Effect of biological therapy on work participation in patients with ankylosing spondylitis: a systematic review
- 1Department of Internal Medicine, Division of Rheumatology, Maastricht University Medical Centre, Maastricht, The Netherlands
- 2Department of Rheumatology, VU University Medical Center, Amsterdam, The Netherlands
- Correspondence to Margaretha M ter Wee, VU University Medical Center, Department of Rheumatology, room 3A52, PO Box 7057, 1007 MB Amsterdam, The Netherlands;
- Received 23 April 2012
- Accepted 2 August 2012
- Published Online First 6 September 2012
Objectives To review systematically the effect of biological treatment in patients with ankylosing spondylitis (AS) on three work outcomes: work status, absence from paid work and at-work productivity.
Methods A systematic literature search was performed (Pubmed, Embase, Cochrane Library) to identify relevant articles. Risk of bias of included studies was assessed using the Cochrane guidelines for cohorts and randomised controlled trials (RCTs). Data were extracted using a self-composed data extraction form. Owing to extensive interstudy heterogeneity, narrative summaries were used to present the data.
Results Nine studies were included (six uncontrolled cohorts, one population-controlled cohort and two RCTs) that reported on 39 comparisons. Overall, 961 patients were treated with three different tumour necrosis factor α inhibitors (etanercept, infliximab, adalimumab). For presenteeism and absence from work, most comparisons showed improvement in favour of biological agents, but not all comparisons were statistically significant and they usually concerned before–after analyses. For work status, changes were less often positive, but studies dealt with patients with longstanding AS, lacked power and had a relatively short follow-up.
Conclusions Although trends towards beneficial effects of biological agents in longstanding AS were seen on all work outcomes, the methodological limitations in the studies included hampers clear conclusions. Since the majority of studies were (extensions of) controlled trials, the generalisability of the effect of biological agents on work participation in real life should be further studied in larger (population-controlled) studies. The effect of biological agents in patients with early disease has not yet been examined.
Several studies have shown substantial impact of ankylosing spondylitis (AS) on work participation before the introduction of biological treatment.1–6 Since the onset of AS is generally in the third decade of life it possibly has a large impact on patients and society owing to its long-term limitations. In population-controlled studies, a Dutch study showed that withdrawal from paid work in AS was three times higher than expected in the general population.1 In Argentina, 26% of working age patients with AS were unemployed compared with 4.5% in the control group.2
Differences in employment among countries were confirmed in a European three-nation study that reported the highest work-disability rate in the country with the most favourable work-disability allowances.4 When studying absence from paid work, 50% of patients incurred sick leave in a European three-nation study during the 2-year follow-up period, and yearly sick leave days were higher than expected in the general population.5 ,6 About 53% of working patients in the years before pre-biological agents reported that discomfort adversely influenced their work. Patients scored their at-work efficiency during hours worked as 7.7 on a scale of 0–10 (where 10 is normal). An extra 1.91h/2 weeks was perceived to be necessary to compensate for loss of production due to work inefficiency.5
With the introduction of biological agents for AS in 2002, the possibility of controlling disease activity and improving physical function and health-related quality of life in the short and long term, improved dramatically.7–10 However, costs of biological therapies are high. Indirect costs are the major part of the total costs of AS, especially the costs of work disability. Savings in indirect or productivity costs might offset drug costs.
To provide a comprehensive view of the possible effects of biological agents on work outcome in AS, all published articles were reviewed that aimed to quantify the effect of biological therapy on three domains of work participation that are relevant for productivity costs, including reduced productivity at work, absence from paid work and long-term work disability irrespective of study design and approach of outcome assessment.
A systematic search of the literature (Pubmed/Medline, Embase and Cochrane Library) was performed by two authors (LRAvdB, MMtW) to identify all articles from 2000 up until 12 April 2012 that quantified the effect of anti-tumour necrosis factor (TNF) treatment on presenteeism, absenteeism or work status in patients with AS. The search strategy is available as online supplementary text. Abstracts of articles were limited to studies in the Dutch, English, French or German language. References of retrieved articles were scanned for additional studies.
Two authors carried out an initial screening of all titles and abstracts retrieved from the search (LRAvdB, MMtW). Articles were eligible for full-text assessment if they reported original data on patients with AS receiving biological therapy, and reported outcomes reflecting participation in paid work: either presenteeism, absence from paid work or employment/work disability status. In the next step, full-text articles were assessed for those in which the title and abstracts had not provided sufficient information, using the same selection criteria. If there was doubt about whether or not an article should be included, the third author (AB) was consulted to reach consensus.
Risk of bias assessment
The risk of bias of included studies was assessed using two checklists from the Cochrane Collaboration: a checklist for cohort studies and one for randomised controlled trials (RCTs) (see online supplementary text).11 Two criteria were added to the checklists as the authors considered these criteria to be important in a study on work outcomes: (1) if a recall period for absenteeism, presenteeism and work disability was reported, and (2) if authors reported whether the work participation outcome could be attributed to AS, general health or another disease. All criteria had three possible answers: yes (adequate information and appropriate approach), no (no adequate information or inappropriate approach) or unclear (insufficient information to determine that criterion). If a specific criterion could not be answered, and a reference was made to another article for the same study, the referenced article was retrieved and checked. Two authors (LRAvdB, MMtW) assessed the risk of bias in the included studies independently. In total, the two authors disagreed on 15 items of the 92 items (16.3%) that were assessed. Discrepancies between both authors were discussed with the third author (AB) until consensus was reached.
Data were extracted by one author (LRAvdB), and checked by another (MMtW) using a self-composed form to extract the following: (a) study design and country of study; (b) intervention population: number of patients included, gender, mean age, number of patients employed, disease duration, eligibility criterion for biological therapy; (c) type of biological agent; (d) control population: number of patients included, gender, mean age, number of patients employed, disease duration; (e) work-related outcome measure and follow-up period; (f) statistical method used to quantify the effect of the biological agent on work outcome and (if applicable) confounders that were adjusted; and (g) results. Whenever possible the changes were presented to reflect differences between two groups (biological therapy compared with control) or two time points (after treatment compared with before treatment).
Owing to extensive interstudy heterogeneity in study populations, outcome measures and approaches to the reporting and analysis of outcomes, a meta-analysis to calculate an overall effect of biological treatment on work participation in patients with AS could not be performed. Therefore narrative summaries are provided.
Results of the search
A total of 124 titles were found through an electronic search on Pubmed/Medline, Embase and the Cochrane Library (figure 1). After deleting duplicates and screening titles and abstracts, 107 articles were excluded leaving 17 articles for full-paper review. Main reasons for exclusion were no work-related outcomes reported (n=59), and no TNF therapy administered (n=27). After a detailed full-paper review, another eight were excluded. Thus nine studies were included in this review: seven cohort studies, of which one was a controlled cohort study,12–17 ,20 and two randomised clinical trials.18 ,19 Scanning references of included studies retrieved no additional articles.
Characteristics of included studies
In total 961 patients were treated with biological agents: 851 patients (89%) in cohort studies and 110 patients (11%) in RCTs. Only TNFα inhibitors were used: infliximab in three articles,12 ,15 ,19 etanercept in two,18 ,20 adalimumab in one,16 and the above three TNF inhibitors considered as one group in three articles.13 ,14 ,17 Of the seven cohort studies, five studies were performed in a single country (UK, Germany or Sweden),12–15 ,17 and two were multinational studies (USA and Europe).16 ,20 Four of the seven cohort studies were open-label extensions of an RCT,12 ,15 ,16 ,20 two were registers of biological agents14 ,17 and one was a retrospective study.13 Of the two RCTs, one was a single country study (UK),18 and the other a multinational study (USA, Canada and Europe).19 In the cohorts, the average age of patients ranged from 39.5 to 49.9 years and disease duration from 10.6 to 16.4 years. For the RCTs this was 40.3 years and 11.0 years, respectively. Tables 1 and 2 show characteristics and extracted data for cohorts and RCTs separately.
Risk of bias in included studies
As can be seen in tables 3a and 3b, most risk of bias items were considered to be appropriately accounted for in the included studies. In four of the seven cohort studies an inappropriate approach was used in the selection of study participants (selection bias).12 ,15 ,17 ,20 Selective loss to follow-up was not appropriately accounted for in two cohort studies, and insufficiently reported in one.13 ,16 ,17 Four cohort studies did not adjust for potential confounders in the analysis.13 ,15 ,17 ,20 Four cohort studies did not report whether the work participation outcome could be attributed to AS, general health or another disease.14 ,15 ,17 ,20 In one RCT insufficient information on whether all groups were similar at baseline was provided.19 In both RCTs, it was unclear if all groups were treated equally, apart from the biological agent. In one RCT, it was unclear which recall period for the work participation outcome measures was used.18
In an open extension study (n=205, 156 weeks) a decrease in percentage of impairment while working due to AS, compared with baseline, was seen of 17.2% and 23.7% after 24 and 156 weeks treatment, respectively. Assuming a working week of 40 h, the end of study result would translate into 9.5 h of improved productivity per week. No significance level was presented.16 An open-label study (n=101, 28 weeks) showed a significant decrease in impairment in work productivity of 2.9 points on the Work Productivity and Activity Impairment Specific Health Problem questionnaire (WPAI-SHP) compared with baseline (p<0.001).12 Finally, in a register of biological agents, patients indicated retrospectively that work capacity had improved by 4.2 points, but no significance level was presented.13
One RCT (n=94, 24 weeks) reported a significant improvement of 2.1 points on a Visual Analogue Scale (VAS) (62%) on self-reported productivity at work (p<0.05).19 No significant improvement was found in the placebo group. The between-group difference was not reported. In the second RCT, in which presenteeism was expressed as an AS-related Work Instability Scale (AS-WIS) score, no significant improvement was found in patients treated with etanercept for 12 weeks compared with placebo (n=20, p=0.109).18
Absence from paid work
Absence from paid work was the outcome most often studied. Six out of seven cohorts (five uncontrolled12 ,13 ,15 ,16 ,20 and one controlled cohort study14), and both RCTs18 ,19 quantified the effect of biological therapy on absence from paid work. The follow-up varied from 12 to 156 weeks.
All cohorts reported a reduction in absence from paid work, although this was not always statistically tested. One small uncontrolled cohort (n=28, 102 weeks) showed that the percentage of people on sick leave and number of days on sick leave significantly reduced when patients continued to receive treatment (p<0.001, and p=0.002, respectively).15 These reductions were not seen in those who had to stop treatment at a certain point during the study (n=11). In another uncontrolled cohort, the mean number of sick days decreased significantly from 37.4 to 15.1 days after treatment (p=0.008).20 Two uncontrolled cohorts used the WPAI-SHP to assess absence from work. One of these found a decrease of 4.24% of work time missed while at work owing to AS after 156 weeks’ treatment (n=205), but the significance level was not tested. Based on a working week of 40 h, this improvement would correspond to 1.7 h of work time gained per week.16 The other cohort applying the WPAI (n=62, 28 weeks) reported a non-significant decrease of two missed work hours due to AS compared with baseline.12 A retrospective biologics register with variable follow-up found that patients who were currently receiving anti-TNF treatment had 14 fewer days mean sickness absence in the previous 12 months as compared with before treatment, but this was not tested for significance.13 The population-controlled cohort (nt = 139) showed that the relative risk (RR) of being on sick leave, compared with the background population, was reduced after 1 year of treatment (RR=4.0, 95% CI 2.0 to 6.3) as compared with treatment start (RR=9.2, 95% CI 5.4 to 15.7). The overlapping CI indicates that the change is not significant. It should be mentioned that 1 year before starting a biological agent, the RR for sick leave was 8.0 (95% CI 4.6 to 13.9), suggesting regression to the mean after the start of TNF inhibition. Within patients, the amount of sick leave, expressed as the percentage of full-time at work, was significantly reduced (mean change −8.1%, p<0.001) as compared with treatment start (18.1% sick leave).14
One RCT (n=163, 24 weeks) showed a significant reduction in work days missed (−1 day in the 6 weeks before a visit, p<0.01) compared with baseline. In the placebo group a non-significant reduction of 0.5 days was reported. The between-group difference was not reported.19 The second RCT showed that the proportion of patients losing work hours owing to AS was 40% (6/15 patients) in the etanercept group, compared with 30% (3/10 patients) in the placebo group. However, follow-up was only 12 weeks and the difference between groups was not significant (p=0.610).18
Of the three prospective cohorts, one uncontrolled study (n=101, 28 weeks) showed a significant improvement in the proportion of patients employed (+7.9%, p<0.001).12 In a prospective biologics register, the number of patients who were work disabled at the start (n=94) and became employed after 3 years (n=8; +8.5%), was higher than the proportion who were employed at baseline (n=94) and became work disabled (n=5; 4.0%).17 In an open extension study (n=84, 52 weeks) the full-time employment rate increased by 6.8%. Additionally, the mean number of work days increased for employed patients (n=58) from 226 days to 241 days (p=0.144).20 In the retrospective register (n=65) it was shown that after 3– 56 months the proportions of patients working full time increased (+9.2%), while the proportion of patients working part time and being temporarily disabled decreased. Since the proportion of patients in retirement remained stable, the observation indicates that the number of patients improving their work status outweighs those for whom work status worsened.13 Finally, the prospective population-controlled cohort showed that after the start of treatment, the risk of receiving work disability pension compared with the general population continued to increase slightly to a RR of 5.1 (95% CI 3.3 to 8) after 1 year compared with a RR of 4.4 (95% CI 2.8 to 6.7) at the start of the TNF inhibition.14
This study reviewed all published articles assessing the impact of biological therapies on work participation in patients with AS, comprising short-term productivity loss at work (presenteeism), absence from paid work (sick leave) and long-term employment status. In total, 39 comparisons were explored in nine articles comparing end points either with baseline or with a comparator group, and using different outcomes or assessing different time points after initiation of TNF. As can be seen in table 4 most comparisons suggest a positive work outcome (32/39), but these positive effects were often not tested for significance (14/32) or proved not significant (7/32). As in a comparable review of the effect of biological agents on work participation in RA, heterogeneity in study design, outcome measures and included patients, and the quality of studies hampered in-depth interpretation of results and the ability to reach strong conclusions.21 ,22 The lack of standardisation of outcomes is striking and was the main reason a meta-analysis could not be performed. A second important methodological problem concerns the sample sizes since work participation was never a primary outcome and therefore sample size calculations were not based on work outcomes. This probably explains why the observed changes were often not significant or the significance testing was not presented.
Despite the above-mentioned restrictions, a summary of the results shows that presenteeism in prospective studies improved 1 year after initiation of TNF by 1.7–2.9 points on a 0–10 VAS, the proportion of patients with sick leave decreased by 10–20% after 1 year of treatment and days absent from work decreased from 8.7 to 22.3 days per year.12 ,16 ,18–20 These results are impressive. However, it should be emphasised that the above data are usually based on before–after comparisons and do not take into account a placebo effect and regression to the mean. Both RCTs either showed no significant changes between groups or did not test statistical differences between groups.
The largest variation in outcome measurement was seen for work status. Although difficult to summarise, overall effects in work status were much smaller than in the other two outcomes. It is important to emphasise that follow-up was often too short to assess long-term changes. Only two studies assessing work status had a follow-up period of more than 1 year.13 ,17 Even more important, patients included in all studies had long disease duration (8–14 years) and the proportion of patients not employed at baseline was also considerable in all studies. This suggests that the negative impact of the disease on employment status had already occurred before starting treatment with TNF inhibitors, leaving limited opportunity for improvement.
The study performing a stratified analysis based on disease duration, showed better effects on sick leave, but especially on work disability, in those with shorter disease duration.14 The use of a population as a control group is in our view the most acceptable solution to performing long-term employment studies using a comparator with adjustment for secular trends. Therefore, the study by Kristensen et al is of considerable interest. While the RR to be on sick leave compared with the general population improved dramatically after the start of TNF inhibition, they also showed that the increased risk of sick leave returned to the same level as 1 year before the start of treatment. This suggests that regression to the mean in patients with flare-ups, accounts for part of the large effect. TNF inhibition could not prevent a small (insignificant) increase in the risk of work disability compared with the background population.14
One paper compared the possible gain in working time due to presenteeism with sick leave by assuming a linear relation between self-reported productivity and time lost.16 This suggests that potential savings in productivity costs due to TNF inhibition by improving presenteeism exceed the savings due to reduction in absence from work. It should be emphasised that there is still discussion as to whether self-reported productivity on a VAS is a valid proxy for ‘true’ productivity loss. Scarce research suggests that the relation is not linear and probably highly dependent on job and workplace characteristics.23 ,24 Therefore, while improvements in presenteeism are definitely important from a patient perspective, one should be cautious about converting self-reported productivity linearly to changes in productivity costs.
In summary, within the methodological limitations of the included studies, it remains unclear whether TNF inhibition in patients with longstanding AS improves presenteeism, absence from paid work or work status. Studies in patients with shorter disease duration, longer follow-up periods and preferably adjusted for secular trends using the general population as a control group to understand employment outcome, are urgently needed.
LRAvdB and MMtW contributed equally.
Contributors All authors made substantial contributions to design, execution and reporting of the manuscript, and saw and approved the final version. None of the authors have any conflict of interest with the content of this manuscript.
Competing interests None.
Provenance and peer review Not commissioned; externally peer reviewed.
Correction notice This article has been corrected since it was published online first. The second author's name has been corrected; the correspondence email address has been corrected; and a change has been made in the reference list.