74 Does work relatedness of an injury influence time on disability benefits in Brazil, a cohort study
基于巴西83,114名背痛工人数据，发现工伤认定与津贴领取时间弱相关，模型解释力低，需更多临床和工作场所因素来改进预测。
<h3>Objectives</h3> Evidence on predictors of time on benefits is mainly from developed countries. Evidence from emerging economies is lacking. In this cohort study predictors for time on disability benefits were identified within a Brazilian workers compensation insurance. Workers can claim benefits for either work related or work relevant conditions. This provided us with the unique opportunity to examine the role of work relatedness of back pain on time on benefits <h3>Methods</h3> In 2008, 83,114 workers diagnosed with back pain were claiming benefits. Claimants had > 15 days away from work. The analysis was adjusted for sex, age (/ 10 year), back pain benefits in 2007, claim rate of the industrial sector in 2007, and ICD-10 diagnosis (as agreed by two physicians). Duration of follow up was 52 weeks. Predictors for time on benefits in the first episode were identified by means of Cox regression analysis. Explained variance and c-statistic were calculated. <h3>Results</h3> Median time on benefits was 55 days (Inter Quartile Range (IQR) = 33–86). 1.49% of workers was on benefits after 52 weeks. Work relatedness of the claim was associated with the outcome: those with a non work related claim returned to work 1.04 times faster compared to those with a work related condition. Only age and diagnosis had hazard rate ratios over 1.2. Explained variance of the model was 3% (c-statistic<0.6). <h3>Conclusions</h3> Work relatedness is weakly associated with the outcome. The factors in this database have little explanatory power. More information on factors like: functional status, pain, recovery expectations, availability of workplace accommodations, physical demands and health care use might result in prediction that has utility in risk stratification an referral to early and appropriate intervention. In large administrative databases, statistical significance is easily attained therefore relevance criteria should be given and model fit should always be reported.