Validity of the older people quality of life-7 domains (OQoL-7) scale

The OQoL-7 is a 28-item questionnaire that was developed to assess the multidimensional QoL of community-dwelling older people. The questionnaire begins with a brief description of QoL to ensure that respondents have a common understanding of the construct (see Additional File 1). Respondents are then asked to rate each item on its perceived importance for their own QoL (very low; quite low; quite high; very high) as well as on potential discomfort or dissatisfaction currently perceived (not at all; a little; a lot). Answers to perceived discomfort or dissatisfaction on the 28 QoL items are reverse coded to express satisfaction (not at all = 2, a little = 1, a lot = 0) and then summed up. This total is divided by the maximum possible total (number of completed items multiplied by 2) and then multiplied by 100 to yield a QoL score ranging from 0 to 100, with higher scores indicating higher QoL. The same procedure is applied on the constituent items of each QoL domain to obtain its specific subscore. The QoL score is considered as missing if more than half of the items are missing. For each of the seven domains, the QoL subscore is considered as missing if more than one of its constituent items are missing. A previous article focused on importance ratings [14], which were used in the present study only for weighting purposes (see paragraph “Weighting procedures” section).

Overall QoL was assessed by a single-item measure: “How do you rate your current quality of life? (excellent, very good, good, fair, poor)”. A single global rating of QoL is a simple and suitable instrument to measure QoL in a broad sense [11]. Self-rated health was reported as very good, good, average, poor, or very poor. A single question is a valuable indicator of the overall health status, and a strong predictor of morbidity and mortality [18, 19].

The SF-12 v2 was completed during an in-person interview conducted by trained medical research assistants. Norm-based physical (PCS) and mental (MCS) component scores were obtained using linear transformations (mean = 50; SD = 10) [20]. Norm-based scores of the eight SF-12 dimensions were also calculated (General Health, Physical Functioning, Role Physical, Role Emotional, Bodily Pain, Vitality, Mental Health, and Social Functioning).

Twenty events from the geriatric adverse events life scale (GALES [21]) were selected for their suitability in old age. Respondents were asked whether they faced any of these stressful events during the previous 12 months.

Information about respondents’ age and sex were obtained from the Population Office at the stage of study sampling and recruitment. Additional information was gathered by means of a postal questionnaire that provided information about highest level of education achieved (Basic compulsory (International Standard Classification of Education (ISCED [22]) level 0–2); Apprenticeship (ISCED level 3); Baccalaureate/professional degree (ISCED level 4–5); University/high school (ISCED level 6–8)) and living arrangement (Alone; With spouse (married or not); Other living arrangement). Information on household gross monthly income was collected during the in-person interview conducted by trained medical research assistants. It was divided by the household size (collected by postal questionnaire) to calculate gross monthly income.

The proportion of missing values was calculated for each QoL item, for the seven domain subscores and for the QoL score. The proportion of participants at the minimum (floor) and maximum (ceiling) values was also calculated.

In the absence of a gold standard, the OQoL-7 was compared to other measures of the same concept (single QoL item) or related concepts (self-rated health, SF-12). Spearman correlations were calculated between the QoL score and the seven QoL subscores on the one hand, and the single QoL item, self-rated health, SF-12 PCS, and SF-12 MCS, on the other hand. These analyses were conducted on the 2011 sample and repeated on the 2016 sample. Correlations were interpreted as small (> 0.1), medium (> 0.3), or large (> 0.5) [23].

The OQoL-7 was tested for construct validity against stressful events during the previous 12 months. Whereas most of these events are likely to influence several QoL domains simultaneously, two events were analyzed more in depth based on their expected impact on a specific QoL domain. A strong association was expected between (1) “Financial difficulties” and the QoL domain “Material resources”; and (2) between Separation, Divorce, or Other difficulties in the couple (labelled “Relationship difficulties”) and the QoL domain “Close entourage”. Spearman correlations were calculated between stressful events and QoL subscores in the seven domains. In addition, the mean and standard deviation of QoL subscores in the seven domains were calculated in the presence and absence of both stressful events. The effect size was estimated using Cohen’s d and was interpreted as small (> 0.2), medium (> 0.5), or large (> 0.8) as proposed by Cohen [23]. To estimate 95% confidence intervals (CI) for Cohen’s d, bootstrapping was performed using 200 bootstrap replications.

To further assess construct validity, Spearman’s correlations were calculated between each of the 28 satisfaction QoL items and QoL subscores in the seven domains. In this particular analysis, for each item, its corresponding QoL subscore was recalculated without the item itself (i.e. considering only the other items in the domain concerned). It was expected that each item would be most strongly associated with its corresponding domain.

In all analyses assessing data quality, concurrent validity, and construct validity, the OQoL-7 total score and the seven subscores were calculated without taking into account importance ratings. In the following analyses, four different weighting procedures were performed, each consisting in the multiplication of importance and satisfaction scores, divided by the sum of importance scores. In the first three methods, satisfaction was weighted by importance at the item level and importance was coded in three different ways. First, items whose importance was rated very low were given a zero weight (weighted QoL score 1: very low = 0; quite low = 1; quite high = 2; very high = 3). The particularity of this method is that items deemed of very low importance have no influence on the total score. Second, items importance had still a linear increasing weight (weighted QoL score 2: very low = 1; quite low = 2; quite high = 3; very high = 4). This method makes the assumption of constant intervals between importance ratings. Third, items importance had a quadratic increasing weight (weighted QoL score 3: very low = 1; quite low = 4; quite high = 9; very high = 16). Like in method 2, the influence of items on the total score increases as their importance increases, but this increase is not linear. In other words, in method 3 the total score is even more influenced by items with a high importance than in method 2. In the fourth method (weighted QoL score 4), satisfaction was weighted by importance at the domain level. The seven domains’ satisfaction subscores were first multiplied by their respective importance subscores according to participant’s rating, then summed-up, and finally divided by the sum of importance subscores. Spearman correlations were calculated between weighted and unweighted QoL scores, and between weighted QoL scores and the single QoL item. In addition, moderated regression analysis was conducted [16]. It consisted in regressing the single QoL item on satisfaction subscores of the seven QoL domains (step 1), then additionally on importance subscores of the seven QoL domains (step 2), and on the seven interaction terms of satisfaction by importance (step 3). A significant increase in explained variance (R²) from one step to the next step was tested using an F-test. The magnitude of this increase was interpreted as small (> 0.02), medium (> 0.13), or large (> 0.26) as proposed [23].