Recent concern about the effectiveness of alternative treatments in medicine and health, in education, in psychology, and in the social sciences has led to a consideration of how to combine or synthesize the results of independent studies. Historically, integration of independent results focused on how to combine p-values.
More recently the emphasis has been on estimating effect sizes, which in turn has motivated a variety of alternative models depending on the experimental conditions. We here review the development of the field with an emphasis of diagnostics and future research.
I would like to discuss meta-analysis from a different point of view; namely, that the task of combining information is inherent in many areas of research. Because methods of analysis may have been developed in other fields, we need to become aware of these methods for possible adaptation to medical analyses.
One of the prime factors drawing the increase in the number of meta-analyses is the information explosion. I am not sure that the degree of this information explosion is fully recognized. From the 1940s to the 1990s there was a ten-fold increase in the number of publications, from 2300 to about 25000 biomedical journals. In psychology there was an increase from 91 to 1100 journals; in mathematics, from 91 to 920 journals. I have reviewed the figures in a number of different fields and the factor of the information explosion is in the order of 9 to 12, depending on the field. A consequence of this increase is a need for synthesis, which occurs in almost every field. Several examples suffice. In weather modification there are now about 900 projects to detect the effectiveness of cloud seeding. A point that I want to emphasize is that the problems in weather modification are very similar to those that occur in medicine. For example, experiments are performed in different months, by different organizations using different methods, by different agents, and with different endpoints. Variability in experimentation is common to almost all fields.
Agriculture is one of the earliest fields in which meta-analysis was used. One of the agricultural areas in which I have collaborated in pomology. This field includes fruits and nuts but excludes vegetables, and is a very important industry in California. There are many experiments related to pest control, to storage, to increasing yield, and so on. In these studies there are diverse measures: appearance; firmness; yield; fruit size; colour; flavour; shelf life; time to production, and so on. What you may notice is that some endpoints are hard endpoints, others are soft endpoints, similar to the measures that you find in medicine and other fields. This suggests a commonality of issues across fields.