Prevention and Communication Research
Communicating risk probabilities

(February 4, 2003) Epidemiological research aims to find probabilities—to determine not only which treatments and behaviors affect risk but to what extent. These probabilities, however, are seldom communicated to the public and rarely understood, even among well-educated adults. In a recent lecture, Dr. Neil Weinstein, Professor of Human Ecology and Psychology at Rutgers University, described his efforts to develop better formats for conveying risk probability information to the public. The overarching goal of his research is to improve the ability of patients to understand and participate in health decisions—and to improve their satisfaction with the decision-making process.

Risk probability information can be presented in one of three basic formats: numerical, graphical, or, more ambiguously, with verbal labels such as "small risk." Within each format, there is a range of options for presenting risk information. For example, the same probability can be expressed as a percentage (12.5%), as 1 in N (1 in 8), or as N in base (125 in 1000). The various options for both graphic and numerical formats have been compared in a number of studies. However, these studies have several limitations, and so researchers have not been able to determine which format works best. For example, most studies only examine whether people can correctly identify the larger of two risks but do not examine the other ways that people use probabilities in their decision-making. In addition, the study populations are often small and homogeneous. As a result, the studies do not have the power to detect significant differences between formats or to determine whether different formats work better with different populations.

Dr. Weinstein has begun to address some of these limitations with a series of experiments he is conducting using the online risk assessment Your Cancer Risk. Study participants are volunteers who visit Your Cancer Risk and agree to answer questions about risk interpretation. This allows for a very large study population that has a demonstrated interest in health and risk and that is diverse in education, gender, ethnicity, and age. In a typical experiment, participants are given numerical information about their hypothetical risk of a particular cancer and told how a new drug could decrease this risk but increase the risk of another type of cancer. They are then asked to determine whether the drug will increase or decrease their total risk of cancer. Answering the question correctly requires them to successfully perform a mathematical operation, such as addition or multiplication. What is notable about these experiments is that they are testing a range of mathematical operations, numeric formats, and risk levels to determine which presentation of risk works best for which people.

Several key findings have emerged from these experiments. In terms of mathematical operations, people are most proficient at comparing two risks and indicating which one is larger. They are least proficient at adding two risks and also have moderate difficulty interpreting a tradeoff in risks (e.g., when a drug cuts one risk in half but doubles another) or a sequence of risks (e.g., when there’s a probability of a side effect to consider, as well as a probability that the side effect is serious). In terms of numerical format, people interpret risk correctly most often when it is presented as a percentage or as N in base rather than as 1 in N. Even though education level influences overall proficiency at interpreting risk, it doesn’t seem to affect which numerical format works best within each mathematical operation. For example, at all education levels, people interpreting a sequence of risks do best with percentages, and people interpreting a tradeoff do best with N in base.

Overall, Dr. Weinstein’s research suggests that with specific types of information, people can use probabilities to make important health decisions. Next steps in his research will involve determining how particular types of graphics might improve risk interpretation and how risk communication might be improved and/or implemented in the clinical setting.

written by Catherine Tomeo Ryan