br Determining cancer specific mortality and
Determining cancer-specific mortality and survival in cancer registry data In cancer research, defining a valid outcome parameter as an indicator of clinical relevance has its challenges. Overall survival (or its related outcome measure, all-cause mortality) is generally considered the gold standard for the demonstration of clinical benefit in cancer patients, as it is measured objectively, it is a direct clinical benefit, and it can be determined unequivocally . However, when the studied disease has relatively high survival rates, and/or the studied MIK665 (S-64315) has high mortality rates from other causes (i.e., all deaths unrelated to the studied disease or intervention), differences in survival between groups due to the intervention may be overshadowed by discrepancies in deaths due to causes unrelated to the disease or intervention. This limitation is particularly evident in non-randomized cohort and case-control studies, in which imbalances in baseline characteristics are likely to occur. Therefore, cancer-specific mortality may be a more informative outcome in such populations, as it directly measures the effect of the intervention on disease-specific outcome [8,9]. Cancer-specific mortality and survival are frequently used as primary outcomes in studies using cancer registry data, often without concomitant reporting of all-cause mortality or overall survival. Any outcome parameter other than overall survival has a level of subjectivity. Unlike overall survival, whose definition precludes the need to ponder on the cause, defining cancer-specific mortality and survival requires ascertaining the cancer in question or its consequential clinical outcomes as the cause of death. Such interpretation is dependent on the individuals evaluating the cause of death, as well the data available to the examiner, such as autopsy reports and a full medical history . Furthermore, assessing cancer-specific deaths in settings where interventions are unblinded may lead to biases, as previously reported in cancer screening data. Black et al. reported that in various large randomized trials of cancer screening, use of cancer-specific mortality led to paradoxical interpretation of the effects of the screening interventions . The authors ascribed the issue to misclassification of the cause of death, both as deaths from the disease had been attributed to other causes, and as deaths from other causes in screened patients had falsely been attributed to the disease detected upon screening. A well-known example of the latter is the transient increase in prostate cancer-related deaths in the US in the late 1980s and early 1990s after the introduction of screening for prostate-specific antigen. Optimally, causes of deaths would be determined by in-depth assessment by a blinded panel of experts who have access to the full medical history of the patient, and if conducted, autopsy reports. However, such evaluations are generally limited to large clinical trials as this requires a massive investment in resources. The costs of conducting such analyses in cancer registries would outweigh the gain in accuracy of the reported cause of death. Furthermore, since cancer registry data is available to the public, adaptive radiation is crucial that confidentiality of the patients included in the registry is protected. Therefore, most cancer registries, including the SEER program, and most observational studies in general, collect mortality statistics from death certificates [, , ]. On death certificates, medical examiners report the direct cause of death. They may also report underlying causes of death, as well as other significant conditions that did not directly contribute to the chain of events leading to death, if known. If two underlying causes of death added to the direct cause of death, the least direct cause of death is reported as another significant condition. Such evaluations are increasingly complicated when patients have substantial comorbidity, especially when the medical examiner did not treat the patient during his/her lifetime and when no autopsies are conducted. When comparing causes of death as registered on death certificates with those in autopsy reports, major disagreements occurred in 15 to 35% of cancer cases [15,16]. Studies report similar discrepancy rates when comparing causes of death between death certificates and blinded evaluation by an expert panel: for instance, a recent Norwegian report found that over- and underreporting of prostate cancer deaths on death certificates occurred in 33% and 19% of all prostate cancer patients, respectively [, , ].