Literature Sources

The primary sources of literature were six of the most widely used computerized bibliographical databases (search years shown in parentheses): MEDLINE (1975-February 2000), HealthSTAR (1975-February 2000), CINAHL (1983-February 2000), CancerLit (1983-February 2000), the Cochrane Library (Issue 3 1999), and EMBASE (1980-January 2000). Searches of these databases were supplemented by secondary searches that included e-mail subscriptions to the tables of contents of current medical journals and thorough searching of the reference lists of all included articles and review articles.

Search Strategy

We developed the basic search strategy using the National Library of Medicine medical subject headings (MeSH) key word nomenclature developed for MEDLINE. The same strategy was used to search HealthSTAR, CINAHL, and CancerLit. A Duke University Medical Center librarian checked the strategy and assisted with its translation to the key word structure used by EMBASE.

The initial search was performed in MEDLINE (and then duplicated in the other databases) and was limited to articles in English and with human subjects. It employed a previously validated (Dickersin, Scherer, and Lefebvre, 1994) search strategy that identifies three subsets: (a) high specificity for randomized controlled trials (RCTs), using terms such as "randomized"; (b) moderate specificity, using terms such as "blinding"; and (c) low specificity, using terms such as "followup studies." The overwhelming majority of RCTs identified by this initial search examined the use of gonadotropin-releasing hormone (GnRH) agonists, frequently as adjunctive treatment prior to surgery. The remainder of the citations identified were either uncontrolled case series, case series with historical or nonrandomized controls, or, in a few instances, prospective cohort studies. A subsequent search (performed initially in EMBASE and then duplicated in MEDLINE and the other databases) was targeted to surgical interventions and included other, less robust, study designs. Finally, because few of the targeted articles on hysterectomy and fibroids provided data relevant to Question 9 (on the effects of hysterectomy on the aging process), we performed a supplemental search on hysterectomy without limiting studies to those including patients with fibroids.

The basic search strategy, the surgery search strategy, and the hysterectomy search strategy are reproduced in Tables 4-6.

Table

Table 4. Basic fibroids search strategy.

Table

Table 5. Fibroids surgery search strategy.

Table

Table 6. Hysterectomy search strategy.

Screening Criteria

Screening Results

The main searches (the basic search and the surgery search) yielded 1,084 English-language articles. The abstracts of these articles were reviewed against the inclusion/exclusion criteria by five physician investigators. A team of two physicians reviewed each abstract. Abstracts were available for more than three-fourths of the citations; however, when no abstract was available, the title, source, and MeSH words were reviewed. At this stage, articles were included if requested by one member of the review team.

Each abstract in the first set to be reviewed (n = 309) was reviewed by four physicians, and a kappa statistic was calculated to determine the strength of the agreement to include or exclude articles and the reason for exclusion. As seen in Table 7, the results showed strong agreement between reviewers, representing consistency in applying the inclusion and exclusion criteria. These kappa values are the highest first-round results observed by the Duke EPC in our experience performing systematic reviews.

Table

Table 7. Results of kappa test.

At the full-text screening stage, each article was independently reviewed by two physicians, who forwarded their decisions to the task order manager for recording and comparison. When necessary, the reviewers were asked to reconcile differences of opinion. Overall, the teams initially disagreed on about 20-25 percent of their decisions, but all disagreements were resolved.

The articles identified by the supplemental hysterectomy search were reviewed by a single reviewer with a special interest in gynecological surgery and the aging process. Eighty-eight of these articles were included in the discussion of Question 9; these articles were added to the project database.

The task order manager coded the records in the bibliographical database at each screening stage. Results are summarized in Tables 8 and 9.

Table

Table 8. Number of records identified in main literature searches, by database.

Table

Table 9. Results of abstract screening and full-text article reviews.

We also performed a review of Issue 3 of the 1999 Cochrane Database of Systematic Reviews. Fifteen citations for "fibroids" were found, three for "myomectomy," and 28 for "hysterectomy." However, only one review specifically addressed hysterectomy or myomectomy for fibroids, and the focus of this review was on preoperative treatment with GnRH agonists (Lethaby, Vollenhoven, and Sowter, 1999). Nineteen trials identified in this review met our inclusion criteria, and all 19 were identified in our online database searches. Two other Cochrane reviews focused on adjunctive treatments to prevent adhesions at the time of infertility surgery (including myomectomy). Although there were four Cochrane reviews of trials of medical therapies for heavy menstrual bleeding, all but one of the included trials explicitly excluded women with fibroids; this trial was identified in our online database search.

During the peer review process, additional articles were suggested by many of the reviewers. All suggested articles went through the same screening process as the original articles and were added to the evidence tables if they met our criteria. Several reviewers also pointed out case reports of particularly severe complications of certain procedures that were excluded by our screening criteria. In reviews of surgical procedures, excluding case reports or small case series clearly leads to omission of these reports. However, for the purposes of this review, we are more concerned with the question "How likely is something to happen?" than the question "Can something happen?" since almost any conceivable complication can occur with any procedure. In addition, it is likely that there is a bias toward both reporting small numbers of complications of newer procedures and publishing those results, since similar complications of more established procedures might be deemed less "reportable." Therefore, we have mentioned case reports of complications where more than one reviewer felt that discussion of these complications was important, but we did not revise our search and screening criteria or systematically review all case reports and small case series that had been previously excluded.

Internal Validity

The criteria for assessing internal validity were as follows:

Randomized allocation to treatment:

• Rationale: By randomly assigning groups to the intervention of interest, other factors that may confound the results are equally distributed between groups (assuming a large enough sample size). This equal distribution minimizes the chances of over- or underestimation of treatment effect based on unequal distribution of confounding factors.
• Operational definition: Criterion met if (and only if) the trial report explicitly stated that treatment was randomly allocated.

If randomized, appropriate randomization methods used:

• Rationale: "Pseudo-randomization" methods may be susceptible to bias, as demonstrated by evidence of unequal distribution of subject characteristics (Schulz, Chalmers, Grimes, et al., 1994) and larger effect sizes compared with studies using more rigorous methods (Schulz, Chalmers, Hayes, et al., 1995). In addition, methods of allocation concealment are also important in preventing bias (e.g., use of preprepared sealed envelopes).
• Operational definition: Criterion met if randomization methods were not susceptible to bias, such as computer-generated numbers in sealed envelopes. Criterion not met by studies that either used methods more prone to bias, such as alternate medical record numbers, or did not describe randomization methods or methods of allocation concealment.

Adequate description of patients and controls:

• Rationale: Patient characteristics that might affect outcomes (such as obesity, prior surgery, medical comorbidities) are likely to differ between two interventions. If these differences are not characterized, then erroneous conclusions may be drawn. For example, comparison of outcomes from a series of laparoscopic appendectomies with those from concurrent open appendectomies found better outcomes with the laparoscopic procedure (Sosa, Sleeman, McKenney, et al., 1993). These differences were not seen when the same group performed a randomized trial, a finding attributable to differential patient selection criteria in the nonrandomized study (Martin, Puente, Sosa, et al., 1995).
• Operational definition: Criterion met if (a) inclusion and exclusion criteria for participation in the study were described or (b) for nonrandomized studies, description of the rationale for selecting a particular intervention was given. Criterion not met if (a) inclusion/exclusion criteria were not described or (b) description of the rationale for selection of the interventions was not given (e.g., a nonrandomized comparison of concurrent laparoscopic and abdominal myomectomies that did not describe why patients received one or the other procedure).

Description of length of followup:

• Rationale: Patient and provider decision-making is dependent on knowledge of the likelihood of benefits and risks over time.
• Operational definition: Criterion met if a summary statistic (mean, median, or minimum) and range of followup were given for studies reporting nonhospital outcomes. Criterion not met if summary statistic and range not given. Criterion not applicable for studies reporting only hospital-based outcomes.

Description of loss to followup:

• Rationale: Failing to account for patients lost to followup may lead to erroneous conclusions, especially if the loss to followup is related to either the underlying disease or the intervention (e.g., patients seeking care elsewhere because of continuing symptoms or unacceptable side effects of treatment).
• Operational definition: Criterion met if (a) loss to followup was explicitly reported, (b) number of subjects for whom data was presented was equal to number of subjects receiving intervention at start of study, or (c) for studies reporting only hospital-based outcomes, number of missing charts or records was reported. Criterion not met if loss to followup was not reported and number of subjects at beginning and end of study was not equal.

Description of dropout rates:

• Rationale: Dropout rates may reflect differences in clinically important variables, such as side effects or treatment response. Failure to account for dropouts may result in erroneous conclusions similar to those seen with failure to account for loss to followup.
• Operational definition: Criterion met if (a) patients dropping out of the study prior to completion were reported or (b) number of subjects at beginning and end of study were equal. Criterion not met if patients dropping out were not reported and numbers of subjects at beginning and end of study were not equal. Criterion not applicable for studies reporting only hospital-based outcomes.

Recognition and description of statistical issues:

• Rationale: Use of inappropriate tests may lead to misleading conclusions. For example, variables such as blood loss, length of stay, or costs are often not normally distributed; use of means instead of medians when data may be affected by outlying observations can be misleading. Many studies, especially case series, may lack sufficient power to detect clinically important differences in outcomes or patient characteristics.
• Operational definition: Criterion met if (a) appropriate statistical tests were used (e.g., nonparametric methods for variables with nonnormal distributions, or survival analysis techniques to account for loss to followup and dropouts) and (b) potential study limitations regarding design and analysis, especially sample size and power issues, were discussed. Criterion not met if (a) inappropriate statistical tests were used or (b) study limitations were not discussed.

External Validity

The criteria for assessing external validity were as follows:

Description of age of study population:

• Rationale: The outcomes of many interventions are affected by patient age. Age is especially important in studies of reproductive disorders in women, since childbearing potential and ovarian hormone production, both key components in decisionmaking regarding management of fibroids, are directly related to age.
• Operational definition: Criterion met if summary statistics of subject age were given. Criterion not met if summary statistics were not given.

Description of racial/ethnic distribution of population:

• Rationale: The epidemiology, and possibly the biology, of fibroids clearly varies between white and black women. Additionally, there is widespread racial variation in the United States in utilization and outcomes of a wide variety of interventions (Fiscella, Franks, Gold, et al., 2000).
• Operational definition: Criterion met if (a) racial/ethnic distribution was described or (b) the geographical setting of the study strongly implied the racial/ethnic background of the entire population (e.g., studies of hysterectomy outcomes in Japan or Nigeria). Criterion not met if (a) racial/ethnic distribution was not described and (b) geographic setting was likely to include subjects of diverse racial/ethnic background.

Description of pregnancy history of population:

• Rationale: Pregnancy history may affect the natural history or biology of fibroids (Parazzini, Negri, La Vecchia, et al., 1996). For surgical interventions, pregnancy history may affect the technical difficulty of a procedure; for example, prior vaginal delivery may facilitate vaginal hysterectomy, while prior cesarean section, by increasing the risk of adhesions, may make either abdominal or vaginal hysterectomy more difficult.
• Operational definition: Criterion met if (a) summary statistics on gravidity or parity were given or (b) percentage of women with prior pregnancy was given. Criterion not met if (a) no summary statistics were given and (b) no distribution data on prior pregnancies were given.

Description of prior surgery:

• Rationale: A history of prior surgery for fibroids might reflect differences in the natural history or biology between patients. Additionally, previous abdominal surgery might increase the risk of complications by increasing the likelihood of intraperitoneal adhesions.
• Operational definition: Criterion met if (a) any description of history of intra-abdominal surgery was given or (b) proportion of women with prior surgery for fibroids was given. Criterion not met if no description of prior surgery was given.

Adequate characterization of fibroid and/or uterine size:

• Rationale: Individual fibroid size, or aggregate uterine size, may affect the nature or severity of symptoms, the response to various treatments, and the risk of complications of surgical treatments.
• Operational definition: Criterion met if data given on (a) uterine size in weeks gestational age; (b) uterine volume, area, or length as estimated by radiologic techniques; (c) uterine weight in grams (for hysterectomy specimens); (d) fibroid diameter or volume as estimated by radiologic techniques; or (e) fibroid dimensions or weight based on pathological examinations. Criterion not met if none of the above were provided.

Adequate characterization of fibroid number:

• Rationale: The number of fibroids may affect the nature or severity of symptoms, the response to various treatments, and the risk of complications of surgical treatments.
• Operational definition: Criterion met if summary statistics or distribution of number of fibroids was provided. Criterion not met if no data were provided on number of fibroids.

Adequate characterization of fibroid location:

• Rationale: The location of fibroids may affect the nature or severity of symptoms, the response to various treatments, and the risk of complications of surgical treatments.
• Operational definition: Criterion met if (a) distribution of fibroids by location (subserosal, intramural, submucosal, or pedunculated) was given or (b) other anatomical descriptions were given (e.g., anterior, posterior, fundal, or within the broad ligament). Criterion not met if no anatomical description was given.

Adequate characterization of baseline symptoms:

• Rationale: Because fibroids may present with a variety of symptoms, assessing the effectiveness of therapy requires an adequate description of the nature and severity of symptoms prior to institution of therapy.
• Operational definition: Criterion met if distribution of specific symptoms or symptom classes associated with fibroids were provided. Criterion not met if specific symptoms were not described (e.g., if the only description of inclusion criteria was "symptomatic fibroids").

Adequate description of timing of outcome measurement:

• Rationale: Outcome measures may vary depending on when they are obtained. Description of when outcomes were measured facilitates comparison between studies.
• Operational definition: Criterion met if (a) time after initiation of therapy at which outcomes were measured was reported or (b) study was limited to hospital-based outcomes. Criterion is not met if (a) time was not reported and (b) study was not strictly hospital-based.

Adequate description of methods used for outcome measurement:

• Rationale: Comparison between studies requires common methods of measurement, which in turn requires adequate description of the methods used to assess comparability.
• Operational definition: Criterion met if (a) methods used to measure outcomes were adequately described or referenced (e.g., pain or bleeding scales), (b) definitions were given (e.g., description of outcomes classified as "complications"), or (c) outcomes were unambiguous (e.g., pregnancy, need for hysterectomy). Criterion not met if (a), (b), or (c) was not present.

Adequate description of validity and reliability of outcome measurement:

• Rationale: Measurements of outcomes are only useful if changes in the outcome being measured are reflected in changes in the measurement (validity) and if these changes are reasonably consistent between the same observer measuring at different times or between different observers (reliability). For example, changes in a scale to assess menstrual blood flow should correlate with some other physiological measure of menstrual blood loss, and this correlation should be consistent when different women apply the same scale.
• Operational definition: Criterion met if (a) a description of the methods used to assess validity and reliability of at least one outcome measure was provided, (b) a reference to another article documenting validity and reliability was provided, or (c) only unambiguous outcomes such as pregnancy were included. Criterion not met if (a), (b), or (c) was not present.

Adequate description of clinical care provided to subjects:

• Rationale: The ability to replicate study results is dependent on adequate description of methods. Additionally, readers should be aware of aspects of clinical care that might influence outcomes.
• Operational definition: Criterion met if (a) a detailed description of the therapy (dose, dosing schedule, and route of administration for medications and/or techniques for invasive therapies) was provided; (b) a reference to another publication describing the procedure was provided; or (c) statistical adjustment was made for likely sources of variation in clinical care (e.g., site where care was given, type of specialist providing care, individual provider). Criterion not met if (a), (b), or (c) was not provided.

Use of previously validated and standardized measures:

• Rationale: Use of measures used by other researchers enhances the ability to compare results across studies. Use of measures used with other medical conditions enhances the ability to compare the impact of uterine fibroids to that of other common conditions, which may be important when setting research and resource allocation priorities.
• Operational definition: Criterion met if at least one measure previously used by another group was used. Criterion not met if all measures were internally developed.

Reporting of Quality Scores

These criteria were not aggregated into an overall quality score but were considered and reported individually. This approach was preferred for several reasons:

• Previous work has shown that numeric grading systems may not discriminate well between high- and low-quality studies, even randomized trials (Juni, Witschi, Bloch, et al., 1999; Moher, Jadad, and Tugwell, 1996).
• Development and use of a new quality score would require additional work for validation.
• Identification of specific weaknesses in each study will be helpful in identifying trends, which in turn will assist with recommendations for future research.

Our approach of describing key study design components, rather than assigning a single aggregate score, is also consistent with recent recommendations from an expert panel on meta-analysis of observational studies (Stroup, Berlin, Morton, et al., 2000).

Summaries of each evaluation are provided in the evidence table entry for each abstracted article. A "+" indicates that a given criterion was met, a "−" denotes that the criterion was not met. Grades were assigned by the primary reviewer and confirmed by the over-reader.

Nationwide Inpatient Sample

The NIS contains administrative discharge data from more than 1000 hospitals in 22 States, representing a stratified sample of 20 percent of U.S. hospitals (NIS, 1997). Weights are provided to permit estimation of national data based on this sample. We used this data set to provide supplemental data on frequency and resource utilization for hysterectomy and myomectomy by age and race (Question 7 on cost and Question 8 on age and racial variation). Because previous work has shown that administrative data may lack sufficient clinical detail to compare outcomes (Myers and Steege, 1999), we did not attempt to directly compare complication rates between these procedures (Question 1).

Hysterectomies for fibroids were identified by searching the 1997 NIS data set for all patients with International Classification of Diseases, Revision 9 (ICD-9) procedure codes of 68.3, 68.4, and 68.5 (all nonradical hysterectomies) who also had diagnosis codes of 218.0, 218.1, 218.2, and 218.9 (submucous, intramural, subserous, and unspecified fibroids).

Myomectomies were identified by searching the 1997 NIS data set for all patients with ICD-9 procedure codes of 68.29 (uterine lesion destruction) who also had diagnosis codes of 218.0, 218.1, 218.2, and 218.9. We used these codes to identify myomectomies at Duke after discussion with hospital coders; only three nonmyomectomy cases were identified using this strategy. Further validation of this coding strategy, especially trying to estimate the proportion of myomectomies not identified, would be useful for future health services research.

Data from the sampled hospitals were converted to national estimates using the weighting variables provided by AHRQ. To assess differences in the proportion of white and black women undergoing each procedure, logistic regression was performed using myomectomy versus hysterectomy as the dependent variable, race as the main independent variable, and controlling for age, payer type, and median income.

Duke University Medical Center Data

Data on hospital costs and clinical characteristics affecting cost and complications of patients undergoing myomectomy were obtained from Duke University Medical Center. Subsequent analyses will include additional chart data on patients undergoing hysterectomy for fibroids. This data source was used to provide additional information for Question 1 (risks and benefits of hysterectomy and myomectomy), Question 7 (costs of interventions), and Question 8 (racial differences in outcomes).

Meta-analysis

Decision Analysis

Decision-analytic models can be a useful means of synthesizing information from a variety of sources, and they have been used by the Duke group in studies of stroke (Matchar, Samsa, Matthews, et al., 1997) and cervical cytology (McCrory, Matchar, Bastian, et al., 1998). We had planned on using a decision model to synthesize the disparate data on fibroid management to help answer the key questions or at least to identify key areas of uncertainty to direct future research. We were able to use decision modeling for some purposes (e.g., estimating cumulative incidence of fibroids based on reported age-specific rates). However, the methodological limitations of the available literature precluded us from using decision analysis as a formal synthesis tool. In particular, heterogeneity of methods for measuring and reporting patient outcomes and for reporting clinically important preintervention patient characteristics limited our ability to perform even preliminary decision analyses.

Developing a decision model in parallel with our literature review did enable us to consider the nature of the data and the potentially important parameters that would be needed to use such a model to assist patients, clinicians, or policymakers. Therefore, we have provided a detailed description of the framework for such a model, along with some potential examples of its use, in Chapter 4.

Evidence Tables

The evidence tables summarize each study meeting the inclusion criteria described above; as such, they represent studies reporting on outcomes of prophylactic or symptomatic treatments for fibroids. Articles cited in the text but not in the evidence tables include (1) review articles; (2) articles addressing background issues such as basic biology, epidemiology, or health services research; and (3) articles either describing methods or illustrating relevant methodological issues.

Many of the studies included in the review of the effects of hysterectomy on the aging process (Question 9) also were not abstracted in evidence tables. Because our initial searches focused on patients with fibroids, many of the articles discussed under Question 9 did not meet our original inclusion criteria. Late in the review process, after it became apparent that the vast majority of studies addressing this question would be excluded, we revisited the question and performed a revised search. Because a large proportion of women, especially premenopausal women, who undergo hysterectomy do so because of symptomatic fibroids, we assumed that these articles would have some validity for women with fibroids. As mentioned above, the review of these articles did not follow the same strict methodology used in the review of articles from the basic search and the surgery search. For that reason, and because they did not explicitly focus on women with fibroids, these articles were not abstracted into the evidence tables.

Because formal data synthesis techniques were not applicable to the majority of the literature on management of fibroids, we elected to summarize the results of our review without attempting quantification. We did attempt to identify qualitatively similar findings that could be suggestive of consistent results. We also tried to identify the characteristics of data that would be needed to address the question adequately and, in some cases, to identify conceptual issues that should be considered by patients and providers given the lack of data (issues surrounding prophylactic surgery for asymptomatic fibroids, for example).