Statistical modelling for real-time spatial surveillance and forecasting
Lead Research Organisation:
Lancaster University
Department Name: Division of Health Research
Abstract
Public health outcomes are typically associated with both a specific place and a specific time, and therefore generate spatio-temporal data. Timely analysis of these data can assist in both the design and the prompt delivery of public health interventions. However, most currently available statistical methods for the analysis of such data operate off-line, and often with considerable delay between the occurrence of the events in question and completion of the analysis of the resulting data.
In this project, we propose to develop novel statistical methods that can be run in real-time, and to implement these in software systems that respond automatically to newly acquired data as they accrue in real-time.
We will embed the methodological developments within two contrasting applications. The first is to the forecasting and control of emerging meningitis epidemics in sub-Saharan Africa; Meningitis is a major public health problem in this part of the world, where between 3% and 30% of the population carries the infective agent, and the mortality rate amongst cases can be as high as 10%. Our second application is to the monitoring and reporting of calls to NHS Direct, classified by place, time and syndrome; the aim is to strengthen current surveillance systems based on NHS Direct data, by providing information at finer spatial and temporal resolutions than hitherto.
In this project, we propose to develop novel statistical methods that can be run in real-time, and to implement these in software systems that respond automatically to newly acquired data as they accrue in real-time.
We will embed the methodological developments within two contrasting applications. The first is to the forecasting and control of emerging meningitis epidemics in sub-Saharan Africa; Meningitis is a major public health problem in this part of the world, where between 3% and 30% of the population carries the infective agent, and the mortality rate amongst cases can be as high as 10%. Our second application is to the monitoring and reporting of calls to NHS Direct, classified by place, time and syndrome; the aim is to strengthen current surveillance systems based on NHS Direct data, by providing information at finer spatial and temporal resolutions than hitherto.
Technical Summary
Statistical methods for analysing separately the patterns of variation in health outcomes over spatial and over time are now well-established. Methods for the integrated analysis of spatio-temporal variation are less well developed, especially in forms suitable for real-time implementation. At the same time, spatially and temporally referenced health outcome data is becoming more common, but the potential to interrogate the resulting data-bases in real-time is under-exploited. The core aim of this project is to develop novel stochastic models and statistical methods for the real-time analysis of spatio-temporal
data, and to link these to real-time health outcome data-streams. The methodology will be generically applicable, but its development will be embedded in two contrasting exemplar applications, one each in developing and developed country settings. The first of our applications will be to the forecasting and control of emergent epidemics of Meningococcal
meningitis in sub-Saharan Africa. Previous work in the Sahel has explored the potential for statistical modelling to assist in predicting the incidence of meningitis by regressing meningitis incidence on a number of satellite-derived environmental and meteorological variables. The recent MERIT (Meningitis Environmental Risk Information Technologies) meetings in Ethiopia (Dec 1-3 2008) recommended the development of an empirical predictive model for meninigits incidence, capable of being implemented in real-time at the finest spatial and temporal resolutions consistent with the availability of reliable incidence and risk-factor data.
Our second area of application will be to the real-time analysis and reporting of information generated by calls to NHS Direct, the UK s 24/7 phone-in health advice service. The normal patterns of variation in the use of this service are complex, depending on spatial, temporal and socio-demographic factors. Successful modelling of these patterns would enable unexplained variations in use to be used in an early warning system for emergent public health problems, at very fine spatial and temporal resolutions We have previously demonstrated the feasibility of this using gastro-enteric symptoms in Hampshire; we now wish to extend the system to other syndromes, notably flu-like illness, and to national coverage.
data, and to link these to real-time health outcome data-streams. The methodology will be generically applicable, but its development will be embedded in two contrasting exemplar applications, one each in developing and developed country settings. The first of our applications will be to the forecasting and control of emergent epidemics of Meningococcal
meningitis in sub-Saharan Africa. Previous work in the Sahel has explored the potential for statistical modelling to assist in predicting the incidence of meningitis by regressing meningitis incidence on a number of satellite-derived environmental and meteorological variables. The recent MERIT (Meningitis Environmental Risk Information Technologies) meetings in Ethiopia (Dec 1-3 2008) recommended the development of an empirical predictive model for meninigits incidence, capable of being implemented in real-time at the finest spatial and temporal resolutions consistent with the availability of reliable incidence and risk-factor data.
Our second area of application will be to the real-time analysis and reporting of information generated by calls to NHS Direct, the UK s 24/7 phone-in health advice service. The normal patterns of variation in the use of this service are complex, depending on spatial, temporal and socio-demographic factors. Successful modelling of these patterns would enable unexplained variations in use to be used in an early warning system for emergent public health problems, at very fine spatial and temporal resolutions We have previously demonstrated the feasibility of this using gastro-enteric symptoms in Hampshire; we now wish to extend the system to other syndromes, notably flu-like illness, and to national coverage.
