Establishing the minimal clinically important difference of the Brief Fatigue Inventory for brain or CNS cancer patients undergoing radiotherapy

Abstract

Background

Minimal clinically important differences (MCIDs) quantify the clinical relevance of quality of life results at the individual patient and group level. The aim of this study was to estimate the MCID for the Brief Fatigue Inventory (BFI) and the Worst and Usual Fatigue items in patients with brain or CNS cancer undergoing curative radiotherapy.

Methods

Data from a multi-site prospective registry was used. The MCID was calculated using distribution-based and anchor-based approaches. For the anchor-based approach, the fatigue item from the PROMIS-10 served as the anchor to determine if a patient improved, deteriorated, or had no change from baseline to end of treatment (EOT). We compared the unadjusted means on the BFI for the 3 groups to calculate the MCID. For the distribution-based approaches, we calculated the MCID as 0.5 SD of the scores and as 1.96 times the standard error of measurement.

Results

Three-hundred and fifty nine patients with brain or CNS tumors undergoing curative radiotherapy filled out the 9-item BFI at baseline and EOT. The MCID for the BFI was 1.33 (ranging from 0.99 to 1.70 across the approaches), 1.51 (ranging from 1.16 to 2.02) and 1.76 (ranging from 1.38 to 2.14) for the usual and worst fatigue items, respectively.

Conclusions

This study provides the MCID ranges for the BFI and Worst and Usual fatigue items, which will allow clinically meaningful conclusions to be drawn from BFI scores. These results can be used to select optimal treatments for patients with brain or CNS cancer or to interpret BFI scores from clinical trials.

Primary central nervous system (CNS) tumors are a diverse collection of tumors that can either be benign or malignant and develop from cells within the CNS. Although uncommon they account for a disproportionate burden of cancer mortality.¹ Both tumor progression and treatment effects in the CNS can cause substantial morbidity and decreased quality of life.² One significant and distressing side effect patients with brain tumors experience is cancer-related fatigue,³ which affects a patient’s quality of life and even survival.⁴ To measure fatigue, physicians often rely on patient-reported outcomes (PROs) like the Multidimensional Fatigue Inventory (MFI)⁵ or Functional Assessment of Cancer Therapy—Fatigue (FACT-F).⁶ Another popular measure of fatigue for people with any type of cancer is the Brief Fatigue Inventory (BFI), which assesses the severity and impact of fatigue experienced by cancer patients.⁷ An advantage of the BFI over other fatigue PROs is that it maintains good psychometric properties with only 9 items.⁸ There is debate as to whether the 9 items measure 1 factor or 2 factors. While some have concluded the first 3 items of the BFI define a severity factor and the remaining 6 items define an interference factor,^9,10 most conclude only 1 factor underlies the 9 items.^7,11–13

Most studies in neuro-oncology rely on statistical significance to conclude if quality of life (QOL) has improved or is better for patients in 1 treatment group versus another. For instance, conformal avoidance of the hippocampal neural stem cells led to a significant difference in cognitive factor differences (eg less difficulty remembering things) for patients undergoing whole-brain radiotherapy.¹⁴ Statistical significance, however, has limitations such as being overly reliant on sample size. Additionally, a statistically significant change does not always lead to a patient experiencing a clinically meaningful change in their QOL. The minimal clinically important difference (MCID) is the smallest numerical change in a PRO that a patient perceives as beneficial or detrimental or could warrant a change in treatment.¹⁵ Accordingly, there could be a statistically significant change in PRO results from one time point to another that is of small magnitude and does not reach the MCID, and therefore is not deemed clinically relevant. The MCID can be calculated with a multitude of methodologies and there is currently no gold standard.¹⁶ The data-driven methods for calculating the MCID can be categorized into 2 categories: distribution-based approaches and anchor-based approaches. Distribution-based approaches use statistics that measure the variability of scores, such as standard deviation (SD) or standard error of measurement (SEM), to calculate the MCID. Because distribution-based approaches rely on the statistical properties of the sample at a single time point, the MCIDs calculated by these approaches do not measure the importance of change over time.¹⁷ Anchor-based approaches, on the other hand, calculate the MCID by mapping the changes of the PRO of interest onto a variable known as an anchor (can be another PRO or a clinical variable) that has an established scale of importance. For instance, change in occupationally productive hours was used as an anchor to establish the MCID for the MFI.^18,19 Because there are pros and cons to each method for calculating the MCID, it is best practice to calculate the MCID using multiple methods and report a range of scores in addition to a single score to reflect the uncertainty of the estimate.^20,21

The MCID is measure-specific, disease-specific, and even treatment-specific (eg chemotherapy and radiation therapy).²² Wang, Hao, Wang, Guo, Jiang, Mendoza, and Cleeland¹⁰ used only distribution-based approaches at one time point to calculate the MCID for the severity and interference factors of the Chinese version of the BFI for patients with any type of cancer who were getting radiotherapy, medical oncology, or surgery. Currently, there are no MCIDs for the overall scale of the BFI specific to CNS radiation oncology patients. Thus, this study will use distribution-based and anchor-based approaches to calculate the MCID of the BFI for patients with brain or CNS cancer receiving curative radiotherapy.

Materials and Methods

Patients with primary CNS tumors were enrolled on a large multi-site prospective registry from 2017 to 2022. This study was approved by the Institutional Review Board (No. 22-001237) and was conducted in accordance with the Declaration of Helsinki of 1975 (as revised in 1985). Patients filled out PROs prior to radiotherapy (Baseline) and filled out the same PROs again at the End of Treatment (EOT). To be included in the analysis, the patient must have undergone curative radiotherapy for brain or CNS cancer (the patient could also have undergone any adjuvant therapy) and completed the 9-item BFI and fatigue item of the PROMIS-10 at Baseline and EOT.

Measures

The 9 items of the BFI are on an 11-point scale (0–10) with higher scores indicating more fatigue. The first 3 items ask the patient to rate their current, usual, and worst fatigue in the past 24 h. The remaining 6 items ask how fatigue has interfered with recent activities (ie general activity, mood, walking ability, normal work, relations with other people, and enjoyment of life) in the past 24 h. The scale score is calculated as the average of the 9 items.⁷ In addition to using the overall mean scale score of the BFI, some clinicians look at the responses to the usual or worst fatigue items to assess improvement or deterioration and make treatment decisions based on patient responses to those items.¹⁰ Thus, there is utility in calculating MCID for those 2 individual items.

The fatigue item from the PROMIS-10 was used as an anchor in this study. This item asked participants, “In the past 7 days, how would you rate your fatigue on average?” This item was measured at baseline and EOT on a 5-point scale with 1 being “None” and 5 being “Very Severe.”

Statistical Analysis

Because there is debate as to the factor structure of the BFI, we first conducted an exploratory factor analysis (EFA) using maximum likelihood estimation to determine the number of factors.²³ The number of factors was determined based on the Kaiser criterion (ie number of eigenvalues above 1), the scree test, and clinical interpretability.^24,25 We calculated the MCID for each factor determined by the EFA.

As is recommended, we calculated the MCID using both distribution-based and anchor-based approaches. For the distribution-based approaches, we calculated the MCID as 0.5 SD and 1.96 SEM. While there are other values that have been proposed, we chose 0.5 for the SD method because it is the value where most meaningful changes occur.^26,27 The SEM was calculated as $1−reliability×SD$⁠. We multiplied by 1.96 to calculate the score that is significantly greater than the expected random variation at the 5% alpha level.²⁸ Reliability was measured via Cronbach’s alpha. Because the SEM relies on the reliability of a scale, SEM does not apply to individual items. Thus, we only calculated the SEM for the overall scale and not the Usual Fatigue and Worst Fatigue items. We calculated the MCID using these distribution-based approaches at baseline and EOT and then averaged across time points so that each approach was associated with only 1 value.

For the anchor-based approach, we used the fatigue item from the PROMIS-10 as the anchor. If a patient selected the same score on the anchor at both time points, they were labeled as experiencing no change in fatigue. If a patient selected a score at EOT that was higher than their score at baseline, they were labeled as deteriorated. Finally, if a patient selected a score at EOT that was lower than the score at baseline, they were labeled as improved. The MCID for deterioration was calculated as the difference between the unadjusted means of the BFI for the no-change and deteriorated groups and the MCID for improvement was calculated as the difference between the no-change and improved groups.²⁸

All analyses, except the EFA, were performed using SAS v9.4 software.²⁹ CEFA was used to conduct the EFA.³⁰ The recommended MCID range was calculated as the minimum and maximum of the various approaches.

Results

Patient Characteristics

Of the 635 patients who had primary brain or CNS tumors undergoing curative radiotherapy, 359 completed the BFI and the anchor at baseline and EOT. Table 1 provides the baseline patient and treatment characteristics of the full sample and is split by patients who have missing data on the BFI at EOT. The median age was 52 (range 18–94). The sample was about evenly split between those who underwent photon radiotherapy and those who underwent proton radiotherapy. The median total dose was 57 Gy (range 40.0–76.0) over a median of 30 fractions (range 10–41).

Factor Analysis

To determine the number of factors underlying the 9 items of the BFI, we conducted an EFA using maximum likelihood estimation and performed the scree test, which identifies the “elbow” of a scree plot and retains all factors above the elbow.²⁴ As shown in Figure 1, the scree plot supported a unidimensional construct of the 9 items. Additionally, the Kaiser criterion also supported a unidimensional construct as only 1 eigenvalue was greater than 1.00. Further, the scale was developed assuming a unidimensional construct and most clinicians interpret the BFI using the scale score.⁷ Thus, in contrast to Wang and colleagues,¹⁰ we concluded there was only 1 factor underlying the 9 BFI items and calculated the MCID for the overall scale of the BFI.

Distribution-Based Approaches

At baseline, the mean and SD of the BFI scale score were 2.68 and 2.09, respectively. At EOT, these values were 3.15 and 2.31, respectively. Thus, the MCID was calculated as 1.05 and 1.16 at baseline and EOT using the 0.5 SD method for an averaged value of 1.10. The reliability of the scale was 0.944 at baseline and 0.952 at EOT. Thus, using the 1.96 SEM method, the MCID was 0.97 at baseline and 1.00 at EOT for an averag value of 0.99. Table 2 presents these results and the results for the usual and worst fatigue items.

As a sensitivity analysis to assess the potential effect of missing data, the distribution-based approaches were calculated at baseline for the full 635 sample of patients. The MCID values and psychometric properties for this sample are presented in Table 3.

Anchor-Based Approaches

The correlations of the PROMIS-10 fatigue item with the BFI scale score and usual and worst fatigue items were 0.67,0.63, and 0.64, respectively. Because these values were above the standard 0.40 threshold, we concluded the PROMIS-10 fatigue item was an appropriate anchor for all 3 metrics.

According to the anchor, 77 patients improved from baseline to EOT with 69 decreasing by 1 point, 7 decreasing by 2 points, and 1 decreasing by 3 points. One-hundred and twenty-seven patients deteriorated from baseline with 106 increasing by 1 point, 19 increasing by 2 points, 1 increasing by 3 points, and 1 increasing by 4 points. Finally, 155 patients indicated no change according to the anchor. Because of the small sample sizes for some of the change values, patients were categorized as either improved, deteriorated, or no change.

The 77 patients who improved from baseline to EOT had an initial mean of 3.61 and a mean change of –1.32 on the BFI. The 127 patients who deteriorated from baseline had an initial mean of 2.26 and a mean change of 1.90 on the BFI. Finally, the 155 patients who indicated no change according to the anchor had an initial mean of 2.55 and a mean change of 0.20 on the BFI. Thus, the MCID was calculated as 1.51 for improvement and 1.70 for deterioration. Table 4 presents these results and the MCID values and mean change for the usual and worst fatigue items.

Recommendations

The MCID for the BFI scale ranged from 0.99 to 1.70, with a mean value of 1.33. The MCID for the usual fatigue item ranged from 1.16 to 2.02, with a mean value of 1.51. Finally, the MCID for the worst fatigue item ranged from 1.38 to 2.14, with a mean value of 1.76. We recommend researchers use the means as the overall MCID when conducting power analyses or comparing groups but also conduct sensitivity analyses using the minimum and maximum of the range of values for the MCID.

Discussion

In this study, both distribution- and anchor-based approaches were used to calculate the MCID for the Brief Fatigue Inventory and 2 of its items for brain and CNS cancer patients. These values can be used to compare independent groups of brain and CNS cancer patients on fatigue, to assess change in fatigue over time, or to aid in sample size planning when designing a clinical trial with cancer-related fatigue as an endpoint.

The MCIDs calculated using the anchor-based approaches were larger compared to those calculated using the distribution-based approaches. For instance, for the usual fatigue item, the MCID was 1.16 using the 0.5 SD approach but was almost double for the deterioration approach (2.02). A possible explanation for the disparate MCID calculations is that the anchor used in this study is a single item from a PRO scale, which contains measurement error. Thus, the categorization of patients into improved, deteriorated, and no change may not be accurate and bias the calculations of the MCIDs. Ideally, the anchor should not contain measurement error, which is why some researchers use more objective measures for the anchor. For instance, WHO performance status and CTCAE scales were used to establish the MCIDs for EORTC QLQ-C30 scales.¹⁹ However, objective measures of neurocognition do not always have strong correlations with subjective experiences of neurocognitive issues as measured by PROs and thus do not reach the correlation of 0.4 thresholds needed for an anchor.³¹ Another possible explanation is that the context of when to use distribution-based and anchor-based approaches may differ. Distribution-based approaches do not measure change as they are calculated using data at a single time point whereas anchor-based approaches can measure change. Thus, distribution-based approaches may only be appropriate for between-group comparisons whereas anchor-based approaches can be used for within-group mean change over time.¹⁹

Within the anchor-based approach, the improvement method indicated a smaller MCID value compared to the deterioration method for the scale and both items. This is likely due to a floor effect. The BFI scale score and individual item scores can range from 0 to 10; however, most of the mean scores of the scale and items were between 2 and 4 points. Taking the scale score as an example, the improved group of patients had a mean of 3.61 on the BFI at baseline. So, their max change, on average, could only be 3.61. Alternatively, the deteriorated group of patients had a mean of 2.26 at baseline. Thus, their max change, on average, could be 10 – 2.26 = 7.74.

Wang et al.¹⁰ calculated slightly higher MCID values for the 6-item interference subscale and 3-item severity subscale of the Chinese version of the BFI compared to our MCID values for the 9-item BFI scale scores. They reported a larger SD (2.50 for both subscales) and a smaller reliability (0.90). Thus, their MCID values were larger compared to our distribution-based approach MCID values. Distribution-based approaches are highly reliant on the diversity of the sample so a possible explanation for this discrepancy is our sample is more homogenous as the patients were all treated with radiotherapy and have either brain or CNS cancer whereas the sample in Wang et al.¹⁰ consisted of patients with any type of cancer who could have been treated with radiotherapy, medical oncology, surgery, and any combination. Thus, a limitation of this study is the generalizability of the results. Previous literature suggests that MCIDs differ across disease sites, so these MCID values may not apply to other cancer types.¹⁹ Another limitation of this study is that we used listwise deletion to handle missing data. Of the 635 patients who responded to the baseline PROs, only 359 of them also filled out EOT PROs, resulting in a 56.6% response rate. A potential explanation for the high rate of missingness is that the patients in this study were enrolled on a registry rather than registered for an interventional trial. More than likely, the data are not missing completely at random so the mean and SD of the BFI and its items in the listwise deletion sample may be biased. We conducted a sensitivity analysis for the distribution-based approaches using the complete sample at baseline. The SD values were slightly higher for the scale score and individual items, resulting in larger MCID values (see Table 3). Even though there are modern missing data handling techniques that can mitigate some of the effects of missing data (eg full information maximum likelihood and multiple imputation), they cannot completely compensate for such a high missing data rate.³² Efforts to increase the response rate and understand the reasons why the response rate is low would be important to identify to decrease the effect of data that are not missing at random.³³

Conclusions

QOL is an important outcome to assess when treating CNS cancers.³⁴ The BFI is a tool to assess cancer-related fatigue and can measure the impact of radiotherapy on a patient’s QOL. This study provides the MCID range for the BFI (0.99–1.70), usual fatigue item (1.16–2.02), and worst fatigue item (1.38–2.14), which will allow clinically meaningful conclusions to be drawn from BFI scores, and inform the selection of optimal treatments for these patients. Future research should explore what demographic characteristics, clinical data, or treatment variables may be predictive of which patients experience a change greater than the MCID threshold.

Conflict of interest statement

All authors declare no conflict of interest with respect to this work.

Funding

This study was institutionally (Mayo Clinic) funded.

Acknowledgments

The authors thank the patients and families along with the research staff of Mayo Clinic. Clinical, treatment, and patient-reported outcome data is supported through an institutionally supported curation of patient electronic records.³⁵

Authorship statement

Conception and design of the study: H.J.G., I.Z., W.G.B., T.L., P.D.B., T.A.D. Data collection: A.M., P.D.B., E.Y., S.A.V., K.W.M., S.L.S., J.B.A., J.L.P., J.L.L., Z.C.W., N.N.L. Analysis and/or interpretation of data: H.J.G., I.Z., T.L., T.A.D. Drafting the manuscript: H.J.G. Review and revision of the manuscript: H.J.G., I.Z., W.G.B., A.M., P.D.B., K.W.M., S.L.S., Z.C.W., B.S.L., T.A.D. Approval of the version of the manuscript to be published: H.J.G., I.Z., W.G.B., T.L., A.M., P.D.B., E.Y., S.A.V., K.W.M., S.L.S., J.B.A., J.L.P., J.L.L., Z.C.W., B.S.L., N.N.L., T.A.D.

Data availability

The data will be made available upon reasonable request.

References