Methods

Systematic Review of Action Rules in Chronic Disease

Project Lead: Dr Rafael Perera

There are several areas in which we are developing new methods for monitoring chronic diseases. One project that is nearing completion is a systematic review of action rules for the management of chronic diseases. Action rules and control charts are ways of distinguishing natural variability from true changes, and may be useful in avoiding unnecessary adjustment. We have also investigated the influence of measuring BNP on guiding treatment for patients with heart failure in general practice, criteria for monitoring, and the psychology of monitoring.

Criteria for Monitoring

Investigators: David Mant, Jeffrey Aronson, Paul Glasziou, Les Irwig

Not all monitoring is worthwhile. For example, a monitoring strategy designed to detect sudden and catastrophic events, but relying on infrequent assessment, is unlikely to improve the safety of treatment (carbimazole and agranulocytosis). Nor is there much use in a monitoring strategy designed to ensure therapeutic drug concentrations when the drug is usually administered in doses sufficient to ensure this. This piece of work aims to set the bases of when monitoring should be considered, developing and selecting the best strategy, and identify the issues for the practical use of monitoring.

Methods: Review of Control Chart Methodology

Investigators: Rafael Perera, Helen Doll, Yi Wan, Carl Heneghan

Statistical process control (SPC), is an umbrella term for several statistical methods that aim to achieve some form of quality control in a given process. The basic assumptions in SPC are the following: a) individual measurements from any process under control will exhibit variation – “common variation”, b) this variability arising from data of a process under control is predictable within a knowable range and can be modeled using a statistical distribution – e.g. Gaussian, binomial, Poisson, etc.-, c) if the control of the process is affected in some way, the new data generated will show some deviation from the statistical distribution used to model it – “external variation”, d) the statistical models obtained when the process is under control will allow us to establish control limits to detect when external variability (process out of control) is present. This study will involve a systematic review of the literature for chronic disease action rules from serial measurements that have been published in peer review journals.

Methods: Control chart simulation work

Investigators: Rafael Perera, Richard Stevens

Tests for Control (SPC rules) designed for manufacturing processes are being modified for their use in healthcare. The data generating processes have received little attention in the creation of these rules which could have costly implications (both health and economic costs). Therefore the SPC rules require some form of validation to verify that patients using control charts obtain benefit from their use (more adequate control). Based on our work modeling processes in areas like high cholesterol, diabetes (both types), hypertension, and ocular hypertension, we are now able to verify the validity of the SPC rules and propose adjustments based on a simulation strategy.

Psychology of Monitoring

Investigators: Alison Ward, Carl Heneghan, Rafael Perera

We aim to identify psychological factors associated with successfully self monitoring of oral anticoagulation therapy (SMOAT). To achieve this we will review all interventions in trials of SMOAT based on socio-cognitive theories and identify any psychosocial factors associated with successful SMOAT. We will also assess the effects of education interventions based on self-determination theory and/or autonomous self regulation.

Statistical Model for Medical Monitoring

Investigators: Rafael Perera, Richard Stevens, Jason Oke

The statistical methods of CII (cholesterol), CIII (cholesterol pre-treatment), HIV monitoring (see below) have much in common. In this paper we formalise the methods in a general statistical framework for these monitoring problems, review their use to date in our papers and others, and hence isolate and address some previously unresolved issues, such as the assumptions used about the source data and the need for confidence intervals.

Clinical Prediction Rules

Project Leads: Dr Matthew Thompson and Dr Richard Stevens

We are currently investigating whether it is possible to use published data to validate clinical prediction rules. Our early results, in validating coronary risk scores such as the Framingham equations in a population with diabetes, are encouraging. Meanwhile, individual-level data sets remain the gold standard for validating prediction rules, and our international collaboration funded by the HTA on prediction rules in acute childhood infections has progressed to the stage of data sharing between countries, which will allow a definitive analysis of prediction rules in this area.