Developing statistical methodology
Scientists in the MRC Biostatistics Unit collaborate with scientists here at the Institute to explore ways of analysing the enormous quantities of data collected in medical research.
The Cardiovascular Epidemiology Unit on the Emerging Risk Factors Collaboration (ERFC) established a central database of information on emerging cardiovascular risk factors in more than two million participants, from over 125 cohort studies. This database presents a number of possibilities to:
• Accurately quantify associations of risk factors with cardiovascular disease
• Find out if associations are stronger in certain groups of people
• Describe the ‘shape’ of associations
• Help us answer the questions of whether an association is confined
“This level of information provides us with the potential to explore new ways to predict the risk of cardiovascular disease,” says Ian White programme leader within the MRC Biostatistics Unit. “Which of course has implications for prevention and the impact of cardiovascular disease on wider public health.”
Dealing with such a large amount of data presents a number of challenges; including the difficulties of exploring and accounting for associations between risk factors and disease, when these associations differ between studies, handling unrecorded risk factors and understanding when a new risk factor is worth including in a predictive model. Data analysis must also allow for the different designs of the studies, and for measurement error in the risk factors.
The ERFC makes use of meta-analysis (the combining of summary results of several studies) to address a set of related research questions. In this study, it quickly became clear that the standard meta-analysis procedure was inadequate. A multivariate meta-analysis was needed, where observation and analysis of more than one statistical summary at a time could be performed. To achieve this, the MRC Biostatistics Unit developed the ‘mvmeta’ software, which has gone on to be employed by users of the statistical software Package, Stata, around the world. The Unit developed a model for assessing the health impact of measuring a new risk factor – to aid improved decisions about who should receive cholesterol lowering drugs.
Statistical epidemiology in ageing research
The large longitudinal cohort study (MRC CFAS) is used as a rich data resource with actual transitions between health states recorded. These transitions from active to disabled, and from disabled to dead, are modelled using modern statistical methodology to estimate the health of the general population.
“These estimates enable us to understand the health of the general population, and investigate whether the population’s health is changing over time”, says Fiona Matthews who leads this programme within theMRC Biostatistics Unit. “Ultimately these estimates allow policy makers to predict and understand the future health and care needs of the population”.
A number of statisticians within the MRC Biostatistics Unit have been involved in developing this methodology including solutions to problems such as missing data during the longitudinal phases and misclassifying the status of an individual.