Diffusion-weighted MRI for evaluating perianal fistula activity: Feasibility study

European Journal of Radiology 81 (2012) 2049–2053

Contents lists available at ScienceDirect

European Journal of Radiology journal homepage: www.elsevier.com/locate/ejrad

Diffusion-weighted MRI for evaluating perianal fistula activity: Feasibility study Takeshi Yoshizako a,∗ , Akihiko Wada a , Taro Takahara b , Thomas C. Kwee b , Megumi Nakamura a , Koji Uchida a , Shinji Hara a , Peter R. Luijten b , Hajime Kitagaki a a b

Department of Radiology, Shimane University Faculty of Medicine, P.O. Box 00693-8501, 89-1 Enya Izumo, Japan Department of Radiology, University Medical Center Utrecht, Q 02.2.314, P.O. Box 85500, 3508 GA Utrecht, The Netherlands

a r t i c l e

i n f o

Article history: Received 5 March 2011 Received in revised form 27 April 2011 Accepted 5 June 2011 Keywords: Diffusion-weighted magnetic resonance imaging Perianal fistula Apparent diffusion coefficient Inflammation activity Abscess

a b s t r a c t Purpose: To assess the feasibility of using diffusion-weighted magnetic resonance (MR) imaging (DWI) for evaluating perianal fistula activity. Materials and methods: This study retrospectively assessed 24 patients with clinically suspected perianal fistula who underwent a total of 28 MR examinations after conservative treatment with antibiotics. DWI was performed at 1.5 T, using b-values of 0 and 1000s/mm2 . Apparent diffusion coefficient (ADC) maps were created and ADCs of the lesions were measured. Lesions were classified into two groups based on the need for surgery and surgical findings: positive inflammation activity (PIA) and negative inflammation activity (NIA). ADCs of both groups were compared using an unpaired t-test, and Receiver Operating Characteristic (ROC) analysis was performed. Results: The ADC (in 10−3 mm2 /s) of the PIA group (0.908 ± 0.171) was significantly lower (P = 0.0019) than that of the NIA group (1.124 ± 0.244). The area under the ROC curve was 0.749. An optimal cut-off ADC of 1.109 yielded a sensitivity of 95.7%, a specificity of 50%, a positive predictive value of 71%, and a negative predictive value of 90%. Conclusion: DWI is a feasible method for evaluating perianal fistula activity. The diagnostic performance of this technique should be established in future, larger prospective studies. © 2011 Elsevier Ireland Ltd. All rights reserved.

1. Introduction Perianal fistulas and abscesses are frequently seen in Crohn’s disease, but also in the post-operative pelvis, and in chronic inflammatory processes such as acquired immune deficiency syndrome (AIDS). Perianal fistulas occur in 30–50% of patients with Crohn’s disease at some stage during their lifetime. When a patient develops a fistula, it requires a very long time to achieve healing and in most cases relapses can be expected in the future [1,2]. Accurate assessment of fistula activity is an important factor influencing therapeutic strategy such as the selection of the number of required drugs and type of surgical approach [1]. Perianal fistulas are evaluated by a variety of methods including anal ultrasound, examination under anesthesia, and magnetic resonance (MR) imaging. Anal ultrasound and (preoperative) examination under anesthesia are relatively simple and easy procedures, but often miss the presence of an abscess and fail to demonstrate the underlying perianal muscle layers [3–5]. MR imaging is the most accurate noninvasive method for determining the presence

∗ Corresponding author. Tel.: +81 853 20 2289; fax: +81 853 20 2285. E-mail address: [email protected] (T. Yoshizako). 0720-048X/$ – see front matter © 2011 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.ejrad.2011.06.052

and course of perianal fistulas [3–9]. At present, surgical treatment planning is based on clinical symptoms and the main roles of MR imaging are to define the fistula tract(s) and to determine whether there are any abscesses beyond the detection level of clinical (digital) examination. However, MR imaging may have additional value beyond the mere anatomic evaluation of perianal fistulas, in that it may be used to assess disease activity. It has been reported that active inflammation can be detected using T2-weighted and contrast-enhanced T1-weighted imaging. Active fistulas are filled with pus and granulation tissue and, as such, appear as hyperintense structures on T2-weighted images [3–9]. Occasionally, some hyperintensity may be seen even in the fibrous area, probably reflecting edema [8]. On contrast-enhanced T1-weighted images, active granulation tissue will enhance while fluid in the fistula itself remains hypointense [8,9]. However, hyperintensity may also extend beyond the fistula and its fibrous sleeve, which corresponds to adjacent inflammation [8]. Recently, Horsthuis et al. [10] investigated the usefulness of dynamic contrast-enhanced MR imaging for the diagnosis of disease activity in perianal Crohn’s disease, and concluded that rapid enhancement and maximum enhancement during dynamic series correlate to disease activity. However, they also reported that the number of slices is limited with this method due to the need for a high temporal-spatial resolution, as a result of which the extent of active lesions cannot be well evaluated.

2050

T. Yoshizako et al. / European Journal of Radiology 81 (2012) 2049–2053

pected of having a perianal fistula on the basis of clinical symptoms (anal pain or discomfort), findings at examination under anesthesia, or pelvic computed tomography, were retrospectively assessed. The general exclusion criteria for MR imaging (e.g. claustrophobia, pregnancy, and implanted pacemaker) were applicable. Patients with contraindications for hyoscine-N-butylbromide (Buscopan, Boehringer Ingelheim), including hypersensitivity to anticholinergic drugs, benign prostate hyperplasia, paralytic ileus, closed angle glaucoma, and shallow anterior chamber, were also excluded. Four of the 24 patients underwent MR imaging twice due to recurrent symptoms after healing from an earlier fistula. The patients consisted of 19 men and 5 women, and ranged in age from 22 to 75 years (mean age, 45 years). Furthermore, 10 patients had Crohn’s disease, three patients had rectal cancer, two patients had prostate cancer, one patient had ulcerative colitis, one patient had AIDS, and the remaining seven patients had no relevant underlying disease. All 24 patients received antibiotics for about one week (mean, 6.3 days; range 5–7 days) before the MR examination. Fig. 1. ROC curve for ADC-based discrimination between active and inactive perianal fistulas. Area under the ROC curve was 0.749.

2.2. MR imaging

Another potentially useful MR technique is diffusion-weighted MR imaging (DWI). DWI yields excellent contrast between lesions, such as cancer or inflammatory processes, and surrounding tissue. Although DWI outside the brain is currently mainly used for oncological applications, the evaluation of abscesses can be another very important target [11–20]. Recently, one small study has reported the potential utility of DWI for the detection of perianal fistulas [21]. However, to our knowledge, it is still unknown whether DWI can also be used as a marker for disease activity [21]. The purpose of this study was therefore to assess the feasibility of using DWI for evaluating perianal fistula activity. 2. Materials and methods 2.1. Patients Institutional review board approval was obtained and informed consent was waived for this retrospective study. A total of 28 MR examinations of 24 consecutive patients who were clinically sus-

All patients were examined with a 1.5-T system (Signa CV/i ver. 9.1 GE Medical System Milwaukee, WI) using a pelvic phased-array coil. All patients were given an intramuscular injection of hyoscineN-butylbromide (Buscopan, Boehringer Ingelheim) to minimize artifacts due to peristalsis. No oral or rectal contrast agents were administered. Non-contrast-enhanced T1-weighted imaging (repetition time/echo time [TR/TE], 600/14 ms) and fat-suppressed T2weighted imaging (TR/TE 4000/100 ms; echo-train length, 12) were performed in the axial plane (i.e. true axial plane). Other imaging parameters included a slice thickness of 5 mm with an interslice gap of 0.5 mm, a matrix size of 256 × 224, and a field of view (FOV) of 220 mm × 220 mm. Scan time for T1-weighted imaging was 2 min 13 s and that of fat suppressed T2-weighted imaging was 2 min 14 s. DWI was performed in the axial plane (i.e. true axial plane), using a single-shot spin-echo echo-planar-imaging (SE-EPI) sequence. Imaging parameters included a TR/TE of 4000/83.7 ms, a flip angle of 90◦ , a slice thickness of 5 mm with an interslice gap of 0.5 mm, number of slices of 16, an actual matrix size of 128 × 192 with reconstruction to 256 × 256, a FOV of 420 mm × 210 mm, a bandwidth of 110 kHz, number of signal averages of 8, b-values of 0 and 1000s/mm2 . Three orthogonal directions of motion probing gradients were used. Scan time for DWI was 2 min 28 s. 2.3. Data processing

Fig. 2. Scatterplot with ADCs of lesions in the positive inflammation activity (PIA) group and ADCs of lesions in the negative inflammation activity (NIA) group. The horizontal line in the scatterplot denotes the cut-off ADC that yields the combination of the highest sensitivity and specificity. The optimal cut-off ADC (in 10−3 mm2 /s) was 1.109. This cut-off ADC yielded a sensitivity of 95.7%, a specificity of 50%, a positive predictive value of 71%, and a negative predictive value of 90%.

Axial ADC maps were created using the b-values of 0 and 1000s/mm2 . One board-certified radiologist (-.-. with more than 15 years of experience in abdominal MR imaging) who was blinded to clinical course and other imaging findings, placed a freehand region of interest (ROI) in each lesion on the ADC map (encompassing as much of the lesion as possible), and ADCs were automatically calculated. Only the slice containing the largest portion of each lesion was selected for ROI analysis. The positions of the ROIs were determined by reviewing both diffusion-weighted and fat-suppressed T2-weighted images. If a patient had multiple lesions, all lesions were analyzed. MR imaging criteria for perianal fistula were based on shape and signal intensities [3–10]; a perianal fistula appears as a linear or oval structure surrounded by an irregular area at MR imaging. In addition, the lesion appears hypointense to isointense on T1-weighted images and isointense to hyperintense on T2-weighted images, relative to muscular structures. All fistulas were classified according to system that was proposed by Parks

T. Yoshizako et al. / European Journal of Radiology 81 (2012) 2049–2053

et al. [22] and locations of major extensions and abscesses were described. 2.4. Reference standard As mentioned previously, all 24 patients received antibiotics for about one week before the MR examination. Therefore, the results of this study are only applicable to this specific patient population. Antibiotics were given as first-line treatment as part of routine patient care. If antibiotics did not promote healing, fistula surgery was considered. A surgeon decided the need for fistula surgery (either open surgery or the use of setons) based on the presence of several symptoms, as follows: identification of pus, severe pain, reddish and swollen skin, elevated serum C-reactive protein (CRP) levels (>2 mg/L), and no improvement of symptoms during observation. Fistulas were confirmed to be active if pus was identified during surgery. Lesions which needed surgery within one week after the MR examination, and which were confirmed to be active during surgery, were classified into the positive inflammation activity (PIA) group. A follow-up time of one week was used, because new (active) lesions may have appeared if a longer followup time would be applied. On the other hand, surgery was not deemed necessary in case of mild pain, improvement of symptoms during observation, and normal serum CRP levels (<2 mg/L). Lesions which did not require surgery within one week of the MR examination, and lesions which were not considered active during surgery, were classified into the negative inflammation activity (NIA) group. The surgeon was not blinded to MR imaging, but MR images were only used for localizing the (suspected) lesions and not for assessing disease activity. The surgery was performed by one of two colorectal surgeons with more than 10 years of experience.

2.5. Statistical analysis Kolmogorov–Smirnov tests were used to check whether ADCs of the two different groups (i.e. PIA and NIA) were normally distributed. ADCs of both groups were compared using an unpaired t-test. If the unpaired t test revealed a significant difference in ADCs between both groups, additional Receiver Operating Characteristic (ROC) analysis was performed to determine the area under the ROC curve and the optimal cut-off ADC with corresponding sensitivity, specificity, positive predictive value, and negative predictive value. P values less than 0.05 were considered to indicate a statistically significant difference. Statistical analyses were executed using MedCalc version 10.4.5.0 software (MedCalc, Mariakerke, Belgium).

3. Results Table 1 shows the classification of fistulas according to the system that was proposed by Parks et al. [22] and locations of major extensions and abscesses. In the 24 patients, a total of 41 lesions were observed (lesions per patient: mean, 1.46; range, 1–3/10 patients with multiple lesions), of which 23 lesions were classified in the PIA group, and the remaining 18 lesions were classified in the NIA group. The latter group included 4 fistulas that were considered inactive during surgery because of the absence of pus and a scar-like appearance. Mean size of ROIs used for ADC measurements was 9.3 mm2 (range, 6–18.2 mm2 ). The ADC of PIA lesions (in 10−3 mm2 /s) was 0.908 ± 0.171 (mean ± SD), while that of NIA lesions was 1.124 ± 0.244. There was a significant difference in ADCs between PIA and NIA lesions (P = 0.0019). The area under the ROC curve was 0.749 and the optimal cut-off ADC was 1.109. This

2051

optimal cut-off ADC yielded a sensitivity of 95.7%, a specificity of 50%, a positive predictive value of 71%, and a negative predictive value of 90% (Figs. 1 and 2). Representative examples are shown in Figs. 3 and 4.

4. Discussion The results of the present study show that DWI with ADC measurements may be useful for discriminating active from inactive perianal fistulas in patients who had previously received conservative treatment with antibiotics. Because the negative predictive value of ADC measurements was relatively high in the present study, ADC measurements may particularly be helpful in excluding active inflammation. DWI is an MR technique that is sensitive to the random (Brownian) motion of water molecules, which, in biologic tissue, is thought to mainly depend on the amount of water that is located in the extracellular space [20]. Cancerous lesions frequently have relatively less extracellular space because of high cellular density, which corresponds to a low diffusivity and consequent high signal intensity at DWI. However, an abscess, which contains inflammatory cells, a matrix of proteins, cellular debris, and bacteria in high-viscosity pus [14], can also exhibit a low diffusivity. In a retrospective study including 13 patients with 20 perianal fistulas, Hori et al. [21] recently investigated the value of DWI for perianal fistula detection. Eighteen (90%) fistulas were detected on T2-weighted images, and 19 (95%) and 19 (95%) were detected on diffusion-weighted and T2-weighted images combined and on contrast-enhanced and T2-weighted images combined, respectively. There was no statistically significant difference in sensitivity of the techniques (P > 0.5 for all comparison pairs). Confidence scores with diffusion-weighted and T2-weighted images combined or those with contrast-enhanced and T2-weighted images combined were significantly greater than those with T2-weighted images alone (P = 0.0047 and P = 0.014, respectively). Hori et al. [21] concluded that DWI is a useful sequence and can be a helpful adjunct to T2-weighted imaging for perianal fistula detection, especially in patients with risk factors for contrast agents. However, Hori et al. [21] did not investigate whether DWI could also be used to assess disease activity, as was shown in the present study. Important advantages of DWI over dynamic contrast-enhanced MR imaging [10] are that no contrast agents have to be administered, it can readily be implemented at commercial scanners, and data post-processing is relatively easy. In addition, DWI well visualizes the three-dimensional configuration of lesions thanks to its high lesion-to-background contrast [13,21]. Although our study showed that ADCs of the PIA group were significantly lower than those of the NIA group, a significant overlap was still observed. Interestingly, Guo et al. [15] reported that small abscesses have higher ADCs than large abscesses. A possible explanation for Guo et al.’s observation [15] may be that the positioning of ROIs and corresponding ADC measurements can be difficult in small abscesses. Another explanation may be that small abscesses really have a higher diffusion coefficient than large abscesses because of different contents. In the present study, a relatively high number of small abscesses was included (mean size of ROIs used for ADC measurements was 9.3 mm2 [range, 6–18.2 mm2 ]) which may in part explain the considerable overlap in ADCs between PIA and NIA lesions. Other investigators [16] have suggested that variable concentrations of inflammatory cells and bacteria are present with different pathogenic organisms and that the host immune response and the age of an abscess might influence the viscosity of the pus, resulting in variations in ADCs. Another report that analyzed a series of hepatic abscesses stated that changes in ADCs could

2052

T. Yoshizako et al. / European Journal of Radiology 81 (2012) 2049–2053

Table 1 Classification of fistulas according to the system that was proposed by Parks et al. and locations of major extensions and abscesses. Type of fistula

Number of lesions

Superficial Intersphincteric Transsphincteric Suprasphincteric Extrasphincteric Total

15 12 6 4 4 41

Locations of major extensions or abscessesa Intersphincteric horseshoe

Ischio-anal fossa

Supralevator

9 (8) 8 (8) 0 0 0 17 (16)

0 0 5 (5) 2 (2) 2 (2) 9 (9)

0 0 0 2 (2) 2 (2) 4 (4)

a 11 lesions had no major extension. () shows the number of placed ROI for the ADC measurement.

Fig. 3. A 73-year-old man (with previous rectal cancer resection) with a perianal fistula without active inflammation. (A) Axial fat saturated T2-weighted imaging shows a perianal lesion (arrow) that exhibits high signal intensity, corresponding to a fistula. (B) At axial DWI (gray-scale inverted display), the lesion (arrow) exhibits spotty slightly high signal intensity. The lesion was determined to be a non-active lesion by the surgeon and healed without surgical intervention. (C) At the axial ADC map (including ROI), the ADC (in 10−3 mm2 /s) of the lesion was 1.199 ± 0.164.

Fig. 4. A 46-year-old man (who had no relevant underlying disease) with a perianal fistula with active inflammation. (A) Axial fat saturated T2-weighted imaging shows a perianal lesion (arrow) that exhibit high signal intensity, corresponding to a fistula. (B) At axial DWI (gray-scale inverted display), the lesion (arrow) exhibits high signal intensity. The lesion was judged to be an active inflammation by the surgeon and was surgically treated. (C) At the axial ADC map (including ROI), the ADC (in 10−3 mm2 /s) of the lesion was 0.921 ± 0.147.

T. Yoshizako et al. / European Journal of Radiology 81 (2012) 2049–2053

occur during disease course [19]. All these issues can cause variations in ADCs, which may result in an overlap in ADCs between PIA and NIA lesions. Another explanation may be that both active inflammation and fibrotic changes (due to chronic inflammation and sequels) may lead to decreased ADCs that overlap with each other. Our study had several limitations. First, our retrospective study included a relatively low number of subjects. Therefore, our imaging findings and conclusions should be interpreted as preliminary and subject to further prospective investigation. This is also underlined by the fact that the optimal cutoff ADC value was retrospectively determined, and corresponding measures of diagnostic accuracy were only moderate (AUC of 0.75, sensitivity and specificity of 95.7% and 50%, and positive predictive value and negative predictive value of 71% and 90%). Second, all patients received antibiotics before the MR examination, which may have influenced fistula activity (active fistulas may have become inactive) and ADCs before the MR examination. Antibiotics were given as routine firstline treatment and it would be unethical to withdraw patients from using antibiotics in the interest of this study. It should also be emphasized that all patients received antibiotics; there was no patient selection bias in this respect. Nevertheless, it should be emphasized that the results of this study are only applicable to patients who are clinically suspected of having a perianal fistula and who had previously received antibiotics. Third, only two b-values (0 and 1000s/mm2 ) were used for ADC calculation. It may have been desirable to acquire more b-values in order to obtain more accurate ADCs. Fourth, only one observer performed the ADC measurements. Future studies should assess the observer agreement of this method. Fifth, surgeons were not blinded to MR imaging in our study. Nevertheless, the surgeons used MR imaging only for localizing (suspected) lesions and not for assessing disease activity. Sixth, the applied reference standard, which was based on the need for surgery (as was determined clinically) and surgical findings, may be regarded as imperfect. Seventh, our MR protocol did not include a contrast-enhanced sequence, whereas it would be of interest to assess whether the ADC can replace contrast enhancement as an indicator of disease activity. Eighth, the applied slice thickness for DWI was relatively large (5 mm with 0.5 mm gap), which could have compromised the evaluation of small perianal fistulas. Nevertheless, this slice thickness is fairly common for current body DWI studies [13], and thanks to its high lesion-to-background contrast, small fistulas with a size below the spatial resolution may still be visible at DWI. 5. Conclusion In conclusion, there was a significant difference in ADCs between PIA and NIA lesions in patients with perianal fistula who had previously received conservative treatment with antibiotics. Therefore, DWI may be of value for evaluating perianal fistula activity.

2053

Conflict of interest We declare that I do not have any affiliation with or financial relationship/interest in a commercial organization that could pose a conflict of interest. References [1] Keshaw H, Foong KS, Forbes A, Day RM. Perianal fistulae in Crohn’s disease: current and future approaches to treatment. Inflamm Bowel Dis 2010;16(5):870–80. [2] Schwartz DA, Loftus Jr EV, Tremaine WJ, et al. The natural history of fistulizing Crohn’s disease in Olmsted County, Minnesota. Gastroenterology 2002;122(4):875–80. [3] Bell SJ, Halligan S, Windsor AC, Williams AB, Wiesel P, Kamm MA. Response of fistulating Crohn’s disease to infliximab treatment assessed by magnetic resonance imaging. Aliment Pharmacol Ther 2003;17(3):387–93. [4] Schwartz DA, Wiersema MJ, Dudiak KM, et al. A comparison of endoscopic ultrasound, magnetic resonance imaging, and exam under anesthesia for evaluation of Crohn’s perianal fistulas. Gastroenterology 2001;121(5):1064–72. [5] Buchanan G, Halligan S, Williams A, et al. Effect of MRI on clinical outcome of recurrent fistula-in-ano. Lancet 2002;23(9346):1661–2. [6] Spencer JA, Ward J, Beckingham IJ, Adams C, Ambrose NS. Dynamic contrast-enhanced MR imaging of perianal fistulas. Am J Roentgenol 1996;167(3):735–41. [7] Beets-Tan RG, Beets GL, van der Hoop AG, et al. Preoperative MR imaging of anal fistulas: does it really help the surgeon. Radiology 2001;218(1):75–84. [8] Halligan S, Stoker J. Imaging of fistula in ano. Radiology 2006;239(1):18–33. [9] Szurowska E, Wypych J, Izycka-Swieszewska E. Perianal fistulas in Crohn’s disease: MRI diagnosis and surgical planning MRI in fistulazing perianal Crohn’s disease. Abdom Imaging 2007;32(6):705–18. [10] Horsthuis K, Lavini C, Bipat S, Stokkers PC, Stoker J. Perianal Crohn disease: evaluation of dynamic contrast-enhanced MR imaging as an indicator of disease activity. Radiology 2009;251(2):380–7. [11] Unal O, Koparan HI, Avcu S, Kalender AM, Kisli E. The diagnostic value of diffusion-weighted magnetic resonance imaging in soft tissue abscesses. Eur J Radiol 2011;77(3):490–4. [12] Vollmann R, Hammer G, Simbrunner J. Pathways in the diagnosis of prevertebral tendinitis. Eur J Radiol 2011, doi:10.1016/j.ejrad.2011.02.061. [13] Takahara T, Imai Y, Yamashita T, Yasuda S, Nasu S, Van Cauteren M. Diffusion weighted whole body imaging with background body signal suppression (DWIBS): technical improvement using free breathing, STIR and high resolution 3D display. Radiat Med 2004;22(4):275–82. [14] Cartes-Zumelzu FW, Stavrou I, Castillo M, Eisenhuber E, Knosp E. Diffusionweighted imaging in the assessment of brain abscesses therapy. Am J Neuroradiol 2004;25(8):1310–7. [15] Guo AC, Provenzale JM, Cruz Jr LC, Petrella JR. Cerebral abscesses: investigation using apparent diffusion coefficient maps. Neuroradiology 2001;43(5):370–4. [16] Kim YJ, Chang KH, Song IC, et al. Brain abscess and necrotic or cystic brain tumor: discrimination with signal intensity on diffusion weighted MR imaging. Am J Roentgenol 1998;171(6):1487–90. [17] Thiriat S, Riehm S, Kremer S, Martin E, Veillon F. Apparent diffusion coefficient values of middle ear cholesteatoma differ from abscess and cholesteatoma admixed infection. Am J Neuroradiol 2009;30(6):1123–6. [18] Chan JH, Tsui EY, Luk SH, et al. Diffusion-weighted MR imaging of the liver: distinguishing hepatic abscess from cystic or necrotic tumor. Abdom Imaging 2001;26(2):161–5. [19] Holzapfel K, Rummeny E, Gaa J. Diffusion-weighted MR imaging of hepatic abscesses: possibility of different apparent diffusion coefficient (ADC)-values in early and mature abscess formation. Abdom Imaging 2007;32(4):538–9. [20] Le Bihan D. Molecular diffusion nuclear magnetic resonance imaging. Magn Reson Q 1991;7(1):1–30. [21] Hori M, Oto A, Orrin S, Suzuki K, Baron RL. Diffusion-weighted MRI: a new tool for the diagnosis of fistula in ano. J Magn Reson Imaging 2009;30(5):1021–6. [22] Parks AG, Gordon PH, Hardcastle JD. A classification of fistula-in-ano. Br J Surg 1976;63(1):1–12.