Software tools and online resources for the self-controlled case series method and its extensions

The self-controlled case series (SCCS) method is a statistical technique to quantify the association between an exposure, such as a drug, and an outcome, such as an adverse event that might or might not be related to the exposure. The method is an alternative to standard techniques such as cohort studies and case-control studies which are commonly used to quantify such associations.

The SCCS method can be attractive in certain circumstances because, unlike standard techniques, it is not subject to bias due to confounding by variables that are fixed in time for the duration of the study (for example, genetic factors, socio-economic context, or underlying health condition). Confounding can occur when a variable is associated with both the exposure and the outcome, and distorts the apparent association of interest. The method is particularly attractive for use with data from large databases that may have been assembled for reasons unrelated to the study, as is the case with administrative databases, for example. In such databases, information on important confounders, such as whether a person smokes or not, or their socio-economic background, or general state of health, may not be available. This means that the effect of these variables can't be allowed for in standard methods - whereas in the SCCS method, this is achieved automatically.

The SCCS method has gained in popularity over the past decade, and is now often used in the area of medicine that deals with the effects of pharmaceutical drugs in populations, known as pharmacoepidemiology. However, the advantages of the SCCS method come at a price in the form of strong assumptions, that may or may not be valid in any particular setting. Over the past years, extensions of the SCCS method have been developed to cater for situations in which some of these assumptions are violated - while still retaining the essential feature of controlling automatically for fixed confounders.

However, most of these extensions are a lot more complicated to apply than the basic SCCS method, and are not available in standard commercial or academic software packages. This greatly limits the use of these extensions of the SCCS method by researchers whose primary expertise is not statistics. The aim of the project is to provide programs and documentation within such software packages (notably R, STATA and SAS) to enable researchers to make use of these techniques more readily.

We aim to provide comprehensive online resources including software programs, examples of their application including suitable data, examples of what might go wrong when assumptions are not met, together with documentation to describe how to run the programs, and information on how others have used the method. All materials and programs will be provided free of charge. We will also develop new methodology, as required, to plug gaps in the methods available, and to help users make informed choices about the SCCS models they should use.

We expect the results to be of direct use to medical researchers and applied statisticians working in pharmacoepidemiology, and in epidemiology more widely. Our experience with the resources we have so far made available for the basic SCCS method shows that they are well used. The major impact we expect from this work is to improve the quality and quantity of studies undertaken with the SCCS method, and hence to contribute to providing better evidence underpinning medical decisions.

The main objective of the project is to program extensions of the self-controlled case series (SCCS) method in standard statistical software, namely R, STATA and SAS. These extensions broaden the applicability of the basic SCCS method to situations in which key assumptions may be violated.

The extensions we plan to program include (a) event-dependent exposures, where the conditional Poisson likelihood framework of the basic method is replaced by one based on estimating equations, with confidence intervals obtained by bootstrapping; (b) events carrying high mortality, resulting in event-dependent observation periods, which requires a 2-stage model incorporating an explicit model for post-event survival; (c) more flexible temporal adjustments including seasonal as well as age effects, and smooth age effects and post-exposure risk functions using splines or kernel smoothing; (d) dependent recurrent events, where the SCCS model can be retained when the dependence is expressed additively rather than multiplicatively on the baseline hazard. We shall also provide software to aid the design and interpretation of SCCS studies, incuding a sample size calculator and graphical tools to plot output.

In addition we shall undertake a comprehensive review of how the SCCS method has been used, and of where any problems in its application might lie, so as better to address these in the new online documentation we shall produce. This will include practical data-based examples, instances of what can go wrong if assumptions are violated, and suggestions for sensitivity analyses and diagnostic procedures.

These materials will be freely available on a new SCCS website, aimed at epidemiologists and applied statisticians wishing to use and learn about the SCCS method. Past experience suggests that such resources are well used. Our ultimate goal is to increase the quality and quantity of studies using the SCCS method, and hence contribute to better evidence-based medical decisions.

The main users of the project will be epidemiologists and scientists working in pharmacoepidemiology within the academic, commercial and public health sectors. This assessment is based on our knowledge of who already uses the self-controlled case series (SCCS) method, obtained by from the home institutions of the authors of published studies that have made use of the method.

The most prominent users will be international and national public health agencies who have used or recommended the SCCS method in the past, which include the World Health Organisation, the European Centre for Disease Prevention and Control, the Centres for Disease Control and Prevention, Health Protection England, the Robert Koch Institute, the Institut National de la Sante et de la Recherche Medicale, and others.

Commercial enterprises that are likely to use the project outputs include pharmaceutical companies, several of which have used the SCCS method in the past, and custodians of large computerised databases, such as Kaiser Permanente in the USA and other healthcare maintenance organisations.

The ultimate beneficiaries will be members of the public, who will benefit from better treatment decisions resulting from appropriately conducted epidemiological studies.

The benefits of using the SCCS method appropriately include more timely and cheaper studies, and effective control of major sources of confounding. These benefits may translate into economic and public health advantages: for example, timeliness is particularly important in the case of vaccine safety issues; and control of confounders is particularly tricky when using electronic databases assembled for reasons other than medical research. Improving the accessibility of the SCCS method and its extensions will contribute to making these benefits more widely available, while improving understanding of the method will improve the quality of its application. These benefits will be lasting, and available immediately upon completion of the project.

In the longer term, the project will also help maintain the leading position of the UK in the development of this type of methodology.

The project will help the career development of the Co-I and the Research Assistant, by providing them with unique expertise and a leadership position in an area of statistical methodology that is attracting increasing interest from both epidemiologists and applied statisticians worldwide.

Additionally, any peer-reviewed publications arising from this grant will be registered on the Open University's open access institutional repository - Open Research Online (ORO) at http://oro.open.ac.uk. ORO is one of the largest repositories in the UK. The site receives an average of 40,000 visitors per month from over 200 different countries and territories and has received over 2.5 million visitors since 2006. It enables access to research outputs via common search engines including Google, by using the OAI (Open Archives Initiative) Protocol for Metadata Harvesting.