9.2.4  Effect measures for ordinal outcomes and measurement scales

Ordinal outcome data arise when each participant is classified in a category and when the categories have a natural order. For example, a ‘trichotomous’ outcome with an ordering to the categories, such as the classification of disease severity into ‘mild’, ‘moderate’ or ‘severe’, is of ordinal type. As the number of categories increases, ordinal outcomes acquire properties similar to continuous outcomes, and probably will have been analysed as such in a clinical trial.

 

Measurement scales are one particular type of ordinal outcome frequently used to measure conditions that are difficult to quantify, such as behaviour, depression, and cognitive abilities. Measurement scales typically involve a series of questions or tasks, each of which is scored and the scores then summed to yield a total ‘score’. If the items are not considered of equal importance a weighted sum may be used.

 

It is important to know whether scales have been validated: that is, that they have been proven to measure the conditions that they claim to measure. When a scale is used to assess an outcome in a clinical trial, the cited reference to the scale should be studied in order to understand the objective, the target population and the assessment questionnaire. As investigators often adapt scales to suit their own purpose by adding, changing or dropping questions, review authors should check whether an original or adapted questionnaire is being used. This is particularly important when pooling outcomes for a meta-analysis. Clinical trials may appear to use the same rating scale, but closer examination may reveal differences that must be taken into account. It is possible that modifications to a scale were made in the light of the results of a study, in order to highlight components that appear to benefit from an experimental intervention.

 

Specialist methods are available for analysing ordinal outcome data that describe effects in terms of proportional odds ratios, but they are not available in RevMan, and become unwieldy (and unnecessary) when the number of categories is large. In practice longer ordinal scales are often analysed in meta-analyses as continuous data, whilst shorter ordinal scales are often made into dichotomous data by combining adjacent categories together. The latter is especially appropriate if an established, defensible cut-point is available. Inappropriate choice of a cut-point can induce bias, particularly if it is chosen to maximize the difference between two intervention arms in a clinical trial.

 

Where ordinal scales are summarized using methods for dichotomous data, one of the two sets of grouped categories is defined to be the event and intervention effects are described using risk ratios, odds ratios or risk differences (see Section 9.2.2). When ordinal scales are summarized using methods for continuous data, the intervention effect is expressed as a difference in means or standardized difference in means (see Section 9.2.3). Difficulties will be encountered if studies have summarized their results using medians (see Chapter 7, Section 7.7.3.5).

 

Unless individual patient data are available, the analyses reported by the investigators in the clinical trials typically determine the approach that is used in the meta-analysis.