More comprehensively measuring quality of life in life-threatening illness: the McGill Quality of Life Questionnaire – Expanded

SRC, AL, and the late Terry Bunston, PhD, with feedback from a larger research team, developed initial items to capture three new domains (Environment; Cognition; Healthcare) and the construct of (feeling like a) Burden. These were combined with select MQOL items to create a 32-item questionnaire. Like MQOL and MQOL-R, each MQOL-E item has a 2-day timeframe. Sixty palliative care patients from inpatient and home care settings in three Canadian cities and 10 palliative care clinicians then rated each item for importance to patient QOL (0–10 scale) and indicated items they found unclear (criterion of ≤10% for retention), redundant, or upsetting. For example, an earlier item stem, “I feel like a burden”, was upsetting and unclear (“burden to whom?”) to some participants, but they deemed the content important. It was revised to “I felt badly about how my situation affected the people I care about”. Rejected items were replaced. The new items were tested with another 63 patients, who completed the revised 26-item questionnaire to evaluate item distribution and again assess clarity and redundancy. New revisions resulted in a 20-item questionnaire. A validation study with this new questionnaire found the factor structure to be unsatisfactory since items representing different domains did not load as predicted (N = 304; unpublished). Various new changes and 11 experimental items, including one for the new financial domain, were explored in 23–27 item questionnaires in the three studies that provided datasets for this report. Since other instruments were completed at the same time in each of these studies [13, 20], to respect the physical condition of the participants, not all potential replacement items could be tested at once. To create the domain of quality of healthcare, Study G (the chronologically earliest dataset used in this report) used the two items from the immediately preceding 20-item version and also tested three new items, since we were hoping to create items with improved distributions. The new items, which did not have substantially improved distributions, were not included in subsequent studies. Only the two older items, which had been reviewed by patients during the development of the 20-item version and for which more data were available (since they were used in studies E and F), were included in the present analyses.

Items were initially developed in English, then translated into French by a professional translator. These two versions were compared by bilingual patients and clinicians; slight discrepancies between the two versions were reconciled. Sometimes this required a different word in French, other times the French translation suggested that the English wording needed clarifying, which was done. These changes did not affect the initial intent of the item. The resulting French version was successfully professionally back-translated into English with no further changes required. Both language versions were used in collecting datasets E and G, where participants chose the language version they preferred. Study F used only the English version since it was part of a larger study conducted in English in a city where English is the language of daily interaction.

To develop the final version of MQOL-E, we tested the seven new items representing the new domains and burden construct described in the Introduction and shown in Table 1 (the original 11 minus the four for quality of healthcare that were removed as explained above). As with MQOL-R, all items are measured on an 11-point scale from 0 to 10, anchored as indicated.

The datasets used in the analyses reported here (E-G) were collected from people with life-threatening illnesses at all stages of the disease trajectory or aged > 80 years. They were recruited from 15 Canadian healthcare sites in seven provinces. Study E (N = 219) was designed to test the construct validity of a 24-item pre-cursor instrument to MQOL-E [13]. The primary purpose of Study F (N = 368) was to develop a self-report instrument to measure satisfaction with healthcare. A 23-item pre-cursor instrument to MQOL-E was used in that study to assess the construct validity of the satisfaction with care instrument [20]. Study G (N = 216) used a 27-item pre-cursor to MQOL-E in a longitudinal study of the QOL of palliative care patients and was also intended to be used to assess construct validity (unpublished). Table 2 provides demographic information about the participants in each study. Datasets E and G include only people with cancer, while F includes people with various end-stage diseases and/or advanced age. Participants were recruited from a range of settings: palliative home care and inpatient units, acute care units, and oncology outpatient clinics. In studies F and G, the mean age was over 65; similar numbers of women and men participated. Study E recruited from oncology clinics, where the participants were somewhat younger (mean 58 years) and 64% were women. More detailed information about the three study samples is provided in Table 2.

In studies F and G, written informed consent was provided; participants had the choice of completing the questionnaire on their own or as a structured interview. In study E, with permission of the Research Ethics Committee, potential participants were provided with an information sheet and questionnaire to be completed and returned anonymously; consent was implied if the questionnaire was completed. All of the studies for item development and the current analysis were approved by the research ethics boards of all of the institutions from which patients were recruited.

Candidate items and domains were tested as additions to MQOL-R in four steps by conducting Confirmatory Factor Analyses (CFAs) using maximum likelihood robust estimation. The procedures, criteria for determining model fit, and goal of each step are described in Table 3. Analyses were conducted using the Mplus software [21]. Model fit was assessed via the Comparative Fit Index (CFI; >.95 indicates good fit) and the Root Mean Square Error of Approximation (RMSEA; <.06 indicates good fit) using criteria from Hu and Bentler [22]. The chi-square test was not used as it is sensitive to sample size [23]. To ensure that the items intended to represent new domains represented a construct different from existing MQOL-R domains, we first tested the factor loading of each item on existing domains. A standardized factor loading ≥ .6 was considered to form part of an existing domain. This is higher than the usual minimum criterion of ≥ .4 [24] because these items were designed to measure different aspects of QOL than the MQOL-R domains, therefore, adding them to an existing domain instead would require strong theoretical support. Nevertheless, items with loadings between 0.4 and 0.6 were closely scrutinized for conceptual fit on existing MQOL-R domains. To assess whether it is justified to create a summary score of all of the domains, the usual criterion of a factor loading ≥0.4 was used. Missing data were accommodated via full information maximum likelihood estimation. The total percentage of missing data was 18% when including all variables and 13% when including the variables in the final model.

Although McCaffrey et al (2016) included the concept of ‘feeling like a burden’ in the social domain (as did Cohen and Leis [8]), we were not certain whether it belonged in the social, psychological, or existential domain, or whether it would need to represent a separate domain. Assessment of this during creation of MQOL-R indicated it did not load sufficiently on any of the MQOL-R factors to justify its inclusion on any original domain, therefore it was included in the MQOL-E analyses as a separate factor. To assess the relevance of the financial domain, models both with and without the financial domain item were tested to determine the extent to which it made a unique contribution to the measurement of QOL.

MQOL-E subscale scores are calculated as the mean of the items measuring that domain. The MQOL-E summary score is the mean of the subscale scores (it does not include the global item). The Pearson correlations between the latent domains, second order factor, and a global measure of QOL were calculated for descriptive purposes. Cronbach’s alpha (α) was calculated to assess internal consistency reliability; an alpha ≥0.7 is considered acceptable [25].