A multi-state model for sick-leave data applied to a randomized control trial study of low back pain

Abstract

Aims: Analysing and presenting data on different outcomes after sick-leave is challenging. The use of extended statistical methods supplies additional information and allows further exploitation of data.

Methods: Four hundred and fifty-seven patients, sick-listed for 8-12 weeks for low back pain, were randomized to intervention (n=237) or control (n=220). Outcome was measured as: "sick-listed'', "returned to work'', or "disability pension''. The individuals shifted between the three states between one and 22 times (mean 6.4 times). In a multi-state model, shifting between the states was set up in a transition intensity matrix. The probability of being in any of the states was calculated as a transition probability matrix. The effects of the intervention were modelled using a non-parametric model.

Results: There was an effect of the intervention for leaving the state sick-listed and shifting to returned to work (relative risk (RR)=1.27, 95% confidence interval (CI) 1.09- 1.47). The nonparametric estimates showed an effect of the intervention for leaving sick-listed and shifting to returned to work in the first 6 months. We found a protective effect of the intervention for shifting back to sick-listed between 6 and 18 months. The analyses showed that the probability of staying in the state returned to work was not different between the intervention and control groups at the end of the follow-up (3 years).

Conclusions: We demonstrate that these alternative analyses give additional results and increase the strength of the analyses. The simple intervention did not decrease the probability of being on sick-leave in the long term; however, it decreased the time that individuals were on sick-leave.