In a recent issue of Cancer, Renshaw et al. reported on the greatest dimension of prostate carcinoma in prostatectomy specimens as a potential predictor of prostate specific antigen (PSA) failure.1 Significance of the research question that the authors addressed is indisputable. Unfortunately, some of the main results were presented in a manner that was not quite straightforward. The authors used univariate and multivariate analyses to evaluate the ability of each of the measured factors (independent variables) to predict time to postoperative PSA failure (dependent variable). Because the time to PSA failure (continuous variable) rather than whether each patient experienced PSA failure in a fixed time period (binary variable) was used, I assume proportional hazards models were used. Some of the independent variables were included in statistical models as continuous variables (e.g., greatest tumor dimension in cm) whereas others were included as categorical variables (e.g., Gleason score > 7 vs. Gleason score ≤ 7). Table 2 of the article by Renshaw et al. shows the “risk ratio” estimate and associated P value for each of the seven independent variables based on univariate analyses. What is confusing are the reported “risk ratio” estimates for continuous variables. For example, the risk ratio estimate for “(tumor) area” was 1.007 (P value = 0.0011). The reader who is less literate in biostatistics may wonder why the risk ratio so close to unity could give such a highly significant P value, especially given a small sample size of the study. The reader with more background in biostatistics would guess that the authors must have calculated the increase in risk of PSA failure for a unit increase in the independent variable being analyzed. For a continuous independent variable, the risk ratio (strictly speaking, hazard ratio if a proportional hazards model was used, but this does not affect my argument here) between two individuals who differ by the quantity Δ on the i-th independent variable (xi) and are the same for all other independent variables is represented by the quantity exp (βiΔ), in which βi is a regression coefficient for the i-th independent variable from the proportional hazards model.2 Alternatively, the hazard ratio also can be interpreted as the instantaneous relative risk of an event per unit time for an individual with risk factor level xi + Δ compared with an individual with risk factor level xi, given that both individuals have survived to time t. Therefore, the interpretation of the above cited result will be that the risk of PSA failure increased by 0.7% with an increase in “(tumor) area” by 1 cm2. Accordingly, the risk ratio of the patient with the largest tumor area (3.7 cm2) compared with the patient with the smallest tumor area (0.06 cm2) within the study population is estimated as exp[ln(1.007) × (3.7–0.06)] = 1.03, in which βi is calculated as the natural logarithm of the reported “risk ratio.” This means that the risk of PSA failure was 3% higher for the patient with the largest tumor area when compared with the patient with the smallest tumor area. I performed similar calculations for the other continuous variables (Table 1). I used maximum and minimum values for each variable for the sake of making my point, but alternative reference values could be chosen. I believe this format of data presentation, with clear specification of comparisons being made, would facilitate the reader's understanding and critical reading. P values are useful in sifting data for associations that are more likely to be real or worth pursuing. However, they do not communicate the magnitude of the association. Presenting the results of statistical analyses in a reader-friendly manner is essential to putting them in a context to which the reader can relate. Data from clinical investigation, such as that by Renshaw et al., would be more appreciated if the reader could interpret them without having to convert the results into a different format to evaluate their implications. Clearly, my critique does not concern the article by Renshaw et al. alone. I happened to read the article because of my interest in its subject matter, and wondered whether busy urologists would have time to do the calculation shown in Table 1 here, even if they knew how. Given the emphasis of Cancer on reporting clinical and epidemiologic data, for which proportional hazards models and logistic regression models frequently are used, I hope the authors and editors try to leave as little of the burden of making sense of data as possible with the reader. Atsuko Shibata M.D., Ph.D.*, * Department of Health, Research and Policy, Stanford University, School of Medicine, Stanford, California