Revisiting the ovarian volume as a diagnostic criterion for polycystic ovaries

Human Reproduction Vol.20, No.10 pp. 2893 2898, 2005 Advance Access publication July 8, 2005. doi:10.1093/humrep/dei159 Revisiting the ovarian volume as a diagnostic criterion for polycystic ovaries Sophie Jonard 1, Yann Robert 2 and Didier Dewailly 1,3 1 Department of Endocrine Gynaecology and Reproductive Medicine, and 2 Department of Radiology, Hôpital Jeanne de Flandre, CHRU, 59037 Lille, France 3 To whom correspondence should be addressed. E-mail: ddewailly@chru-lille.fr BACKGROUND: This study revisited the ovarian volume (OV) as a diagnostic criterion for polycystic ovaries (PCO). Indeed, a threshold of 10 cm 3 for the OV, chosen at the polycystic ovary syndrome (PCOS) international consensus held at Rotterdam in 2003, was to date not based on appropriate studies such as receiver operator characteristic (ROC) curve analysis. METHODS: This prospective study included 154 women with PCOS, selected by using the former National Institutes of Health criteria, who were compared with 57 women with normal ovarian function. Ultrasound examination was performed between cycle days 2 and 7 with a 7 MHz transvaginal transducer. RESULTS: Mean OV, ovarian area (OA) and follicle number (FN) values were significantly higher in the PCOS group than in controls. The area under the ROC curve (AUC) was >0.9 for all three criteria, indicating a satisfactory diagnostic potency for each. Concerning the OV, setting the threshold at 7 cm 3 offered the best compromise between specificity (91.2%) and sensitivity (67.5%). In comparison, specificity and sensitivity were 98.2 and 45%, respectively, with a threshold at 10 cm 3. Nevertheless, the highest AUC was obtained for FN (0.956) and then for OA (0.941). CONCLUSIONS: OV is a good diagnostic criterion for PCO but, on the basis of the present data, we propose to lower its threshold to 7 cm 3. The FN >12 still appears as the best diagnostic criterion. The OA could be used as a surrogate for OV in difficult situations. Key words: follicle number/ovarian area/ovarian volume/polycystic ovaries Introduction Since the polycystic ovary syndrome (PCOS) international consensus held at Rotterdam in 2003, ultrasound criteria are now included in the definition of this disease (Balen et al., 2003; Rotterdam ESHRE/ASRM-sponsored PCOS consensus workshop group, 2004). These criteria include the ovarian volume (OV) and the follicle number (FN), which have been considered the key features of polycystic ovaries (PCO) for >15 years (Adams et al., 1985). They are combined according to the following definition: either 12 or more follicles measuring 2 9 mm in diameter or increased ovarian volume (>10 cm 3 ). This choice was based on a literature review and on a discussion between experts. However, data from the literature that were available at the time of the consensus conference were stringent to various degrees. Only appropriate procedures such as the receiver operator characteristic (ROC) curve analysis (Zweig and Campbell, 1993) allow setting the cut-off value of a given parameter in order to obtain the best compromise between specificity and sensitivity. At the time of the Rotterdam conference, such analysis was available for the FN exclusively, in only one study performed by ourselves (Jonard et al., 2003). This study showed that a threshold set at 12 follicles per ovary offered the best compromise between sensitivity (75%) and specificity (98%) to distinguish PCO from normal ovaries. On the other hand, the choice of the threshold of 10 cm 3 for the OV was more empirical. It was based on the synthesis of results from many studies in which the upper normal threshold was simply defined as being either the maximal value of controls or the 95th percentile of the control range (see Table I). Although this guarantees a specificity near to 100%, it does not necessarily offer the best sensitivity. These studies have reported a mean OV >10 cm 3 for normal ovaries, but some found higher values (Adams et al., 1985; Fulghesu et al., 2001; see Table I). Furthermore, in studies reporting on large-scale ultrasonography in general populations of premenopausal women, the highest normal OV values span up to 18 (Van Nagell et al., 1990) or even 20 cm 3 (Pavlik et al., 2000). However, in some studies, the upper limit was defined as the mean + 2SD, without logarithmic transformation of data (Pavlik et al., 2000; Fulghesu et al., 2001). This yields an overestimation when the parameter is not normally distributed, as is the case for the OV. Therefore, beside the issue of diagnosing PCO, the upper normal limit of the OV still suffers from some variability in the literature. To clarify these issues, we decided to revisit the OV as a criterion for PCO. To do so, we prospectively collected ovarian The Author 2005. Published by Oxford University Press on behalf of the European Society of Human Reproduction and Embryology. All rights reserved. 2893 For Permissions, please email: journals.permissions@oupjournals.org

S.Jonard, Y.Robert and D.Dewailly Table I. Results of some ultrasound studies described in the literature concerning the ovarian volume Reference Ultrasound examination route Volume threshold indicative of PCOS Percentage of patients with clinical PCOS having the volume criterion Percentage of controls having the volume criterion No. of studied patients No. of studied controls Adams et al. (1985) TA >15 cm 3 33 0 76 17 Yeh et al. (1987) TA >10 cm 3 70 0 108 25 3 Pache et al. (1992) TV >8 cm 70 0 52 29 Van Santbrink et al. (1997) TV >10.7 cm 3 41 0 330 48 3 Atiomo et al. (2000) TV >9 cm 70 45 32 40 Fulghesu et al. (2001) TV >13.2 cm 3 21 5 53 30 TA = transabdominal; TV = transvaginal. measures in a new series of patients and controls who were selected without using ultasonographic criteria. We applied the ROC procedure to the data in order: (i) to check whether 10 cm 3 is the best threshold for OV; and (ii) to challenge the diagnostic potency of OV against that of FN and of ovarian area (OA), OA being the criterion that we used instead of OV to define PCO in all our previous studies, prior to the Rotterdam consensus (Jonard et al., 2005). Materials and methods Patient and control populations This study was approved by the Institutional Review Board of the Lille University Hospital. Informed consent was obtained from all patients and controls before their inclusion in the study. Controls. The control population consisted of 57 women with normal ovarian function. These women were referred to our Department for IVF because of tubal and/or male infertility. Exclusion criteria included a history of menstrual disturbances (i.e. cycle length either <25 days or >35 days), hirsutism [as assessed by the modified Ferriman and Gallway (F G) score >6], serum level of prolactin >20 mg/ml, serum testosterone higher than our previously reported upper normal threshold, i.e. 0.5 ng/ml (Pigny et al., 1997), and hormonal treatment during the 3 months prior to the study. Patients with PCOS. From 154 patients consecutively recruited by our Department and defined as having PCOS according to the Rotterdam criteria (Rotterdam ESHRE/ASRM-sponsored PCOS consensus workshop group, 2004), 98 were selected by using the National Institutes of Health (NIH) criteria (Zawadski and Dunaif, 1992), i.e. association of an oligo-anovulation (our definition of oligo-anovulation differs slightly from the NIH definition, i.e. oligomenorrhoea or amenorrhoea or cycle length either <25 days or >35 days and/or ovulatory disturbances as assessed by basal body temperature chart and/or serum progesterone level <3 ng/ml in the luteal phase at day 24 in the cycle), and clinical and/or biochemical signs of hyperandrogenism (F G score >6 and/or testosterone level >0.5 ng/ml). The diagnosis of PCOS was retained after exclusion of other aetiologies (congenital 2894 adrenal hyperplasia, androgen-secreting tumour or Cushing s syndrome). We used this former definition of PCOS in order not to use the OV, OA or FN as selection criteria, since our goal was to study their diagnostic potency, free from any inclusion bias. Serum sampling Blood sampling was performed during the early follicular phase (i.e. between days 2 and 7 after the last menstrual period) in both PCOS patients and control women, as described previously (Pigny et al., 1997). In PCOS patients, the last menstrual period was either spontaneous or induced by the administration of didrogesterone (10 mg/day during 7 days). Pelvic ultrasound examination We carefully applied the guidelines recommended by the Rotterdam consensus conference (Rotterdam ESHRE/ ASRM-sponsored PCOS consensus workshop group, 2004). Ultrasound examination was performed between cycle days 2 and 7 with a 7 MHz transvaginal transducer (Logic 400 General Electric, Milwaukee, USA). Ultrasound measurements were taken in real time, according to a standardized protocol. The highest possible magnification was used to examine the ovaries. After the longest medial axis of the ovary was determined, the length and thickness were measured and the OA was calculated using a manual or automatic ellipse to outline the ovary as described previously (Dewailly et al., 2002). The OV was estimated after the measurement of the length, width and the thickness and use of the classical formula for a prolate ellipsoid: L W T 0.523 (Sample et al., 1977; Adams et al., 1985; Orsini et al., 1985). The FN was established as described previously (Jonard et al., 2003) by scanning each ovary from the inner margin to the outer margin in longitudinal cross-section. All follicles of <9 mm, but >2 mm, were counted. Follicle diameter was the mean of two recorded diameters in the longitudinal and antero-posterior planes. Patients in whom transvaginal ultrasonography was inappropriate (virgins or patients who refused) were excluded from the analysis, as well as those in whom no follicle was seen in either the right or the left ovary and/or in whom the ovarian area was below the lower normal limit, i.e. 2.5 cm 2. Patients with at least one follicle >9 mm in diameter at ultrasonography, or a serum estradiol level >80 pg/ml, were also excluded from the study so as not to confound the data with the presence of a dominant follicle or a corpus luteum.

Ovarian volume and polycystic ovaries Statistical methods For the OA, OV and the FN, the data used for statistical analysis were the mean of recorded values for the left and right ovaries. Statistical significance between mean values was attributed to two-tailed P < 0.05. ROC curves (Zweig and Campbell, 1993) were constructed to examine the diagnostic test performance, i.e. the ability to discriminate between controls and patients with PCOS. Sensitivity against (1 specificity) was plotted at each level, and the area under the curve (AUC) was computed by the non-parametric Wilcoxon statistical test (Zweig and Campbell, 1993). The AUC represents the probability of correctly identifying controls and patients with PCOS. A value of 0.5 means that the test result is no better than chance. The statistical analysis was performed using Statview 4.5 (Abacus Concepts Inc., Berkeley, CA) and SPSS 11.5 (SPSS Inc., Chicago, IL) for the ROC analysis. Results Table II shows the main clinical, hormonal and ultrasound features of patients and controls. Analysis of variance indicated no significant effect of the day of sampling and ultrasound examination (from 2 to 7 days after the last menstrual) on the OA, OV or FN values (P = 0.97, P = 0.86 and P = 0.82, respectively). Mean OA, OV and FN values were significantly higher in the PCOS group than in controls (Table II). Figure 1A and B shows that the 10th 90th percentile ranges of OA, OV and FN for women with PCOS overlapped those of controls. The highest individual values in controls were 5.7 cm 2, 10.3 cm 3 and 14, respectively. Diagnostic potency of OV The AUC curve for OV (see Materials and methods) was 0.905 (Figure 2), indicating that the tested variable had a satisfactory diagnostic potency. The ROC analysis also allowed estimation of the sensitivity and specificity for a given threshold (Table 3). A threshold at 10 cm 3 offered a good specificity (98.2%) but a weak sensitivity (45%). Setting the threshold at 7 cm 3 offered the best compromise between specificity (91.2%) and sensitivity (67.5%) (Table III). 25 22,5 20 17,5 15 12,5 10 7,5 5 2,5 0 60 50 40 30 20 10 0 PCO CONTROLS Mean Ov Area (cm2) PCO CONTROLS 2-9mm FN Mean Ov Vol (ml) Figure 1. Box-and-whisker plots showing the values of (A) OA and OV, and (B) FN. Whiskers represent the 10 90th percentile ranges, and the boxes indicate the 25 75th percentile ranges. The horizontal line in each box corresponds to the median. A B Table II. Main clinical and hormonal features in controls and in patients with PCOS PCOS (n = 98) Controls (n = 57) P a Age (years) 27.2 (19.5 33.0) 29.0 (24.5 35.0) NS Body mass index 27.9(20.1 40.8) 22.9 (19.0 31.5) <0.002 (kg/m 2 ) Testosterone 0.63 (0.37 0.94) 0.27 (0.13 0.42) <0.0001 (ng/ml) Mean OA (cm 2 ) 6.25 (4.50 8.65) 3.90 (2.90 4.80) <0.0001 Mean OV (cm 3 ) 9.48 (5.69 16.78) 4.75 (3.11 6.86) <0.0001 2 9 mm FN 16.75 (10.0 30.0) 6.5 (4.5 10.5) <0.0001 Values are expressed as medians with 10 90th percentile range in parentheses. a P level by Student t test; NS = non-significant. Diagnosis potency of FN The AUC curve for the FN (0.956) (Figure 2) indicated a better diagnostic performance than for the OV. The present data on this new series confirm that a threshold set at 12 for the FN offers the best compromise between specificity (97%) and sensitivity (79%) (Table III). Diagnosis potency of OA The AUC curve for the OA yielded an intermediate value (0.941) (Figure 2), between those for OV and FN. The best compromise between specificity (77.6%) and sensitivity (94.7%) was obtained with a threshold set at 5.0 cm 2 (Table III). 2895

S.Jonard, Y.Robert and D.Dewailly ROC curve for OV Table III. ROC curve data Area under the ROC curve Threshold Sensitivity Specificity Figure 2. ROC curves for OV, OA and FN. Discussion In this study based on a new series of controls and patients with PCOS, we confirm that an increased OV is a reliable ultrasound criterion for the diagnosis of PCOS, as proposed by the Rotterdam consensus conference (Rotterdam ESHRE/ASRMsponsored PCOS consensus workshop group, 2004). However, our results indicated that the 10 cm 3 threshold proposed by this conference yielded an almost 100% specificity but a sensitivity <50%. This was expected since this cut-off value was far above the 90th percentile of our control range. We obtained better results when the threshold was lowered to 7 cm 3, but it must be stressed that gaining an acceptable sensitivity for this parameter (i.e. close to >70%) brings it close to unacceptable specificity (i.e. close to <90%). 2896 Sensitivity Sensitivity Sensitivity 1 - Specificity ROC curve for OA 1 - Specificity ROC curve for FN 1 - Specificity OA (cm 2 ) 0.941 4 95.9% 55.4% 4.5 90% 75% 5.0 77.6% 94.7% 5.5 67% 100% 6 53.1% 100% OV (cm 3 ) 0.905 6 89.8% 77.2% 7 67.5% 91.2% 8 63.3% 94.7% 9 53.1% 98.2% 10 45.9% 98.2% 2 9 mm FN 0.956 10 91.5% 85% 11 89.4% 90.2% 12 79% 97% 13 76.3% 100% 14 68.9% 100% 15 65% 100% The upper normal threshold proposed here for OV is much lower than reported by some authors (Adams et al., 1985; Yeh et al., 1987; Van Santbrink et al., 1997; Atiomo et al., 2000; Fulghesu et al., 2001; see Table I) but it agrees with others (Pache et al., 1992). Indeed, there is still a lot of controversy in the literature about the normal upper limit of OV, which varies from 8 to 15 cm 3, and even more (Adams et al., 1985; Yeh et al., 1987; Pache et al., 1992; Van Santbrink et al., 1997; Atiomo et al., 2000; Fulghesu et al., 2001; see Table I). The main reason for this appears to be the lack of standardization of the ultrasound procedure. In many studies, OV was measured at random during the menstrual cycle (van Nagell et al., 1990; DiSantis et al.,1993; Atiomo et al., 2000; Pavlik et al., 2000). The mean OV can therefore have been overestimated in some patients because of the presence of growing follicles >10 mm or a corpus luteum (Granberg et al., 1990). Indeed, Christensen et al. (1997) showed by ultrasonography that OV changes during the normal menstrual cycle, with the lowest values in the early follicular phase and the highest values in the luteal phase. When data were restricted to the very early follicular phase of natural cycles, mean normal OV was only 4.5 cm 3. Therefore, setting the normal upper limit for OV requires strict adhesion to the Rotterdam consensus conference guidelines (Rotterdam ESHRE/ASRM-sponsored PCOS consensus workshop group, 2004), which recommend scanning cycling women in the early follicular phase for the correct measure of OV and FN. Also, OV diminishes with age from 25 years onwards (Pavlik et al., 2000). Consequently, any study reporting on OV should use a control group that is age-matched to the patient population. Lastly, in some studies, the control group may have included non-symptomatic patients with PCO, and this may have skewed the distribution of OV to excessive highest values (Christensen et al., 1997). The substantial intra- and inter-observer variability observed with OV measurement (Balen et al.,2003) is another reason explaining the large differences between the upper limits of the normal OV from one study to the other. This is due to the technical difficulty of obtaining strictly longitudinal ovarian cuts,

Ovarian volume and polycystic ovaries which is an absolute condition for accurate measures of the ovarian axis in each of the three planes (length, width and thickness). This is particularly true in cases of non-ellipsoid ovaries, and it must be recognized that any assimilation of OV to a sphere or prolate ellipse is, at best, an estimate because of the irregular shape of the ovary. This explains why OA yielded better results than OV in the present study. Indeed, measuring the OA is less operator dependent, providing the examination is performed on a strictly longitudinal plane and outlining of the ovary is done by hand with automatic calculation of the outlined area. This last technique must be preferred to fitting an ellipse to the ovary when the area is given by the machine, since PCO are frequently non-ellipsoid (Jonard et al., 2005). In support of the good reproducibility of OA measurement, we found exactly the same figure as the one that we reported 10 years ago with an OA threshold empirically set at 5.5 cm 2, i.e. 100% specificity and 66% sensitivity (Robert et al., 1995). However, this threshold was a little too high as indicated by the ROC data in the present study where the optimal cut-off value for OA was 5.0 cm 2. It must be stressed that in all studies, including the present one, the thresholds for ultrasound measures have been established by comparing normal women with patients with PCOS. All have omitted the difficult issue of multifollicular ovaries (MFO), as did also the Rotterdam consensus conference. Actually, there is no consensual definition for MFO, although they have been described as ovaries in which there are multiple ( 6) follicles, usually 4 10 mm in diameter, with normal stromal echogenicity and, for some authors, slightly increased ovarian sizes (Adams et al., 1985). Since no histological data about MFO are available, it is not known whether all stages of follicular growth are involved and, hence, whether or not they represent an entity distinct from PCO. Clinically, MFO are characteristically seen in situations other than PCOS, mainly during normal puberty and in women suffering from hypothalamic amenorrhoea (HA). Although the clinical picture of HA is theoretically different, the combination of amenorrhoea and PCO-like features could erroneously lead to the diagnosis of PCOS if one applied the Rotterdam definition too inflexibly. Such an overlap between PCO and MFO emphasizes the need for a careful exclusion diagnosis before applying the Rotterdam criteria as well as for considering ultrasound criteria other than FN and OV in difficult situations. This point raises the question as to whether the OA should be preferred to OV. Besides providing an accurate estimation of ovarian size, freezing a longitudinal plane also allows estimation of the amount of ovarian stroma, which has been shown to be closely related to serum androgen level, an excess of which is specific of PCO (reviewed in Balen et al., 2003). In conclusion, this study confirms that measurement of the OV yields good diagnostic accuracy to distinguish normal ovaries from PCO. However, it shows that the best compromise between sensitivity and specificity is obtained with a threshold set at 7 cm 3, instead of the 10 cm 3 threshold proposed by the Rotterdam consensus conference on the basis of empirical assessments. In addition, we confirm on a different series that priority should be given to an FN >12 to define PCO by ultrasound (Jonard et al., 2003), as proposed by the Rotterdam consensus conference in 2003 (Rotterdam ESHRE/ASRM-sponsored PCOS consensus workshop group, 2004). Finally, although less used than OV, the OA has a higher diagnostic power than OV. Whether this could help in distinguishing PCO from MFO deserves further studies. Acknowledgements We thank Mrs Lydie Lombardo and Sylvie Vanoverschelde for their assistance in collecting the blood samples, and Miss Celine Vandaele for her assistance in computing the clinical, biological and ultrasound data. Supported by a grant from the Délégation à la Recherche du CHU de Lille, France. References Adams J, Franks S, Polson DW, Mason HD, Abdulwahid N, Tucker M, Morris DV, Price J and Jacobs HS (1985) Multifollicular ovaries: clinical and endocrine features and response to pulsatile gonadotropin-releasing hormone. Lancet 2,1375 1378. Atiomo WU, Pearson S, Shaw S, Prentice A and Dubbins P (2000) Ultrasound criteria in the diagnosis of polycystic ovary syndrome (PCOS). Ultrasound Med Biol 26,977 980. Balen AH, Laven JSE, Tan SL and Dewailly D (2003) Ultrasound assessment of the polycystic ovary: international consensus definitions. Hum Reprod Update 95 514. Christensen JT, Boldsen J and Westergaard JG (1997) Ovarian volume in gynecologically healthy women using no contraception, or using IUD or oral contraception. Acta Obstet Gynecol Scand 76,784 789. Dewailly D, Robert Y, Helin I, Ardaens Y, Thomas-Desrousseaux P, Lemaitre L and Fossati P (1994) Ovarian stromal hypertrophy in hyperandrogenic women. Clin Endocrinol 41,557 562. Dewailly D, Robert Y, Lions C and Ardaens Y (2002) Ultrasound examination of polycystic and multifollicular ovaries. In Chang RJ, Heindel JJ and Dunaif A (eds), Polycystic Ovary Syndrome. Marcel Dekker, New York, pp. 63 76. DiSantis DJ, Scatarige JC, Kemp G, Given FT, Hsiu JG and Cramer MS (1993) A prospective evaluation of transvaginal sonography for detection of ovarian disease. Am J Roentgenol 161,91 94. Granberg S, Wikland M and Friberg LG (1990) Tumor marker CA 125 and ovarian volume at different cycle day periods and in postmenopause. Int J Gynaecol Obstet 33,149 152. Fulghesu AM, Ciampelli M, Belosi C, Apa R, Pavone V and Lanzone A (2001) A new ultrasound criterion for the diagnosis of polycystic ovary syndrome: the ovarian stroma/total area ratio. Fertil Steril 76,326 331 Jonard S, Robert Y, Cortet-Rudelli C, Pigny P, Decanter C and Dewailly D (2003) Ultrasound examination of polycystic ovaries: is it worth counting the follicles? Hum Reprod 18,598 603 Jonard S, Robert Y, Ardaens Y and Dewailly D (2005) Imaging polycystic ovaries. In Kovacs GT (ed.), Polycystic Ovary Syndrome, 2nd edn. Cambridge University Press, Cambridge, in press. Orsini LF, Venturoli S and Lorusso R (1985) Ultrasonic findings in polycystic ovarian disease. Fertil Steril 43,709 714. Pache TD, Wladimiroff JW, Hop WCJ and Fauser BCJM (1992) How to descriminate between normal and polycystic ovaries: transvaginal US study. Radiology 183,421 423. Pavlik EJ, DePriest PD, Gallion HH, Ueland FR, Reedy MB, Kryscio RJ and van Nagell JR (2000) Ovarian volume related to age. Gynecol Oncol 77,410 412. Pigny P, Desailloud R, Cortet-Rudelli C, Duhamel A, Deroubaix-Allard D, Racadot A and Dewailly D (1997) Serum alpha-inhibin levels in PCOS: relationship to the serum androstenedione level. J Clin Endocrinol Metab 82,1939 1943. Robert Y, Dubrulle F, Gaillandre L, Ardaens Y, Thomas-Desrousseaux P, Lemaitre L, Dewailly D (1995) Ultrasound assessment of ovarian stroma hypertrophy in hyperandrogenism and ovulation disorders: visual analysis versus computerized quantification. Fertil Steril 64,307 312. RotterdamESHRE/ASRM-sponsored PCOS consensus workshop group (2004) Revised 2003 consensus on diagnostic criteria and longterm 2897

S.Jonard, Y.Robert and D.Dewailly health risks related to polycystic ovary syndrome (PCOS). Hum Reprod 19, 41 47. Sample WF, Lippe BM and Gyepes MT (1977) Grey-scale ultrasonography of the normal female pelvis. Radiology 125,477 483. Van Nagel JR, Higgins RV, Donaldson ES, Gallion HH, Powell DE, Pavlik EJ et al. (1990) Transvaginal sonography as a screening method for ovarian cancer. A report of the first 1000 cases screened. Cancer 65,573 577. Van Santbrink EJP, Hop WC and Fauser BCJM (1997) Classification of normogonadotropic infertility: polycystic ovaries diagnosed by ultrasound versus endocrine characteristics of polycystic ovary syndrome. Fertil Steril 67,452 458. Yeh HC, Futterweit W and Thornton JC (1987) Polycystic ovarian disease: US features in 104 patients. Radiology 163,111 116 Zawadski JK and Dunaif A (1992) Diagnostic criteria for polycystic ovary syndrome; towards a rational approach. In Dunaif A, Givens JR and Haseltime F (eds), Polycystic Ovary Syndrome. Blackwell Scientific, Boston, pp. 377 384. Zweig MH and Campbell G (1993) Receiver-operating characteristic (ROC) plots: a fundamental evaluation tool in clinical medicine. Clin Chem 39,561 577. Submitted on March 29, 2005; resubmitted on May 18, 2005; accepted on May 24, 2005 2898