点击显示 收起
1 From the Department of Radiology, Georgetown University Hospital, 3800 Reservoir Rd NW, GC201, Washington, DC 20007-2197. Received March 23, 2001; revision requested April 30; revision received June 8; accepted July 5. Supported in part by research grants from Siemens Medical Systems and the Edward Bennett Williams Interventional Radiology Research and Education Fund.
| ABSTRACT |
|---|
| Top ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES |
|---|
MATERIALS AND METHODS: Two hundred consecutive patients who underwent uterine artery embolization at one institution were prospectively examined. Baseline clinical variables measured included age, race, prior oral contraceptive use or progesterone treatment, prior gonadotropin-releasing hormone agonist treatment, and prior births. Imaging parameters were baseline uterine volume, baseline leiomyoma volume and location, and number of leiomyomas. After treatment, follow-up imaging and questionnaire data were obtained at 3 and 12 months. Associations between baseline characteristics and outcome variables of interest were assessed by using linear regression, logistic regression, Pearson product moment correlation coefficients, and Kendal correlation coefficients, with adjustment for confounding variables when indicated.
RESULTS: Regression models indicated that larger dominant leiomyoma volume was associated with a smaller percentage reduction in volume at 3 months (P = .03). A submucosal leiomyoma location was associated with a greater volume reduction at 3 months (P = .04), but this difference did not persist at 12 months (P = .09). The odds of reported menstrual bleeding improvement at 3 months were higher with a submucosal leiomyoma location (P = .04); however, this association was not statistically significant after adjustment (P = .07). The odds of improved bulk-related symptoms were not associated with leiomyoma volume change or location.
CONCLUSION: Smaller baseline leiomyoma size and submucosal location are more likely to result in a positive imaging outcome. There are limited associations between other baseline parameters and either symptom change or imaging outcome.
Index terms: Arteries, therapeutic embolization, 854.1264, 98.1264 • Leiomyoma, 854.315, 854.318 • Uterine neoplasms, 854.315 • Uterine neoplasms, MR, 854.1214 • Uterine neoplasms, US, 854.1298
| INTRODUCTION |
|---|
| Top ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES |
|---|
As a part of a larger study designed to prospectively evaluate UAE, we gathered clinical and imaging data at baseline and at appropriate follow-up intervals. The purpose of the study we present here is to determine if outcome can be predicted on the basis of preprocedural characteristics.
| MATERIALS AND METHODS |
|---|
| Top ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES |
|---|
Each patient treated presented with heavy menstrual bleeding, bulk-related symptoms, or a combination of symptoms. For the purposes of this study, bulk-related symptoms were defined as pelvic pain, pelvic pressure or heaviness, menstrual cramps, urinary frequency, and/or back or leg pain referable to the fibroids on the basis of their location. Measured baseline clinical variables included age, race, prior oral contraceptive use or progesterone treatment, prior gonadotropin-releasing hormone agonist (Lupron; TAP Pharmaceuticals, Deerfield, Ill) treatment, and prior births. None of the patients had received gonadotropin-releasing hormone treatment for at least 6 weeks prior to UAE. African American and white were the dominant racial groups (African American, 50.0% [n = 100]; white, 45.0% [n = 90]; Hispanic, 1.5% [n = 3]; Asian, 2.5% [n = 5]; and other, 1.0% [n = 2]) in the study population and the only groups with sufficient numbers of patients to be included in the analysis of ethnicity.
Baseline imaging variables were uterine volume, dominant (largest) leiomyoma volume, dominant leiomyoma location, and number of leiomyomas. For determination of these parameters, each patient underwent either pelvic ultrasonography (US) or MR imaging of the pelvis. After examination of the first 14 patients, the protocol was changed such that pelvic MR imaging was required before UAE because of difficulty in obtaining accurate uterine and leiomyoma measurements and in accurately assessing the dominant leiomyoma location with US.
The measurements provided in the radiologic reports were used to calculate volumes. Dominant leiomyoma and uterine volumes both before and after treatment were calculated by using the formula for a prolate ellipse, length x width x diameter x .5233, as described by Orsini et al (12). Leiomyoma location was determined on the basis of the location of the center of the leiomyoma. The numbers of leiomyomas were categorized into the following groups: one dominant, two to five, and greater than five. A research nurse calculated the uterine and leiomyoma volumes on the basis of measures documented in the radiologic report. Leiomyoma size and location were confirmed by one of the authors (J.B.S., S.A.A., or R.C.J.). Follow-up MR imaging studies were reviewed in the same fashion, usually with initial images available for comparison.
Follow-up imaging was performed in 174 patients at 3 months and 116 patients at 12 months after treatment. The uterine volume and dominant leiomyoma volume reductions from baseline values at each follow-up interval were calculated.
Bilateral embolization was performed in each case. Polyvinyl alcohol particles (500–710 µm) (Contour, Boston Scientific, Boston, Mass; Ivalon, Cook, Bloomington, Ind; and Trufill, Cordis, Miami, Fla) were embolized into each uterine artery until the leiomyoma vasculature was occluded and there was slow flow or near stasis in the main uterine artery. No supplemental embolic agents were used. Uterine artery catheterization was performed by using 5-F catheters (Glidecatheters; Boston Scientific, Natick, Mass) in most cases. For small vessels or when there was flow-limiting spasm, microcatheters (Fastracker 325; Target Therapeutics, Fremont, Calif) were used. Antispasm medications such as nitroglycerin were rarely used. No ovarian or other pelvic branches were embolized.
Before models were fit to the data, preliminary scatter plots of the outcome variables against each predictor variable were used to assess the nature of the associations and to detect extreme values that may have had undue influence in the model fitting. The predictor variables were those that were analyzed to determine if they affected outcome and included the demographic variables, baseline uterine volume, dominant leiomyoma volume and location, and number of leiomyomas. Outcome variables were the outcome measures of interest in any given analysis, such as volume reduction and symptom change. Scatter plots of predictor variables against other predictor variables were used to assess associations that could have led to potential confounding. After the fitting of each model, model assumptions were evaluated by means of graphical analysis of residuals, and in the case of the logistic regression models, model fit was assessed by using the Hosmer Lemeshow goodness of fit test. Stepwise model building was used to help identify important baseline variables to be included in the models. Cases were excluded from a particular model only when they were missing data for a variable included in that model.
To assess whether the observed associations between the predictors and outcome variables were more than would be expected by random chance, we performed a hypothesis test of the null hypothesis of no change. In the context of the linear regression model, no change corresponds to a change equal to 0, whereas in the context of the logistic regression model, no change corresponds to an odds ratio of 1. The results of the hypothesis tests are presented as P values, and a value of less than or equal to .05 indicates a significant finding. In addition, estimates were calculated as 95% CIs.
For purposes of these analyses, symptom change in menstrual bleeding and bulk-related symptoms, as well as patient satisfaction with symptom change, was assessed by using questionnaires completed by patients at 3 and 12 months after treatment. The results were dichotomized for symptoms as improved or not improved and for satisfaction as satisfied or not satisfied. To evaluate the associations between self-reported symptom change and satisfaction, logistic regression techniques were used.
The continuous outcome variables were the volume reductions of the uterus and of the dominant leiomyoma at each follow-up interval, and the associations of these variables were assessed by using linear regression techniques. These models estimated the amount of change in the response variable that would be expected between subjects that differed in the predictor variable by one unit. In the volume measurements, one unit equaled 1 cm3.
The predictor variables measured on a nominal scale, such as dominant leiomyoma location, were entered into the models by using indicator variables. An indicator variable is a variable of either 1 or 0. One of the nominal variables is arbitrarily chosen as the reference level with which all the other nominal variables in the analysis are compared. A separate indicator variable is introduced for each level of the nominal variable except the reference level, and the model effectively compares each level of the variable with the reference level.
When only one predictor variable was included in a model, the estimated change in the response variable for a 1-cm3 change in the predictor variable was without regard to the values of other potentially confounding variables. For example, the change in dominant fibroid volume associated with higher baseline uterine volumes may have been confounded if it had been found that patients with large baseline uterine volumes tended also to have larger baseline dominant leiomyoma volumes. Therefore, we might consider a model that includes both baseline volume measurements. In this way, estimates of the effect of baseline uterine volume can be interpreted as the amount of change in percentage dominant leiomyoma volume reduction associated with a one-unit change in baseline uterine volume, when all other variables in the model are held constant. By making this adjustment for potentially confounding predictors, we are effectively making comparisons within groups of patients who are similar with respect to the other variables in the model. This adjustment was made only in those models that detected potentially confounding variables. These were the models that evaluated dominant leiomyoma change at 3 and 12 months and the models that evaluated symptom change at 3 months. No confounding variables were found in the 12-month symptom change models.
To assess the relationship between symptom and satisfaction outcomes and the relationship between symptom outcome and uterine and leiomyoma volume changes, Pearson product moment correlation coefficients were determined for outcomes measured on a continuous scale and Kendal correlation coefficients were determined for outcomes measured on an ordinal scale. Like the Pearson product moment correlation coefficient, the Kendal is always between -1 and 1, with 0 representing no monotonic association.
| RESULTS |
|---|
| Top ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES |
|---|
|
|
|
|
|
At 3 months after therapy, patient satisfaction was associated with prior gonadotropin-releasing hormone agonist use only (a negative association: odds ratio, 0.24; P = .02), but this association disappeared by 12 months (odds ratio, 0.55; P = .70). Leiomyoma location was the only variable with a moderate association with satisfaction at the 12-month interval: Satisfaction was more likely with intramural leiomyoma (odds ratio, 3.16; P = .05), and there was a trend toward satisfaction with a submucosal location (odds ratio, 7.23; P = .07).
Pearson product moment correlation coefficients for the relationship between volume reduction and dominant leiomyoma and uterine volume were 0.50 (P < .001) at 3 months and 0.51 at 12 months (P < .001), indicating a moderate to strong association between the two variables. The results of Kendal correlation analyses are presented in Table 6. There were very significant associations between bleeding and bulk-related symptom (ie, pain) improvements and symptom outcome and between these symptom improvements and satisfaction outcome. There was weak association between symptom or satisfaction outcome and either absolute uterine or dominant leiomyoma volume or percentage volume reduction at each follow-up interval. Bleeding improvement was weakly negatively associated ( = -0.11; P = .05) with dominant leiomyoma volume reduction, with greater absolute volume change (in cm3) negatively associated with bleeding improvement (P = .05). On the other hand, in the second analysis, there was a weak positive association ( = 0.16; P = .01) at 3 months between percentage dominant leiomyoma volume reduction and bleeding improvement. At percentage volume reduction analysis, there also were weak positive associations with satisfaction at 3 months ( = .17; P = .004) and bleeding improvement at 12 months ( = .21; P = .03).
|
| DISCUSSION |
|---|
| Top ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES |
|---|
In our analysis, we did not find any demographic characteristics that influenced leiomyoma volume change. Larger baseline dominant leiomyoma volume predicted less volume reduction at both 3 and 12 months after therapy, and submucosal leiomyomas were more likely to shrink in the short term, even after adjustment for leiomyoma volume.
Bleeding outcome also demonstrated a trend toward improvement with submucosal leiomyoma location and smaller baseline uterine and leiomyoma volumes. Bulk-related symptoms were not associated with baseline leiomyoma volume or location. There was poorer bulk-related symptom outcome with prior oral contraceptive use and prior births, but the physiologic basis of these associations—if they are real—is unclear. Since these associations were not present at 12 months, we doubt their clinical importance. Similarly, patient satisfaction was negatively associated with prior gonadotropin-releasing hormone agonist therapy at 3 months but not at 12 months. Again, we doubt the existence of an important association when that association is neither strong nor persistent beyond 3 months.
One point of possible confusion in our analysis is related to the findings of the Kendal correlation analysis of the relationships of absolute dominant leiomyoma volume reduction, percentage leiomyoma change, and outcome. There was a weak negative association between absolute volume change and bleeding improvement, whereas there was a weak positive association between percentage volume reduction and bleeding improvement. The reason for these seemingly contradictory findings might be determined by reviewing the perspectives of the two analyses. In the first analysis, a greater absolute volume change (in cm3) favored a larger rather than smaller leiomyoma, because even with a smaller percentage change, a larger leiomyoma often will have a substantially greater absolute change in volume (in cm3) than will a small leiomyoma.
Results of the earlier analysis (Table 4) indicate that larger baseline leiomyoma volume negatively affects outcome; thus, the negative association may reflect that finding. At assessment of percentage volume change, fibroids of all sizes were on an equal footing in the model, and as a result, a positive association was detected. Alternatively, leiomyoma location also might have influenced this analysis, given that bleeding symptoms are often associated with a submucosal location and even small submucosal leiomyomas can cause severe bleeding. A small submucosal leiomyoma might have a small change in volume that equates to a large percentage change. Regardless of the reason for this apparent discrepancy, the associations of volume and percentage reduction with symptom outcome were weak.
Burn et al (10) assessed imaging characteristics and percentage leiomyoma volume reduction. However, their study focused primarily on the signal intensity characteristics of leiomyoma and their relationship to percentage leiomyoma volume reduction. They noted that increased signal intensity on T1-weighted images was associated with a poor leiomyoma volume percentage reduction and increased signal intensity on T2-weighted images was associated with a greater percentage reduction. Their study included 18 patients, and no association between baseline leiomyoma volume and outcome was detected. Jha et al (11) also focused on the imaging characteristics both before and after therapy. In a group of 31 patients, a submucosal leiomyoma location strongly predicted outcome: There was a 30%–40% greater volume reduction compared with the reduction associated with an intramural or subserosal location. It should be noted that the patients included in the analysis of Jha et al (11) were among the 200 patients examined in the current study.
With the larger sample size in the current study, we found that submucosal leiomyoma location was associated with a greater percentage dominant leiomyoma reduction at 3 months after therapy, but this association was much weaker (13.4% greater change; P = .04 adjusted) and this difference did not persist at 1 year, although this may have been because of the smaller sample size in this analysis at 12 months (n = 102 vs n = 159 at 3 months). Jha et al (11) did not find any correlation between percentage volume reduction of the dominant leiomyoma and clinical outcome. Similarly, the baseline volume of the dominant leiomyoma did not predict bleeding or bulk-related symptom outcome. Finally, Jha et al observed a negative association between volume reduction and increasing age.
As part of an assessment of a group of 60 patients, Goodwin et al (8) analyzed some variables in relation to hysterectomy and clinical failure. They noted that patients with a smaller percentage reduction in uterine volume were more likely to undergo hysterectomy. Those patients who had undergone a previous myomectomy were less likely to undergo hysterectomy than those who had not undergone previous myomectomy. Goodwin et al (8) also observed that younger patients were more likely to have clinical failures, a finding that is opposite of that observed by Jha et al (11). We found no association between age and outcome.
McLucas et al (13), in assessing some of the variables associated with UAE, compared a series of patients separated into failure and success groups. Although the details of their analysis are limited, these investigators identified only baseline leiomyoma diameter greater than 8.5 cm as a predictor of failure (P = .05). However, it is not clear what proportion of patients in this group had poor outcome compared with the larger patient group and how this diameter was determined to be significant. The results of that study did not show any association between age, parity, technical details of the procedure, or baseline uterine volume and outcome.
When assessing the studies just mentioned, including our own, we observed that some of the reported results conflict with other reported data. There are several potential reasons for these disparities, and these reasons reflect the limitations of regression analysis in this setting.
The type of analysis described herein is exploratory. With multiple comparisons, it is likely that some of the identified statistically significant findings may be due to chance alone. The rate of this type of error, type 1, for any single analysis is 5%, but the potential for this type of error increases as the number of analyses increases. On the other hand, the possibility of a type 2 error—that of detecting no difference when one exists—is unknown and may be statistically significant. A power analysis to determine the sample size necessary to detect a difference is difficult to perform in exploratory studies because this type of analysis requires an estimate of the degree of the effect of each variable on the outcome before the study is started. Since the purpose of our analysis was to estimate the effect of each variable, a priori power analysis was not possible.
Another potential hindrance to the interpretation of findings was missing data. Although there were few missing data elements at baseline in our study, more missing elements became apparent as the study progressed. This was particularly true with regard to imaging follow-up: Just over 50% of patients underwent follow-up imaging studies 1 year after therapy. Similar limitations have existed in other studies (8,10,11,13). If missing measurements are assumed to occur at random, then they merely reduce the power of the analysis to enable the detection of differences. If, however, missing elements are more likely to occur in those patients with better or worse results than those in the group as a whole, then bias in the form of differential misclassification that can lead to erroneous conclusion enters the analysis.
Confounding variables are another source of potential error in analysis. For example, both leiomyoma location and baseline volume may influence outcome, and drawing conclusions about one without controlling for the other might lead to error in the estimate of the effect of the variable. In those analyses in which clinically important variables might have confounded the estimates, we controlled for these and reported adjusted estimates.
Perhaps the most important limitation of the present study, as well as in the investigations of others who have attempted this type of research, was the method used to assess treatment outcome. The accuracy of measurements obtained in imaging studies has not been reported for this setting, and the interobserver variability of uterine leiomyoma measurements obtained before and after embolization is not known. To our knowledge, there has been one published study (14) with results indicating that MR imaging is more reliable than abdominal US for measurements of volume in leiomyomatous uteri, and this disparity increases with increasing uterine volume.
There are similar problems with measuring symptomatic change after UAE and other therapies for leiomyoma. There are currently no disease-specific validated measures of symptom outcome for any leiomyoma therapy in the public domain. Although Spies et al (15) reported on the results of UAE by using a validated symptom and quality-of-life questionnaire with a small number of patients and menorrhagia measures have been developed (16–18), no such measures have been used yet in a large study of UAE outcome. Without valid measures, the accuracy of the symptom change assessment cannot be estimated. For these reasons, the results of any exploratory study of predictors of outcome must be interpreted with caution. The results of this study might best be used as a means of estimating the potential degree of effect of baseline variables in future investigations of UAE.
Bearing these cautions in mind, we can draw several tentative conclusions. Although baseline demographic variables do not appear to predict outcome, it is likely that both baseline dominant leiomyoma size and baseline leiomyoma location may affect volume reduction and symptom control after uterine embolization. Additional study is needed to reach definitive conclusions.
| REFERENCES |
|---|
| Top ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES |
|---|