Current issues of ACP Journal Club are published in Annals of Internal Medicine


Meta-analysis: Bone mineral density measurement predicts risk for fractures in women

ACP J Club. 1996 Sept-Oct;125:48. doi:10.7326/ACPJC-1996-125-2-048

Source Citation

Marshall D, Johnell O, Wedel H. Meta-analysis of how well measures of bone mineral density predict occurrence of osteoporotic fractures. BMJ. 1996 May 18;312:1254-9.



To determine whether bone mineral density measurements can predict subsequent fractures in women.

Data sources

Studies published from 1985 to 1994 were identified with MEDLINE, EMBASE, and SweMed databases using the terms bone and bones, bone density, bone mineral content, and densitometry; bibliography scans; and contact with colleagues.

Study selection

English-language studies were selected if the participants were adult women; bone density had been measured with absorptiometry, quantitative computed tomography, quantitative magnetic resonance imaging, or ultrasound scanning; bone density had been done at baseline, and women were followed for fracture occurrence; and women were not receiving treatment for bone or hormonal disorders. 3 reviewers independently screened the citations.

Data extraction

Data were extracted on participant selection and follow-up, age at baseline, bone density measurements (proximal or distal radius, hip lumbar spine, and calcaneus), identification of fractures, and fracture type (forearm, hip, and vertebra). The main outcome was a pooled relative risk (RR) for fracture associated with a decrease of 1 standard deviation (SD) in bone density adjusted for age.

Main results

11 cohort studies (90 000 person y and > 2000 fractures) and 8 case-control studies were included. Combining all sites, with measurements by all methods, the RR for fractures was 1.5 (95% CI 1.4 to 1.7) for a standard deviation decrease in bone density. Results for predicting all fractures from measurement at specific sites (radius, hip, lumbar spine, calcaneus) were similar. Measurements at the hip gave better predictions for hip fractures (RR 2.6, CI 2.0 to 3.5) and measurements of the lumbar spine gave better predictions for vertebral fractures (RR 2.4, CI 1.8 to 3.2). The combined case-control studies had weighted odds ratios for hip fracture predicted by a decline of 1 SD in bone density measurement at the femoral neck of 2.7, at the trochanter of 2.8, at the Ward triangle of 2.1, and at the lumbar spine of 1.8.


Bone density measurements can predict the risk for fracture in adult women.

Source of funding: Swedish Council on Technology Assessment in Health Care.

For article reprint: Professor H. Wedel, Department of Epidemiology and Biostatistics, Nordic School of Public Health, Box 12133, S-40242, Gothenburg, Sweden. FAX 46-31-69-1777.


The role of bone density measurement in postmenopausal women is controversial because it fails to perfectly discriminate between women with and without fractures. Low bone density is a risk factor for fractures, however, and assessing bone density is claimed to be of value for predicting fracture risk. The meta-analysis by Marshall and colleagues provides good evidence from prospective studies to quantitate how well lower bone density estimates the risk for future fractures.

The methods used by Marshall and colleagues are of high quality for systematically reviewing the literature on the measurement of bone density and the statistical pooling of study findings. The review clearly shows that measuring bone density is at least as good for predicting fractures as high blood pressure is for predicting stroke and high serum cholesterol is for predicting coronary heart disease.

Do the findings of this study indicate that population-based screening for osteoporosis should be encouraged? The answer is no. Unlike hypertension, no adequate evidence from large, randomized trials exists to show that treating osteoporosis prevents fractures. Some promising interventions for preventing hip fracture include bisphosphonate and alendronate; combined calcium and vitamin D supplements; and external hip protectors.

As for the timing of bone density measurements, the meta-analysis by Marshall and colleagues provides limited evidence to show that a single bone density measurement can predict fractures equally well for 10 years as for 2 years.

On a methodological note, meta-analyses of observational epidemiologic studies have often been criticized. This criticism, however, is only relevant to meta-analyses of studies of causal relations, where the possibility of uncontrolled confounding always exists. Confounding is not an issue in predictive studies and, therefore, meta-analyses of such predictive studies as this one areappropriate.

Robert G. Cumming, MB, BS, MPH, PhD
University of SydneySydney, New South Wales, Australia