Calcified Tissue International

, Volume 94, Issue 5, pp 560–567

A Meta-Analysis of Reference Markers of Bone Turnover for Prediction of Fracture

  • Helena Johansson
  • Anders Odén
  • John A. Kanis
  • Eugene V. McCloskey
  • Howard A. Morris
  • Cyrus Cooper
  • Samuel Vasikaran
  • IFCC-IOF Joint Working Group on Standardisation of Biochemical Markers of Bone Turnover
Original Research

DOI: 10.1007/s00223-014-9842-y

Cite this article as:
Johansson, H., Odén, A., Kanis, J.A. et al. Calcif Tissue Int (2014) 94: 560. doi:10.1007/s00223-014-9842-y

Abstract

The aim of this report was to summarize the clinical performance of two reference bone turnover markers (BTMs) in the prediction of fracture risk. We used an updated systematic review to examine the performance characteristics of serum procollagen type I N propeptide (s-PINP) and serum C-terminal cross-linking telopeptide of type I collagen (s-CTX) in fracture risk prediction in untreated individuals in prospective cohort studies. We excluded cross-sectional studies. Ten potentially eligible publications were identified and six included in the meta-analysis. There was a significant association between s-PINP and the risk of fracture. The hazard ratio per SD increase in s-PINP (gradient of risk [GR]) was 1.23 (95 % CI 1.09–1.39) for men and women combined unadjusted for bone mineral density. There was also a significant association between s-CTX and risk of fracture, GR = 1.18 (95 % CI 1.05–1.34) unadjusted for bone mineral density. For the outcome of hip fracture, the association between s-CTX and risk of fracture was slightly higher, 1.23 (95 % CI 1.04–1.47). Thus, there is a modest but significant association between BTMs and risk of future fractures.

Keywords

Marker of bone turnoverFractureSerum procollagen type I N propeptideSerum carboxy-terminal cross-linking telopeptide of type I collagenMeta-analysis

Introduction

The development of markers of bone turnover (commonly termed “bone turnover markers” [BTMs]) has provided an important tool in the clinical and preclinical assessment of bone active interventions [1]. Attractive features of the markers are that samples of blood or urine are easily collected and a variety of assays are available, are relatively noninvasive, and provide information that is complementary to bone mineral density (BMD).

Despite an extensive research base, uncertainties remain over their use in routine clinical practice to assess fracture risk and/or monitor treatment [15]. Limitations variously include their biological variability [6] and, in some cases, the multiple methodologies used for the same analyte. In the specific case of fracture risk prediction, there is heterogeneity in the fracture outcomes reported (e.g., spine, hip, nonspine, and all fractures) [7] and heterogeneity in expressing fracture risk. For example, risk can be expressed as the hazard ratio (HR) for fracture per standard deviation (SD) increase in BTM, as a BTM lying within the top quartile (compared to the rest of the population), as the HR in tertiles, or as values more than 2 SD above the premenopausal reference interval [1]. This variation poses problems in comparing the clinical value of different markers and comparing the same marker between studies.

A working group of the International Osteoporosis Foundation and the International Federation of Clinical Chemistry and Laboratory Medicine has recently recommended one bone formation marker (serum procollagen type I N propeptide [s-PINP]) and one bone resorption marker (serum C-terminal cross-linking telopeptide of type I collagen [s-CTX]) as reference markers to be used in future studies of fracture risk assessment so that sufficient data are accumulated for the assessment of their potential for inclusion in fracture risk-assessment tools [1]. The aim of the present study was to summarize by meta-analysis the current information available on these reference analytes for fracture risk prediction.

Methods

We examined the performance characteristics of s-PINP and s-CTX in fracture risk prediction from systematic literature searches [1], which were updated from 2010 to February 2012. Eligible studies of the performance characteristics in untreated individuals were taken from a systematic review of the English-language literature in the PUBMED database between the years 2000 and 2010 [1], which incorporated the 2001 tabulated evidence from the Agency for Healthcare Research and Quality on BTMs [8], was based on a MEDLINE systematic review, and provided the source of relevant prospective studies up to the year 2001. To ensure completeness, the PUBMED search was updated to February 2012 for new publications and the bibliographies of recent reviews studies were examined for studies of potential interest [25, 9, 10]. No new studies were identified that met the criteria for inclusion.

Studies eligible for inclusion were prospective cohort studies of s-PINP or s-CTX measured at baseline in untreated individuals. Nested case–control and case–cohort studies were also allowed. The primary outcome was the first incident fracture in middle-aged or older men and women. We excluded cross-sectional studies and abstracts. We excluded studies that did not provide separate data for men and women. Ten publications were identified that examined the association between incident fracture and s-PINP or s-CTX.

The risk of fracture was reported in several different ways—as the HR between the highest quartile and the three lowest quartiles, the HR per SD, and the HR per unit of measurement and by tertile of BTM. In order to merge the results, a uniform metric was required. The metric that we chose was the HR for fracture per SD difference in BTM (the gradient of risk [GR]). Where the results were reported in more than one way, the GR was chosen. If the GR was unreported, we used the HR per unit of measurement; and if neither was available, the ratio of tertiles or quartiles was used. The available measurements were transformed into gradients of risk using methods described in the Appendix. The association between the risk ratio comparing the highest quartile and the three lowest quartiles and corresponding risk ratio per SD is shown in Fig. 1.
https://static-content.springer.com/image/art%3A10.1007%2Fs00223-014-9842-y/MediaObjects/223_2014_9842_Fig1_HTML.gif
Fig. 1

The relationship between the hazard ratio per SD (gradient of risk) and the risk ratio of the highest quartile (Q4) and the three lowest quartiles (Q1–Q3). Dashed line is the line of identity

The 10 publications identified in the review are summarized in Table 1. Bauer [11], Szulc et al. [12], and Yoshimura et al. [13] reported that there was no significant relationship between BTMs and the risk of fracture (p > 0.05) but provided no relevant numerical information and were therefore discarded. Gerdhem et al. [7] and Ivaska et al. [14] studied the same cohort. Ivaska et al. studied the cohort for 9.0 years, and Gerdhem et al. had a shorter follow-up time of 6.5 years. The study with the shorter follow-up was selected for inclusion in the meta-analysis because the study of Ivaska et al. [14] reported a time interaction; i.e., the HRs for any fracture were highest at the start of the follow-up period and waned with time. In a sensitivity analysis, the study of Gerdhem et al. was replaced by that of Ivaska et al. Among the remaining six papers, the associations between fracture and s-PINP were reported in three studies [1517] and the results for s-CTX were described in all six [7, 1519].
Table 1

Details of publications identified

Study

BTM

Follow-up (years)

Sex

Age (years)

Fracture outcome

Population/setting

Assay used for s-PINP

Assay used for s-CTX

Fasting

Type of unit

Included in meta-analysis

 Bauer et al. [15]

s-PINP, s-CTX

4.6

M

>65

Hip and nonspine

Advert and mass mailing

Roche diagnostics (Mannheim, Germany)

Roche diagnostics

Yes

Quartilesa

 Chapurlat et al. [18]

s-CTX

4.9

F

>75

Hip

Population-based registers

 

Osteometer Biotech (Herlev, Denmark)

Quartilesa, >2 SD vs. the rest

 Dobnig et al. [19]

s-CTX

2

F

>70

Hip, nonvertebral

Nursing home

 

Elecsys β-CrossLaps

No

HR per 1 unit

 Garnero et al. [16]

s-PINP, s-CTX

5

F

50–89

Osteoporotic (vertebral + appendicular)

Healthy untreated postmenopausal

Farmos Diagnostica (Uppsala, Sweden)

Osteometer Biotech (Ballerup, Denmark)

Yes

Quartilesa, >2 SD vs. the rest

 Gerdhem et al. [7]

s-CTX

6.5

F

75

Any, hip, clinical vertebral

Population-based, nursing home

 

Roche Diagnostics

No

Quartilesa

 Meier et al. [17]

s-PINP, s-CTX

6.3

M

>70

Low-trauma

All in city

Orion Diagnostica (Espoo, Finland)

Nordic Bioscience Diagnostics (Herlev, Denmark)

No

HR per SD, quartiles

Excluded from meta-analysis

 Bauer [11]

s-CTX

8

F

>65

Hip and vertebral

Population-based listing

   

No results given

 Ivaska et al. [14]

s-CTX

9.0

F

75

Any, hip, clinical vertebral

Population-based, nursing home

   

HR per SD, tertiles

 Szulc et al. [12]

s-PINP, s-CTX

7.5

M

50–85

Any, major fragility

Insurance-based

   

No results given

 Yoshimura et al. [13]

s-PINP, s-CTX

10

M + F

40–79

Osteoporotic

Population-based

   

No results given

aHighest quartile versus three lowest quartiles

Only one study reported the HRs per SD, a population-based cohort of men 70 years and older followed for 6 years [17]. In one female cohort aged more than 70 years from a nursing home with 2 years’ follow-up, the result was reported as the HR per unit [19]. In the remaining four studies, three in women and one in men, the HRs were reported as quartiles (the highest quartile vs. the three lowest quartiles) [7, 15, 16, 18].

Distribution

When we had access to the HR between Q4 and Q1–Q3, we determined the β-coefficient of a Cox or Poisson regression model for a corresponding risk variable, which has a standard normal distribution. The method used in this report was not dependent on the distribution of the risk variable, BTM.

Merging

All results were then transformed to a GR with 95 % confidence intervals (CIs) and then expressed as a β-coefficient (ln[GR]) and its variance. The variance was the mean of the two variances calculated from the two confidence interval limits. The β-coefficients were then weighted and merged according to the variance. Heterogeneity between cohorts was tested by means of the I2 statistic [20]. Low heterogeneity was noted for the results in Tables 2, 3, and 4, I2 = 0 % (95 % CI 0–98), I2 = 10 % (95 % CI 0–51), and I2 = 8 % (95 % CI 0–45), respectively; and a fixed effects rather than a random effects model was used.
Table 2

The relationship between s-PINP and fracture risk

Cohort

Sex

n

Outcome fracture

Adjustment

HR per SD

Bauer et al. [15]

M

1,005

Nonspine

Age and clinic

1.31 (1.12–1.54)

Garnero et al. [16]

F

435

Osteoporotic

Age, previous fracture, and physical activity

1.17 (0.81–1.69)

Meier et al. [17]

M

151

Osteoporotic

No adjustment

1.10 (0.88–1.37)

Merged result

    

1.23 (1.09–1.39)

Table 3

The relationship between s-CTX and fracture risk

Cohort

Sex

n

Outcome fracture

Adjustment

HR per SD

Bauer et al. [15]

M

1,005

Nonspine

Age and clinic

1.16 (0.99–1.37)

Chapurlat et al. [18]

F

408

Hip

No adjustment

1.48 (1.03–2.12)

Dobnig et al. [19]

F

1,664

Nonvertebral

Age, BMI, mobility, previous fracture, and creatinine

1.10 (0.93–1.32)

Garnero et al. [16]

F

435

Osteoporotic

Age and physical activity

1.75 (1.13–2.71)

Gerdhem et al. [7]

F

1,040

Any

No adjustment

1.10 (0.88–1.38)

Meier et al. [17]

M

151

Low-trauma

No adjustment

1.20 (0.94–1.54)

Merged result

    

1.18 (1.08–1.29)

Table 4

The relationship between s-CTX and hip fracture risk

Cohort

Sex

n

Outcome fracture

Adjustment

HR per SD

Bauer et al. [15]

M

1,005

Hip

Age and clinic

1.41 (1.02–1.95)

Chapurlat et al. [18]

F

408

Hip

No adjustment

1.48 (1.03–2.12)

Dobnig et al. [19]

F

1,664

Hip

Age, BMI, mobility, previous fracture, and creatinine

1.07 (0.79–1.45)

Gerdhem et al. [7]

F

1,040

Hip

No adjustment

1.01 (0.65–1.55)

Merged result

    

1.23 (1.04–1.47)

Results

s-PINP

Three publications reported an association between s-PINP and the risk of fracture (Table 2). Note that the three studies have slightly different fracture outcomes and different settings for adjustment. When the results were merged, the HR per SD was 1.23 (95 % CI 1.09–1.39); i.e., when s-PINP was higher by 1 SD, the risk of fracture was increased by 23 % (Fig. 2).
https://static-content.springer.com/image/art%3A10.1007%2Fs00223-014-9842-y/MediaObjects/223_2014_9842_Fig2_HTML.gif
Fig. 2

Forest plot for the relationship between s-PINP and fracture risk

s-CTX

The six publications investigating the association between s-CTX and the risk of fracture [7, 1519] are summarized in Table 3. Note again that the studies have different fracture outcomes and different settings for adjustment. When the data from all cohorts were merged, the HR per SD was 1.18 (95 % CI 1.05–1.34), so that a 1 SD increment in s-CTX was associated with an increased risk of fracture of 18 % (Fig. 3). When the results for women only were merged (four cohorts), the HR per SD was 1.19 (95 % CI 1.05–1.34). When the results adjusted for age were merged (three cohorts), the HR per SD was 1.17 (95 % CI 1.04–1.31). When the study of Ivaska et al. [14] with 9 years’ follow-up was included instead of that of Gerdhem et al. [7] with 6.5 years of follow-up of the same study (Table 3), the HR per SD was 1.19 (1.08–1.32).
https://static-content.springer.com/image/art%3A10.1007%2Fs00223-014-9842-y/MediaObjects/223_2014_9842_Fig3_HTML.gif
Fig. 3

Forest plot for the relationship between s-CTX and fracture risk

The results of the four publications that reported an association between s-CTX and the risk of hip fracture are summarized in Table 4. Note that the four studies have different settings for adjustment. When the results were merged, the HR per SD was 1.23 (95 % CI 1.04–1.47), so that a 1 SD increment in s-CTX was associated with an increased risk of fracture of 23 % (Fig. 4). When the results for women only were merged (three cohorts), the HR per SD was 1.17 (95 % CI 0.95–1.44).
https://static-content.springer.com/image/art%3A10.1007%2Fs00223-014-9842-y/MediaObjects/223_2014_9842_Fig4_HTML.gif
Fig. 4

Forest plot for the relationship between s-CTX and hip fracture risk

Three papers reported results for CTX when adjusted for BMD [7, 15, 16]. The summary estimate of the HR/SD was above unity in all these studies but fell short of statistical significance. When the data were merged, the HR/SD was 1.12 (95 % CI 0.97–1.29).

Discussion

To our knowledge, this is the first meta-analysis of BTMs which was made possible by standardizing the expression of risk as the gradient of fracture risk per SD difference in BTM. Our results indicate a statistically significant but modest association between BTMs and future fracture risk for the recommended reference analytes for bone formation, s-PINP, and for bone resorption, s-CTX not adjusted for BMD. Fracture risk increased by approximately 20 %, depending on the analyte and the outcome fracture that was studied. These gradients of risk are substantially lower than those reported for the use of femoral neck BMD in the prediction of fracture [2123]. For example in a large meta-analysis, the GR for hip fracture with BMD was 2.21 (95 % CI 2.03–2.41) and that for other osteoporotic fractures was 1.56 (95 % CI 1.49–1.64) [22]. The comparative value of BTMs and BMD should be cautiously interpreted, however, since the GRs with BMD for fracture risk are age-dependent [22]. Thus, for osteoporotic fractures, the GR for BMD is 1.62 at the age of 80 years, falling progressively to 1.19 at the age of 50 years. It is not known whether there is any age interaction between the BTMs and fracture risk. Also, we were not able to determine to what extent fracture risk prediction was independent of BMD. BTMs would be particularly helpful if their association with fracture risk were independent of BMD. The literature on this point is inconsistent, with some studies reporting that the prediction of fracture with BTMs was independent of BMD [7, 17, 24, 25] and others failing to show such an effect [15, 26].

A strength of the study is that we were able to standardize the metric of predictive power (ability). The metric used was the GR—namely, the increase in fracture HR between two individuals, who differ by 1 SD in BTM. This has the advantage of maximizing the use that can be made of publications that used differing indices of risk. The approach is robust in that it works independently of the distribution of BTMs. The limitation of this approach is that the assumption is made that the GR is similar over the entire range of BTM values. This assumption has not been formally demonstrated [1], and in the absence of access to primary data, we were unable to test this in the present study. For the same reason, we were unable to examine the ratio of the two reference analytes.

Further limitations include the disparate fracture outcomes used in the present study, and future studies would benefit from standardizing the fracture outcomes reported. An important limitation is publication bias. Indeed, bias is evident in the published literature in the sense that no useful numerical data were supplied for nonsignificant results in three of the papers identified [1113]. Another consideration that we were unable to address is the interaction with time. Ivaska et al. [14] reported with s-CTX that the HRs were highest for any fracture in the beginning of the follow-up period, 2.5 years from baseline. Loss of predictive value with time also has been reported in the case of another BTM [27]. With a 5-year follow-up period, the ratio of carboxylated to total serum osteocalcin was a very strong risk factor for fracture (HR = 5.32, 95 % CI 3.26–8.68) over the period of observation, but the predictive value of the marker lasted only 3 years. The length of follow-up is different in the various publications used in this meta-analysis, exceeding 5 years in four cohorts [7, 14, 15, 17]. Since we have previously shown that a long follow-up can hide an association that is stronger in the beginning of the follow-up [28], the association may have been stronger in the present study if all cohorts had contributed with shorter follow-up or if consideration of this phenomenon had been incorporated at primary analysis. Also, we did not have information about recent fracture in the cohorts. BTMs are known to increase after a fracture [1]. Thus, the association between markers and fracture risk may be confounded by a history of prior fracture. Finally, the sample handling was ill reported, and suboptimal sampling and processing times may have negatively affected the associations found.

We conclude that there is a moderate but significant association between the BTMs studied (s-CTX and s-PINP) and risk of future fractures not adjusted for BMD. Clearly, more studies are required to better evaluate the independent role of BTMs in fracture risk prediction. The use of common reference BTMs in prospective cohort studies with the standardization of their measurements, as recommended by the International Osteoporosis Foundation and the International Federation of Clinical Chemistry and Laboratory Medicine, will help to address these important issues.

Acknowledgements

H. J. was supported by an ESCEO-AMGEN Osteoporosis Fellowship Award. Amgen had no input into the analysis plan or the writing of this report.

Copyright information

© Springer Science+Business Media New York 2014

Authors and Affiliations

  • Helena Johansson
    • 1
  • Anders Odén
    • 1
  • John A. Kanis
    • 1
  • Eugene V. McCloskey
    • 1
  • Howard A. Morris
    • 2
  • Cyrus Cooper
    • 3
    • 4
  • Samuel Vasikaran
    • 5
    • 6
  • IFCC-IOF Joint Working Group on Standardisation of Biochemical Markers of Bone Turnover
  1. 1.WHO Collaborating Centre for Metabolic Bone DiseasesUniversity of Sheffield Medical SchoolSheffieldUK
  2. 2.School of Pharmacy and Medical SciencesUniversity of South AustraliaAdelaideAustralia
  3. 3.The MRC Epidemiology Resource CentreSouthampton General Hospital, University of SouthamptonSouthamptonUK
  4. 4.NIHR Musculoskeletal Biomedical Research Unit, Institute of Musculoskeletal SciencesUniversity of OxfordOxfordUK
  5. 5.Department of Core Clinical Pathology and Biochemistry, PathWest Laboratory MedicineRoyal Perth HospitalPerthAustralia
  6. 6.School of Pathology and Laboratory MedicineUniversity of Western AustraliaNedlandsAustralia