In rodent models, undercarboxylated osteocalcin (ucOC) acts as a hormone that promotes insulin sensitivity and secretion. If ucOC plays a similar role in humans, then antiresorptive therapies, which reduce ucOC levels, may increase the risk of insulin resistance and diabetes. We tested whether antiresorptive therapies result in higher fasting glucose, increased weight, or greater diabetes incidence in post hoc analyses of three randomized, placebo-controlled trials in postmenopausal women: Fracture Intervention Trial (FIT) (N = 6151) of alendronate (4 years), Health Outcomes and Reduced Incidence with Zoledronic Acid Once Yearly Pivotal Fracture Trial (HORIZON-PFT) (N = 7113) of zoledronic acid (3 years), and Fracture Reduction Evaluation of Denosumab in Osteoporosis Every 6 Months (FREEDOM) trial (N = 7076) of denosumab (3 years). Fasting glucose was measured annually in FIT and HORIZON in a subset of women, and every 6 months in FREEDOM in all participants. Weight was measured annually in all trials. Incident diabetes was identified from adverse event reports, initiation of diabetes medication, or elevated fasting glucose. Differences in fasting glucose changes from randomization to trial conclusion between treatment and placebo groups were not statistically significant: −0.47 mg/dL in FIT, 0.20 mg/dL in HORIZON-PFT, and 0.09 mg/dL in FREEDOM, all p > 0.6. Weight change differed between treatment and placebo groups in FIT (0.32 kg, p = 0.003) and FREEDOM (0.31 kg, p = 0.023) but not in HORIZON-PFT (0.15 kg, p = 0.132). In the three trials combined, diabetes occurred in 203 and 225 women assigned to treatment or placebo, respectively. Diabetes incidence was not increased in any of the treatment groups or in the pooled estimate (pooled relative risk [RR] = 0.90; 95% confidence interval [CI] 0.74–1.10). Antiresorptive therapy does not have a clinically important effect on fasting glucose, weight, or diabetes risk in postmenopausal women. Contrary to predictions from mouse models, reduced bone turnover does not appear to play a significant role in glucose metabolism in humans.
Findings in rodent models indicate that osteocalcin (OC) acts as a hormone that promotes insulin sensitivity in peripheral tissues and insulin secretion by the pancreas.[1, 2] OC knock-out mice have increased fat mass and insulin resistance, and are more likely than wild-type littermates to develop diabetes on a high-fat diet. In the rodent models, undercarboxylated osteocalcin (ucOC) was identified as the active form of osteocalcin. OC, a noncollagenous protein in bone, is produced by the osteoblasts and undergoes posttranslational carboxylation. In humans, circulating levels of ucOC as well as total OC, a standard marker of bone formation, increase with greater osteoblast activity. Levels of ucOC may also be affected by bone resorption because of the removal of carboxyl groups in the acidic conditions of resorption, although this remains controversial.[1, 4]
Antiresorptive therapies for osteoporosis reduce bone formation markers, including OC, as well as bone resorption markers. Bisphosphonates and denosumab reduce circulating levels of total OC by 30% to 50%.[5-7] Alendronate has been shown to reduce ucOC by 56%. The effects of other bisphosphonates or denosumab on ucOC have not been investigated, but given their effects on OC and other bone formation markers, it is highly likely that they substantially reduce ucOC levels. In this context, the results from rodent models have led to the hypothesis that the reductions in OC and ucOC with antiresorptive therapy could lead to insulin resistance, lower insulin secretion, weight gain, and an increased risk of diabetes. If this was so, it would place individuals being treated for osteoporosis at increased risk of diabetes-related morbidity. Although these widely used therapies are not known to cause weight gain or impaired glucose metabolism, there has been little direct investigation of the effects of osteoporosis therapies on glucose or energy metabolism.
At the same time, the findings from rodent models, if proven in humans, suggest new possibilities for the prevention of insulin resistance and diabetes through alterations in bone metabolism. OC levels are lower in those with diabetes in cross-sectional studies, but this may be because of negative effects of diabetes on bone formation. Results from observational studies of the longitudinal effects of OC or ucOC levels on glucose metabolism have been inconsistent.[11-15] Randomized placebo-controlled trials of antiresorptive therapies provide an optimal setting to determine whether reductions in bone turnover affect glucose metabolism. To test whether these therapies increase the risk of weight gain, glucose intolerance, or diabetes, we used data from three large randomized placebo-controlled trials of alendronate, zoledronic acid, and denosumab, antiresorptive therapies that substantially reduce bone remodeling.
Materials and Methods
The results of three randomized placebo-controlled trials of antiresorptive therapies were analyzed separately to determine the effect of each intervention on changes in weight, fasting glucose, and incident diabetes compared with placebo. Effects of the therapies were also analyzed separately among overweight and obese women who are at higher risk of developing elevated fasting glucose and diabetes. Women with a body mass index (BMI) at baseline of 25 kg/m2 or more were considered overweight/obese. Analyses excluded women with diabetes, or suspected diabetes, at baseline. Each trial is described below.
Fracture Intervention Trial (FIT)
In FIT, 6459 postmenopausal women aged 55 to 80 years were assigned to receive daily alendronate (ALN) or placebo.[16-18] Women with low dietary calcium intake at baseline (<1000 mg/day) were provided daily oral calcium (500 mg) and vitamin D (250 IU). Eligible women had femoral neck (FN) bone mineral density (BMD) T-score ≤ −1.6. The vertebral fracture arm (FIT I) enrolled women (n = 2027) who also had an existing radiographic vertebral fracture for up to 3 years. The clinical fracture arm (FIT II) enrolled women (n = 4432) without radiographic evidence of a vertebral fracture for up to 4 years. The daily ALN dose in both arms was 5 mg for the first 2 years of the trial after which the dose was increased to 10 mg.
Weight was measured at baseline and annual clinic visits in all participants. Glucose was also obtained annually in all women. In a subset of participants (n = 1162) identified in advance for evaluation of bone turnover markers, blood was drawn after fasting, and this subset is included in our analysis of fasting glucose (FG). Women who reported a history of diabetes or use of diabetes medication at baseline were excluded as prevalent diabetes cases. Those with an elevated FG (≥126 mg/dL) or spot glucose (≥200 mg/dL) at baseline were also excluded from analyses. A total of 308 women were excluded. Incident diabetes was identified based on adverse event reports, reported use of diabetes medication at a follow-up visit, or, in the subset with FG available, an elevated FG during follow-up. Women with a single elevated glucose but no other indication of diabetes were excluded (n = 174). Whole-body dual-energy X-ray absorptiometry (DXA) scans were obtained at baseline and the final clinic visit (3 or 4 years later) for all participants. For these analyses, total soft-tissue mass (bone-free mass) was calculated as total mass minus total bone mineral content.
Health Outcomes and Reduced Incidence with Zoledronic Acid Once Yearly Pivotal Fracture Trial (HORIZON-PFT)
In HORIZON, 7736 postmenopausal women aged 65 to 89 years were stratified into two groups based on current use of other osteoporosis medications and were randomized within the two strata to receive a once-yearly infusion (5 mg) of zoledronic acid (ZOL) or placebo for up to 3 years. All women also received oral daily calcium (1000 to 1500 mg) and vitamin D (400 to 1200 IU). Eligible women had a FN BMD T-score ≤ −2.5, with or without existing radiographic vertebral fracture, or FN BMD T-score ≤ −1.5 with radiologic evidence of at least one moderate or two mild vertebral fractures. Weight and glucose were measured at baseline and annually during the trial on all participants. A fasting blood draw was obtained for prespecified subsets of participants for studies of bone turnover markers and renal safety. During the trial, 5213 women had at least one FG measurement, including 4130 women with a measurement at baseline. At baseline, women (n = 623) who reported a history of diabetes or use of any diabetes medications or had an elevated FG were excluded from analyses. Incident diabetes was identified based on adverse event reports, reported use of diabetes medication, or an elevated FG during follow-up. Women with a single elevated glucose but no other indication of diabetes were excluded (n = 140). Dates were not available for the initiation of concomitant medications, including diabetes medications. As a result, the timing of incident cases identified by use of diabetes medications during follow-up was not known.
Fracture Reduction Evaluation of Denosumab in Osteoporosis Every 6 Months Trial (FREEDOM)
In FREEDOM, 7762 postmenopausal women aged 60 to 90 years were stratified into 5-year age groups and then randomized within strata to receive denosumab (DMAB) or placebo every 6 months for up to 3 years. Women received daily oral calcium (at least 1000 mg) and vitamin D (400 to 800 mg). Eligible women had BMD T-scores < −2.5 at the total hip or lumbar spine. Weight was measured at baseline and annually. Fasting glucose was measured at baseline, the 1-month visit, 6-month visit, and every 6 months thereafter. At baseline, women (n = 686) who reported a history of diabetes or use of any diabetes medication or who had an elevated FG were excluded from analyses. Incident diabetes was identified based on adverse event reports, reported use of diabetes medication at a follow-up visit, or an elevated FG during follow-up. Women with a single elevated glucose but no other indication of diabetes were excluded (n = 163).
We used linear mixed models to estimate and compare changes in fasting glucose and weight in the treatment and control groups, net of any differences at randomization occurring by chance. These models accounted for within-subject correlation of the repeated outcomes using random intercepts and slopes, with unstructured covariance. Trends were modeled as linear, and between-group differences were captured by a time-by-treatment interaction. Estimates for FIT were adjusted for study arm (presence or absence of baseline vertebral fracture), estimates for HORIZON-PFT were adjusted for baseline strata and region, and estimates for FREEDOM were adjusted for 5-year age strata.
The rate of incident diabetes by treatment assignment was compared with proportional hazards models in FIT and FREEDOM. The odds of incident diabetes were compared with logistic models in HORIZON-PFT because some incident cases, identified through initiation of a diabetes medication, could not be assigned a date of occurrence. We estimated the pooled effect of assignment to an antiresorptive therapy on incident diabetes with an unadjusted logistic regression model. Because incident diabetes is a relatively rare outcome, we report the odds ratio from logistic regression models as a relative risk.
Baseline characteristics of women without prevalent diabetes in the FIT, HORIZON-PFT, and FREEDOM trials are provided in Table 1. Between 44% and 55% of women in the trials were overweight or obese at baseline. In the three trials, average age, weight, fasting glucose, proportion overweight or obese, and proportion with FN BMD T-score ≤ −2.5 did not differ by treatment group.
|No. of women||3084||3067||3537||3576||3535||3541|
|Age (years)||68.0 (6.2)||68.2 (6.1)||73.0 (5.3)||73.0 (5.4)||72.2 (5.2)||72.3 (5.2)|
|Weight (kg)||64.3 (10.9)||64.4 (10.9)||59.7 (10.9)||60.4 (11.2)||63.7 (10.1)||63.4 (10.3)|
|Overweight/obesea||1360 (44)||1406 (46)||1612 (46)||1744 (49)||1927 (55)||1917 (54)|
|FN BMD T-score|
|≤−2.5||1300 (42)||1295 (42)||2574 (73)||2536 (71)||1262 (36)||1257 (36)|
|>−2.5||1783 (58)||1770 (58)||941 (27)||1024 (29)||2257 (64)||2271 (64)|
|Mean (mg/dL)||93.3 (10.2)||93.0 (10.6)||94.9 (10.0)||95.4 (10.0)||91.8 (9.2)||91.8 (9.5)|
Fasting glucose change
The placebo groups experienced an increase in mean fasting glucose during follow-up that was statistically significant in FIT (3.87 mg/dL, p = 0.0002) and HORIZON-PFT (1.74 mg/dL, p ≤ 0.0001) but not in FREEDOM (0.52 mg/dL, p = 0.074). Differences in average fasting glucose changes between treatment and placebo groups were not statistically significant for any of the three trials: −0.47 mg/dL for FIT, 0.20 mg/dL for HORIZON-PFT, and 0.09 mg/dL for FREEDOM, all p > 0.6 (Table 2).
|Fasting glucose changea (mg/dL)||Weight changea (kg)|
|Difference (treatment – placebo) (95% CI)||p value||Difference (treatment – placebo) (95% CI)||p Value|
|FIT (alendronate)||−0.47 (−3.37, 2.43)||0.752||0.32 (0.10, 0.54)||0.003|
|HORIZON-PFT (zoledronic acid)||0.20 (−0.60, 1.00)||0.634||0.15 (−0.05, 0.35)||0.132|
|FREEDOM (denosumab)||0.09 (−0.32, 0.50)||0.676||0.31 (0.04, 0.58)||0.023|
In the placebo groups, average weight during the trial was stable in FIT (mean change −0.06 kg, p = 0.431) and declined slightly in HORIZON-PFT (−0.50 kg, p < 0.0001) and in FREEDOM (−0.44 kg, p < 0.0001). In FIT and FREEDOM, but not HORIZON-PFT, the group assigned to the study drug had less weight loss than the placebo group (Table 2). The difference in average weight change in FIT over 4 years, comparing alendronate and placebo groups, was 0.32 kg (p = 0.003). The difference in average weight change over 3 years, comparing treatment and placebo groups, was 0.15 kg (p = 0.132) in HORIZON-PFT and 0.31 kg (p = 0.023) in FREEDOM.
The contribution of treatment-induced increases in bone mass to the greater weight gain in the ALN group in FIT was assessed using whole-body DXA results. Among FIT participants with whole-body DXA scans at baseline and last visit (n = 5010), total bone mineral content (BMC) increased in the alendronate compared with the placebo group (+0.014 versus −0.003 kg/year, p < 0.001). However, soft-tissue mass, calculated as total mass minus total BMC, also increased in the alendronate compared with the placebo group (+0.02 versus −0.05 kg/year, p = 0.014). The greater increase in soft-tissue mass in the alendronate compared with the placebo group included greater increases in total fat mass (−0.005 versus −0.054 kg/year, p = 0.059) and total lean soft-tissue mass (0.028 versus 0.004 kg/year, p = 0.056), although these differences were not statistically significant.
Incident diabetes occurred in 69 alendronate and 72 placebo-assigned women in FIT; 68 zoledronic acid and 75 placebo-assigned women in HORIZON-PFT; and 66 denosumab and 78 placebo-assigned women in FREEDOM. The incidence rate of diabetes was 6.2 per 1000 person-years in the FIT placebo group and 8.2 per 1000 person-years in the FREEDOM placebo group. A rate could not be calculated in HORIZON because of the lack of data on the time of diabetes diagnosis. However, the proportion with incident diabetes was similar to the other two trials. The risk of incident diabetes was not statistically different between treatment and placebo groups in any of the three trials (Fig. 1). In an unadjusted pooled estimate, the odds ratio for incident diabetes was 0.90 (95% CI 0.74–1.10) (p = 0.31).
Effects of therapy among overweight and obese women
Among overweight and obese women, a group with higher risk of developing elevated fasting glucose and diabetes, the effects of the three antiresorptive therapies on fasting glucose, weight, and incident diabetes were similar to those found in the full cohorts. Change in fasting glucose did not differ by treatment status in models that included only overweight and obese women. The treatment groups tended to gain weight relative to the placebo groups, but the difference was only statistically significant in FIT. The difference in average weight change, comparing treatment and placebo groups, was 0.53 (p = 0.003) in FIT, 0.31 (p = 0.51) in HORIZON, and 0.39 (p = 0.054) in FREEDOM among overweight and obese women. The relative risks for incident diabetes among overweight and obese women were 1.07 (95% CI 0.73–1.59) in FIT, 1.00 (95% CI 0.64–1.55) in HORIZON, and 0.82 (95% CI 0.56–1.21) in FREEDOM.
In contrast to the hypothesized increases in fasting glucose and diabetes expected from osteocalcin knock-out models in rodents, we did not find evidence of an effect of antiresorptive therapies on these outcomes in three large randomized placebo-controlled trials over periods of 3 to 4 years. The upper 95% confidence limit of 1.10 for the pooled hazard ratio for the effects of antiresorptive therapies on incident diabetes allows us to rule out increases in risk of more than 10% with considerable confidence. Treatment with alendronate and denosumab, but not zoledronic acid, was associated with a small positive difference in weight change compared with placebo, consistent with the findings in rodent models. However, the magnitude of the weight change over several years was probably not clinically meaningful.
Antiresorptive therapy increases bone density, relative to placebo, and this might contribute to the lack of weight loss in the treated groups. However, analyses in FIT using whole-body composition showed that the increases in bone mass with alendronate treatment accounted for less than half of the greater weight gain in the alendronate, compared with the placebo, group.
Previous analyses have established that bone turnover markers were substantially reduced in these trials. In FIT 1 year after randomization, the alendronate group had substantial reductions, compared with placebo, in formation markers, bone alkaline phosphatase (ALP) (22.7%), and procollagen type I N-terminal propeptide (PINP) (41.3%), and in the resorption marker, serum C-telopeptide (sCTX) (35.0%). In HORIZON, 3 years after randomization (1 year after the last infusion), bone ALP (30%), PINP (56%), and sCTX (51%) were reduced in women assigned to zoledronic acid. In FREEDOM, comparing denosumab and placebo, PINP was reduced by 18% at 1 month, by 50% before treatment was administered at 6 months, and by 76% at 36 months. At the same time points, sCTX levels were reduced by 86%, 72%, and 72% in the treated compared with placebo group. Measurements of OC or ucOC were not obtained in any of the contributing trials, so it was not possible to directly examine the relationship between changes in these markers and in glucose metabolism.
Other studies have established that each of these antiresorptive therapies reduce circulating levels of total OC by 30% to 50%[5-7] and also reduce bone resorption markers.[6, 19, 20, 24] Effects on ucOC have only been investigated for alendronate, and published studies have reported decreased levels with treatment.[8, 25, 26] In postmenopausal women, alendronate (10 mg/day) reduced ucOC by 29% (95% CI 8–45%) after 3 months and by 56% (95% CI 40–68%) after 12 months. It is most likely that zoledronic acid and denosumab, both shown to reduce total OC and other bone turnover markers, also reduce ucOC levels.
The women in these trials were recruited on the basis of osteoporosis, associated with lower weight, and the prevalence of overweight and obesity in the trials was lower than in broader populations of a similar age. Thus, trial participants may have had a lower risk of developing diabetes than the general population. The incidence of diabetes in the placebo groups was similar to reported incidence rates in European populations but lower than rates reported in the United States.[28-30] However, when we investigated the effects of the therapies in those who were overweight or obese at baseline, a group with higher risk of developing diabetes, we did not find evidence of differences compared with the cohort as a whole.
Vestergaard examined the effect of antiresorptive therapies (alendronate, etidronate, and raloxifene) on the risk of developing diabetes over an average follow-up of 3.8 years, using national Danish databases of prescriptions to identify osteoporosis therapy and in-patient/out-patient records to identify incident diabetes. The study found a reduced risk of developing type 2 diabetes with use of these therapies. For alendronate, the estimated hazard ratio was 0.71 (95% CI 0.59–0.85). However, the patients prescribed osteoporosis therapy may have had lower BMI and therefore a reduced risk of diabetes. The current analysis provides a more rigorous assessment of the effect of antiresorptive therapies on risk of diabetes because each of the trials was randomized and placebo-controlled, thus controlling for potential confounding by differences in baseline BMI or other risk factors for diabetes.
In general, trials of antiresorptive therapies have not reported an increase in adverse event reports of incident diabetes, although this was not a hypothesized adverse effect. An exception is the Multiple Outcomes of Raloxifene (MORE) trial, which had more reports of new or worsening diabetes in the raloxifene (1.2%) than in the placebo (0.5%) group (p = 0.009). However, there was an imbalance at baseline with higher average FG in the raloxifene group. In addition, raloxifene has much smaller effects on bone turnover than the interventions assessed in the present analysis, suggesting that other mechanisms are involved in any effect on diabetes.
In the PTH and Alendronate Trial (PaTH) comparing treatment with PTH (1–84) and alendronate, Schafer and colleagues measured serum ucOC and assessed changes in weight and fat mass. During the first year of the trial, the PTH group lost weight (mean change −0.4 kg) relative to the alendronate group (mean change +0.2 kg), but the difference was not statistically significant (p = 0.21). In analyses that combined the two treatment groups and assessed correlations between changes in ucOC over the first 3 months of the study with 12-month changes in weight, fat mass, and adiponectin, those with the greatest increases in ucOC had the greatest decreases in weight (r = −0.27, p = 0.01) and fat mass (r = −0.17; p = 0.10), and the greatest increases in adiponectin (r = 0.29; p < 0.01), consistent with predictions from the rodent models.
Our results indicate that these antiresorptive therapies do not affect fasting glucose levels or diabetes risk and probably do not have a clinically important effect on weight. These results also provide evidence that substantial reductions in bone turnover do not have a net effect on glucose metabolism in humans, a finding that is counter to the effect hypothesized from rodent models. The lack of effect on glucose metabolism is consistent with findings from two randomized trials of vitamin K1 supplementation, a treatment that substantially lowers ucOC levels without reducing bone turnover. These trials reported no effect on fasting glucose or insulin resistance over 1 year and 3 years. The small differences in weight change that we observed with alendronate and denosumab are consistent in direction with findings from rodent models. Although differences in weight change observed with zoledronic acid were not statistically significant, this may be a chance finding because the results for zoledronic acid were of a similar magnitude and the confidence intervals for differences in weight change in the three trials overlapped. These results and the negative correlation between ucOC levels and weight gain reported by Schafer and colleagues suggest that the mechanisms identified in the rodent models may remain operative in humans. It is also possible that antiresorptive therapies affect weight change through a different, unidentified mechanism. The physiological and clinical implications of these small differences in weight change are unclear. Weight loss in older women is associated with frailty and increased mortality. A meta-analysis reported that antiresorptive therapies reduce mortality in osteoporotic women. The relative maintenance of body weight we observed in the current analyses might be a contributing factor. There is a need for further research to more fully understand the relationship between bone turnover, weight change, and energy metabolism in humans, particularly in circumstances such as PTH therapy that increase ucOC dramatically.
Our failure to find support for the predictions from the rodent models may be because of differences in the extent of suppression of osteocalcin. Although the three antiresorptive therapies studied cause marked reductions in bone turnover, they do not completely suppress osteocalcin production, in contrast to the absence of osteocalcin in the knock-out mouse models. It is possible that only very low or very high levels of ucOC have an appreciable effect on glucose metabolism, although substantial lowering of ucOC in postmenopausal women by vitamin K1 did not alter fasting glucose or insulin resistance. Future research might focus on effects in patients with the largest changes in ucOC in response to osteoporosis therapies. Additionally, the trial participants were postmenopausal women, whereas the rodent studies were conducted in male and female mice with normal gonadal function. The timing and duration of exposure to low OC also differed. The knock-out mice lacked OC throughout the life span, including embryonic development.
The randomized and placebo-controlled design and large size of these trials provides the best available context in which to test the hypothesis from rodent models that reductions in bone turnover will affect insulin sensitivity and secretion. And, our results were generally consistent across three different antiresorptive therapies in trials conducted in different populations. However, the three trials were all conducted in postmenopausal women, so results might not apply to men or to other age groups. In addition, our data do not include effects of these therapies beyond the 3- to 4-year duration of the trials.
In conclusion, suppression of bone turnover with antiresorptive therapy does not have a clinically important effect on fasting glucose, weight, or diabetes incidence. These results suggest, contrary to findings in mouse models, that reduced bone turnover, and by implication lower undercarboxylated osteocalcin, does not play a significant role in the regulation of insulin sensitivity in humans. They also provide reassurance to patients receiving such osteoporosis treatments that they are not at increased risk of impaired glucose metabolism or diabetes.
AVS has received consulting fees from Amgen and GlaxoSmithKline, and research support from Amgen, GlaxoSmithKline and Merck.
LP has received consulting fees from Nycomed.
RBW received consulting fees from Merck.
DMB has received consulting fees from Amgen, Eli Lilly, Merck, Nycomed, and Zosano, and research grants from Merck, Novartis, and Roche.
IRR has received board membership, consulting fees, honoraria, and research grants from Amgen, Merck, and Novartis, and has provided expert testimony for Novartis.
ALS, AG, EV, LLL, DCB, and SRC state that they have no conflicts of interest.
ALS was supported by a career development award from the Department of Veterans Affairs (5 IK2 CX000549-02). The clinical trials were supported by their respective sponsors. The Fracture Intervention Trial was sponsored by Merck & Co., Inc.; Health Outcomes and Reduced Incidence with Zoledronic Acid Once Yearly Pivotal Fracture Trial (HORIZON-PFT) was sponsored by Novartis Pharma; and the Fracture Reduction Evaluation of Denosumab in Osteoporosis Every 6 Months (FREEDOM) trial was sponsored by Amgen Inc. This study was not financially supported by the trial sponsors.
Authors' roles: Study design: AVS, ALS, AG, and IR. Study conduct: AVS. Data analysis: LP, LL, and EV. Data interpretation: AVS, ALS, AG, IR, EV, RW, LP, LL, DCB, DMB, and SC. Drafting manuscript: AVS and EV. Revising manuscript content: AVS, ALS, AG, IR, EV, RW, LP, LL, DCB, DMB, and SC. Approving final version of manuscript: AVS, ALS, AG, IR, EV, RW, LP, LL, DCB, DMB, and SC. AVS takes responsibility for the integrity of the data analysis.