Data from a cohort of women treated with bisphosphonates were used to illustrate that multi-state models may be the useful tools of analysis where two causes of treatment failure are the ultimate outcome of interest, and the sequence of recurrent episodes of treatment starting and discontinuing is also of concern.
All the 11 863 women resident in the Italian Region of Lombardy, aged 45 years or over and who received bisphosphonates for the first time during 2003 entered into the study and were followed up until December 2005. Multi-state models with in-treatment and treatment-free periods as transient states, and fractures and gastrointestinal-related events as competing causes of failure, were fitted to the data. The effect of several covariates on transition rates between states was estimated.
One half of the women was treated with bisphosphonates for less than 27% of the period of observation. Negative prognostic factors for treatment discontinuation were younger age and the use of alendronate at once-daily dosing, with low refill compliance during follow-up. The risk of fracture was higher for older women who experienced fracture before entry in the cohort, and for those who switched between drugs, had low refill compliance, experienced episodes of treatment discontinuation and used corticosteroids during follow-up.
The use of multi-state models in pharmacoepidemiology provides non-biased estimates of the effect of covariates in predicting treatment discontinuation, restarting and failures. Our analyses emphasize the importance of maximising compliance in order to reduce the risks of both treatment discontinuation and fracture. Copyright © 2008 John Wiley & Sons, Ltd.