2.Although the Method of Measurement of the CRP Concentration Has Been
We want to thank all three reviewers for their valuable suggestions and have included them in the manuscript to the best of our capacities. We believe their feedback has resulted in an improved version.
Reviewer #1: This is a very interesting and well written manuscript, which presents a very interesting methodological approach.
1.The presentation of the results "per unit of log-transformed CRPconcentration" is not informative. What does one "unit of log-transformed CRP concentration" correspond to? I suggest that the Authors use the standard deviation of the log-transformed CRP concentration. This is informative from the epidemiological point of view and provides the results comparable with other variables in this or other studies.
We presented the results using log-transformed values as a methodological preference for taking into account the skewed distribution of CRP-levels. Nevertheless, we have opted to follow the reviewer’s suggestion to analyze the data per standard deviation as we agree it facilitates comparison across studies and facilitates the understanding of the findings. Most important we verified that both approaches lead to essentially the same results and conclusions. Changes have been applied and results are now presented per standard deviation along the main paper while we also present those on a log-transformed CRP scale in the supplementary material. This is mentioned in the Methods section on page 6 lines 129-132: “All effect estimates are reported per standard deviation of increase in CRP. Additionally, CRP values were log-transformed to account for the non-normal distribution and effect estimates expressed per unit log-transformed CRP (Results presented in Supplementary Materials)”.
2.Although the method of measurement of the CRP concentration has been
described previously, more information should be provided here (detection limit, intra- and inter-assay variation), because this is the main parameter investigated in this study.
We agree and have included this information in the Methods section. Page 5 lines 91-96 now reads as follows: “Serum CRP measurement: At the baseline visit, high-sensitivity CRP was measured in non-fasting frozen serum samples of 6,658 study participants using a rate near-infrared particle immunoassay (Immage Immunochemistry System, Beckman Coulter, Fullerton, CA, USA). This system measures concentrations from 0.2 to 1.440 mg/l, with a within-run precision <5.0%, a total precision <7.5%, and a reliability coefficient of 0.995.”
3.Higher CRP level may be associated with frailty and/or sarcopeniawhich themselves are associated with higher risk of fall, and consequently, they may be associated with higher risk of fracture. Can the Authors adjust their analyses for the history of fall or grip strength or some other parameter related to frailty? If this is not
possible, the "lower leg disability" should be defined in more detail. In fact, this term is frequently used in the papers from the Rotterdam study, but not clearly described.
We agree with the reviewer that examining indicators of falling propensity, frailty and/orsarcopeniaconstitute important mediators that should be consider. We adjusted the analyses for fall risk and found identical results as now described in the manuscripton page 8 lines 172-174and in Table 2 (HR = 1.07; 95% CI: 1.02-1.12; p = 0.003) suggesting falling is not likely to be mediating the relationship. In addition, we have further described the lower leg disability definition in the Methods section on page 6 lines 114-119as following:“One intensively trained interview assistant carried out a one hour home interview including the Stanford Health Assessment Questionnaire (HAQ)[1]to assess lower limb disability, as described previously[2]. Briefly, the assessment measures the ability or disability and difficulty to perform different categories of daily life activities, such as rising, walking, bending and getting in and out of a car, summarized in an index ranging from 0 (no impairment) to 3 (severe impairment).” Though relevant, grip strength measurements were not analyzed, because they were included in the Rotterdam Study in recent follow-up visits many years after the baseline examinations used in this paper.
References:
- Pincus T, Summey JA, Soraci SA, Jr., Wallston KA, Hummon NP (1983) Assessment of patient satisfaction in activities of daily living using a modified Stanford Health Assessment Questionnaire. Arthritis Rheum 26:1346-1353
- Odding E, Valkenburg HA, Algra D, Vandenouweland FA, Grobbee DE, Hofman A (1995) Association of locomotor complaints and disability in the Rotterdam study. Ann Rheum Dis 54:721-725
4.I am afraid that the use of the calculated parameters of geometry of
the femoral neck in 2013 is somewhat obsolete. The calculation of cortical thickness of the shaft of a tubular bone may make sense. Bycontrast, the equations are based on several assumptions and, for the femoral neck, all of them are false.
We agree with the reviewer that DXA-based bone geometry represents 2D assessments of a 3D structure, whose measurement relies on several assumptions and several of the parameters are not independent of BMD. Unfortunately, we do not have more optimal measurements like QCT or MRI, which accurately measure the underlying bone geometry. We have included a statement about the limitation of the HSA in the Discussion on page 14 lines 324-327: “Finally, DXA-based bone geometry represents 2D assessments of a 3D structure, whose measurement relies on several assumptions and several of the parameters are not independent of BMD. Unfortunately, we currently do not have more optimal measurements like QCT or MRI available, which accurately measure the underlying bone geometry”. In line with the reviewer’s perspective and plead for more accurate technological approaches, pQCT measurements are since recently being collected in the Rotterdam Study and will be used in the future. Nevertheless, we do believe that the hip structure analysis (despite its limitations) is an informative tool to interpret the associations (or lack of them) with DXA-based bone mineral density of the hip. The approach has been used across a very large number of papers whose conclusions are robust and some of which have been replicated using 3D technologies. These include examining the effect of treatment on bone [3-7]; age related declines in BMD; adaptation to skeletal loading [8, 9]; and the interpretation of the effect of BMD on fracture risk[10-12] among others. BMD provides quantitative information on the amount of bone mineral per unit area but provides no indication of mass distribution, which is shown to be different across individuals by HSA, even in those matched for the same BMD levels. We believe that the geometry analysis presented in the paper provides further insight into the skeletal involvement in individuals with high CRP levels. In fact, as pointed out by reviewer 2, the finding of decreased bending strength has also been found in a recent paper[13]appearing in the literature while our paper was under review.
5.Line 277 - typographic error.
Thank you, we have corrected the error.
Reviewer #2: This mauscript examines the relationship between
high-semsitivity CRP (hs-CRP) and bone mineral density (BMD), bone
geometry derived from DXA scans and fracture risk. It confirms an
association between hs-CRP and fracture risk. BMD was not associated
with CRP though there was an association with bone geometry parameters.
The novel aspect of this analysis is the use of molecular techniques to
characterise the genetic determinants of hs-CPR and the evaluation of
whether serum CPR, effectively adjusted for genetic variance, explains
these association. This analysis indicates that the known
polymorphisms contributes very little to the variance in serum hs-CRP
values and, not surprisingly, suggests that hs-CRP is not in the casual
pathway of fragility fractures.
The principal limitation of this analysis relates to the inclusion of
an unspecified number of glucocorticoid-treated participants. This
confounder has been adjusted for statistically in the analyses
examining the association between hs-CPR and fracture, BMD and bone
geometry but has not been accounted for in the genetic analyses. As
glucocorticoid exposure and systemic inflammatory disorders can effect
both the outcome and the exposure of interest, a sensitivity analysis
should be done excluding glucocorticoid-treated participants, those
with systemic inflammatory disorders or both. Table 1 should also
enumerate the number of participants in the various categories of
glucorticoid exposure (ever and current) and the number of participants
with systemic inflammatory disorders (such as rheumatoid arthritis, SLE, inflammatory bowel disease).
Glucocorticoid exposure and the presence of systemic inflammatory disorders are indeed important confounders, which can affect the relationship between CRP levels and skeletal outcomes. The genetic analysis makes part of a Mendelian Randomization (MR) approach used to infer the presence (or not) of causal relationships. One important intrinsic property of the MR approach is that non-genetic confounding factors (e.g. environmental exposures) will be “randomized” across groups of individuals assembled using a sufficiently large number of genetic instrumental variables (i.e. 29 SNPs) not related to the confounders. Just as the study groups in a randomized trial, distribution of variables are very well balanced across bins of the genetic score. Therefore, the relationship between the genetic score with the outcome of interests will be unaffected by (i.e. independent of) confounders. The number of glucocorticoid-treated participants is 1.19% as included in the baseline characteristics presented on Table 1. Under the MR principles one will expect a similar distribution of glucocorticoid use across the bins of the genetic score. As expected, adjusting the genetic analysis of CRP levels for glucocorticoid treatment and/or performing a sensitivity analysis excluding glucocorticoid users does not modify the results: (HR = 1.00; 95% CI: 0.99-1.00; p = 0.23 before with adjustment and HR = 1.00; 95% CI: 0.99-1.00; p = 0.20 after). This is also the case for unmeasured variables including those we do not have information on like the distribution of inflammatory disorders across participants (such as rheumatoid arthritis, SLE, inflammatory bowel disease).
Minor Points:
1. The manuscript should be updated to include the paper by Ishii S, Cauley J, Greendale GA et al. J Bone Miner Res 2013;28(7):1688-98 that reports the association between bone geometry variable and hs-CPR.
We thank the reviewer for bringing this reference to our attention, which supports our findings on bone strength. The Discussion on page 11 lines 270-273 now reads: “More specifically, recent work in the Study of Women's Health Across the Nation (SWAN) showed that CRP levels were inversely associated with hip geometry strength indices (a finding in line with those of our study) which partially explained the increased fracture risk observed in SWAN [13]”.
2. A citation for the paper by Zacho et al is required
Thank you for pointing this fact out, we have added the reference to the text.
3. Page 12 - the association between buckling ratio and hs-CRP is weak and does not achieve statistical significance. It is not a trend.
We have reworded the statement on page 10 lines 214-217: “Increased CRP levels were associated with a weak increase of the buckling ratio instability index though not achieving statistical significance”.
Reviewer #3: Oei et al. showed that higher CRP is associated with
increased fracture risk, narrower femoral neck width, and a lower
bending strength. Although the associations between chronic
inflammation and poor bone health outcomes are already well-known, the
fact that this study included hip bone geometry data and Mendelian
randomization analyses is quite interesting. However, I have several
critical concerns, especially about the adjustment strategy.
1. <Page 9, Lines 46-49 The authors concluded that "In this study we
report an association between serum CRP levels and any-type of fracture
risk for both men and women". However, in this study, they did not show
the associations in each gender group and did just adjust for sex as a
confounding factor. Therefore, this conclusion is inappropriate.
We appreciate the reviewers concern regarding the effect of gender. We have thoroughly considered the effect of gender both as a potential confounder and stratifying by gender. This is described in the Methods page 7 lines 137-138: “Models for fracture, BMD and hip geometry were corrected for potential confounders, including: sex, age and BMI;”; and in the Results on page 8 lines 166-169: “For the gender-stratified analyses effect estimates did not all reach statistical significance, but were of similar magnitude (male HR = 1.11; 95% CI: 1.02-1.22; p = 0.02; female HR = 1.05; 95% CI: 1.00-1.11; p = 0.05).”
Furthermore, the biologic differences in bone metabolism between men
and women are well-known. So, I suggest that the results across the
whole analyses should be presented separately in men and women. In
addition, if possible, menopausal status, which is one of the most
important factors on bone metabolism, should be adjusted in women as well.
As stated by the reviewer the sexual dimorphism of bone is well described. Yet, as mentioned before we do present separate analyses for men and women, and cannot identify any indication of sex specificity in our findings. If there are residual sex-specific differences in our study these will be of extremely low magnitude if not negligible. We also agree with the reviewer that menopausal status is a very strong determinant of bone metabolism. Yet, the minimal age of inclusion is 55 years, resulting in only few pre-menopausal women (0.3% of all female participants) included in the study making very unlikely that menopausal status plays an influential role in the association observed across the whole spectrum of our study population.
2. In Table 2, Mode 4 included age, sex, BMI, myocardial infarction,
DM, dementia, cancer and lower limb disability, whereas Model 5
included age, sex, BMI, and use of systemic corticosteroids. Is there
any reason why the use of systemic corticosteroids was not included in
Model 4? The use of systemic corticosteroids should be included in
Model 4, and then Model
5 can be deleted.
The rationale to construct the different models was based on initially correcting by co-morbidities and disability (model 4); followed by model 5 where we assessed specifically the potential influence of systemic corticosteroid use. (with potential for affecting the relation under study or serving as an indicator of inflammatory/immune conditions). Also, the study sample that has complete data available was smaller for model 4.
Nevertheless, we have followed the advice and found the following, very similar, estimate with this model (HR = 1.06; 95% CI: 1.01-1.11;p = 0.018), as now included in the revised manuscript text and table.
3. Smoking status can affect both inflammatory process (Arnson Y el al.
J Autoimmun. 2010;34:J258-265) and bone metabolism, and can be a
serious confounding factor. Therefore, smoking status should be
included in the adjustment model.
Smoking status indeed exerts influence on the inflammatory process. On the results section, pages 8-9 lines 182-186 we did assess the potential influence of smoking on the reported associations and did not observe any indication of a confounding effect: “Analyses adjusting for smoking status (HR = 1.06; 95% CI: 1.01-1.11; p = 0.01) or stratifying according to smoking status did not display clear differences between groups (“never smokers”: HR = 1.07; 95% CI: 1.01-1.14; p = 0.03; “past smokers”: HR = 1.09; 95% CI: 0.99-1.19; p = 0.08; “current smokers”: HR = 1.07; 95% CI: 0.90-1.27; p = 0.47).”
4. Taken together from Comment #1 to #3, I suggest that the final
adjustment model should include age, BMI, smoking status, menopause
status (in women), myocardial infarction, T2DM, dementia, prevalent
cancer, lower limb disability ± FN-BMD (or LS-BMD) in each gender.
As discussed above we see no indication of sex-specificity of the associations and practically all female participants of our study were post-menopausal. Further, power (reduce in sample size) and multiple testing will be issues making difficult the interpretation of the associations and their relation with covariates. We have performed the requested analyses, which are presented below but opted for not including in the paper considering the space constraints and leaving this decision to the editor’s discretion.
male 270 fractures in 1,863 participants; HR = 0.96; 95% CI: 0.85-1.17; p = 0.96
female 795 fractures in 2,586 participants; HR = 1.07; 95% CI: 1.01-1.14; p = 0.03
5. In Table 2, the authors presented the results of hip fracture and
vertebral fracture only after adjustment for age, sex, BMI. The results
after additional adjustment for other confounding factors, such as
smoking status, myocardial infarction, T2DM and so on, should be
presented in Table 2 as well.
We have performed the requested analyses to see that covariates did not essentially modify these associations:, hip fracture (HR = 1.09; 95% CI: 1.02-1.17; p = 0.008)vs. after (HR = 1.10; 95% CI: 1.01-1.20; p = 0.03); wrist fracture (HR = 1.06; 95% CI: 0.97-1.16; p = 0.21) vs. after (HR = 1.08; 95% CI: 0.98-1.19; p = 0.15); vertebral fracture(OR = 1.34; 95% CI: 1.14-1.58; p = 0.0004) vs. after (OR = 1.32; 95% CI: 1.09-1.59; p = 0.004). We added a sentence in the results section on page 8 lines 178-180:“As observed for any type of fracture, correction of the associations with hip and vertebral fracture were not affected by the inclusion of covariates in the models”.
6. The results about the associations between serum CRP levels and hip
bone geometry were presented from Page 8, Line 51 to Page 9, Line 7.
Are these results adjusted for confounding factors or not? The authors
should clarify the adjustment model.
The descriptions of the adjustments included in the models are specified in the Methods section on page 7 lines 135-139, we have procured to clarify the wording: ”Models for fracture, BMD and hip geometry were corrected for potential confounders, including: sex, age, and BMI;fracture analyses had additional adjustments for myocardial infarction, prevalent type 2 diabetes mellitus, dementia, prevalent cancer, lower limb disability, use of systemic corticosteroids and risk of falling” We have also specified in the results that the associations were corrected by covariates.
7. The distinctive point of this study is that it included the hip bone
geometry data. Therefore, the results about the association between
serum CRP levels and hip bone geometry should be emphasized, and thus
presented as a Table after applying for various adjustment models.
We appreciate the suggestion of the reviewer but we realize the paper is already lengthy and have chosen to present the results concisely in the text. At consideration of the editor we can include all the results in an additional table.