The perspective I am bringing to this panel reflect our work with OMERACT – an informal international group that has been meeting since 1992 with the goal of improving outcomes used in arthritis research. They look for consensus on what to measure and how to do so in order to allow better communication across many stakeholders: clinicians, consumers, researchers, regulators and industry. OMERACT meets every other year with lots of work taking place in between. The measurement of work productivity began being addressed by a subgroup of us at OMERACT 8 (in 2006).
Decisions on outcomes at OMERACT are based on whether the instruments have passed the “OMERACT Filter” of Truth (that is content and construct validity), Feasibility (is it practical to use from a cost and ease of use perspective) and Discrimination (is it sensitive to difference between treatment and control arms in clinical trials – either change scores or responder analyses). Evidence must be gathered from the literature or new studies and presented at OMERACT meetings for discussion and a vote. The decisions guide the instruments that researchers will use in trials and this will help with communication between trials, and their application in clinical practice.
Today I would like to give you a sense of our progress to date by summarizing some of our challenges and advances through the OMERACT initiative.
There are many terms being used in measuring work productivity. And the first things we needed to do was define our target. First we considered the perspective. Workplace productivity measures from the level of the workplace, cars off the line and is important for societal level analyses. But it is affected by many different factors. The perspective we chose was workER productivity – the experience or contribution of the individual worker. Within that, arthritis could impact the worker productivity by causing the person to be absent from their job. But also by presenteeism or at-work productivity loss the loss of efficiency or effectiveness on the job. Both are important outcomes in arthritis research
But beyond that it became apparent that some of us needed an outcome state measure – perhaps to measure the impact of an ergonomics program on absenteeism or at-work productivity. And others had an equal interest in measuring measuring worker productivity as a cost indicator to quantify the costs associated with either absenteeism or at-work productivity loss for economic analyses
we decided we needed to be sensitive to these four domains: absenteeism and presenteeism or at-work productivity loss and measures of outcome state and cost estimate. *But further, we also know that people transition between absenteeism and being at-work with some difficulties, and ideally we would want a way to measure that could measure both in one metric.
This leads to the second challenge we encountered: the contextual nature of work. The difficulty experienced with work is, in part, defined by the job that person is at – the physical, mental and time demands and even things like where the work needs to be done. Our outcome measures don’t know this so they will give us what they can and we get a number that tells us about amount of difficulty. So a person having difficulty at work, say functioning at 75% of maximum productivity in his ideal job changes jobs to a less demanding one where he can function well. His number say he has gone from 75% to 100% productivity. But does that mean he is better at work? Work productivity is always a balance of person ability and job demands, but in this case we have totally changed jobs, perhaps to a less satisfying one. The numbers cannot catch that.
There are many contextual factors that may influence the numbers we get on our outcome instruments. Some are more environmental – job as we have discussed, but also the flexibility of the workplace to be able to accommodate – some small workplaces can not accommodate lighter duties. Also the organizational practices around disabilities – availability of ergonomists, therapy or different equipment at work. But there are also individual factors as our colleague Monique Gignac and others have described as the many adaptations and coping behaviours people with arthritis make to stay at work – adjusting their time at work, using vacation for appointments, getting help from coworkers, modifying tasks to make them less demanding, and anticipating and avoiding problems.
Recognizing the challenges of this measurement, I want now to focus on our advances. And there have been advances.
At OMERACT 9 we agreed that absenteeism was more than days absent from work. That we had to consider half days off, or use of vacation instead of sick time. We decided what should be considered in absenteeism.
We voted that arthritis research should consider a broader set of indicators of absenteeism that includes not only sick days off due to arthritis, but also vacation days taken to manage arthritis, part days off, change in number of hours worked (reduced hours), temporary cessation of work on disability or prolonged sick leave and permanent cessation (leaving work force). We were asked to consider further, perhaps with sensitivity analyses, the impact of using absence and cessation of work due to any cause, including by choice.
In terms of at-work productivity, we found 21 different scales that included the measurement of at-work productivity loss. In the literature there were only five head to head comparisons of their measurement properties, with concerning results – the scales described different levels of work limitations in the same patients. Lavigne found one instrument said 53% of the sample had work limitations, and another said 10%. There were low correlations between measures and little information on arthritis.
To help with gathering evidence, we conducted a head to head comparison of five measures of productivity in a cohort of 250 persons with either OA or inflammatory arthritis. We were able to show some evidence of validity, responsiveness and predictive validity of the individual scales – but the correlations between the scales were only moderate at best. From a cost perspective, Wei Zhang compared cost estimations and found them to vary up to seven fold depending on the instrument you chose. We compared a wider array of measures and continued to find clinically significant differences in the way they would quantify the impact of arthritis on work.
We believe one of the likely reasons is that the instruments are targetting different concepts so they have different strengths. These are three instruments we have been looking at, the first two developed for arthritis. The WALS focuses on difficulties at work based on HAQ type questions. The WIS or work instability scale focuses on the worry about mismatch between job and the person’s abilities and an estimates that risk or concern about loss of work due to arthritis. The WLQ by contrast quantifies the proportion of time a person is having any level of difficulty with their work, and is easily converted into hours of productivity lost, and from their to costs.
At OMERACT 9 we reviewed this evidence believed there were six measures that were gathering evidence towards the filter and we encourage people to focus on these and try them. There was no clear winner yet, as shown by the percent endorsing each questionnaire, but a clear direction to focus on these 6 of the 21 for now. We asked people to work with some of these measures and come back with more evidence about their relative strengths as outcome measures.
The other advance worth mentioning is in the blending of absenteeism and presenteeism into one metric. Currently we measure absenteeism or at-work difficulties separately. And at-work issues are assigned &quot;missing data status&quot; if a person is absent. Losing important information, as they are clearly at one end of the at-work productivity spectrum. We need to find ways of integrating them into one measure. The application of newer statistical approaches can allow us to blend information on both into one outcome. Hierarchical modelling can handle absenteeism without assigning it to missing or zero. SEM allow absenteeism and presenteeism to be blended into a single indicator or latent trait.
Sustained at work (24.9%): consistently at-work and low dropout rate; at-work productivity improved significantly ( p = .006); on average at-work disability decreased 20.7% after 12 months. High level of presenteeism but more likely dropout (9.6%): consistently at-work if not dropout; no significant change in the level of at-work disability ( p = .92). Sustained absenteeism (25.9%): consistently off-work; unable to evaluate its productivity loss or improvement over time. Usually at work (8.7%): working at most of the time points and returned to work at 12-month assessment; at-work productivity improved significantly ( p <.001); on average at-work disability decreased 44.3% after 12 months. Return to work (11.8%): initial off-work but return to work; at-work productivity improved significantly ( p <.001); on average at-work disability decreased 80.5% after 12 months. Working then off (19.2%): initial at-work but off at follow-up. At-work disability increased dramatically and significantly over time. Dropout is non-ignorable missing (not MAR or MCAR) Those sustained at work and with less change of at-work disability over time were more likely dropout of the study. Group dropout rates were 54%, 75%, and 86% at 3-, 6-, and 12-month follow-up, respectively. Those initial off-work then return to work at follow-up (with decreased at-work disability) were more likely dropout. Group dropout rates were 46%, 60%, and 76% at 3-, 6-, and 12-month follow-up, respectively. Conversely, those sustained absenteeism and those initial at-work then off-work at follow-up were less likely dropout. Group dropout rates were less than 15%.
SEM can also allow the potential to build more complex models to explain the differences we are seeing in our tools. Perhaps at work difficulties picked up in the WALS will lead to worker productivity loss which might lead to absence. And work instability might pick up when work productivity loss transitions to absenteeism. But these models need large sample sizes, and we also need solutions applicable to smaller studies and to clinical practice. We also need solutions for these situations.
I would conclude by saying that work IS an important outcome in persons with arthritis not only for financial reasons but self-esteem and social roles. We have many outcomes available for absenteeism and at-work productivity loss, but they are not equivalent and likely should not be compared across studies. The OMERACT type approach tries to achieve a consensus across stakeholders based on evidence of “truth, feasibility and discrimination” – key measurement properties for choosing the best way to measure work.
Measuring worker productivity: outcome or cost?
Dorcas Beaton, BScOT, PhD
Health Policy Management & Evaluation, University of Toronto,
Mobility Program Clinical Research Unit, St Michael’s Hospital,
Amount of difficulty at work (at-work productivity loss or presenteeism)
As an indicator of cost
Cost of days off work +/- replacement workers (absenteeism costs)
Cost of at-work productivity loss (presenteeism costs)
Target: the impact of arthritis on work at the level of the worker Goal: to have solid indicators of each of these cells. Solid in terms of breadth & depth, measurement properties Possibly same measure for Cost and Outcome goals! YES YES Cost estimate YES YES Outcome State Presenteeism/ At-work productivity loss Absenteeism
Dropout is non-ignorable missing. The likelihood of dropout, the level of presenteeism and level of at-work disability were significantly associated.
6 5 4 3 2 1 <.001 Increased over 60% in 3 months <10%, <10%, <15% Initially at work, later off (19.2%) <.001 Decreased 80.5% over 12 months 46%, 60%, 76% Initially off work with RTW (11.8%) <.001 Decreased 44.3% over 12 months <10%, <10%, <15% Working at most time points (8.7%) NA Can not evaluate (no WLQ measures) <10%, <10%, <15% Sustained off work (25.9%) 0.92 No significant change 54%, 75%,86% High presenteeism (9.6%) 0.006 Decreased 20.7% over 12 months <10%, <10%, <15% Sustained at work (24.9%) P-value for change in at-work disability At-work disability Dropout rate at 3-, 6-, 12M Class
Structural equation modeling At work difficulty At work productivity loss Absent from work (WALS) (WLQ) Person-job instability (WIS) Potential to build more complex models to explain the relationship between different measures