Revised 15-item Glasgow Benefit Inventory with five factors based on analysis of a large population study of medical and surgical otorhinolaryngological interventions

Objectives: To review, using confirmatory factor analysis, the widely used 18-ques-tion Glasgow Benefit Inventory [GBI] that has three factors. Thereafter to develop, using exploratory factor analysis, a more coherent, revised version of the GBI. Design: Confirmatory and exploratory factor analysis of a large national GBI data set of ORL interventions. Setting:

performed, it is easy to use and adaptable to a variety of clinical situations. Since its original description, the GBI has been used on a range of ORL surgical interventions with Hendry et al reviewing 117 reports with a search date of January 2015. 2 Since then there have been a further 79 publications of surgical series, not only in ORL but facial plastic surgery and ophthalmology [OVID Medline search July 2019 using search terms "GBI" and "Glasgow Benefit Inventory"].
The original report of the GBI in 1996 included a principal component analysis (PCA) that derived three sub-scales: General, Social Support and Physical Health. The practice of statistics has evolved materially since then and there is agreement that while PCA can reduce data by combining variables into a smaller number of components, it is not always effective in understanding the constructs that underlie the data. It therefore should not be used for exploratory factor analysis. The exploratory element requires that as few assumptions as possible are made at the outset about the nature of the factors and the relationship between them, which is not the case for PCA.
In addition, the "General" component in the 1996 analysis is not interpretable as a construct, containing as it does a diverse range of questions with no coherent pattern. The overall GBI scores are the average across all 18 responses. However, a composite total score is only justifiable if the factors all converge. This has not been shown.
Given the frequency of use of the GBI in research and clinical contexts, it is critical that the psychometric properties of the instrument are revisited, and if necessary revised.

The Scottish Ear Nose and Throat Outcome study (SENTOS)
was a prospective cohort study of 9005 patients attending outpatient ORL clinics at a number of Scottish NHS hospitals. 3 The study involved the administration of two outcome measures: the Health Utilities Index mark 3 (HUI-3) and the GBI. Only the results from the HUI-3 have been reported in detail. 3 GBI questionnaires were completed by 4799 SENTOS participants, giving a data set for analysis which is considerably larger than any other obtained thus far. This gives us an excellent opportunity to reassess the factor structure within the data and to revise the GBI if required.
The purpose of this paper is not to report the many medical and other non-surgical interventions that are common to ORL practice and in SENTOS. These will be reported elsewhere.

| ME THODS
The aim in this study was primarily to test the stability of the factor structure reported in the original 1996 description of the GBI 1 by using more current, robust statistical methods in a newer large data set.
The data set comprised GBI responses obtained from adult patients (aged 16 years or older) who attended an NHS Academic ORL outpatient appointment and who completed the GBI as part of the SENTOS study. Details of this patient cohort have been published previously. 3 Our plan was to perform a confirmatory analysis to see whether the original three-factor structure provided an appropriate fit for the data. Should this not provide a good fit for the data, we planned to proceed to exploratory factor analysis to uncover whatever underlying variables exist in the data. 4 Further aims of the study were to assess the internal consistency of the item/factor structure and to report methods for calculating and analysing GBI scores.
Confirmatory analysis was performed using principal component analysis, with and without covariance (orthogonal). Exploratory analysis was done using parallel analysis with minimal residual extraction and oblimin rotation.
All analyses were performed using The R Project for Statistical Computing (R Foundation for Statistical Computing, Vienna, Austria).
For brevity and ease of reading, only an outline of the statistical methods and results is presented here. Those readers who wish to know more detail are directed to the Appendix S2.

| Missing SENTOS data
Of the 9005 subjects in SENTOS, 68 were aged 14 or 15 years and therefore excluded as being non-adults. 4799 adults answered at least one of the 18 GBI questions.
Question 9 ("job opportunities") was not answered by 8.3% of participants compared with the 1%-2% for the other seventeen questions. This question could be irrelevant to a large proportion of any study population, particularly those who are retired or not actively seeking employment. The age of participants leaving question 9 unanswered was higher than those who completed it (mean age 67.25 vs 53.16 years, t test, P < .001). This is in keeping with an effect of retirement from employment. Its removal from the scores of 3436 adults that had completed question 9 had no material clinical

• The Scottish Ear Nose and Throat Outcome Study
[SENTOS] produced a large data set of GBI responses which were used to explore and refine the scoring and reporting of the GBI.
• Q9 on employment opportunities was unanswered by 8% of participants. As such it should be removed.
• An exploratory factor analysis then produced a 15-item, five-factor GBI which has been named GBI-5F.  Accordingly, question 9 was removed from all further analysis.

| Categorisation of active interventions
In the data set, interventions were limited to two for each patient: one primary and, where applicable, one secondary intervention. In many instances, a patient had two active interventions that made it impossible to differentiate which intervention was the primary reason for any reported benefit. Middle ear surgery along with provision of a hearing aid is such an example. Accordingly, patients with more than one recorded intervention were evaluated by two authors [GGB and SR] to exclude, in a systematic manner, patients in whom it would be difficult to attribute the benefit to a specific intervention group. This left 3658 patients for analysis. This remains a sufficient number of patients to perform a valid analysis for this paper.
There were two large categories of "reassurance" and "self-management" with a total of 1532 patients. These categories are interesting and warrant further study, but the absence of any active (medical or surgical) intervention means that their reported benefit was small and therefore unhelpful for this study. Excluding this, "reassurance" and "self-management" group leaves a total of 2126 patients who underwent some kind of active intervention. One thousand nine hundred eighty of these had complete GBI data with no missing items. It is their data that form the basis of this study.

| Confirmatory factor analysis
Our confirmatory factor analysis suggests that the three-factor model originally described in 1996 1 is not a good fit for the data.
It appears that not enough factors were properly elaborated, with the General factor derived from nine questions, potentially obscuring multiple dimensions. This makes sense given the large number of heterogeneous questions encompassed by the General factor. We therefore proceeded to a new exploratory factor analysis in order to discern a more acceptable factor structure for our data.

| Exploratory factor analysis
Parallel analysis suggested six factors be extracted on the active interventions sample of 1980 adults. Four factors clearly met the criteria for retention (Table 1). We reviewed the questions in each factor and interpreted to represent Quality of Life (factor 1), Self-Confidence (factor 2), Support (factor 3) and General Health (factor 4).
The fifth factor (which we interpret as Social Involvement) was found to have two items above the 0.40 loading criterion and one that just fell short of this threshold at 0.38 (Stevens, 1992). Despite just missing the coefficient criterion, the factor had a cogent theme, which we interpreted as Social Involvement, and additional information that justified its retention. The sixth factor was discarded due to having only one element (Q14), and that this one question was below the loading criterion (ie was empty). We decided to remove Q14 as it did not load significantly onto any of the first five factors.
Further analysis revealed that the Self-Confidence factor could be made more reliable by discarding Q10 (Cronbach's α increase from 0.92 to 0.96).While this is only a small increase in reliability, we decided that the subject matter of Q10 (self-consciousness) was unlikely to be of importance in most otolaryngological interventions. Removing Q10, along with Q9 and Q14 would then result in five factors, each with three question items, and the revised GBI would then be more "balanced" with each question item contributing to the same degree to the overall score and the factor score.
The next analysis was to measure the change in the "Total" GBI scores when the results of questions 10 and 14 were removed in addition to question 9. The Total GBI score with 15 questions was 5.8 ± 17.1, which is not clinically of any material difference from the Total GBI score of 18 questions of 5.6 ± 17.0.

| Scale reliability
Based on analysis of the full active interventions sample of 1980 patients, Cronbach's α was found to be 0.862. Item total correlations (see Table 2) were all greater than 0.3 apart from two questions in the Support factor: Q11 "people really care about you" and Q15 "support from your family." For both of these questions, alpha increased if the item was deleted, although the difference in each case was small. Removing these questions would remove the Support factor TA B L E 2 Reliability analysis of the 15 items in the GBI-5F entirely so the decision was made to retain them (see Discussion for justification of the inclusion of the Support factor).

| New five-factor Glasgow Benefit Inventory (GBI-5F)
Deleting Q9, Q10 and Q14 leaves us with a new 15-item questionnaire as shown in Appendix S1, with three questions loading onto  Table 3.

| Comparing the original GBI outcomes with the new GBI-5F outcomes using two interventions
The distribution of scores for patients who underwent tonsillectomy and first hearing aid fitting in the SENTOS data set is shown in Figures 1 and 2

| Synopsis of key findings
• A confirmatory analysis was performed on the original 18-question GBI using a large data set covering all otorhinolaryngological interventions from a prospective, national study.
• One question was removed at the start of the analysis because it had not been completed by 8% of respondents, most likely because the question was about "ability to work." Such a question would not be relevant to a high proportion of ORL patients. Its removal made no material difference to the total GBI score.
• Exploratory factor analysis identified five factors, each consisting of three questions. These were named; Quality of life, Selfconfidence, Social involvement, General Health and Support. The first three were previously combined as the General subscale while the latter two were already sub-scales. Removal of the two

| Main implications of implementation of 15 questioned GBI with five factors
The major strength of the new GBI lies in the five-factor structure which provides much richer data for interpretation. Just on the two examples shown in Figures 1 and 2, it can be seen that the pattern of factor scores varies considerably between the two interventions. For example, General Health has a large positive effect from tonsillectomy that is absent after hearing aid fitting, as one might expect. That tonsillectomy produces a greater benefit for Social Involvement than hearing aid fitting is unexpected and interesting.

| The overall role of the GBI
The GBI is a "generic" outcome measure, meaning that it does not have any system-or symptom-specific questions. It is often useful, therefore, to accompany it with other condition-or symptom-specific questionnaires.
The main generic alternative for ORL conditions is the SF-36. 5 This was rejected for use in the SENTOS for many reasons. It has eight factors (vitality, physical functioning, general health perception, physical role functioning, emotional role functioning, social role functioning and mental health) the majority of which one would not expect to be modified by the majority of ORL interventions. There is a shorter SF-12 6 that has single questions for these factors and is only able to be reported as a total score. This is not informative of the areas of benefit.
The HUI-3 was used in the SENTOS study alongside the GBI and the results have already been reported. 3 Since the HUI-3 has a specific section on hearing, it is not surprising that significant health benefit was shown for patients undergoing ear surgery or provision

F I G U R E 1
The new GBI-5F total and factor scores for patients undergoing tonsillectomy (n = 94) in the SENTOS data set, with the equivalent total and factor scores from the old 18-item, three-factor version of the GBI for comparison. The scores from the new 15-item, five-factor version are shown in red, and those from the old 18-item, three-factor version are shown in blue (and labelled "old 3FM" as an abbreviation for three-factor model). The total scores are shown first on the left. Support is mathematically identical to Social support from the old three-factor model, but we have chosen to rename it. Similarly, General health is mathematically identical to Physical health from the old three-factor model, but again we have renamed it. The General factor score from the old three-factor model has now been broken up into Quality of life, Self-confidence and Social involvement and it can be seen how these add new information. It is important, then, that the GBI-5F provides reliable and useful information in the individual domains that are contained within it. In seeking to provide a more robust and justifiable factor structure, the risk is of creating a completely new instrument that renders previous studies obsolete or uninterpretable. The GBI-5F does not do this.
Psychometric rigour is important but needs to be carefully weighed alongside the relevance of the information and the ease of clinical interpretation.

| Strengths and weaknesses of the study
The SENTOS data set covers a very large number of adult patients commonly accepted criteria for a robust factor but we made the decision to keep Social involvement because we feel that it contains

F I G U R E 2
The new GBI-5F total and factors scores for patients undergoing their first provision of a hearing aid (n = 416) in the SENTOS data set, with the equivalent total and factor scores from the old 18-item, three-factor version of the GBI for comparison information that is useful and informative in a clinical context. As an example, it can be seen from Figures 1 and 2 that tonsillectomy produces some improvement in Social involvement in a way that hearing aid provision does not. This is unexpected and worthy of further study.
The factor Support is also of concern from a psychometric point of view, in that two of its three questions have a poor correlation with the remainder and the overall internal consistency of the GBI would be improved by their removal. This would remove the Support factor completely, which is something we have The fact that the Support factor contains two questions items that have a poor correlation with the remaining items suggests that the total GBI-5F score should be used with caution because the data do not support the assumption that all questions relate to a single, coherent, underlying concept of "benefit." There is a richness of information in the five factors which should be reported separately.
Clinicians find a total score intuitive and every published paper so far has reported a total score, so it is likely that it will continue to be used, but we would suggest that the emphasis should always be on the factor scores.

| Advantages of the new five-factor structure
When we examine Figure 1 using the original three-factor analysis, it can be seen that the main benefit of tonsillectomy is an improvement in General Health with no real change in the Support score. These two factors remain in the new five-factor analysis, but the previous General factor is broken down into three new factors of Quality of Life, Self-confidence and Social involvement.
This adds information that, along with their better general health, they become more socially involved. As younger people (mean age of 29 years), the lack of benefit regarding self-confidence and the gaining of more support from others can be noted, as it might not have been anticipated. An important benefit of the new, shorter GBI-5F is that removing Q9 eliminates a major source of incomplete data as patients not in current employment often struggled to know how to complete the item and therefore left it blank. Many then failed to complete the rest of the questionnaire, resulting in large amounts of missing data.

| Clinical applicability of the study: recommendations for reporting the GBI-5F
1. If an 18-question GBI has been used, it should be reduced to a 15 question GBI-5F with the deletion of Q9, Q10 and Q 14 and consequent renumbering of the remaining question items in italics q1-15 (Appendix S1 and Table 3).

2.
With GBI-5F questionnaires, the data are unlikely to be normally distributed, so the mean and standard deviation are unlikely to be good descriptors for the data. The factor scores are reported as medians with quartiles and ranges.
3. Total GBI-5F scores should be used with caution if at all. The emphasis should be on reporting the benefit scores for each of the five factors and interpreting these individually and alongside each other.
4. If comparisons require to be made between the original threefactor GBI and the new GBI-5F (for example, in a systematic review), then the following guidance can be followed. The 15-item total score (still scaled from −100 to +100) differs little from the total 18 item score and all previous studies reporting total scores can still be interpreted and compared. The fifteen individual question items retained in the GBI-5F have not been changed in any way and can still be directly compared, as can the two factors Support (which is identical in all respects except name to Social support in the original GBI) and General Health (which is identical in all respects except name to Physical Health in the original GBI). The three new factors Quality of life, Social involvement and Self-confidence should be reported where available. Taking the arithmetic mean of these three new factor scores will produce a close approximation to the old General factor from the original three-factor GBI if such a comparison needs to be made.

ACK N OWLED G EM ENTS
Professor Michael A Akeroyd and Dr Graham Naylor are thanked for supporting the analysis within the Glasgow department. Ms Suzanne Ross performed many of the original analyses. Ms Jaclyn Farrell performed some intervention analysis. Mr Iain RC Swan clarified several aspects of the SENTOS data.

CO N FLI C T O F I NTE R E S T
None of the authors has any conflict of interest to declare.

AUTH O R CO NTR I B UTI O N
GGB initiated analysis, supported the development of the different themes in interpretation and wrote substantial parts of the paper.
HK vetted previous analyses and re-performed many of these using stricter criteria. He was substantially involved in writing the results and discussion section of the paper. WW reviewed the analyses and contributed to writing the paper.

E TH I C A L A PPROVA L
This is a reanalysis of previously collected data from a large multicentre study (SENTOS) which was done with Scottish Multi-centre Research Ethics approval. All data are anonymised. No new approvals are required for this study and it raises no new ethical issues.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.