Phalangeal Osteosonogrammetry Study: Age-Related Changes, Diagnostic Sensitivity, and Discrimination Power



Phalangeal osteosonogrammetry was introduced as a method for bone tissue investigation in 1992. It is based on the measure of the velocity of ultrasound (amplitude-dependent speed of sound [AD-SoS]) and on the interpretation of the characteristics of the ultrasound signal. In this study we have collected a database of 10,115 subjects to evaluate the performance of AD-SoS and to develop a parameter that is able to quantify the signal characteristics: ultrasound bone profile index (UBPI). The database only includes females of which 4.5% had documented vertebral osteoporotic fractures, 16% lumbar spine dual X-ray absorptiometry (DXA), and 6% hip DXA. The analysis of the ultrasound signal has shown that with aging the UBPI, first wave amplitude (FWA), and signal dynamics (SDy) follow a trend that is different from the one observed for AD-SoS; that is, there is no increase during childhood. In the whole population, the risk of fracture per SD decrease for AD-SOS was odds ratio (OR) 1.71 (CI, 1.58-1.84). The AD-SoS in fractured subjects was significantly lower than in a group of age-matched nonfractured subjects (p < 0.0001). In a small cohort of hip-fractured patients UBPI proved to be lower than in a control age-matched group (p < 0.0001). When the World Health Organization (WHO) working group criteria were applied to this population to identify the T score value for osteoporosis, for AD-SoS we found a T score of −3.2 and for UBPI we found a T score of −3.14. Sixty-six percent of vertebral fractures were below the AD-SoS −3.2 T score and 62% were below UBPI −3.14. We observed the highest incidence of fractures (63.6%) among subjects with AD-SoS who had both DXA T score values below the threshold. We conclude from this study that ultrasound investigation at the hand phalanges is a valid methodology for osteoporosis assessment. It has been possible to quantify signal changes by means of UBPI, a parameter that will improve the possibility of investigating bone structure.


THE SUITABILITY of ultrasound-based investigations for the diagnostic assessment of bone tissue, particularly as an aid in the diagnosis of osteoporosis, has been addressed in multiple studies over the past decade.(1–4) The interest in this technology is partly based on unresolved issues concerning current diagnostic means in osteoporosis but mostly based on its undoubted practical advantages, as compared with conventional X-ray-based methods. It is relatively inexpensive, free of ionizing radiation, and equipment is small in size, permitting easy transport and use in places distant from fully equipped medical centers. Nevertheless, often it has been claimed that bone ultrasound (US) methods are “easy to use,” but such statements are overly optimistic in view of the practical problems encountered in the increasing use of the wide-ranging equipment available today, and the essential need for adequate training of the operators should always be stressed.(5)

One of the initial starting points in the research on ultrasound and bone tissue was that ultrasound possibly could provide more and different information on the physical properties of bone tissue, as compared with dual X-ray absorptiometry (DXA).(6,7) This goal has not been pursued in depth, and recently the ultrasound results have been proposed as “estimated BMD.”(8) Finally, in most instances ultrasound measurement has been indicated as a method to identify subjects that require further investigation by DXA, thereby the possible intrinsic advantage of ultrasound technology risks being largely neglected.

The interaction of ultrasound energy with living bone tissue is extremely complex as density, structure, and elasticity affect ultrasound transmission, that is, velocity, absorption, scattering, and signal characteristics.(9,10) Thus, a more sophisticated approach, including the analysis of the pattern of the transmitted US signal,(11,12) could be fruitfully adopted, as it has been for most of the diagnostic applications of ultrasound technology, for example, the echography. Among the techniques that have focused on signal pattern analysis the bone ultrasound investigation at the phalanx plays a central role.

The phalangeal osteosonogrammetry method was introduced in Europe in 1992–1993 and many studies have suggested its validity in clinical settings. Bone resorption is associated with significant changes both in ultrasound velocity and in the characteristics of the ultrasound signal once it has crossed the phalanx.(13–22)

Phalangeal osteosonogrammetry is computer-assisted and all measurement data are routinely stored in a personal computer (PC). This feature makes it possible to collect large amounts of unprocessed raw data. Thus, we have been able to pool and analyze data collected in over 20 European centers that are involved in the buildup of normative data, the discrimination of subjects with osteoporotic fractures, and the comparison with DXA findings at the spinal and femoral sites. All the centers are characterized by their homogeneity in training and patient recruitment criteria. In this way, a cross-sectional database was created containing the osteosonogrammetry readings of over 10,000 European women of all ages.

This large database would enable the investigation of ultrasound velocity changes with aging and osteoporosis and give us enough statistical power to analyze and quantify the changes occurring to the ultrasound signal in consequence of its interaction with bone tissue. The final aim was to ascertain the diagnostic relevance of the ultrasound parameters to be utilized in the clinical practice.


Patient selection

At all centers the measurements were performed after approval by the local ethical committee and all patients gave their consent to undergo ultrasound examination. For children the consent was obtained from parents. To pool the data, each center was interviewed as to the criteria and procedures for patient recruitment for phalangeal osteosonogrammetry. The center was then asked to supply its original database that also should contain the patient's age, menopausal status, and body mass index. Data only referring to female white subjects of all ages were collected. For adult females alcohol abuse or smoking were exclusion criteria adopted by all centers.

Data sets referring to patients with intercurrent diseases influencing bone tissue homeostasis, such as rheumatoid arthritis, renal impairment, and hyperparathyroidism, were excluded. Patients undergoing treatment with corticosteroids or drugs that could interfere with bone metabolism were excluded, as well as subjects having over 6 months of treatment with osteotrophic drugs.

When possible, we also included spine and hip bone mineral density (BMD) measurements. We collected data from eight different DXA devices, six Hologic (Waltham, MA, U.S.A.) (five quantitative digital radiography [QDR] and one ODX) and two Lunar (Lunar Corp., Madison, WI, U.S.A.); a cross calibration program was not foreseen. Nevertheless, we verified that for each center the mean values of BMD in the adult premenopausal population were not significantly different (data not shown). To be able to compare the results from different devices, we used the American National Health and Nutrition Educational Survey (NHANES) tables to obtain the equivalent T score values from the actual BMD measurements.(23) In one case (ODX), to be able to compare data we used a different algorithm developed in that center.(24)

T score values were used for both quantitative ultrasound (QUS) and DXA measurements; they were defined according to the following formula:

equation image

Subjects were included in a fractured group if the presence of at least one fracture was documented (25% reduction in any height of the vertebra) by lateral X-ray of the thoracolumbar spine. Nevertheless, no central review of the X-rays was performed.

The final database, referring to measurements performed from 1993 to 1998, included 10,450 females, between the ages 0 and 109 years. Before any in-depth analysis, the database was first inspected and subjects were excluded on the basis of the following criteria:

  • Menopausal age greater than chronological age

  • Menopause before 30 years of age

  • Body mass index extremes (<19 and >40), only referring to adult subjects (over 20 years of age)

  • Mistakes in ultrasound measurement, that is, missing data from one or more fingers, and ultrasound speed in bone lower than in soft tissue

Cases excluded (335) were documented and reviewed in cooperation with the supplying center (Table 1).

Table Table 1.. Measurements Excluded
original image

The final validated database was made up of 10,115 subjects. The entire database was then sorted by test date, to ensure randomization. Then the database was blinded as to the source of the data and code, and subjects were identified exclusively by sequential numbering.

Ultrasound measurement

The phalangeal osteosonogrammetry method used in this study (DBM Sonic 1200 IGEA; Carpi, Italy) is based on the transmission of US through the proximal phalanges (digits II-V), probes are applied to the lateral surfaces of the finger. The coupling of the probes with the skin is mediated by gel. The device calculates the amplitude-dependent speed of sound (AD-SoS) and it automatically averages the AD-SoS values of the four fingers.(16)

At each measuring session the reference speed of the patient's soft tissue is measured by applying the probes to the soft tissue area between the base of the thumb and the index finger. For bone tissue measurement probes are positioned at the distal metaphysis of the first phalanges in the proximity of the condyles. For newborn babies and children below 3 years of age the velocity was measured at the distal humerus with probes in the proximity of the condyles, because the diameter (14 mm) of the US probes was too large to allow the measurement at the fingers. The distal humerus is the most accessible and easy to measure with the DBM Sonic 1200.

All osteosonogrammetry data were stored on a PC connected to the device. For most of the patients it was possible to collect the ultrasound graphic trace, that is, the characteristics of the electrical signal generated by the ultrasound at the receiving probe after crossing the phalanx (see Fig. 1). To perform an in depth analysis of the graphic trace we considered the part of the US signal with a speed higher than 1570 m/s (speed in soft tissue), because at a lower velocity the signal processing is not possible because amplitude values saturate. Furthermore, analysis was limited to the data relating to digits II-IV, because the little finger often fails to yield a sufficient signal for analysis.

Figure Figure 1.

Physical parameters that were considered for graphic trace analysis.

Based on previous experiences(18,22,25,26) the following parameters on each graphic trace (Fig. 1) were considered:

  • Fast wave amplitude (FWA; in mV)

  • Dynamics of the ultrasound signal (SDy; mV/μs2)

  • Time interval between the first received signal and the speed value of 1700 m/s (time frame [TF]; in μs)

  • Signal energy (EN) normalized (in mV2μs)

  • Maximum signal amplitude (UPA) in the TF (in mV)

Device calibration and operator training

All devices had been calibrated by the manufacturer using a composite mother phantom. The devices were calibrated for AD-SoS (m/s) and the amplitude of the first peak (FWA in mV). In addition, each center was provided with a phantom to control the ultrasound velocity in Plexiglas on a weekly basis (reference value, 2760 m/s).

All the operators had been trained by the personnel of the manufacturer and the measurement method and as the device calibration were verified twice a year. Whenever a device was returned to IGEA it was always compared with the “mother phantom” to verify the amplitude calibration.


Five centers supplied data sets, which were collected to evaluate the reproducibility of the technology, these were kept separate from the large database. This second database included a total of 29 young adult subjects measured at least three times for intraoperator reproducibility. Interoperator reproducibility involved two operators per center and 14 young adult females. Intraoperator reproducibility studies foresaw the repositioning of the probes. Operators were blinded from previous measurements.

The root mean square (RMS)_CV% was obtained by

equation image

where xi is the mean value obtained for each patient i, m is the number of subjects (i = 1,2,… ., m), and SD is calculated according to the following formula(27):

equation image

Statistical analysis

Statistical methods included Student's t-test and analysis of variance (ANOVA) for the comparison between groups of data. The random distribution of the measurements was guaranteed by the final ordering by test date. When comparing fractured and nonfractured subjects, we created subgroups of the same size and age from the nonfractured population, which was usually much larger than the fractured one. Thus, we considered for each age range (2 years) the number of subjects in the fractured and nonfractured group; these were sorted according to the sequential numbering of the whole database. For each age range we calculated the ratio (n) between nonfractured and fractured subjects. The first subject of each sequence of n subjects was selected in the nonfractured group for each age range.

The linear regression analysis was used to detect the associations between variables yielding correlation coefficients. Multiple logistic regression was used to determine the relative risk of fracture with a CI of 95%, referred either to densitometric or ultrasonic variables and to compare the discriminatory ability of the different methods. Relative risk of fracture was expressed as the odds ratio (OR) per SD decrease of each variable.

The logistic regression analysis also was used to test the diagnostic effectiveness of the investigated US variables and, from these variables, to find an optimized model able to discriminate fractured from nonfractured subjects. The assessment of the optimum combination of variables was made using Wald's test, by selecting the variables that jointly contribute in a significant way to the improvement of the diagnostic effectiveness of the model.(28,29) The variables in the optimized model should then show a significance level of p < 0.05.

Receiver operating characteristic (ROC) analysis was done to assess the discrimination ability of different parameters between fractured and nonfractured subjects by calculating the area under the ROC curve (AUC).

Standardized precision errors (SPEs) also were calculated for QUS parameters using lumbar spine BMD as a reference technique, according to the criteria proposed by Glüer.(30)

All statistical analyses were done with the software Statistical Package for Social Sciences (SPSS, Inc., Chicago, IL, U.S.A.) and the ROC analyses were done with LABROC software (Dept. Of Radiology, University of Chicago, Chicago, IL, U.S.A.).

The World Health Organization (WHO) Study Group on Osteoporosis(31,32) has proposed criteria for the definition of osteoporosis on the basis of a BMD T score threshold level, to identify subjects at elevated risk of fracture. This definition is based on the lower 17% of the overall distribution of postmenopausal (aged over 50 years) readings at one given skeletal site. The same criteria have been applied in this study to find out T score values for QUS parameters, which can identify osteoporotic subjects. Nevertheless, to extend the WHO Study Group criteria to our population, we had to take into account the differences in the distribution of subjects in the age ranges considered, as reported by the WHO Study Group, in which the incidence of osteoporosis was calculated separately for different age ranges in the postmenopausal population.(31)


Characteristics of the devices

At the start of the investigation, the mean speed of ultrasound in the composite mother phantom was 2565 ± 12 m/s and the mean amplitude of the first peak was 2.17 ± 0.17 mV for 26 devices utilized. With the mother phantom we were able to verify 24 out of 26 devices after completion of the ultrasound measurements by the centers and we could confirm that calibration was within the accepted range, that is, 2565 ± 13m/s for AD-SoS and 2.14 ± 0.16mV for amplitude of the first peak. All values were not statistically different from those recorded at the beginning of the ultrasound measurement.

Device calibration was checked routinely and recorded at each center with the Plexiglas phantom. The device calibration was maintained by means of a resident program through a guided calibration procedure. For all devices the average ultrasound speed in the phantom was 2758 ± 8 m/s.

Characteristics of the study group

The overall database was built up with the contributions of 22 centers, using 26 individual DBM Sonic devices for the measurement of a total of 10,450 subjects; of these, 335 (3.2%) measurements were subsequently excluded on account of inconsistencies, as listed in Table 1. The average number of measurements per center was 421 ± 308, the smallest group being represented by the 45 newborns.

The demographics of the population in age decades are shown in Table 2, where the incidence of subjects with radiological evidence of atraumatic osteoporotic fractures is indicated.

Table Table 2.. Distribution According to Age Groups of the Subjects
original image

Out of the 10,115 subjects for whom the average AD-SoS value was available, 7693 (76.1%) also had the graphic trace recorded. In fact, in the early stages (1993-1994) the signal data were not regularly stored. Within the graphic trace group 307 females had a documented osteoporotic fracture.

Spine BMD and hip BMD measurements were available for 1675 and 592 subjects, respectively; all subjects having hip measurement also had spine measurement. DXA results were supplied by eight centers; the average number of measurements per center were 221 ± 206 for spine and 137 ± 101 for hip DXA. Nevertheless, both AD-SoS and graphic trace information was available for 1549 (92%) lumbar spine and 549 (93%) hip BMD.

Nevertheless, we could verify that for both QUS and DXA measurements, there were no statistically significant differences in the values of adult premenopausal women among different centers.

Ultrasound measurements

In the premenopausal adult population the comparison with ANOVA test of AD-SoS values among the centers showed no statistically significant differences (data not shown). Mean speed of sound in soft tissue for the whole population was 1565 ± 13 m/s.

The peak value for AD-SoS was found in the group aged 25–30 years, 2119 ± 68 m/s; this peak value virtually overlaps the one of the device reference curve (2124 ± 70 m/s). The lowest AD-SoS reading of 1594 m/s was observed in a 74-year-old woman having multiple vertebral fractures and a soft tissue speed of 1570 m/s.

Figure 2 shows the average AD-SoS trend in the nonfractured population (9665 subjects). AD-SoS correlation with age was positive from 3 to 30 years of age (r = +0.75). Between 31 and 40 years of age (premenopause), the correlation coefficient was r = −0.16. We then evaluated women after the age of 50 years and in postmenopause (because after the age of 50 years 95% of the whole population was postmenopausal); among these, the correlation with age was r = −0.58. Furthermore, we calculated the annual AD-SoS loss in early and late postmenopause. In the first 5 years after menopause the mean annual AD-SoS loss is 10.4 ± 1.02 m/s, whereas in late menopause (more than 5 years after menopause) the mean annual AD-SoS loss is significantly lower (3.86 ± 0.80 m/s; p < 0.0001).

Figure Figure 2.

AD-SoS trend versus age in nonfractured women. For subjects below 3 years of age data were collected using the humerus as the site of measure (dotted outlined box).

For 7386 nonfractured subjects graphic traces were available, three for each subject (digits II, III, and IV) adding up to a total of over 22,000 graphic traces. We evaluated the trend versus age of the five parameters characterizing the graphic trace. Three of these, EN, TF, and UPA show a trend similar to the AD-SoS, growing from birth to adulthood, constant over adulthood, declining after menopause. Figure 3 shows the trend of EN and TF. On the other side, Fig. 4 shows that SDy and FWA are stable from childhood to adulthood and then they decline significantly after menopause, this effect being more evident for SDy compared with FWA. These findings imply that although young and aged postmenopausal females may have similar AD-SoS values, they should differ for SDy and FWA values.

Figure Figure 3.

Trends of EN and TF versus age in women (mean ± 1 SD). These trends are AD-SoS-like. No data on newborns are shown because it was not possible to analyze US signal.

Figure Figure 4.

Trends of FWA and SDy versus age in women (mean ± 1 SD). These trends are not AD-Sos-like. No data on newborns are shown because it was not possible to analyze US signal.

Fracture discrimination

The population included 450 subjects with radiological evidence of osteoporotic fractures: 414 (92%) vertebral fractures and 36 (8%) hip fractures, average age 68.8 ± 10 and 75.7 ± 17 years, respectively. The mean AD-SoS value was 1860 ± 101 m/s for the vertebral fractures (−3.8 ± 1.4 T score) and 1843 ± 97 m/s for the hip fractures (−4.0 ± 1.4 T score). Considering the low number of hip fractures and the lack of BMD data for these patients, we will focus most of our analysis on vertebral fracture cases.

We then considered only the subjects over 20 years of age. By multiple logistic regression analysis, we calculated the risk of vertebral fracture per SD decrease of AD-SoS, controlled for age; the OR was 1.71 (CI, 1.58-1.84). Within the subjects for whom the graphic trace was available, we compared the whole vertebral fracture group (284 subjects, 71 ± 8.7 years) with a nonfractured age-matched group also made of 284 subjects (70.9 ± 10 years). Their graphic trace data have been utilized to select a combination of parameters, which can discriminate fractured subjects.

The optimum logistic multivariate model derived from this analysis allows us the identification, from the five parameters considered, of the set of three signal parameters in which mathematical combination best discriminated the fractured subjects from controls. This optimum combination was called Ultrasound Bone Profile Index (UBPI), based on the following mathematical equation:

equation image

UBPI thus describes the probability that the tested subject belongs to the nonfractured group, derived from the variables inserted in the equation; then UBPI has been normalized and its values range from 0–1, 1 being attributed to the highest value obtained.(29) The average values of UBPI and AD-SoS in the nonfractured group (0.48 ±0.0.16 and 1926 ± 29, respectively) were significantly higher than in the fractured group: UBPI, 0.35 ± 0.15 (p < 0.0001); and AD-SoS, 1864 ± 98 (p < 0.0001).

The calculation of the UBPI parameter includes the SDy and the FWA values that we showed do not show the same trend versus age observed for AD-SoS. Thus, we should expect that UBPI contains information that is different from AD-SoS. As a confirmation of this Table 3 shows that children and aged women with the same AD-SoS can instead be discriminated by UBPI.

Table Table 3.. Ultrasound Parameters in Young and Adult Females
original image

When UBPI was tested over the whole population we found that among the subjects aged 25–30 years the average UBPI is 0.83 ± 0.14. Figure 5 shows the changes of UBPI with aging. UBPI correlation with subjects aged 3–30 years was r = 0.13. UBPI declined after menopause and its correlation with age was r = −0.57.

Figure Figure 5.

UBPI changes with aging.

We calculated the UBPI CV using the database that included reproducibility measurements at different centers and we found 2.85% CV intraoperator and 2.97% CV interoperator. When the CV was calculated for AD-SoS in the same data set we found 0.75% intraoperator and 0.87% interoperator.

SPE was calculated for AD-SoS and UBPI, using lumbar spine DXA as a reference technique (SPE, 0.90%) with the following results: SPE (AD-SoS vs. DXA) = 1.38%; SPE (UBPI vs. DXA) = 0.89%.

The correlation between AD-SoS and UBPI for subjects over 20 years of age was r = 0.74.

UBPI could be calculated for 307 fractured subjects; UBPI T score value among 284 vertebral fractures was −3.43 ± 1.04 and among 23 hip fractures was −3.28 ± 0.72.

We then tested the efficiency of UBPI by logistic regression using the hip-fractured group (76 ± 15.4 years) and a group of 74 nonfractured females (75.4 ± 14.3 years), yielding an OR of 2.11 (CI, 1.02-4.57) and an AUC of 0.69 ± 0.05. In the same group the OR calculated for AD-SoS was 1.93 (CI, 1.09-3.42) and the AUC was 0.71 ± 0.05.

Osteosonogrammetry and densitometry

We considered 1549 subjects for whom lumbar spine BMD, AD-SoS, and UBPI were available. Among these 549 had hip BMD measured too; Table 4 shows the positive correlation r values among densitometric and osteosonogrammetry measurements, all being highly significant (p < 0.0001).

Table Table 4.. Linear Correlation Between AD-SoS, UBPI, BMD L2-L4, BMD Hip in the Group of 549 Subjects (p < 0.0001)
original image

Table 5 shows the mean values of AD-SoS and UBPI according to BMD T score levels at the spine and at the hip, respectively.

Table Table 5.. Mean Values for US Parameters Grouped According to T Score Values of Spine and Hip
original image

We selected, within the 1433 nonfractured subjects, an age-matched subgroup made up of 114 subjects; Table 6 shows that both US and BMD values remain significantly different between fractured and nonfractured subjects. Our analysis was limited to lumbar spine BMD data because hip measurements were too few in this subgroup.

Table Table 6.. Average Values of US Measurement and Lumbar Spine BMD in Age-Matched Groups
original image

We then calculated the ability to discriminate between fractured subjects of DXA and QUS in the whole population, between young premenopausal women and women with fracture, and finally in age-matched groups; Table 7 reports the AUC values for each group and the methodology.

Table Table 7.. AUC for QUS and Lumbar DXA Parameters for Subjects With and Without Fractures
original image

Threshold levels

To find the AD-SoS T score threshold level the WHO criteria were applied to the 5203 postmenopausal women over 50 years of age; we found a T score cut-off level for AD-SoS of −3.2 corresponding to 1900 m/s. We then studied the distribution of 414 vertebral and of 36 hip-fractured subjects around the T score threshold level. Sixty-six percent of the vertebral fractures and 75% of the hip fractures fell below −3.2 SD. Proceeding in a similar way for UBPI, a T score cut-off level of −3.14 was found, corresponding to a value of 0.38. Again, the distribution of 284 vertebral and 23 hip-fractured subjects was studied; 62% of vertebral fractures and 52% of hip fractures fell below this threshold. These results compare favorably with DXA measurement (1675 measurements available) because, out of 116 subjects with vertebral fractures for whom DXA measurement was available, 63% had a T score value below −2.5 SD.

We then considered the whole adult population (6961 females over 20 years of age) for which both AD-SoS and UBPI were calculated and that included 284 documented vertebral fractures. We evaluated the distribution of subjects below the −3.2 T score level for AD-SoS and −3.14 for UBPI and above the −1 T score level to identify the frequency of fractured subjects in each group. Out of 804 subjects with AD-SoS values below −3.2 T score, 180 (22.4%) were fractured. On the other hand, out of 2546 with AD-SoS T score values > −1 SD, only five (0.2%) were fractured. UBPI had T score values < −3.14 SD in 515 subjects; 169 (32.8%) of these were fractured. Among 3431 subjects with normal UBPI (T score > −1) seven (0.2%) were fractured. Both parameters were below the cut-off values (T score, −3.2 for AD-SoS and 3.14 for UBPI) in 456 subjects; 157 (34.4%) were fractured. Among the 2073 subjects, with both AD-SoS and UBPI T score > −1 SD, only one (0.05%) was fractured.

To compare these results with DXA performances, we used the group of 1549 subjects with lumbar spine BMD; among the 466 subjects with lumbar spine BMD T score < −2.5, 82 (17.5%) were fractured; out of 468 subjects with BMD T score values > −1, six (1.3%) were fractured.

Among 108 subjects that cumulate the contemporary presence of AD-SoS and spine BMD parameters below their respective threshold level, the percent of fractures increases up to 35.3% (Fig. 6). Furthermore, when hip BMD was considered, out of 68 subjects with a T score below −2.5 SD, 30 (44%) were fractured; nevertheless, out of 22 subjects with AD-SoS and both BMD values below the threshold, 14 (63.6%) were fractured. When AD-SoS, UBPI, hip, and lumbar spine BMD were below the threshold level, 68.4% of subjects were fractured. Furthermore, we evaluated the 80 fractured women for whom hip and spine DXA and finger US data were available. Values below the thresholds were found for hip DXA in 30 (38%), for spine DXA in 60 (75%), and for AD-SoS in 46 (58%) of women with fractures. In 75 (94%) of these we found that the test results were below threshold values in at least one of the three skeletal sites investigated.

Figure Figure 6.

Percent of fractured subjects when AD-SoS, spine, and hip BMD T score values are below threshold. *Data referred to a 549 subject population for which hip BMD data were available.


In this study 10,115 phalangeal osteosonogrammetry measurements performed in European women of the white race of all ages were pooled for cross-sectional analysis. This was made possible because the selected test centers were homogeneous with regard to equipment calibration, measurement procedures, and operator training. Moreover, study aims in all centers exclusively addressed the issues of (i) the precision of the method, (ii) the collection of normality data, (iii) tests of the discrimination ability regarding subjects with osteoporotic fractures, and (iv) in some 16% of the adults included in the population, the comparison with densitometric methods.

The speed of ultrasound through the distal metaphyseal portions of the first phalanges of digits II-V was measured over an age range of 3–109 years and for a small group of newborn babies and children less than 3 years of age by measuring the distal humerus. It is still to be seen how this site compares with finger measurement in older groups. Nevertheless, the AD-SoS showed very large dynamic variations over a lifetime, in agreement with previous radiogrammetry data witnessing the largest range of deviation from peak adult bone mass for the finger phalanges.(33–35)

Soft tissue ultrasound speed was homogeneous and stable over a lifetime, thus forming the baseline value for bone ultrasound measurement. Peak adult AD-SoS values were recorded in the 25- to 30-year age group (2119 m/s), which virtually overlaps the age reference value of 2124 m/s supplied with the device. Considering that there should be no AD-SoS values lower than the soft tissue readings, the resulting dynamic range of the phalangeal ultrasound measurements thus is given by the difference between soft tissue baseline speed and the peak adult velocity plus 2 SD (CI, 95%), that is, 695 m/s.

The access to such a large number of measurements and particularly to over 20,000 graphic traces made it possible to investigate, besides the speed parameter AD-SoS, also a series of parameters making up the ultrasound signal trace. The study of the graphic trace is justified both by the empirical observation of the changes occurring to the graphic trace in osteoporotic patients(36) and, most important, by the recent experimental findings showing that graphic trace changes can be extremely specific and informative (Fig. 7).(22,26)

Figure Figure 7.

Examples of US graphic traces of (A) healthy normal child (8 years of age); (B) adult healthy woman (35 years of age); (C) postmenopausal nonfractured woman (67 years of age); (D) postmenopausal osteoporotic woman with vertebral fractures (69 years of age).

Using multiple logistic regression, graphic trace-associated parameters could be combined to develop the UBPI. UBPI, conceived as a “fracture-predictive value,” has shown a good sensitivity and specificity in discriminating hip-fractured from -nonfractured subjects of the same age.

UBPI is calculated on the results of the analysis of the graphic trace of 3 fingers, whereas AD-SoS is calculated in 4 fingers. Here, we decided to continue calculating AD-SoS in this way to allow comparison of our AD-SoS results with those present in the literature, because all are based on 4-finger measurements.

We have shown that this new parameter discriminates between young and adult females having the same AD-SoS values. The increase of AD-SoS with age (r = 0.75) before 30 years reflects the increase in bone mineralization; in fact it has been shown that ultrasound velocity is largely dependent on BMD rather than on bone width.(37) On the other hand, the independence of UBPI from the age before 30 years (r = 0.13) cannot be explained by phalanx thickness, bone marrow distribution (in fact, they change significantly during that period of life (38)), or by the amount of soft tissue because it does not influence the UBPI calculation.

After the adult age the natural occurring decrease in mineral content is associated with changes in structural and elastic properties of the bone. This explains the high correlation between AD-SoS and UBPI (r = 0.74) and the similar correlation level of both parameters to lumbar spine BMD. Before the adult age the correlation between AD-SoS and UBPI is poor (r = 0.46) when low BMD values are not associated to any deficit in structural or elastic properties of bone. Based on all of the above considerations, we hypothesize that UBPI may be related mostly to bone tissue characteristics like elasticity and structure rather than density. Nevertheless, it has been shown in an animal model that the characteristics of the US graphic trace can discriminate between osteomalacia and castration-induced osteoporosis, whereas AD-SoS could not.(12)

UBPI values were found to be reproducible, with a calculated intraoperator precision error of 2.85% in the same population where AD-SoS precision error was calculated as 0.75%.

Among subjects with radiographically documented fractures AD-SoS was found far below young adult reference values and was able to discriminate between groups who fractured and those who did not fracture (p < 0.0001, Table 6). Furthermore, for each SD decrease from adult age the calculated OR was 1.7. When UBPI was tested on hip-fractured patients we calculated an OR of 2.11 per SD decrease. Our findings confirm the results of radiographic absorptiometry showing the sensitivity of the phalanx as a site predictive of fracture risk.(35)

When we analyzed subjects for whom BMD values were available, we found that ultrasound (AD-SoS and UBPI) was able to identify subjects with a BMD T score below −2.5 and those with a BMD T score above the threshold value. Ultrasound and densitometric results were moderately but positively correlated.

Within the group with BMD measurements, when fractured and nonfractured subjects were compared, it was seen that there were no statistically significant differences among the AUC calculated by ROC analyses for any of the diagnostic procedures. Actually, our findings for QUS and DXA yield similar results to those reported by Greenspan when comparing young adults and fractured subjects whereas in this study the performance of both DXA and QUS is slightly better in age-matched groups.(39)

Finally, using the WHO Study Group criteria to identify osteoporotic subjects, cut-off levels of −3.2 T score for AD-SoS and −3.14 T score for UBPI were obtained. The AD-SoS T score threshold translates into a speed of 1900 m/s, which is very close to the cut-off values calculated in several studies done on much smaller populations.(14,15,17) Among those subjects with AD-SoS T score below −3.2 SD, 22% had an osteoporotic fracture, this percentage increased to 35.3% when both AD-SoS and spine BMD T score values were below the calculated thresholds. When hip BMD was considered also the percent of fractures cases increases to 63.6%. These results indicate that having more than one skeletal site with test results (QUS or BMD) below threshold increases the probability of being fractured. On the other hand, measuring different skeletal sites allows the identification of a higher number of fractured subjects; 75 out of 80 (94%) women with vertebral fractures had test results below threshold values in at least one of the three skeletal sites investigated, finger, spine, or hip.

Overall, the results of this large-scale analysis confirm that QUS investigations are indeed useful for the study of bone tissue and confirm the positive clinical experience previously reported in smaller studies. Threshold levels have been calculated for ultrasound parameters (AD-SoS and UBPI) using the criteria developed for DXA; they have been shown to be valuable; nevertheless, their application in clinical practice should always be integrated with all other information, both clinical and instrumental.

The main limitations of the study are (i) it only considers European white women; (ii) vertebral fractures were assessed at each center but no central review was performed but, nevertheless, all centers are experienced in osteoporosis investigation; (iii) the limited number of vertebral fractures representing 4.5% of the population; (iv) the number of hip fractures is low and limited investigation on peripheral fracture has been carried out; (v) the number of DXA measurements is limited to 16% of the population but, nevertheless, subgroup analysis showed similar results to the large total group; (vi) no cross-calibration of DXA devices was performed; (vii) the lack of information about the actual absence of osteoporotic fractures in the nonfractured group, in fact the fracture rate does not follow the expected age-related increase. This, especially when age-matched groups are compared, will have a negative impact on the performances of both QUS and DXA; nevertheless, it actually results in a conservative evaluation of the efficiency of US technology.

Our data indicate that the analysis of multiple osteosonogrammetry parameters, pertaining to the ultrasound signal modifications, will contribute to the improvement of the efficacy of QUS procedures in bone assessment, beyond the established value of speed of sound and ultrasound attenuation. The further study of ultrasound signal features using a signal processing approach successfully used for many years in echography may also in the near future provide new ways of looking at diverse disorders affecting bone tissue metabolism.


We thank Richard Eastell and Bridget Ingle for their determinant contribution to the redaction of this paper.