Using climate to predict disease outbreaks: a review:
Previous page: Climate-based early warning systems for infectious diseases |
General discussion and conclusions
It is generally accepted that the transmission of many infectious diseases is affected by climatic conditions. Diseases caused by pathogens which spend part of their life cycle outside of human, or other warm-blooded, hosts are particularly climate-sensitive. Some of these diseases are among the most important global causes of mortality and morbidity, particularly in poorer populations in developing countries. In many environments, these diseases occur as epidemics, possibly triggered by changes in climatic conditions favouring higher transmission rates.
Efforts to develop climate-based disease EWS date back to the work of Gill and co-workers in India in the 1920s. Interest has been rekindled in recent years, however, reflecting in part increasing levels of concern over possible future impacts of climate change on human society. At the same time climate and other environmental data have become widely available and relatively inexpensive, as have GIS and other tools required to link these observations with disease data. There is, therefore, clear justification for in-vestigating climate based EWS’ potential to allow advance planning of control interventions. The case for such EWS has been made repeatedly in review papers, particularly in the context of malaria.
In this report, we have reviewed the degree to which important infectious diseases are sensitive to climate variations, and used this as a basis for identifying diseases for which climate EWS may be most useful. We have adapted existing work on malaria to form a generalized framework for developing EWS for infectious diseases. Subsequently we review the extent to which existing systems provide accurate advance warnings of the likelihood and size of epidemics, which are useful in making control decisions.
These sections show that there is considerable research activity in this area. Of the diseases that meet our criteria for having the potential for climate-based EWS, only a few (African trypanosomiasis, leishmaniasis, yellow fever and Murray Valley encephalitis) have no reports of an EWS being developed. For others (St. Louis encephalitis and West Nile virus in the United States) there are operational and effective
warning systems, but these rely solely on rapid detection of virus activity – i.e. similar to the strategy employed for early detection and pre-diction of influenza outbreaks. It remains unclear whether adding climatic predictors would improve predictive accuracy or the lead-times associated with these systems. For the remaining diseases (cholera, malaria, meningitis, dengue, Japanese encephalitis, Rift Valley fever and Ross River virus) research projects have demonstrated a temporal link between climatic factors and variation in disease rates. In some of these projects the power to predict epidemics has been tested already, although in many cases the tests are preliminary, based either on a very limited dataset, or with little description of the methods used. There are no published reports indicating that any of these systems currently are used for influencing control decisions (see Table 3), although efforts are being made to set up and validate such EWS for malaria (Southern Africa Malaria Control (SAMC), unpublished reports).
It is not clear why such systems are not widely used, but we suggest a number of likely explanations. Firstly, affordable and accessible data and analytical tools have become widely available only recently so that the field is at a relatively early stage of development. Many more studies should be available in the next two to three years as systems are completed and tested in other locations. Secondly, as few studies have been published there are no generally agreed criteria for accessing predictive accuracy. Consequently it is difficult to judge the utility of existing systems. Thirdly, most research projects have been carried out on relatively limited resources and therefore have not been tested in locations outside of the original study area. Fourthly, most studies in this area focus solely on climate factors and do not explicitly test other explanations for variations in disease rates through time. Finally, as such studies are often under-taken as pure research it is not clear to what extent they address specific control decisions and are of use to health policy-makers.
Of the several possible ways to help to address these issues, perhaps the most urgent is the need to maintain and strengthen systems for reporting incidence of epidemic diseases. High-quality, long-term disease data are essential for generating models relating climate to infectious disease. It is probably true to say that development of EWS for some diseases has stalled because of a shortage of suitable epidemiological data. More commonly, disease/climate modelling has been restricted to discrete datasets for relatively small areas. These exercises are useful for exploring methodological issues and in many cases have produced promising results although there are questions concerning the extent to which findings from these studies can be generalized. The implications of this are that before EWS can be widely tested or applied, usually it will be necessary to bolster existing disease surveillance systems. In some cases there may be a need to begin this process from scratch – in others, viable systems may exist but require modification to ensure timely transfer of data from the point of collection to the point of analysis. For diseases such as malaria, which often are diagnosed clinically, further work needs to be carried out to determine the extent to which quality of diagnosis affects our ability to recognise (and predict) epidemics.
There is a need for clear definitions of terminology and methods for assessing predictive accuracy. If the aim of an EWS is to predict epidemic versus non-epidemic time periods, the definition of an epidemic (i.e. number of cases in a specific population over a specified time) should be determined before the modelling process is carried out. The accuracy of such systems could be measured using standard epidemiological measures such as sensitivity, specificity, positive and negative predictive value, and kappa statistics. The accuracy of models which attempt to predict case numbers could be measured as the root mean square error, or as correlation coefficients between observed and predicted case numbers. In all cases, model accuracy should be assessed against independent data (i.e. not included in the original model building process) to give an accurate replication of an attempt to predict a future epidemic. Predictive models ideally should be tested in a wide range of locations and, if necessary, adjusted to take account of geographical variations in climate-disease relationships.
It is important for any EWS to test for non-climatic influences (e.g. the effects of population immunity, migration rates, drug resistance) on variations in disease rates. Thorough testing of alternative explanatory factors should avoid incorrectly attributing disease variations to climate. More importantly in practical terms, measurements of all relevant factors for which data are available should allow the generation of more accurate predictive models.
As research into EWS moves beyond the pure research stage, it becomes increasingly important to include health policy-makers in all stages of system design. For example, local disease control personnel should be involved in defining an epidemic and in determining the most appropriate lead-times over which predictive accuracy should be assessed (e.g. whether it is more important to have an accurate prediction with a lead-time of one to two weeks, or a more uncertain prediction with a lead-time of several months). These discussions should take place in relation to specific control decisions, and consider local (particularly resource) constraints on the implementation of the EWS. Experience with the famine EWS in the 1990s showed that its effectiveness depended less on the accuracy of warnings than on political factors.
The final decision over whether an EWS should be implemented ideally should be made on the basis of a cost-effectiveness analysis. This should measure the value of information in collecting data on the various climatic and non-climatic influences, in terms of both predicting the occurrence and size of epidemics and increasing the effective use of control resources. In some situations, for example, adding climatic information to an early warning system may give only a small increase in predictive power and therefore cost-effectiveness of control: however it may be sufficiently cheap and simple to collect to justify inclusion.
Table 3 . Summary of the development of EWS for infectious diseases: current state of the art and future requirements [pdf 31kb]
Using climate to predict disease outbreaks: a review:
| Next page: Bibliography