Development and evaluation of personalized risk assessments for osteoporotic patients

DEVELOPMENT AND EVALUATION OF
PERSONALIZED RISK ASSESSMENTS FOR
OSTEOPOROTIC PATIENTS

by

Le Phuong Thao Ho

A thesis submitted to the University of Technology Sydney in partial fulfilment
of the requirements for the Degree of Doctor of Philosophy

School of Biomedical Engineering,
Faculty of Engineering and Information Technology,
University of Technology Sydney
Australia
2018


CERTIFICATE OF ORIGINAL AUTHORSHIP
I, Le Phuong Thao HO, declare that this thesis is submitted in fulfilment of the
requirements for the award of Doctor of Philosophy in the School of Biomedical
Engineering at the University of Technology Sydney.
I also declare that the intellectual content in this thesis is the product of my own work,
except to the extent that assistance from others in the project’s design and conception or
in style, presentation and language expression is acknowledged. In addition, I certify
that all information sources and literature used are indicated in the thesis.
The work in this thesis has not previously been submitted for a degree, nor has it been
submitted as part of the requirements of a degree, at any other academic institution,
except as fully acknowledged within the text.
This research is supported by the Australian Government Research Training Program.

Signature of Candidature:

Date: 6 October 2018

i


ABSTRACT
Osteoporosis is a skeletal disease characterized by reduced bone strength
and deterioration in bone microstructure, leading to increased risk of fragility
fracture. Bone strength is mainly determined by bone mineral density (BMD). The
variation in BMD is partly determined by genetic factors. An individual's risk of
fracture is determined by the individual's genetic structure and environmental
exposures. While several genetic variants associated with BMD have been
identified, the contribution of these variants to fracture risk prediction has not been
well-documented. In this thesis, I sought to (i) construct an osteogenomic profile
from BMD-associated genetic variants; (ii) assess the association between the
profile and fracture risk and bone loss; (iii) determine the clinical utility of the
osteogenomic profile in terms of fracture risk assessment; and (iv) improve the
accuracy of hip fracture prediction in postmenopausal women by using artificial
neural network approach.
The work in this thesis was based on the Dubbo Epidemiology Osteoporosis
Study, which is designed as a population-based longitudinal prospective cohort
investigation that involved more than 4000 men and women aged 60+years. The
individuals had been followed up to 27 years. The incidence of fracture was
ascertained during the follow-up period. A unique osteogenomic profile was
constructed for each individual from 68 BMD-associated genetic variants. The
osteogenomic profile was significantly associated with BMD, fracture risk, and
BMD changes. Incorporating the osteogenomic profile into existing prognostic
ii

model improved the prognostic performance over and above of traditional clinical
risk factors models (age, gender, prior fracture, and history of fall). The area under
the receiver operating characteristic curve for model with the osteogenomic profile
was 71.1%, an increase of 0.5% compared with the model without the profile.
More importantly, reclassification analysis showed that compared with the clinical
risk factor (CRF) model, adding GRS resulted in 16% of individuals moving
correctly from one risk category to another. In decision curve analysis, I found that
for risk threshold greater than 15%, the osteogenomic profile could help reduce
the number of unnecessary treatments. I also demonstrated that the predictive
accuracy of fracture prediction using artificial neural network model was improved
to 87% (AUC 0.94), which was significantly higher than that for the Cox's
proportional hazards model (Accuracy 82%, AUC 0.85) or the Garvan model
(Accuracy 83%, AUC 0.86).
In conclusion, this thesis shows that an osteogenomic profile constructed
from multiple BMD associated genetic variants is associated with fracture risk,
and that the incorporation of osteogenomic profile could enhance the accuracy of
fracture risk assessment for an individual men and women.

iii


PUBLICATIONS ARISING FROM THIS THESIS
Published Papers
1.

Mai, H. T., Tran, T. S., Ho-Le, T. P., Pham, T. T., Center, J. R., Eisman, J. A., &
Nguyen, T. V. (2018). Low-trauma rib fracture in the elderly: Risk factors and
mortality consequence. Bone, 116, 295-300. doi:10.1016/j.bone.2018.08.016

2.

Ho-Le, T. P., Pham, H. M., Center, J. R., Eisman, J. A., Nguyen, H. T., & Nguyen,
T. V. (2018). Prediction of changes in bone mineral density in the elderly:
contribution of "osteogenomic profile". Arch Osteoporos, 13(1), 68.
doi:10.1007/s11657-018-0480-2

3.

Ho-Pham, L. T., Ho-Le, T. P., Mai, L. D., Do, T. M., Doan, M. C., & Nguyen, T.
V. (2018). Sex-difference in bone architecture and bone fragility in Vietnamese. Sci
Rep, 8(1), 7707. doi:10.1038/s41598-018-26053-9

4.

Ho-Le, T. P., & Nguyen, T. V. (2018). Mathematics Research in Association of
Southeast Asian Nations Countries: A Scientometric Analysis of Patterns and
Impacts.
Frontiers
in
Research
doi:10.3389/frma.2018.00003

Metrics

and

Analytics,

3(3).

5.

Ho-Le TP, Center JR, Eisman JA, Nguyen HT, Nguyen TV. Prediction of Hip
Fracture in Post-menopausal Women using Artificial Neural Network Approach.
39th Annual International Conference of the IEEE Engineering in Medicine and
Biology Society (EMBC), 11-15 July 2017; 10.1109/EMBC.2017.8037784

6.

Nguyen TV, Ho-Le TP, Le UV. International collaboration in scientific research
in Vietnam: an analysis of patterns and impact. Scientometrics. journal article, Feb
2017;110(2):1035-51.

7.

Ho-Le TP, Center JR, Eisman JA, Nguyen HT, Nguyen TV. Prediction of Bone
Mineral Density and Fragility Fracture by Genetic Profiling. Journal of Bone and
Mineral Research. Feb 2017;32(2):285-93.

8.

Pham HM, Nguyen SC, Ho-Le TP, Center JR, Eisman JA, Nguyen TV. Association
of Muscle Weakness With Post-Fracture Mortality in Older Men and Women: A 25Year Prospective Study. Journal of bone and mineral research. Nov 2016. Epub
2016/11/20.

iv

Manuscripts Submitted or In Preparation
9.

Ho-Le, T. P., Thach S. Tran, Jacqueline R. Center, John A. Eisman, Tuan V.
Nguyen (2018), Contribution of Multimorbility to Post-Fracture Mortality : A
Latent Class Analysis Approach – In preparation.

10.

Ho-Le, T. P., Center, J. R., Eisman, J. A., Nguyen, H. T., & Nguyen, T. V. (2018).
Genetic prediction of lifetime risk of fracture, Journal of Bone and Mineral
Research – In preparation.

11.

Ho-Le, T. P., Center, J. R., Eisman, J. A., Nguyen, H. T., & Nguyen, T. V. (2018).
Assessing the clinical utility of genetic profiling in fracture risk prediction: A
Decision Curve Analysis Approach. Journal of Osteoporosis International – In
preparation.

Conference Presentations (* indicates oral presentation)
1.

Thao P. Ho-Le, Thach S. Tran, Jacqueline R. Center, John A. Eisman, Tuan V.
Nguyen (2018), Contribution of Multimorbility to Post-Fracture Mortality: Result
of a Long Term Population Based Study, 40th Annual Scientific Meeting of the
American society for Bone and Mineral Research, Quebec, Canada.

2.

Thao P. Ho-Le, Jacqueline R. Center, John A. Eisman, Tuan V. Nguyen (2018),
Assessing Clinical Utility of Genetic Profiling in Fracture Risk Assessment: A
Decision Curve Analysis, 40th Annual Scientific Meeting of the American society
for Bone and Mineral Research, Quebec, Canada.

3.

*Thao P. Ho-Le, Thach S. Tran, Jacqueline R. Center, John A. Eisman, Tuan V.
Nguyen (2018), Post-Fracture Mortality: A Latent Class Analysis of
Multimorbidities, 28th Annual Scientific Meeting of the ANZBMS, Queenstown,
New Zealand.

4.

*Thao P. Ho-Le, Jacqueline R. Center, John A. Eisman, Tuan V. Nguyen (2018),
Clinical Utility Assessment of Genetic Profiling in Fracture Risk Prediction: A
Decision Curve Analysis Approach, 28th Annual Scientific Meeting of the
ANZBMS, Queenstown, New Zealand.

5.

Thao P. Ho-Le, Jacqueline R. Center, John A. Eisman, Hung T. Nguyen, Tuan V.
Nguyen (2017), Prediction of Hip Fracture in Postmenopausal Women using
v


Artificial Neural Network Approach, 39th Annual Scientific Meeting of the
American society for Bone and Mineral Research, Denver.
6.

Thao P. Ho-Le, Jacqueline R. Center, John A. Eisman, Hung T. Nguyen, Tuan V.
Nguyen (2017), Genetic Prediction of Lifetime Risk of Fracture, 39th Annual
Scientific Meeting of the American society for Bone and Mineral Research, Denver.

7.

*Thao P. Ho-Le, Jacqueline R. Center, John A. Eisman, Hung T. Nguyen, Tuan V.
Nguyen (2017), Determination of Risk Threshold for Osteoporosis Therapy: A
Decision Curve Analysis Approach, 39th Annual Scientific Meeting of the
American society for Bone and Mineral Research, Denver.

8.

Thao P. Ho-Le, Jacqueline R. Center, John A. Eisman, Hung T. Nguyen, Tuan V.
Nguyen (2017), Burden of fractures attributable to low bone mineral density, 39th
Annual Scientific Meeting of the American society for Bone and Mineral Research,
Denver.

9.

*Thao P. Ho-Le, Jacqueline R. Center, John A. Eisman, Hung T. Nguyen, Tuan V.
Nguyen (2017). Determination of risk threshold for osteoporosis therapy: a
decision curve analysis approach. 27th Annual Scientific Meeting of the ANZBMS,
Brisbane.

10.

*Ho-Le TP, Center JR, Eisman JA, Nguyen HT, Nguyen TV (2017). Accurate
Prediction of Hip Fracture by Machine Learning Approach. 27th Annual Scientific
Meeting of the ANZBMS, Brisbane.

11.

*Thao P. Ho-Le, Jacqueline R. Center, John A. Eisman, Tuan V. Nguyen, Hung
T. Nguyen (2017). Prediction of Hip Fracture in Post-menopausal Women using
Artificial Neural Network Approach. 39th Annual International Conference of the
IEEE Engineering in Medicine and Biology Society, Jeju.

12.

Ho-Le TP, Center JR, Eisman JA, Nguyen HT, Nguyen TV (2017). Genetic
prediction of lifetime risk of fracture. WCO-IOF-ESCEO conference, Florence.

13.

Thao P. Ho-Le, Hanh M. Pham, Jacqueline R. Center, John A. Eisman, Hung T.
Nguyen, Tuan V. Nguyen (2016). Genetic Profiling Predicts Bone Loss and Bone
Mineral Density, 38th Annual Scientific Meeting of the American society for Bone
and Mineral Research, Atlanta.

14.

*Ho-Le TP, Center JR, Eisman JA, Nguyen HT, Nguyen TV (2016). Prediction of
bone mineral density and fragility fracture by genetic profiling, 38th Annual
Scientific Meeting of the American society for Bone and Mineral Research, Atlanta.

vi


15.

*Ho-Le TP, Center JR, Eisman JA, Nguyen HT, Nguyen TV (2016). Prediction of
fracture by genetic profiling. Vietnamese Osteoporosis conference, Nha Trang,
Vietnam.

16.

*Thao P. Ho-Le, Hanh M. Pham, Jacqueline R. Center, John A. Eisman, Hung T.
Nguyen, Tuan V. Nguyen (2016). Genetic Profiling of 68 SNPs is Associated with
Femoral Neck Bone Loss. Annual Meetings of the Endocrine Society of Australia
and Society for Reproductive Biology and Australia and New Zealand Bone and
Mineral Society, Gold Coast, Australia.

17.

*Ho-Le TP, Center JR, Eisman JA, Nguyen HT, Nguyen TV (2016). Prediction of
osteoporotic fracture by genetic profiling. New Horizons conference, Sydney,
Australia.

18.

*Ho-Le TP, Center JR, Eisman JA, Nguyen HT, Nguyen TV (2015). Polygenic
Risk Score improves fracture risk prediction: the Dubbo osteoprorosis
epidemiology study. 25th Annual Scientific Meeting of the ANZBMS, Tasmania,
Australia.

vii


AWARDS AND TRAVEL GRANTS

1.

The Christine and T Jack Martin Travel Grant (2018), the 28th Annual
Scientific Meeting of the ANZBMS, Queenstown, New Zealand.

2.

Second oral presentation prize (2018), The Faculty of Engineering and
Information Technology Research Showcase, UTS, Australia

3.

Eleventh Fellows Forum on Metabolic Bone Diseases Fellowship (2017),
Endocrines Fellow Foundation, Denver, USA

4.

Young Investigator Award (2017), the 39th Annual Meeting of American
Society of Bone and Mineral Research, Denver, USA

5.

Sol Posen Best Research Paper Award (2017), Brisbane, Australia

6.

High Quality Publication Award (2017), The University of Technology,
Sydney (UTS)

7.

Best Poster Presentation Award (2017), 4th EMBL Australia Postgraduate
Symposium, Sydney

8.

Outstanding Abstract (2016), Vietnamese Osteoporosis Society conference,
Nha Trang, Vietnam

9.

Amgen Outstanding Abstract Award (2015), Annual Australia and New
Zealand Bone and Mineral Society (ANZBMS) conference

viii


TRAVEL AWARDS

1.

ASBMR Young Investigator Travel Award, Montreal, Canada, 2018

2.

Australian and New Zealand Bone and Mineral Society 28th Annual Meeting
Travel Grant, Queenstown, New Zealand

3.

ASBMR Young Investigator Travel Award, Denver, USA, 2017

4.

UTS Vice Chancelor Travel Grant, 2017

5.

Australian and New Zealand Bone and Mineral Society 27th Annual Meeting
Travel Grant, Brisbane, Australia, 2017

6.

ASBMR Young Investigator Travel Award, Atlanta, USA, 2016

7.

UTS Vice Chancelor Travel Grant, 2016

8.

Australian and New Zealand Bone and Mineral Society 26th Annual Meeting
Travel Grant, Gold Coast, Australia, 2016

9.

Australian and New Zealand Bone and Mineral Society 25th Annual Meeting
Travel Grant, Tasmania, Australia, 2015

ix


ACKNOWLEGEMENTS
Firstly, I would like to thank Vietnamese Ministry of Education and
Training (MOET) and University of Technology Sydney (UTS) for their
University of Technology Sydney – Vietnam International Education
Development (UTS-VIED) scholarship so that I can do this research.
I express my deepest sincere thanks to my supervisor Professor Tuan Van
Nguyen for his great patience, guidance, and especially his persistent
encouragement. During my PhD journey, I have had valuable opportunities to learn
not only from his wide academic knowledge but also from the kind philosophy of
his life. He has sharpened my belief that working hard and treating others kindly is
the not easy but worth to do in my own life. Without his support and faith, I would
never have finished my degree with various honor awards and precious
opportunities to attend national and international conferences in the bone field.
I honestly thank my co-supervisor Professor Hung Tan Nguyen for his
enthusiastic sharing, quick response and kind support whenever and whatever I
need to make my research smoothly. I would like to deliver a grateful thank to
Professor John Eisman and Professor Jacqueline Center for providing treasured
comments, suggestions, and helps on my research studies.
I thank Dr Minh Tuan Ho and Da Nang University for their support during
my visiting collaboration. I would like to appreciate Dr. Bich Tran, Dr. Thach
Tran, Dr Hanh Pham, and Dr. Angela Shu for their supports in data preprocessing and other questions throughout the study. I gratefully acknowledge
Sr. Janet Watters, Shaye Field, Donna Reeves and Jodie Ratleg for the clinical
data collection. I also appreciate the assistance of Denia Mang and the IT group
at the Garvan Institute of Medical Research for database management.
Especially, I would like to acknowledge Dr Pavel Bitter and research team at
the Australian Cancer Research Foundation (ACRF) for their assistant in doing
genotyping.

I appreciate all my colleagues and staff in the School of Biomedical
Engineering, UTS and Bone Biology Division, Garvan Institute of Medical
x


Research, who supported me in my study with encouragement and confidence. I
would like to thank my close friends for their sincere friendship and continuous
support during my research years.
My heartfelt gratitude goes to my parents, my sister and brother, especially
my husband Tuan and my son Phu An for their endless love, understanding,
sharing and tireless support over the years that enabled me to complete this thesis.

xi


TABLE OF CONTENTS
CERTIFICATE OF ORIGINAL AUTHORSHIP ........................................................... i
ABSTRACT ........................................................................................................................ ii
PUBLICATIONS ARISING FROM THIS THESIS...................................................... iv
AWARDS AND TRAVEL GRANTS ............................................................................ viii
ACKNOWLEGEMENTS .................................................................................................. x
TABLE OF CONTENTS ................................................................................................. xii
LIST OF FIGURES ....................................................................................................... xviii
LIST OF TABLES ........................................................................................................... xix
LIST OF ABBREVIATIONS .......................................................................................... xx
CHAPTER 1.

INTRODUCTION AND BACKGROUND ......................................... 1

1.1

BACKGROUND............................................................................................... 2

1.2

MOTIVATION OF THE THESIS ................................................................. 6

1.3

AIMS OF THESIS ........................................................................................... 9

1.4

CONTRIBUTIONS ........................................................................................ 11

1.5

STRUCTURE OF THESIS ........................................................................... 13

CHAPTER 2.
2.1

LITERATURE REVIEW................................................................... 16

OSTEOPOROSIS AND OSTEOPOROTIC FRACTURE ........................ 17

2.1.1 Aetiology of osteoporosis ................................................................................. 17
2.1.2 Definition of osteoporosis ................................................................................ 17
2.1.3 Bone loss .......................................................................................................... 21
2.1.4 Osteoporotic fracture ....................................................................................... 22

2.2

RISK FACTORS FOR OSTEOPOROSIS FRACTURES ......................... 22

2.2.1 Advancing age .................................................................................................. 24
2.2.2 Gender .............................................................................................................. 25
2.2.3 Bone mineral density (BMD) ............................................................................ 26

xii


2.2.4 Falls.................................................................................................................. 27
2.2.5 Personal history of fracture ............................................................................. 28
2.2.6 Lifestyle factors ................................................................................................ 29
2.2.7 Nutritional factors ............................................................................................ 30
2.2.8 Genetic factors ................................................................................................. 32
2.3

CURRENT RESEARCH OF FRACTURE RISK ASSESSMENT ........... 35

2.3.1 Garvan Fracture Risk Calculator .................................................................... 35
2.3.2 FRAX ................................................................................................................ 37
2.3.3 QResearch Database’s QfractureScores.......................................................... 39
2.3.4 Other predictive models for fracture risk ......................................................... 40
2.4

THE GAP IN LITERATURE ....................................................................... 40

2.4.1 Translation of genetic factors........................................................................... 41
2.4.2 Predictive value of risk factors ......................................................................... 42

2.4.3 Performance evaluation of predictive models .................................................. 43
2.4.4 Decision curve analysis .................................................................................... 44
2.4.5 Machine learning approach ............................................................................. 44
2.5

PROPOSED STRATEGY OF OSTEOPOROSIS ASSESSMENT ........... 46

CHAPTER 3.

MATERIALS AND METHODS ....................................................... 50

3.1.

STUDY DESIGN AND SETTINGS ............................................................. 51

3.2.

STUDY POPULATION ................................................................................. 52

3.3.

DATA COLLECTION................................................................................... 52

3.3.1. Anthropometric data......................................................................................... 53
3.3.2. Bone mineral density measurement .................................................................. 54
3.3.3. Co-morbidities .................................................................................................. 55
3.3.4. Medication........................................................................................................ 55
3.3.5. Lifestyle factors ................................................................................................ 56
3.3.6. History of falls .................................................................................................. 57

xiii


3.3.7. Ascertainment of fracture ................................................................................. 57
3.3.8. Assessment of mortality data ............................................................................ 58
3.3.9. Collection of blood samples ............................................................................. 59
3.4.

DATA ANALYSIS ......................................................................................... 59

3.4.1. Descriptive analysis and variable selection ..................................................... 59
3.4.2. Association analysis and modelling ................................................................. 60
3.4.3. Model performance evaluation ........................................................................ 60
CHAPTER 4.
PREDICTION OF BONE MINERAL DENSITY AND
FRAGILITY FRACTURE BY GENETIC PROFILING ............................................. 63
4.1.

INTRODUCTION .......................................................................................... 66

4.2.

MATERIALS AND METHODS................................................................... 68

4.2.1 Study design and setting ................................................................................... 68
4.2.2 Ascertainment of fracture ................................................................................. 69
4.2.3 Clinical risk factors .......................................................................................... 69
4.2.4 Genotyping and genetic risk score ................................................................... 70
4.2.5 Data analysis .................................................................................................... 71
4.2.6 Model evaluation .............................................................................................. 72

4.3.

RESULTS ........................................................................................................ 74

4.3.1 Baseline characteristics ................................................................................... 74
4.3.2 GRS and BMD .................................................................................................. 79
4.3.3 GRS and fracture risk ....................................................................................... 82
4.3.4 Discrimination and reclassification analysis ................................................... 86
4.4.

DISCUSSION ................................................................................................. 90

CHAPTER 5.
PREDICTION OF CHANGES IN BONE MINERAL DENSITY IN
THE ELDERLY: CONTRIBUTION OF “OSTEOGENOMIC PROFILE” ............. 96
5.1

INTRODUCTION .......................................................................................... 98

5.2

MATERIALS AND METHODS................................................................... 99

5.2.1 Study design and setting ................................................................................... 99

xiv


5.2.2 Clinical measurements ..................................................................................... 99
5.2.3 Bone mineral density measurements .............................................................. 100

5.2.4 Genotyping and genetic risk score ................................................................. 101
5.2.5 Data analysis .................................................................................................. 102
5.3

RESULTS ...................................................................................................... 104

5.3.1 Baseline characteristics and relative rate of BMD change ............................ 104
5.3.2 Factors associated with relative rate of BMD change ................................... 108
5.3.3 Prediction of rapid bone loss ......................................................................... 114
5.3.4 Random forest analysis .................................................................................. 119
5.4

DISCUSSION ............................................................................................... 120

CHAPTER 6.
ASSESSING THE CLINICAL UTILITY OF GENETIC
PROFILING IN FRACTURE RISK PREDICTION: A DECISION CURVE
ANALYSIS APPROACH .............................................................................................. 124
6.1

INTRODUCTION ........................................................................................ 126

6.2

MATERIALS AND METHODS................................................................. 128

6.2.1 Study design and setting ................................................................................. 128
6.2.2 Ascertainment of fractures ............................................................................. 128
6.2.3 Bone mineral density (BMD) and clinical risk factors ................................... 129
6.2.4 Genotyping and genetic risk score ................................................................. 129

6.2.5 Data analysis .................................................................................................. 131
6.3

RESULTS ...................................................................................................... 133

6.3.1 Baseline characteristics ................................................................................. 133
6.3.2 Decision curve analysis .................................................................................. 134
6.4

DISCUSSION ............................................................................................... 141

CHAPTER 7.
PREDICTION OF HIP FRACTURE IN POST-MENOPAUSAL
WOMEN USING ARTIFICIAL NEURAL NETWORK APPROACH ................... 144
7.1.

INTRODUCTION ........................................................................................ 147

7.2.

METHODOLOGY ....................................................................................... 148

xv


7.2.1. Study design and settings ............................................................................... 148
7.2.2. Measurements................................................................................................. 149
7.2.3. Building of ANN ............................................................................................. 150
7.2.4. Evaluation of ANN model performance ......................................................... 154
7.3.

RESULTS ...................................................................................................... 156

7.3.1. Baseline characteristics ................................................................................. 156
7.3.2. Prediction of hip fracture by ANN ................................................................. 161
7.3.3. Comparison between ANN model and other models ...................................... 165
7.3.4. Relative importance evaluation ...................................................................... 168
7.4.

DISCUSSION ............................................................................................... 169

7.5.

CONCLUSION ............................................................................................. 171

CHAPTER 8.
8.1.

CONCLUSIONS AND FUTURE DIRECTION ............................ 173

SUMMARY................................................................................................... 174

8.1.1. Construction of an “Osteogenomic Profile” (GRS) ....................................... 175
8.1.2. Association between GRS and BMD. ............................................................. 176
8.1.3. Contribution of GRS into the prediction of fracture risk ............................... 177
8.1.4. Contribution of GRS in the prediction of bone loss. ...................................... 178
8.1.5. Clinical utility of “Osteogenomic Profilie” ................................................... 179
8.1.6. Prediction of hip fracture in post-menopausal women using artificial neural
network approach. .......................................................................................... 179
8.2.

CONCLUSIONS AND FUTURE DIRECTION........................................ 180

8.2.1. Rare variants .................................................................................................. 181
8.2.2. Gene-gene interactions................................................................................... 182
8.2.3. Gene-environment interactions ...................................................................... 183
8.2.4. Epigenetics ..................................................................................................... 184
8.2.5. Application of machine learning approaches................................................. 186
8.2.6. Clinical utility of machine learning predictive models .................................. 187

xvi


8.2.7. Future of the application of genetics in individualised prognosis of fracture
using machine learning approaches............................................................... 188
APPENDIX A: SUPPLEMENTAL TABLES ............................................................. 190
APPENDIX B: QUESTIONAIRES ............................................................................. 204
REFERENCES ............................................................................................................... 219

xvii


LIST OF FIGURES
Figure 2-1: Normal bone (left) and osteoporotic bone (right) ........................................ 19
Figure 2-2: The online tool - Garvan Fracture Risk Calculator ...................................... 36
Figure 2-3: The online tool (FRAX®) for calculation of 10-year absolute risk of fracture
(assessed in September 2018) ......................................................................................... 38
Figure 4-1: Relationship between GRS.FN and femoral neck BMD (left panel) and
lumbar spine (right panel) for men and women. ............................................................. 81
Figure 4-2: Cumulative risk of fracture for women and men stratified by median of

GRS.FN. .......................................................................................................................... 85
Figure 4-3: Calibration analysis: predicted and observed probability of fracture for
model with clinical risk factors and GRS. ...................................................................... 89
Figure 5-1: Rate of bone mineral density loss for men and women ............................. 106
Figure 5-2: Association between GRS and rate of BMD loss in men and women ....... 112
Figure 5-3: Osteogenomic profile stratified by non-rapid bone loss and rapid bone loss
groups in women and men. ........................................................................................... 118
Figure 6-1: Net benefit curves for women along threshold of (A) 0-30% and (B) 1530%. .............................................................................................................................. 139
Figure 6-2: Net benefit curves for men along threshold of (A) 0-30% and (B) 15-30%.
....................................................................................................................................... 140
Figure 7-1: A typical Artificial Neural Network with 1 input layer, 1 hidden layer, and 1
output layer.................................................................................................................... 151
Figure 7-2: Boxplot of key risk factors stratified by hip tracture status; N: Non-fracture
group; Y: Hip fracture group......................................................................................... 159
Figure 7-3: Pairwise correlation betwee risk factors. ................................................... 160
Figure 7-4: Discrimination of hip fracture in three ANN models ................................. 162
Figure 7-5: ANN model with BMD and non-BMD risk factors ................................... 163
Figure 7-6: Discrimination of predictive models .......................................................... 167
Figure 7-7: Relative importance of risk factors in ANN Model III .............................. 168

xviii


LIST OF TABLES
Table 2-1: Classification of bone density based on T-score. .......................................... 20
Table 2-2: Classification of risk factors for fracture ....................................................... 24
Table 4-1: Key baseline characteristics of 902 women and 557 men ............................. 75
Table 4-2: Baseline characteristics of 902 women stratified by fracture status .............. 77
Table 4-3: Baseline characteristics of 557 men stratified by fracture status ................... 78
Table 4-4: Association between polygenic score (GRS) and bone mineral density in

men and women: Results of multiple linear regression analysis .................................... 80
Table 4-5: Association between genetic risk score and fracture risk: Results of the Cox's
proportional hazards analysis .......................................................................................... 84
Table 4-6: Concordance index (and 95% confidence interval) of predicted and observed
fracture risk for model with GRS with and without adjustment for covariates .............. 87
Table 4-7: Reclassification analysis: Percent of reclassification compared with model
with clinical risk factors .................................................................................................. 88
Table 5-1: Number of BMD measurements during up to 27-yr follow-up ................... 105
Table 5-2: Baseline characteristics of 860 women and 524 men .................................. 107
Table 5-3: Determinants of rate of bone loss: Univariate linear regression analysis.... 110
Table 5-4: Determinants of rate of bone loss: Multiple linear regression analysis ...... 111
Table 5-5: Mixed-effects analysis on 860 women ........................................................ 113
Table 5-6: Baseline characteristics of 860 women and 524 men stratified by rate of bone
loss ................................................................................................................................ 115
Table 5-7: Risk of rapid bone loss in women: Multiple logistic regression analysis ... 116
Table 5-8: Risk of rapid bone loss in men: Multiple logistic regression analysis ........ 117
Table 6-1: Baseline characteristics of participants........................................................ 136
Table 6-2: Calculated net benefit of the four models using different treatment threshold
....................................................................................................................................... 137
Table 6-3: Net reduction in unnecessary osteoporosis treatments ................................ 138
Table 7-1: KEY BASELINE CLINICAL CHARACTERISTICS STRATIFIED BY 10YEAR HIP FRACTURE STATUS .............................................................................. 158
Table 7-2: Neural network classification results ........................................................... 164
Table 7-3: Predictive performance of four algorithms for predicting hip fracture ....... 166

xix


LIST OF ABBREVIATIONS
ANOVA

analysis of variance

ANN

artificial neural network

AR

absolute risk

AUC

area under the receiver operating characteristic curves

BMC

bone mineral content

BMD

bone mineral density

CI

confidence interval

DNA

deoxyribo-nucleic acid

DOES

Dubbo Osteoporosis Epidemiology Study

DXA

dual X-ray absorptiometry

DZ

dizygotic twins

FNBMD

femoral neck bone mineral density

GWAS

genome-wide association study

HR

hazard ratio

HWE

Hardy Weinberg equilibrium

LD

linkage disequilibrium

LOD

logarithm of the odds

LSBMD

lumbar spine bone mineral density

MAF

minor allele frequency

MZ

monozygotic twins

OR

odds ratio

PAR

population attributable risk

PCR

polymerase chain reaction

QCT

quantitative computed tomography

QUS

quantitative ultrasound

RCT

randomised controlled trial

RNA

ribonucleic acid

ROC

receiver operating characteristic curve

RR

relative risk

SD

standard deviation

SNP

single nucleotide polymorphism
xx


CHAPTER 1 - INTRODUCTION

CHAPTER 1. INTRODUCTION AND
BACKGROUND

Osteoporosis is a skeletal disease that affects a large proportion of the elderly
population. In Australia, osteoporosis affects 1.2 million people aged 50 and older, and
most of them do not have a diagnosis until the occurrence of a bone fracture (Henry et
al., 2011). Individuals with an osteoporotic fracture are more likely to have re-fracture,
recurrent falls, less mobility, and excess mortality. Given the aging of the population, it
is important to predict fracture early for appropriate intervention. However, the prediction
is difficult because osteoporotic fracture is affected by multiple risk factors, including
anthropometric characteristics, clinical risk factors, and genes. Existing tools for fracture
risk assessment based on non-genetics risk factors have good performance, but there is
room for further improvement. Genome-wide association studies (GWAS) have
identified many genetic variants that are associated with low bone mineral density
(BMD), but the utility of these genes in terms of fracture prediction is not clear. Because
low BMD is associated with higher risk of fracture, improving the accuracy of fracture
risk assessment by using of BMD-related genetic variants is a promising approach.
Chapter 1 provides a brief background of osteoporosis and osteoporotic fracture,
highlighting the seriousness of osteoporosis, followed by the overall structure of the thesis
research. Specifically, section 1 discusses various definitions of osteoporosis and
osteoporotic fracture, their risk factors and current predictive models. Section 2 provides
the motivation for my doctoral research and the need for improvement of fracture risk
assessment. The aims and contributions of my thesis are presented in sections 3 and 4 of
this chapter. The overall structure of the thesis and key research findings are provided in

the last section of Chapter I.

1


CHAPTER 1 - INTRODUCTION

1.1

BACKGROUND
Osteoporosis is defined as “a skeletal disorder characterized by compromised

bone strength predisposing a person to an increased risk of fracture” (NIH, 2001). In this
definition, bone strength refers to both bone quality and bone density. Bone quality is
commonly assessed by micro-architecture, bone turnover and mineralisation, whereas
bone density, also called “bone mineral density” (BMD), can be measured as the amount
of bone mass of mineral per defined area of a bone. In the sense of physics, BMD is the
amount of mineral per volume of bone in a specific site. However, the current technology
can only measure bone density per squared-centimetre of bone surface, commonly at the
lumbar spine, femoral neck, or hip (Kanis et al., 1994; Seeman, 2002). Currently, bone
densitometry is the “gold stand” technology for bone density measurement, where dualenergy X-ray absorptiometry (DXA) is used to quantify the amount of X-ray absorption
by minerals inside the bone (Watts, 2004). DXA is quick, painless, non-invasive and is
regarded as the most accurate method for diagnosing osteoporosis.
The World Health Organisation defined osteoporosis based on BMD.
Specifically, a BMD T-score is operationally derived from the difference between an
individual’s BMD and the peak BMD of young healthy adult reference population,
divided by standard deviation. The Third National Health and Nutrition Examination
Survey (NHANES III) reference database in White women aged 20–29 years was
recommended as reference range for femoral neck measurements (Kanis, Johnell, et al.,
2008). DXA instruments have been programmed to compute these values using predefined references. WHO defined osteoporosis as a value for BMD at the femoral neck

of 2.5 SD or more below the young female adult mean (T-score ≤ -2.5) (WHO, 1994).
The T-score femoral neck BMD ≤ -2.5 indicates severe osteoporosis in the presence of
2


CHAPTER 1 - INTRODUCTION

one or more fragility fractures. Data from an Australian study (Henry, 2011) showed that
approximately 6% of men and 23% of women over 50 years old having osteoporosis, and
these numbers increased to 13% and 43%, respectively, for those over 70 years old.
The ultimate outcome of osteoporosis is osteoporotic (or fragility) fracture.
Operationally, osteoporotic fracture is defined as any loss of continuity of the bone at any
sites. The fracture is commonly occurred in people aged 50 or above, associated with low
BMD, and caused by low trauma such as fall from standing height or less, and not related
to underlying diseases (bone or cancer) (Kanis et al., 2001). However, this definition may
underestimate the prevalence of osteoporotic fracture in the community; because fractures
due to high trauma could also be osteoporotic (Huang et al., 1996; Nguyen et al., 2007;
Nguyen et al., 1997a). The most common sites of bone affected by osteoporosis are the
hip, spine, and distal forearm (Cummings & Melton, 2002; Johnell & Kanis, 2005).
One of the most important risk factors for fracture is advancing age (Chang et
al., 2004; Eisman, et al., 1994; Sanders et al., 1999). Compared with pre-menopausal
women, those aged 60 and above have higher risk of fracture, with the odds ratios being
10.2 (Jiang et al., 2013). In Caucasian women, every five-years of advancing age is
associated with a 2-fold increase in the risk of fracture (Nguyen et al., 2007; Taylor et al.,
2004) and 1.5 folds higher risk of hip fracture (Cummings et al., 1995). In Caucasian
men, advancing age is associated with increased hip fracture risk at a similar magnitude
to that observed in women (Kanis et al., 2005). Moreover, the risk of fracture is
approximately 2.8-fold for humerus, and 1.6-fold for forearm for every 5 years increasing
age in men (Nguyen et al., 2001). The exact pathway of the association between aging
and fracture is still unknown, but it may relate to the cumulative deterioration of physical

health associated with aging (Kelsey & Samelson, 2009).
3


CHAPTER 1 - INTRODUCTION

Low BMD is another important risk factor for fracture. Each standard deviation
decrease in BMD is associated with 1.5 – 3.0 fold increase in the risk of fracture (Marshall
et al., 1996; Johnell et al., 2005; Nguyen et al., 1993). BMD is more predictive of hip
fracture than non-hip fracture, with the gradient of risk being 2.6 (Kanis et al., 2002).
However, (Siris et al., 2006) had indicated that even though the risk of fracture is much
higher in individuals with osteoporosis, the greatest absolute number of fractures occurred
in individuals with osteopenia, roughly five times more than individuals with
osteoporosis. This leads researchers to investigate more risk factors for fracture risk over
and above BMD measurement.
Gender, history of fall, and prior fracture are also major risk factors for fracture
(Nguyen et al., 2008). Women are more likely to have fracture (40-50% of lifetime risk)
at any sites compared with men (13-22%) (Johnell & Kanis, 2005). In addition, people
with a fall are more likely to have fracture, due to the strong association between fall and
weak muscle strength and increased body sways. Residents of nursing homes had had two
to six falls each year, with up to 5% resulting in fracture (Oliver et al., 2010; Rubenstein
et al., 1996). More importantly, the risk for an osteoporotic fracture is approximately

doubled in the presence of a prior fracture (Unnanuntana et al., 2010). The presence of
an asymptomatic vertebral fracture even leads to a higher (7- to 10-fold) increase risk of
subsequent vertebral fracture (Ross et al., 1991).
Apart from the factors discussed above, there are multiple risk factors, including
but not limited to, family history of fracture, lifestyle factors, nutrition, and genetic
factors. Individuals having frequent alcohol consumption, smoking, long-term use
glucocorticoid steroids, low body weight, low calcium intake, early menopause and

certain medications that interfere with bone metabolism (McCloskey et al., 2012). In
4