How important is the ‘minimal clinically important change’?

Patient-based outcome measures have been developed to measure the health status of patients suffering from many conditions found in musculoskeletal medicine. Many types have been developed. Generic measures can be used across a broad spectrum of illness and can compare the health of an affected group with that in the general population.¹ Condition-specific instruments on the other hand measure the effect of a single condition on health. Compared to generic measures they have a narrower focus and are more sensitive to small but clinically significant changes in health status over short periods. Examples include the Roland–Morris questionnaire² for low back pain and Western Ontario McMaster universities arthritis index (WOMAC) for osteoarthritis.³

In order to be useful these instruments need to be reliable, valid, and responsive. Reliability is the extent to which they are free from random error and is measured in terms of internal consistency and reproducibility. Validity is the extent to which they measure what they intend to measure. Responsiveness is the ability to detect important changes over time, which is particularly important in randomized controlled trials (RCTs). They should also be precise, feasible to administer, acceptable to patients, appropriate, and easy to interpret.

Interpretability is difficult because when an outcome measure improves by, say, five points it is not immediately apparent what this means. One method is to divide the change by its standard deviation to give a standardized difference. The following benchmarks have been proposed to describe the relative size of this standardized change: 0.2 is considered small, 0.5 medium, and 0.8 or greater large.⁴

Another approach is to identify a clinically important difference that patients perceive as beneficial. Several methods have been proposed to estimate this, two of which are presented later in this edition of this journal.⁵ This clinically important difference is an important part of sample size calculations for RCTs, which are calculated on the basis that there is sufficient statistical power to test the hypothesis that a clinically important difference in the primary outcome is achieved.⁶

Many clinicians have taken this concept a step further when interpreting the results of RCTs by stating that this is the minimal clinically important difference that should be considered worthwhile.⁷ For the Roland–Morris questionnaire it has been suggested that this is five points.⁸ As the standard deviation of this instrument is 4 then this effect size is large with a standardized difference of 1.25, and would mean that RCTs of spinal manipulation^9,10 have an effect size for manipulation that is less than the threshold for what is clinically worthwhile.

Is it justified to have a single estimate of a minimal clinically important change for each outcome measure? The estimates will vary according to which method of calculation is used and also the context in which they were measured. First of all, this minimal clinically important difference varies depending upon whether one considers groups or individuals. Secondly, clinical experience suggests that there is individual variation between patients as to what degree of clinical improvement is considered acceptable. Some patients would accept any level of improvement, while others would only accept a complete cure. Thirdly, health economic analysis does not specify a minimal clinically important difference. Indeed, it would be difficult to do so as the main result of a health economic analysis is a ratio of costs compared with consequences. In this ratio, the cost of an intervention is just as important as the size of its clinical effect. Thus, an inexpensive treatment with a small effect has the same ratio as an expensive treatment with a large effect. Finally, the size of effect that is important depends upon the invasiveness of the intervention. To be acceptable an invasive procedure such as surgery would need to have a large effect, but smaller effects for non-invasive interventions such as educational booklets may still be acceptable. For example, osteoarthritis guidelines¹¹ recommend that paracetamol should be used first line for analgesia even though the effect size is small with a standardized difference of only 0.2.¹²

In conclusion, clinically important differences are useful for calculating sample sizes for trials and for interpreting the size of effect of interventions, but specifying a minimum worthwhile change is controversial and dependent upon the context in which it is measured.