In the final part of this series, Christopher Ball explains how to apply evidence about prognosis to your patient
Many clinical decisions depend on doctors making predictions about the likely consequences of diseases on patients. Unfortunately, risk can be difficult to understand
--people and governments often behave illogically in response to perceived risk. Extremely rare conditions like variant Creutzfeldt-Jakob disease led to uproar in the media and to the government spending billions of pounds. Yet common causes of death among young people--for example, road crashes--are relatively ignored and underfunded.
Prognostic studies rarely help because they typically use complex statistical analysis to identify the relevant risk factors and then present multiple factors using numbers that do not seem to relate closely to patients' outcomes. Worse still, as we shall see,
these numbers are difficult to manipulate or customise.
Evidence can help with uncertainty
Integrating prognostic information into clinical care requires care and an understanding of your patient's ideas and concerns. Everyone likes certainty, and patients often have false expectations that doctors can divine the future. Helping your patient to understand that at best clinical prediction is rudimentary may destroy some of their faith in the medical profession and lead to increased anxiety but ultimately can improve shared decision making.
Doctors and medical students need to remember that prognostic data relate to large samples of patients, and not individuals. 60% of patients in a group may have had a stroke, but this does not mean an individual patient has a 60% chance of having one--he or she will either have one (100%) or not (0%). You should, therefore, take care to talk about the effect on groups: "In patients similar to you, 60% have a stroke," rather than make predictions about your patient: "You are likely to have a stroke."
Your perception of low or high risk may be different from your patient's. As a doctor, 20% risk seems low to me, but when I was faced with an outcome of 20% as a patient, it felt risky, and I was anxious. When discussing the future with patients, it is important to find out their expectations and how they feel about the likely outcomes.
Clinical scenario
Your team admitted a previously healthy 80 year old man with congestive cardiac failure from new onset atrial fibrillation two days ago. On today's ward round, your consultant tells him that he needs to start taking aspirin to reduce his risk of having a stroke.
After the ward round, he calls you over and asks how likely it is that he will have a stroke, since he has never had one before. You tell him you think his chance of a stroke is low, but you will check for him.
Clinical question -- In a patient with atrial fibrillation (patient) what clinical features (intervention) help predict a stroke (outcome)?
Resource selected--Evidence-based On-call
Search strategy--Atrial fibrillation stroke
Result--26 articles were found. The 10th structured abstract title gave the answer
Answer--A clinical prediction rule based on five clinical factors (congestive heart failure, older than 75, diabetes mellitus, hypertension, and a history of stroke or transient ischaemic attack) helped identify patients at increased risk1
Time to answer--22 seconds
Outcome measures--As you read through the structured abstract, you find table 1.
Table 1: From Evidence-based On-call
| Outcome |
Time to outcome |
No of patients/total No(%) |
95% CI |
NNF(95% CI) |
| First ischaemic stroke |
1.2 years |
94/1733 (5.4) |
4.4 to 6.5 |
18(15 to 23) |
The information in table 1 is easy to understand, but you may not be familiar with number needed to follow (NNF). This is the reciprocal of the proportion of patients with a given outcome. In this example, you would need to follow 18 patients with atrial fibrillation to see one of them have a first ischaemic stroke within 1.2 years. When the risk is small (<10%), patients and doctors can immediately grasp the size.
All measures of an outcome should be accompanied by a measurement at follow up. The chance of an event happening may change over time--for example, the risk of a recurrent stroke is less after the first year, so any number can only measure the risk of an outcome at a specific time. Prognosis curves--the risk of an outcome against time--provide more detailed information, but for most purposes percentages and numbers needed to treat are sufficient.
Measures of risk
We are all familiar with the important cardiovascular risk factors, but is there a way of working out how likely it is that a patient with any risk factor will go on to develop a stroke or myocardial infarction? If two risk factors are worse than one, by how much? Is there any way of customising the risk to individual patients?
Two common measures of risk are odds ratios and relative risks. They both quantify a difference in risk between two groups, but they differ. An odds ratio is a comparison of risk incurred by a control group with that incurred by an intervention group; if the odds in both groups are the same then the odds ratio will be 1. The relative risk compares the odds of two groups without a control. Hence a prognostic factor with an odds ratio of 3.0 does not necessarily produce the same increase in risk as one with a relative risk of 3.0.
Odds ratios and relative risks present a number of difficulties. Both provide only a relative measure of the increase in risk. Just as with relative risk reduction, unless you know the baseline risk of an outcome, it can be difficult to advise your patient on the actual percentage increase in risk due to any one risk factor.2 Unfortunately, most prognostic studies do not provide enough information to calculate this absolute increase in risk. Also, odds ratios and relative risk usually give an increase in risk due to a single factor. There is no simple way of combining multiple risk factors to generate one overall measure related to a number of risk factors.
Table 2: Clinical prediction rule - allocating points
| Prognostic factors |
1 |
| Congestive cardiac failure |
1 |
| hypertension |
1 |
| Aged 75 or older |
1 |
| Diabetes mellitus |
1 |
| History of stroke or transient ischaemic attack |
2 |
Fortunately, you can use two other methods to overcome this inability to customise prognosis using odds ratio or relative risk. The first is to use clinical audit. Studying local figures can help you determine how patients similar to your own have performed in the past at your local institution. Obviously this information needs to be kept up to date to reflect advances in clinical practice.
Table 3: Clinical prediction rule - chance of stroke at 1.2 years and numbe needed to follow
| Score |
Stroke at 1.2 years (95% CI) |
NNF (95% CI) |
| 6 |
40% (0.0% to 83%) |
3 (1 to ƒ)
|
| 5 |
9.2% (2.2% to 16%) |
11 (6 to 46) |
| 4 |
8.6% (4.9% to 12%) |
12 (8 to 20) |
| 3 |
7.4% (4.6% to 10%) |
13 (10 to 22) |
| 2 |
4.4% (2.6% to 6.2%) |
23 (16 to 38) |
| 1 |
3.7% (2.0% to 5.4%) |
27 (19 to 51) |
| 0 |
1.7% (0.0% to 4.0%) |
60 (25 to ƒ) |
The most effective alternatives are clinical prediction rules. Similar in nature to clinical diagnosis rules, these allow clinicians to combine a set of prognostic factors using a point-scoring system, and then rank patients for their risk of a clinical outcome. Unfortunately clinical prediction rules are rarely as good as clinical diagnosis rules; they are usually better at predicting patients at low risk for an outcome than patients at high risk. As with clinical diagnosis rules, you need to make sure that the clinical prediction rule you are using has been tested out and validated in multiple settings and is not just an experimental tool. Keep a look out for them, though--they can make your clinical work much faster and easier.
Back to the clinical scenario
You find a clinical prediction rule (tables 2 and 3). Your patient scores 2, which translates into a 1 in 23 chance of a stroke within the next year. You note that if he had a stroke or a transient ischaemic attack, the risk of another stroke would double to 1 in 12.
You explain that in similar patients the risk of a stroke is greater compared with other patients but is still low. If he went on to have even a minor stroke, he might be at greater risk of another. On reflection, he decides to take the aspirin and thanks you for your advice.
Further reading
Sackett DL, Straus S, Richardson WS, Rosenberg W, Haynes RB. Evidence-based medicine: how to practice and teach EBM. London: Churchill-Livingstone, 2000.
For more examples of assessing cardiovascular risk, go to:
