Age

Patients enrolled in most neurosurgical clinical studies are described by age, but this is particularly essential in brain tumor reports because age is such an important prognostic factor for most types of brain tumor. For malignant and low-grade glioma, for example, older patients have much shorter survival for reasons that are still not entirely clear.¹⁰ Older patients with malignant glioma are less likely than younger patients to have known low-grade tumors preceding their diagnosis of malignancy, more likely to have PTEN (phosphatase and tensin homologue from chromosome 10) abnormalities, and less likely to have p53 mutations.¹¹ Pediatric malignant glioma patients have longer median survival than do young adults.¹² Conversely, in medulloblastoma, patients younger than 2 years have shorter survival, partially because of reluctance to subject such young children to aggressive central nervous system radiotherapy, but also probably because of more biologically aggressive disease.^13,14 For both acoustic neuromas and hemangioblastomas, younger patients with apparently isolated sporadic tumors are much more likely to have occult neurofibromatosis type 2 and von Hippel-Lindau disease, respectively, which potentially affects long-term outcome.^15,16 Age at diagnosis or initiation of treatment and age limitations on eligibility for prospective studies should be carefully documented in all brain tumor reports.

Histology

In 1932, Harvey Cushing announced to a hushed crowd that he would never again attempt to report on the subject of brain tumor outcomes as a whole.¹⁷ Almost all subsequent brain tumor outcome studies have followed Cushing’s example in being limited to a single tumor histology and most to a single point in the course of the disease, such as at diagnosis or first recurrence. This seems straightforward enough, except that histologic definitions have changed over time, some diagnoses are highly inconsistent from pathologist to pathologist, and some are dependent on the amount of tissue sampled at surgery (or when diagnoses are based on imaging alone).

The present standard reference for describing the histology of brain tumors is the 2007 histologic classification of the World Health Organization (WHO).^18–20 In interpreting older studies, temporal changes in grading systems should be kept in mind. For glioblastoma, newer grading systems no longer mandate the presence of tissue necrosis, which was formerly required to make the diagnosis.^18,21,22 This change should have little effect on most studies because the majority of glioblastomas still contain necrosis and prognoses are similar for patients whose glioblastomas do or do not contain necrosis.²³ However, for studies that focus specifically on long-term glioblastoma survivors, the evolution of neuropathologic criteria constantly forces a re-evaluation of earlier research. This is so because some patients in whom glioblastoma was originally diagnosed are now recognized to have other tumors by current criteria, such as some anaplastic oligodendrogliomas that contain vascular proliferation or benign tumors whose histopathologic appearance formerly qualified for a diagnosis of glioblastoma. For example, increased awareness of the relatively new diagnosis of pleomorphic xanthoastrocytoma required removal of some patients from one long-term glioblastoma survivor series,²⁴ and recognition of malignant oligodendroglioma elements caused another group to reclassify a third of their long-term glioblastoma survivors.²⁵ Only 17% of long-term glioblastoma survivors in Sweden from 1958 to 1977 were not reclassified on later re-examination.²⁶

At the other end of the glioma spectrum, the identification of pilomyxoid astrocytomas in 1999 caused some poor-prognosis patients to be reclassified from earlier pilocytic astrocytoma series,^27,28 just as increasing recognition of atypical teratoma-rhabdoid tumors has removed some of the worst-prognosis patients from modern medulloblastoma series.^29,30 Medulloblastoma series are also variably affected by categorization into subdiagnoses, such as desmoplastic, anaplastic, and nodular medulloblastoma.¹⁹

For lower grade gliomas, changes in formal grading systems rather than the development of new diagnoses have been more significant. Although the classic histologic criteria for defining oligodendroglial portions of a tumor have generally been established, the subjective nature of their application (in the absence of an immunohistochemical marker), in the context of intense clinical interest in oligodendroglial sensitivity to chemotherapy,³¹ has resulted in important epidemiologic shifts in the frequency of diagnosing oligodendroglioma in the past 2 decades, with a remarkable increase in the proportion of low-grade tumors now thought to contain oligodendroglial elements.³² The most recent WHO revision has introduced a new grade IV diagnosis of “glioblastoma with an oligodendroglioma component” for many tumors that were previously diagnosed as grade III anaplastic oligodendrogliomas.^18,19 The impact of these changes on outcome studies is complex and difficult to predict. Revisions in the WHO definition of atypical meningioma have also been shown to have a concrete impact on the recommended treatment of many patients at certain centers.^33–35

Interpathologist disagreement on brain tumor diagnoses is sufficiently common to require central review for many studies. Discordant results on central review are more common with diagnoses originating at community hospitals, especially those without a neuropathologist, and affect some diagnoses more than others.^36–38 For example, in the San Francisco Bay area, Aldape and coauthors reported more than 50% discordant results on central review of tumors diagnosed as containing an oligodendroglial component, thus indicating that the distinction between oligodendroglial and astrocytic histology is an especially subjective one.³⁸ The diagnosis of glioblastoma in this study was much more reliable. For studies using data sources that include many community hospital diagnoses, such as population-based case registries, it has been recommended that all diagnoses except glioblastoma and meningioma be subjected to central pathology review whenever possible.³⁶ Most multicenter randomized studies of brain tumor treatment now require central pathology review before registration.

An important consideration in studies on glioma surgery is that the pathologic diagnosis may vary with the extent of tumor resection because of the combination of sampling bias and the high degree of spatial pathologic heterogeneity in many gliomas. Glioma grading systems are based on counting individual malignant features that may be seen in successive microscopic fields examined, so the idea that more tissue examined leads to a more malignant diagnosis makes intrinsic sense. This has been demonstrated in clinical studies comparing rates of high-grade diagnoses in patient cohorts that differ in the extent of resection,³⁹ as well as in studies comparing needle biopsy diagnoses with later resections, especially in larger tumors.^40–42 Similarly, within glioblastoma, detection of necrosis is more common with more extensive resection.²³ Less widely recognized is the tendency for more extensive resections to produce more histologic diagnoses that include an oligodendroglial component. Perry and colleagues described a cohort of grade 3 gliomas in which the percentage containing an oligodendroglial component increased from 3% in biopsied tumors to 29% in subtotal resection specimens and 43% in gross total resections.⁴³ Similar results have been described for grade 2 gliomas, both in a single-institution study⁴⁴ (46% oligodendroglial-containing tumors in biopsy specimens, 89% in gross total resections) and in a population-based study⁴⁵ (32% oligodendroglial-containing tumors in biopsy specimens, 62% in resection specimens). Because of sampling bias, when one group in a treatment comparison selectively contains patients with less extensive surgical resection, the pathologic diagnoses in that group will be biased toward lower grade histology and less common oligodendroglial diagnoses than will the true pathology.

Studies in which some or all patients are enrolled on the basis of imaging diagnoses alone, without histologic confirmation, face special problems in reliably defining diagnostic categories. Radiosurgery series and series of tumors undergoing observation without other treatment are particularly dependent on accurate imaging diagnoses. Randomized clinical trials (RCTs) in which patients are registered before surgery provide some insight into the error rates of imaging diagnoses. For example, in three RCTs of patients undergoing resection for single metastases, all of whom had single brain lesions and established diagnoses of systemic cancer, 7 of 127 presumed metastases (6%) proved to be other lesions (gliomas, abscesses, or inflammatory changes)^46–48; in a phase II trial involving implantation of carmustine (BCNU)-impregnated wafers for single metastases with presurgical enrollment that did not require a previous history of cancer, 6 of 35 patients (17%) had pathology other than metastasis.⁴⁹ These results are relevant to radiosurgical series for single brain metastases. Nonenhancing cerebral masses are usually thought to represent low-grade gliomas, but up to a third will be malignant gliomas, especially in older patients,⁵⁰ and many low-grade gliomas enhance on magnetic resonance imaging (MRI), a feature that characteristically portends a poorer prognosis.^51,52 An observational study of “low-grade gliomas” that is limited to nonenhancing tumors will therefore include some malignant gliomas and exclude some poor-prognosis low-grade tumors, thus making the combined effect of the two biases difficult to predict. Treatment of malignant gliomas without a histologic diagnosis is less common, but knowing the rate of misdiagnosis based on imaging can assist in power calculations for trials. Of 30 patients enrolled in an RCT of glioblastoma resection based on characteristic images,⁵³ 7 (23%) had nonglioma lesions and another 4 had grade 3 gliomas. In a larger study that required an MRI diagnosis of contrast-enhancing glioma amenable to complete resection for enrollment, 34 of 322 patients (11%) had nonglioma pathology; 97% of the remaining tumors were grade 4 astrocytomas.⁵⁴

Defining the stage of disease (i.e., at diagnosis versus recurrent) is generally straightforward. However, care should be exercised in defining the date of diagnosis in glioblastoma studies that include (or exclude⁵⁵) patients with previous known low-grade tumors (“secondary glioblastomas”). Defining the date of recurrence in treated malignant gliomas is discussed later in the chapter.

Composite Baseline Risk Scores—Recursive Partitioning Analysis and Others

Several specially developed composite baseline risk classifications have found widespread use in reporting treatment of glioma, metastasis, and medulloblastoma. Baseline prognostic classifications or scores are useful in constructing historical control groups for case series without internal controls and for stratifying randomization in RCTs so that both treatment groups will contain patients with comparable prognoses given standard treatment.

For malignant glioma patients, the 1993 Radiation Therapy and Oncology Group (RTOG) recursive partitioning analysis (RPA)⁵⁶ is the most commonly used prognostic classification. This scheme uses patient age, functional status (Karnofsky performance scale [KPS] and work ability), mental status and neurological function, length of symptoms, type of surgery, radiation dose delivered, and tumor histology to define six patient groups with distinctly different survival prognoses. The U.K. Medical Research Council developed and validated a similar six-group classification of malignant glioma patients that has been less used widely,^57,58 and a glioblastoma-specific RPA has also been published.⁵⁹ For patients with low-grade glioma, Pignatti and coworkers identified five baseline prognostic factors (age >40, astrocytoma histology, tumor diameter >6 cm, tumor crossing the midline, and presence of a neurological deficit before surgery) and split patients into two prognostic groups, low risk (two or fewer factors) and high risk (three or more factors).⁶⁰ Brain metastasis patients are classified with another RTOG RPA scheme that uses patient age, presence of metastases outside the brain, and state of control of the primary tumor to define three prognostic groups.^61,62 A 2008 revision of RPA, called graded prognostic assessment (GPA), was developed by the RTOG to be easier to use and less subjective than RPA.⁶³ GPA uses patient age, KPS, number of brain metastases, and the presence of known extracranial metastases to define four prognostic classes, as opposed to three for RPA. The authors found that GPA was as prognostic as RPA and superior to other published prognostic indexes for brain metastasis patients, but GPA has not yet been validated on other data sets. For medulloblastoma, various means of combining baseline risk factors, including age, dissemination, and residual disease after surgery, have been used to construct “average-risk” and “poor-risk” groups.¹⁴

Prognostic indexes are only valid to construct historical control patient sets that share diagnostic, prognostic, and treatment characteristics with the original cohort from which the index was developed, and they should not be used as comparisons for patient cohorts that differ substantially from the original cohort. For example, the RTOG conducted a phase II trial of interferon beta-1a after adjuvant radiation therapy for malignant glioma patients in which eligible patients were those who had stable disease at the completion of radiation treatment.⁶⁴ Initial analyses showed superiority over expected survival with RPA, but the comparison was invalid because patients whose tumors progressed during radiation therapy (a marker for poor prognosis)⁶⁵ were included in the RPA but excluded from the trial (about a third of the original cohort). A matched-pair analysis that accounted for disease stability during radiation therapy showed no difference between survival of interferon-treated patients in the trial and appropriate historical controls. Similarly, malignant glioma trials that stipulate gross total resection for entry (such as many immunotherapy trials) require a special prognostic index constructed from similar patients.⁶⁶ When the standard of care shifts for a disease, as it has recently for malignant glioma because of the introduction of adjuvant temozolomide,⁶⁷ new indexes must be developed.

A basic characteristic of all such prognostic indexes in current use is that they were developed from patients enrolled in prospective clinical trials. In the United States, only 2% to 3% of patients with malignant glioma enter clinical trials (Barker FG, Curry WT, unpublished data). It is well known that trial patients have decidedly favorable prognostic factors in comparison to population-based cohorts because of trial eligibility requirements, self-selection, and social factors.^68–70 Valid historical controls for population-based patient cohorts cannot be generated with the standard trial-based prognostic indexes. Conversely, single-center trial patients usually have even better baseline prognostic factors than do patients in multicenter trials, which partially accounts for the tendency of many therapies that appear promising in early single-center trials to fail in RCTs^70–72; this renders the use of multicenter trial–based prognostic indexes suspect as historical controls for single-center trial reports that use survival end points.

Other Factors

All brain tumor studies should include functional status in baseline descriptions because it is a very strong predictor of both outcome and survival.¹⁰ Commonly used scales include the KPS⁷³ and European Cooperative Oncology Group (ECOG) scale.⁷⁴ For special tumor types, other baseline neurological status needs to be specified. An example is preoperative hearing status for acoustic neuromas, which is usually described in terms of the Gardner-Robertson classification,⁷⁵ a strong predictor of hearing preservation results. This same scale is used for studies of radiosurgical treatment. Many specific neurological measures are more often used to describe complications of surgery (such as facial nerve weakness)⁷⁶ and are described later. Tumor size, location, focality and multiplicity, and proximity to or involvement of eloquent structures all contribute to the concept of “resectability,” an important prognostic factor described subsequently.

In metastasis studies, additional important baseline variables include histology and time since diagnosis of the primary tumor (longer intervals suggest a better outcome),⁷⁷ “expected survival” at enrollment (most surgical trials require a minimum expected survival, although physicians are notoriously bad at such predictions^78–80), and whether the systemic cancer is stable or progressive (a very subjective determination with no known guidelines). Despite their subjective and imprecise nature, all these factors are strong predictors of actual patient survival.

End Points for Cancer Studies

A bewildering variety of end points have been used in cancer outcome studies, many of which are useful in brain tumor reports.^5,100 Careful selection of end points for both efficacy and toxicity is one of the most important aspects of planning a brain tumor trial. Appropriate end points for reports of disease treatment typically shift over time as those treatments improve. For example, in the treatment of acoustic neuroma, operative mortality has been supplemented in turn by complete resection, anatomic facial nerve preservation, good facial function, and hearing preservation as additional important end points, and with the introduction of radiosurgery, tumor stability has been added.¹⁰¹ Because many cancer innovations eventually involve approval of a drug or product by regulatory authorities, the perspective of the U.S. Food and Drug Administration (FDA) on appropriate end points for cancer product approval will be touched on briefly in this chapter as an example.^102–104

Survival, Operative Mortality, and Disease-Specific Mortality

The original criterion for success of brain tumor surgery was operative mortality, and overall patient survival is still the “gold standard” for success of therapy for malignant brain tumors, including low-grade gliomas. A survey of neurosurgeons showed that mortality was considered the most important end point far more often than morbidity or quality of life.¹⁰⁵ Survival (or overall survival) is usually defined as the time from initial surgery, date of diagnosis, or date of trial registration until death. Events (such as loss to follow-up) that will be considered censoring events in survival analysis should be carefully defined.¹⁰⁶ The obvious advantages of overall survival as an end point are its undeniable importance to the patient (within reasonable limits imposed by quality of life) and lack of ambiguity in definition. Governments typically maintain death indices that can be used to obtain follow-up even for patients who have gone missing during the study.^107,108 Survival is an ideal end point for diseases with high acute mortality, such as glioblastoma. In diseases with longer expected survival, trials with survival end points must plan for lengthy follow-up (with an increasing amount of patient dropout over time), and when effective salvage therapies are available, the survival benefit of the initial treatment becomes difficult to distinguish. In these situations, progression-free survival (see later) may be a more attractive end point.

In surgical studies, operative mortality is a short-term end point of importance to many investigators. Cushing, in an era when many patients had prolonged hospitalization after brain tumor surgery, used mortality before hospital discharge as his definition of operative mortality.¹⁷ Later investigators substituted 30-day mortality as the standard measure to capture events occurring soon after hospital discharge (such as death from pulmonary embolism). With the increasing use of administrative databases for brain tumor outcome studies,^109–111 mortality before hospital discharge is again frequently reported by investigators. Use of this end point can introduce bias into analyses. For example, studies showed a substantial decrease in in-hospital mortality for patients undergoing brain tumor craniotomy in the United States from 1988 to 2000.^109–111 However, a Medicare analysis showed that although in-hospital mortality after craniotomy (all types) fell 8.5% between 1998 and 2003, 30-day mortality remained almost unchanged. This strongly suggests that the observed decrease in in-hospital mortality is largely an artifact of decreasing length of hospital stay over the study period.¹¹² In general, when patients with acute conditions remain in the hospital longer (even for administrative reasons), their observed in-hospital mortality will be higher.

In malignant glioma patients, studies show that virtually all deaths result from the primary tumor, so the difference between overall mortality (death from any cause) and disease-specific mortality (death from the brain tumor) is negligible. Although the same holds true for death in patients with neurofibromatosis type 2,¹¹³ for patients who undergo operations for sporadic benign tumors such as acoustic neuromas, meningiomas, or pituitary adenomas, most eventual deaths are not directly related to the brain tumor or its treatment. In fact, there is a bias toward better survival in such patients than in the general population because of preferential selection of healthy patients for surgery, as seen in patients with other benign neurosurgical conditions.^114–116 This bias is related to the “healthy worker” effect in epidemiology—when cases are selected only from a working population, a spurious increase in survival over general population controls is seen.¹¹⁷ In these situations, analysis of overall survival can produce confusing or misleading results. Not surprisingly, patient age at diagnosis is the strongest prognostic factor for survival in studies where few deaths result from the disease under study, such as meningioma.¹¹⁸ Disease-specific survival is the appropriate end point for determining relevant prognostic survival factors in the treatment of benign tumors: in one study of radiation therapy after subtotal resection of benign skull base meningiomas, the 15-year cause-specific survival rate was 92%, but the overall survival rate was just 62%.¹¹⁹

In detailing results, all time-dependent end points, such as survival, completeness of follow-up, and extent of patient dropout, should be reported. Most reports simply state the number of patients lost to follow-up, but when the loss occurred is also important. Better available methods include calculating the maximal theoretically possible follow-up (i.e., all patients monitored from enrollment to the time of analysis) and reporting the actual follow-up achieved as a percentage of the former.¹²⁰ Another method is the “reverse Kaplan-Meier” method in which the indicator for events and censoring is reversed.¹²⁰ Loss to follow-up in survival studies is usually nonrandom. Younger patients are more likely to be lost, as are those who have experienced a complication of therapy or other poor outcome,^114,121,122 so including only patients with complete follow-up can bias a study toward better results.

Progression-Free Survival, Time to Progression, and Time to Treatment Failure

In many brain tumor treatment studies, progression-free survival is an important end point. It typically reflects a period of relative freedom from symptoms for the patient, in addition to being an excellent surrogate for overall survival.^123,124 However, defining tumor progression in brain tumors (especially malignant gliomas) is complex and subjective; definitions of tumor progression and response are discussed later in this chapter. Progression-free survival starts at diagnosis, surgery, or trial enrollment and ends at either tumor progression or death. In contrast, time to progression ends only when tumor progression is actually observed (typically with an imaging study). The difference is important because many patients fail to keep clinical or imaging appointments when their tumors recur, so tumor progression is never formally demonstrated before death. Such patients are censored observations in time-to-progression analyses, which limits the usefulness of this end point. Similarly, “time to treatment failure” ends either at death, at tumor progression, or when treatment is stopped for any other reason, such as toxicity or patient choice. This is only rarely a useful end point in brain tumor trials because efficacy and toxicity are generally better assessed separately.

Other Time-Dependent End Points

Besides the survival measures already discussed, the concepts of independent or functional survival, time to neurological progression, and neurological death are important in some brain tumor trials. Definitions of functional or independent survival and time to neurological progression are highly dependent on the functional status scale and definition of neurological symptoms used; these scales are discussed later. Neurological death is an end point used in some brain metastasis trials.^46,125 In surgical or radiosurgical trials, it reflects cause-specific mortality for the treated lesion or lesions or from other, new brain metastases.

Quality of Life, Functional Status, and Measures of Symptoms

In recent years there has been increasing emphasis on patient-reported outcomes (PROs) in cancer therapy in addition to length of survival.^{105,126–133} PROs include, in order of increasing specificity, health-related quality of life (HRQoL), functional status, and measures of specific neurological deficits. In the WHO framework for discussing health-related problems, decreases in overall HRQoL correspond roughly to “handicap,” deficits in functional status to “disability,” and specific symptoms to “impairment.” In the United States, the increasing value placed on PROs is reflected by their acceptance by the FDA as supporting evidence in the approval process for a growing number of anticancer drugs.¹²⁸

Methods for measuring HRQoL are still in the course of development, and specialized texts are available.^5,100,134 HRQoL is a measure of an individual’s satisfaction with life insofar as it is affected by health. It excludes many domains of “quality of life” defined generally, such as financial security, political and religious freedom, quality of interpersonal relationships, and so on, except as they are affected by the person’s health status. Generic HRQoL measures are applicable to all individuals regardless of health state; such measures include familiar instruments such as the short form health survey SF-36 (www.sf-36.org). Other measures are designed to capture HRQoL in cancer patients generally, such as the Functional Assessment of Cancer Therapy—General version (FACT-G; www.facit.org) and the European Organization for Research and Treatment of Cancer (EORTC) Quality of Life Questionnaire (QLQ) C30 (www.eortc.be/home/qol/ExplQLQ-C30_1.htm). The highest level of specificity is achieved by HRQoL measures designed specifically for brain tumor patients, such as the brain tumor subscale of the FACT-G, called the FACT-BR,¹³⁵ and the EORTC QLQ Brain Cancer Module (BN 20).^136,137 HRQoL and cognitive functioning are strong predictors of survival in glioma patients, both at diagnosis^136–139 and at recurrence.¹⁴⁰ Most large trials of brain metastasis treatment now report HRQoL or cognitive function measures, or both,¹²⁶ and HRQoL reports in other tumor types such as meningioma,^141–143 acoustic neuroma,^131,133,144 and some types of pituitary adenoma^145,146 are becoming increasingly common. Such studies sometimes use disease-specific HRQoL scales that have not been developed or validated with rigorous psychometric methods.

Despite their obvious merits, HRQoL and cognitive function have been tricky end points in brain tumor therapy trials for several reasons. The main obstacle has been the expense and difficulty of administering the instruments themselves. Combined with the declining functional status seen in the late stages of many brain tumor patients’ lives, this causes missing data to be a frequent problem in HRQoL analyses. HRQoL data tend not to be missing at random, but selectively for patients who are doing poorly—thereby falsely inflating the HRQoL of the remaining compliant population. Caregivers can be asked to rate HRQoL for patients under these circumstances (proxy rating). The reliability of proxy ratings as a surrogate for PROs is poor, especially in patients with cognitive deficits, but they are preferable to missing data.^147,148 Other problems have included difficulty defining the minimum clinically significant change in HRQoL measures (as opposed to the smallest statistically significant change) and the possibility that some supportive medications commonly used in brain tumor patients (such as antiseizure medications and corticosteroids) can interfere with test validity.¹⁰² To date, HRQoL has not been used by the FDA as the primary evidence for approving any anticancer product, although HRQoL data are often included in applications.^127,128

Functional status is an individual’s capacity to perform the normal activities of daily living; for adults this includes self-care and work. The oldest and most commonly used measure of functional status in cancer studies is the KPS (Table 113-2);⁷³ a closely allied scale is the ECOG performance scale, which is very similar to the WHO and Zubrod scales (Table 113-3).⁷⁴ Performance status is an important prognostic factor for survival in brain tumor patients, and most prospective studies include performance status as a baseline variable, usually with a minimum score required as an entry criterion for the study (e.g., KPS score of 70). Examining the KPS itself, it is clear that the difference between some levels is subjective and can be difficult to assign reliably, especially the highest levels of the scale. For example, a history of a single seizure curtails driving in most jurisdictions even if the seizures are controlled; whether such a patient can be assigned a KPS score of 100 (normal) or even 90 (capable of normal activity) is ambiguous. In a general cancer practice setting, interrater agreement on the KPS is generally good; in one study, raters agreed on KPS scores 59% of the time and disagreed by one level (10 points) 32% of the time, with 7% of ratings disagreeing by 20 or 30 points.¹⁴⁹ In brain tumor patients the KPS score is highly correlated with age and is sensitive to depression; it is most sensitive to changes in HRQoL in glioma patients for KPS scores in the 50 to 80 range and performs less well for higher scores.¹⁵⁰ KPS scores can be difficult to assign in retrospective studies. They are also highly confounded by the extent of resection in glioma studies, probably reflecting a combination of relief of mass effect from extensive resections with deficits caused primarily by unresectable tumors involving eloquent areas. Despite these drawbacks, KPS and other functional status scores are widely used in reporting baseline status in patient cohorts and in expressing immediate improvement or deterioration after brain tumor surgery by comparing presurgical and postsurgical scores.¹⁵¹ In longer term outcome studies, the KPS is used for defining “independent” or “functional” survival, and a 10-point decline in KPS score has been shown to correlate with increased risk for death in glioma patients.¹⁰

TABLE 113-2 Karnofsky Performance Scale

From Karnofsky DA, Burchenal JH. The clinical evaluation of chemotherapeutic agents in cancer. In: MacLeod CM, ed. Evaluation of Chemotherapeutic Agents. New York: Columbia University Press; 1949:196.

TABLE 113-3 Eastern Cooperative Oncology Group Performance Scale

From Oken MM, Creech RH, Tormey DC, et al. Toxicity and response criteria of the Eastern Cooperative Oncology Group. Am J Clin Oncol. 1982;5:649-655.

Cognitive function has already been mentioned in the context of HRQoL^140,152 and is most often measured with the Mini-Mental Status Examination (MMSE).^153,154 A multitude of other measures are available to express changes in neurological function generally or in specific functions.⁴ For example, after acoustic neuroma surgery, the Glasgow Benefit Inventory can be used to measure otolaryngologically related health status.¹⁵⁵ More specifically, facial nerve function can be graded by using the House-Brackmann scale,⁷⁶ hearing by using the Gardner-Robertson classification,⁷⁵ tinnitus by using the Tinnitus Handicap Inventory,¹⁵⁶ and vertigo by using the Dizziness Handicap Inventory.¹⁵⁷ After glioma surgery, the National Institutes of Health (NIH) Stroke Scale,^158–161 which measures 15 aspects of neurological function, has been used to detect new neurological deficits.⁵⁴ Other investigators have used less formal criteria, such as improved/stable/worsened motor function¹⁶² or detailed listings of adverse events,¹⁶³ to express neurological damage from glioma surgery quantitatively. Although the NIH Common Toxicity Criteria (www.fda.gov/cder/cancer/toxicityframe.htm) are widely used to measure the toxicity of chemotherapy, including brain tumor chemotherapy, the scales provided are not readily adapted to neurosurgical complications and are not as widely used as in reports of drug treatment. After pituitary surgery, endocrine end points are often the most useful in measuring cure and recurrence. These specialized tests are covered elsewhere in this textbook.

Tumor Response Rate

The tumor response rate (sometimes called the radiographic or objective response rate) is the proportion of patients whose tumors shrink in size as a direct response to therapy, an end point that is common in oncology studies. In brain tumor trials, tumor response is assessed with imaging studies, although tumors elsewhere in the body (e.g., plexiform neurofibromas) are sometimes measured directly by inspection.¹⁶⁴ The response rate has historically been considered to be a valid surrogate variable for improvement in survival or progression-free survival. The advantage of using the response rate as an end point is that it is more rapidly assessed than survival, with the median time to best response for most tumor/drug combinations being in the range of a few months. Assumptions behind the choice of response as an important oncology trial end point are that most anticancer drugs work through a cytotoxic mechanism, more cytotoxicity is reflected by greater macroscopic tumor shrinkage, and spontaneous tumor shrinkage is not part of the natural history of most cancers and must represent an effect of treatment. In addition, some tumor responses in general oncology have direct clinical benefits through reduction of symptoms. The FDA considers the tumor response rate in a single-arm trial to be usable as the major criterion for cancer drug approval under appropriate circumstances, especially if supported by additional evidence of clinical benefit.¹⁰² The most common use of response as a major end point in oncology is in phase II trials in which a single-arm design is used to test drugs for preliminary evidence of drug activity; in brain tumor trials, malignant glioma that progresses after initial radiation treatment (with or without adjuvant chemotherapy) is the usual setting for phase II drug testing.

With few exceptions (lymphoma, germinoma, and to a lesser degree, oligodendroglioma), most brain tumors have relatively low response rates to radiation treatment or chemotherapy, and the responses that are observed tend to be partial rather than complete. In general oncology, although drugs that cause higher response rates generally result in greater improvement in survival, large improvements in the tumor response rate tend to correspond to relatively modest survival benefits. For example, in advanced colorectal cancer, a 50% improvement in tumor response rate was associated with only a 6% improvement in survival.¹⁶⁵ Few drugs have ever been shown to cause such a large improvement in response rate for any brain tumor.

Tumor response rates have also been shown to be difficult to interpret in malignant glioma trials for several reasons. First, the likelihood of tumor response in drug trials for recurrent malignant glioma has been shown to depend on several patient- and tumor-related factors such as age, KPS score, tumor histology, and previous treatment.¹⁶⁶ This makes defining a target response rate that would indicate warranting a drug for further testing difficult.¹⁶⁷ Second, some “tumor responses” after surgical and radiation treatment of malignant glioma probably represent spontaneous regression of treatment-related imaging changes rather than the effect of anticancer drugs. Postoperative enhancement of the resection margin is common even after cerebral resection for non-neoplastic pathology unless imaging is performed within 48 to 72 hours after surgery.^{89–91,168–170} New enhancement within or adjacent to malignant tumors after radiation therapy may represent treatment-related necrosis rather than progression (“pseudoprogression”); this phenomenon is more common in patients treated with concomitant radiation therapy and temozolomide.^171,172 Third, increases in corticosteroid dose have been shown to reduce tumor enhancement, thereby confounding assessment.¹⁷³ Fourth, newer agents now entering testing may not work through cytotoxic effects or may have other beneficial effects on tumors that are not reflected by decreases in enhancement. Cytostatic agents might cause tumor stability rather than response, and this benefit could be missed if response were the only end point assessed. Some vasoactive drugs may provide direct clinical benefit through reducing peritumoral edema, an effect that is missed when reduction in enhancing volume is the end point.¹⁷⁴ Fifth, because enhancing volume itself is only a surrogate for tumor volume (reflecting the volume of blood-brain barrier breakdown characteristic of malignant glioma) and not a direct measure of solid tumor or relatively preserved brain tissue containing infiltrating tumor, vasoactive drugs could potentially cause spurious responses that reflect changes in vascular biology rather than a true antitumor effect.^174,175

For these reasons, recent trends in trial design have largely replaced tumor response with 6-month progression-free survival (PFS6) as the major end point in phase II drug trials for recurrent malignant glioma.¹⁰² Analysis of two large trial databases showed that PFS6 is a good surrogate for overall 1-year survival.^123,124 In addition to being free of several of the problems noted earlier in assessing response, PFS6 is completely assessible by 6 months after the last patient is enrolled in a phase II trial, thus facilitating early reporting.

Measurement of response for malignant glioma has changed over the past 2 decades and is still in the course of evolution. The earliest response scale (“Levin criteria”), based on computed tomography, used investigators’ subjective assessment of progression, stability, and response to define a five-level scale ranging from definite response to definite progression, with a sixth possible score indicating development of a new lesion.¹⁷⁶ The 1990 “Macdonald criteria” used the area calculated by multiplying the largest cross-sectional diameter of the tumor by the largest diameter perpendicular to it as the major determinant of response, in conjunction with neurological status and corticosteroid dose.¹⁷⁷ A 50% reduction in area was necessary to achieve a response, and a 25% increase in area indicated tumor progression. Corticosteroid dose was included in the criteria because of evidence that dose changes can affect tumor enhancement.¹⁷³ Although in general oncology a unidimensional response criterion based on changes in the greatest linear dimension of the tumor or tumors has recently been adopted,^178,179 widespread use of this method for primary brain tumors would have obvious drawbacks because of the highly irregular shape of most gliomas. Most scales for measuring response also require that the response be durable, specifically, that it be maintained over some minimum period, such as 2 months.

Current trends in brain tumor response assessment favor measurements based on MRI using semiautomated segmentation software, edited manually as needed by the radiologist, to define the measured volume because inter- and intrarater variability is reduced in comparison to linear- or area-based measurements.^180,181 Although no standard change in volume to define progression or response has yet emerged, a 50% change in volume to define both thresholds has been suggested.¹⁸² Volumetric assessment is particularly attractive for radiosurgery studies because the treatment planning process typically supplies a baseline volume that can easily be related to volumes at follow-up. Although novel imaging methods that reflect changes in tumor metabolism or vascular biology, such as perfusion or permeability, are in early stages of testing as possible response measures in brain tumor trials,¹⁸³ none has yet reached the level of routine clinical use at the time of writing. Volumetric methods also hold promise for assessment of benign tumor responses, such as in meningiomas or acoustic neuromas.¹⁸⁴ Whether unidimensional methods will gain future currency in brain metastasis trials remains to be seen, although to date response assessment has not been a major end point in these studies.

Early-Phase-of-Development Studies (Phase I, Phase II, “Phase 0”)

The characteristics of early- and late-phase studies in drug or product development are covered in depth in Chapter 11. In brief, a phase I trial is intended to prove the basic safety of a new treatment and, when applicable, to find the maximal tolerated dose of a new drug. Phase II trials are designed to detect the first evidence of a new treatment’s activity against tumors in patients and to select promising treatments for definitive proof of efficacy in phase III testing. Detailed guidelines for reporting phase I and II trials of brain tumor drugs¹⁸⁵ and surgically implanted brain tumor treatments¹⁸⁶ have been published.

Phase I trials of interest to surgeons primarily involve surgically administered treatments, such as drug-impregnated polymer wafers,^{49,163,187–189} gene¹⁹⁰ or oncolytic virus therapy,¹⁹¹ convection-enhanced delivery (CED) of targeted toxins or other drugs,^192–194 or brachytherapy.^195–197 In phase I studies the main goal is to define the maximal safe dose of the novel agent or, for some agents, the maximal feasible dose (when toxicity is negligible and cost or other factors prevent delivery of the agent above a certain practical ceiling). The number of patients treated in phase I studies is sharply limited to expose the fewest possible patients to harm. Evidence of efficacy is not expected in phase I trials, although tumor response is typically monitored and many drugs that eventually prove effective do show responses in phase I testing.

The typical structure of a phase I trial is a dose escalation model with three to six patients treated per dose stratum and escalation to a higher dose as each successive stratum proves safe. Either the dose of the implanted therapy^187,198 or the dose of a concomitant systemic therapy^199,200 may be escalated. Dose escalation phase I trials of surgically implanted brain tumor treatments have typically reported major toxicities thought to be at least potentially related to the intracranial implantation, including seizures, confusion, focal neurological deficits, cerebral edema, wound-healing problems, and fatigue. Infection is another important concern in these studies, particularly when foreign bodies are implanted or percutaneous catheters must remain in place for several days while treatments are being infused, although experience to date has not shown increased clinical infection rates to be a common problem.

Perhaps the two major challenges for phase I trials of surgically implanted treatments are related to the small number of patients treated, which makes it possible to miss uncommon but important toxicities, and the frequent difficulty distinguishing an elevated toxicity profile from the range of possible outcomes after routine craniotomy. In general oncology trials, well-established dose escalation and reduction rules have been developed to guide the course of a phase I trial and have been shown to be good predictors of the experience expected when a larger cohort is treated in phase II or III trials or in general use.⁶ Most toxicities in these trials are hematologic and readily reversible. In contrast, toxicity from implanted brain treatments may not be reversible, and an expected level of one such toxicity per three to six patients treated may exceed what would be acceptable in practice. Brain tumor surgery trials, then, may sometimes be better structured as phase I/II trials in which a larger cohort of patients are treated at the projected maximal tolerated dose under the same conditions of rigorous toxicity monitoring that characterize phase I trials. Alternatively, initial phase II trials may incorporate a pilot phase in which special attention is paid to toxicity in the first small cohort treated while further accrual is placed on hold.

In addition, when thrombocytopenia or neutropenia occur in phase I trials of systemic therapies, they are usually readily attributable to the experimental drug, whereas the normal range of postoperative morbidity after tumor craniotomies that do not use novel therapies includes some patients who have seizures, new neurological deficits, and transient confusion. Some trials, therefore, not only define dose-limiting toxicities by the NIH Common Toxicity Criteria level but also specify that seizures that are similar in character and frequency to preoperative patterns should not count as dose-limiting toxicity.¹⁹² Additional ad hoc toxicities not included in the NIH Common Toxicity Criteria may be monitored, such as increases in intracerebral pressure, Glasgow Coma Scale deterioration, or persistent (>2 weeks) new focal deficits.¹⁹² For some drug delivery methods, such as CED, reversible mass effect symptoms related to local volume infusion have been described and characterized.¹⁹⁴ In some cases, investigators have chosen phase II doses based on phase I trials in unconventional ways, such as that certain dose cohorts had longer or shorter survival. In one such example,¹⁸⁷ a later phase I trial using more standard dose escalation (made possible by drug manufacturing improvements) demonstrated that much higher doses of the agent could be safely delivered than had originally been thought possible.¹⁹⁸

Phase II trials are designed to detect the first evidence of an anticancer drug’s activity against tumors in patients. They are typically single-arm open-label studies enrolling 40 to 80 patients, although other designs are occasionally used. The purpose of a phase II trial is not to provide definitive evidence of efficacy, that being the role of phase III testing. In general oncology, the typical end point in phase II drug trials is the tumor response rate, but survival or progression-free survival can be used if measurement of response is problematic. Phase II trial results are compared with a historical control group to make the decision whether to proceed to phase III testing of the drug.¹⁶⁷ Many single-arm brain tumor surgical case series can informally be considered to be phase II trials as a way of examining the strength or weakness of the study design.

The major problem specific to brain tumor surgery phase II trial design is defining an appropriate historical control group. The major ways of constructing a standard-treatment historical control group were mentioned earlier, such as using results from previously conducted large clinical trials stratified by a validated prognostic index, such as the RTOG RPA for malignant glioma treatment.⁵⁶ The difficulty in using RPA to predict patient survival in surgical studies is that surgical patients are highly selected based on tumor features that favor surgical resection and the RPA classification does not fully account for these features. Similar bias is seen when using RPA to match patients selected for brachytherapy, radiosurgery, or intra-arterial chemotherapy.²⁰¹ For each of these treatments, initial phase II trials showed highly favorable response or survival results.^197,202,203 However, RCTs subsequently showed each of the treatments to be ineffective.^{195,196,204,205} For each treatment, later study showed that the historical control patients selected for eligibility for the treatment (but who did not receive the treatment) had a survival advantage based solely on their eligibility, which explained the overoptimistic phase II testing results.^206–210

For surgically implanted therapies and high-dose radiation studies (using brachytherapy or radiosurgery), imaging changes in or around the tumor site that mimic tumor progression but actually represent treatment-related brain necrosis are frequent.^211–214 This prevents the use of response or time to tumor progression as reliable phase II end points for these types of studies; survival end points are a better choice.¹⁰² Similarly, when gross total resection is a criterion for trial entry, response cannot be used as an end point, and progression-free survival or overall survival is used in phase II trials.⁶⁶ Some therapies progress directly from phase I to phase III testing because of the difficulty of designing a phase II trial.^187,215

The “phase 0” trial²¹⁶ is a new addition to the standard phases of drug development described in Chapter 11. Phase 0 trials enroll a small number of patients (typically <15), with limited drug exposure, and have no therapeutic or diagnostic intent. They clarify drug pharmacokinetics or pharmacodynamics during first-in-human use of new agents. In brain tumor studies, assessment of whether the novel agent achieves the desired modulation of its intended target is commonly the goal of phase 0 studies. In a typical design, a patient with recurrent glioma receives a dose of a novel, usually molecularly targeted agent immediately before planned surgical resection. After surgery, the resected tissue is submitted for analysis of tissue drug levels and the drug’s biologic target. This tests the drug’s penetration into the tumor, as well as drug activity. Problems with this design include ethical barriers (because of the lack of intended benefit to the patient, combined with concrete risks) and the possibility that novel agents could increase the risk associated with surgery (e.g., because of a direct effect on tumor vasculature²¹⁷). This trial design has been used to measure gene transduction after gene therapy,²¹⁸ inactivation of methylguanine-deoxyribonucleic acid methyltransferase (MGMT) after systemic O⁶-benzylguanine administration,¹⁹⁹ and epidermal growth factor receptor inhibitor effects on recurrent malignant gliomas.²¹⁹ It will probably gain wider currency because of the expense of testing molecularly targeted agents and the difficulty of submitting the many available new agents to extensive clinical testing in glioma, a relatively rare disease.²²⁰

Phase III Clinical Trials

The validity of phase III trial results is critical because it is these trials that most often cause a change in general clinical practice. Details of assessing the reliability of RCTs is covered in Chapter 11 of this textbook, and special points to consider in surgical RCTs are reviewed there. Among the most important problems for designing and conducting surgical RCTs are difficulty defining a patient population for which there is community equipoise (uncertainty in the medical community about the best treatment),^221–223 overcoming preexisting patient preferences in enrolling patients to an RCT,^224,225 difficulty blinding patients and outcome assessors,^226,227 dependence on the technical skill of the operator and related issues such as the learning curve,^192–194228 and the ongoing evolution characteristic of new technologies.^229,230

In surgical RCTs on brain tumor treatment, besides these issues that confront internal validity, comparability of the trial population to patients in general practice and the feasibility (and cost) of widespread adoption of complex procedures are special problems. RCT entry criteria are typically restrictive in comparison to general practice, and some examples indicate that relatively few patients in unselected series would actually qualify for surgical brain tumor RCTs. Whittle and coauthors²³¹ reported that only 25% of malignant glioma patients seen in an academic neuro-oncologic surgical practice met the entry criteria for an RCT of drug-impregnated wafers that was open at their center. Trial-eligible patients had better prognostic factors, including younger age, better clinical grade, and more extensive resections, and were more likely to undergo postoperative adjuvant radiation therapy. Generalizing the results of the RCT to the larger ineligible population is therefore problematic. Similarly, when therapies include resection of recurrent malignant glioma or are tested specifically in this population, it must be remembered that only 15% of patients will qualify for resection at time of glioma recurrence and that these patients likewise have relatively favorable prognostic factors when compared with patients who do not undergo second operations.²³²

Surgical treatments of malignant glioma are becoming increasingly complex with the introduction of CED, treatment plans that demand imaging-complete resections, and other technically demanding procedures. Proper training and credentialing may be necessary for surgeons who participate in trials of some complex therapies, such as CED, to prevent high rates of unsatisfactory technical performance from threatening the validity of the trial.^192,194 Technical aspects of performing the treatment may require study before or during the trial: investigators found that more CED catheters placed at a second, separate procedure met quality criteria than did catheters placed at the time of resection.¹⁹⁴ Ideally, these matters would be straightened out during phase I or II testing. Outcomes in patients undergoing complex surgical procedures, including craniotomy for tumor,^109–111233 are better with high-volume providers (hospitals and surgeons), and specialist surgeons have higher rates of complete tumor resection and fewer neurological complications.^234,235 However, with pretrial credentialing, the effects of surgeon volume on technical outcomes can be mitigated or prevented,²³⁶ and such trials offer an educational opportunity to begin the diffusion of advanced surgical technologies into the broader community.^237,238

Extent of Surgical Resection as a Prognostic Factor for Survival

One of the oldest and most popular study designs in brain tumor surgery is a comparison of survival in patients who receive extensive resection and those who do not. Nazzaro and Neuwelt published an influential rebuttal of much of this research in 1990 in which the flaws in trial design and statistical analysis nearly universal in these studies at that time were pointed out: failure to adjust the analyses for other important prognostic factors that might not be equally distributed between biopsied and resected patients (age, functional status, tumor location, tumor pathology); differential use of adjuvant therapies after biopsy or resection, such as radiation therapy, chemotherapy, and resection at recurrence; and general design flaws, such as consistent use of retrospective design and, frequently, failure to use “any form of statistical analysis.”²³⁹ Recent reviews confirm that this literature is still deeply flawed.^83,240

The main limitation of biopsy-versus-resection studies is that they compare “apples to oranges.” Comparisons of results of randomized and nonrandomized studies on the same treatments have shown that nonrandomized studies often give similar answers as RCTs on treatment questions, but only when the nonrandomized studies closely mimic the RCTs with which they are being compared.^241–243 Specifically, the control group in nonrandomized studies should be treated concurrently with the treatment group, and all patients in both the treatment and control groups should be equally eligible for the treatment being studied. This criterion is almost never met in studies that compare patients with different degrees of surgical resection: patients undergo gross total resection, less extensive resection, or biopsy based largely on the resectability of their tumors rather than by randomization. When eligibility for treatment (i.e., resectability) is a prognostic factor for outcome, confounding by indication prevents a valid nonrandomized comparison between treatments.

Some alternative trial designs are available that can avoid or adjust for this bias. First, an actual RCT can be performed, such as that reported by Vuorinen and associates on resection of glioblastoma.⁵³ Second, randomized studies of the addition of technical adjunctive treatments to surgery that are intended to improve the extent of resection (such as neuronavigation or intraoperative imaging) can provide a relatively pure means of randomizing patients to different chances for total resection. In the three such trials reported to date, when the adjunctive treatment improved the degree of resection, survival was improved as well.^54,162,244 The third design, which has not yet been used in a published study, would be to start with patients who were eligible for complete resection but in whom the initial resection proved to be subtotal. These patients, all of whom would have imaging-detected residual disease that was thought to be resectable, would then be randomized to second-look surgery or to immediate treatment with adjuvant therapies (radiation therapy and chemotherapy). A fourth design would be a nonrandomized comparison of subtotal versus total resection adjusted for “resectability” by using stratification or a propensity score model. Although several glioma resectability scales have been published that could be used for this purpose,^{151,245–248} the design has not yet been used. Fifth, studies could use a cohort²⁴⁷ or case-control²⁴⁹ design for patients who were considered eligible for gross total resection to compare those who did or did not undergo total removal. In a variant on this design, Shaw and coworkers used the well-known unreliability of the surgeon’s grading of the extent of resection in a constructive manner: in a cohort of prospectively observed, low-grade glioma patients in whom surgeons graded surgical excision as total, better progression-free survival was seen when the resection was deemed truly complete by imaging than when imaging showed measurable residual disease.⁸⁵

Studies on Technologic Adjuncts for Improving the Extent of Resection

Another popular research theme in brain tumor surgery is technologic advances to improve the extent of resection. Examples include intraoperative imaging techniques (MRI, ultrasound) to detect residual tumor, fluorescent dye visualization of residual tumor through specially adapted microscopes, and stereotactic neuronavigation based on preoperative imaging. The strongest design for studying these methods is an RCT in which patients eligible for resection (or for gross total resection) are randomized to receive or not receive the surgical adjunctive treatment being tested, with an end point measuring the extent of resection achieved.^54,162,244 A problem with such studies is the impossibility of blinding surgeons to the type of resection to be used. For intraoperative imaging or fluorescent dye studies, the best time for randomization would be after completion of the best possible resection via conventional surgery, although this is not usually done. For any technique promoting more aggressive surgery, careful and objective documentation of any additional risk for permanent neurological deficits is mandated.

A common, but much weaker study design is the routine use of an intraoperative tumor visualization technique at the end of conventional resection, with an end point of whether residual tumor is demonstrated (in which case additional resection is performed).^250–253 This design obviously biases toward artificially conservative initial efforts at resection with consequently high rates of “usefulness” of the technique in detecting residual tumor. An unrelated, but characteristic weakness of these nonrandomized studies is comparison against a historical control group demonstrating shorter length of stay as a purported benefit of the technique. Confounding by the strong temporal trend toward shorter length of stay for all neurosurgical treatments over the past 2 decades,²⁵⁴ primarily driven by insurance considerations, cannot be adjusted away by using statistical techniques.

Analysis of Survival in Recurrent Tumor Studies

Two special problems in recurrent tumor studies deserve special attention. The first is the common tendency to compare survival in patient groups who received different treatments at recurrence (such as reoperation) by using the original diagnosis as the starting point for the survival measurement. It is obvious that the later treatment cannot influence survival before it is even applied. The second problem is that the starting point for such studies, tumor recurrence, is a very subjective end point/starting point. In such studies, the means of distinguishing tumor progression from treatment-related imaging changes (such as with advanced imaging or biopsy^255,256) should be described carefully. Many studies of recurrent glioma exclude patients who had earlier treatment with therapies such as brachytherapy, radiosurgery, or drug-impregnated wafers because the incidence of peritumoral enhancement that does not represent tumor progression is so high after these treatments.

Cause-Specific Survival in Metastatic Tumor Studies

Metastatic tumor studies are subject to a complicated form of bias as a consequence of intercurrent death in some patients from progression of systemic disease. Imagine two groups of patients with single intracranial metastases treated in a uniform manner. If one group has uncontrolled systemic disease, few “neurological deaths” will occur as a result of intracranial treatment failure because overall survival will be very short. In contrast, a group with systemic disease that is cured will have a high eventual rate of neurological death as a result of treatment failure at the intracranial site. This “competing risks” problem merits consultation with a biostatistician when designing or reporting such studies.

Health Services Research: Volume Outcome and Disparities Studies

General considerations relevant to these types of studies are given in Chapter 11 of this textbook. With specific relevance to brain tumor studies, special consideration should be given to the influence of cultural differences on the detection and treatment of cancer among socially defined groups, which is particularly strong because cancer care is so expensive, invasive, and potentially toxic.²⁵⁷ In addition, the underrepresentation in cancer clinical trials of patients who are underinsured or uninsured, members of racial or ethnic minorities, older, female, less well educated, and unmarried^68,69,258 limits the confident assumption that new treatments will be equally effective in these patient groups.