Introduction
Clinical, research, and administrative interest in quality of life (QOL) has increased over the past three decades. QOL is generally defined as an individual’s subjective, holistic view of life circumstances across physical, psychological, and social domains [
1‐
4]. It is a valuable predictor of patients’ overall health status, their perceptions of health services, and how to improve those services [
3,
5,
6]. To that end, measures of patient-reported outcomes that emphasize patients’ subjective perspectives like QOL are increasingly used in psychiatry [
6] and relevant given the growing attention to self-defined recovery as the goal of healthcare [
7]. Many tools that measure QOL have been developed and examined to determine their reliability, validity, feasibility, and sensitivity to change [
8].
One such self-report tool is the Quality of Life Enjoyment and Satisfaction Questionnaire (Q-LES-Q), designed to measure a patient’s satisfaction and enjoyment in different areas of daily functioning. The original scale consists of 93 questions, which were grouped into eight subscales on the basis of expert clinical opinion: physical health, subjective feelings, leisure time activities, social relationships, work, school/coursework, household duties, and general activities [
9]. The abbreviated version (Q-LES-Q-SF) consists of 14 items derived from the long form’s general activities subscale, plus two questions about medication and overall life satisfaction. Both versions are among the most frequently used QOL measures in psychopharmacology and clinical trials [
10], and have been translated into several languages.
A number of studies have assessed the reliability, validity, and factor structure of the Q-LES-Q and Q-LES-Q-SF to date (see Table
1). Notably, there is little consensus regarding the factor structure for the Q-LES-Q-SF [
10], with research indicating one factor [
2,
9], two factors [
5], or an equal chance of one or two factors [
11]. Psychometric studies have involved work with versions in Chinese, French, and other languages and cultures, which may explain the lack of consensus in the results [
2,
3]. A variety of methods (exploratory factor analysis using Pearson’s or polychoric correlations, principal components analysis, confirmatory factor analysis, structural equation modeling) may have contributed to diverse findings [
2,
5,
9,
11]. Moreover, investigations of the psychometric properties of the Q-LES-Q-SF in English have been sparse, and none have addressed its factor structure.
Table 1
Review of peer-reviewed literature on the Q-LES-Q (all forms)
Short form: Q-LES-Q-SF |
| English US | C, O, O (mail) | 67 | 32.4 | 65.8% | 0.90 | – | At 1–2 weeks (ICC = 0.86)c | – |
| English US | Mx, Mx, I | 150 ADHDd | 36.2 ± 10.8 | 46% (n = 81) | 0.88 | – | – | SAS total T (r = 0.72) |
134 Non-ADHD adults | 30.0 ± 8.7 | 53% (= 63) | 0.84 | – | CGI |
173 Placebo | 37.1 ± 9.5 | 47% (n = 91) | CGI |
| English US | O, P, I | 2588 | 40.0 ± 12.1 (Range 18–65) | 64.6% | 0.86 | – | – | HAM-A, CGI-S, MADRS, PSQI: Week 1(r = − 0.25 to − 0.36), Week 8 (r = − 0.25 to − 0.36) |
| Serbian Serbia | I, P, I | 57 | 47.16 ± 9.22 | 33.3% | 0.90 | All items but work (3) correlated to total (r = 0.41–0.81) | At 1 week (ICC = 0.93, n = 54) | CGI-S (r = 0.89) PGIs (r = 0.43) CGI-I (r = 0.47) |
| Chinese Taiwan | O, M, I | 1482 | 54.65 | 58.57% | 0.87 | EFA, CFAe 2 factors Psychosocial (eigenvalue = 5.24, 37.41% variance) Physical (eigenvalue = 1.27, 9.07% variance) | At 2 weeks (ICC = 0.75, n = 199) | SF-12 (r = 0.35–0.38) PHQ-9 (r = 0.37) HAM-D-17(r = 0.49)
Psychosoc. subscale: MCS-12 (r = 0.37) PCS-12 (r = 0.27)
Physical Subscale: MCS-12 (r = 0.28) |
| English Australia | O, Mx, I | 843 | 32.06 | 78.4% | 0.88 | – | – | – |
| English US | E, P, I | 4041 | 42.6 ± 13 | 62.8% | 0.87 | – | (ICC = 0.74) | – |
| French France | O, S, I | 124 | 39.2 ± 11.2 Median 36 | 16.7% | 0.90 | CFA 1 factor: (RMSEA = 0.077) (90%CI [0.054–0.098]),CFI = 0.968 TLI = 0.962, loadings (r = 0.523 and 0.851) CFA 2 factors: (RMSEA = 0.076) (90%CI [0.054–0.098]), CFI = 0.969 TLI = 0.963, loadings (r = 0.525 and 0.870) (a) PSI high for unidimensional construct (IRT = 0.902, item residuals − 2.5 to + 2.5) All items significantly correlated to total (r = 0.47–0.79) (b) |
Long-form Q-LES-Q | | |
| English US | O, P, I | 95 | 39.1 ± 10.7 Range 18–63 | 59% | – | General activities (GA)f (α = 0.90, n = 83) Other subscales (α = 0.92–0.96) | GA (r = 0.74, n = 54) Other subscales (r = 0.63–0.89) | HAM-D: (r = − 0.61), SCL-90 (r = − 0.67) BDIb (r = − 0.36)
Subscale level
CGI-S: (r = 0.34–0.68) GA (r = − 0.66) |
| English US | I, P, I | 151 | 18–65 | 55% | 0.6 | PCAg on 44 items from 4 original domains confirmed 4 subscales Physical health (α = 0.93), Subjective feelings (α = 0.95,) Leisure activities (α = 0.89), Social relationships (α = 0.89) |
| Italian Italy | O, P, I | 404 | 38.9 ± 12.3 | 55.9% | – | GA (α = 0.92) Other subscales (α = 0.82–0.96) | At 1 week (ICC > 0.8, n = 62) GA (ICC = 0.89, n = 62) | WSAS (r = − 0.56) |
| English US | C, O (healthy), I | 529 | 33 Range 18–84 | 58% | – | GA (α = 0.90, n = 69); correlated with other subscales (r = 0.41–0.62); Other (α = 0.82–0.93) | GA (ICC = 0.86, n = 69) Other subscales (ICC = 0.58–0.89, n = 69) | – |
| Portuguese Brazil | I, S, I | 100 | 20.1 ± 6.2 | 22% | – | GA (α = 0.84) Other subscales (α = 0.78–0.93) |
WHOQOL-BREF areas with subscales
Physical: physical (r = 0.46), feelings (r = 0.54), leisure (r = 0.28), GA (r = 0.6) Psychological: physical (r = 0.4), feelings (r = 0.6), leisure(r = 0.34), social (r = 0.49), GA(r = 0.61) Social: physical (r = 0.3), feelings (r = 0.48), social (r = 0.44), GA (r = 0.48) Environment: physical (r = 0.37), feelings (r = 0.5), household (r = 0.3), social (r = 0.4), GA (r = 0.5) |
| Finnish Finland | I, P, I | 190 | 38 ± 13 (range 18–65) | 41% | 0.89 | One factor (41% variance) Three factors (56% variance) | EQ-5D (r = 0.445; p < 0.001) |
Other forms |
| Hebrew Israel | Mx, P, I | 339 Initial 133 Confirm. 175 control 133 Model construction 124 follow-up 38 Inpatient | 38.5 ± 10.3 39.6 ± 9.3 38.4 ± 9.9 39.6 ± 9.3 39.9 ± 9.4 36.6 ± 8.6 | 29.2% 23.3% 72.9% 23.3% 21.8% 18.4% | 0.92 (GA) | 18-Item scale of Long-Form EFA 4 factors (α = 0.74–0.97) Q-LES-Q domains (α = 0.69–0.87) | At 2 weeks: (ICC = 0.79–0.90) | QLS (r = 0.40–0.65, p < 0.001) (b) |
In addition, samples used for the majority of Q-LES-Q-SF factor analyses have been recruited during an episode of inpatient or outpatient care or when patients arrived at medical facilities for treatment. It is unclear whether the factor structure remains stable for populations who complete the survey removed from the point of care, that is, outside of a clinic and not necessarily actively care-seeking. Testing the factor structure on patients outside the point of care would thus entail analysis in a more diverse sample of patients. Measuring QOL within this population is important for population-based healthcare systems, such as national health services and accountable care organizations (ACOs) which must proactively manage patient care, as well as for payers and organizations monitoring healthcare delivery [
11,
12]. Furthermore, there are no psychometric data on telephone administration of this survey which allows for interactive self-reports outside the point of care. Given the variation in Q-LES-Q-SF results to date and the limitations in the populations studied, the purpose of our current study is to advance understanding of the possible reasons for previous findings of both uni- and bi-dimensional latent structures elsewhere by exploring possible Q-LES-Q-SF factor structures in a population of individuals outside the point of care enrolled in treatment. General mental health clinics across nine United States Department of Veteran Affairs medical centers (VAMCs) provide treatment. These analyses aim to contribute to understanding the dimensionality of the Q-LES-Q-SF in enrolled populations with varied psychiatric diagnoses and physical comorbidities and to the sparse literature on the reliability of the English language version.
Discussion
We investigated the factor structure of the Q-LES-Q-SF based on data collected during telephone interviews with 568 Veterans enrolled in general mental health services but not at the point of care at the time of survey completion. Notably, our sample had substantial physical and mental health burden (Table
2). Using EFA and MTA in a split-half approach to consider both uni- and bi-dimensional solutions, we identified single- and dual-factor solutions, the latter with a strong psychosocial factor (
k = 10) and a possible weaker physical health factor (
k = 3).
Interpreting our bi-dimensional results is informed by reviewing the two prior studies of Q-LES-Q-SF factor structure, one in Chinese [
6] and the other in French [
2,
16]. The former, using a primary care locus of recruitment, identified both a psychosocial and a physical factor, while the latter, recruiting from a substance abuse facility, identified a single, overall factor and a possible second factor. Ethnic/cultural variability in the expression of general mental illnesses, particularly major depressive disorder, is well documented; for example, compared to depressed Western populations, Chinese patients often endorse physical rather than psychological symptoms [
37]. Such tendencies may partly explain the difference in the Q-LES-Q-SF factor structure identified by Chinese and French studies. Further, the degree to which differences in language, culture, locus of sample recruitment, or a combination of these factors contribute to the differences in findings between those previous studies and the present one cannot be determined for certain.
Our results provide insight into this divergence of findings. Similar to both prior studies, we found a strong primary factor in our sample representing a psychosocial subscale. The physical health factor was much weaker and, in our estimation, equivocal. Recalling that the original Q-LES-Q-SF was constructed based on expert opinion without formal psychometrics, it is not surprising to observe some instability in factor structure across languages and populations. However, from our and prior [
2,
6] analyses, it is clear that a strong psychosocial factor can be distinguished. In fact, relying solely on the EFA results, one could make a strong case for a unidimensional factor—Factor 1 in this study—which also includes item 1 (physical health). Indeed, the MAP and Horn’s Parallel Analysis tests support the existence of a single factor.
In contrast, the presence of a separate, distinctly physical health factor is uncertain. The Q-LES-Q was initially reported in the psychopharmacologic literature [
9] and extensively used in medication treatment trials. The three specific physical health items included by experts (physical ability, vision, sex) correspond to side effects frequently encountered in the medications typically investigated at that time (tricyclic antidepressants and first generation antipsychotics), and this may be a reason behind the inclusion of these specific symptoms. Changes in medication usage and their side effects in the years since the introduction of the Q-LES-Q, combined with differences in population characteristics across studies, may mitigate the usefulness of these three items to distinguish differences in patient experience.
Nonetheless, the pattern of convergent and discriminant validity of the proposed Q-LES-Q-SF factors with the VR-12 suggest the possibility of a two-factor structure. The proposed Q-LES-Q-SF psychosocial factor correlates more strongly with the MCS than the PCS. Conversely, the proposed Q-LES-Q-SF physical factor correlates more strongly with the PCS than the MCS.
Correlations of the full Q-LES-Q-SF with the MCS and PCS reveal differences in relative strength that suggest a unidimensional interpretation that emphasizes the psychosocial content of the measure. The Q-LES-Q-SF overall score correlated strongly with the VR-12 MCS and less so with the PCS. Consistent with our findings, Bourion-Bédès and colleagues’ [
2] also found a strong correlation between Q-LES-Q-SF total score and the MCS from the SF-12 (from which the VR-12 derives), and weaker correlation with the SF-12 PCS. In Lee and colleagues’ [
6] study, their psychosocial factor correlates more strongly with the MCS, and the physical factor with the PCS as in Bourion-Bédès’ and our Western samples. The entire Q-LES-Q-SF in Lee’s study shows modest, nearly identical correlations with both the SF-12 PCS (
r = 0.35) and MCS (
r = 0.38).
All three studies, across three distinct populations, cultures, and languages, converge around a strong psychosocial factor. In the present study and that of Lee and colleagues [
6], this is complemented by a weaker, separate physical factor, while Bourion-Bédès and colleagues’ study [
2,
16] resulted in a single overall factor that is heavily psychosocially weighted.
Thus researchers and program evaluators can have confidence in using the Q-LES-Q-SF as a single, psychosocial factor. The two-factor solution can be used with little confidence due to equivocal support for a distinct physical factor in the measure as it currently exists.
Limitations
Our study has several limitations. We used Pearson’s rather than polychoric correlations to allow for closer comparison to Lee and colleagues (2014) bi-dimensional results. Future research in this area could focus on alternative methods utilizing polychoric correlations for ordinal scales. We also note that the Kaiser criterion and scree plot method used here to mirror the procedures used by Lee and colleagues (2014) and many others may lead to overdimensionalization, especially when Likert scales are involved [
24‐
26]. To at least partially mitigate this possibility, we also applied Velicer’s minimal average partial (MAP) and Horn’s parallel analysis (PA) and recommend similar multiple procedures be used in future research.
We also had high missing data rate for item 3 (work), likely due to high rates of retirement, disability, and unemployment in our sample. However, this has been found in other studies involving mental health and substance use populations [
2,
4,
10,
16,
38]. Following Bishop and colleagues’ [
5] example, we excluded questions related to work because morbidity rates in their populations were too high.
It is also possible that mode of administration affected results. Patients may be less willing to disclose about sensitive topics in real-time conversation compared to mail-out surveys. However, both Lee and colleagues [
6] and Bourion-Bédès and colleagues [
2,
12] administered surveys within clinics, which may feel even less anonymous than phone interviews. Additionally, paper administration for clinical purposes includes instructions on circling specific facets of the topic in question that cause dissatisfaction within items, i.e., Item 9: “sexual desire, interest, and/or performance” (Endicott, personal communication). When the survey is administered aloud, all three elements of sexual experience must be mentally combined in some fashion and judged together. Similar conflicting interpretations of questions with options may explain the low communalities of items 11 (“living
or housing situation?”) and 13 (“vision, in terms of work
or hobbies?”) and the unusually high communality for item 12 (“able to get around without feeling dizzy
or unsteady
or falling?”). Although the aspects of items chosen on paper do not affect the total Q-LES-Q-SF score, the options inherent in the items mean that these responses may be more variable than other items.
The higher proportion of males within our sample may have also affected how item 9 (sex) loaded in factor analyses. Even when oversampling for females, 79.6% of the participants were male (Table
1). For item 9 in particular, differential item functioning has been observed based on sex [
2].
Finally, our study may be limited in that results may differ in populations treated in systems different from the VA, a large, publicly funded healthcare system. Patients receiving care within the VA comprise an aging population with a high proportion of males who have high rates of physical comorbidities [
39]. These comorbidities are shown to contribute to perceptions of QOL and overall health outcomes [
40]. However, this type of population will become increasingly relevant as a higher proportion of the world-wide population ages and as general mental health care becomes more integrated into primary and specialty services [
41‐
43].
Conclusions
This study confirms that the Q-LES-Q-SF is a valid and reliable recovery-oriented self-report instrument within a general mental health population assessed not at the point of care. The factor structure may be best described as one clear psychosocial factor and one possible weaker physical factor, a structure which may vary according to factor structure extraction methods, treatment of Likert scales as ordinal versus categorical, degree of disease burden, culture, language, and mode of administration. While the psychosocial factor is notably stable across three populations, cultures, and languages, future research may reduce the instability of the second physical factor, perhaps even incorporating additional items. In addition, further assessment of the effect of administration mode (i.e., paper versus phone) on the Q-LES-Q-SF responses and factor structure is needed.
Based on current evidence, researchers and program evaluators can be secure in using the full Q-LES-Q-SF score or the single 10-item psychosocial factor. In the evaluation of interventions or other studies with a particular focus on the physical quality of life, breakout scoring of the physical subscale might be examined with considerable caution. However, in those circumstances where the assessment of physical quality of life is critical, the use of supplemental validated measures of that dimension is recommended.