM is the exposure, model 4 adjust s for the other two G Ms. Both combined -sex and sex -156 stratified analyses were conducted to account for known disparities in fracture risk 22 and hip 157 shape23 between males and females. All statistical analys es were performed using STATA 158 version 18 (Stata Corp, College Station, TX, USA). 159 Composite models 160 To investigate the overall at-risk hip shape for fracture a composite HSM figure was plotted by 161 combining all HSMs. Briefly, to do this, unadjusted beta coefficients for the associations 162 between HSMs and fracture were first computed. Each beta was then multiplied by 10 to 163 enhance the visualisation of shapes, and subsequently multiplied by the HSM -specific SD to 164 account for the contribution of each HSM to the overall shape variance. These adjusted values 165 were combined into a single vector to assess the collective impact of hip shape on hip fracture. 166 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 9

167 Baseline characteristics 168 This study included 38,128 UKB individuals with complete data and a left hip DXA image 169 available (Table 1). The mean age was 63.7 years, and 52% of participants were female. Mean 170 BMD of the left femur was 0.99 g/cm2, with females having a lower mean BMD (0.93 g/cm2) 171 compared to males (1.06 g/cm2). A total of 133 participants (0.35%) had a hip fracture, with a 172 higher prevalence among females (89 cases, 0.45%) compared with males (44 cases, 0.24%). 173 HSMs 1-10 had a mean value of 0 by design (Figure 2). Mean HSM values differed between 174 sexes, with the greatest difference seen in HSM1, HSM3, and HSM9. For the GMs, the 175 combined sex mean for FNW was 31.6 mm, FHD was 45.9 mm and HAL was 96.7 mm. Males 176 had a greater mean FNW, FHD, and HAL than females. 177 Geometric measures and their inter-relationships 178 FHD, FNW, and HAL were all highly correlated with each other (r 0.81-0.89) and with height 179 (r 0.75-0.81) (Supplementary Figure 1). Weight was moderately correlated with FHD, FNW, 180 HAL, height, and BMD (r 0.52-0.57). The HSMs were orthogonal by design . Similarly, no 181 correlation was observed between the HSMs and the other covariates. 182 Geometric measures and their association to hip fracture 183 Femoral Neck Width 184 In the unadjusted analysis of all participants (Figure 5, Supplementary Table 2), FNW was not 185 associated with hip fracture (Model 1: 1.15, 0.97 -1.36, 0.11). However, a strong association 186 was seen between a wider FNW and hip fracture following adjustment for demographic 187 characteristics and BMD (Model 3: 1.77, 1.30 -2.43, 3.26×10 -4). In sex -stratified analysis 188 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 10 (Supplementary Table 2), a wider FNW showed a strong association with hip fracture in both 189 sexes, both in the unadjusted model and following adjustment for demographic characteristics. 190 In males, the strongest association was observed in the unadjusted model (Model 1: 2.17, 1.44-191 3.25, 1.99×10-4). The association weakened with adjustment for BMD ( Model 3: 1.75, 1,08 -192 2.82, 0.02). A similar trend was noted in females, with the strongest association being in the 193 unadjusted model (Model 1: 2.88, 2.05-4.06, 1.40×10-9). Further adjustment for BMD resulted 194 in attenuation (Model 3: 1.70, 1.11 -2.59, 0.01). The effect sizes were greater in females 195 compared to males in models 1 and 2, with a similar effect size seen in both sexes in model 3. 196 Femoral Head Diameter 197 In the unadjusted analysis of all participants , t here was little evidence for an observed 198 association between FH D and hip fracture (Model 1: 1.12, 0.95 -1.53, 0.17) (Figure 5, 199 Supplementary Table 2). However, a strong positive association was seen when adjus ting for 200 BMD (Model 3: 1.89, 1.39-2.57, 4.48×10-5). In the unadjusted sex-stratified analysis, a larger 201 FHD demonstrated a greater effect size in females compared with males (Model 1: females - 202 2.43, 1.73-4.30, 2.60×10 -7; males - 2.30, 1.54-3.44, 4.50×10 -5). When adjusting for BMD, a 203 larger effect size was seen in males compared with females (Model 3: males - 2.01, 1.28-3.14, 204 2.26×10-3; females - 1.70, 1.11-2.60, 0.01). 205 Hip Axis Length 206 Similar to FNW and FH D, HAL did not show an association with hip fracture in unadjusted 207 analysis of all participants (Model 1: 1.08, 0.91-1.28, 0.39) (Figure 5, Supplementary Table 2). 208 However, an increased HAL was associated with hip fracture after adjusting for BMD (Model 209 3: 1.61, 1.18 -2.21, 3.08×10-3). When compared with FNW and FH D, HAL exhibited the 210 smallest effect size across all models. A strong positive association was seen only in the 211 unadjusted sex-stratified analysis (Model 1: males - 2.07, 1.37-3.11, 4.84×10-4; females - 2.05, 212 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 11 1.48-2.85, 1.58×10 -5) (Supplementary Table 2), with the associations seen diminishing after 213 further adjustment for BMD (Model 3: males – 1.84, 1.11-3.04, 0.02; females - 1.40, 0.92-2.13, 214 0.11). 215 Mutual Adjustment 216 When e ach G M was mutually adjusted for the other two G Ms, along with demographic 217 characteristics and BMD, there was less evidence for an association with hip fracture in both 218 combined and sex-stratified analysis. All results fell below the statistical significance threshold 219 for multiple testing (Figure 5, Supplementary Table 2). 220 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 12 Association between HSMs and hip fracture 221 Each HSM was initially assessed for its association with hip fracture. In the unadjusted 222 combined-sex analysis, there was evidence of a strong positive association between HSM2 and 223 hip fracture ( Model 1: HR 1.32, 95% CI 1.11 -1.56, P= 1.47×10-3) (Figure 4 , Table 2). This 224 association persisted upon adjustment for demographic characteristics and BMD ( Model 3: 225 1.31, 1.11-1.55, 1.51×10-3). HSM2 captures features of a narrower FNW, a higher NSA, and 226 reduced acetabular coverage (Figure 2). No other HSMs were found to be associated with hip 227 fracture in combined-sex analysis. 228 In female sex-stratified analysis (Supplementary Table 1), HSM2 showed a positive association 229 with hip fracture when adjusted for demographic characteristics (Model 2: 1.37, 1.11 -1.68, 230 2.79×10-3). Apart from this, sex -stratified analyses failed to show statistical evidence for an 231 association with hip fracture potentially because they were underpowered. 232 To evaluate the association between each HSM and hip fracture risk, independent of the hip 233 shape components captured by GMs, each HSM was further adjusted for all three GMs (FNW, 234 FHD, HAL) (Figure 4, Table 2). Analysis of all participants showed that the associations seen 235 in Models 1, 2, and 3 were maintained after adjusting for demographic characteristics, BMD, 236 and GMs. HSM2 emerged as the only HSM to show strong evidence of an association with hip 237 fracture in this model (Model 4: 1.30, 1.09 -1.55, 3.27×10 -3). In sex -stratified analysis 238 (Supplementary Table 1) , none of the associations met the Bonferroni -adjusted p -value 239 threshold. However, HSM2 showed weak evidence of an association with hip fracture when 240 fully adjusted in both females and males (Model 4: females - 1.27, 1.03-1.57, 0.02; males - 241 1.34, 0.96-1.84, 0.06). Additionally, in males, HSM9 continued to show weak evidence of a 242 negative association with hip fractures after full adjustment (Model 4: 0.66, 0.48-0.89, 0.01). 243 No other HSM was associated with hip fracture when fully adjusted in either sex. 244 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 13 Composite model 245 The composite model (Figure 3) showed that the overall at-risk shape, which is represented by 246 the solid line, included a narrower FNW, reduced acetabular coverage, smaller greater 247 trochanters, and a smaller FH D. This closely reflects HSM2 , which shares these shape 248 characteristics. 249 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 14

250 This large, longitudinal cohort study explored the relationship between DXA-derived HSMs 251 and GMs with hip fracture risk. The findings indicate that HSM2, characterised by a narrower 252 FNW, higher NSA, smaller femoral head, and reduced acetabular coverage, was positively 253 associated with hip fractur e risk, even after adjusting for age, sex, height, weight and BMD. 254 While GMs (FNW, FHD, HAL) also showed associations with hip fracture when adjusted for 255 the same covariates, these relationships attenuated upon mutual adjustmen t, confirming their 256 inter-relatedness. In contrast, HSM2 retained its association with hip fracture after accounting 257 for GMs, suggesting that HSM2 captures additional information beyond these three measures 258 of hip geometry. 259 Currently, t here are few comparative studies in the literature that have investigated the 260 association between SSM-derived hip shape and hip fractures. Furthermore, these studies have 261 derived their SSM from different populations , meaning it is not possible to draw direct 262 comparisons between specific HSMs. For instance, Gregory et al. applied a SSM consisting of 263 29 points outlining the femoral head and neck to standard radiographs in a small group of 264 females (26 cases, 24 controls) 9. They found that SSM -derived hip shape predicted fracture 265 risk after adjusting for height and weight. Specifically, a HSM with a longer, narrower femoral 266 neck and a higher NSA was more likely to fracture, reflecting the at -risk shape identified in 267 this study. However, their sample size was considerably smaller than that of our current study 268 and the outline points on the radiographs used to perform SSM did not include the lesser 269 trochanter. Baker-LePain et al. used a similar approach in a nested case-control study involving 270 Caucasian females (168 cases, 231 controls)8. They employed a larger number of outline points 271 (n=60) than Gregory et al. (n=29), and their model included the lesser trochanter. They found 272 that hips exhibiting extreme values of HSM4, characterised by a narrower FNW, increased 273 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 15 femoral neck length, and a smaller femoral head, were associated with hip fractur e. These 274 features closely resemble the at-risk hip shape identified in this study (narrower FNW and 275 smaller femoral head). Although Baker-LePain et al. adjusted for age, body mass index , and 276 femoral neck BMD , they, like Gregory et al., only included females within their analys es, 277 leaving it unclear whether the observed relationships are sex -specific. Goodyear et al. 278 performed SSM using 72 outline points on DXA scans of females aged over 75 years (182 279 subjects, 364 controls)24. The authors identified a hip shape associated with fracture that also 280 closely resembles the findings of our study, including a narrower FNW, greater NSA, reduced 281 acetabular coverage, and smaller greater trochanters. This study offers the closest comparison 282 to the present analysis as it was performed on DXA scans and used similar outline points , 283 including the acetabulum and lesser trochanter . However, the sample size was smaller, and 284 analysis focused on females only. This limitation is significant because HSMs are known to be 285 influenced by sex 25, and our study found notable differences in HSMs between the sexes. 286 Furthermore, none of the studies adjusted for GMs. 287 Although the at-risk hip shape (HSM2) identified in this study was characterised by a narrower 288 FNW, the analysis of GMs and hip fracture revealed that a wider FNW was associated with hip 289 fracture ( Figure 5, S upplementary Table 2 ). This finding has been reported in other 290 observational studies26-28, including a recent genetic analysis29 that found that individuals with 291 a genetic predisposition to a greater FNW were at higher risk of fracture . The seemingly 292 contradictory findings between HSM and G Ms regarding FNW and fracture risk may be 293 attributable to several factors. GMs objectively quantify individual aspects of hip morphology, 294 meaning that bone size can impact the magnitude of the measurement. For example, larger 295 individuals are likely to have a bigger femur across all dimensions; thus, a taller and heavier 296 person would be expected to have a larger FNW and HAL. Moreover, FNW is highly correlated 297 with height and moderately correlated with weight (Supplementary Figure 1 ), a relationship 298 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 16 that has been consistently reported in other studies11,30, highlighting the significant influence 299 of demographic characteristics, such as height and weight, on FNW. In contrast, SSM employs 300 Procrustes analysis to align and scale hip outlines based on shape, effectively capturing bone 301 morphology while excluding the influence of individual size. This ability to isolate shape from 302 size is important because HSM2 remained associated with hip fracture risk , independent of 303 FNW, FHD, and HAL . This suggests that these individual measures are not independently 304 driving hip fracture risk. Instead, the interactions and combined influence of these factors, 305 effectively captured by SSM, likely contribute to fracture risk . Ratios of GMs have been 306 suggested as an alternative, aiming to reduce the influence of correlation by standardising 307 measures against one GM13. However, SSM still outperformed ratio values in a previous small 308 study9. 309 Previous research has explored sex differences in hip shape23,25,31,32; few studies have examined 310 these differences within the context of hip fractures. HSM2 showed similar effect sizes between 311 the sexes, but a notable difference was seen with HSM9 (Supplementary Table 1). Studies of 312 individual hip shape measures have shown that females tend to have a smaller FHD, narrower 313 FNW, and shorter femoral neck length compared to males32, which likely reflects that females 314 are typically smaller than males. Similarly, Frysz et al. found sex differences in HSMs, with 315 females exhibiting a narrower FNW and smaller lesser trochanter compared with males25. This 316 finding is noteworthy, particularly given the weak evidence of a negative association with hip 317 fracture seen with HSM9 in males (Supplementary Table 1). HSM9 was characterized by a 318 larger lesser trochanter but a narrower femoral neck, suggesting that a larger lesser trochanter, 319 a feature more common in male hip shapes , could offer some protective effect against hip 320 fracture. Since the lesser trochanter serves as the insertion point for hip flexor muscles33 its size 321 could be indicative of muscle mass. Given that sarcopenia (loss of muscle mass and 322 function)34,35 is a known risk factor for hip fracture36-39, a larger lesser trochanter may represent 323 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 17 a proxy for muscle strength and function , potentially reducing fracture risk in males . 324 Additionally, innate female hip shape characteristics may predispose females to a higher 325 fracture risk, as they often exhibit features linked to fractures, such as a narrower FNW. 326 Interestingly, although HSM9 included a narrower FNW, similar to the fracture-prone HSM2, 327 this reinforces the idea that fracture risk is influenced by multiple interacting shape constituents 328 rather than any single measurement. 329 This study has several key strengths. Its large sample size and population-based design greatly 330 enhances the representativeness of the findings, thereby improving the reliability of effect 331 estimates. The study also simultaneously examined the relationship between SSM-derived hip 332 shape and GMs with hip fracture, allowing for a direct comparison of these two methods and 333 an evaluation of their independent associations with fracture risk . One of the limitations of 334 SSM is that each study uses a different population to derive their HSMs, thus you cannot 335 compare across models. This UK Biobank model could provide a reference for other 336 populations. Both the SSM -derived hip shape and the G Ms were semi-automatically derived 337 from DXA scans, requiring minimal manual point correction. Given the widespread use of 338 DXA scans in clinical practice for assessing osteoporosis, this approach makes accommodating 339 SSM-derived hip shape measures through tools like FRAX a feasible option. Additionally, the 340 inclusion of both combined and sex-stratified analyses represents a significant strength of this 341 study. While many studies primarily examine post-menopausal females, this study also 342 included m ale participants , providing valuable insights into male hip shape and its role in 343 fracture risk. 344 There are limitations to this study. As an observational study, it cannot establish causality. 345 Further research to understand the factors driving the association between HSM2 and hip 346 fracture risk is needed, although a recent study using genetic data found evidence of a causal 347 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 18 association between HSM2 and hip fracture in the same population40. NSA could not be derived 348 from the DXA scans due to the limited view of the femoral shaft. Given that prior studies have 349 linked higher NSA to hip fracture , and HSM2 represents a higher NSA, we were unable to 350 determine if the association between HSM2 and hip fracture was independent of NSA10,41. The 351 predominantly Caucasian study population may limit the generalisability of the findings . 352 Notably, differences in hip shape have been reported between the UK Biobank cohort and the 353 exclusively Chinese Shanghai Changfeng cohort 42. The mean age of participants (63.7 years) 354 may have reduced the study’s power, as hip fractures predominantly occur in older 355 individuals43. However, as participants continue to be followed-up and additional DXA images 356 from UKB become available, analysis can be repeated with more hip fracture cases, potentially 357 strengthening findings. Since the analysis focused only on left hip DXA scans, and the side of 358 the body the hip fracture occurred on is unknown, it is plausible that effect estimates could be 359 biased towards the null. As a result, the true effect of hip shape on fracture risk may be 360 underestimated or not fully captured in the analysis. Furthermore, using 2-dimensional DXA 361 scans to assess the shape of a 3-dimentional structure may result in the loss of spatial detail ; 362 however, SSM can help mitigate these limitations by using proportional rather than absolute 363 values of hip shape as described by GMs9. 364 In conclusion, this study examined SSM -derived hip shape using high -resolution DXA scans 365 from a large cohort of UKB participants, demonstrating risk of incident hip fracture is higher 366 with a narrower femoral neck, a higher femoral neck angle and reduced acetabular coverage. 367 Importantly these associations were independent of a wide range of covariates including 368 established measures of femoral geometry. Given that DXA scans are already routinely used to 369 assess osteoporosis risk, it is conceivable that SSM -derived measures of hip shape could be 370 accommodated into existing fracture risk tools such as FRAX® to improve prediction. This 371 approach could facilitate targeted preventative treatments for individuals with hip shapes 372 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 19 resembling HSM2, thereby reducing the risk of hip fractures and alleviating the resultant 373 morbidity and mortality. 374 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 20

This study has us ed the UK Biobank resource, access application 17295. The authors would like to thank Dr Monika Frysz, who was instrumental in deriving the hip shape modes in UK Biobank. For the purpose of Open Access, the author has applied a Creative Commons Attribution (CC BY) licence to any Author Accepted Manuscript version arising from this submission. NCH is supported by the UK Medical Research Council (MRC) [MC_PC_21003; MC_PC_21001] and National Institute for Health Research (NIHR) Southampton Biomedical Research Centre, University of Southampton, and University Hosp ital Southampton NHS Foundation Trust, UK. Author contributions Contribution to study conception and design: SS, JHT, BGF, RAB Contribution to acquisition of data: SS, AH, RE, FRS, JSG, RMA, CL, TC, NCH, JHT, RAB, BGF Contribution to analysis and interpretation of data: SS, AH, RE, FRS, JSG, RMA, CL, TC, NCH, JHT, RAB, BGF Drafting the article: SS, BGF, RAB Reviewing the final manuscript: SS, AH, RE, FRS, JSG, RMA, CL, TC, NCH, JHT, RAB, BGF Conflict of interest statement The authors have no conflicts of interest to disclose. . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 21 Data access statement All data variables are available from UK Biobank. The BoneFinder® search model and the SSM can be requested via the BoneFinder® website for independent validation: https://bone- finder.com/ Ethics statement This study is overseen by the UKB Ethics Advisory Committee, and ethical approval was given by the National Information Governing Board for Health and Social Care and North -West Multi-centre Research Ethic s committee (11/NW/0382). All participants provided informed consent for their data to be used in the study. Funding statement SS and AH were self-funded undergraduate students. BGF is supported by an NIHR Academic Clinical Lectureship and an Academy of Medical Sciences Starter Grant (SGL030\1057). RB, RE, FS and MJ were supported by a Wellcome Trust collaborative award (209233/Z/17/Z). CL is funded by a Sir Henry Dale Fellowship jointly funded by the Wellcome Trust and the Royal Society (223267/Z/21/Z). For the purposes of open access, the authors have applied a CC BY public copyright licence to any Author Accepted Manuscript version arising from this submission. . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 22

1. Harris E, Clement N, MacLullich A, Farrow L. The impact of an ageing population on future increases in hip fracture burden. Bone Joint J. Jan 1 2024;106-b(1):62-68. doi:10.1302/0301-620x.106b1.bjj-2023-0740.r1 2. Downey C, Kelly M, Quinlan JF. Changing trends in the mortality rate at 1-year post hip fracture - a systematic review. World J Orthop. Mar 18 2019;10(3):166-175. doi:10.5312/wjo.v10.i3.166 3. Schuit SCE, van der Klift M, Weel AEAM, et al. Fracture incidence and association with bone mineral density in elderly men and women: the Rotterdam Study. Bone. 2004/01/01/ 2004;34(1):195-202. doi:10.1016/j.bone.2003.10.001 4. Kanis JA, Harvey NC, Cooper C, Johansson H, Odén A, McCloskey EV . A systematic review of intervention thresholds based on FRAX : A report prepared for the National Osteoporosis Guideline Group and the International Osteoporosis Foundation. Arch Osteoporos. Dec 2016;11(1):25. doi:10.1007/s11657-016-0278-z 5. Hippisley-Cox J, Coupland C, Brindle P. The performance of seven QPrediction risk scores in an independent external sample of patients from general practice: a validation study. BMJ Open. Aug 28 2014;4(8):e005809. doi:10.1136/bmjopen-2014-005809 6. Gregson CL. CFracture, an alternative to QFracture that accounts for mortality to better predict fragility fracture risk. The Lancet Healthy Longevity. 2023;4(1):e6-e7. doi:10.1016/S2666-7568(22)00293-8 7. Kanis JA, Compston J, Cooper C, et al. SIGN Guidelines for Scotland: BMD Versus FRAX Versus QFracture. Calcified Tissue International. 2016/05/01 2016;98(5):417-425. doi:10.1007/s00223-015-0092-4 8. Baker-LePain JC, Luker KR, Lynch JA, Parimi N, Nevitt MC, Lane NE. Active shape modeling of the hip in the prediction of incident hip fracture. J Bone Miner Res. Mar 2011;26(3):468-74. doi:10.1002/jbmr.254 9. Gregory JS, Testi D, Stewart A, Undrill PE, Reid DM, Aspden RM. A method for assessment of the shape of the proximal femur and its relationship to osteoporotic hip fracture. Osteoporos Int. Jan 2004;15(1):5-11. doi:10.1007/s00198-003-1451-y 10. Fajar JK, Taufan T, Syarif M, Azharuddin A. Hip geometry and femoral neck fractures: A meta-analysis. J Orthop Translat. Apr 2018;13:1-6. doi:10.1016/j.jot.2017.12.002 11. Heppenstall SV , Ebsim R, Saunders FR, et al. Hip geometric parameters are associated with radiographic and clinical hip osteoarthritis: Findings from a cross-sectional study in UK Biobank. Osteoarthritis Cartilage. Dec 2023;31(12):1627-1635. doi:10.1016/j.joca.2023.09.001 12. Broy SB, Cauley JA, Lewiecki ME, Schousboe JT, Shepherd JA, Leslie WD. Fracture Risk Prediction by Non-BMD DXA Measures: the 2015 ISCD Official Positions Part 1: Hip Geometry. Journal of Clinical Densitometry. 2015/07/01/ 2015;18(3):287-308. doi:10.1016/j.jocd.2015.06.005 13. Gregory JS, Aspden RM. Femoral geometry as a risk factor for osteoporotic hip fracture in men and women. Medical Engineering & Physics. 2008/12/01/ 2008;30(10):1275- 1286. doi:10.1016/j.medengphy.2008.09.002 14. Bredbenner TL, Mason RL, Havill LM, Orwoll ES, Nicolella DP, Study ftOFiM. Fracture Risk Predictions Based on Statistical Shape and Density Modeling of the Proximal Femur. Journal of Bone and Mineral Research. 2014;29(9):2090-2100. doi:10.1002/jbmr.2241 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 23 15. Ambellan F, Lamecker H, von Tycowicz C, Zachow S. Statistical Shape Models: Understanding and Mastering Variation in Anatomy. Adv Exp Med Biol. 2019;1156:67-84. doi:10.1007/978-3-030-19385-0_5 16. Johnson LG, Bortolussi-Courval S, Chehil A, et al. Application of statistical shape modeling to the human hip joint: a scoping review. JBI Evid Synth. Mar 1 2023;21(3):533- 583. doi:10.11124/jbies-22-00175 17. Frysz M, Faber BG, Ebsim R, et al. Machine Learning-Derived Acetabular Dysplasia and Cam Morphology Are Features of Severe Hip Osteoarthritis: Findings From UK Biobank. J Bone Miner Res. Sep 2022;37(9):1720-1732. doi:10.1002/jbmr.4649 18. Bycroft C, Freeman C, Petkova D, et al. The UK Biobank resource with deep phenotyping and genomic data. Nature. 2018/10/01 2018;562(7726):203-209. doi:10.1038/s41586-018-0579-z 19. Littlejohns TJ, Holliday J, Gibson LM, et al. The UK Biobank imaging enhancement of 100,000 participants: rationale, data collection, management and future directions. Nat Commun. May 26 2020;11(1):2624. doi:10.1038/s41467-020-15948-9 20. Lindner C, Thiagarajah S, Wilkinson JM, Wallis GA, Cootes TF. Development of a fully automatic shape model matching (FASMM) system to derive statistical shape models from radiographs: application to the accurate capture and global representation of proximal femur shape. Osteoarthritis and Cartilage. 2013/10/01/ 2013;21(10):1537-1544. doi:10.1016/j.joca.2013.08.008 21. Faber B. Geometric Parameters Python 3.0 Code. 2022. https://zenodo.org/badge/latestdoi/518486087 22. Holroyd C, Cooper C, Dennison E. Epidemiology of osteoporosis. Best Practice & Research Clinical Endocrinology & Metabolism. 2008/10/01/ 2008;22(5):671-685. doi:10.1016/j.beem.2008.06.001 23. Wang SC, Brede C, Lange D, et al. Gender differences in hip anatomy: possible implications for injury tolerance in frontal collisions. Annu Proc Assoc Adv Automot Med. 2004;48:287-301. 24. Goodyear SR, Barr RJ, McCloskey E, et al. Can we improve the prediction of hip fracture by assessing bone structure using shape and appearance modelling? Bone. 2013/03/01/ 2013;53(1):188-193. doi:10.1016/j.bone.2012.11.042 25. Frysz M, Gregory J, Aspden RM, Paternoster L, Tobias JH. Sex differences in proximal femur shape: findings from a population-based study in adolescents. Sci Rep. Mar 12 2020;10(1):4612. doi:10.1038/s41598-020-61653-4 26. Fajar JK, Rusydi R, Rahman S, Alam AIN, Azharuddin A. Hip geometry to predict femoral neck fracture: only neck width has significant association. Apollo Medicine. 2016;13(4):213-219. doi: 10.1016/j.apme.2016.05.005 27. Han J, Hahn MH. Proximal Femoral Geometry as Fracture Risk Factor in Female Patients with Osteoporotic Hip Fracture. J Bone Metab. Aug 2016;23(3):175-82. doi:10.11005/jbm.2016.23.3.175 28. El-Kaissi S, Pasco JA, Henry MJ, et al. Femoral neck geometry and hip fracture risk: the Geelong osteoporosis study. Osteoporos Int. Oct 2005;16(10):1299-303. doi:10.1007/s00198-005-1988-z 29. Tobias JH, Nethander M, Faber BG, et al. Femoral neck width genetic risk score is a novel independent risk factor for hip fractures. Journal of Bone and Mineral Research. 2024;doi:10.1093/jbmr/zjae002 30. Alonso CG, Curiel MD, Carranza FH, Cano RP , Pérez AD. Femoral bone mineral density, neck-shaft angle and mean femoral neck width as predictors of hip fracture in men and women. Osteoporosis International. 2000;11(8):714. doi:10.1007/s001980070071 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 24 31. Peacock M, Liu G, Carey M, et al. Bone Mass and Structure at the Hip in Men and Women over the Age of 60 Years. Osteoporosis International. 1998/05/01 1998;8(3):231-239. doi:10.1007/s001980050059 32. Edwards K, Leyland KM, Sanchez-Santos MT, et al. Differences between race and sex in measures of hip morphology: a population-based comparative study. Osteoarthritis and Cartilage. 2020;28(2):189-200. doi:10.1016/j.joca.2019.10.014 33. Siccardi MA, Tariq MA, Valle C. Anatomy, Bony Pelvis and Lower Limb: Psoas Major. StatPearls. StatPearls Publishing; 2024. 34. Cruz-Jentoft AJ, Sayer AA. Sarcopenia. Lancet. Jun 29 2019;393(10191):2636-2646. doi:10.1016/s0140-6736(19)31138-9 35. Rosenberg IH. Sarcopenia: origins and clinical relevance. Clin Geriatr Med. Aug 2011;27(3):337-9. doi:10.1016/j.cger.2011.03.003 36. Hida T, Ishiguro N, Shimokata H, et al. High prevalence of sarcopenia and reduced leg muscle mass in Japanese patients immediately after a hip fracture. Geriatr Gerontol Int. Apr 2013;13(2):413-20. doi:10.1111/j.1447-0594.2012.00918.x 37. Hong W, Cheng Q, Zhu X, et al. Prevalence of Sarcopenia and Its Relationship with Sites of Fragility Fractures in Elderly Chinese Men and Women. PLoS One. 2015;10(9):e0138102. doi:10.1371/journal.pone.0138102 38. Harvey NC, Orwoll E, Kwok T, et al. Sarcopenia Definitions as Predictors of Fracture Risk Independent of FRAX(®) , Falls, and BMD in the Osteoporotic Fractures in Men (MrOS) Study: A Meta-Analysis. J Bone Miner Res. Jul 2021;36(7):1235-1244. doi:10.1002/jbmr.4293 39. Testa G, Vescio A, Zuccalà D, et al. Diagnosis, Treatment and Prevention of Sarcopenia in Hip Fractured Patients: Where We Are and Where We Are Going: A Systematic Review. J Clin Med. Sep 17 2020;9(9)doi:10.3390/jcm9092997 40. Faber BG, Frysz M, Zheng J, et al. The genetic architecture of hip shape and its role in the development of hip osteoarthritis and fracture. Human Molecular Genetics. 2024;doi:10.1093/hmg/ddae169 41. Gómez Alonso C, Díaz Curiel M, Hawkins Carranza F, Pérez Cano R, Díez Pérez A. Femoral Bone Mineral Density, Neck-Shaft Angle and Mean Femoral Neck Width as Predictors of Hip Fracture in Men and Women. Osteoporos Int. Sep 2000;11(8):714-20. doi:10.1007/s001980070071 42. Zheng J, Frysz M, Faber BG, et al. Comparison between UK Biobank and Shanghai Changfeng suggests distinct hip morphology may contribute to ethnic differences in the prevalence of hip osteoarthritis. Osteoarthritis and Cartilage. 2023/11/05/ 2023;doi:10.1016/j.joca.2023.10.006 43. Johnell O, Kanis JA. An estimate of the worldwide prevalence and disability associated with osteoporotic fractures. Osteoporos Int. Dec 2006;17(12):1726-33. doi:10.1007/s00198-006-0172-4 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 25 TABLES Table 1 Descriptive statistics of UK Biobank participants included in this study Combined Female Male N = 38,123 N = 19,820 (52%) N = 18,303 (48%) Exposures Mean [SD, Range] Mean [SD, Range] Mean [SD, Range] Age (years) 63.7 [7.6, 44-82] 63.0 [7.4, 45-82] 64.3 [7.7, 44-81] Height (cm) 170.2 [9.4, 135-204] 163.7 [6.4, 135-196] 177.2 [6.6, 150-204] Weight (kg) 75.4 [15.1, 34-171] 68.2 [12.9, 34-169] 83.2 [13.4, 47-171] Left femoral bone mineral density (g/cm2) 1.0 [0.2, 0.0-1.7] 0.9 [0.1, 0.1-1.7] 1.1 [0.1, 0.0-1.7] Hip shape mode 1 0.0 [1.0, -4.6-3.9] 0.3 [0.9, -3.8-3.9] -0.3 [1.0, -4.6-3.6] Hip shape mode 2 0.0 [1.0, -4.7-4.5] -0.0 [1.0, -4.7-4.2] 0.0 [1.0, -4.5-4.5] Hip shape mode 3 0.0 [1.0, -4.1-4.3] -0.3 [0.9, -4.1-4.0] 0.3 [1.0, -3.6-4.3] Hip shape mode 4 0.0 [1.0, -4.4-4.0] -0.1 [1.0, -4.4-4.0] 0.1 [1.0, -3.8-4.0] Hip shape mode 5 0.0 [1.0, -4.5-3.5] 0.0 [1.1, -4.4-3.4] -0.0 [0.9, -4.5-3.5] Hip shape mode 6 0.0 [1.0, -4.6-5.0] 0.2 [1.0, -3.4-5.0] -0.2 [1.0, -4.6-3.9] Hip shape mode 7 0.0 [1.0, -4.9-5.0] 0.1 [1.0, -4.9-4.7] -0.1 [1.0, -4.6-5.0] Hip shape mode 8 0.0 [1.0, -4.4-4.5] -0.1 [1.0, -4.4-4.0] 0.1 [1.0, -4.0-4.5] Hip shape mode 9 0.0 [1.0, -4.1-5.0] -0.3 [0.9, -4.1-4.5] 0.3 [1.0, -3.7-5.0] Hip shape mode 10 0.0 [1.0, -4.1-3.8] 0.0 [1.0, -4.1-3.8] -0.0 [1.0, -4.1-3.8] Narrowest neck width (mm) 31.6 [3.5, 21.4-45.8] 29.0 [2.0, 21.4-37.8] 34.5 [2.4, 22.9-45.8] Diameter of femoral head (mm) 45.9 [3.8, 33.4-64.4] 43.0 [2.3, 33.4-53.7] 49.0 [2.6, 34.7-64.4] Hip axis length (mm) 96.7 [8.0, 68.1-127.1] 90.8 [4.8, 68.1-115.5] 103.1 [5.5, 76.9-127.1] Number fractured [%] Number fractured [%] Number fractured [%] Hospital diagnosed fracture 133 [0.35] 89 [0.45] 44 [0.24] Mean [SD, Range] Mean [SD, Range] Mean [SD, Range] Time to end of study (years) 5.0 [1.5, 0.2-8.5] 5.0 [1.5, 0.1-8.5] 5.0 [1.5, 0.2-8.5] Population characteristics of the UK Biobank participants included in this study with complete data for all covariates. . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 26 Table 2: Cox proportional hazard results for the association between each hip shape mode and hip fracture in combined sex analysis Hazard ratios (HR) with 95% confidence intervals (CI) and p -values are shown for each hip shape mode and their association to hip fracture. Model 1 = unadjusted; model 2 = adjusted for age, sex, height, and weight; model 3 = adjusted for model 2 plus bone mineral density; model 4 = adjusted for model 3 plus the geometric measures. Model 1 Model 2 Model 3 Model 4 Exposure HR [95% CI] p-value HR [95% CI] p-value HR [95% CI] p-value HR [95% CI] p-value Hip shape mode 1 1.01 [0.85-1.20] 0.91 0.94 [0.79-1.13] 0.52 1.11 [0.92-1.33] 0.27 1.12 [0.90-1.40] 0.33 Hip shape mode 2 1.32 [1.11-1.56] 1.47 × 10-3 1.36 [1.15-1.62] 3.3 × 10-4 1.31 [1.11-1.55] 1.51 × 10-3 1.30 [1.09-1.55] 3.27 × 10-3 Hip shape mode 3 0.99 [0.84-1.18] 0.94 1.13 [0.94-1.35] 0.19 1.16 [0.97-1.39] 0.10 1.10 [0.91-1.31] 0.33 Hip shape mode 4 0.87 [0.74-1.03] 0.11 0.84 [0.71-1.00] 0.05 0.88 [0.74-1.05] 0.17 0.88 [0.71-1.07] 0.20 Hip shape mode 5 0.97 [0.82-1.15] 0.74 0.98 [0.84-1.16] 0.86 1.02 [0.87-1.21] 0.79 0.99 [0.83-1.17] 0.89 Hip shape mode 6 1.09 [0.92-1.29] 0.31 1.00 [0.84-1.19] 1.00 0.96 [0.80-1.14] 0.63 0.96 [0.79-1.16] 0.68 Hip shape mode 7 0.96 [0.81-1.14] 0.66 0.99 [0.84-1.17] 0.93 1.05 [0.88-1.24] 0.60 1.11 [0.93-1.31] 0.25 Hip shape mode 8 1.00 [0.84-1.19] 0.99 1.02 [0.86-1.21] 0.82 0.95 [0.80-1.12] 0.53 1.00 [0.83-1.21] 0.98 Hip shape mode 9 0.89 [0.75-1.06] 0.19 0.96 [0.81-1.15] 0.68 0.99 [0.82-1.18] 0.87 0.90 [0.75-1.08] 0.26 Hip shape mode 10 0.99 [0.84-1.18] 0.93 0.95 [0.80-1.13] 0.59 0.97 [0.82-1.15] 0.74 0.91 [0.76-1.09] 0.30 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 27 FIGURES Figure 1: An example hip DXA scan from UKB showing the points placed around the hip joint. . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 28 Figure 2: The ten hip shape modes (HSMs). The solid line shows the shape +2 standard deviations ( SD) from the mean, and the dotted line shows the shape -2 SDs from the mean. . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 29 Figure 3: Composite image of the ten hip shape modes. The solid line shows the shape at risk of fracture, the dotted line shows the mean shape. . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 30 Figure 4: Cox proportional hazard results for the association between each hip shape mode (HSM) and hip fracture in combined sex analysis. Hazard ratios (HR) with 95% confidence intervals (CI) are plotted. Square = unadjusted (model 1); circle = adjusted for age, sex, height, and weight (model 2); triangle = adjusted for model 2 plus bone mineral density (model 3) ; diamond = fully adjusted for model 3 plus the three geometric measures (model 4). . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 31 Figure 5: Cox proportional hazard results for the association between each geometric measure (GM) and hip fracture in combined sex analysis. Hazard ratios (HR) with 95% confidence intervals (CI) are plotted. Square = unadjusted (model 1); circle = adjusted for age, sex, height, and weight (model 2); triangle = adjusted for model 2 plus bone mineral density (model 3) ; diamond = fully adjusted for model 3 plus the other two geometric measures (model 4). FNW = femoral neck width, FHD = femoral head diameter, HAL = hip axis length . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 32 SUPPLEMENTARY TABLES AND FIGURES Supplementary Table 1: Cox proportional hazard results for the association between each hip shape mode and hip fracture in sex- stratified analysis Model 1 Model 2 Model 3 Model 4 Exposure HR [95% CI] p-value HR [95% CI] p-value HR [95% CI] p-value HR [95% CI] p-value Males Hip shape mode 1 1.12 [0.83-1.52] 0.47 1.09 [0.80-1.47] 0.59 1.28 [0.94-1.75] 0.12 1.32 [0.89-1.97] 0.17 Hip shape mode 2 1.43 [1.06-1.93] 0.02 1.39 [1.03-1.88] 0.03 1.38 [1.02-1.86] 0.04 1.34 [0.98-1.84] 0.06 Hip shape mode 3 1.20 [0.88-1.62] 0.25 1.25 [0.92-1.69] 0.16 1.27 [0.94-1.73] 0.12 1.17 [0.86-1.60] 0.33 Hip shape mode 4 0.79 [0.59-1.07] 0.13 0.77 [0.57-1.04] 0.09 0.79 [0.58-1.07] 0.13 0.72 [0.50-1.03] 0.08 Hip shape mode 5 0.84 [0.62-1.15] 0.28 0.85 [0.63-1.16] 0.32 0.92 [0.67-1.26] 0.59 0.87 [0.63-1.20] 0.40 Hip shape mode 6 1.00 [0.73-1.36] 1.00 0.99 [0.73-1.35] 0.97 0.96 [0.70-1.31] 0.79 0.94 [0.68-1.31] 0.73 Hip shape mode 7 1.10 [0.82-1.47] 0.53 1.14 [0.85-1.52] 0.39 1.20 [0.89-1.61] 0.24 1.29 [0.96-1.74] 0.09 Hip shape mode 8 0.98 [0.73-1.32] 0.89 0.98 [0.73-1.31] 0.89 0.90 [0.67-1.21] 0.48 0.96 [0.70-1.32] 0.81 Hip shape mode 9 0.72 [0.54-0.98] 0.04 0.70 [0.52-0.95] 0.02 0.73 [0.54-0.99] 0.04 0.66 [0.48-0.89] 0.01 Hip shape mode 10 0.95 [0.72-1.26] 0.73 0.94 [0.71-1.25] 0.68 0.97 [0.74-1.28] 0.84 0.91 [0.67-1.23] 0.53 Females Hip shape mode 1 0.83 [0.67-1.03] 0.09 0.87 [0.70-1.09] 0.22 1.02 [0.82-1.28] 0.83 1.04 [0.80-1.36] 0.76 Hip shape mode 2 1.27 [1.04-1.56] 0.02 1.37 [1.11-1.68] 2.79 × 10-3 1.26 [1.03-1.54] 0.02 1.27 [1.03-1.57] 0.02 Hip shape mode 3 1.05 [0.84-1.31] 0.68 1.07 [0.86-1.34] 0.54 1.13 [0.90-1.40] 0.29 1.08 [0.86-1.35] 0.52 Hip shape mode 4 0.97 [0.78-1.19] 0.75 0.87 [0.71-1.08] 0.21 0.94 [0.76-1.15] 0.54 0.96 [0.75-1.23] 0.76 Hip shape mode 5 1.01 [0.83-1.24] 0.89 1.05 [0.86-1.27] 0.65 1.05 [0.86-1.29] 0.60 1.03 [0.84-1.26] 0.78 Hip shape mode 6 1.03 [0.84-1.27] 0.78 1.01 [0.82-1.24] 0.96 0.96 [0.78-1.18] 0.68 0.96 [0.76-1.22] 0.75 Hip shape mode 7 0.87 [0.71-1.07] 0.18 0.94 [0.77-1.15] 0.56 0.99 [0.81-1.22] 0.96 1.04 [0.84-1.27] 0.73 Hip shape mode 8 1.05 [0.85-1.29] 0.68 1.04 [0.85-1.28] 0.70 0.98 [0.80-1.21] 0.87 1.03 [0.82-1.29] 0.82 . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 33 Hip shape mode 9 1.14 [0.91-1.41] 0.25 1.15 [0.92-1.43] 0.22 1.15 [0.92-1.44] 0.21 1.07 [0.85-1.35] 0.56 Hip shape mode 10 1.01 [0.81-1.26] 0.92 0.97 [0.78-1.21] 0.80 0.96 [0.77-1.19] 0.68 0.91 [0.72-1.14] 0.40 Hazard ratios (HR) with 95% confidence intervals (CI) and p -values are shown for each hip shape mode and their association to hip fracture. Model 1 = unadjusted ; model 2 = adjusted for age, height, and weight ; model 3 = adjusted for model 2 plus bone mineral density ; model 4 = adjusted for model 3 plus the geometric measures. . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 34 Supplementary Table 2: Cox proportional hazard results for the association between each geometric measurement and hip fracture. Model 1 Model 2 Model 3 Model 4 Exposure HR [95% CI] p-value HR [95% CI] p-value HR [95% CI] p-value HR [95% CI] p-value Combined sex FNW (mm) 1.15 [0.97-1.36] 0.11 2.45 [1.80-3.33] 1.17 × 10-8 1.77 [1.30-2.43] 3.26 × 10-4 1.31 [0.88-1.96] 0.19 FHD (mm) 1.12 [0.95-1.33] 0.17 2.42 [1.76-3.32] 5.20 × 10-8 1.89 [1.39-2.57] 4.48 × 10-5 1.47 [0.96-2.25] 0.07 HAL (mm) 1.08 [0.91-1.28] 0.39 1.85 [1.34-2.55] 1.93 × 10-4 1.61 [1.18-2.21] 3.08 × 10-3 1.21 [0.84-1.74] 0.31 Males FNW (mm) 2.17 [1.44-3.25] 1.99 × 10-4 2.13 [1.33-3.43] 1.74 × 10-3 1.75 [1.08-2.82] 0.02 1.17 [0.63-2.18] 0.62 FHD (mm) 2.30 [1.54-3.44] 4.50 × 10-5 2.36 [1.45-3.84] 5.54 × 10-4 2.01 [1.28-3.14] 2.26 × 10-3 1.61 [0.85-3.06] 0.14 HAL (mm) 2.07 [1.37-3.11] 4.84 × 10-4 2.01 [1.19-3.37] 0.01 1.84 [1.11-3.04] 0.02 1.39 [0.78-2.45] 0.26 Females FNW (mm) 2.88 [2.05-4.06] 1.40 × 10-9 2.71 [1.80-4.08] 1.63 × 10-6 1.70 [1.11-2.59] 0.01 1.39 [0.82-2.35] 0.22 FHD (mm) 2.43 [1.73-3.40] 2.60 × 10-7 2.46 [1.62-3.74] 2.36 × 10-5 1.70 [1.11-2.60] 0.01 1.35 [0.77-2.36] 0.29 HAL (mm) 2.05 [1.48-2.85] 1.58 × 10-5 1.73 [1.15-2.62] 0.01 1.40 [0.92-2.13] 0.11 1.09 [0.67-1.75] 0.73 Hazard ratios (HR) with 95% confidence intervals (CI) and p -values are shown for each geometric measure and their association to hip fracture. Model 1 = unadjusted; model 2 = adjusted for age, sex, height, and weight (no sex adjustment in sex -stratified analysis); model 3 = adjusted for model 2 plus bone mineral density; model 4 = adjusted for model 3 plus the remaining 2 geometric measures. FNW = femoral neck width, FHD = femoral head diameter, HAL = hip axis length . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint 35 Supplementary Figure 1: Pearson’s correlation matrix (r) showing the correlation between each HSM, GM (FHD, HAL, FNW), height, weight, age and BMD within the cohort. Green shows a strong correlation (r ≥0.7 -1), orange shows a moderate correlation (r ≥0.5 -<0.7), red shows a weak correlation (r <0.5). FHD = femoral head diameter, HAL = hip axis length, FNW = femoral neck width, BMD = bone mineral density, HSM = hip shape mode, GM = geometric measure. . CC-BY 4.0 International licenseIt is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in(which was not certified by peer review)preprint The copyright holder for thisthis version posted October 2, 2025. ; https://doi.org/10.1101/2025.09.30.25336960doi: medRxiv preprint