A joint meeting of the Royal Statistical Society Merseyside local group & Young Statisticians Section

2nd February 2011, 10.00-16.00

University of Liverpool Lecture Theatre C, University Lecture Block

Prognostic Modelling

10:00 - 10:45 Eleni Rapsomaniki (Research Associate, University of Cambridge)

A framework for comparing prognostic models based on net health benefits

New prognostic models are traditionally evaluated using measures of discrimination and risk reclassification, but these do not take full account of the clinical and health economic context. We propose a framework for comparing prognostic models in terms of the public health impact (net benefit) of the treatment decisions they support. This framework should form part of health economic evaluations that assess the value of adding information on new risk factors to a prognostic model. Our approach assumes a pre-defined treatment threshold so we compare risk predictions by their impact on treatment allocation. We extend previous work in this area by quantifying net benefits in life years, thus linking prognostic performance to health economic measures: by taking full account of the occurrence of events over time; and by considering estimation and crossvalidation in a multiple-study setting. The method is illustrated in the context of cardiovascular disease risk prediction using an individual participant data meta-analysis.

10:45 - 11:30 Lisa Pennells (Statistician, University of Cambridge)

Assessing discrimination across multiple studies

There is little agreement on the most appropriate measure to assess the predictive ability of a prognostic model. Completing such model assessments using data from multiple studies adds further complexity. In this work, different approaches for assessing the predictive ability of a Cox proportional hazards model across multiple studies are investigated. The focus is on measures of discrimination, Royston and Sauerbrei's D and Harrell's C-index, and examples are given using data from two large collaborations of prospective observational studies of cardiovascular disease. Reasons behind heterogeneity in results across studies are discussed and the appropriateness of pooling heterogeneous results is considered.

11:30-11:45 Coffee break

11:45 -12:30 Deborah Stocken (Biostatistics Lead, University of Birmingham)

Development and validation of a prognostic index for advanced pancreatic cancer
(Abstract to follow)

12:30 -13:15 Laura Bonnett (Research Assistant, University of Liverpool)

Factors influencing the risk of seizure recurrence in patients with epilepsy and the implications for driving

Driving regulations currently differ between EU member states. Consequently minimum standards to drive must be identified and implemented. In the UK the DVLA use a risk based approach to estimate when patients can regain their licenses - following an unprovoked seizure this equates to a year while new EU legislation requires only six months off driving. There are similar discrepancies concerning treatment withdrawal for patients in remission. UK patients are advised not to drive while withdrawing and for six further months and if a recurrence occurs and treatment is reinstated they must wait twelve months but in the EU the wait is only three months. Regression modelling was used to investigate how antiepileptic treatment and several clinical factors influence the risk of seizure recurrence following a single unprovoked seizure, completion of drug withdrawal, and treatment reinstatement for patients with recurrence. Several methods for external validation were employed to validate the model for time to second seizure following a single, unprovoked, seizure including May's comparison of deviances. Methods for handling covariates with all entries missing were also considered in the context of external validation via a simulation study. Advice is needed as to how this data should be utilised: in particular whether a population approach should be taken with a focus on the unadjusted results or whether attempts should be made to individualise risk. Guidance is also required as to whether the focus should be on risk estimates only or on the confidence interval as well. If the focus is on the estimate only our unadjusted estimates suggest that UK legislation concerning time off driving after all three events may be too conservative while EU standards may be too liberal. If the focus is placed on the confidence intervals, direction is needed as to whether a conservative or liberal approach should be taken.

13:15-14:15 Lunch break

14:15 -15:00 Willi Sauerbrei (Professor, Universitaetsklinikum Freiburg)

A new strategy for meta-analysis of continuous covariates in prognostic factor studies
(Abstract to follow)

15:00 -15:45 Richard Riley (Senior Lecturer, University of Birmingham)

Individual participant data meta-analysis of prognostic factor studies: state of the art?

Prognostic factors are associated with the risk of a subsequent outcome in people with a given disease or health condition. Meta-analysis using individual participant data (IPD), where the raw data are synthesised from multiple studies, has been championed as the gold-standard for synthesising prognostic factor studies. We assessed the feasibility and conduct of this approach, using a systematic review of currently published IPD meta-analyses of prognostic factor studies. Forty-eight published IPD meta-analyses of prognostic factors were identified up to March 2009. Only three were published before 2000 but thereafter a median of four articles exist per year, with traumatic brain injury the most active research field. Availability of IPD offered many advantages, such as checking modelling assumptions; analysing variables on their continuous scale with the possibility to assessing for non-linear relationships; and obtaining results adjusted for other variables. However, researchers also faced many challenges, such as large cost and time required to obtain and clean IPD; unavailable IPD for some studies; different sets of prognostic factors in each study; and variability in study methods for measurement. In this talk I will summarise our review and highlight the advantages and disadvantages of the IPD approach. I will also suggest where improvements are needed; in particular, continuous variables are often categorised without reason; publication bias and availability bias are rarely examined; and reporting standards are often sub-standard.

ALL ARE WELCOME. ALTHOUGH THE MEETING IS FREE TO ATTEND, THERE IS A CHARGE OF £6 FOR LUNCH.

Please register for this event at: https://sites.google.com/site/rssmerseyside/research-meetings/prognostic-modelling/registration-form

Further details at: https://sites.google.com/site/rssmerseyside/research-meetings/prognostic-modelling
Contact: Laura Bonnett ([log in to unmask]) or Jamie Kirkham ([log in to unmask])

You may leave the list at any time by sending the command

SIGNOFF allstat

to [log in to unmask], leaving the subject line blank.