Precision measures the reproducibility of a bone densitometry technique, and is usually expressed as a coefficient of variation (CV) or standardised CV, which takes into account the range of measurements of the particular method (Gluer et al. 1995). To be clinically useful the precision needs ideally to be in the region of 1 %, and certainly better than 3 %. The precision for total hip and lumbar spine is approximately 1 %; for femoral neck and trochanter CV it is 2.5 % and for Ward’s area it is 2.5–5 %. In peripheral sites precision is 1 % in the distal radius, 2.5 % in the ultra-distal radius and 1.4 % in the calcaneus (Grampp et al. 1993; Pacheco et al. 2002; Shepherd et al. 2006). The measurement sites generally used in clinical diagnosis, in contrast to research studies, are therefore lumbar spine (L1–4), femoral neck and total hip (Kanis and Gluer 2000). Precision can be measured in either phantoms, normal individuals or in patients with osteoporosis. Precision is optimum in phantoms, and will be less good in patients with osteoporosis than in normal people, because positioning is more problematic in the former. Precision can be calculated by making repeat BMD measurements in the same individual after repositioning (usually a minimum of 10, but preferably 20 individuals or patients), although the International Society for Clinical Densitometry recommends 100 repeat scans (ISCD Position Development Conference—Writing Group 2004). Departments performing bone densitometry should ideally calculate their own precision but this is not always feasible. Precision is optimised by using the minimum number of expert and highly motivated and well trained technical staff; it is not ideal to have a large number of staff who rotate through different departments and perform bone density scanning only infrequently.

Accuracy is how close the BMD measured by densitometry is to the actual calcium content of the bone (ash weight). The accuracy of DXA lies between 3 % and 8 %. The inaccuracies are related to marrow fat and DXA taking soft tissue as a reference (Tothill and Pye 1992; Blake et al. 1999; Blake et al. 2009). Although DXA measurements are affected by accuracy errors similar inaccuracies are present when other basic clinical measurements are made of blood pressure and body temperature (Blake et al. 2012).

Specificity is the ability of the measurement to discriminate between patients with and without fractures, and to measure small changes with time and/or treatment. A statistically significant change in BMD is calculated from the precision of the measurement technique. To reach a statistically significant change, the BMD has to increase or decrease by at least 2.77 times the precision error. This is termed the least significant change (LSC) and implies that changes in BMD of 3–4.5 % in the lumbar spine and total hip, and of 6–7.5 % in the femoral neck BMD (Gluer 1999; Bonnick et al. 2001) have to be present for change in BMD to be significant. Changes in bone density are generally small; even in the early post-menopausal period in women, when bone loss is greatest, bone density decrements may only be in the region of 1–2 % per annum. Therefore, when performing follow-up BMD measures in an individual patient it is essential to leave an adequate time interval between measures, usually 18–24 months (Gluer 1999), unless particularly large changes in BMD are expected, for example after large doses of oral glucocorticoids (International Society for Clinical Densitometry 2007).

Since whether a patient develops a fracture depends on factors in addition to BMD (whether the patient falls, the nature of the fall, the patient’s age and response to the fall), it is impossible for BMD techniques to completely discriminate between those with and without fractures (Stone et al. 2003). However, approximately 60–70 % of bone strength depends on BMD (Ammann and Rizzoli 2003), the lower the BMD the more at risk the patient is of suffering a fracture and fracture prediction is optimum from the site-specific DXA BMD measurement (Marshall et al. 1996).

DXA provides an “areal” density (g/cm²) of integral (cortical and trabecular) bone. The cortical/trabecular ratios vary in different sites (Eastell et al. 1989; Faulkner et al. 1991), being approximately:

As a result of the different composition of bones and rates of change in these various skeletal sites, it is not surprising that measurements in different sites in the same individual will not give the same results (Eastell et al. 1989). The correlations between the BMD measurements made in the same patient vary between r = 0.4 and r = 0.9; it is not possible to predict from a DXA BMD measurement made in one site, what the BMD will be in another site using DXA or other bone densitometry methods (Grampp et al. 1997). In research studies, BMD measurements in different anatomical sites and by various bone density methods (DXA, QCT, QUS) may be complementary.

These quantitative photon absorptiometric techniques involve very low radiation doses, which are similar to those of natural background radiation (NBR; 2400 μSv per annum; about 7 μSv per day) (Kalender 1992; Huda and Morin 1996; Blake et al. 2006; Damilakis et al. 2010) (Table 1). For the first-generation pencil-beam scanners the dose per site scanned was about 1 μSv, but the dose may be up to 6 μSv for the hip scan in pre-menopausal women, when the ovaries are included in the scan field (Lewis et al. 1994). The doses are a little higher for the fan-beam scanners (generally 3–9 μSv), but were up to 62 μSv with some scanners (Lunar Expert, which is no longer manufactured) (Eiken et al. 1994; Njeh et al. 1996). Radiation doses for forearm and calcaneus scans are extremely low (0.5 and 0.03 μSv respectively). These DXA doses are less than one-quarter of the dose of a chest radiograph (20 μSv), and considerably lower than doses involved in other radiographic examinations carried out to confirm the diagnosis of osteoporosis in patients at risk; lateral spinal radiographs involve doses in the region of 300–700 μSv per projection, depending on patient size and exposure factors (Damilakis et al. 2010).

There has been much debate concerning the appropriate use of bone densitometry, particularly in population screening in women at menopause (Melton et al. 1990), and the cost-effectiveness of such a programme has not been established. However, there is now consensus that DXA bone densitometry is the quantitative method of choice for the diagnosis of osteoporosis before fractures occur (Compston et al. 1995; Kanis et al. 1997, 2009; Adams 2013). DXA bone densitometry has high specificity but low sensitivity. Selection of patients who would most appropriately be referred for DXA bone densitometry was in the past based on a case-finding strategy in those who have had an insufficiency (low trauma) fracture or have other strong risk factors (Royal College of Physicians 1999), and there were national differences in such referral guidelines.

A new approach for appropriate DXA scan referral and interpretation was introduced by the World Health Organisation (WHO) in 2010. This tool is based on using clinical risk factors (Table 2), with or without femoral neck BMD, to estimate a patient‘s 10-year probability of a fracture at the hip or one of the major osteoporotic sites (Johnell et al. 2005; Kanis et al. 2007; Compston 2009a). The latter are defined as the hip, spine, forearm and humerus. The FRAX WHO fracture risk assessment tool is accessed using the web site http://www.shef.ac.uk/FRAX (World Health Organisation 2010), is applicable in women and men aged 40–90 years, is currently available for use in 30 different countries and has been incorporated into manufacturers’ DXA scan reports (Fig. 5a, b). The FRAX algorithm was developed from data based on meta-analyses from nine different international (North America, Europe and Asia) fracture studies. These included 46,000 men and women with 190,000 person-years of follow-up, and 850 cases of hip fracture and 3,300 other osteoporotic fractures (Kanis et al. 2007; Blake et al. 2012). FRAX is increasingly being incorporated into guidelines on the appropriate referral for DXA and treatment of osteoporosis, but these vary between nations. In the United States the National Osteoporosis Foundation (NOF) has added FRAX to the criteria for treating patients using therapeutic threshold fracture risk levels of 3 % for hip fracture and 20 % for a major osteoporotic fracture (National Osteoporosis Foundation 2010). In the United Kingdom a different treatment algorithm was produced by the National Osteoporosis Guidelines Group (NOGG), which recommended that FRAX is used to select the patients who would most appropriately be referred for a DXA scan (Kanis et al. 2008; Compston et al. 2009) (Fig. 5a, b). FRAX should lead to a more appropriate utilisation of DXA scanning and results, and more cost-effective intervention strategies (Johansson et al. 2012). Whether other clinical risk factors and additional quantitative skeletal assessments can be added to the FRAX calculator is being explored (Binkley and Lewiecki 2010; Lewiecki et al. 2011).

When a BMD measurement has been made in a patient, this has to be interpreted as normal or abnormal and a report formulated that will be of assistance to the referring clinician (Miller et al. 1996). For this it is essential that age-, sex- and racially matched reference data are available. The scanner manufacturer supplies such normal reference databases. These databases are predominantly, but not exclusively, drawn from a white, Caucasian, American-based population. There is a paucity of appropriate reference ranges for children and certain ethnic minorities (e.g. Asians; Afro-Caribbean). A patient’s results can be interpreted in terms of the standard deviations (SD) from the mean of either sex-matched peak bone mass (PBM) (T-score) or age-matched BMD (Z-score) (Parfitt 1990). Alternative methods of interpretation are as a percentage or percentile of expected PBM, either gender- and age-matched or just gender-matched.

The WHO has defined osteoporosis in terms of bone densitometry. A T-score of less than −2.5 defines osteoporosis (World Health Organisation 1994). This was arbitrarily defined as the level of BMD in post-menopausal women that identified approximately 30 % of that population as having osteoporosis in their lifetime (this is thought to be the percentage of post-menopausal women who will suffer a vertebral fracture in their lifetime). The definition applied to DXA measurements made in the lumbar spine, the proximal femur and the forearm. The definition does not apply to other techniques (e.g. QCT, QUS) or other anatomical sites (e.g. calcaneus) (Grampp et al. 1997; Faulkner et al. 1999; Miller 2000), nor is it yet confirmed to be applicable to younger women and men.

T-scores are calculated by taking the difference between a patient’s measured BMD and the mean BMD in healthy young adults matched for gender and ethnicity, and expressing the difference relative to the young adult population SD:

A T-score measurement of less than, or equal to, −2.5 at the spine, distal 1/3 radius, femoral neck or total hip sites is taken to indicate osteoporosis. Measurements between −2.5 and −1.0 are interpreted as indicating osteopenia, while measurements greater than −1.0 at all three sites are classified as normal (World health Organisation 1994). A limitation of this T-score method for making decisions about patient therapies is that factors such as age and a history of previous insufficiency fracture are independent risk factors that are as important as BMD in determining the 10-year risk of fracture. In the FRAX fracture risk calculator a selected list of clinical risk factors (Table 2) is used together with femoral neck BMD to improve fracture risk prediction.

Z-scores compare the patient’s BMD with the mean BMD for a healthy subject matched for age, gender and ethnicity (International Society for Clinical Densitometry 2007; Chun 2011).

Until PBM has been reached (i.e. in children and young adults up to approximately 19 years) interpretation can be made only by comparison to the age-matched mean (Z-score) (Faulkner et al. 1993b; Gordon et al. 2008).

BMD measurements in the whole skeleton can be performed using total body scans (Nuti and Martini 1992). These are usually interpreted after excluding the head from the scan analysis, particularly in children (Lewiecki et al. 2008; Ward et al. 2007).

Variations in the mean and standard deviation of reference ranges may alter the number of patients identified as osteoporotic (Ahmed et al. 1997); in DXA of the hip, use of the NHANES reference database is preferred for white Caucasians (Looker et al. 1998).

In calculating change over time, the absolute BMD values (g/cm²) have to be used. To be statistically significant the change in BMD has to be 2.77 times greater than the precision to reach the least significant change (LSC); in longitudinal studies in individual patients one needs to leave an intervening period of at least 18–24 months between measures to ensure significant change in individual patients (Gluer 1999).