Could a clinical prediction model help to reduce asthma mis-diagnosis?

Luke Daines speaking at the AUKCAR Annual Scientific Meeting, with npj Primary Care Respiratory Medicine logo inset
9 May 2019

To coincide with the publication of their systematic review in npj Primary Care Respiratory Medicine today, in the blog below Luke Daines provides more information for those less familiar with clinical prediction models!

Asthma (mis)diagnosis

Making a diagnosis of asthma can be difficult. One recent study reported that asthma was wrongly diagnosed in up to a third of cases in the community.1 Mis-diagnosis of asthma can lead to untreated symptoms, incorrect treatment and risk of asthma attack. There are several reasons making asthma difficult to diagnose. Asthma has different underlying causes, and symptoms vary between people and over time. Tests are helpful but can’t correctly identify asthma 100% of the time. Consequently, experts disagree about how best to diagnose asthma – which symptoms and tests are most useful, and in what combination?

What about a data-driven solution?

In this era of digital healthcare, data-driven solutions such as clinical prediction models, are increasingly used to support healthcare decisions. A clinical prediction model is a data-driven algorithm that combines information, such as elements from a clinical history, physical examination, test results or response to treatment, to estimate the likelihood of a disease being present or happening in the future.

How could a clinical prediction model help asthma diagnosis?

As information is gathered about a person (clinical history, physical examination and results from tests) a clinical prediction model can provide a likelihood that asthma is present, and help doctors and nurses decide what to do next. Can a diagnosis be confirmed? Is more information needed? Is asthma unlikely?

What was the point of this study?

This study reviewed clinical prediction models for the diagnosis of asthma. We looked for clinical prediction models for use in primary care as most asthma diagnoses occur in non-specialist settings. It was a systematic review, meaning that searching, checking and weighing up the quality of each article was done to particular standards. For example, two people checked all 13798 records found by our searches to make sure no important articles were missed.

What did the study find?

There were seven clinical prediction models for asthma diagnosis in primary care. Unfortunately, each of the clinical prediction models had one or more flaws in the way they were developed or tested. This means that if the clinical prediction models were used in their current form, there is a high chance that asthma diagnosis would be over or under estimated which could mislead doctors and nurses. We were also surprised that most clinical prediction models were developed using only information from a clinical history. So, with all things considered, none of the clinical prediction models can be recommended for use in everyday healthcare.

What’s next?

The systematic review taught us some important lessons which will help as we develop a new clinical prediction model for asthma diagnosis in primary care. Firstly, to develop an accurate clinical prediction model its important to follow particular methods, as laid out in the TRIPOD guidelines.2 Secondly, to be relevant to everyday healthcare we will use as wide a range of information as possible, including clinical history, physical examination, and the results from tests. Finally, clinical prediction models only work if they are used! So, we are actively seeking the views of doctors, nurses and people with asthma to help us.

If you’d like any more information about the project, please contact me:
Email: luke.daines@ed.ac.uk | Twitter: @ljdaines

Read the article

The published article is available from npj Primary Care Respiratory Medicine

Cite as

Daines, L et al., Systematic review of clinical prediction models to support the diagnosis of asthma in primary care, npj Primary Care Respiratory Medicine. 29:19 (2019). doi.org/10.1038/s41533-019-0132-z

References

1 Aaron, SD et al., JAMA. 317(3), 269-279 (2017) doi:10.1001/jama.2016.19627

2 Collins, GS et al., BMC Med. 13(1), 1 (2015) doi.org/10.1186/s12916-014-0241-z

© 2015 AUKCAR