Development of a Health Rating scale – Emoticon Health Rating Scale (EHRS) – Selection of items
Integrated Study Report on Study 1:1 and 1:2
Stress related illness constitute an overwhelming majority of sick leave in the Swedish working population. Needless to say the cost for society, employer and employee is enormous.
The overall aim of the current project is to develop a fast, reliable, and web-based health rating scale with the purpose to monitor the respondent’s health, giving them the possibility for self-reflection and thereby preventing or minimizing stress related illness. An analysis module will be couple to the instrument for further data analysis of the respondent’s health. This module will also include a system indicating of an individual’s improvements or deterioration in health (e.g. person’s entering into a risk zone for burnout).
Investigators and Study Administrative Structure
The study was conducted by Nextconsulting in cooperation with PFM Rese
The study was financed and administered by Hälsokällan/Ginsam AB Sverker Månsson
Thorsten Klint, Ph.D.; Associated Professor
Göran Granath, Ph.D
Date of study 1:1
Study was performed from 21 to 26 January 2016
Date of study 1:2
Study was performed from 25 to 26 April 2016
To select variables/questions from the literature reflecting different dimensions in the spectrum of health versus illness. These variables are submitted to a web-based panel in order to find out which variables are correlated and thus give similar information. That in contrast to variables which are not correlated and thus could possibly give other types of information. The purpose of the first phase is mainly to reduce the number of variables assessed useful for the final rating scale, by empirical testing in a panel of test persons, and subsequent analysis by multivariate statistics.
Overview and Rationale
Study Design Our aim is to reflect health from a salutogenic perspective and thereby using the WHO´s original definition of health ”Health is a state of complete physical, mental and social well-being and not merely the absence of disease or infirmity.” The intention was to select variables reflecting different aspects of health, like mood, sleep, stress and physical activity, from validated rating scales in the literature. A major problem during the selection was that a large majority of rating scales are concerned with a disease perspective. Despite that we were able to identify a number of domains reflecting health and the kind of variables likely to reflect health from a salutogenic perspective. A major source in the identification of rating scales was Measuring health by McDowell (2006), and about 50 other references were also consulted. In the process, to develop a web-based health rating scale for self- monitoring, we first selected about 40 variables (questions) from the literature (only 35 were used in the first study, because 5 variables were missing for technical reasons) and 47 in the second study; original variables plus some extra variables to cover domains we though were of special interest. The next step was to test the selected variables in a web-based panel to find out how the variables were correlated and clustered.
A telephone recruited web panel from Nextconsulting was used, where each participant had to rate his/her health status at one occasion. The participants were randomly selected from a pool of 70 000 persons living in Sweden, with a representative distribution of the Swedish population concerning age, sex, education, geography etc.
Variables for Evaluation
The purpose of this study was to determine which variables are correlated and give similar information and which variables are not correlated and could thus possibly give other types of information. The purpose of the first step is mainly to reduce the number of variables useful for the final rating scale, by means of multivariate statistics. The variables used are described in Table 1 and 2. Baseline information about perceived general health, age, gender, education, working situation, and physical exercise were collected.
The variable layout was designed as in fig 1.
No formal sample size calculation was performed, but the number of subjects intended for enrollment was set to 3 times the number of variables (questions in the questionnaire).
Descriptive statistics of age, gender, education etc. was done in order to get a general view of the populations in the two studies.
The general way of analysis was first to study the correlations between the variables, ie the questions and the background data (age, gender etc.). After that PCA and factor analysis was done. The factor analysis was based on PCA and completed with ‘Varimax’ rotations. That was done in order to find meaningful underlying factors. Non-hierarchical cluster analysis, in variable space, was used in order to group the variables in the number of clusters obtained from the factor analysis. In each cluster, the variable closest to the cluster centre, was chosen. These ‘type variables’ were the subject to different analyses like correlations and PCA.
All statistical programming with plots was done using version 13 of the STATISTICA software produced by Dell.
Disposition of Subjects
In study 1:1 121 participants were included. 16 participants never finished the questionnaire and 3 of them scored “I don’t know” in more than three question and was subsequently disregarded from the analysis. In study 1:2 125 participants were included. 20 participants never finished the questionnaire and 3 of them scored “I don’t know” or nothing in more than three question and was subsequently disregarded from the analysis.
Primary Objectives Study 1:1
In the first place all 35 variables were subject to a factor analysis. Three factor loadings were isolated which explained about 21 % of the total variance (Table 3). Thereafter the variables were Varimax-rotated and three new factors isolated. Out of the 35 variables factor-loadings > 0.65 were used as a cut-off, which generated 14 variables (Table 4). These variables were then used in a PCA (Table 5). At least three variable groupings were identified. A tentative labelling of the groups could be: 1. Stress variables, 2. Mood variables and “feeling good“ variables. The fourth grouping consists of variables reflecting “meaning of life” (Fig 1). The same grouping was confirmed by a hierarchical cluster analysis (Wards method) (Fig 2).
Study 1:2 In the first place all 47 variables were subject to a factor analysis. Three factor loadings were isolated which explained about 29 % of the total variance. Thereafter the variables were Varimax-rotated and three new factors isolated. All variables were then used in a PCA (Fig 3). A number of tentative variable groups could be suggested, but the grouping was not very distinct since a large number of variables very highly correlated. Therefor a none-hierarchical cluster analysis was used as an alternative strategy. 9 clusters were used in the final analysis, but with variable Q32.2 (Variable 6 in Table 1; erotic feelings) excluded since it was it was significantly different between genders. Nine variables were selected based on the importance in the cluster analysis (Fig 4). Indicator variables are marked with yellow in Table 5. The same grouping was also illustrated by hierarchical cluster analysis (Wards method) (Fig 5).
Comparisons between study populations
Using the common 35 variables the distribution of background variables was compared between the two studies. There were no significant differences concerning gender, age, working situation and sleep, however, there were significant differences concerning levels of education and physical exercise. The number of participants completing gymnasium was higher in the second study but the number of participants with > 2 years university education was highest in the first study. Comparing the variables (questions) between the two studies only two variables were significantly different (analysis of variance). Variable Q37.2 (15) ”Today I feel stressed” and Q37.5 (18)
The overall aim of the first phase was mainly to reduce the number of variables useful for development of a fast and reliable rating scale reflecting a subject’s general health versus illness. This was done by letting persons from a web-based panel scoring the variables at one moment in time to find out which variables potentially were correlated or divided in clusters possibly giving different kind of information. The first study showed that a large number of variables, as expected, were correlated but also that clear clusters emerged based on the PCA and cluster analysis. Interesting and important was that the correlations and groupings seemed to make common sense. In the second study the number of variables was increased from 35 to 47 to cover more aspect of the health spectra and compensate for 5 variables missing in the first run. In the first PCA most variables were very well correlated except erotic feelings (Q32.2) and the two stress-related variables (Q27.3 and 37.2). Q27.1 physical activity could possibly also be considered to separate from the rest in p2. Both the none-hierarchical and hierarchical cluster analysis and gave a basis for nine useful variables to be used in further testing. To be noticed the question about erotic feelings was not included in this analysis. To want extent the differences in background variables education and physical exercise between the studies reflect on differences in the variables “Today I feel stressed” and “Today I have had time for myself” is difficult to evaluate.
The overall aim to reduce the number of variables from the original 35 and 47 was substantiated by the two studies. The analysis gives a firm ground for further studies with approximately 9 variables or more. The data shows reasonable similarity between studies and the clustering of variables describes domains of health versus illness which make intuitive sense.
Antonovsky A. Health, Stress and Coping. San Francisco: Jossey-Bass; 1979
Antonovsky A. Unraveling the Mystery of Health. How People Manage Stress and Stay Well. San Francisco: Jossey-Bass; 1987.
Lundgren-Nilsson et al. Internal construct validity of the Shirom-Melamed Burnout Questionnaire (SMBQ). BMC Public Health 2012, 12:1
McDowell I. Measuring Health: A Guide to Rating Scales and Questionnaires, OXFORD UNIVERSITY PRESS, Third Edition 2006 (Book)
Nutbeam, D. Health promotion glossary. Health Promotion, 1, 113–127, 1986.
Smith B. J. , K. C. Tang and D. Nutbeam. WHO Health Promotion Glossary: new terms. Health Promotion International, Vol. 21 No. 4, 2006.