Introduction

Approximately 20–40 % of cancer patients will eventually develop brain metastases [1]. It is the most common type of brain neoplasms in comparison to primary brain cancers. The most common primary cancer sites to metastasize to the brain are the breast, colorectal, lung, melanoma, and kidney [1]. Brain metastases are associated with a multitude of symptoms such as, but not limited to, headaches, motor weakness, balance problems, altered mental status, visual problems and seizures, with headaches and focal weakness being the most common symptoms [2–4].

Age, performance status (PS), extent of brain metastases and extracranial diseases ultimately aid in the decision-making process in determining the choice of treatment for patients with brain metastases [5]. The standard of care for multiple brain metastases is to offer whole brain radiotherapy (WBRT) and dexamethasone [2]. For solitary brain metastasis, in patients with good PS, brain metastases may be treated with surgical resection or with stereotactic radiotherapy [5]. In certain instances, such as in patients with poorer PS, supportive care such as symptom management with medication only may be offered [2]. Chemotherapy is not commonly considered as a treatment option for patients with brain metastases due to the complexity of the blood-brain barrier, which limits the entry of the treatment to the brain or limits its concentration in the blood to the brain [6].

Despite multiple treatment options, patients stricken with brain metastases have limited survival with an observed medial survival of 3–6 months following WBRT. Improved survival has been observed in patients treated with radiosurgery or surgical resection [5]. Due to limited survival, the improvement of health-related quality of life (HRQOL) through palliation of symptoms takes priority [7].

To assess the impact of brain metastases on HRQOL, tools have been employed [7]. One such tool is the European Organization for Research and Treatment of Cancer Quality of Life Questionnaire brain module (EORTC QLQ-BN20) in conjunction with the EORTC QLQ-C30, the general questionnaire [8].

Determining the minimal clinically important difference (MCID) in quality of life (QOL) instruments is important for healthcare practitioners and researchers alike. MCID is the smallest difference in score which is clinically important [9, 10]. Currently, clinical trials determine effectiveness of treatments on HRQOL through establishing statistical significance. However, statistical significance may not reflect a clinically relevant change, as statistical significance can be reached in a large enough sample size [11, 12]. The objective of this study was to determine the MCIDs in the EORTC QLQ-BN20 in patients with brain metastases.

Methods

Patient population

This analysis included pooled data from two prospective data sets. From October 2009 to July 2010, patients with brain metastases referred for WBRT from four Canadian centres and one Spanish centre were included in this study. All research was conducted upon approval by the respective research ethics board at each institution. Patients with brain metastases referred for WBRT, radiosurgery/gamma knife or neurosurgical resection were approached for the study. Patients over the age of 18 years with histologically or radiologically documented brain metastases were included.

Data collection

At baseline consultation, patients completed the EORTC QLQ-BN20 and either the EORTC QLQ-C30 or the EORTC QLQ-C15 PAL. Demographic information was also collected. This included age, gender, primary cancer site, Karnofsky performance status (KPS), cancer and radiation treatments. Patients also completed the BN20 and C30/C15-PAL at 1-month follow-up.

European Organization for Research and Treatment of Cancer—brain module (EORTC QLQ-BN20) instrument and scoring

The EORTC QLQ-BN20 is the brain cancer supplementary module to the EORTC QLQ-C30, the general “core” questionnaire. It contains 20 items, broken into four multi-item scales (13 items) and 7 singular items. The four subscales of the BN20 are future uncertainty (4 items), visual disorder (3 items), communication deficit (3 items), and motor dysfunction (3 items). The remaining 7 items assess symptoms of disease such as headaches, seizures, drowsiness, weakness of legs, bladder control and treatment toxic effects such as hair loss and itchy skin. Each item is rated on a Likert scale from 1 to 4 (1 = ‘not at all’, 2 = ‘a little’, 3 = ‘quite a bit’, and 4 = ‘very much’). All items were scored based on EORTC scoring procedure. Raw scores were standardized using linear transformation so that scores ranged from 0 to 100. For all four scales and 7 single items, a higher score represents poorer QOL.

Statistical analysis

Descriptive analysis was summarized as mean, standard deviation (SD), median, range for age and KPS and as proportions for categorical variables such as gender, primary cancer site, etc.

The methodology described by Maringwa et al. was used to determine MCIDs in the EORTC QLQ-BN20. This has been applied to other EORTC QOL tools [11, 13]. There are two different methods for determining MCIDs: anchor-based and distribution-based. Both anchor- and distribution-based approaches were utilized in the determination of MCID in the EORTC QLQ-BN20.

In the anchor-based approach, an “anchor” or external criterion is used. There are certain requirements for the selection of anchor: independently interpretable, clinically relevant, definable and moderately correlated with the instrument. Ultimately, for the anchor to be effective, it must be associated with HRQOL as determined by a moderate correlation (r > 0.30) and relevance to the disease [14]. Commonly used anchors are PS or progression of disease. The anchor is used to determine meaningful change. For our current analysis, patient-reported overall QOL as rated by the question: “How would you rate your overall quality of life during the past week?” (scale 1–7; the higher the scale, the better overall QOL) as taken from the EORTC QLQ-C30 or C15-PAL was used as the clinical marker to which the MCID was calculated.

Mean scores and mean change scores for all EORTC QLQ-BN20 scales and single items and overall QOL were calculated for each patient. As only four and nine patients had two units of increase and decrease from baseline, respectively, and three and six patients had more than two units of increase and decrease from baseline, respectively, one-unit change was considered as the overall QOL anchor. Patients with greater than one unit change in overall QOL were excluded from the MCID calculation as these changes were considered to be greater than minimal changes. Changes in overall QOL were categorized into three groups: deteriorated by one unit, no change and improved by one unit. Each individual patient’s QLQ-BN20 change scores were assigned to one of the three “clinically meaningful” categories as defined by the overall QOL anchor. To control for the amount of change in QLQ-BN20 that occurred to patients who did not change according to the overall QOL, MCID and its 95 % confidence interval (CI) were estimated by calculating the difference in the mean scale change between adjacent categories (e.g. ‘improved’ vs. ‘no change’ and ‘no change’ vs. ‘deteriorated’). MCIDs were determined to be statistically significant if the 95 % CI did not include zero.

Spearman correlation analysis was conducted between the overall QOL item from the accompanying C30 or C15-PAL and the BN20 scores at baseline and follow-up, respectively. Clinical relevance of the BN20 scale to overall QOL scale is represented by moderate correlation between QOL and BN20 scales and singular items which is represented by |r ≥ 0.3|.

The distribution-based approach is based on the idea that MCIDs can be estimated based on the distribution of scores within a sample of patients. As such, distribution-based approach is calculated by determining the standard deviations of scores and change scores and also the standard error of measurement (SEM). The SEM measures the precision of the EORTC QLQ-BN20 or other HRQOL instruments [15]. Standard deviations of scores and change scores were divided by fractions. In this current analyses, 0.2 SD, 0.3 SD, 0.5 SD and SEM (estimated by \( SD\times \sqrt{1-\mathrm{reliability}\kern0.5em \mathrm{coefficient}} \), describing the error associated with the measure) for the QLQ-BN20 were calculated and considered to be MCID estimates. All analyses were conducted using Statistical Analysis software (SAS version 9.3 for Windows; SAS Institute Inc., Cary, NC, USA).

Results

A total of 99 patients who completed both the baseline and follow-up BN20 and C30 or C15-PAL were included for analysis (Table 1). Patient’s age ranged from 22 to 83 years with an average of 61 ± 11.0 years. KPS of patients ranged from 40 to 100, with a median of 80. Of the included patients, 56 (57 %) were females. The most common primary cancer sites were the lung (54 %) and breast (22 %). Most patients received radiation, either radiosurgery/gamma knife (36 %) or WBRT (57 %). The remainder of patients received neurosurgery (6 %), and one patient received symptom management treatment (1 %). Most patients were outpatients (96 %). Our study had patients with one brain metastasis (33 %), two to three brain metastases (37 %) and multiple (more than three) brain metastases (30 %).

Table 1 Patient demographics (n = 99)
Full size table

Of the four EORTC QLQ-BN20 subscales at baseline, future uncertainty (r = −0.45) and motor dysfunction (r = −0.31) had moderate or better correlation with the anchor. While of the seven symptoms and treatment toxicity, only headache (r = −0.31) had a Spearman correlation of at least 0.30 with the overall QOL anchor. At follow-up, of the four EORTC QLQ-BN20 subscales, only future uncertainty domain (r = −0.48) had significant correlation with overall QOL anchor. There was a lower correlation (r < |0.30|) between any BN20 scale changes and overall QOL anchor changes (Table 2).

Table 2 Baseline, follow-Up and changes in EORTC QLQ-BN20 scores and correlation with overall quality of life
Full size table

All four subscales and seven symptoms and treatment toxicities on the EORTC QLQ-BN20 were included in the MCID calculation. At follow-up, there were 36 patients with no overall QOL change, 21 patients with overall QOL increase by one unit from baseline (improved) and 20 patients with one-unit decrease from baseline at 1 month follow-up (deteriorated). The mean scores for all EORTC QLQ-BN20 scales and single items for patients improved by one unit in overall QOL, no change in overall QOL and deteriorated by one unit in overall QOL are shown in Table 3. The mean change scores between adjacent anchor-based groups are shown in Table 3. For improvement, the mean change scores represented the difference between the improvement and no-change group. For deterioration, the mean change scores represented the difference between the deterioration and no-change group.

Table 3 Minimal clinically important differences (MCIDs) for the EORTC QLQ-BN20 subscales—the mean (standard deviation) of patient’s EORTC QLQ-BN20 scores categorized by their anchor status (improved, no change or deteriorated) and differences in the mean change scores (95 % CI) between adjacent categories (improvement is between “improvement” and “no change”, and deterioration is between “deterioration” and “no change”)
Full size table

MCID for improvement and no change groups did not demonstrate statistical significance for the four subscales and seven symptoms or treatment toxicity. However, for deterioration, two of the seven symptoms and treatment toxicity demonstrated statistically significant MCID. A change of 6.1 (95 % CI, 0.8 to 11.4) units in seizures and 13.8 (95 % CI, 0.2 to 27.4) units in weakness of legs, respectively, were required to establish clinical significance (Table 3). When compared to anchor-based MCID estimates, the distribution-based MCID estimates of 0.5 SD were the closest in comparison (Table 4).

Table 4 Distribution-based approach to determine minimal important differences in the EORTC QLQ-BN20
Full size table

Discussion

Since 1990, a shift in clinical trial endpoints has occurred [16]. Previously, clinical trial endpoints have mainly consisted of concrete endpoints that are patient-independent such as radiological or physical changes/progression or survival/progression-free survival. However, patient-reported outcomes such as QOL have gained importance and become more widely incorporated into clinical trials as exemplified by the National Cancer Institute of Canada policy requiring HRQOL to be incorporated into all their phase III trials unless otherwise indicated [17]. This change, which has been shown to be useful in determining the best treatments for patients and evaluating effects of treatment toxicity on patients [18], is also of benefit to evaluating treatments for advanced cancer patients of which HRQOL is the primary treatment endpoint. With such increasing importance placed on HRQOL, further emphasis is required to evaluate how to interpret this endpoint to ultimately assess treatments for patients. As such the application of MCID analysis is conducted to determine clinically accurate and substantial evaluations of treatments. The intention of our study was to determine the MCID in the EORTC QLQ-BN20 scales in brain metastases patients.

The most informative HRQOL questionnaires are disease-specific or condition-specific [18], an example of which is the EORTC QLQ-BN20. A literature search was conducted using the keywords “meaningful change” or “minimal clinically important difference” or “minimal important difference” and “neoplasm” or “cancer” to determine whether a similar research has been conducted in meaningful changes seen in patients with brain metastases. This is the first study evaluating MCIDs in the EORTC QLQ-BN20 for patients with brain metastases.

One other study was conducted by Maringwa et al. on MCIDs in the EORC QLQ-C30 and EORTC QLQ-BN20 for patients with primary brain cancer; this study included 941 patients and used the mini-mental status exam (MMSE) and World Health Organization performance status as an anchor. In their analysis, PS was used as a clinical anchor for the motor dysfunction scale from the BN20 and found that a 5.2-unit change in the scale represented clinically significant deterioration. Similarly, the MMSE was used as another anchor to determine the MCID for the communication deficit scale of the BN20, which found that 9.1-unit changes represented clinically significant improvement [11]. Although most similar to our current study, the findings in Maringwa et al.’s study differ from our current findings. This may be due to the difference in patient population and also the clinical anchors used.

In an analysis conducted by Bedard et al. in determining minimal important differences in advanced cancer patients using the EORTC QLQ-C15PAL, a sub-analysis was conducted on patients with different metastases. In patients with brain metastases, it was found that at 1-month follow-up, a change of 18.0 units in the scale of emotional functioning was statistically significant to represent clinically significant improvement using the overall QOL question as an anchor. A change of 27.2, 40.8 and 36.5 units in physical functioning, fatigue and pain, respectively, was required to represent a clinically significant improvement in patients [19]. Our study however, did not find any MCIDs for improvement in the EORTC QLQ-BN20 scales and symptoms.

MCID analyses have also been conducted in a general population of advanced cancer patients on a variety of symptoms and QOL assessments. However, due to the heterogeneity of patients included in the patient population, this could lead to a difference in MCIDs due to the symptom experience of patients with different sites of metastases. Bone metastases patients would be more likely to experience pain or bone pain, and brain metastases patients may be more likely to cite symptoms of fatigue [20]. Similar to our current investigation, Liang et al. have also conducted a MCID analysis on the EORTC QLQ-C30 and EORTC QLQ-BM22, which is the bone metastases module, to more specifically evaluate HRQOL in patients with bone metastases [21]. Comparably, our study is supported by the specificity of our patient population. However, these MCIDs are based on the clinical anchor used and are relative only to our chosen anchor.

The distribution-based MCID for the BN20 in brain metastases patients was close to the 0.5 SD. In accordance with our results, in a previous validation study by Norman et al. conducted in multiple studies using other QOL questionnaires, they found the 0.5 SD to be the closest value which represents most meaningful change [22].

Further investigation in MCID in the brain metastases population is useful to determine the required sample sizes for future brain metastases-related clinical trials by determining the required amount of patients to be treated [23]. This can also avoid the limitations of evaluating statistical significance in deciding treatment options.

In certain other studies, a larger magnitude of change has been found to be associated with meaningful improvement [13, 21]; however, other studies have also found the opposite: a larger magnitude of change is required to find a clinically meaningful deterioration [10, 24, 25]. Maringwa et al. have speculated that this may be attributed to subconscious physician bias in reporting stable or worsening PS, which ultimately explains why improvement requires greater QOL change [13]. However, our study avoids this possible bias through using a patient-reported clinical outcome in which patients themselves judge their QOL. This may explain why in our study, only meaningful change in deterioration was detected but not detected in patients who improved. Future studies may consider using such patient-reported clinical anchors as the patients themselves are most aware of their conditions, and such physician bias may also be avoided.

Limitations

A significant limitation to our current study is the small sample size. A larger sample size is required to reduce the possibility that the determined MCIDs for deterioration are not due to sampling variation and determine whether MCIDs for improvements exist. Our second limitation is the usage of one anchor (overall QOL) in our current study. Other conducted studies on MCIDs have utilized multiple anchors in the anchor-based approach. Thirdly, as suggested by Cella et al. and Crosby et al., the appropriateness of the anchor depends on the level of correlation between the QOL scales and the anchor [15, 26]. The anchors should also have at least a correlation of r > 0.30 with HRQOL [14, 26]. However, in our correlation analysis, most scores had weak correlation with overall QOL at baseline or follow-up, except for future uncertainty, headaches and motor dysfunction. Although significantly correlated, the MCID estimates for future uncertainty, motor dysfunction and headache were not significant. The absence of strong correlation with the anchor may suggest that the chosen anchor is inappropriate for MCID estimates; however, other studies have also found moderately strong correlation at best between their chosen anchor and HRQOL scores, with the reason of this currently unknown [10, 11, 24]. As such our findings should be taken with caution. Due to the weak correlation with our chosen anchor, this advocates for further analyses with multiple anchors, especially since MCIDs are relative to the clinical anchor chosen [11]. Future studies should conduct MCID analysis in a larger population size with multiple anchors, such as KPS or changes in MMSE [11], to avoid such problems.

Conclusion

In our current study, we determine the required change, which represents clinically meaningful change in the EORTC QLQ-BN20 in brain metastases patients undergoing treatment. Understanding MCIDs allows physicians to determine the impact of treatment on patients’ QOL and sample sizes for clinical trials. Future studies should be conducted to validate in a larger population of patients with brain metastases.