Abstract
Objective We aimed to identify threshold values of presenteeism measurement instruments that reflect unacceptable work state in employed patients with rheumatoid arthritis (RA) and whether those thresholds can predict future adverse work outcomes (AWOs). Additionally, we assessed the performance of presenteeism thresholds previously established in axial spondyloarthritis (axSpA) among patients with RA for the same instruments.
Methods Data from the multinational Patient-Reported Outcomes in Employment Study in Rheumatoid Arthritis (RA-PROSE) study were used. Thresholds to determine when patients consider themselves in an “unacceptable work state” were calculated at baseline for 4 instruments assessing presenteeism and for the patient global assessment of RA-related pain. Different approaches derived from the receiver-operating characteristic methodology were used. Accuracy of thresholds to predict AWO throughout 12 months was assessed and previously developed presenteeism thresholds for axSpA were also tested.
Results A total of 104 employed patients were included: 15% of the patients considered themselves in an unacceptable work state, of which 7 (7%) had at least 1 AWO over 12 months. Thresholds of all instruments specifically developed in RA showed good performance vs the external criterion (area under the curve [AUC] > 0.75), except for the Quantity and Quality (QQ) method (AUC 0.62). The available axSpA thresholds were more accurate by reducing overestimation. The final optimal thresholds were Work Productivity and Activity Impairment Questionnaire (WPAI)-presenteeism ≥ 40, QQ method < 97, Workplace Activity Limitations Scale ≥ 0.75, 25-item Work Limitations Questionnaire with modified physical demands scale ≥ 29, and pain intensity ≥ 4. For AWO over 12 months, pain and WPAI performed best in predicting AWO.
Conclusion The final thresholds to assess unacceptable presenteeism for axSpA were also chosen as most accurate for use in RA. In addition, accurate thresholds of pain reflecting unacceptable work state are available.
Rheumatoid arthritis (RA) is a chronic systemic inflammatory disease characterized by polyarthritis and potentially joint destruction. Generally, RA affects adults of working age, leading to productivity loss as a result of being absent from work (absenteeism, sick leave, work disability, or job loss) or by having a limitation in or reduced productivity while at work (presenteeism).1-3
Productivity loss may occur early in the disease course because of limitations in physical function related to inflammatory disease activity or later in this process because of additional joint destruction, among other potential causes. Despite the availability of new and more effective disease-modifying treatment strategies that reduced the effect of RA on work participation,4 patients still report different levels of work ability and productivity at work when compared to the general population.5,6
Of a large number of self-reported measurement instruments for presenteeism, Outcome Measures in Rheumatology (OMERACT) selected 6 candidate measures for further assessment of their psychometric properties: the Work Ability Index,7 Quantity and Quality (QQ) method,8 Work Productivity and Activity Impairment Questionnaire (WPAI)-presenteeism,9 the rheumatoid arthritis-specific Work Productivity Survey,10 Workplace Activity Limitations Scale (WALS),11 and the 25-item Work Limitations Questionnaire with modified physical demands scale (WLQ-25 PDmod).12 The WPAI-presenteeism questions have already been endorsed by OMERACT, as they fit requirements for truth, reliability, and feasibility.13 Thresholds of meaning for 4 of these instruments have been developed for patients with axial spondyloarthritis (axSpA), namely for the WPAI, QQ method, WALS, and WLQ-25 PDmod. These thresholds have demonstrated good performance in identifying patients unsatisfied with their work state and, albeit with slightly lower accuracy, in predicting the occurrence of adverse work outcomes (AWOs) over the next year.14
Using thresholds of meaning for presenteeism measurement instruments, despite having a lower precision, is more practical, and these thresholds are easier to interpret in comparison to a continuous score. If feasible, it would be convenient to have the same thresholds for different diseases. This requires thresholds to be accurate and to perform similarly across different diseases. Moreover, identifying thresholds of presenteeism measurement instruments to reflect specifically when patients with RA consider themselves in an unacceptable work state is crucial from both a clinical and academic perspective (standardizing research, selection of appropriate instruments, study of more meaningful outcomes, etc.).
Further, for reasons of feasibility, when selecting instruments for clinical studies and evaluation in daily clinical practice, thresholds for a traditional health outcome instrument—namely pain assessment—could be useful, particularly if they provide equally accurate information regarding acceptable work state or future AWO. This might facilitate the identification of persons at risk in daily practice when no presenteeism instruments are available.
The aim of this study was to identify the thresholds of 4 instruments for presenteeism and for the pain intensity rating scale that most accurately reflect a patient’s health state considered as an “unacceptable work state” in RA, and to explore whether these thresholds predict future AWO. Additionally, this study aimed to compare the performance of the specifically developed presenteeism thresholds in RA with the thresholds previously established in axSpA for the same instruments. Identifying common thresholds in both axSpA and RA would enable comparability across work outcomes studies in the 2 diseases.
METHODS
Study design and patient recruitment. Data from the multinational, prospective, observational study on Patient-Reported Outcomes Survey of Employment in Rheumatoid Arthritis (RA-PROSE) were used. Briefly, the RA-PROSE study includes data on work participation in patients with a diagnosis of RA (according to their treating rheumatologists) from 4 countries: Canada, the Netherlands, the United Kingdom, and the United States. Data were collected from October 2008 to December 2013. For the current study, data of working patients aged between 18 and 65 years (working age) with paid work at baseline were used; for the thresholds analysis, patients reporting sick leave at baseline were excluded.
Data collection. Participants were asked to complete an online survey every 3 months during a 1-year period (at baseline and at 3, 6, 9, and 12 months). Following a similar protocol as the Patient-Reported Outcomes Survey of Employment in Ankylosing Spondylitis (AS-PROSE) study,15 data collected included baseline sociodemographics (eg, age, sex, ethnicity); education (superior/no superior education); and marital status; and, at each visit, information on work (ie, being employed or not, days currently on sick leave, presenteeism); disease characteristics (ie, symptom and disease duration, treatment); health outcomes (eg, pain, multidimensional Health Assessment Questionnaire)16; scores from several instruments to measure health-related quality of life (eg, Medical Outcomes Study 36-item Short Form survey17 and Patient Acceptable Symptom State18); work-related context (eg, nature of work or job type, dichotomized into blue or white collar and full or part-time work); and lifestyle factors (ie, BMI, past or current smoking).
Instruments for which thresholds were determined. Four self-reported instruments proposed by OMERACT to address presenteeism or including presenteeism as a subscale were included in RA-PROSE. Two assess the global effect of health on work (WPAI-presenteeism scale and QQ method) and 2 are multiitem, multidimensional instruments (WALS and WLQ-25 PDmod) that address the effect of health on various aspects of work. The WPAI-presenteeism scale scores limitations in productivity while at work (0-100%; 100% = worst productivity), the QQ method multiplies global assessment of quality and quantity of work (0-100; 100 = best quality and quantity), the WALS contains 12 questions on varying (dis)ability related to work (0-3; 3 = worst ability), and the WLQ-25 PDmod has 25 items across 4 subscales addressing the percentage of time at work with various limitations (each scale ranges from 0-100 [maximally limited]). These 4 subscales can also be converted into the WLQ-25 PDmod by calculating the average of the 4 and providing the estimated productivity loss (0-100; 100 = maximal productivity loss).19 A more detailed description of the instruments is shown in Supplementary Text S1 (available from the authors upon request).
Additionally, the threshold value that indicates unacceptable work state was determined for the pain intensity assessment, obtained from a 0-10 numerical rating scale, with a higher score indicating higher pain.
Anchor questions. Two predefined external criteria were chosen, as they were considered relevant when implementing thresholds in clinical studies and in daily clinical practice. First, the RA-PROSE case report form included a question on the Patient Acceptable Work State (PAWS). The specific formulation for the item in the current study was in line with considerations by OMERACT.13,20,21 In this survey, question 1 from PAWS asked participants the following: “Considering all the different ways your disease is affecting you, if you would stay in this state for the next few months, do you consider that: your ability to perform your current job is satisfactory? (yes/no).” To align the direction of the anchor question with the presenteeism instruments, the answer was inverted, allowing the assessment of an unacceptable work state. Second, as a long-term outcome, future AWO was used, defined as any sick leave, or short-term or long-term disability over the 12 months of follow-up.
Statistical analysis: threshold analyses. Receiver-operating characteristics (ROC) analysis was used to define the thresholds for unacceptable work state.22 This statistical approach allows determination of the threshold by balancing sensitivity (SE) and specificity (SP) for each measurement instrument. We applied the following 4 approaches to find the optimal cutoff:
The 75% percentile method identifies the 75th percentile of the distribution of the presenteeism scores. In order to favor more SE in the selection of the thresholds, we calculated this in patients who considered themselves in an acceptable work state, so that a value above the cutoff in each of the instruments reflects an unacceptable work state. This approach has been validated as a comparable alternative to the ROC analysis, and is much easier to derive.23,24
The Liu method maximizes the product of the SE and SP.25
The Youden index assumes that SE and SP are equally important and subtracts the value 1 from the sum of SE and SP, so that the maximum value of the index becomes 1 when there is perfect agreement.26
The nearest to 0.1 approach identifies the point where the shoulder of the ROC curve is closest to the left upper corner of the graph (the point with perfect SE and SP), and hence the point of highest accuracy of the test.27
Since the last 3 methods balance SE and SP, we added an alternative ROC-based approach in which we gave more weight to SE and higher positive likelihood ratio, taking into account that there are limited consequences of an overdiagnosis (ie, low SP) for patients or the healthcare system and that we consciously want to have higher SE in identifying patients at risk. When 2 thresholds seemed to perform well, the final decision was based on performance; that is, the proportion correctly classified combined with absolute numbers of false positives vs false negatives for the conflicting thresholds. The threshold with the highest proportion of correctly classified patients and the best balance between overestimation and underestimation was the one chosen. We also adapted, when necessary, theoretical numbers from the output of the different methods to the closest real number of the scale of the measurement instrument with the same SE and SP.
Comparative analysis. An additional comparative analysis was performed to assess the performance of thresholds previously developed in axSpA to identify patients with RA having an unacceptable work state or at risk for AWO over the 12 months of follow-up.15
All the analyses were performed using Stata SE V.17.
RESULTS
Baseline characteristics. Of 239 patients participating in RA-PROSE, 104 (44%) were included in the current analyses as they were aged ≤ 65 years and had paid work at baseline. Of these, 80 (77%) were women, with a mean age of 48 (SD 9) years and a mean symptom duration of 9.9 (SD 8.6) years. Mean baseline pain was 3.2 (SD 2.2); 78 (75%) patients were on treatment with conventional synthetic disease-modifying antirheumatic drugs (DMARDs) and 36 (35%) with biologic DMARDs. In total, 60% of the patients had a full-time job and 20% were blue-collar workers. Baseline characteristics for the total population as well as stratified by satisfaction with work state are shown in Table 1. Supplementary Tables S1 and S2 (available from the authors upon request) provide a detailed description by origin country and by presence of AWO during the 12-month follow-up.
Baseline characteristics of the total population and by satisfaction with work statea.
The mean baseline values of the 4 presenteeism instruments were as follows: WPAI-presenteeism, 23.2 (SD 23.7); QQ method, 81.9 (SD 25.6); WALS, 0.65 (SD 0.45); and WLQ-25 PDmod, 22.2 (SD 14.8). Based on the PAWS anchor question, 15% (n = 16) of the patients indicated they were unsatisfied with their work state, and 10% (n = 10) already reported an AWO (Supplementary Table S3, available from the authors upon request).
Thresholds for “unacceptable work state.” The final selected optimal thresholds in RA for each presenteeism measurement instrument and the pain intensity scale, with the proportion of persons correctly classified and SE/SP, are presented in Table 2.
Performance assessment of each presenteeism instrument when classifying unacceptable work status.
In general, the results of all the ROC methods were similar and thresholds were appropriate for differentiating between patients with unacceptable and acceptable work state when looking at the proportion that were correctly classified. The only exception was the Youden method, particularly when applied to the QQ method (Supplementary Table S4, available from the authors upon request).
Supplementary Table S4 (available from the authors upon request) presents more detail on the results for the different ROC methods (thresholds, area under the curve [AUC], SE/SP) for each measurement instrument. The AUC by ROC for all the instruments was close to 0.80, with the exception of the QQ method (AUC 0.62). The pain assessment performed similarly to the presenteeism instruments (Figure). After the dichotomization of the measurement instruments by our proposed threshold value, the AUC remained acceptable, although, again, with lower values for the QQ method and pain.
ROC curves for excessive presenteeism from different measurement instruments according to unacceptable work state. QQ: Quantity and Quality; ROC: receiver-operating characteristic; WALS: Workplace Activity Limitations Scale; WLQ-25: 25-item Work Limitations Questionnaire; WPAI: Work Productivity and Activity Impairment Questionnaire.
Testing thresholds derived in axSpA. Table 2 also depicts, where applicable, the performance of the previously developed thresholds among patients with axSpA (second row) in the RA population. As can be seen, the selected WPAI-presenteeism thresholds were identical for both RA and axSpA cohorts. For the other presenteeism instrument, the axSpA thresholds performed better in identifying work status as unacceptable and in predicting AWO in RA, mainly by reducing the false positives (ie, avoiding overestimation; Table 3). In the end, the thresholds for the presenteeism instruments in RA performing the best are the same as the ones in axSpA.
Performance assessment of each presenteeism instrument when predicting AWO over 12 months.
DISCUSSION
This study proposes thresholds for presenteeism that accurately identify unacceptable work state in working persons with RA. These thresholds have proved to be the same as the ones previously identified for persons with axSpA. Thresholds were ≥ 40 for WPAI-presenteeism, < 97 for the QQ method, ≥ 0.75 for WALS, and ≥ 29 for the WLQ-25 PDmod. The threshold for pain intensity in RA was ≥ 4. Overall, the proposed thresholds performed poorly in terms of predicting AWO over 12 months, with the exception of WPAI-presenteeism and pain assessment, which correctly identified at baseline three quarters of the patients with an AWO over the subsequent 12 months.
It was striking that the threshold previously developed in patients with axSpA performed better than the threshold specifically developed within the RA population. However, it should be noted that only 104 of the 239 patients in the RA-PROSE study were actively working and only 15% reported an unacceptable work state. Consequently, the somewhat more stringent thresholds for axSpA reduced the absolute numbers of those incorrectly classified as “unacceptable” by the RA-specific thresholds. In the final selection of the thresholds, we thus prioritized the numbers correctly classified, reducing overdiagnosis and thereby increasing specificity. The somewhat lower values of the RA-specific thresholds might relate to the higher proportion of female individuals. Also, in the axSpA study, when sample size was sufficient to compare performance of the thresholds across contextual factors (age, sex, education, among others), the overall axSpA threshold resulted in overdiagnosis in women and persons with a lower educational level.
When comparing the performance of the currently proposed thresholds in the RA-PROSE and AS-PROSE populations, the threshold for WPAI-presenteeism had the best discriminatory capacity in both populations, with 82% of patients being correctly classified; this was very similar to the performance in the AS-PROSE cohort (83%).15 Aside from being highly accurate in both diseases, the fact that the WPAI-presenteeism instrument is the more widely used,28 and the only one endorsed by OMERACT13 to date, leads to the conclusion that WPAI-presenteeism is an excellent tool to be used regardless of the disease. On the other hand, threshold performances for the QQ method were also similar for RA and axSpA (65% and 63%, respectively), and both were considered insufficiently accurate. For the 2 multidimensional presenteeism questionnaires (WALS and WLQ-25 PDmod), threshold performances were rather different among axSpA and RA (76% vs 79% for WALS, and 68% vs 61% for WLQ-25 PDmod). As mentioned earlier, the sex ratio between populations might have influenced the results.
Regarding the capacity for predicting AWO in the following year, the general low accuracy was consistent with the axSpA results; although, once again, WPAI-presenteeism correctly classified at least 75% of patients, promising some value in identifying patients at higher risk of presenting with future absenteeism or work disability. Of note, the results of both analyses are influenced by the low occurrence of AWO. Nevertheless, this is a limitation that cannot be easily overcome, and such analyses still allow the comparison of the performance, in terms of predictive validity, of the thresholds for the different instruments.
One of the main limitations of this study is the low sample size, which could affect the stability of the thresholds. Even though this was the reason for testing the performance of the axSpA thresholds, further external validation would be important to confirm their accuracy. The major disadvantage when handling thresholds is the lower precision in comparison with continuous variables. This was mainly seen within the pain thresholds, in which the AUCs for the continuous variable decreased considerably after dichotomization (AUC 0.77 to 0.67; Supplementary Table S4, available from the authors upon request). Notwithstanding, thresholds are more useful for the identification of patients at risk, and therefore the aim of this analysis was to identify them, despite knowing that continuous scores perform better than any dichotomization. Lastly, another limitation is the potential overlap between the PAWS and presenteeism questionnaires, particularly the single-item questions, such as those in the WPAI. However, the objective of this study was specifically to determine the cutoff points for each of the 4 presenteeism instruments, without necessarily comparing the accuracy of the instruments to that of the PAWS, which was used here as an external criterion.
In summary, thresholds for presenteeism instruments representing an unacceptable work state have been established for RA, which are the same as the previously developed thresholds for axSpA: WPAI-presenteeism ≥ 40, QQ method < 97, WALS ≥ 0.75, and WLQ-25 PDmod ≥ 29. Similar performance has also been demonstrated for an instrument assessing health outcomes used in daily clinical practice like pain (≥ 4). WPAI-presenteeism is the instrument that performed best and it can be used to identify patients requiring more tailored care to avoid future AWOs.
ACKNOWLEDGMENT
HMO is supported by the National Institute for Health Research (NIHR) Leeds Biomedical Research Centre. The views expressed are those of the authors and not necessarily those of the National Health Service, NIHR, or UK Department of Health.
Footnotes
CONTRIBUTIONS
Conceptualization and methodology: SR, AB, EN, DC, WPM; data curation and formal analysis: DC; supervision: SR, AB, EN: writing – original draft: DC, SR, AB, EN; writing – review & editing: WPM, HMO, MNM.
FUNDING
No funding was provided for the current analysis. The AS-PROSE study was conducted with a grant from AbbVie (named Abbot at the time of the RA-PROSE development).
COMPETING INTERESTS
SR has received research grants or consultancy fees from AbbVie, Lilly, Galapagos, Janssen, MSD, Novartis, Pfizer, Sanofi, and UCB. EN has received speaker honoraria from or been on advisory boards for Celltrion, Pfizer, Sanofi, Gilead, Galapagos, AbbVie, Lilly, and Fresenius Kabi, and holds research grants from Pfizer and Lilly. WPM has received research or educational grants from AbbVie, Novartis, Pfizer, and UCB; honoraria or consulting fees from AbbVie, BMS, Boehringer Ingelheim, Celgene, Lilly, Galapagos, Janssen, Merck, Novartis, Parexel, Pfizer, and UCB; and is the Chief Medical Officer for Care Arthritis Ltd. MNM has received consultancy fees from Novartis, AbbVie, UCB, Lilly, Janssen, and Pfizer, and received research grants from BMS and Amgen. HMO has received research grants from Janssen, Novartis, Pfizer, and UCB, and speaker or consulting fees from AbbVie, Amgen, Biogen, Lilly, Janssen, Moonlake, Novartis, Pfizer, and UCB. AB has received research grants from AbbVie and consulting fees from AbbVie, Galapagos, Novartis, and Pfizer, all to her department. DC declares no conflicts of interest relevant to this article.
ETHICS AND PATIENT CONSENT
Ethics approval was received from the University of Alberta (no. Pro00123727). Written informed consent was obtained from all patients before enrollment, and the ethics committees from the individual participating centers approved the study.
- Accepted for publication January 13, 2025.
- Copyright © 2025 by the Journal of Rheumatology
