Targeting care at those at highest risk of an asthma attack is an attractive concept. Asthma attacks are at best unpleasant, at worst catastrophic and even fatal. Asthma attacks drive health care costs (largely via hospitalisation) and costs to society (via loss of work),1 and reducing hospitalisation is the key to reducing the overall costs of asthma care.2 The concept of good asthma control fundamentally includes the notion of an individual's likelihood of experiencing an attack.3

Risk prediction is growing in importance — both to the individual and at the public health level. Risk scores already exist to predict, for example, the risk of future cardiovascular events,4 development of diabetes,5 and lung cancer.6 UK general practices now routinely use risk-stratification tools to identify — and then target with preventive care — patients at highest risk of hospitalisation.

In this issue of the PCRJ, Hyland and colleagues7 report an observational study testing the intriguing hypothesis that a person's risk of a future asthma exacerbation is related to their past attendance in primary care for problems other than asthma. They assessed a range of variables in 166 people with asthma at an asthma clinic in a single UK general practice, and then gathered data on attendance for the following five years from their medical records. Data included medication adherence (both self-reported and assessed from prescriptions), severity (as medication step), and demographic factors. They found that non-asthma visits to primary care were a stronger predictor of future asthma exacerbations than either asthma severity or medication adherence. This relationship held not just for attendances for ENT problems (which might suggest respiratory dysfunction) but also for gastrointestinal and psychological problems. The authors propose that higher numbers of non-asthma visits are a marker of a ‘dysregulated’ or ‘dysfunctional’ patient. They speculate as to whether the relationship reflects a behavioural or biomedical, perhaps inflammation-mediated, mechanism. A third potential explanation is that higher numbers of non-asthma visits are a marker of a dysfunctional clinical system that does not adequately cope with multimorbidity8 — i.e. the coordination of care and consideration of multiple problems in making treatment decisions may be lacking. Data on multimorbidity in the study population would allow this idea to be explored.

Hyland and colleagues acknowledge the limitations of a study set in a single UK general practice.7 Clearly it would be worthwhile testing these relationships in other larger, and more demographically diverse, datasets. Although not part of their hypothesis, it would have been interesting to report the relationship between prior and future asthma exacerbations. Nonetheless, the authors are confident they have identified a potentially useful predictor of risk of future emergency consultation for asthma. They make the point that a narrow focus on adherence or severity may not fully capture this risk.

A number of approaches have been taken to predicting risk of future asthma exacerbations.9 Some have focused on monitoring fluctuations in peak expiratory flow rate,10 symptoms and questionnaire scores,11 or biomarkers of inflammation; others have focused on the frequency of past consultations.7 Some have tried to predict exacerbations themselves,10 others the risk of an attendance with an asthma exacerbation.7

Predicting the risk of attendance with an asthma exacerbation is likely to be more problematic than predicting the risk of harder disease outcomes like lung cancer, diabetes and myocardial infarction. Firstly, asthma is recognised as a condition with multiple, potentially distinct phenotypes.9 Each may vary in its likelihood of deterioration. Second, some attacks, even severe ones, do not lead to a consultation with a health care professional. As the American sociologist Irving Zola pointed out, help-seeking is a social process.12 Whether or not a person ultimately attends health care with an asthma attack will depend on the context and their interpretation of their symptoms,12 their abilities to self-care,13 and their past experience of consultations,14 as well as practical, economic and logistical aspects of accessing care.14 These are factors that are difficult to quantify, but may go some way to explaining the very large variations in unscheduled care for asthma seen across different ethnic groups.15

Just as with cardiovascular disease, ultimately the most useful approach to risk prediction for asthma is likely to come from analyses of very large primary care datasets.4 Predictor variables will need to capture a diversity of demographic factors including social deprivation and ethnicity, biological measures of disease such as lung function, aspects of management including medication and adherence, lifestyle — notably smoking status, and consultation behaviour including frequencies of attendance for asthma and non-asthma related care. Validation of a model's predictive abilities in a subsequent dataset will strengthen its credibility. Finally, the test of its usefulness should include comparing the impact of targeted care using the risk prediction tool with usual asthma care, probably using a design incorporating cluster randomisation of primary care practices. The study would need to be large enough to assess benefits for those targeted as well as potential poor outcomes for those whose care might be less effective because they are in a low risk group.

This highlights an essential aspect of an important debate: the attraction of risk prediction is that it enables targeting of those at highest risk of an exacerbation. However, it is clear from studies of asthma deaths16,17 that severe and fatal attacks can appear out of the blue, affecting patients whose risk scores might place them in a group which would mean they would be neglected if care was driven entirely by risk prediction. Interestingly, the approach taken in the Finnish Asthma Programme, which successfully reduced hospitalisation and health care costs, was largely untargeted, emphasising better care and anti-inflammatory treatment for all asthma patients.2

In summary, two challenges remain: first, we need to produce better risk prediction tools for people with asthma; second, we need to prove they lead to better care.