Methods: A demographically diverse sample of 1,461 Internet users read about a hypothetical disease and then were randomized to view either survival or mortality graphs that showed either 5 years of data or 15 years of treatment outcomes data for two possible treatments and a no treatment group. All risks were compounded annually, yielding exponential curve shapes. Participants identified the most effective treatment, provided ratings comparing the effectiveness of the two treatments, and answered comprehension questions.
Results: Responses on comprehension measures were mixed: viewers of mortality graphs were less able to identify which treatment was more effective (85% vs. 94%, p<0.001) but better able to correctly report individual data points in a series of comprehension questions. However, after screening out all respondents who could not correctly identify the more effective treatment, treatment effectiveness ratings varied significantly between respondents seeing the 5 year and 15 year survival graphs (M=5.63 vs. 6.73, t=-6.04, p < 0.0001), even though the yearly relative and absolute risk reduction was the same in both cases. When participants viewed the analogous 5 year and 15 year mortality graphs, however, this variation was significantly reduced (M=6.02 & 6.42, ANOVA interaction F=7.65, p=0.006). Qualitatively similar results are observed in full sample analyses.
Conclusions: The number of years of data provided in a survival curve can change beliefs about treatment effectiveness by itself. Presenting data in mortality graph format significantly reduces this unwanted effect, although special care will need to be taken to ensure that readers of mortality graphs recognize that the optimal treatment is shown by the lowest curve. All line-based risk graphics (whether framed in survival or mortality terms) should highlight duration information to facilitate improved comprehension of treatment effectiveness.