Year: 2014
Type: letter
Abstract: Pinsky questions the methodology in our article on prostate cancer mortality in areas with high and low prostate cancer incidence (1) and asks for “standard analyses appropriate for ecologic studies” by use of regression analysis of change in incidence in all counties as a continuous exposure. In our study, exposure was based on change in prostate cancer incidence in the population that is an imperfect proxy for PSA uptake. Pinsky suggests that exposure for PSA testing can be compared with exposure to a causative agent, eg, ultraviolet radiation, for which a higher and longer exposure (dose) will cause an increase in risk (response). However, in a population exposed to PSA screening, incidence will first increase drastically by detecting prevalent cases, and then fall in subsequent screening rounds (longer exposure); thus the analogy between a causative agent and PSA testing is inappropriate (2). The end-point in our study was prostate cancer mortality in the population in two groups of counties with highest and lowest exposure. However, we also demonstrated that mortality in the intermediate counties was in between the high and low groups (Figure 3) (1). Importantly, we used an incidence-based mortality analysis (3) based on deaths in the population among cases diagnosed during the study period only. This method has rarely been applied to prostate cancer previously, and it is important to keep in mind that this method concerns the population and not individual men. By use of this method, we avoided diluting risk estimates by including deaths in cases in the population that had been exposed (or not exposed) and diagnosed prior to the study period, as would have occurred with standard methodology. Stimulated by Pinsky, we performed additional analyses using Poisson models (Table 1). In analyses at county level, we found a statistically significant decrease in mortality both for time period and exposure by group similar to that reported in our article (1). In analyses with exposure as a continuous variable, higher exposure, both as a change in incidence and as a rank, was associated with lower prostate cancer mortality in the later time period (P < .001) and when both periods were analyzed. Because the relative risk estimates from these models with continuous exposures are highly dependent on the selected scale, the P values are the most informative results. Incidence based prostate cancer mortality with county as a categorical covariate * Analysis was made by use of Poisson log-linear regression models including both time periods (1990–1999 and 2000–2009) and all 24 counties as categorical covariates. Estimates of intercept and risk ratio for 23 counties not shown. CI = confidence interval. † The interaction terms between time period and county group (intermediate- and high-incidence counties) measured decrease in prostate cancer mortality related to exposure. ‡ County exposure rank measured decrease in prostate cancer mortality. County exposure was ranked from low (1) to high (24) during 2000–2009 and was set to one in 1990–1999. Incidence based prostate cancer mortality with county as a categorical covariate * Analysis was made by use of Poisson log-linear regression models including both time periods (1990–1999 and 2000–2009) and all 24 counties as categorical covariates. Estimates of intercept and risk ratio for 23 counties not shown. CI = confidence interval. † The interaction terms between time period and county group (intermediate- and high-incidence counties) measured decrease in prostate cancer mortality related to exposure. ‡ County exposure rank measured decrease in prostate cancer mortality. County exposure was ranked from low (1) to high (24) during 2000–2009 and was set to one in 1990–1999. In conclusion, the additional analyses with exposure measured on county level as a categorical and a continuous variable support the results in our article and suggest that prostate cancer mortality is lower in a population that is exposed to higher levels of PSA testing.
DOI: 10.1093/jnci/dju293
Authors: Pär Stattin, Sigrid Carlsson, Håkan Jonsson
Language: English
Subfield: Oncology
Field: Medicine
Domain: Health Sciences
Open Access status: bronze
OpenAlex ID: W578812416