Most health risk assessments say if a person is at high, medium or low risk of either dying from or developing a given medical condition. Most also indicate what lifestyle factors contribute to this risk. What they do not say is the magnitude of each risk for an individual and how much that person’s risk will decrease if they change their lifestyle. For example, if one is at moderate risk of two diseases, say bowel cancer and heart disease, most people would be unaware that their risk of heart disease is still five times higher than their risk of bowel cancer.
In order to construct an electronic risk assessment tool for health and disease states, it is necessary to provide supporting research evidence and a method of encapsulating the best estimate of relative risk. For each medical condition, it is necessary to present credible estimates of risk, based on evidence from relevant, peer reviewed medical research. Important features of the risk assessment tool are:
- The tool gives numerical estimates of risk, rather than an imprecise statement such as "increased risk" or "reduced risk".
- The tool has the capability for interaction, allowing users to explore the impact on their personal risk of changing individual risk factors.
- The tool utilizes best available medical evidence
The aim of this project is to provide healthy people with a quantitative assessment of their personal risk of developing some important diseases and some of the factors that influence their risk. This is an ambitious task and we would not claim to have produced the definitive approach. Although we believe this is the most informative collection of disease prediction equations available at the present time they do have limitations. The ones we are aware of are outlined below.
Someone looking at their risk of lung cancer until the age of 50 should read this model as saying, "Assuming survival to age 50 the chance of developing lung cancer during that time would be (some predicted value)". This approach has the appeal that changing risk factors will have the expected impact on cumulative risk and the mathematics remains transparent. We chose the risk of developing a certain condition rather than the risk of dying from it because for many people the fear of living and dealing with a disabling disease is as frightening as dying from it.
This is different than lifetime risk calculations, which generally calculate the risk of dying from a given condition. Lifetime risk must take account of the fact that we all die of something in the end and calculating the relative contribution of common competing causes of death at various ages is difficult. Not only that, but the interpretation by users is complex. For example, a user of an interactive model predicting lifetime risk of lung cancer would see their individual risk of lung cancer fall with increasing cigarette consumption, because they would be dying of heart disease and chronic lung disease before they could get lung cancer.
How accurate are these percentages?
These models are good for illustrating the change in risk due to the presence or absence of single risk factors for prediction times of up to 5 years. They are likely to be reasonably good for 15 or 20 years and for combinations of several risk factors. For longer prediction times and varying more than, say, four risk factors the results should be regarded as illustrative rather than precise. The absolute level of risk for an individual may also be wide of the mark because the majority of overall risk remains unexplained in most research studies. This is why "confidence intervals" have not been included. That said these prediction equations do calculate the best estimate of risk that can be provided on the data given.
Is this useful in the end? We believe it is. We believe that putting some quantification on risk allows users to explore the possible impact on their health of altering what they do. We find this approach more informative than a bland statement of "high risk" that is often value laden or that a certain action will "cut down" a risk without any indication of by how much.
Is risk really reversible?
This is a difficult question to answer, but in many cases the answer seems to be, "yes". This is good news for people with high risks who are older. Intuition might tell you that you are constantly doing damage to your body that accumulates over time, and in many cases that may be true. An example of this is in skin cancer, where the earlier and more often you are badly burned in life, the higher your risk of skin cancer. Staying out of the sun when you are old cannot reverse this risk.
However, there is good evidence that for heart disease, for example, your risks can be significantly reduced no matter what your age. Cholesterol reduction by medications called "statins" reduces the risk of heart attack, angina or sudden death from heart problems by up to 30%, and this is entirely independent of age. Similarly, blood pressure reduction by drugs reduces the risk of stroke and heart disease by 25% - again entirely independent of age. Because in general it is older people who have the highest risks, they actually stand to benefit the most from treatment.
The risk for developing heart disease in tobacco users has been shown to decline to a level comparable with a person who has never smoked within 2-3 years of giving up. Furthermore, the risk of having a stroke is reversed after 5-10 years of stopping. Studies have also shown that life expectancy improves even in people who stop smoking later in life (i.e. at 65 years or older).
The reduction of risk that can be obtained from changing lifestyle habits such as diet, alcohol consumption and exercise is largely unknown. Therefore, the amount of risk reduction that can be expected from optimizing these habits needs to be viewed with caution. Certainly they should not take the place of blood pressure control, cholesterol control, and smoking cessation as goals.
How good is the evidence?
Our aim in searching for evidence was to identify up to ten high quality, relevant research studies for each topic. We used Medline to search using free text, MeSH terms and thesaurus search terms specific to each medical condition. To narrow the documents we used filters using "risk" and study design type; cohorts, case control, longitudinal, follow up. Searches were limited to studies published in English language and human studies. Although a comprehensive systematic review of the literature on each disease was not possible due to the scope of this project, we feel that the evidence used represents a reasonable cross-section of high-quality literature on the subjects in question.
What we have done is to seek out plausible values of relative risk to use in the prediction equations. We have used an approach that searches for high quality research studies and have then applied our judgment tempered by Austin Bradford Hill's criteria for causation when selecting which risks to use. Hill's criteria are: strength, consistency, specificity, temporality, biological gradient, plausibility, coherence, experimental evidence and analogy.
If this sometimes appears somewhat subjective then that is because at times it is a matter of judgment. The judgments have seldom altered the relative risk by more than a small amount. For each risk factor we had to choose a value to use in the model and have been faced at times with a range from which to choose. While a meta-analysis may provide the best point estimate, one is not always available and would be spurious to conduct on the sample of studies we have used for each condition. Given the level of uncertainty surrounding an individual's absolute personal risk we are comfortable with a comparatively lesser degree of uncertainty regarding a risk factor's relative risk.
What is the mathematical model that is used?
The actual mathematical and statistical models and risk coefficients that are used to determine risk are proprietary at this time, but have been validated by the authors and reviewers to be appropriate for use in this setting.
