Clinical prediction rules: friend or foe?

How can we use evidence-based medicine for diagnosis and prognosis?

One way is to use clinical prediction rules, which look at a patient’s symptoms, clinical signs, and sometimes diagnostic tests, – and then uses some maths – to try and quantify the probability that a patient has of having a particular disease or condition.

One of the most well-known examples is the Ottawa Ankle Rule, which combines a series of clinical features (being able to walk and tenderness over particular ankle and foot bones) to help clinicians determine if someone with an ankle injury is likely to have a fracture and will therefore need an X-ray.

There are many prediction rules out there for a whole host of conditions, such as for infection, assessing cardiovascular risk, anxiety and depression, to name but a few. And new ones are being developed and published all the time.

But what we wanted to know is whether a bunch of general practitioners (GPs) actually know anything about them, do they use them, and if so, do they find them useful?

To answer this question, we did a survey of 400 GPs across the UK, gave them a list of clinical prediction rules that we thought would be relevant and asked them if they knew of them and/or used them.  Just to add more evidence we also did a systematic review of guidelines to have a look if they mentioned or recommended relevant clinical prediction rules.

What we found is that for cardiovascular disease most GPs knew about and used clinical prediction rules, and most guidelines also recommended using them.  GPs mostly used them to guide therapy, but also to comply with guidelines.

Depression was another condition where most GPs reported using a clinical prediction rule, mostly to assess severity or, again, to comply with guidelines.

We were surprised to find out that although the Ottawa Ankle Rule to assess ankle injuries was only recommended by a few guidelines, it was nonetheless used by several of the GPs in our survey.  On the flip side, although several guidelines recommended using a clinical prediction rule in the diagnosis of breast cancer, almost none of the GPs we asked used one or had even heard of one.

Quite a few GPs told us they preferred using their clinical judgement rather than following a clinical prediction rule, and some even questioned what the evidence was behind these rules (good question!)… Well we haven’t looked at that systematically (yet!….I feel another study coming on…), but we do know that there are some that are well validated with plenty of evidence, but at the moment are not used.

The bottom line is there are many clinical prediction rules out there, but many of them are not known to clinicians and many don’t seem to be useful.  For those researchers interested in prediction rules, maybe there is a need to get together more and find out which clinical prediction rules would be most beneficial to clinicians and patients…and make sure we produce the necessary evidence to help implement them.  So what do you think? Do you know of any? Do you use them? Do you find them useful? Answers on a postcard…or better yet in the comment box at the bottom of this page.

Annette Pluddemann

About Annette Pluddemann

View more posts by Annette Pluddemann

One comment on “Clinical prediction rules: friend or foe?

  1. Dear Annette,

    Thank you for posting this on this interesting topic. I definitively believe that clinical prediction rules could be a possible solution to bridge the current ‘know-do gap’ and could contribute to integrating evidence-based medicine into daily clinical practice.

    In a research paper by Toll and colleagues in 2008, it was shown that the number of papers discussing prediction rules has more than doubled in recent years. Most publications, however, concern the development of new prediction rules, with few articles describing validation and almost none confirming their clinical impact. In addition, a large amount of prediction rules is scattered throughout the internet, often hosted on local servers that lack transparency and quality control.

    I therefore completely agree with your statement: ‘maybe there is a need to get together more and find out which clinical prediction rules would be most beneficial to clinicians and patients…and make sure we produce the necessary evidence to help implement them’.

    Clinicians and researchers should unite to solve the abovementioned problem. As a start, some of my colleagues and myself have been working on an online medical prediction platform (www.evidencio.com), with which clinicians and researchers can create new prediction rules themselves based on already published studies or based on their own research data. In addition, Evidencio facilitates external validation of existing models (including assessment of model discrimination and calibration) using anonymized institutional data. All models on Evidencio are reported in a transparent format, including information regarding the underlying study population, formulas used, predictor coefficients, model performance, and names of publishing authors/institutions.

    Each clinician and researcher with an interest in medical prediction rules is welcome to join or online research community. They can simple sign up for a free account on our platform and start contributing to the community.

Leave a Reply to Pleijhuis RG, MD, PhD Cancel reply

Your email address will not be published. Required fields are marked *