Effect of MRI on Therapeutic Decisions in Invasive Cervical Carcinoma

Gynecologic Oncology 79, 485– 489 (2000) doi:10.1006/gyno.2000.5986, available online at http://www.idealibrary.com on

Effect of MRI on Therapeutic Decisions in Invasive Cervical Carcinoma Direct Comparison with the Pelvic Examination as a Preoperative Test Sandra Postema, M.D.,* Peter M. T. Pattynama, M.D.,† Annette van den Berg-Huysmans, M.Sc.,* Lex W. Peters, M.D.,‡ Gemma Kenter, M.D.,‡ and J. Baptist Trimbos, M.D.‡ *Department of Radiology and ‡Department of Gynecology, Leiden University Medical Center, Postbus 9600, 2300 RC Leiden, The Netherlands; and †Department of Radiology, Erasmus University Medical Center Rotterdam, Dr. Molewaterplein 40, 3015 GD Rotterdam, The Netherlands Received May 15, 2000

a useful, more verifiable, and more reproducible method for staging of cervical carcinoma [1–5]. MRI can usually be performed on an outpatient basis, thus without the costs and inconvenience of hospitalization and anesthesia [1]. To our knowledge, however, there are few objective data on the effects of using MRI on the accuracy of treatment planning, especially not in comparison with the generally accepted clinical test, the pelvic examination. Beyond that, most of the previous MRI studies have been performed with low- and midfield magnets and using relatively outdated imaging sequences. The primary aim of this study, then, was to determine the accuracy of MRI at 1.5 T for treatment planning in patients with invasive cervical carcinoma, in a direct comparison with the standard practice of pelvic examination (including examination under general anesthesia in selected patients). If MRI would prove to be as reliable as the pelvic examination or even better for treatment planning, this would be a persuasive argument to use MRI as an alternative staging method, especially as compared to the pelvic examination under general anesthesia. The secondary aim was to assess the accuracy and reproducibility of MRI for parametrial extension of the tumor and involvement of the bladder.

Objectives. Our aim was to compare magnetic resonance imaging (MRI) with the current standard clinical practice (pelvic examination including general anesthesia in selected patients) with regard to treatment planning in invasive cervical carcinoma. It was of particular interest to compare the accuracy of both methods for allocating the patients to the appropriate treatment modality: surgery versus primary radiotherapy. Methods. One hundred and three consecutive patients with primary invasive cervical carcinoma underwent both MRI at 1.5 T and pelvic examination. The gold standard for comparing treatment decisions was based on the surgico-pathologic data: tumor confined to the cervix (treatment decision for surgery) or extracervical tumor spread (treatment decision for primary radiotherapy). Results. A gold standard was available in 91 patients. The pelvic examination made correct treatment decisions in 89% of patients. However, the sensitivity for extracervical spread was only 44% (8/18 patients). MRI was better at identifying extracervical tumor spread: 67 and 89% for observers 1 and 2, respectively. MRI, however, had more false positive results and correct treatment decisions were made in 69 – 84% of patients (observer 1, 76/91; observer 2, 63/91). Conclusion. Treatment decisions based on the pelvic examination were correct in 89%, with MRI not bringing improvement. MRI, however, is better in diagnosing extracervical spread, but at the cost of more false positives. © 2000 Academic Press

SUBJECTS AND METHODS INTRODUCTION

Patients MRI was performed in 103 consecutive patients (26 –79 years old, mean 49.5 years) who were referred to the Department of Gynecology of our hospital for treatment of primary invasive cervical carcinoma. All patients underwent a pelvic examination by an experienced gynecologic oncologist during the initial outpatient visit. In 78 (76%) patients, this was extended to include a pelvic examination under general anesthesia (performed in cooperation with an experienced radiation oncologist) because the preliminary examination proved to be

In staging for invasive cervical carcinoma the clinically relevant issue concerns the choice of treatment: surgery versus primary radiotherapy. Currently, as the standard clinical practice, the pelvic examination has the central place in tumor staging—in all patients this takes place in the gynecological office during an outpatient visit. If necessary the pelvic examination is extended to include an examination performed under general anesthesia, requiring a short hospital stay. Magnetic resonance imaging (MRI) has been introduced as 485

0090-8258/00 $35.00 Copyright © 2000 by Academic Press All rights of reproduction in any form reserved.

486

POSTEMA ET AL.

TABLE 1 Summary of FIGO Staging Classification of Carcinoma of the Cervix and Corresponding MR Findings Stage 0 I IA IB

II

Summary of FIGO classification In situ Confined to cervix (extension to corpus disregarded) Preclinical Clinically invasive

IIIA IIIB IVA

Extending beyond cervix but not to pelvic wall or lower third of vagina No parametrial involvement Parametrial involvement Extending to pelvic wall and/or lower third of vagina No extension to pelvic wall Extension to pelvic wall Spread to bladder or rectum

IVB

Spread to distant organs

IIA IIB III

MR findings No evidence of mass lesion

No evidence of mass lesion Definitive: tumor completely surrounded by low-signal-intensity ring of the cervix (axial images) Suggestive: sharply marginated tumor disrupting low-signal-intensity ring of the cervix (axial images)

Disruption of low-signal-intensity vaginal wall (sagittal images) Protrusion of tumor through disrupted low-signal-intensity ring of the cervix (axial images)

Same finding as for IIA in the lower third of the vagina Same finding as for IIB beyond lateral margin of cardinal ligament or to pelvic muscles Segmental disruption of the low-signal-intensity bladder wall or thickened high-signalintensity rectal wall Evidence of mass lesions in distant organs

inconclusive with regard to extracervical tumor extension. MRI was performed in all 103 patients within 2 weeks of the initial pelvic examination. The patients were treated based on the clinical stage according to the rules of the International Federation of Gynecology and Obstetrics (FIGO). The results of the MRI examinations were not revealed to the treating gynecologists and did not play a role in treatment planning. The study was approved by the ethical review board of our institution and all patients gave informed consent. MRI Protocol The MRI examinations were performed on a 1.5 T magnet (Gyroscan NT-15, Philips Medical Systems, Best, the Netherlands) with use of the system body coil. The imaging protocol consisted of T2-weighted fast spin-echo (SE) images in transverse and sagittal planes (TR/TE, 2500/120 ms; field-of-view (FOV), 22 cm; matrix size, 256 ⫻ 256; 8 excitations; echotrain-length (ETL), 18), as well as a T1-weighted sequence in the transverse plane (TR/TE, 450/15 ms; FOV, 22 cm; matrix size, 256 ⫻ 256; 4 excitations). Multiple 5-mm-thick sections with 1 mm intersection gap covered the area of interest. Previously we have shown that the here-used fast SE sequence yields optimal contrast-to-noise ratio, when compared with conventional SE MRI or alternative fast SE sequences including GRASE [6, 7]. Intravenous Buscopan was given to reduce motion artifacts by bowel movements. The MRI examinations were read independently by two experienced radiologists, who were blinded to patient identity, the results of the physical examination, and the surgico-pathologic findings. The following items were scored: vaginal wall involvement, parametrial ingrowth, pelvic wall ingrowth, bladder wall involvement, and rectal wall involvement. Based on

this evaluation a MRI-based FIGO stage (Table 1) was attributed that was used for allocating the patients to different treatment groups: surgery for stage IA–IIA and radiotherapy for stage IIB–IV. Scoring was done on 5-point confidence scales to allow receiver-operating-characteristic (ROC)analysis. Pelvic Examination Pelvic examinations were performed by experienced gynecologic oncologists who stated their opinion on the FIGO disease stage, based on assessment of tumor ingrowth into the vagina, parametria, and/or pelvic wall. The pelvic examination under general anesthesia was performed in cooperation with an experienced radiation oncologist. Patients were allocated to treatment groups based on the evaluation of local tumor spread as explained above. For the purpose of this study, additional examinations (IVU, cystoscopy, urine cytology, etc.) were disregarded. Gold Standard and Statistical Analysis It was of interest to compare the accuracy of MRI with that of pelvic examination with regard to treatment planning. The optimal treatment for each patient was determined in retrospect, considering the surgico-pathological data on the local pelvic extension of tumor mass. These surgico-pathological data were accepted as the gold standard for treatment planning. The lymph node status was not included in the analysis. Primary radiotherapy was considered the appropriate treatment whenever the postoperative evaluation resulted in local extracervical tumor extension. In the patients with tumor confined to the cervix and/or in the upper 1/3 of the vagina, surgery was

487

EFFECT OF MRI OF THERAPEUTIC DECISIONS

TABLE 2 Sensitivity (True Positive Fraction of Patients Requiring Radiotherapy) and Specificity (True Positive Fraction of Patients Requiring Surgery) of the Pelvic Examination and of MRI, with Regard to Treatment Decisions in Invasive Cervical Carcinoma

Physical examination MRI Observer 1 Observer 2

Sensitivity (%)

Specificity (%)

44

100

89 89

82 64

the appropriate treatment, irrespective whether or not additional postoperative radiotherapy was necessary. Postoperative radiotherapy was given whenever the histologic invasion depth was ⬎15 mm, and/or whenever lymph node metastasis was present [8], and/or in the presence of positive surgical margins. Patients treated with primary radiation were only included in the analysis if a gold standard for treatment decision was available. In this patient group hydronephrosis (in n ⫽ 2 patients) was accepted as gold standard evidence for stage III disease (ureteral involvement) and positive cystoscopy/urine cytology (n ⫽ 7) was accepted as gold standard evidence for stage-IVA disease (bladder involvement). We compared the various diagnostic methods with regard to sensitivity and specificity and with regard to the proportion of patients that were allocated to the correct treatment. Differences were tested for statistical significance using McNemar ␹ 2 tests at P ⫽ 0.05. In this context, sensitivity was defined as the true positive fraction of patients requiring radiotherapy that was identified by the test and specificity as the true positive fraction requiring surgery. The interobserver reproducibility for MRI readings was determined using the ␬-index. From the 5-point confidence scales ROC-curves were constructed and the areas under the curves (AUC) were calculated with ROCFIT software (C. E. Metz, M.D., Chicago, IL). Validation of MRI with regard to parametrial and bladder wall involvement was based on comparison with surgicopathological data. When analyzing the data for parametrial involvement, the left and right parametria were regarded separately (a single patient thus contributed two data points to the comparison). Sensitivity and specificity of MRI was determined for each observer and were tested for significant differences using the McNemar ␹ 2 test at P ⫽ 0.05. ROC curves were constructed and compared. RESULTS The results of the pelvic examination as performed on an outpatient’s basis in the gynecological office were accepted as reliable enough for treatment planning in 25 women. In the other 78 patients an additional pelvic examination under general anesthesia was performed because some doubts remained

after the initial examination (more complete muscle relaxation during general anesthesia improves the accuracy of the pelvic examination). Of the 103 consecutive patients, 82 initially underwent radical hysterectomy. In 41 patients surgery alone was considered to be adequate treatment. In 9 patients extracervical spread of the disease was found (bladder wall involvement, n ⫽ 3; invasion of parametria, n ⫽ 5; involvement of bowel, n ⫽ 1) for which primary radiotherapy would have been the treatment of choice. Postoperative radiotherapy was also required in 32 other patients. The reasons for this were positive surgical margins (n ⫽ 10), and/or pelvic lymph node involvement (n ⫽ 21), and/or tumor invasion depth exceeding 15 mm (n ⫽ 16). The remaining 21 patients were treated with primary radiotherapy; in 15 cases because of high FIGO-stage (IIB–IV) and in 6 because of reasons unrelated to tumor staging making them unfit for radical surgery (e.g., obesity). A gold standard was available in all 82 surgical patients. In the patients who underwent primary radiotherapy, a gold standard was available only in 9 patients who had involvement of the ureter (n ⫽ 2) or the bladder (n ⫽ 7) (see under Subjects and Methods). A gold standard was thus available in 91 patients, in whom the retrospectively determined optimal treatment would have been primary radiotherapy in 18 patients and surgery (with/without additional postoperative radiotherapy) in 73 patients. Accuracy of the Pelvic Examination The sensitivity of the pelvic examination for extracervical tumor spread was only 44% (8/18), with a specificity of 100% (73/73, Table 2). Treatment decisions based on the pelvic examination were correct in 81/91 patients (89%). Ten patients were understaged (11%); they were assigned to surgery but in fact required primary radiotherapy (Table 3). Accuracy of MRI For MRI, sensitivity was 89% (16/18) and specificity was 82% (60/73) for observer 1 and 89% (16/18) and 64% (47/73), respectively, for observer 2 (Table 2). MRI observer 1 made 76/91 (84%) correct decisions but allocated 2 patients (2%) wrongly to surgery and 13 (14%) wrongly to radiotherapy. Observer-2 was correct in 63/91 patients (69%) and made 28 (31%) errors by allocating 26 patients (28%) wrongly to ra-

TABLE 3A Number of Patients, for Each Test/Observer That Is Correctly Assigned a Specific Treatment, out of a Total of 91 Consecutive Patients with Invasive Cervical Carcinoma

Correctly to surgery Correctly to radiotherapy Total correct

Physical examination

MRI-obs 1

MRI-obs 2

73 8 81 (89%)

60 16 76 (84%)

47 16 63 (69%)

488

POSTEMA ET AL.

TABLE 3B Number of Patients, for Each Test/Observer That Is Incorrectly Assigned a Specific Treatment, out of a Total of 91 Consecutive Patients with Invasive Cervical Carcinoma

Incorrectly to surgery Incorrectly to radiotherapy Total incorrect

Physical examination

MRI-obs 1

MRI-obs 2

10 0 10 (11%)

2 13 15 (16%)

2 26 28 (31%)

diotherapy and 2 (2%) wrongly to surgery (Table 3). Interestingly, of the 10 patients who were understaged by the pelvic examination, 6 and 8 were correctly assigned to primary radiotherapy by MRI-observers 1 and 2, respectively. Accuracy and Reproducibility of MRI for Extracervical Extension For observer 1 sensitivity and specificity for bladder involvement were 77% (7/9) and 97% (71/73), respectively; for observer 2 sensitivity and specificity were 67% (6/9) and 93% (68/73), respectively. The McNemar’s ␹ 2 test was not significant. The interobserver agreement was good with a gamma of 0.75. Considering parametrial invasion of the tumor sensitivity and specificity were 20% (2/10) and 97% (134/138) for observer 1 and 60% (6/10) and 73% (101/138) for observer 2, respectively. There was only a reasonably good interobserver agreement (gamma ⫽ 0.57). Observer Variability in MRI The interobserver agreement for treatment allocation with MRI was only moderate, with ␬ ⫽ 0.48 (the kappa-values based on readings in all 103 patients). An explanation for this relatively poor observer agreement was provided by the ROC analysis. This analysis showed that the area under the ROC curve, and thus, overall accuracy, was not significantly different for the two observers, with areas under the ROC curve of 0.90 and 0.84, respectively. However, the poor observer agreement reflected that the two observers operated on different parts of the ROC curve: The operating points at which categories 1–3 (surgical treatment) and categories 4 –5 (radiotherapy as treatment) of the 5-point scale were grouped together were significantly different between the two observers, P ⬍ 0.001 (false positive fraction, 0.07; true positive fraction, 0.65), for observer 1 and (0.37; 0.88) for observer 2, respectively. Observer 1 thus tended toward more “conservative” scoring, favoring surgery, whereas observer 2, based on the same MR images, more frequently considered local tumor invasion present and hence leaned toward primary radiotherapy. DISCUSSION The present study investigated the accuracy of MRI at 1.5 T, as a convenient method for allocating patients with invasive

cervical carcinoma to the correct treatment. From the present data it appears that MRI is as accurate for treatment planning as the standard pelvic examination. However, the pelvic examination is relatively biased toward surgery since extracervical tumor spread will often go unnoticed. Extracervical spread was missed by the pelvic examination in 10 of 18 patients (56%). This finding was in accordance with earlier studies, with reported staging errors of the pelvic examination ranging 17–32% in stage IB disease and 50 – 64% in stage IIA–IIIB disease [9 –14]. On the other hand, pelvic examination was more accurate than MRI in allocating the optimal treatment modality (89%). By contrast, extracervical spread was more accurately assessed with MRI. Extracervical extension was recognized by MRI in 12/18 patients (67%) for observer 1 and in 16/18 (89%) for observer 2. MRI has relatively good accuracy for detecting bladder involvement but only moderate accuracy for detecting parametrial tumor ingrowth. On the downside, MRI more often falsely diagnosed extracervical spread. Overstaging with MRI seems a problem especially for parametrial involvement and not for bladder involvement. In particular, there was considerable observer variability for overstaging with MRI especially with regard to parametrial ingrowth (good specificity of 97% for observer 1 versus low specificity of 73% for observer 2). In the current study, the accuracy of MRI for treatment allocation in 91 patients was 69 – 84%. These results are in good agreement with the previously reported MRI results for staging of cervical carcinoma. This despite the fact that the various MRI studies have been done with different equipment, at different magnetic field strengths and with different T2weighted imaging sequences. By using MRI at 1.5 T in 67 women, Togashi et al. reached an overall staging accuracy of 76%, whereby errors were made, resulting in both under- and overstaging [3]. In a study in 57 patients, most of which were studied with low-field MRI at 0.35 T, Hricak et al. found that MRI staging was accurate in 81%. However, for lesions with extracervical spread, the accuracy of MRI was only 74%. Again overstaging and understaging occurred, although most errors were made in overstaging. In the series of Sironi et al. in 57 patients examined with MRI at 0.5 T, the overall staging accuracy was 85% [4]. Kim et al., in a study involving 99 patients studied with MRI at 0.5 T, the overall accuracy of MRI for preoperative staging was 77% [5]. In the present study we observed a considerable interobserver variability of MRI. The two MRI observers operated on different parts of the ROC curves and they apparently used different levels of confidence to score local tumor invasion. This relatively large observer variability for MRI underscores the need for more standardized MRI reading, focused training, and feedback from the gynecology and pathology departments. Furthermore, from the data it is unclear on what part of the ROC curve one would preferably “be.” The absence of congruity in our data between the two observers regarding the

489

EFFECT OF MRI OF THERAPEUTIC DECISIONS

proper reference point on the ROC curve demonstrates the lack of normative characteristics of the ROC theory. A decision on the optimal part of the ROC curve would depend on the relative weight that one assigns to the respective errors: erroneous allocation to either radiotherapy or to surgery [15]. A limitation of this study relates to the composition of our patient group. It is likely that our patient series contained a preselection bias. As a tertiary referral center, we see a relative large proportion of “surgical cases,” the obvious advancedstage cervical carcinomas being treated with radiotherapy by the referring centers themselves. Looked at it in this way, the present study may overrepresent the patients with more subtle extracervical spread. The effect of this bias, however, seems limited: Because MRI is more accurate than the physical examination in detecting extracervical spread it seems unlikely that these large, advanced-stage carcinomas would be misinterpreted. A second limitation is that, whereas the MRI examinations were scored by blinded observers, the physical examination was performed by nonblinded observers who had knowledge of all previous clinical data. This bias would most likely result in underperformance of MRI relative to the physical examination. Finally, not all 103 patients contributed to the analysis. Twelve patients without a gold standard were excluded from analysis because they underwent primary radiotherapy, and this may have constituted a selection bias toward low FIGO stages. Yet it was considered unethical to subject these patients to surgery, which, according to good clinical practice, would be inappropriate/unsuitable treatment. As was mentioned earlier, the grounds on which these patients received primary radiotherapy were a clinical FIGO stage IIB made with a high degree of certainty (n ⫽ 6), suboptimal cardiovascular/pulmonary status (n ⫽ 3), obesity (n ⫽ 2), and patient refusal of surgery (n ⫽ 1). In conclusion, the findings of this study do not support a routine introduction of MRI instead of pelvic examination for staging of cervical carcinoma. For the detection of parametrial involvement MRI is relatively insensitive. We believe that, as yet, MRI should be considered as an additional tool that should be used in combination with pelvic examination. When extracervical tumor spread is suspected, MRI can be a useful diagnostic test, especially with respect to possible bladder wall involvement.

REFERENCES 1. Hricak H, Powell CB, Yu KK, Washington E, Subak LL, Stern JL, Cisternas MG: Invasive cervical carcinoma. Role of MR imaging in pretreatment work-up. Cost minimization and diagnostic efficacy analysis. Radiology 198:403– 409, 1996 2. Hricak H, Lacey CG, Sandles LG, Chang YCF, Winkler ML, Stern JL: Invasive cervical carcinoma: comparison of MR imaging and surgical findings. Radiology 166:623– 631, 1998 3. Togashi K, Nishimura K, Sagoh T, Minami S, Noma S, Fujisawa I, Nakano Y, Konishi J, Ozasa H, Konishi I, Mori T: Carcinoma of the cervix: staging with MR imaging. Radiology 171:245–251, 1989 4. Sironi S, DeCobelli F, Scarfone G, Colombo E, Bolis G, Ferrari A, DelMaschio A: Carcinoma of the cervix: value of plain and gadoliniumenhanced MR imaging in assessing degree of invasiveness. Radiology 188:797– 801, 1993 5. Kim SH, Choi BI, Han JK, Hong DK, Lee HP, Kang SB, Lee JY, Han MC: Preoperative staging of uterine cervical carcinoma: comparison of CT and MRI in 99 patients. J Comp Assist Tomogr 17:633– 640, 1993 6. Postema S, Peters LAW, Hermans J, Trimbos JB, Pattynama PMT: Cervical carcinoma: Do fast SE and fat suppression techniques improve MR tumor staging at 0.5T? J Comput Assisst Tomogr 20:807– 811, 1996 7. Postema S, Pattynama PMT, van Rijswijk CSP, van Erkel A, Tjin A Ton ER: MR imaging of uterine cervical carcinoma: comparison between fast spin-echo MRI and GRASE. Eur Radiol 8:45– 49, 1998 8. Hatch KD: Cervical cancer, in Berek JS, Hacker NF (eds): Practical Gynecologic Oncology, 2nd ed. William and Wilkins, Baltimore, 1994, pp 259 –261 9. van Nagell JRJ, Roddick JWJ, Lowin DM: The staging of cervical cancer: inevitable discrepancies between clinical staging and pathologic findings. Am J Obstet Gynecol 110:973–977, 1971 10. Averette HE, Ford JH Jr, Dudan RC, Robert E, Girtanner WJ, Hoskins MHL: Staging of cervical cancer. Clin Obstet Gynecol 18:215–232, 1975 11. Lagasse LD, Creasman WT, Shingleton HM, Blessing JA: Results and complications of operative staging in cervical cancer: experience of the gynecologic oncology group. Gynecol Oncol 9:90 –98, 1980 12. Delgado G, Bundy B, Zaino R, Sevin B-U, Creasman WT, Major F: Prospective surgical-pathological study of disease-free interval in patients with stage IB squamous cell carcinoma of the cervix: a gynecologic oncology group study. Gynecol Oncol 38:352–357, 1990 13. Dargent D, Frobert JL, Beau G: V factor (tumor volume) and T factor (FIGO classification) in the assessment of cervix cancer prognosis: the risk of lymph node spread. Gynecol Oncol 22:15–22, 1985 14. Chung CK, Nahhas WA, Zaino R, Stryker JA, Mortel R: Histologic grade and lymph node metastasis in squamous cell carcinoma of the cervix. Gynecol Oncol 12:348 –354, 1981 15. Van Erkel AR, Pattynama PMT: Receiver operating characteristic (ROC) analysis: Basic principles and application in radiology. Eur J Radiol 27:88 –94, 1998