Non-randomised studies of the effects of interventions are critical to many areas of healthcare evaluation, but their results may be biased. It is therefore important to understand and appraise their strengths and weaknesses. We developed ROBINS-I (“Risk Of Bias In Non-randomised Studies - of Interventions”), a new tool for evaluating risk of bias in estimates of the comparative effectiveness (harm or benefit) of interventions from studies that did not use randomisation to allocate units (individuals or clusters of individuals) to comparison groups. The tool will be particularly useful to those undertaking systematic reviews that include non-randomised studies.

Although renal osteodystrophy and vitamin D analogs may be related to survival in maintenance hemodialysis (MHD) patients, most studies have examined associations between baseline values and survival without accounting for variations in clinical and laboratory measures over time. We examined associations between survival and quarterly laboratory values and administered paricalcitol in a 2-year (July 2001-June 2003) cohort of 58,058 MHD patients from all DaVita dialysis clinics in USA using both time-dependent Cox models with repeated measures and fixed-covariate Cox models with only baseline values. Whereas hypercalcemia and hyperphosphatemia were robust predictors of higher death risk in all models, the association between serum calcium and mortality was different in time-varying models. Changes in baseline calcium and phosphorus values beyond the Kidney Disease Outcome Quality Initiative recommended targets were associated with increased mortality. Associations between high serum parathyroid hormone and increased death risk were masked by case-mix characteristics of MHD patients. Time-varying serum alkaline phosphatase had an incremental association with mortality. Administration of any dose of paricalcitol was associated with improved survival in time-varying models. Controlling for nutritional markers may introduce overadjustment bias owing to their strong collinearity with osteodystrophy surrogates. Whereas both time-dependent and fixed-covariate Cox models result in similar associations between osteodystrophy indicators and survival, subtle but potentially clinically relevant differences between the two models exist, probably because fixed models do not account for variations of osteodystrophy indices and changes in medication dose over time.