Effect of Precision Error on T-scores and the Diagnostic Classification of Bone Status

Transcription

1 Journal of Clinical Densitometry, vol. 10, no. 3, 239e243, 2007 Ó Copyright 2007 by The International Society for Clinical Densitometry /07/10:239e243/$32.00 DOI: /j.jocd Original Article Effect of Precision Error on T-scores and the Diagnostic Classification of Bone Status Gary M. Kiebzak,*,1 Kenneth G. Faulkner, 2 Wynn Wacker, 3 Ronald Hamdy, 4 Edith Seier, 5 and Nelson B. Watts 6 1 Center for Orthopaedic Research and Education, St. Luke s Belmont Center, Houston, TX; 2 Synarc, San Francisco, CA; 3 GE Healthcare, Madison, WI; 4 Osteoporosis Center, College of Medicine, East Tennessee State University and VAMC, Johnson City, TN; 5 Department of Mathematics, East Tennessee State University, Johnson City, TN; and 6 University of Cincinnati College of Medicine, Cincinnati, OH Abstract We quantified confidence intervals (CIs) for T-scores for the lumbar spine and hip and determined the practical effect (impact on diagnosis) of variability around the T-score cutpoint of 2.5. Using precision data from the literature for GE Lunar Prodigy dual-energy X-ray absorptiometry (DXA) systems, the 95% CI for the T-score was G0.23 at the lumbar spine (L1eL4), G 0.20 at the total hip, and G0.41 at the femoral neck. Thus, T-score variations of G0.23 or less at the spine, G0.20 at the total hip, and G0.41 at the femoral neck are not statistically significant. When diagnosing osteoporosis, T-scores in the interval 2.3 to 2.7 for spine or total hip (after rounding to conform to guidelines from the International Society for Clinical Densitometry) and 2.1 to 2.9 for femoral neck are not statistically different from 2.5. Better precision values resulted in smaller 95% CIs. This concept was applied to actual clinical data using Hologic DXA systems. The study cohort comprised 2388 white women with either normal or osteopenic spines in whom the densitometric diagnosis of osteoporosis would be determined by hip T- scores. When evaluating actual patient T-scores in the range 2.5 G 95% CI, we found that the diagnosis was indeterminate in approximately 12% of women when T-scores for femoral neck were used and in 4% of women when T-scores for total hip were used, with uncertainty as to whether the classification was osteopenia or osteoporosis. We conclude that precision influences the variability around T-scores and that this variability affects the reliability of diagnostic classification. Key Words: Bone mineral density; dual-energy X-ray absorptiometry; DXA; precision error; T-score. Introduction The least significant change (LSC) value for bone mineral density (BMD) accounts for measurement variability (precision) and determines when change in BMD between 2 measurements is statistically significant (1e3). Although not commonly taken into account, precision also influences Received 01/29/07; Revised 03/07/07; Accepted 03/10/07. *Address correspondence to: Gary M. Kiebzak, PhD, Center for Orthopaedic Research and Education, St. Luke s Belmont Center, 2909 W. Holcombe Blvd, Houston, TX gkiebzak@ sleh.com a single-point BMD measurement, resulting in statistical uncertainty in the BMD and therefore uncertainty regarding the T-score (4). This variability around a T-score introduces uncertainty of diagnosis when evaluating bone status for patients with T-scores near the transition threshold between normal and osteopenia (T-score 5 1) and between osteopenia and osteoporosis (T-score 5 2.5) (5). According to basic statistics, there is a range of T-scores above and below 2.5 within which T-scores are not significantly different from 2.5 and from each other. That is, confidence intervals (CIs) around the T-score of 2.5 exist, based on the precision of the measurement, which define a range within which T-scores are statistically equivalent. Within 239

2 240 Kiebzak et al. that range, T-scores are indeterminate with respect to diagnostic classification; i.e., diagnosis can be statistically uncertain using a T-score from a single dual-energy X-ray absorptiometry (DXA) scan. The primary objective of this study was to define the 95% CI around the diagnostic T-score threshold of 2.5 and to show how precision error can affect the size of the 95% CI. The secondary objective was to estimate the percentage of patients who may have an uncertain diagnosis (either osteopenic or osteoporotic) as a result of the indeterminate range of T-scores defined by the 95% CI. Methods The description of methods and the results each consist of 2 sections. The first section describes the calculation of the 95% CI for a T-score and demonstrates the effect of precision error using actual data collected in the field from GE DXA systems. The second section applies the concept to a cohort of women who were scanned on a Hologic DXA system to show the percentage of patients who would have an indeterminate diagnosis based on hip T-scores. Use of both GE and Hologic data demonstrates the generalizability of the concept of a 95% CI for a T-score. The Concept of Variability Around a T-score: Defining the 95% CI for a T-score The 95% CI for a T-score depends on both accurate and precise measurements. For a properly calibrated DXA system using an accepted reference population, accuracy errors will not be significant, and the CI for the T-score will be solely a function of the precision error. Assuming BMD is normally distributed, the 95% CI for a single T-score measurement can be estimated from the properties of the normal distribution (6). The 95% CI of a single T-score assessment is determined from the equation: 95% CI 5 T-score ð1:96þðprecision errorþ Population SD where Precision 5 the root mean square standard deviation (SD) for a set of repeat measures. This value is obtained by a precision assessment, which must be done at every DXA center. Population SD 5 the SD of the manufacturer s reference population used to calculate the T-score. The 1.96 coefficient in the equation represents the 95% range in the normal distribution necessary to compute the 95% CI. Reported precision errors for expert DXA centers (centers that are research oriented and conduct clinical trials, etc.) and clinical DXA centers (such as small private practice centers) were used (Table 1) (7). Using the GE Lunar Prodigy DXA system (GE Healthcare, Madison, WI), the expert precision error (g/cm 2 ) in women (age, 63 G 9 yr, mean G SD) was for the L1eL4 spine, for the femoral neck, Table 1 Expert and Clinical Precision Errors (g/cm 2 ) With the GE Lunar Prodigy Expert Clinical Note: See Ref. (7). and for the total hip region. At clinical centers using the GE Lunar Prodigy, precision error in women (age, yr) was reported as at the L1eL4 spine, at the femoral neck, and at the total hip. Table 2 shows the population SDs used in the GE Lunar EnCore software (8). Application of Concept to a Cohort of Patients The center s measurement precision using a Hologic Delphi DXA system (software version 11.1; Waltham, MA) was determined using triplicate measurements in 15 patients with repositioning between scans. Patients used in the center s precision assessment were representative of the center s patient population. The precision (g/cm 2 ) was for the femoral neck and for the total hip. The 95% CIs for femoral neck and total hip were calculated as above using our center s precision data and the manufacturer s reference population SDs, which are g/cm 2 for femoral neck and g/ cm 2 for total hip(8). We retrospectively reviewed DXA data collected from 1997 to Data were evaluated for all Caucasian women O50 yr of age who had lumbar spine and bilateral hip scans performed in the same scanning session. Women were excluded if BMD was affected by documented pathology or technical issues such as positioning of the proximal femur for scanning. A total of 3012 patients were identified. The mean age was 64 yr, range, 50e92 yr. T-scores were calculated for lumbar spine L1eL4, right and left total hip, and femoral neck from the standard Hologic reference database (manufacturer-specific spine database and NHANES III for hip). Bone status was determined from the lowest T-score, and patients were classified as normal, osteopenic, or osteoporotic using International Society for Clinical Densitometry (ISCD) and World Health Organization (WHO) criteria (9,10). Table 2 Population Standard Deviations (SDs) Used to Calculate T-scores for GE Lunar Prodigy L1eL4 spine Femoral neck Total hip Population SD (g/cm 2 ) Note: See Ref. (8).

3 Confidence Limits for T-Scores 241 From this cohort, we identified women with normal and osteopenic spines for whom a diagnosis of osteoporosis would be based on T-scores from the hip scan. There were 1229 women with normal spines and 1159 women with osteopenic spines. We determined the percentage of women with total hip and femoral neck T-scores within the range defined by the 95% CI around 2.5 (after rounding to conform to the guidelines of the ISCD) (11), namely, 2.3 to 2.7 for total hip and 2.2 to 2.8 for femoral neck for this particular DXA system. Results Defining 95% CIs for T-scores Using representative precision values for Prodigy DXA systems from clinical centers, the 95% CI for T-scores at each skeletal site were G0.23 at the lumbar spine, G0.20 at the total hip, and 0.41 at the femoral neck (Table 3). Around the T-score of 2.5, these 95% CIs defined ranges of 2.27 to 2.73 for the spine, 2.30 to 2.70 for the total hip, and 2.09 to 2.91 for the femoral neck, respectively. By statistical standards, T-scores within each range would not be significantly different from each other, and not different from 2.5. The influence of precision values on the calculated 95% CIs and T-score ranges around 2.5 are summarized in Tables 3 and 4 and Fig. 1, showing that the better the precision, the smaller the indeterminate range around 2.5. Application of Concept to Study Cohort Table 5 summarizes the percentage of women with T- scores in the range defined by 2.5 G 95% CI. The larger the 95% CI, the larger the indeterminate range, as seen by comparing the percentage of women identified using femoral neck vs total hip T-scores. Compared with women with normal spines, more women with osteopenic spines had hip T-scores in the indeterminate range. Discussion This study illustrated the concept that variation exists around a single T-score, just as it exists for any single-point measurement. The degree of variation is predominantly determined by a center s measurement precision; the better the precision, the smaller the indeterminate range of T-scores Table 3 Average T-score 95% Confidence Intervals for Expert and Clinical Centers Using GE Lunar Prodigy Dual-Energy X-Ray Absorptiometry Systems Expert Clinical Table 4 Indeterminate T-Score Range Around T-score 2.5 Defined by 95% Confidence Intervals for Expert and Clinical Centers Using GE Lunar Prodigy Dual-Energy X-Ray Absorptiometry Systems Expert Clinical 2.34 to to to to to to 2.70 Note: In clinical practice, T-scores would normally be rounded to one decimal place per International Society for Clinical Densitometry guidelines (11). and vice versa. When applied to the diagnostic thresholds T-score and 2.5, there will be patients whose diagnostic classification will be uncertain (indeterminate) based on a single DXA scan. In such cases, clinical judgment will be necessary with respect to management decisions. Although not commonly taken into account, this concept should be considered in clinical decision making. T-scores have been adopted by the field of densitometry in an effort to standardize reporting of DXA scan results, as absolute BMD values differ substantially at different skeletal sites and between DXA systems from different manufacturers (12). The WHO subsequently used T-scores to designate a densitometric definition of osteoporosis as a T-score of 2.5. However, despite being originally intended as a population definition, the WHO definition of osteoporosis has become a diagnostic standard for individuals as well. Having a threshold definition (i.e., a T-score defining a category) creates the problem of interpretation of values close to the threshold. Some physicians may adhere to strict definitions and not classify a patient as osteoporotic until the lowest T- score is 2.5. In addition, strict interpretation of threshold guidelines may restrict commencement of treatment until specific T-score thresholds have been reached. T-score Accuracy Limit (95% CI) Expert Sites Clinical Sites L1-L4 Spine Femoral Neck Total femur Fig. 1. Average T-score accuracy errors (95% confidence intervals) for expert and clinical centers. See Ref. (7).

4 242 Kiebzak et al. Table 5 Percent of Patients Who Could Not Be Definitively Diagnosed as Osteoporotic Due to Indeterminate T-Scores a Spine status Normal (n ) Osteopenic (n ) All patients (n ) Left total hip Percent of patients for each skeletal site Right total hip Left FN Right FN Abbr: FN, femoral neck. Notes: Patients had both left and right hips scanned in the same session. If the spine was normal or osteopenic based on L1eL4 T- score, then the densitometric diagnosis of osteoporosis would be determined by the status of the hips if a hip T-score was 2.5. T-scores were rounded to one decimal place per International Society for Clinical Densitometry guidelines (11). a Indeterminate T-score ranges are defined as 2.5 G 95% confidence interval: total hip, 2.3 to 2.7; FN, 2.2 to 2.8. Our results suggest that when managing an individual patient, clinicians should not become overly focused on either a single T-score result or a fixed threshold defining categories of bone status. In bone densitometry, rigid adherence to thresholds could negatively impact patient management. For example, suppose a center s 95% CI for T-scores at femoral neck is G0.4 with an indeterminate range of T-scores of 2.1 to 2.9. If an untreated postmenopausal woman has a femoral neck T-score of 2.3, then there is little justification to withhold treatment until lower T-scores are obtained because 2.3 is not significantly different from 2.5. All of the patients in our clinical cohort who we considered to have indeterminate T-scores had T-scores! 2.0, which is the threshold suggested by the National Osteoporosis Foundation for initiating treatment. Thus, these patients could have commenced treatment without requiring a T-score of 2.5 or lower. Finally, although we have not provided examples in this article, the concept of variation around T-scores should also be applied to the WHO definition of osteopenia at a T-score threshold of 1.0. Recognition of variation around a single T-score raises 2 important issues. First, densitometrists must realize that the skeletal site with the worst precision at their center will also have the greatest diagnostic uncertainty. For many centers, femoral neck precision is not as good as total hip precision (13). This is somewhat problematic in that femoral neck has been designated the preferred site to define osteoporosis (14). Second, consideration should be given to the benefit of duplicate scans and routine bilateral hip scans as a method to help resolve classification uncertainties when a T-score from a single scan is in the indeterminate range (5,15). Increasing the number of scans and taking the average value will narrow the 95% CI and improve the measurement confidence (1). In summary, T-scores are subject to measurement variation just like any other test. For clinical DXA centers, the 95% CI can be reduced by reducing the precision error to levels that have been demonstrated at expert DXA centers. Our results underscore the need for each DXA center to determine its own precision error. Not only is knowledge of precision important for determining the LSC, which is used to monitor change, but as we have shown, precision is also important for determining the certainty of a diagnosis. That is, if a patient s T-score is within the indeterminate range of values around a diagnostic threshold (T-score 5 1 or 2.5), then it may not be possible to definitively classify bone status as normal, osteopenia, or osteoporosis. Finally, in any case, a diagnosis of osteoporosis and fracture risk assessment should not rely entirely on a single T-score: it is important to take into account patient factors such as age and history of atraumatic fracture. References 1. Bonnick SL, Johnston CC Jr, Kleerekoper M, Lindsay R, Miller P, Sherwood L, Siris E Importance of precision in bone density measurements. J Clin Densitom 4(2):105e Gluer CC, Blake G, Lu Y, Blunt BA, Jergas M, Genent HK Accurate assessment of precision errors: how to measure the reproducibility of bone densitometry techniques. Osteoporos Int 5(4):262e Lenchik L, Kiebzak GM, Blunt BA What is the role of serial bone mineral density measurements in patient management? J Clin Densitom 5(Suppl):S29eS Faulkner KG, Weynand LS, Wacker WK, Barden HS Determination of T-score confidence limits for diagnosing osteoporosis. J Clin Densitom 9(2):237. [abstract]. 5. Kiebzak GM, Hamdy RC, Seier E, Watts NB Variability around the T-score cutpoint of -2.5 impacts diagnostic classification. J Clin Densitom 8(2):238. [abstract]. 6. Zar JH Biostatistical analysis. 2nd ed. Prentice Hall, Englewood Cliffs, NJ. 7. Weynand LS, Barden HS, Faulkner KG Precision of fanbeam densitometers in clinical practice. J Bone Miner Res 19(Suppl 1):S235. [abstract]. 8. Favus MJ, ed. Primer on the metabolic bone diseases and disorders of mineral metabolism. 4th ed. Lippincott Williams & Wilkins, Philadelphia, PA. Appendix vi. pp WHO Assessment of fracture risk and its application to screening for postmenopausal osteoporosis WHO Technical Report Series Hamdy RC, Petak SM, Lenchik L Which central dual x-ray absorptiometry skeletal sites and regions of interest should be used to determine the diagnosis of osteoporosis? J Clin Densitom 5(Suppl):S11eS The Writing Group for the ISCD Position Development Conference 2004 Nomenclature and decimal places in bone densitometry. J Clin Densitom 7(1):45e49.

5 Confidence Limits for T-Scores Faulkner KG The tale of the T-score: review and perspective. Osteoporos Int 16:347e Shepherd JA, Lu Y, Wilson K, et al Cross-calibration and minimum precision standards for dual-energy x-ray absorptiometry: the 2005 ISCD Official Positions. J Clin Densitom 9(1):31e Hans D, Downs RW, Duboeuf F, Greenspan S, Jankowski LG, Kiebzak GM, Petak SM Skeletal sites for osteoporosis diagnosis: the 2005 ISCD Official Positions. J Clin Densitom 9(1):15e Hamdy R, Kiebzak GM, Seier E, Watts NB The prevalence of significant left-right differences in hip bone mineral density. Osteoporos Int 17:1772e1780.