A great deal of General Practice is essentially about managing risk. Every time a patient walks through your consulting door you are basically thinking “what are the chances of this patient coming to harm from what they are about to tell me?” Some patients will fall into a category of “high risk” needing immediate or quick treatment. Others may have a degree of risk that necessitates either further investigations or monitoring. Or there may be low risk cases which can be assessed, reassured or treated. This is all supposed to happen within 10 minutes.

But how far should we go to convey that risk to our patients and what is the best way to do it? A recent scenario at my practice made me pause for thought. The cardiologists had advised us to have a discussion with a patient on the merits and risks of aspirin versus warfarin for their atrial fibrillation.

How could that conversation go?

“Right Mr X, the cardiologists have written to me and asked me to help you decide between warfarin or aspirin. They mention in the letter that the risk of you now having a stroke is about 6% per year, however aspirin reduces that risk by about 25% but with warfarin there is about a 45% risk reduction. However, the number needed to treat with warfarin is 37, but bear in mind that warfarin increases the annual absolute risk of major haemorrhage by 2%, so it’s up to you, which one would you prefer? ”

“umm…I’m sorry Doctor, I didn’t understand all of that”

“No neither did I”

So how should we explain risk to patients? The 2002 BMJ clinical review “Explaining risks: turning numerical data into meaningful pictures” by Edwards and colleagues is certainly worth a read. More recently, a study in the Annals of Family Medicine also tried to answer the question. The group surveyed 934 consecutive patients drawn from family practitioners’ waiting rooms in Auckland, New Zealand. Patients were asked to rate how much various modes of communicating the benefits of therapy, to their 5-year CVD risk score, would encourage them to take medication daily. The modes offered to them included: relative risk, absolute risk, odds, number needed to treat, and natural frequencies. The same information was presented in 2 pictorial forms (bar graphs and 10 × 10 people charts). Most patients (61.8%) preferred a doctor to give an opinion than to explain using either numbers or pictures. More than half also preferred a pictorial presentation to numbers; and of the numerical presentations patients found relative risk reduction most encouraging, with absolute risk reduction rated second overall and numbers needed to treat (NNT) the least likely to be persuasive to take their medication.

So should this mean to our practice? Remember EBM is the integration of the best clinical practice, personal expertise and individual patient preference. The latter component is dependent upon the patient fully understanding the risks and benefits of treatment so that a shared management plan can be reached. Having an idea of what those risks mean ourselves is the first step but finding the best way to convey it to our individual patients in as simplest way as possible is perhaps the bigger challenge.

As well as diagnosing disease, it may be important to predict which patients need more investigation or more treatment. The term, “clinical prediction rule”, is often used to describe the best combination of medical signs, symptoms, and other findings in predicting the probability of a specific disease or outcome. Therefore, they can be used in diagnosis as in the last article, or they can estimate the risk of a specific outcome, and therefore help in decision-making.

To form prediction rules, researchers look at a group of patients suspected of having a specific disease or outcome. They then compare the value of clinical findings available to the clinician versus the results of more intensive testing or the results of delayed clinical follow up.

Previous studies have shown that giving clinicians information in the form of risk scores in isolation may not influence clinical practice. GPs routinely underuse risk scores and have suggested that computerized risk prediction for multiple diseases simultaneously and the integration of lifestyle recommendations may improve their uptake.

Clinical prediction rules are most effective when implemented as part of a pathway within hospital policies and guidelines, as shown in the case of pneumonia and the more rigorously the rules are implemented, the more successful they are at improving outcomes.

The CHADS(2) score is an example of a clinical risk score to help in deciding whether to start warfarin in patients with atrial fibrillation to prevent stroke. Patients are classified by their clinical history, getting 1 point each for Congestive cardiac failure, Hypertension, Age>75 years, Diabetes and 2 points for Stroke. Patients with scores of 2 or more should get warfarin, as long as they have no contraindications. A recent improvement on this score, the CHA(2)DS(2)-VASc has added 3 additional factors (history of vascular disase, sex and age 65-74) to improve the prediction of risk so that only people who are really at negligible risk of stroke do not get warfarin.

Other examples of risk scores used in this way are Wells scores for deep vein thrombosis and pulmonary embolism and CURB-65 in pneumonia.

In the first article in this series we looked at the dimension of risk. Expression of risk in terms of an unwanted outcome or event can be described with descriptions or distinctions based on both its quality or on its quantity. The probability can be described in qualitative terms such as rare or infrequent or expressed quantitatively such as 1 in 1000. What is important to acknowledge that patients differ in what they like.

For example, in the context of a study on an invasive diagnostic procedure, 467 patients varied in terms of the way the risk was told to them: 43 % preferred qualitative (verbal expressions) disclosure of information vs. 36% preferring quantitative. The exact number varies form study to study, but roughly it is about half and half for each type of disclosure. In terms of influenza vaccines for children, slightly more parents preferred quantitative information (60%). It is often interesting how we come to decisions without knowing the basic quantitative risk. For instance, ask yourself the question: What is the risk of death from swine flu compared to seasonal flu? Qualitative descriptions are often foremost in our minds and very few will be able to give quantitative information. To be honest, I think I’ve shot myself in the foot as I don’t know the answer to this question myself. But, my premise is correct, as words like rare and infrequent are what come to my mind.

The most useful strategy identified in a systematic review of effective formats for communicating probabilistic information, including 15 randomized controlled trials found for both written and verbal information, patients have a more accurate perception of risk if probabilistic information is presented as numbers rather than words. That means that it is all right to say words like rare or infrequent but without a quantitative number such as 1 in a million people find it difficult to correctly perceive the actual probability.

How to present qualitative information

Qualitative descriptions are appealing because they use common words that seem to be generally well understood. However, qualitative words, of which there are many, have no generally accepted anchoring at specific quantitative levels of frequency, despite efforts to promote such anchoring.

Data about actions sometimes bear little relation to the statistics. Saxe found no connection between the probabilities provided to couples during genetic counselling and the decisions they made. The use of verbal categories with scales of risk, such as very unlikely, was easier to use and represented better their true feelings.

Proposals have been made to standardise the language of risk with standardised terms for specified frequencies ("high" for risks 1 in less than 100 and "moderate" for between 1 in 100 and 1 in 1000). However, it’s all in the interpretation and patients would probably not understand such standardised terms consistently.

There are a number of simple recommendations worth incorporating based on this article:
In describing risk choose either:

1) numeric probabilities, or
2) numeric and qualitative information together

Qualitative descriptions should use common words that seem to be generally understood.

Next time you are met with a risk, whether presenting or reading about another scare story in the media, you might want to consider whether some of these issues outlined have been met. They probably won’t have.