Reduced productivity at work due to ill health or work environment problems in the form of absence from work, as well as reduced performance while at work, could lead to production losses for both society and for companies. Losses related to ill health, referred to as health-related production loss hereafter, have been highlighted recently in research as well as in society. Sickness absence was previously the only factor considered when estimating the cost of production loss from both the societal and employer perspective. However, more recent studies have shown that the costs related to consequences of employees attending work despite being sick, so-called presenteeism, could be at least twice as high as the cost of sickness absence [1, 2]. Thus the cost of production loss due to presenteeism has to be considered when production losses are estimated even from the broader societal perspective.
The practical concern, however, has been how lost production due to ill health is measured [3, 4]. Objective methods, often adopting computer-based measurements of health-related production losses, have been used sparingly due to the lack of generalizability of outcomes and to time constraints on implementation [5]. Instead, the use of self-reported productivity instruments is the most common approach [4, 5]. These self-reported instruments mostly measure indirect costs to the employer arising from health-related absences and reduced performance while sick but at work. To this end, efforts to measure productivity have focused on performance-based approaches to estimate health-related production loss in order to establish the business value of an employee’s health as well as to communicate to employers the reasons for investing in employee health [6–8].
Several productivity valuation instruments have been developed to capture these losses, some validated and others not [5, 8, 9]. The challenges, though, lie in the feasibility of collecting information related to production loss and how these instruments perform with regard to measuring productivity-related information, their applicability to real-world businesses, and the possibility of monetizing health-related production loss [5]. The measurement of production loss as a result of presenteeism seems to pose more challenges than the measurement of production loss due to absenteeism because production is not easily measurable in most cases [8, 10]. Furthermore, some of the instruments have been developed for a specific purpose directed towards particular groups and with varying recall periods also affecting the possibility to capture these losses [5, 8].
The implication of these challenges, from the employer’s perspective, is that measuring production loss is mired in uncertainty as to which measure will accurately capture the losses. To be useful, a measuring instrument must therefore: 1) determine if and to what extent health and work environment problems affect employee performance; 2) be possible to use when evaluating change over time as part of an intervention evaluation [3]; and 3) be capable of measuring the costs arising from the particular problem. Such a tool, of course, needs to be tested for its validity and reliability. A first step is to make sure that the instruments used capture what they are intended to capture, which is evaluated in different validation tests.
Health-related production loss, as measured in this study, has previously been tested for its construct validity in a working population [11]. That study indicated that health-related production loss was explained to a larger extent by health than by work environment-related factors. However, this difference was not statistically significant and the construct validity of this measure could therefore not be stated [11]. A possible explanation for this result is that health-related production loss also is a consequence of work environment-related problems, as shown in previous studies [12, 13], and it is questionable if the hypothesis that health-related production loss can be explained by health alone, without influence from work environment factors, was judiciously chosen. The current study expands the validation process for this Swedish measure developed to capture health-related production loss by examining its construct validity, and also by examining responsiveness validity in a population suffering from persistent back/neck pain or common mental disorders (CMDs). To date, these diagnoses are the two most common causes of sickness absence in Sweden as well as in Europe generally, and there has been a lot of focus on these groups for the prevention of sickness absence or the rehabilitation of those already on sick leave. These groups were also the targets of a large, nationally funded rehabilitation project launched in Sweden in 2009, from which data for this study has been drawn.
The specific aim of this study is to evaluate the construct validity and responsiveness of this measure of health-related production loss. A further aim of the study is to investigate if there is a difference in the level of production loss within a population suffering from persistent back/neck pain and CMDs.
Convergent validity
Construct validity refers to the extent to which a construct measures the construct it is supposed to measure [14]. In this study, construct validity was tested by evaluating expectations about the relationship between different health-related conditions and health-related production loss (convergent validity). The validation of a measurement is a process that includes several steps depending on the characteristics of the measurement and what it is intended to be used for. The measurement evaluated here refers to work performance, or production loss, as well as to those factors that may limit this performance. In this part of the validation process, the focus is on the later part, i.e., whether the measurements are able to reflect or be indicative of health-related factors that may limit work performance. Work performance per se was not included in this data collection and is therefore not included in this validation. However, work ability was measured and is used as an indicator of performance, as it is related to employees’ ability to perform at work.
The measure of health-related production loss was correlated with different measures of health: general health, health-related quality of life, and work ability. If these three constructs correlate with health-related production loss as defined in the hypothesis below, this indicates convergent validity. First, health-related production loss was correlated with a question about general health from the validated questionnaire Short Form-12 (SF-12) [15]. Previous studies have shown that general health is weakly-to-strongly correlated with health-related production loss [9, 16]. This difference in the strength of correlation was found for different time points within the same study population [16], but was also related to the different instruments capturing health-related production loss [9].
Hypothesis H1a
General health is expected to have a moderate negative correlation with health-related production loss (r = −0.30 – -0.50).
Hypothesis H1b
General health is expected to have a moderate negative correlation with health-related production loss in patients suffering from N/B pain (r = −0.30 – -0.50).
Hypothesis H1c
General health is expected to have a moderate negative correlation with health-related production loss in patients suffering from CMD (r = −0.30 – -0.50).
Second, production loss was correlated with health-related quality of life (HRQL), measured by the validated Euroqol questionnaire EQ-5D [17]. HRQL has previously been shown to be associated with productivity [18, 19]. The degree of correlation has not been expressed. However, as HRQL is a measure that can be used to capture general health we assumed that the strength of the correlation for HRQL would equal that of general health.
Hypothesis H2a
HRQL is expected to have a moderate negative correlation with health-related production loss (r = −0.30 – -0.50).
Hypothesis H2b
HRQL is expected to have a moderate negative correlation with health-related production loss in patients suffering from N/B pain (r = −0.30 – -0.50).
Hypothesis H2c
HRQL is expected to have a moderate negative correlation with health-related production loss in patients suffering from CMD (r = −0.30 – -0.50).
Third, health-related production loss was correlated with the person’s overall work ability, measured using a question from the Work Ability Index (WAI) [20, 21]. Work ability is normally used as an indicator of an employee’s ability to perform at work but has seldom been evaluated in relation to a measurement of health-related production loss with data collected from the same population. One study has tried to evaluate if work ability is a robust indicator for assessing production loss [22]. However, this study used estimates of work ability from one population and modeled it with data on production loss from another population. The result from the study suggested that work ability is a robust indicator for assessing production loss. Two other studies [23, 24] evaluated the validity of a different measurement of how people function at work when experiencing health-related problems, The Work Role Functioning Questionnaire (WRFQ), and correlated it with WAI. They found a moderate to strong correlation between the two constructs. Based on the results in these studies we assume that work ability is at least moderately correlated with health-related production loss.
Hypothesis H3a
Work ability is expected to have at least a moderate negative correlation with health-related production loss (r = −0.30 – -0.50).
Hypothesis H3b
Work ability is expected to have at least a moderate negative correlation with health-related production loss in patients suffering from N/B pain (r = −0.30 – -0.50).
Hypothesis H3c
Work ability is expected to have at least a moderate negative correlation with health-related production loss in patients suffering from CMD (r = −0.30 – -0.50).
Responsiveness
Responsiveness refers to the ability of an instrument to detect important changes either in terms of its ability to change over time, or in terms of how changes to it relate to corresponding changes in a reference measurement [25]. The first type of responsiveness is referred to as “internal responsiveness,” whereas the second type of responsiveness is referred to as “external responsiveness.” In this study, we evaluated the responsiveness of the measure using external responsiveness, i.e., whether changes in the scores of health-related production loss should be related to corresponding changes in various measures of health [25]. For all hypotheses (H4a-c) it was expected that health-related production loss would be reduced when health was improved, i.e., the different constructs were expected to have a negative association with health-related production loss.
Hypothesis H4a
Changes in general health are significantly and negatively associated with changes in health-related production loss.
Hypothesis H4b
Changes in HRQL are significantly and negatively associated with changes in health-related production loss.
Hypothesis H4c
Changes in work ability are significantly and negatively associated with changes in health-related production loss.