The following example (from the NHS Centre for Reviews and Dissemination) will help make sense of the information presented in an odds-ratio diagram. The review investigates how stroke patients fare when cared for in a specialist inpatient care facility (or “stroke unit” fulfilling the definition of “a multidisciplinary team with a special interest in stroke”) compared with those cared for in a general medical ward. Select the review and go to the summary of analyses section This screen shows the odds ratios for a number of meta analyses for different combinations of comparisons and outcomes, all within the Stroke Units systematic review.

The title of the figure defines the comparison or intervention under investigation which in this case is specialist inpatient care versus general wards. In other words this is a comparison of the care provided to stroke patients in specialist inpatient facilities with that in general medical wards. The outcome listed below is death at final review. This means that the systematic review or meta analysis is comparing the number of people with a stroke who died when cared for in a specialist inpatient facility with those who died in a general medical ward.

Similar trials within the above review have been grouped together according to particular characteristics, in this case the type of specialist inpatient care (setting) provided, in order to perform several different meta analyses. The first type of specialist inpatient care is a rehabilitation ward or team which is in the top of the diagram and the second is a stroke ward or team which is in the bottom half of the diagram. There were 6 trials which looked at a rehabilitation ward or team as the specialist inpatient care and a further 6 trials which looked at a stroke ward or team as the specialist inpatient care. The trial identifier and some figures for each trial are listed under those sub headings. The odds ratio result for each trial is shown by a blue box and for each meta analysis by a black diamond. The bottom diamond shows the result for the total meta analysis for all 12 trials that measured this outcome and comparison.

Some further characteristics of this odds-ratio diagram are shown in the figure below. Looking at the first trial labelled “Birmingham 1972” note that the odds-ratio is greater than 1 (blue box to the right of the line of no effect). This means that stroke patients in the treatment arm of the trial (specialist inpatient care setting) are more likely to die than those in the control arm (general ward). This is because an odds ratio greater than 1 (to the right of the line of no effect) means that the outcome under investigation is more likely to occur in the treatment group than in the control group. In the case of a bad/negative outcome such as death this result means that this trial is showing a harmful effect. 

The above does not indicate how certain we are of the result of the trial. This is indicated by the horizontal line through the blue box for the individual trial under investigation. The horizontal line represents the confidence interval for the result derived from the trial. The confidence interval is the range in which we are confident that the “real” result lies when the result of the trial is extrapolated to the whole of the population which was sampled. The diagram illustrates 95% confidence intervals which means that the horizontal line encompasses the range of results in which we are 95% confident  that the real result lays. Or in other words in theory we would expect the results of 95 out of 100 trials to lie somewhere along this line.

The vertical solid black line shown above represents an odds ratio of 1. An odds-ratio of 1 means that the people in the treatment group would have been just as likely to experience the stated outcome as those in the control group and so the treatment has no effect. If a confidence interval crosses the line of no effect then we cannot be sure that the real result of the trial when extrapolated to the population sampled for the study lays the same side of the line of no effect as the trial result. In this situation we would describe the result as of uncertain benefit since the real result could be positive or could be negative. (Note that in the Birmingham 1972 study the blue box is very small and the horizontal line is very large.) This is a visual representation of a trial that is very small and therefore has a result with a relatively high degree of uncertainty; the treatment group was only 29 patients and the control group only 23. These numbers are far too small to expect a conclusive or clear result for the size of effect being measured, hence a large confidence interval.

Each of the trials in this meta analysis is illustrated below the Birmingham 1972 trial and a blue box or diamond is shown for each result. Below the blue boxes (under the rehabilitation ward stroke team sub heading) is a large diamond which is a representation of the odds ratio and confidence interval for the meta analysis of the 6 trials above it. That is the patients involved in the 6 trials were combined to give  a treatment arm of 271 patients in which 50 died and a control arm of 228 patients in which 50 also died. This combined result gave an odds ratio of 0.81 with 95% confidence intervals from 0.49 to 1.33. The results are given in figures further along to the right of the display. The tables show the odds-ratio and confidence interval figures. The odds-ratio implies that the result is beneficial, i.e. less patients dying in the treatment group (about 0.81 times as many or about a 19% reduction) compared with the control group, but the result is uncertain because the confidence intervals cross the line of no effect. 

In the bottom half of the screen is a similar meta analysis for trials where the specialist inpatient care was in the form of a stroke ward or team and as for the rehabilitation ward or team each of the individual trials is shown and its odds ratio represented by a blue box and the meta analysis result for those individual trials shown by the second black diamond in the diagram. In the Metaview screen there is a third diamond, which is the odds ratio for the meta analysis of all of the trials above (the 6 trials where the specialist inpatient care was the rehabilitation ward or team plus the 6 trials where the stroke ward or team was the specialist inpatient care setting). This provided a combined treatment arm of 947 patients in which 228 died and a combined control arm of 979 patients in which 297 died. The odds ratio overall for the combined result is 0.77 with 95% confidence intervals ranging from 0.62 to 0.95.

Overall this meta analysis shows that specialist inpatient care for stroke patients results in less deaths at final review than general wards (about 0.77 times as many or about a 23% reduction) and the result, with 95% confidence intervals, is conclusive because the confidence intervals do not cross the line of no effect.