By their very nature, observational studies have an added layer of complexity when compared to Randomised Clinical Trials (RCT). There are often very many more controls than cases and the lack of design in the study may cause the two groups to be very different in terms of measured covariates. Some of the hurdles of analysing observational studies as well as methods to overcome them will be discussed.
Matching methods can be used to implement a 'design' by matching each case to a control or controls, minimising the distance (based on the observed covariates) between each control and case pair. While matching is a useful technique, it can often lead to omitting a large fraction of the dataset, particularly if we use one to one matching. Alternatively, we can use various weighting techniques in the analyses; these weights may be based on the number of controls matched to each case when matching is carried out using all patients, or the inverse propensity score. The propensity score is the conditional probability of being a case given the observed covariates.
The above techniques will be illustrated with data from an observational study carried out on colorectal cancer, where the cases of interest were patients with a secondary condition. The data set contained over 20,000 control patients and 170 patients with the secondary condition and is an extreme example of imbalance between the number of cases and controls.
Keywords: Observational study; Inverse Probability Weighting; Survival Analysis; Matching
Biography: Cara is currently studying for a PhD in NUI, Galway under the supervision of John Hinde. Previous to this Cara, received a Masters in University of Kent and worked for 4 years as a non-clinical statistician, in Industry.