Date: No date given

Loading map...

This event is being held in honour of Professor Doug Altman and will be chaired by Martin Bland

Speakers:

Toby Prevost: “Analysis of serial measurements in medical research and in clinical trials”

Abstract:
In an article published in the BMJ in 1990 (now cited >2400 times), Matthews, Altman and colleagues considered the analysis of serial measurements in medical research and recommended a two-stage approach based on summary measures. In this talk, I will describe the major themes in the article and their application to the design of a clinical trial. Briefly, parallels will be drawn between the role of summary measures in trials of individual and cluster randomised designs, and between analysis by summary measures and by mixed models.
Reference: Matthews JN, Altman DG, Campbell MJ, Royston P. Analysis of serial measurements in medical research. BMJ 1990;300:230-5

Mark Ashworth: “Some Problems with Primary Care Data”

Abstract:
Primary care is a rich source of statistical data. It is also a source of considerable problems for the purposes of statistical analyses. I will give examples of where ‘big data’ from primary care may produce misleading data outputs. 
I will discuss three national sources of primary care data, CPRD, THIN and data from the Royal College of General Practitioners’ monitoring practices, comparing their data formats and their use as assets for data analyses. I will contrast this with locally available patient-level data such as Lambeth DataNet.
Numeric estimates may improve communication and risk perception in general practice. We get different perspectives when risk perception is based on NNT, Absolute Risk, or the still predominant use of Relative Risk, in making clinical decisions. We also need to exercise caution in basing health service decisions on univariable rather than multivariate data in primary care. I will give examples of where decisions have been made using misleading data.
Finally, I will be looking at the use of primary care data analyses from the perspective of patient care in primary care. GPs use data for the communication of risk, balancing evidence-based medicine with patient preference, and I will give examples applying this to breast cancer, prostate cancer and CVD risk. I will then use these arguments to pose the question of why rational patients may reject statistically based advice?

Camila Caiado: “Developing Planning Tools for GP Practices: Using Population and Health Care Records”

Abstract:

We are constructing models and apps that provide health providers and local authorities with powerful tools for better decision making and planning. An app has been developed with GP practices for predicting patient flows and behaviour and help health providers to understand and prepare for both current and future demands. The app also allows managers and clinical teams to see projections of GP practice activity with different population growth scenarios.

Gary Collins: “ISSUES on Prognosis and Prediction in Medical Research”

Abstract:

Decisions are routinely made by healthcare professionals on the basis of an estimated probability for diagnosis and prognosis. Models which combine multiple predictors are used as a basis for estimating this probability. A number of models are now embedded in general practice software, such as the QRISK model for predicting the 10-year risk of developing cardiovascular disease.  Traditionally, prediction models have been developed using regression based approaches such as logistic or Cox regression depending on the outcome being predicted. More recently, there has been a surge in interest in using modern approaches under the umbrella term of machine learning. The interest and uptake in machine learning is often accompanied with enthusiastic claims that they can provide superior predictive accuracy over traditional approaches, however recent systematic reviews have highlighted various shortcomings in the design, conduct and reporting of these comparative studies.

 

In this talk I will discuss, with examples, some of the challenges in developing and validating prediction models using both traditional and machine learning approaches.  I will also critically highlight some of the more common methodological, terminology and reporting issues encountered in prediction models studies.

 

Attendance is free and open to all, but pre-registration is required.

Venue: The Royal Statistical Society

City: London

Country: United Kingdom

Postcode: EC1Y 8LX

Organizer: Royal Statistical Society

Event types:

  • Workshops and courses


Activity log