OBJECTIVES: There are many examples in health outcomes research where inadequate control for comorbidity influence has resulted in effect estimates confounded by disease severity. Selection bias is a common feature of data from routine healthcare setting where the decision to give a particular drug to a patient with a given disease is generally based on patient characteristics, including disease condition. Thus, failure to properly control for the bias could result in false associations. Propensity scores methodology is commonly used despite its limitations because of its potential for minimising the association between exposure and confounding factors. We describe a methodology for assessing drug effect in longitudinal data that minimises confounding by disease severity generally associated with observational studies. METHODS: For a particular outcome of interest, we obtain the profiles of rates ratios from two sets of matched cohorts. In set A, patients with disease X are compared with others free of X in the periods prior to and post diagnosis of X. In set B which involves only patients with disease X, those exposed to treatment Y are compared with those unexposed to the drug in the periods prior to and post exposure. The two sets of profiles are then assessed using simple regression over the respective periods. In effect, we attempt to disentangle the disease and treatment effects. Data from the UK GPRD are used to assess possible association between a particular outcome and treatment in COPD RESULTS: We found evidence of association between the outcome and COPD but none for the drug. CONCLUSIONS: The profile approach utilizes the data collected over the disease natural history and exposure history to assess the relationships between the outcome and both the disease and treatment. This is a key strength often ignored when results are reported as point estimates. By design, it also minimises the effect of selection bias.