Accepted Manuscript Extra Virgin Olive Oil Consumption Reduces The Risk of Osteoporotic Fractures in The Predimed Trial J.F. García-Gavilán, M. Bulló, S. Canudas, M.A. Martínez-González, R. Estruch, S. Giardina, M. Fitó, D. Corella, E. Ros, J. Salas-Salvadó PII: S0261-5614(17)30006-7 DOI: 10.1016/j.clnu.2016.12.030 Reference: YCLNU 3024 To appear in: Clinical Nutrition Received Date: 13 September 2016 Revised Date: 13 December 2016 Accepted Date: 31 December 2016 Please cite this article as: García-Gavilán J, Bulló M, Canudas S, Martínez-González M, Estruch R, Giardina S, Fitó M, Corella D, Ros E, Salas-Salvadó J, Extra Virgin Olive Oil Consumption Reduces The Risk of Osteoporotic Fractures in The Predimed Trial, Clinical Nutrition (2017), doi: 10.1016/ j.clnu.2016.12.030. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT 1 EXTRA VIRGIN OLIVE OIL CONSUMPTION REDUCES THE RISK OF 1 OSTEOPOROTIC FRACTURES IN THE PREDIMED TRIAL 2 García-Gavilán JF1, Bulló M1,2, Canudas S1, Martínez-González MA2,3, Estruch R2,4, Giardina S1, 3 Fitó M5, Corella D2,6, Ros E2,7, Salas-Salvadó J1,2. 4 1 Human Nutrition Unit, Biochemistry and Biotechnology Department, Faculty of Medicine and 5 Health Sciences, University Hospital of Sant Joan de Reus, IISPV, Universitat Rovira i Virgili, 6 C/Sant Llorenç 21, 43201, Reus, Spain. 7 2 CIBERobn Physiopathology of Obesity and Nutrition, Instituto de Salud Carlos III, Madrid, 8 Spain. 9 3 Department of Preventive Medicine and Public Health, University of Navarra, Pamplona, Spain. 10 4 Department of Internal Medicine, August Pi i Sunyer Institute of Biomedical Research 11 (IDIBAPS), Hospital Clinic, University of Barcelona, Barcelona, Spain. 12 5 Cardiovascular Risk and Nutrition (Regicor Study Group), Hospital del Mar Medical Research 13 Institute, Barcelona Biomedical Research Park, Barcelona, Spain. 14 6 Department of Preventive Medicine, University of Valencia, Valencia, Spain. 15 7 Lipid Clinic, Endocrinology and Nutrition Service, IDIBAPS, Hospital Clinic, University of 16 Barcelona, Barcelona, Spain. 17 18 Correspondence: Mònica Bulló and Jordi Salas-Salvadó, Human Nutrition Unit, Faculty of 19 Medicine and Health Sciences, Universitat Rovira i Virgili, C/ SantLlorenc 21, 43201 Reus, Spain. 20 E-mail: monica.bullo@urv.cat, jordi.salas@urv.cat 21 M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT 2 ABSTRACT 22 Background & Aims 23 The incidence of osteoporotic fractures is lower in countries in the Mediterranean basin. Virgin 24 olive oil, a key component of the Mediterranean Diet (MDiet), with recognised beneficial 25 effects on metabolism and cardiovascular health, may decrease the risk of osteoporotic 26 fractures. The aim to this study was to explore the effect of chronic consumption of total olive 27 oil and its varieties on the risk of osteoporosis-related fractures in a middle-aged and elderly 28 Mediterranean population. 29 Methods. 30 We included all participants (n=870) recruited in the Reus (Spain) centre of the PREvención 31 con DIeta MEDiterránea (PREDIMED) trial. Individuals, aged 55-80 years at high 32 cardiovascular risk, were randomized to a MedDiet supplemented with extra-virgin olive oil, a 33 MedDiet supplemented with nuts, or a low-fat diet. The present analysis was an observational 34 cohort study nested in the trial. A validated food frequency questionnaire was used to assess 35 dietary habits and olive oil consumption. Information on total osteoporotic fractures was 36 obtained from a systematic review of medical records. The association between yearly repeated 37 measurements of olive oil consumption and fracture risk was assessed by multivariate Cox 38 proportional hazards. 39 Results. 40 We documented 114 incident cases of osteoporosis-related fractures during a median follow-up 41 of 8.9 years. Treatment allocation had no effect on fracture risk. Participants in the highest 42 tertile of extra-virgin olive oil consumption had a 51% lower risk of fractures (HR:0.49; 95% 43 CI:0.29-0.81. P for trend = 0.004) compared to those in the lowest tertile after adjusting for 44 potential confounders. Total and common olive oil consumption was not associated with 45 fracture risk. 46 Conclusions. 47 M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT 3 Higher consumption of extra-virgin olive oil is associated with a lower risk of osteoporosis-48 related fractures in middle-aged and elderly Mediterranean population at high cardiovascular 49 risk. 50 51 Keywords: Olive oil, Osteoporotic fractures, Prevention, Aging. 52 53 Abbreviations: MedDiet, Mediterranean diet; BMD, bone mineral density; MUFA, 54 monounsaturated fatty acids; PUFA, polyunsaturated fatty acids; EVOO, extra virgin olive oil; 55 FFQ, food frequency questionnaire; BMI, body mass index. 56 M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT 4 INTRODUCTION 57 Osteoporosis is an age-related progressive bone condition characterised by bone mass loss and 58 microarchitecture degradation that increase the risk of potentially serious fractures. It is a major 59 burden for health care systems as osteoporotic fractures and falls by osteoporotic fractures are 60 associated with a high dependence, morbidity and mortality.[1–3] Osteoporosis is estimated to 61 affect 27.5 million people (22 million women and 5.5 million men) aged between 50-84 years 62 worldwide and its prevalence is expected to rise to 33.9 million by 2025.[4] 63 Bone remodeling balance is affected by several factors, such as age, heredity or endocrine 64 diseases.[3] Lifestyle factors, such as smoking, physical activity and diet also affect bone 65 health.[5] Low calcium intake and low exposure to sunlight leading to reduced synthesis of 66 vitamin D have also been identified as common risk factors because of their role in bone mass 67 health.[1,6] In addition, other specific nutrients, foods, or dietary patterns can influence bone 68 health.[6–9] Adhering to a traditional Mediterranean diet (MedDiet), characterized by high 69 intake of fruits, vegetables, nuts and olive oil, has been linked to a lower risk of hip 70 fractures,[10–12] which might partly explain the epidemiological evidence of a geographical 71 variation in the incidence of hip fractures across Europe, the highest rates being in North Europe 72 and the lowest in the Mediterranean basin countries or in United States’ population where it was 73 associated a lower risk of hip fracture with MedDiet pattern.[11,13] These observations might 74 be attributed to the high content of monounsaturated fats (MUFA) and polyphenols in olive oil, 75 the main fat consumed in the Mediterranean diet. The intake of MUFA has been positively 76 correlated with bone mineral density (BMD) in the Greek and Spanish populations[14–16] and 77 higher circulating levels of bone remodelling osteocalcin have been reported after following a 78 MedDiet enriched with extra-virgin olive oil (EVOO).[17] Similarly, a high intake of olive 79 extract has also been linked to higher levels of osteocalcin and stabilization of bone mass loss in 80 osteopenic postmenopausal women.[18] 81 The effect of consumption of olive oil and its varieties on the risk of osteoporotic fractures has 82 not been studied. Our aim was to examine the association between the amount of total olive oil 83 and its varieties (extra virgin and common olive oil) consumed and the risk of osteoporotic 84 M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT 5 fractures in a sub-sample of middle-aged and elderly Mediterranean participants of the 85 PREDIMED trial. We hypothesized that higher consumption of EVOO containing high amounts 86 of polyphenols would reduce the risk of osteoporosis-related fractures. 87 MATERIALS AND METHODS. 88 Study design and population. 89 The present study was carried out in the framework of the PREDIMED study, a large, multi-90 centre, randomized and controlled parallel group trial aimed at assessing the effect of the 91 MedDiet on the primary prevention of cardiovascular diseases in Spain. This trial is registered 92 at http://www.controlled-trials.com as ISRCTN35739639. Osteoporotic fractures were assessed 93 only as part of an ancillary study including all participants (n=870) recruited in the 94 PREDIMED-Reus centre. Full details of the PREDIMED protocol are published elsewhere.[19] 95 Participants (men aged 55–80 years and women aged 60–80 years) were randomly assigned to 1 96 of 3 intervention groups: (1) a MedDiet supplemented with EVOO (MedDiet-EVOO group; 50 97 g or more per day), (2) a MedDiet supplemented with mixed nuts (MedDiet-Nuts; 30 g of nuts 98 daily), or advice on a low-fat diet (Control). Supplemental foods were given for free to 99 participants in the MedDiet groups, while those in the control diet group received non-food 100 gifts. Participants had no history of CVD at baseline but they were at high cardiovascular risk 101 because of the presence of type 2 diabetes or at least three of the following risk factors: current 102 smoker; hypertension; high levels of low-density lipoprotein cholesterol; low levels of high-103 density lipoprotein cholesterol; overweight or obesity and/or a family history of premature 104 cardiovascular disease. Participants excluded were those with a BMI greater than 40 kg/m2, 105 severe chronic illness, drug or alcohol addiction, history of allergy or intolerance to olive oil or 106 nuts, and/or a low predicted likelihood of changing dietary habits according to Prochaska and 107 DiClemente's stages-of-change model.[20] The local institutional review board approved the 108 study protocol, and all participants provided written informed consent. Recruitment took place 109 between 1st October, 2003, and 30th June, 2009 and the intervention was terminated in 2010 with 110 an extended follow-up to August 2015. The study was performed according to Declaration of 111 Helsinki about Ethical Principles for Medical Research Involving Human Subjects. 112 M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT 6 Measurements 113 At baseline and at each annual visit until the end of intervention in 2010, data on lifestyle 114 variables, medical conditions and medication use were recorded. Weight and height were 115 measured with light clothing and no shoes, using calibrated scales and a wall-mounted 116 stadiometer, respectively. Waist circumference was measured midway between the lowest rib 117 and the iliac crest using an anthropometric tape. Blood pressure was measured using a validated 118 oscillometer (Omron HEM705CP; Hoofddorp, The Netherlands) in triplicate with a five-minute 119 interval between each measurement, and the mean of these values was recorded. Trained 120 personnel took fasting blood samples for subsequent biochemical analysis. The validated 121 Spanish version of the Minnesota Leisure-Time Physical Activity questionnaire was given at 122 baseline and yearly.[21] 123 Dietary assessment. 124 A 137-item semi-quantitative validated food frequency questionnaire (FFQ) was given to all 125 participants at baseline and was repeated every year throughout the follow-up period.[22] 126 Energy and nutrient intake were estimated from Spanish food composition tables.[23,24] Data 127 regarding the consumption of different types of olive oil was obtained from the FFQ, which 128 included three different questions on the type of olive oil consumed: (1) EVOO (obtained only 129 by mechanically pressing the olives, acidity <1%), (2) refined oil (refined olive oil, acidity 130 <0.3%) and (3) pomace olive oil (obtained using solvents from the leftovers of pressing the 131 olives and mixed with other refined olive oils, acidity <0.3%). The number of 12 g tablespoons 132 was recorded for each variety in 9 frequency categories as follows: no consumption, one to three 133 times per month, n times per week (n = one, two to four or five to six) or n times per day (n = 134 one, two to three, four to six or more than six). The number of tablespoons stated was converted 135 into grams per day. One FFQ item asked about EVOO intake and two other items asked about 136 refined olive oil and pomace olive oil, and these two values were added together for common 137 olive oil intake. Total olive oil intake was then the sum of all three items. Using the Pearson 138 correlation coefficient (r), reproducibility and validity of the FFQ were 0.55 and 0.60, 139 respectively, for total olive oil consumption, and the intraclass correlation coefficients for 140 M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT 7 reproducibility and validity were 0.71 (P-value: <0.001) in a population similar to the 141 PREDIMED participants.[22] 142 A validated 14-item MedDiet screener was also administered to assess the degree of adherence 143 to the MedDiet.[25] Two of the 14 items were related to olive oil intake. To control for the 144 overall dietary pattern, the 2 items related to olive oil were removed from the total score; thus, a 145 12-point score was used as covariate in the models. 146 Outcome. 147 All osteoporotic fractures were adjudicated according to the criteria defined by Warriner and co-148 workers including fractures scoring over 5, representing those more likely due to osteoporosis 149 This score consider fracture risk groups according to sex, age and race, and scored from 1 to 9 150 with higher scores representing those fractures most likely due to osteoporosis.[26] This was 151 also selected in accordance with previous studies regarding new classification of osteoporotic 152 fractures beyond the classical ones (vertebral, hip and wrist-forearm).[27–29]. According to the 153 International Classification of Diseases Clinical Modification (ICD-CM), open clavicle (ICD-154 CM 810.1-810.3), phalanges (ICD-CM 816.1-816.13 AND 826.0-826.1), tarsal/metatarsal 155 (ICD-CM 825.0-825.39), scapula (ICD-CM 811.0-811.19), and skull/facial (ICD-CM 800.00-156 804.99) fractures were excluded.[26] Incident cases of osteoporotic fractures through 1st 157 December, 2010 were identified initially from a systematic, comprehensive and standardised 158 annual review of all outpatient and inpatient medical records of each participant. Information on 159 osteoporotic fractures was updated yearly using medical records. An independent researcher 160 confirmed all fracture events. 161 Statistical analyses. 162 Participants’ baseline characteristics were described with means (SD) and percentages 163 (number). To take advantage of the yearly dietary assessments, we averaged the food 164 consumption from the baseline to the end of the follow-up or to the last follow-up FFQ before 165 the occurrence of fractures. Then, participants were categorized into tertiles of total olive oil, 166 EVOO or common olive oil consumption using the mean value of all FFQs from the beginning 167 to the last before the incidence of fracture or the end of follow-up in those not suffering a 168 M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT 8 fracture. Dietary variables were adjusted for total energy intake using the residuals method[30] 169 and they are presented in accordance with energy-adjusted tertiles of EVOO intake. Follow-up 170 time was estimated as the interval from the beginning of the study up to the date of fracture 171 events, death (for any reason) or end of follow-up, whichever came first. 172 The associations between energy-adjusted tertiles of total olive oil consumption or its different 173 subtypes and the risk of osteoporotic fractures were assessed using time-dependent multivariate 174 Cox proportional hazards models. We tested the proportionality of hazards with the use log-rank 175 test. Results are expressed as hazard ratios (HRs) and 95% confidence intervals (CI). Model 1 176 was adjusted for age, sex, BMI, education level (primary education, secondary education, 177 academic/graduate), leisure time physical activity (metabolic equivalent of task (MET)-178 minutes/day), smoking status (never, former, current smoker) and the intervention group. As 179 other covariates can interfere with the risk of fractures, Model 2 was additionally adjusted for 180 prevalence of diabetes (yes/no), prevalence of previous documented osteoporotic fractures 181 (yes/no), use of insulin (yes/no), use of oral antidiabetic medications (yes/no), use of diuretic 182 drugs (yes/no), use of oral glucocorticoids (yes/no), use of anti-osteoporotic drugs (yes/no), use 183 of anticoagulants (yes/no), use of oestrogen (yes/no) and baseline MedDiet adherence (12-point 184 score). Covariates were selected based on their biological plausibility of having an association 185 with the risk of fractures. The same models (excluding the baseline 12-point score) were used to 186 assess the risk of osteoporotic fractures according to the dietary intervention group. The 187 associations between MUFA intake, polyunsaturated fatty acids (PUFA) intake and 188 MUFA:PUFA ratio with the risk of fractures were assessed using the covariates included into 189 the Model 3. Nelson-Alen estimator was used to analyse the increasing failure rates. Sensitivity 190 analysis was conducted excluding early cases observed during the first year of intervention. The 191 level of significance was P <0.05 for all statistical tests for bilateral contrast. Statistical analyses 192 were carried using SPSS 21.0 for windows (IBM, Chicago, IL, USA) and STATA 14 193 (StataCorp, College Station, TX). 194 RESULTS. 195 M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT 9 During a median of 5.2 years of intervention and 8.9 years of follow-up, we documented 114 196 incident cases of osteoporosis-related fractures (40 in MedDiet-EVOO group, 37 in MedDiet-197 Nuts group and 37 in control group). Tables 1 and 2 show the baseline anthropometric and 198 dietary characteristics of the study participants according to energy-adjusted tertiles of EVOO 199 consumption. There were not significant differences in age, sex, BMI, previous fractures, 200 prevalence of diabetes, medications, energy intake, protein intake, alcohol intake, vitamin D or 201 fermented dairy products intake between tertiles of EVOO consumption. The mean 202 consumption of total olive oil was 56.5 g/day in participants at the highest tertile and 37.6 g/day 203 in those in the lowest tertile. 204 According to the intervention group, no significant differences in the risk of osteoporotic 205 fractures were observed (HR (95%CI)) 1.13 (0.71-1.79) and 1.05 (0.66-1.67) in the MedDiet-206 EVOO and MedDiet-Nuts groups compared to control group (Supplemental file). 207 Figure 1 shows the survival curve of osteoporotic fractures and the number of participants at 208 risk by energy-adjusted EVOO tertiles at different time points. Table 3 shows the HR and 95% 209 CIs for the association between total olive oil consumption and the specific subtypes and 210 osteoporosis-related fractures. Total olive oil and common olive oil consumption were not 211 associated with a lower risk of fractures despite a non-significant trend to a lower reduction of 212 bone fracture risk was observed in subjects allocated in the highest tertiles of total olive oil 213 consumption. In contrast, a 51% reduction in the risk of osteoporosis-related fractures was 214 observed in the fully-adjusted model for individuals in the highest tertile of EVOO consumption 215 compared to the reference tertile (HR: 0.49; 95% CI: 0.29 to 0.81). The highest tertile compared 216 to the reference tertile of MUFA intake (HR: 1.04; 95% CI: 0.66 to 1.65), PUFA intake (HR: 217 1.20; 95% CI: 0.76 to 1.90) or the MUFA:PUFA ratio (HR: 0.87; 95% CI: 0.55 to 1.38) showed 218 no association with fracture risk. 219 The results of the sensitivity analysis were consistent with the general analysis. When early 220 cases occurred during the first year (7 events were excluded), the risk in the higher tertile of 221 EVOO consumption was relatively 46% lower (HR: 0.54; 95% CI: 0.32 to 0.92, P for trend = 222 0.050) than the reference tertile. 223 M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT 10 DISCUSSION 224 The novel finding of this longitudinal study in an older Mediterranean population at high risk 225 for cardiovascular disease is that high EVOO consumption is associated with a reduced risk of 226 osteoporotic fractures, whereas a non-significant trend to a lower risk was also observed for 227 total olive oil consumption. 228 The prevalence of osteoporosis and osteoporosis-related fractures is highly variable within 229 European regions, with the lowest prevalence in the Mediterranean area.[31] These differences 230 might be attributed to environmental factors and dietary regimens.[10–12,32] The MedDiet is 231 based on a combination of foods comprising a complex array of nutrients and bioactive 232 phytochemicals with anti-inflammatory, antioxidant and alkalinising properties that could all 233 contribute to bone health. Olive oil is one of the key foods in the MedDiet and its consumption 234 accounts for one to two thirds of total vegetable fat intake, where MUFA, in the form of oleic 235 acid, is the most abundant fatty acid consumed. In a cross-sectional study conducted in Greece, 236 MUFA intake was associated with a higher BMD.[33] Another study conducted in adult Greek 237 women found higher total and spine BMD in those whose diet contained a combination of olive 238 oil and fish with little meat, but not in association with the full MedDiet pattern.[34] A higher 239 dietary MUFA:PUFA ratio has also been related to a lower risk of osteoporotic-related fractures 240 produced by a same-level fall in elderly subjects.[16] 241 However, in the present study, we found no associations of MUFA intake or the MUFA:PUFA 242 ratio with fracture risk. These differences might be due to our study population displaying 243 narrow ranges of MUFA intake and the MUFA:PUFA ratio compared to previous studies. In 244 fact, results from prior studies showed no significant protection against fractures from MUFA 245 intake or MUFA:PUFA ratios in the ranges of our study population. Moreover, the differences 246 in the risk of osteoporosis-related fractures between different types of olive oil observed in our 247 study cannot be explained by differences in its fatty acid profile, as the fatty acid composition is 248 not affected by the extraction method used, since all olive oils are produced from the same 249 variety of olives.[35] This suggests that other compounds present in olive oil, beyond the fatty 250 acid composition, might play an important role in bone health. 251 M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT 11 Common olive oil is a mixture of virgin and (usually) more than 80% of refined oil, with fewer 252 antioxidant and anti-inflammatory compounds. In contrast, EVOO is the best quality oil, 253 produced by mechanically pressing ripe olives, and contains the highest amounts of bioactive 254 and antioxidant components, such as polyphenols, that by different mechanisms might exert 255 favourable effects on bone metabolism.[35] Several studies conducted in vitro and in animal 256 models have assessed the beneficial role of olive oil phenols on the formation and maintenance 257 of bone through its modulation of both bone cell differentiation and function.[36–38] 258 Oleuropein, tyrosol and hydroxytyrosol, the most abundant polyphenols in olive oil, have been 259 related to several beneficial effects on bone metabolism in vitro an in animal models.[39] In 260 humans, osteopenic subjects who consumed 250 mg/day of a polyphenol extract from Olea 261 europaea for 12 months significantly increased their osteocalcin levels and stabilized lumbar 262 spine BMD compared to a control group.[18] Similarly, in a prior PREDIMED sub-study, we 263 found higher serum levels of osteocalcin and the bone remodelling marker procollagen amino-264 terminal pro-peptide after 2 years of intervention with the MedDiet-EVOO compared to 265 theMedDiet-Nuts or the control diet.[17] In contrast, we found no significant protective effect 266 on bone fractures in subjects allocated to the MedDiet-EVOO group compared to the control 267 diet, as would be initially expected. This apparent discrepancy could be explained because the 268 difference in the total consumption of either total olive oil or extra-virgin olive oil between 269 participants in the MedDiet-EVOO group or control group was substantially lower than 270 differences between tertiles of olive oil consumption, as participants had a high MedDiet score 271 at baseline with olive oil as the main culinary fat. It is also plausible that exposure time to the 272 intervention diets was not long enough to improve or delay the age-related changes in bone 273 structure. Thus far, no other studies have been conducted to assess the relationship between 274 olive oil consumption and bone-related markers. Our findings extend the potential beneficial 275 role of EVOO consumption demonstrated on bone biochemical markers to a lower risk of 276 osteoporotic-related fractures as clinical outcome. Moreover, our results also suggest a 277 beneficial role of the phenolic compounds present in EVOO, as no association was found for the 278 common refined olive oil, which is depleted of these bioactive compounds. 279 M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT 12 The strengths of our study are a well-characterized cohort with long-term follow-up, controlled 280 by several potential confounders, the analysis of different varieties of olive oil and the use of 281 cumulative mean across all the available FFQs to improve the precision of the exposure. For 282 the fracture identification we used an objective score, however, this classification has some 283 potential limitations as was based on fracture categories identified by standard diagnostic codes 284 which identifies accurately a total of 94% of cases compared with the gold standard of medical 285 record review.[26] There are also limitations to our study. First, the generalizability of our 286 results may be limited, as the study population was made of older Mediterranean individuals at 287 high cardiovascular risk which increased their risk for osteoporotic fractures.[40] Second, 288 because of the observational nature of the study, residual confounding remains a possibility 289 even though our analyses were extensively adjusted for a wide range of potential confounders. 290 Third, no bone biochemical markers or data on BMD were available. Fourth, due to the low 291 number of fractures and the relative small study size, we cannot exclude a potential beneficial 292 effect of total olive oil consumption on the risk of bone fractures as the hazard ratio clearly 293 indicates a lower risk, although not strong as for EVOO. Finally, although the FFQ used was 294 validated, measurement errors cannot be discarded, especially regarding the self-reporting of 295 different varieties of olive oil. Still, our findings are consistent with the potential beneficial 296 effects of olive oil on bone health previously described. 297 In summary, we found that greater consumption of EVOO is associated with a lower risk of 298 osteoporosis-related fractures in an older Mediterranean population at high cardiovascular risk. 299 Our findings highlight the consumption of EVOO, one of the key foods of the MedDiet, in the 300 prevention of osteoporosis-related fractures. 301 FUNDING. 302 The Spanish Ministry of Health (ISCIII), PI1001407, Thematic Network G03/140, RD06/0045, 303 FEDER (Fondo Europeo de Desarrollo Regional), and the Centre Català de la Nutrició de 304 l’Institut d’Estudis Catalans funded part of this study. The Fundación Patrimonio Comunal 305 Olivarero and Hojiblanca SA (Málaga, Spain), California Walnut Commission (Sacramento, 306 CA), Borges SA (Reus, Spain), and Morella Nuts SA (Reus, Spain) donated the olive oil, 307 M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT 13 walnuts, almonds and hazelnuts, respectively, used in the study. We want to acknowledge their 308 collaboration. No funding sources played any role in the design, collection, analysis or 309 interpretation of the data or in the decision to submit the manuscript for publication. CIBERobn 310 is an initiative of ISCIII, Spain. 311 AUTHORS' CONTRIBUTIONS. 312 The authors’ responsibilities were as follows—MB, MAM, RE, MF, DC, ER and JS-S: 313 contributed to the conception, design, and implementation of the project; JGG, SC, SG and MB 314 contributed to data collection and analytical procedures; JGG, SC, SG and MB: conducted the 315 statistical analysis, interpreted data, and wrote the manuscript; and all authors: read and 316 approved the final version of the manuscript. 317 COMPETING INTERESTS. 318 JSS reports grants from RTIC G03/140 ISCIII, Spain, grants from CIBER obn ISCIII, Spain, 319 other from California Walnut Commission, Sacramento CA, USA, other from Patrimonio 320 Comunal Olivarero, Spain, other from La Morella Nuts, Spain, other from Borges S.A., Spain, 321 other from Nut and Dried Fruit Foundation, personal fees from Nuts for Life, other from Nut 322 and Dried Fruit Foundation, other from Nut and Dried Fruit Foundation, during the conduct of 323 the study; personal fees from Danone S.A., personal fees from Font Vella Lanjaron, personal 324 fees from Eroski Distributors and personal fees from Instituto Danone, outside the submitted 325 work. 326 ER reports grants, non-financial support and other from California Walnut Commission, grants, 327 personal fees, non-financial support and other from Merck, Sharp & Dohme, grants, personal 328 fees, non-financial support and other from Alexion, personal fees, non-financial support and 329 other from Aegerion, grants and personal fees from Sanofi Aventis, grants, personal fees, non-330 financial support and other from Ferrer International, grants from Amgen, grants from Pfizer 331 and personal fees from Akcea, outside the submitted work. 332 None of the other authors had a personal or financial conflict of interest. 333 ACKNOWLEDGEMENTS. 334 M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT 14 We thank all the participants and the PREDIMED personnel, including all the staff of the 335 primary centre, for their enthusiastic collaboration in the PREDIMED study. 336 M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT 15 REFERENCES. 337 [1] Azagra R, López-Expósito F, Martin-Sánchez JC, Aguyé A, Moreno N, Cooper C, et al. 338 Changing trends in the epidemiology of hip fracture in Spain. Osteoporos Int 339 2014;25:1267–74. doi:10.1007/s00198-013-2586-0. 340 [2] Cummings SR, Melton LJ. Epidemiology and outcomes of osteoporotic fractures. Lancet 341 2002;359:1761–7. doi:10.1016/S0140-6736(02)08657-9. 342 [3] Schürer C, Wallaschofski H, Nauck M, Völzke H, Cummings SR, Melton LJ, et al. 343 Fracture Risk and Risk Factors for Osteoporosis. Dtsch Arztebl Int 2015;112:365–72. 344 doi:10.3238/arztebl.2015.0365. 345 [4] Hernlund E, Svedbom A, Ivergård M, Compston J, Cooper C, Stenmark J, et al. 346 Osteoporosis in the European Union: medical management, epidemiology and economic 347 burden. A report prepared in collaboration with the International Osteoporosis 348 Foundation (IOF) and the European Federation of Pharmaceutical Industry Associations 349 (EFPIA). Arch Osteoporos 2013;8:136. doi:10.1007/s11657-013-0136-1. 350 [5] Zhu K, Prince RL. Lifestyle and osteoporosis. Curr Osteoporos Rep 2015;13:52–9. 351 doi:10.1007/s11914-014-0248-6. 352 [6] Trzeciakiewicz A, Habauzit V, Horcajada M-N. When nutrition interacts with osteoblast 353 function: molecular mechanisms of polyphenols. Nutr Res Rev 2009;22:68–81. 354 doi:10.1017/S095442240926402X. 355 [7] Shen C-L, Chyu M-C, Yeh JK, Zhang Y, Pence BC, Felton CK, et al. Effect of green tea 356 and Tai Chi on bone health in postmenopausal osteopenic women: a 6-month 357 randomized placebo-controlled trial. Osteoporos Int 2012;23:1541–52. 358 doi:10.1007/s00198-011-1731-x. 359 [8] Dong H, Hutchins-Wiese H, Kleppinger A, Annis K, Liva E, Lammi-Keefe C, et al. 360 Effects of Omega-3 Polyunsaturated Fatty Acid Supplementation on Bone Turnover in 361 M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT 16 Older Women. Int J Vitam Nutr Res Int Zeitschrift Für Vitamin- Und 362 Ernährungsforschung J Int Vitaminol Nutr 2014;84:124–32. doi:10.1024/0300-363 9831/a000199. 364 [9] Willett WC, B H, ML M, SE C, DD W, A T, et al. Mediterranean Diet and Fracture 365 Risk. JAMA Intern Med 2016;176:652. doi:10.1001/jamainternmed.2016.0494. 366 [10] Benetou V, Orfanos P, Pettersson-Kymmer U, Bergström U, Svensson O, Johansson I, et 367 al. Mediterranean diet and incidence of hip fractures in a European cohort. Osteoporos 368 Int 2013;24:1587–98. doi:10.1007/s00198-012-2187-3. 369 [11] Haring B, Crandall CJ, Wu C, LeBlanc ES, Shikany JM, Carbone L, et al. Dietary 370 Patterns and Fractures in Postmenopausal Women: Results From the Women’s Health 371 Initiative. JAMA Intern Med 2016;176:645–52. doi:10.1001/jamainternmed.2016.0482. 372 [12] Byberg L, Bellavia A, Larsson SC, Orsini N, Wolk A, Michaëlsson K. Mediterranean 373 Diet and Hip Fracture in Swedish Men and Women. J Bone Miner Res 2016. 374 doi:10.1002/jbmr.2896. 375 [13] Dhanwal DK, Dennison EM, Harvey NC, Cooper C. Epidemiology of hip fracture: 376 Worldwide geographic variation. Indian J Orthop 2011;45:15–22. doi:10.4103/0019-377 5413.73656. 378 [14] Trichopoulou A, Lagiou P. Healthy traditional Mediterranean diet: an expression of 379 culture, history, and lifestyle. Nutr Rev 1997;55:383–9. 380 [15] Rivas A, Romero A, Mariscal-Arcas M, Monteagudo C, Feriche B, Lorenzo ML, et al. 381 Mediterranean diet and bone mineral density in two age groups of women. Int J Food Sci 382 Nutr 2013;64:155–61. doi:10.3109/09637486.2012.718743. 383 [16] Martínez-Ramírez MJ, Palma S, Martínez-González M a, Delgado-Martínez a D, de la 384 Fuente C, Delgado-Rodríguez M. Dietary fat intake and the risk of osteoporotic fractures 385 in the elderly. Eur J Clin Nutr 2007;61:1114–20. doi:10.1038/sj.ejcn.1602624. 386 M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT 17 [17] Fernández-Real JM, Bulló M, Moreno-Navarrete JM, Ricart W, Ros E, Estruch R, et al. 387 A mediterranean diet enriched with olive oil is associated with higher serum total 388 osteocalcin levels in elderly men at high cardiovascular risk. J Clin Endocrinol Metab 389 2012;97:3792–8. doi:10.1210/jc.2012-2221. 390 [18] Filip R, Possemiers S, Heyerick A, Pinheiro I, Raszewski G, Davicco M-J, et al. Twelve-391 month consumption of a polyphenol extract from olive (Olea europaea) in a double 392 blind, randomized trial increases serum total osteocalcin levels and improves serum lipid 393 profiles in postmenopausal women with osteopenia. J Nutr Health Aging 2015;19:77–86. 394 doi:10.1007/s12603-014-0480-x. 395 [19] Martínez-González MÁ, Corella D, Salas-Salvadó J, Ros E, Covas MI, Fiol M, et al. 396 Cohort profile: design and methods of the PREDIMED study. Int J Epidemiol 397 2012;41:377–85. doi:10.1093/ije/dyq250. 398 [20] Nigg CR, Burbank PM, Padula C, Dufresne R, Rossi JS, Velicer WF, et al. Stages of 399 change across ten health risk behaviors for older adults. Gerontologist 1999;39:473–82. 400 [21] Elosua R, Marrugat J, Molina L, Pons S, Pujol E. Validation of the Minnesota Leisure 401 Time Physical Activity Questionnaire in Spanish men. The MARATHOM Investigators. 402 Am J Epidemiol 1994;139:1197–209. 403 [22] Fernández-Ballart JD, Piñol JL, Zazpe I, Corella D, Carrasco P, Toledo E, et al. Relative 404 validity of a semi-quantitative food-frequency questionnaire in an elderly Mediterranean 405 population of Spain. Br J Nutr 2010;103:1808–16. doi:10.1017/S0007114509993837. 406 [23] Mataix J. Tablas de Composición de Alimentos. 4th ed. Granada, Spain: Universidad de 407 Granada; 2003. 408 [24] Moreiras O, Carvajal A, Cabrera L, editors. Tablas de Composición de Alimentos [Food 409 Composition Tables]. 9th ed. Madrid, Spain: Ediciones Pirámide; 2005. 410 [25] Schröder H, Fitó M, Estruch R, Martínez-González MA, Corella D, Salas-Salvadó J, et 411 M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT 18 al. A short screener is valid for assessing Mediterranean diet adherence among older 412 Spanish men and women. J Nutr 2011;141:1140–5. doi:10.3945/jn.110.135566. 413 [26] Warriner AH, Patkar NM, Curtis JR, Delzell E, Gary L, Kilgore M, et al. Which 414 fractures are most attributable to osteoporosis? J Clin Epidemiol 2011;64:46–53. 415 doi:10.1016/j.jclinepi.2010.07.007. 416 [27] D’Elia G, Roselli G, Cavalli L, Innocenti P, Brandi ML. Severe osteoporosis: diagnosis 417 of non-hip non-vertebral (NHNV) fractures. Clin Cases Miner Bone Metab 2010;7:85–418 90. 419 [28] Stone KL, Seeley DG, Lui L-Y, Cauley JA, Ensrud K, Browner WS, et al. BMD at 420 multiple sites and risk of fracture of multiple types: long-term results from the Study of 421 Osteoporotic Fractures. J Bone Miner Res 2003;18:1947–54. 422 doi:10.1359/jbmr.2003.18.11.1947. 423 [29] Seeley DG, Browner WS, Nevitt MC, Genant HK, Scott JC, Cummings SR. Which 424 fractures are associated with low appendicular bone mass in elderly women? The Study 425 of Osteoporotic Fractures Research Group. Ann Intern Med 1991;115:837–42. 426 [30] Willett W, Howe G, Kushi L. Adjustment for total energy intake in epidemiologic 427 studies. Am J Clin Nutr 1997;65:1220S–1228. 428 [31] García-Martínez O, Rivas A, Ramos-Torrecillas J, De Luna-Bertos E, Ruiz C. The effect 429 of olive oil on osteoporosis prevention. Int J Food Sci Nutr 2014;65:834–40. 430 doi:10.3109/09637486.2014.931361. 431 [32] Feart C, Lorrain S, Ginder Coupez V, Samieri C, Letenneur L, Paineau D, et al. 432 Adherence to a Mediterranean diet and risk of fractures in French older persons. 433 Osteoporos Int 2013;24:3031–41. doi:10.1007/s00198-013-2421-7. 434 [33] Trichopoulou A, Georgiou E, Bassiakos Y, Lipworth L, Lagiou P, Proukakis C, et al. 435 Energy Intake and Monounsaturated Fat in Relation to Bone Mineral Density among 436 M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT 19 Women and Men in Greece 1. Prev Med (Baltim) 1997;26:395–400. 437 [34] Kontogianni MD, Melistas L, Yannakoulia M, Malagaris I, Panagiotakos DB, 438 Yiannakouris N. Association between dietary patterns and indices of bone mass in a 439 sample of Mediterranean women. Nutrition 2009;25:165–71. 440 doi:10.1016/j.nut.2008.07.019. 441 [35] Gimeno E, Castellote AI, Lamuela-Raventós RM, De la Torre MC, López-Sabater MC. 442 The effects of harvest and extraction methods on the antioxidant content (phenolics, α-443 tocopherol, and β-carotene) in virgin olive oil. Food Chem 2002;78:207–11. 444 doi:10.1016/S0308-8146(01)00399-5. 445 [36] García-Martínez O, De Luna-Bertos E, Ramos-Torrecillas J, Ruiz C, Milia E, Lorenzo 446 ML, et al. Phenolic Compounds in Extra Virgin Olive Oil Stimulate Human Osteoblastic 447 Cell Proliferation. PLoS One 2016;11:e0150045. doi:10.1371/journal.pone.0150045. 448 [37] Honda Y, Tanaka T, Tokuda T, Kashiwagi T, Kaida K, Hieda A, et al. Local Controlled 449 Release of Polyphenol Conjugated with Gelatin Facilitates Bone Formation. Int J Mol 450 Sci 2015;16:14143–57. doi:10.3390/ijms160614143. 451 [38] Shen C-L, Chyu M-C, Cao JJ, Yeh JK. Green tea polyphenols improve bone 452 microarchitecture in high-fat-diet-induced obese female rats through suppressing bone 453 formation and erosion. J Med Food 2013;16:421–7. doi:10.1089/jmf.2012.0199. 454 [39] Chin K-Y, Ima-Nirwana S. Olives and Bone: A Green Osteoporosis Prevention Option. 455 Int J Environ Res Public Health 2016;13:755. doi:10.3390/ijerph13080755. 456 [40] Sennerby U, Melhus H, Gedeborg R, Byberg L, Garmo H, Ahlbom A, et al. 457 Cardiovascular diseases and risk of hip fracture. JAMA 2009;302:1666–73. 458 doi:10.1001/jama.2009.1463. 459 460 1 M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT 20 462 Table 1. Baseline characteristics of study participants according to energy-adjusted tertiles of extra virgin olive oil consumption. Variable T1 (n= 290) T2 (n= 290) T3 (n= 290) Age (years)† 67 ± 6 68 ± 6 67 ± 6 Men, % (n) 46.6 (135) 42.1 (122) 45.9 (133) BMI (kg/m2)† 29.7 ± 3.2 29.5 ± 3.2 29.6 ± 3.4 Waist circumference (cm)† 101.9 ± 9.00 100.6 ± 8.3 101.1 ± 9.1 Leisure-time energy expenditure 255.1 ± 265.8 286.3 ± 281.3 244.2 ± 239.8 in physical activity (MET minutes/day)† Smoking status, % (never, current, former) 59.3, 12.8, 27.9 61.7, 14.1, 24.2 64.8, 9.3, 25.9 Educational level, % (n) Primary education 5.9 (17) 6.6 (19) 6.6 (19) Secondary education 14.5 (42) 18.3 (53) 20.0 (58) Academic/graduate 79.6 (231) 75.2 (218) 73.4 (213) History of osteoporotic fractures, % (n) 18.3 (53) 14.5 (42) 19.3 (56) Diabetes, % (n) 51.0 (148) 49.3 (143) 55.5 (161) Hypertension, % (n) 85.2 (247) 86.2 (250) 85.5 (248) Medication use, % (n) Diuretics 26.9 (78) 23.8 (69) 23.8 (69) Insulin 5.2 (15) 5.9 (17) 6.6 (19) Oral glucocorticoids 1.4 (4) 1.0 (3) 1.7 (5) Osteoporosis drugs 9.7 (28) 11.0 (32) 13.1 (38) Oral anticoagulants 1.4 (4) 1.4 (4) 0.3 (1) Oral antidiabetic drugs 36.2 (105) 30.3 (88) 37.2 (108) Oestrogens 1.7 (5) 2.8 (8) 2.4 (7) † Data are expressed as means ± SD. BMI, body mass index; MET, Metabolic Equivalent of Task. M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT 21 Table 2. Baseline dietary characteristics of study participants according to energy-adjusted tertiles of extra virgin olive oil consumption Variable T1 (n=290) T2 (n=290) T3 (n=290) Nutrients Total energy intake (kcal/day)† 2314.3 ± 625.2 2327.2 ± 580.7 2291.5 ± 571.6 Proteins (g/day)† 95.1 ± 21.8 96.3 ± 22.5 93.2 ± 22.4 Carbohydrates (g/day)† 240.9 ± 79.5 234.3 ± 73.4 219.7 ± 67.0 Total fat (g/day)† 100.4 ± 30.2 105.0 ± 30.1 108.9 ± 30.5 Saturated fatty acids (g/day)† 26.9 ± 9.3 27.8 ± 9.3 27.5 ± 9.4 Monounsaturated fatty acids (g/day)† 48.5 ± 15.8 52.4 ± 16.2 56.4 ± 15.9 Polyunsaturated fatty acids (g/day)† 16.7 ± 6.8 16.5 ± 5.9 16.4 ± 6.1 Fibre (g/day)† 22.6 ± 6.9 24.1 ± 8.5 23.6 ± 7.6 Alcohol (g/day)† 9.5 ± 14.9 8.5 ± 13.6 8.4 ± 12.6 Vitamin D (µg/day)† 5.8 ± 3.4 6.0 ± 3.6 5.7 ± 3.1 Calcium (mg/day)† 1044.6 ± 362.7 1051.3 ± 364.4 992.3 ± 341.7 Food Total olive oil (g/day)† 34.9 ± 16.9 40.8 ± 17.6 48.0 ± 15.9 Extra virgin olive oil (g/day)† 20.0 ± 19.0 35.2 ± 19.2 46.3 ± 17.3 Common olive oil (g/day)† 14.8 ± 19.4 5.3 ± 12.7 1.5 ± 6.1 Legumes (g/day)† 17.7 ± 8.0 18.1 ± 9.2 17.4 ± 8.5 Vegetables (g/day)† 284.7 ± 116.3 313.2 ± 137.0 322.1 ± 134.9 Cereals (g/day)† 256.6 ± 101.7 254.0 ± 98.0 238.1 ± 90.4 Fruit (g/day)† 299.5 ± 178.0 315.6 ± 177.2 319.3 ± 160.7 No fermented dairy (g/day)† 274.83 ± 186.98 258.26 ± 186.93 234.96 ± 173.84 Fermented dairy (g/day)† 114.66 ± 96.61 113.54 ± 95.39 105.29 ± 89.80 Meat (g/day)† 142.7 ± 54.9 146.6 ± 55.7 146.3 ± 65.5 Fish (g/day)† 101.0 ± 42.0 103.3 ± 45.6 102.1 ± 42.8 Nuts (g/day)† 10.7 ± 12.6 14.3 ± 14.9 13.6 ± 15.5 M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT 22 463 Modified MedDiet score (12-point score) 6.4 ± 1.6 6.6 ± 1.8 6.6 ± 1.7 † Data are expressed as means ± SD. M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT 23 Table 3. Risk of osteoporotic fracture according to energy-adjusted tertiles of cumulative olive oil intake T1 (n = 290) T2 (n = 290) T3 (n = 290) P for trend Mean total olive oil intake (g/day) 37.60 ± 6.76 48.23 ± 1.99 56.52 ± 4.32 Fracture event, % (n) 13.80 (40) 13.80 (40) 11.70 (34) Mean total energy intake (kcal/day) 2240.19 ± 450.80 2254.16 ± 354.91 2236.28 ± 361.21 Crude model 1 (Ref.) 0.93 (0.60, 1.44) 0.81 (0.51, 1.27) 0.367 Multivariate model 1ᵃ 1 (Ref.) 0.78 (0.49, 1.23) 0.73 (0.45, 1.19) 0.202 Multivariate model 2ᵇ 1 (Ref.) 0.74 (0.47, 1.18) 0.69 (0.42, 1.14) 0.141 Mean common olive oil intake (g/day) -0.13 ± 0.12 0.63 ± 0.85 12.49 ± 8.90 Mean total energy intake (kcal/day) 2000.56 ± 209.33 2516.34 ± 352.67 2213.73 ± 396.01 Fracture event, % (n) 15.90 (46) 10.30 (30) 13.10 (38) Crude model 1 (Ref.) 0.63 (0.40, 1.00) 0.81 (0.53, 1.25) 0.950 Multivariate model 1ᵃ 1 (Ref.) 0.88 (0.54, 1.42) 0.94 (0.61, 1.46) 0.955 Multivariate model 2ᵇ 1 (Ref.) 0.96 (0.59, 1.56) 1.00 (0.64, 1.55) 0.952 Mean extra-virgin olive oil intake (g/day) 28.77 ± 10.27 45.11 ± 2.99 55.35 ± 4.62 Mean total energy intake (kcal/day) 2229.47 ± 446.80 2254.69 ± 352.28 2246.47 ± 368.47 Fracture event, % (n) 15.90 (46) 12.80 (37) 10.70 (31) Crude model 1 (Ref.) 0.73 (0.48, 1.13) 0.63 (0.40, 0.99) 0.037 Multivariate model 1ᵃ 1 (Ref.) 0.62 (0.39, 0.97) 0.52 (0.31, 0.85) 0.007 Multivariate model 2ᵇ 1 (Ref.) 0.59 (0.37, 0.95) 0.49 (0.29, 0.81) 0.004 Cox regression models were used to evaluate the risk of osteoporotic fracture event by energy-adjusted tertiles of total olive oil (g/day), energy-adjusted tertiles of common olive oil (g/day) and energy-adjusted tertiles extra- virgin olive oil (g/day). Results were expressed as Hazard Ratios (95% CI) and means ± SD or percentage (n). ᵃ Model: Adjusted for age (years), sex, body mass index (BMI) (kg/m²), educational level (illiterate/primary M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT 24 464 education, secondary education, academic /graduate), leisure time physical activity (Metabolic Equivalent of Task (MET)-minutes/day), the intervention group and smoking status (never, former, current smoker). ᵇ Model: Additionally adjusted for prevalence of diabetes (yes/no), prevalence of previous fractures (yes/no), use of insulin (yes/no), use of oral antidiabetic drugs (yes/no), use of diuretic drugs (yes/no), use of glucocorticoids drugs (yes/no), use of osteoporotic drugs (yes/no), use of anticoagulant drugs (yes/no), use of estrogen drugs (yes/no) and baseline Mediterranean diet adherence (12-point score). M AN US CR IP T AC CE PT ED ACCEPTED MANUSCRIPT 25 465 Figure 1. Nelson-Aalen curves of cumulative hazard for osteoporotic fracture by tertiles 466 of energy adjusted extra-virgin olive oil intake. 467 0.00 0.05 0.10 0.15 C u m u la ti v e h a za rd f u n c ti o n 290 286 280 275 272 269 264 T3 290 289 286 281 272 265 261 T2 290 280 272 264 251 242 236 T1 year 1 year 2 year 4 year 6year 3 year 5 year 7 0 2 4 61 3 5 7 Follow-up years Tertile 1 Tertile 2 Tertile 3 View publication stats