Skip navigation
Link to Back issues listing | Back Issue Listing with content Index | Subject Index

Prediction from RCTs


Bandolier has long wondered why information from randomised trials is not better used (or even used at all) to inform us about diagnosis or prognosis. When talking to professionals, they will often comment that they want less information about what treatment to use, but much, much more information about which patient to use the treatment on. Diagnosis, or risk assessment, is the key, especially in busy primary care.

The curious thing is that the answer to their prayers probably lies in the very randomised trials that tell them about treatment. Think about it for a moment. Trials of treatment A in disease X have to recruit patients with disease X. Many are screened, and some chosen. So first of all we have a mass of information about those who 'have' the disease according to some contemporary definition, those who 'do not have' the disease, and those who, treated or untreated, have desirable or undesirable outcomes. Despite the complexity of any diagnosis, surely someone could analyse this mass of information and tell us whether there is some simple rule that could guide those of us not involved in clinical trials to decisions about who to treat?

Alas not, or at least only too rarely. The reasons are about as complex as the trials and statistics, but it comes down to the fact that in healthcare companies the Cerebrus-like creatures who guard the data can't see why diagnosis is so difficult. That's what it's like in industrial circles, but other areas are more open, and can show the rest how to do it [1].


The study [1] was based on an individual patient data analysis of antihypertensive intervention trials. This includes all major randomised trials of antihypertensive drugs for which individual patient information was available in 1995. There were 47,000 individuals. Of these, 3,001 had died in the average follow up of 5.2 years, and 1639 had died from cardiovascular causes.

In the eight trials, 16 baseline factors were common, and with prior plausibility as risk factors. These were:

Multivariate analysis showed that BMI, diastolic blood pressure, heart rate and uric acid were non-significant predictors. The final model used the 12 remaining factors, and these were grouped into convenient intervals of blood pressure or cholesterol.


The relationship between the total risk score from adding all the individual factors, and the risk of cardiovascular death over five years, is shown in Figure 1. For instance, a man (12 points) aged 54 (11 points) who did not smoke cigarettes (0 points), was 1.7 metres tall (3 points) with a systolic blood pressure of 130 (2 points), total cholesterol of 5.4 mmol/L (2 points) and creatinine of 80 mmol/L (1 point), and with no history of heart disease, stroke or diabetes, would have a total score of 31 points. That would translate into a five year risk of cardiovascular death of about 1%.

Figure 1: Cardiovascular mortality and risk score

The scoring system is well laid out in the original BMJ article and its Internet site. Best of all is that there is an Internet site ( ) available for users. It is easy to input information and it gives instant calculation of score and risk for an individual and the normal for age and sex.


The results are interesting, and most readers would want to know their own personal risk. If high, especially compared with normal for age and sex, something may be done about some of the risk factors, and the balance of risk reduction and inconvenience of doing anything weighed.

More important is the fact that this work has been done. It is an important exemplar of what professionals want, and how information can be used. Holders of information databases might want to take note, and get a dose of vision.


  1. SJ Pocock. A score for predicting risk of death from cardiovascular disease in adults with raised blood pressure, based on individual patient data from randomised controlled trials. BMJ 2001 323:75-81.
previous or next story in this issue