Teleoanalysis: combining data from different types of studyBMJ 2003; 327 doi: http://dx.doi.org/10.1136/bmj.327.7415.616 (Published 11 September 2003) Cite this as: BMJ 2003;327:616
- 1 Wolfson Institute of Preventive Medicine, Barts and the London, Queen Mary's School of Medicine and Dentistry, University of London, London EC1M 6BQ
- Correspondence to: N Wald
- Accepted 9 July 2003
Teleoanalysis can provide the answer to questions that would be obtained from studies that have not been done and often, for ethical and financial reasons, could never be done
Once a causal link has been established between a risk factor and a disease it is often difficult, and sometimes impossible, to determine directly the dose-response relation. For example, although we know that saturated fat intake increases the risk of ischaemic heart disease, the exact size of the effect cannot be established experimentally because long term trials of major dietary changes are impractical. One way to overcome the problem is to produce a summary estimate of the size of the relation by combining data from different types of study using an underused method that we call teleoanalysis. This summary estimate can be used to determine the extent to which the disease can be prevented and thus the most effective means of prevention. We describe the basis of teleoanalysis, suggest a simple one-step approach, and validate the results with a worked example.
What is teleoanalysis?
Teleoanalysis can be defined as the synthesis of different categories of evidence to obtain a quantitative general summary of (a) the relation between a cause of a disease and the risk of the disease and (b) the extent to which the disease can be prevented. Teleoanalysis is different from meta-analysis because it relies on combining data from different classes of evidence rather than one type of study.
Randomised trials with disease end points are often not enough to determine dose-response relations; their results tend to be limited by factors such as dose, duration of treatment, and a limited age range of subjects. We also need data from observational epidemiological studies (particularly large cohort studies) and often knowledge of the mechanism of action. Short term trials using drugs or vitamins are also helpful because a drug can have a large specific effect that is not otherwise achievable.
It may also be necessary to quantify the individual effects that relate to separate steps in a causal pathway—that is, the effect of factor A on disease C is determined from the estimate of the effect of A on an intermediate factor B and the estimate of the effect of B on C, rather than by directly measuring the effect of A on C. The exercise is like putting together the pieces in a jigsaw puzzle.
The adverse effects of interventions always need to be considered, and including them in the analysis will give an assessment of the benefit to harm ratio. However, inclusion of adverse effects relies more on monitoring than on interpretive analysis, and we will not consider it further in this article.
In contrast to meta-analysis, which increases the precision of summary estimates of an effect within a category of study, teleoanalysis combines different categories of study to quantify the relation between a causative factor and the risk of disease. This is helpful in determining medical practice and public health policy. Put simply, meta-analysis is the analysis of many studies that have already been done; teleoanalysis provides the answer to questions that would be obtained from studies that have not been done and often, for ethical and financial reasons, could never be done.
Examples of teleoanalysis
Dietary cholesterol and ischaemic heart disease
An example of the value of teleoanalysis is in quantifying the relation between diet (A) and the risk of ischaemic heart disease (C). This relation can be fully recognised only through some of the important intermediate factors—for example, serum cholesterol concentration (B). A meta-analysis of randomised trials suggested that a low dietary fat intake had little effect on the risk of ischaemic heart disease.1 But the effect of a significant reduction in dietary fat can easily be underestimated, even when it is based on the results of randomised trials. The difficulty is that diet in the general population cannot readily be changed on a large enough scale for long enough to give interpretable results. Randomised trials answer the pragmatic question of the efficacy of the dietary advice offered but do not indicate the efficacy of some more effective means of reducing dietary fat—for example, dietary stanols, which lower cholesterol concentrations.
The problem can be overcome by using serum cholesterol concentration as an intermediate factor in the causal pathway. We can then examine the effect of lower dietary fat intake on serum cholesterol concentration (which can be done in small scale experimental studies) and the effect of serum cholesterol concentration on the risk of ischaemic heart disease (which can be done in epidemiological cohort studies and in trials of cholesterol lowering drugs).2–4 Data from several sources need to be reconciled quantitatively to assess the effect of dietary fat on disease.
Dietary folate and neural tube defects
Increasing maternal intake of folic acid prevents fetal neural tube defects. In the Medical Research Council vitamin study, women who took folic acid 4 mg/day around the time of conception had an 83% reduced chance of having a child with a neural tube defect.5 Other evidence supported the preventive effect of folic acid,6 7 but the dose-response relation remained unknown. This made it difficult to advise women planning a pregnancy on the appropriate dose and an illogical two dose policy emerged; despite the absence of adverse effects, women who had had an affected child were advised to take 5 mg a day but women in the general population were advised to take only 0.4 mg a day.8 The problem of dose was unresolved and the effect of fortifying flour with folic acid on the risk of a neural tube defect also could not be predicted. Too few satisfactory studies had examined different doses of folic acid supplementation to quantify its preventive effects.
By using teleoanalysis to integrate different sources of data and different classes of evidence, it was, however, possible to resolve the issue.6 Several studies have examined the effect of increasing folic acid intake (A) on serum folate concentrations (B), showing that an absolute increase in folic acid intake leads to an absolute increase in serum folate concentration. A cohort study established the dose-response relation between maternal serum folate (B) and the risk of having a neural tube defect pregnancy (C), showing that a proportional increase in serum folate concentration leads to a proportional decrease in risk of neural tube defects.9 Combining results from these sources in a two-stage model produced a simple quantitative summary that linked folic acid intake to the prevention of neural tube defects (figure).10
Teleoanalysis in one step
Teleoanalysis can also be done with a statistical method that integrates all the available data in one step, in effect performing both steps simultaneously. This model can be implemented in a statistical computer package WinBUGS (based on a bayesian conditional independence model), which is currently available free from the internet.11 The details of the statistical method are available on bmj.com.
Trials to determine the dose-response relation between risk factors and disease are often difficult or impossible to do
Quantification of such dose-response relations therefore relies on combining epidemiological data with data from other sources
Teleoanalysis is an underused method that combines different categories of data to produce a summary estimate of the size of the relation
The method is valuable in public health as it can estimate the extent towhich a disease can be prevented and thus help determine the most effective means of prevention
The table compares estimates of the effect of folic acid on neural tube defects obtained by the one-step method with those derived by the two-step approach.10 The two methods give nearly identical results. In effect, each validates the other and confirms a dose-response relation of considerable public health importance. The models show the extra benefit of using a 5 mg a day supplement in the prevention of neural tube defects instead of 0.4 mg per day, as is currently recommended for women in general. With a background serum folate concentration of 5 ng/ml this would yield an 84% (95% confidence interval 49% to 96%) reduction in the risk of neural tube defects compared with a 35% (14% to 54%) reduction with the currently recommended dose. The one-step method estimates confidence intervals directly but the two-step method requires a separate simulation.
Although the two-step approach is simpler to carry out with conventional software and is easily explained to non-statisticians, the one-step approach may be preferred because it automatically provides an indication of the precision of the estimates. Given the availability of the necessary software, there is no reason why teleoanalysis should not become standard. Use of the technique will formalise a method that has been used in isolated situations over the past decade and should help to encourage the use of intermediate markers in quantitative analyses. Teleoanalysis alone does not prove that a particular agent causes a disease, but once this has been established—for example, in a single dose drug trial—it is a useful method to quantify the effect and indicate the dose-response relation.
We thank Jeffrey Aronson for suggesting the term “teleoanalysis” (from “teleos,” which means complete or thorough in Greek), and Frank Speizer for his comments on the manuscript.
Details of the statistical method for the one-stage process are available on bmj.com
Contributors and sources The authors have worked for many years in aetiological epidemiology seeking to identify preventable causes of disease and quantifying these effects.
Competing interests None declared.