NIPSA: a new scale for measuring non-illness predictors of sickness absence
A/Prof Samuel B Harvey 1, 2, 3,*
Ms Min-Jung Wang 2, 4
Dr Sarah Dorrington 1
Dr Max Henderson 1, 5
Dr Ira Madan6
Dr Stephani L Hatch 1
Prof Matthew Hotopf1, 7
1. Institute of Psychiatry, Psychology & Neuroscience, King’s College London, London, UK
2. School of Psychiatry, University of New South Wales, Sydney, Australia
3. Black Dog Institute, Sydney, Australia
4. School of Public Health, Harvard University, Boston, USA
5. Leeds & York Partnerships NHSFT, Leeds UK
6. Guy’s and St Thomas’ NHS Trust and King’s College London
7. South London and Maudsley NHS Foundation Trust, London UK
* Corresponding Author
A/Prof Samuel B Harvey
Head of Workplace Mental Health Research Program
Black Dog Institute
University of New South Wales
Hospital Road
Randwick NSW 2031 Australia
Word count: 2968
ABSTRACT
Objectives
We describe the development and initial validation of a new scale for measuring non-illness factors that are important in predicting occupational outcomes, called the NIPSA (non-illness predictors of sickness absence) scale.
Methods
Forty-two questions were developed which covered a broad range of potential non-illness related risk factors for sickness absence. 682 participants in the South East London Community Health (SELCoH) study answered these questions and a range of questions regarding both short and long term sickness absence. Factor analysis was conducted prior to examining the links between each identified factor and sickness absence outcomes.
Results
Exploratory factor analysis using the oblique rotation method suggested the questionnaire should contain 26 questions and extracted four factors with eigenvalues greater than one; perception of psychosocial work environment (factor 1), perceived vulnerability (factor 2), rest-focused attitude towards recovery (factor 3), and attitudes towards work (factor 4). Three of these factors (factors 1, 2 and 3) showed significant associations with long-term sickness absence measures (p<0.05), meaning a final questionnaire that included 20 questions with three sub-scales.
Conclusions
The NIPSA is a new tool that will hopefully allow clinicians to quickly assess for the presence of non-illness factors that may be important in predicting occupational outcomes and tailor treatments and interventions to address the barriers identified. To the best of our knowledge, this is first time at scale focused on trans-diagnostic, non-illness related predictors of sickness absence has been developed.
Key words
Sickness absence, return to work, psychosocial work environment, vulnerability, rest, recovery, work, occupational outcomes
SUMMARY BOX
What is already known about this subject?
Regardless of the underlying medical diagnosis, symptom severity alone is not a strong predictor of occupational outcomes, such as sickness absence. There are a range of other factors relating to individual perceptions, response to symptoms and the workplace that can influence an individual’s sickness absence behavior.
What are the new findings?
We have described a new measure, called the NIPSA (non-illness predictors of sickness absence) scale, which consists of 20 questions that can provided reliable measures on three subscales; perception of psychosocial work environment (factor 1), perceived vulnerability (factor 2), and rest-focused attitude towards recovery (factor 3). We have also demonstrated that how an individual’s scores on each of these factors is associated with measures of long term sickness absence behaviour.
How might it impact on clinical practice in the foreseeable future?
Once this new scale has been further validated, it is hoped that clinicians will be able to use NIPSA to identify non-illness factors that are risk factors for short-term episodes of sickness absence progressing to long-term sickness absence or recurrent episodes of sickness absence. Specific interventions can be mapped onto each of the three factors described.
INTRODUCTION
Sickness absence is a major public health and economic problem across the developed world.1Over recent decades there has been a gradual change in the ascribed medical causes of sickness absence, with mental disorders now being the leading diagnosis in cases of long-term sickness absence and disability benefits in most developed countries.2-4 Regardless of whether an episode of sickness absence is due to mental or physical illness, there is increasing evidence that objective measures of symptom severity are not a reliable predictor of the duration of any period of sickness absence.5-7 Indeed, at a population level over the last century, rates of incapacity benefits and long term sickness absence have tended to move in the opposite direction to most overall measures of population health.8 These observations indicate that sickness absence is a complicated behavioural response that can be influenced by many factors other than the medical diagnosis and simple symptom levels.4
Researchers studying individual disorders have begun to identify a number of non-illness related predictors of sickness absence relevant for each disorder. For example, amongst those with back pain, fear-avoidance beliefs and behaviours have been found to consistently predict sickness absence, even after controlling for pain intensity and physical disability.9 Avoidant behaviour in response to symptoms and rest-focused attitudes to recovery havealso been found to predict long-term sickness absence amongst those with chronic fatigue.5Life course studies, which have followed individuals from childhood through to later life haveidentified a number ofother individual factors, such as temperament, perceived health vulnerabilityand general intelligence,that are predictorsof occupational incapacity in later life, independent of the level of physical or mental health symptoms.10-13
Not surprisingly, a range of workplace factors have also been identifiedthat appear to influence rates of sickness absence. In the occupational health literature, there are three main models that are typically used to examine the impact of the psychosocial work environment on workers’ health: the Job Demand-Control-Support (JCDS) model;14 the Effort-Reward-Imbalance (ERI) model;15 and the Organisational Justice model.16 The psychosocial work environment, as measured by these models, is known to be associated with a range of health outcomes, such as depression, cardiovascular disease and overall mortality.17-21Given this, it is not surprising that these same measures of the workplace environment are often found to be robust predictors of sickness absence.22-24 However, there is now emerging evidence that these workplace factors may predict sickness absence independently of their impact on health. A recent study of more than 7000 Norwegian workers followed for 12 months replicated the association between job strain (the combination of high job demand and low decision latitude) and sickness absence, but then went on to show that this effect could not be explained by extensive measures of physical and mental health.25
The importance of recognising non-illness related predictors of sickness absenceis highlighted by the observation that symptom-based treatments alone are often not enough to return an individual to work after a period of sick leave.26 In order to maximise occupational recovery, the individual and workplace factors that are contributing to sickness absence behaviours need to be identified and addressed in addition to symptom-focused treatments. However, to date there has not been a simple way to measure the different factors that may contribute to sickness absence in individual cases. In this paper we describe the development and initial validation of a new scale for measuring non-illness factors that may be important in predicting occupational outcomes, called the NIPSA (non-illness predictors of sickness absence) scale.
METHODS
Questionnaire development
The NIPSA questionnaire was designed to measure a broad range of individual and work-related factors that have been suggested to predict occupational health outcomes. A total of 42 questions were developed which covered the following topics; perceived vulnerability, attitude towards employment, the psychosocial work environment, relationship with employer, coping style at work, and response to symptoms. The choice of which psychosocial work risk factors to include was informed by a systematic review on this topic which our research team was conducting at the same time.27 The selections of the other domains covered were based on a series of conversations with colleagues familiar with each topic area. Each question was posed as a statement (for example “I am better than most people at handling stressful situations”), with participants asked to rate how much they agree of disagree with each statement. A Likert-type scale with five choices (strongly disagree, somewhat disagree, neither agree nor disagree, somewhat agree, strongly agree)each assigned with a numerical value between 1 and 5 was utilised.Participants were instructed to think about their last job when answering the work-related questions. Half of the items were reverse scored in order to reduce response biases such as acquiescence or agreement response tendency.2829
Study population
The South East London Community Health (SELCoH) study is a psychiatric and physical morbidity survey carried out on communities in the south London boroughs of Southwark and Lambeth.30Households (defined as one person or a group of people who share accommodation as their only or main residence) were identified using the Small User Postcode Address File (PAF). Trained interviewers visited each selected household at least four times at different times of the day. Interviews were conducted on as many adults aged 16 years and older, who lived at the given address and were available during any one of the visits. Interviews were conductedusing a computer assisted interview schedule. As part of the interview, participants completed the 42-item NIPSAquestionnaire.
Of the 1439residents who were contacted between 2011 to 2013, interviews were conducted with 1052 participants (73% response rate) using a computer assisted interview schedule. A previous study using SELCoH data has shown that this sample had similar demographic and socioeconomic indicators to UK Census Information for the catchment area.31In order to avoid the potential issues that may arise as a result of clustering by household, only one individual from each household was randomly selected and kept in the analysis, whilst the remaining members of the same household were excluded. In addition, those who were not of working age (aged 16-65 years for men, aged 16-59 years for women) were excluded (156 individuals) leaving a final sample of 682 individuals used for the development and evaluation of the scale.
Factor analysis and selection of questions
Suitability for factor analysis was assessed using a series of tests; Bartlett’s Test of Sphericity, Kaiser-Meyer-Olkin Measure of Sampling Adequacy test,32 and a sample size to variable ratio of more than 5:1.33Once these requirements were adequately satisfied, an exploratory factor analysis (EFA) of principal factors was performed to investigate the latent structure of the questionnaire. The number of unrotated factors to be extracted was determined using Kaiser’s criteria of eigenvalues greater than one34 and the scree plot.35 The point at which the drop of the curve ceases and levels on the scree plot indicates the maximum number of factors to be extracted. As it was established a priori that the proposed factors would likely be correlated, we used the oblique rotation (i.e. promax approach method) to extract the factors. Individual questions that did not load adequately (factor loadings < 0.30) onto any of the extracted factors, or those that loaded strongly (factor loadings ≥ 0.3) onto more than one factor were removed. This process was repeated, with reducing number of questions, until an interpretable factor structure was obtained. Once a factor structure was established, the appropriateness of each question to the overall factor label was assessed by the authors.
The internal consistencies of the extracted factors were calculated using Cronbach’s alpha coefficient.36 An alpha coefficient >0.70 denotes satisfactory reliability for exploratory research.37 Item-total correlation, which is the correlation between an item in the factor and the aggregate of the other items of the same factor, was also inspected; a minimum correlation of 0.20 is the typical threshold used for retaining an item in the factor.38
Factor scores were created by summing the raw scores of the items, with additional Bartlett factor scores created using the score prediction command (predict) in STATA. This refined method for computing factor scores maximizes validity, and produces unbiased estimates of the factor scores that are highly correlated with a given factor.39Reverse scoring was required for the negatively phrased items. Factor score distribution was described using the mean and standard deviation where data was normally distributed.All statistical analyses were performed on STATA v.12.0 for Windows.40
Validation against measures of occupational outcomes
Amongst those who were in employment at the time of the interview, recent sickness absence was enquired about in two ways. Firstly, participants were asked about long-term sickness absence, which for this study was defined as whether an episode of sick leave lasting more than two weeks had occurred over the past two years. Participants were also required to complete questions from the World Health Organization Health and Work Performance Questionnaire concerning absenteeism.41 They were asked about the number of hours they are expected to work every week, as well as the number of full or partial days of work that they missed in the preceding four weeks due to their own physical or mental ill health. Relative absenteeism was calculated as the percentage of work hours missed during the four weeks prior to the interview, where a value of zero indicates no sickness absence and 100% equates to total sickness absence. Previous studies have shown the validity of the questions and the scoring method and have demonstrated that the estimates correlate closely with employer records of absenteeism.41Any participant who reported receiving incapacity benefits was classified as permanently sick / disabled or unemployed respectively.
Spearman’s correlation analysis was performed to examine associations between each of the factors and relative absenteeism. Student’s t-tests were conducted to determine the relationship between the factors and permanent sickness/disability or long-term sickness absence (at least one episode of sick leave of 2 weeks or more in the past 2 years).
RESULTS
Demographic information
The baseline characteristics of the study participants are presented in Table 1. The mean age of the population was 39.2 years, the percentage of female participants was 59% (n=400), and a slight majority had an education level higher than a General Certificate of Secondary Education (GCSE) or equivalent qualifications (n=467).
Construct validity
Exploratory factor analysis using the oblique rotation method extracted four factors with eigenvalues greater than one. Factor 1 had an eigenvalue of 3.89 and explained 45% of the variance; factor 2 had an eigenvalue of 1.89 and accounted for 21% of the variance; factor 3 and 4 had eigenvalues of 1.40 (explained variance 16%) and 1.32 (explained variance 14%) respectively. The four-factor solution, demonstrated in Table 2, explained 96% of the total variance. The scree plot also suggested a four-factor structure for the questionnaire. Factor rotation using the oblique method was administered to extract the four factors, which were labelled as follows: perception of psychosocial work environment (factor 1), perceived vulnerability (factor 2), rest-focused attitude towards recovery (factor 3), and attitudes towards work (factor 4). Of the original 42 proposed questions, 12 were removed by the stepped exploratory factor analysis. The items “At work I feel part of a team that works well,” “I easily feel criticized by my co-workers,” “I am easily embarrassed by health problems,” “I tend to be a bit of an all or nothing person” were removed because they were deemed by the authors as poor matches for their respective factors. After the removal of these items, factor analysis was repeated on the remaining 26 items and 10 items loaded onto factor 1, with loadings ranging from 0.34 to 0.78. Six items loaded onto factor 2 with factor loadings between 0.32 and 0.60. Factor 3 was loaded with 4 items (factor loadings ranging from 0.63 to 0.67), while factor 4 had 6 items with factor loadings between 0.30 and 0.60. Generally, items from each factor had loadings of less than 0.1 on the other factors.
Internal Consistency
Cronbach’s alpha coefficient was calculated for each of the factors to determine the internal consistency of the derived subscales. Factors 1 and 3 showed acceptable internal consistency (α = 0.79 and 0.70 respectively); however, the alpha coefficients for factor 2 (α = 0.61) and factor 4 (α = 0.62) indicated sub-optimal internal consistency. Chronological removal of items from the subscales resulted in no additional improvements to the internal consistencies. None of the items had item-total correlations lower than 0.2, and therefore were all retained in the questionnaire.
Correlations with occupational outcomes
Factor scores were computed for the four extracted factors using both the non-refined (summed raw scores) and refined (Bartlett scores) approaches. The distributions of the summed raw scores are summarised in Table 3. Bartlett factor scores are standardised, and therefore for all factors, the means and standard deviations were approximately 0 and 1 respectively. A higher score for psychosocial work environment represents more positive perceptions of the workplace and less work-based risk factors; a higher score on perceived vulnerability indicates stronger feelings or perceptions of vulnerability at work; a higher rest-focused attitude towards recovery score suggests greater beliefs that rest from work is important for recovery from illness; and a higher score for the factor attitudes towards work represents stronger views on the importance of work. Relative absenteeism was non-normally distributed, and therefore the Spearman’s rank correlation coefficient was calculated to investigate its association with each of the factors (Table 4). There were no significant correlations between any of the factors and relative absenteeism (all p>0.05), regardless of whether the score was calculated using the refined (Bartlett scores) or non-refined (summed scores) method.