br Improved reporting of cancer
Improved reporting of cancer-specific mortality and survival using cancer registry data When studying cancer registry data, the definition of cancer-specific mortality deserves consideration. The SEER program developed and validated an algorithm to optimize the use of death certificates to report cancer-specific deaths. Using this algorithm, cancer-specific deaths are not limited to patients whose death certificates report the cancer to be part of the causal chain leading to death, but to any cause of death that is likely related to the particular cancer, or a consequence of the cancer diagnosis. For instance, when a patient is diagnosed with endometrial cancer as a primary tumor, all deaths related to the uterus and ovaries are considered deaths related to endometrial cancer. This method is not perfect: the NCI acknowledges that mislabeling may occur, for example when a Nutlin-3 is not attributed to its primary tumor. Nevertheless, this algorithm will make ascertainment more conservative by improving the chances that cancer-related deaths will not be missed [14,20]. On the other hand, if researchers aim to prevent misclassification of any deaths unrelated to the studied cancer as cancer-related deaths, a stricter definition may be chosen. In that case, the value of other-cause mortality is limited, as the strict definition leads to misclassification of cancer-related deaths as other-cause mortality [9,21]. Thus, researchers should carefully consider misclassification due to their definition of cancer-related deaths, which may lead to under- and/or over-reporting of cancer-related deaths, depending on the definition chosen. The definition must be reported in the methods, as well as the rationale for any deviation from the SEER-definition of cancer-specific mortality. Potential misclassification bias may occur when using cancer registry data. In comparative studies, non-differential misclassification will generally result in bias towards the null hypothesis. Importantly, differential misclassification may bias the results in either direction, depending on whether there is relative over- or under-reporting in one group. For this reason, standard reporting of cancer-specific mortality or survival should be done in parallel with all-cause mortality or overall survival, respectively. While differences in the magnitude of the effect in these two types of outcome measures are to be expected because of the dilution effect that comes from using all-cause mortality or overall survival , this approach provides the opportunity of verifying whether the directionality of the effects using these two outcomes is coherent with each other and with the underlying hypotheses. A valid approach would be to report cancer-specific survival or mortality using different definitions of cancer-related deaths, akin to a sensitivity analysis, thus examining multiple scenarios for consistency of findings with underlying hypotheses. The most conservative definition would only include all deaths that have been directly attributed to the cancer being studied. A more liberal definition would be the SEER definition of cancer-specific mortality. Furthermore, opinions from clinicians and researchers specialized in the cancer of interest could be used to evaluate which codes in the International Statistical Classification of Diseases and Related Health Problems are definitely, plausibly, and/or possibly related to the studied cancer. Such definitions would be cancer- and case-specific: for example, bone metastases commonly occur in breast and prostate cancer, but are relatively uncommon in skin cancer patients; deaths related to bone disorders are therefore more likely to be related to the primary tumor in breast or prostate cancer patients than in skin cancer patients. Similarly, suicide is more likely cancer-related in a patient who just received a cancer diagnosis than in a patient who already used antidepressants prior to the cancer diagnosis and/or has been living with low-grade cancer for years without evidence of recurrence . Finally, when possible, one should consider the reporting of relative survival in addition to cancer-specific mortality/survival. In recent years, use of this measure has decreased relatively to cancer-specific mortality/survival in studies using cancer registries such as SEER [24,25]. In relative survival, the observed survival rate is corrected on the basis of the expected survival rate in a comparable population without the disease. This measure has its own limitations, for example as non-matching life tables may introduce bias, particularly when comparing subgroups [, , ]. Nevertheless, its advantage is that it does not require the subjective interpretation of death causes. Therefore, concomitant reporting of cancer-specific survival and relative survival may complement each other.