The best way to find out whether a treatment for stroke works is to combine the results of several clinical trials using a statistical process called a ‘meta-analysis’. The current methods for performing a meta-analysis work well with certain types of data but are not that good at evaluating the outcomes that matter to stroke survivors, such as quality of life or the length of hospital stay after stroke. These outcomes are often distributed unevenly across the population, with the majority of patients clustered at one end of the scale. For example, most stroke survivors will be discharged from an acute stroke hospital ward within a week to 10 days, but a small number of patients will stay in hospital for longer than a month. ‘Skewed’ data like these are often analysed inappropriately or have to be left out of meta-analysis completely, leading to biased or weaker conclusions about how well a stroke treatment works.
To address the problem, this study will (1) find alternative statistical approaches for handling skewed data in stroke clinical trials, and (2) explore ways of extracting necessary statistical data when it is missing from clinical trial reports, thus allowing more data to be included in meta-analysis. Thousands of stroke survivors have participated in clinical trials investigating new stroke treatments. The approaches developed in this study will allow the data collected in stroke clinical trials to be evaluated more precisely, providing better information to guide the future treatment of stroke patients.