Significant Differences in UK and US Female Bone Density Reference Ranges

Abstract

Summary

In the United Kingdom (UK) T- and Z-scores are usually calculated using reference ranges derived from United States (US) populations. In the UK arm of a recent randomised trial (IBIS-II) substantially fewer women than expected were recruited into the osteopenic (-2.5 < T-score < −1.0) and osteoporotic (T-score < −2.5) arms of the study. Comparison with data from two independent studies showed that UK women aged > 45 years with a typical body mass index of 28 kg m⁻² have spine and hip BMD 0.6 SD higher than their US counterparts.

Introduction

Dual energy X-ray absorptiometry (DXA) is widely used for the diagnosis of osteoporosis and to investigate the effect of pharmacological treatments on bone mineral density (BMD). In both routine and research settings it is important that DXA results are correctly interpreted.

Methods

Z-scores for the first 650 UK Caucasian women enrolled in the IBIS-II study were compared with data from two independent studies of unrelated, unselected UK Caucasian women: (1) 2382 women aged 18 to 79 recruited to the Twins UK adult twin registry; (2) 431 women aged 21 to 84 with no risk factors for osteoporosis recruited at Guy’s Hospital. All DXA measurements were performed on Hologic densitometers. Subjects were divided into six age bands and Z-scores calculated using the manufacturer’s US reference range for the spine and the NHANES III reference range for the femoral neck and total hip.

Results

The overall mean Z-scores for the IBIS-II, Twin and Guy’s groups were: spine: 0.61, 0.29, 0.33; femoral neck: 0.42, 0.36, 0.45; total hip: 0.65, 0.38, 0.39 (all p < 0.001 compared with the expected value of 0). The mean body weight of subjects in the three studies was 74.4, 65.5 and 65.4 kg respectively. Analysis revealed a highly significant relationship between Z-score and weight at each BMD site with a slope of 0.03 kg⁻¹.

Conclusions

In general US spine and hip reference ranges are not suitable for the calculation of Z-scores in UK women. For some research study designs the differences may significantly influence the pattern of subject recruitment.

INTRODUCTION

Measurements of bone mineral density (BMD) of the spine and hip using dual energy X-ray absorptiometry (DXA) have become the ‘gold standard’ for the diagnosis of osteoporosis [1]. DXA is also used in research studies to investigate the effect of pharmacological treatments on BMD [2-6]. DXA measurements are interpreted using T-scores and Z-scores, where T-scores express the difference in standard deviation units between a subject’s BMD and the mean figure in healthy young adults, while Z-scores express the difference from a healthy age matched reference population [7]. In both clinical and research settings it is important that BMD results are correctly interpreted, and in the United Kingdom (UK) T- and Z-scores are frequently calculated using reference ranges derived from United States (US) populations. These are the manufacturer’s US reference range for the spine [8,9] and the National Health And Nutrition Examination Survey (NHANES) III reference range for hip sites [10]. Accurate calculation of T- and Z-scores relies upon the gender and ethnic-matched reference population being representative of the local population.

The International Breast Cancer Intervention Study II (IBIS-II) is an international, multi-centre, double-blinded, randomised controlled trial of anastrozole vs. placebo in postmenopausal women at increased risk of developing breast cancer. In the IBIS-II Bone Sub-Protocol study a group of subjects enrolled in the main breast cancer prevention trial were invited to take part in an additional study to examine the effectiveness of risedronate at preventing bone loss associated with anastrozole treatment [11]. Women participating in the bone sub-study were enrolled into one of three strata: 300 subjects with osteoporosis (T-score ≤ −2.5) at baseline, who received risedronate 35 mg once weekly; 400 subjects with osteopenia (−2.5 < T-score < −1.0), who were randomised to receive either risedronate or a placebo; and 300 subjects with a normal BMD (T-score ≥ −1.0), none of whom received bone protection therapy.

During enrolment for the bone sub-study recruitment was slower than expected because of a deficit of subjects enrolled in the osteoporotic and osteopenic groups. Instead of the predicted proportions of 30%: 40%: 30% for the normal, osteopenic and osteoporotic groups the ratio was 57%: 33%: 10%. To study the reasons for this, Z-scores in the first 650 UK Caucasian women recruited to the IBIS-II bone sub-study were compared with data from two independent studies of UK women: 2382 women recruited to the Twins UK adult twin registry [12], and 431 women recruited at Guy’s Hospital, London [13]. It was noticed that women enrolled in the IBIS-II bone sub-study were on average 10 kg heavier than the women in the Twin Registry and Guy’s Hospital studies. Since body weight is known to have a significant effect on individual’s BMD measurements the role of body weight and its influence on Z-score values was investigated.

The present study had two aims: (1) to investigate mean T- and Z-scores calculated using US reference ranges in UK Caucasian women enrolled in the IBIS-II bone sub-study and compare them with T- and Z-scores figures found in two independent studies of healthy UK women; (2) to investigate the effect of body weight on Z-score values and its contribution to the BMD differences between the three study groups.

SUBJECTS AND METHODS

IBIS-II study

Women could be enrolled in the IBIS-II breast cancer prevention study if they were postmenopausal, had no radiological evidence of breast cancer, and had a two-fold relative risk increase of developing breast cancer, which was defined by: (i) relatives having developed breast or ovarian cancer; (ii) nulliparous; (iii) radiological evidence of benign disease; and other more specific criteria. Subjects with low bone density could only participate if they agreed to take bisphosphonates. The IBIS-II Bone Sub-Protocol study enrolled an unselected group of subjects from the main prevention study to assess whether risedronate treatment attenuates the deleterious effects of anastrozole on BMD [14]. The present investigation used the BMD data at enrolment for the first 650 UK Caucasian women scanned on Hologic DXA densitometers taking part in the IBIS-II Bone Sub-Protocol study. Women participating in the Bone Sub-Protocol were a representative sample of the main IBIS-II study.

Twin Study

The Twins UK adult twin registry is a volunteer cohort of adult twins aged 18-85 that presently includes approximately 10,000 monozygotic and dizygotic twins, recruited by media campaigns from all over the UK [12]. A database of DXA BMD measurements and demographic data from the registry was used in this study. Dizygotic twins have DNA that is approximately 50% identical to each of their siblings (twin or not), and monozygotic twins have identical DNA. To rule out any genetic bias, results from just one of each pair of twins having BMD measurements were included in the present investigation, leaving data for 2382 singleton Caucasian females. Studies have shown that this population is similar to singleton populations derived from age-sex registers such as the Chingford study [15].

Guy’s Study

This population consisted of 878 Caucasian women recruited from three sources: patients referred by their general practitioner for routine bone density screening by DXA; young hospital personnel; and women from the general population who volunteered to participate in clinical research [13]. Subjects were enrolled for DXA measurements of spine and hip BMD. To obtain a group of healthy women, those with risk factors known to influence bone metabolism or with a history of low-trauma fracture were excluded, leaving 431 subjects [13].

MEASUREMENTS

BMD measurements of the lumbar spine and proximal femur (non-dominant hip) were obtained using Hologic QDR-4500 densitometers (Hologic Inc., Bedford, MA), except for some early data from the Twin study that were obtained using a Hologic QDR-2000. All scans were performed by trained technologists. IBIS-II scans performed at 14 different UK centres were analysed at a central quality control centre at Guy’s Hospital. Regions of interest analysed were the total spine (L1-L4), femoral neck and total hip. Subjects’ age, height and weight were also recorded. All the densitometers were installed and calibrated by the manufacturer’s UK representative. A review of daily scans of the Hologic spine phantom on each densitometer showed precision figures that averaged 0.38% (range 0.30% to 0.54%). Spine phantom BMD measurements averaged 100.02% of the nominal values (range 99.61 to 100.30%). A study of long-term in-vivo precision gave figures of 1.6%, 2.5% and 1.6% for spine, femoral neck and total hip BMD respectively [16]. A separate in-vivo precision study performed for the Twins UK adult twin registry gave precision results of 0.8% at the spine and 1.6% at the femoral neck [17].

DATA ANALYSIS

The data were divided into six age bands: <25, 25-34, 35-44, 45-54, 55-64, and ≥65 years. Weight, height and body mass index (BMI = wt(kg)/ht(m)²) data for the three study groups were compared with recently published figures for the UK female population [18] and the results for non-Hispanic white women from the US NHANES III study [19]. Z-scores were calculated for all subjects by the difference between the subject’s and the age-matched mean BMD, expressed in standard deviation (SD) units. T-scores were calculated with respect to the young adult reference values. Spine T- and Z-scores were calculated using the manufacturer’s US spine reference range [8,9] and hip scores were calculated using the NHANES III range [10]. DXA manufacturers have adapted the NHANES results as their reference data for calculating T- and Z-scores at the proximal femur [20].

The mean and SD of the T- and Z-scores values were calculated for each age group in each study population to demonstrate gross differences from the reference populations for the spine, femoral neck and total hip regions of interest. If the reference range correctly describes the study population, the mean Z-score will be Z=0 and SD=1.0. T- and Z-scores values are related by the following equation [21]:

$$\text{T-score}=\text{Age Specific Mean T-score}+\text{Z-score}$$ (1)

In Equation 1 the Age Specific Mean T-score is the curve of mean T-score as a function of age obtained by expressing the BMD reference range in T-score units [22]. Student’s t-tests were performed to examine whether the mean Z-scores were significantly different from 0 [23]. The Z-score SDs were tested against the reference population SD of 1.0 using the F-test [23]. Statistical significance was based on a p-value of less than 0.05.

To examine the effect of body weight on BMD each subject’s Z-score was plotted against their weight and a linear regression line fitted to the points [23]. The plot was performed for each study group at each BMD site. The weight coefficient – the gradient of this regression line – was calculated for each plot together with the intercept on the weight axis.

RESULTS

Demographic data for the 3 study groups including figures for weight, height and BMI broken down by age are shown in Table 1. The absence of subjects aged <45 y in the IBIS group reflects the entry requirement to be postmenopausal. Ages in the IBIS group ranged from 45 to 71 y, for the Twin study from 18 to 79 y, and for the Guy’s study from 21 to 84 y. The number of subjects in each age group varied from 11 to 742 (Table 1).

TABLE 1.

	IBIS-II Bone Sub-Study				Twin Study				Guy’s Study
Age	n	W	H	BMI	n	W	H	BMI	n	W	H	BMI
<25					142	63.1	1.65	23.1	11	69.8	1.71	23.8
25-34					315	64.8	1.64	24.1	45	64.6	1.66	23.5
35-44					496	65.4	1.63	24.5	63	67.2	1.64	25.0
45-54	183	74.2	1.63	28.0	742	65.9	1.63	25.0	160	64.6	1.64	24.1
55-64	372	74.3	1.62	28.2	534	66.7	1.61	25.6	94	66.8	1.62	25.6
≥65	94	75.3	1.61	29.1	153	64.1	1.59	25.4	58	64.1	1.58	25.6
Tot/Av	650	74.4	1.62	28.3	2382	65.5	1.63	24.8	431	65.4	1.63	24.7

n=number in group W=weight, kg; H=height, metres; BMI=body mass index, kg/m²; Tot/Av=Total n, Average W, H or BMI

The Twin study and Guy’s study have similar overall mean weights of 65.5 kg and 65.4 kg respectively, while the IBIS study was higher at 74.4 kg. Except for the <25 y group in the Guy’s study (n=11), mean weights in the Twin and Guy’s study groups ranged between 63.1 and 67.2 kg. Mean weights in the IBIS study group were between 74.2 and 75.3 kg, consistently higher than the other two studies.

Table 2 summarises recently published weight, height and BMI data for UK women [18] together with similar data for non-Hispanic white females from the US NHANES III study [19]. For the IBIS study women the mean BMI was 28.3 kg/m². For comparison the mean BMI for women aged >45 y was 27.9 kg/m² for UK women, 26.8 kg/m² for US women in the NHANES III study, 25.3 kg/m² for the Twin study, and 24.8 kg/m² for the Guy’s study. For women <45 y in the Twin and Guy’s studies the mean BMI figures were closer to the population values for UK and US women. Mean heights for women >45 y in all 3 studies were similar to the population figures for UK and US women (IBIS study: 162.2 cm; Twin study: 161.7 cm; Guy’s study: 162.0 cm; UK women: 160.6 cm; US women 161.1 cm). Mean weights for women >45 y reflected the differences found in BMI (IBIS study: 74.4 kg; Twin study: 66.0 kg; Guy’s study: 65.2 kg; UK women: 71.5 kg; US women 67.8 kg).

TABLE 2.

UK - NHS [18]					US - NHANES III [19]
Age	n	W	H	BMI	Age	n	W	H	BMI
16-24	294	65.0	1.63	24.4	20-29	452	63.7	1.64	23.8
25-34	453	68.1	1.63	25.7	30-39	579	68.9	1.65	25.5
35-44	649	71.3	1.63	26.8	40-49	477	70.5	1.64	26.4
45-54	527	71.8	1.62	27.4	50-59	484	74.3	1.63	28.2
55-64	538	72.9	1.61	28.2	60-69	501	70.7	1.61	27.3
65-74	393	71.2	1.59	28.0	70-79	643	66.8	1.58	26.7
≥75	281	64.9	1.55	26.9	≥80	621	60.4	1.55	25.1
Total/Av	3135	69.7	1.61	26.8	Total/Av	3757	68.5	1.62	26.0

n=number in group; W=weight, kg; H=Height, metres; BMI=body mass index, kg/m²; Total/Av=Total n, Average W, H or BMI NHANES III data are for non-Hispanic white females

Mean T- and Z-score vs. Age

Spine

Mean T-scores for each age group in all three studies are shown plotted in Figure 1A together with the manufacturer’s spine reference range. Although for the 25-34 y age group there is good agreement with the reference data, at other ages there is a trend for the figures to increasingly diverge from the reference curve with increasing age.

FIGURE 1.

Mean T-score vs. age for A: lumbar spine; B: femoral neck; C: total hip BMD for subjects in the three study groups. Error bars show ± 1 SEM. Dashed curves show the Age Specific Mean T-score (see Equation 1) calculated from the BMD reference ranges. Spine reference range: Hologic manufacturer’s [8,9]; femoral neck and total hip: NHANES III [10]. Asterisks show the statistical significance of the difference of the mean Z-score from zero (* = p < 0.05; ** = p < 0.01; *** = p < 0.001)

The mean overall Z-scores expressing the divergence from the reference range (Equation 1) were +0.61, +0.29 and +0.33 for the IBIS-II, Twin and Guy’s studies respectively. For women >45 y in the Twin and Guy’s studies the mean Z-scores were +0.44 and +0.35 respectively The mean Z-score of −0.28 for the <25 y group in the Twin study was the only statistically significant negative figure in the entire study. For the 35-44 y and older age groups the mean Z-scores were statistically significantly greater than zero in all three studies. For the young normal group (25 to 34 y) the mean T-score (SEM) for the pooled data from the Twin and Guy’s studies (N = 360) was −0.05 (0.05).

Femoral Neck

Mean T-scores for each age group and each study are plotted in Figure 1B. At the femoral neck site the mean Z-scores were all statistically significantly greater than 0 except for the <25 y age group in the Guy’s study. In contrast to the trend seen in the spine the hip sites show a similar offset from the NHANES III reference curve across all age groups. The mean overall Z-scores were +0.42, +0.36 and +0.45 for the IBIS-II, Twin and Guy’s studies respectively. For women >45 y in the Twin and Guy’s studies the mean Z-scores were +0.39 and +0.43 respectively. For the young normal group (20 to 29 y) the mean T-score (SEM) for the pooled data from the Twin and Guy’s studies (N = 276) was +0.32 (0.06).

Total Hip

Mean T-scores for each age group and each study are plotted in Figure 1C and are systematically higher than the NHANES III reference curve. The mean Z-score results for each age group were all statistically significantly greater than 0. The mean overall Z-scores for the total hip site were +0.65, +0.38, and +0.39 for the IBIS-II, Twin and Guy’s studies respectively. For women >45 y in the Twin and Guy’s studies the mean Z-scores were +0.41 and +0.40 respectively. For the young normal group (20 to 29 y) the mean T-score (SEM) for the pooled data from the Twin and Guy’s studies was +0.34 (0.06).

In the IBIS study Z-score data hip SDs were within ±0.18 of 1.0, but the spine showed consistently higher SDs rising with age from 1.29 to 1.84. In the Twin study, the standard deviation of Z-scores was within ±0.12 SDs from 1.0, except in the older age groups at the spine where the SD increased from 1.27 at 45-54 y to 1.40 in the oldest subjects. A small sample size increases the statistical error so results from the <25 y age group in the Guy’s study (n = 11) are not significant. Otherwise the Z-score SD was within ±0.2 SD from 1.0, except for the older spine groups where it increased from 1.31 in the 55-64 y group to 1.47 in the ≥65 y group.

Z-score vs. Weight

Subjects’ Z-scores were plotted against their weight for each study and each ROI and a linear regression line was fitted to these points. Figure 2 is a representative example of these plots. The plots demonstrate a greater spread of Z-scores in the spine across all studies, but no significant differences otherwise.

FIGURE 2.

Scatter plot of total hip Z-score vs. body weight for subjects in the Twin study. The straight line shows the regression line fitted to the data.

Gradients of the regression lines for the Z-score vs. weight plots were calculated and are shown in Figure 3A. All the gradients were statistically significantly different from 0 (p < 0.001) and within the statistical errors approximated to the same figure (Chi-squared = 5.8; df = 8; P = 0.67). The weighted mean gradient was 0.034 kg⁻¹.

FIGURE 3.

A: Plot of the slope of the Z-score vs. body weight regression line for spine, femoral neck and total hip Z-score for subjects in the three study groups (*** = p < 0.001). An example of a Z-score vs. weight regression line is shown as the solid line in Figure 2.

B: Plot of the intercept on the weight axis of the Z-score vs. body weight regression line for spine, femoral neck and total hip Z-score for subjects in the three study groups. The weight intercept is the intercept of the regression line (solid line in Figure 2) on the dashed line (Z-score = 0) in Figure 2.

Intercept points on the weight axis were also calculated from the regression lines in the Z-score vs. weight plots (Figure 3B). These equate to the weight in kg that gives a mean Z-score of 0. The largest of the intercepts was 60.8 kg in the femoral neck from the IBIS study, and the smallest was 50.2 kg again in the femoral neck, from the Guy’s study. Overall the mean intercept was 55.3 kg.

DISCUSSION

Subjects in the Twin and Guy’s studies were non-selected, included a wide range of ages, and were healthy UK Caucasian women. Using subjects from two independent studies makes errors due to DXA calibration, operator technique or subject selection bias less likely. The issue of whether women with a diagnosis of breast cancer have higher than average BMD is controversial [24,25]. However, an increased risk of breast cancer of itself is not known to affect BMD. Although all three groups might be considered representative of the UK female population, a weight difference of almost 10 kg was found between the IBIS subjects and the Twin and Guy’s groups that would be expected to influence the BMD findings. Comparison with UK and NHANES III anthropometric data shows that the mean BMI of 28.3 kg/m² for the IBIS study women is similar to the figure for UK and US women aged >45 y, while the Twin and Guy’s study women >45 y had a significantly lower BMI than the UK average. In contrast for women <45 y the BMI data for the Twin and Guy’s study women were in better agreement with the UK and US female reference data. Average heights for the women > 45 y in the three study groups of around 1.62 m were similar to the UK and US norms, and so the differences in BMI largely reflect the differences in body weight.

In the UK T- and Z-scores are usually calculated using reference ranges derived from US populations. Historically, there was poor agreement between hip T-scores calculated on GE-Lunar and Hologic densitometers [26]. The NHANES III study included over 14,000 randomly selected individuals representative of the whole US population, making subject selection bias in this reference range unlikely [10]. This extensive sampling of the US population led the principal DXA manufacturers to adopt the NHANES III BMD data for the calculation of T- and Z-scores [20]. In contrast the Hologic spine reference range was generated from volunteers scanned at some of the first DXA centres in the US and includes measurements of 605 women [8,9]. The mean body weight for the women aged >45 y was 63.7 kg [9]. Despite the small sample size and coverage compared with the NHANES III study, there is good agreement between Hologic and GE-Lunar spine T-scores calculated using the manufacturers’ reference ranges [26].

Reference ranges used to calculate Z-scores take into account the bone loss associated with aging, and at each age the BMD data are assumed to follow a normal distribution with the same population SD. Therefore BMD measurements from a large population that match the reference data should have a mean Z-score of 0 and an SD of 1.0 at all ages. In Figure 1 the mean Z-score values are shown as the offset of the mean T-score values from the dashed lines showing the reference curves (Equation 1). At both the femoral neck and total hip sites the mean Z-scores were statistically significantly different from 0 for all ages and all studies, except the <25 y group in the Guy’s study (n=11), with overall mean values of 0.36-0.65 that remained relatively constant across the full age range (Figure 1B,C). At the spine the majority of mean Z-scores were statistically significant from 0, though three groups were not significant and one gave a statistically significant negative figure. In contrast to the hip sites, for the spine there was a trend for the mean Z-score to increase with age. When the Hologic spine reference range was constructed spinal radiographs were taken to exclude subjects with degenerative disease [8]. Therefore some of the deviation seen in Figure 1A might be explained by the increased prevalence of degenerative disease in older subjects. The changes associated with degenerative disease have a much smaller effect on the measurement of hip BMD. In the Twin and Guy’s studies women >45 y had a lower overall mean Z-score at the spine and total hip sites than the IBIS-II subjects (spine: 0.35-0.44 vs. 0.61; total hip: 0.40-0.41 vs. 0.65). The demographic data suggests this might be due to their lower body weight compared with the IBIS-II study. For the young adult subjects in the Twin and Guy’s studies the pooled mean T-scores were in close agreement with the expected value of 0 for spine BMD but for the femoral neck and total hip were statistically significantly different at +0.32 and +0.34 respectively.

For the hip regions the Z-score SDs were constant with age and were not statistically significantly different from 1.0. However, in subjects aged >45 y spine SDs differed significantly from 1.0 with the SD figure increasing with age. There were no gross differences in the SD figures between studies.

The slopes of the regression lines fitted to the scatter plots of Z-score vs. weight showed little variation between the three studies or between the different BMD sites, and were consistent with an overall mean figure of 0.034 kg⁻¹ (Figure 3A). This finding confirms previous evidence that body weight significantly influences BMD and therefore Z-score [27]. The mean weight intercept was 55 kg (Figure 3B), which reflects the body weight at which the US reference ranges would give a mean Z-score figure of 0 in UK women.

Several previous studies have reported similar Z-score discrepancies in UK women. Petley et al. investigated BMD measurements from GE-Lunar densitometers in 700 women from Southampton [28]. They found higher BMD results by about 0.5 SDs in their study population compared with the manufacturer’s US reference ranges for spine, femoral neck and total body BMD (all age points p<0.05 or less). Similar results were found for a large multi-centre database of UK women [29], and also for the Chingford study [15]. The latter included approximately 900 women aged 45-65 y recruited from general practice who were found to have a mean spine Z-score of +0.35 and a mean femoral neck Z-score of +0.16 [15], figures that are both statistically significantly different from zero.

Studies of other Caucasian populations show geographical variation in spine and hip BMD [9,30-32]. The strongest evidence is from the DXA bone density measurements performed during the European Vertebral Osteoporosis Study (EVOS) [30]. A comparison of BMD measurements in 2228 men and 2581 women aged 50-80 y from 16 centres in 12 European countries showed highly significant differences between centres after adjustment for age, height and weight. The between centre SD was approximately 25% of the population SD and the difference in mean BMD between the highest and lowest centres was about 1 SD. The authors concluded that the study demonstrated clear differences between European populations in BMD at the spine and hip. Kaptoge et al reported similar geographical BMD variations among young European men and women [31]. In a study of 1163 men and 329 women aged 19-29 y from 17 centres in 12 countries they reported significant variation in adjusted mean spine and hip BMD levels arising from national differences. Although mean European values for hip BMD were sometimes different to US figures, the SDs were indistinguishable. Flicker et al reported BMD differences between Australian and North American white female populations based on measurements performed on Hologic instruments [9]. Measurements of 411 Australian women aged 15-74 y gave a mean spine Z-score of +0.63 and a mean femoral neck Z-score of +0.37 when interpreted using the manufacturer’s North American spine reference range and the NHANES III hip reference range respectively. A recent study of Australian men from the Geelong Osteoporosis Study showed femoral neck BMD levels similar to those obtained in the US non-Hispanic white males participating in the NHANES III study. However, spine BMD levels differed from figures from France, Canada and Spain even after exclusion of men with vertebral abnormalities on spinal radiographs [32].

This study has emphasised body weight as an important and often neglected factor influencing subjects’ spine and hip BMD results. The two principal DXA manufacturers calculate Z-scores differently, GE-Lunar making an allowance for body weight but not Hologic [33]. The long-term trend towards increasing obesity raises the question of what effect this might have on BMD measurements. Data for UK women show that over the 12-year period from 1993 to 2004 mean body weight over all age groups increased from 66.6 to 69.7 kg, height from 161.1 to 161.4 cm, and BMI from 25.7 to 26.8 kg/m² [18]. There is a steady trend for weight to increase by 0.3 kg/y, corresponding to a predicted Z-score increase of 0.1 per decade.

The results of the present study make it clear that the reason for the slow recruitment of UK women to the osteoporotic and osteopenic arms of the IBIS-II bone sub-study was the unexpectedly high Z-scores of UK women in the >45 y age group. With an average body weight of 74 kg the IBIS-II subjects were the typical weight of UK women in their age group, and similar to the non-Hispanic US white women in the NHANES III study. A difference in body weight is therefore not the explanation for the Z-score differences between UK and US women. In contrast, the mean body weight of 65 kg found in the Twin and Guy’s study women >45 y is not typical of older UK women [18]. With the weight coefficient of 0.03 kg⁻¹ found from the Z-score vs. weight scatter plots, had the women in the Twin and Guy’s studies had the higher body weight found in the IBIS-II and NHANES III studies the Z-score differences between UK and US women would have increased by about 0.2-0.3 SD from around 0.4 SD to around 0.6 SD. However, higher body weight did not explain all the differences between the IBIS-II study Z-scores and the other two studies. At all three BMD sites there was a trend for mean Z-score to increase with age in the IBIS-II women despite the fact that their mean body weight did not change.

Even at the lower than average body weight of 65 kg found in the Twin and Guy’s studies mean spine, femoral neck and total hip Z-scores in UK women calculated using US reference ranges lie in the range 0.3 to 0.4. Only for groups of women with a mean body weight as low as 55 kg would the US reference ranges give appropriate Z-score values in UK women. With the majority of mean Z-score results significantly greater than 0 in all three study populations, we conclude that in general US reference ranges are not reliable for calculating Z-scores values for UK Caucasian women.

In the IBIS-II bone sub-study and other similar research studies involving the selection of subjects based on a BMD measurement, the assignment of individuals to different study groups is often based on the interpretation of the BMD findings in terms of T-scores. It is better that a common reference range such as NHANES III is used for such selection procedures, since the use of different reference ranges for studies performed in different countries can only add to the confusion. Nevertheless, whether the adopted reference range is actually a good description of the population screened for the study can have a significant effect on the enrolment of subjects. In the IBIS-II bone study the unexpectedly high overall mean Z-scores of 0.42-0.65 were sufficient to delay significantly recruitment into the osteoporotic and osteopenic groups.

The choice of reference ranges for calculating T-scores for the diagnosis of osteoporosis has always been a cause of controversy. It is not the intention of the present article to suggest that UK DXA centres should not continue to use US reference ranges for the diagnosis of osteoporosis based on T-scores. In this application it is more important that an accepted reference range such as NHANES III is adopted as widely as possible rather than cause confusion by using inconsistent local reference ranges for calculating T-scores. For example, the use of NHANES III femoral neck T-scores is a requirement when T-scores are used in the World Health Organisation (WHO) FRAX tool for evaluating patients’ 10-year risk of osteoporotic fracture [34]. The availability of the FRAX tool renders the previous controversies over which reference to use less relevant, since DXA measurements can be entered as raw BMD results and the relationship between BMD and fracture risk is derived from population and ethnicity dependent epidemiological data rather than comparison with reference ranges. Z-scores are used to compare patients with their peers in terms of healthy controls matched for age, gender and ethic origin and for GE-Lunar DXA systems body weight too. To this extent clinicians should be aware that the use of US reference ranges to calculate spine and hip Z-scores for UK women with body weights typical of the population may overestimate the true Z-score by 0.6 SD.

This study has several limitations. We used Z-scores calculated from Hologic densitometers only since insufficient data were available to evaluate Z-scores on GE-Lunar systems. Also, this investigation was limited to data from Caucasian women alone and there were no data for male subjects.

In conclusion, the US spine and hip reference ranges are generally not suitable for the calculation of Z-scores in UK women, and for some research study designs these differences may significantly influence the pattern of subject recruitment. The results confirmed that body weight has an important influence on spine and hip BMD measurements, and that for studies enrolling groups of subjects with atypical BMI this may be an additional factor affecting recruitment.