Determining cause of death in prostate cancer: are death certificates valid?

Accurate assessment of cause of death (COD) is important for determining cause-specific survival in cancer research. It is possible to reliably ascertain COD by meticulous review of inpatient and outpatient medical records with the use of predetermined clinical algorithms (1). Unfortunately, this method, although useful for small retrospective studies, is impractical for large datasets and national tumor registries that are commonly used for cancer research. When these large databases are used, COD is assigned with a standardized decision algorithm that uses International Classification of Diseases, ninth revision (2), codes to assign both immediate and underlying COD (3). However, this methodology is unreliable (4,5), particularly when patients are older or have considerable comorbidity, as is the case in prostate cancer (6,7). In a study of mortality trends, Grulich et al. (8) estimated that inaccuracies in death certification and coding accounted for up to 46% of the noted increase in prostate cancer mortality seen in England and Wales from 1970 through 1990. By contrast, in men with prostate cancer identified through the Connecticut Tumor Registry, Albertsen et al. (9) found a high level of agreement between the underlying COD, determined by a review of the medical records, and death certificate data. It is important that prostate cancer-related mortality ascertained by death certificate be reliable because studies that use large datasets, such as the Surveillance, Epidemiology, and End Results (SEER) Program, may be used to determine whether interventions (e.g., screening, radical prostatectomy, or radiotherapy) are effective. This study assessed whether the underlying COD on death certificates for men with prostate cancer agreed with an independent review of inpatient medical records in a sample of prostate cancer patients who died in King County, WA, in 1995. Our goal was to assess the validity of the coding system currently used by each state for determining underlying COD from death certificate data that are submitted to the National Center for Health Statistics, Hyattsville, MD. All study procedures were approved by the Institutional Review Board of the Fred Hutchinson Cancer Research Center, Seattle, WA. With the use of the Seattle–Puget Sound SEER Cancer Registry, we identified all 538 men (aged 58–98 years at time of death) in King County who were diagnosed with prostate cancer from 1973 through 1995 and who died during 1995. Any subject who died at one of 11 selected hospitals in King County was eligible for the study. Of the 171 men diagnosed with prostate cancer who died in any hospital in King County, 133 (78%) died in one of the 11 selected hospitals included in the study. Of these 133 case subjects, medical records could be reviewed for 128 (96%). A trained abstractor used a standardized data form developed for a previous study (9) to review medical records and evaluate clinical course before death and comorbidity. Although autopsies were performed on 12 (9%) case subjects, these reports were not available for review by the abstractor. Following abstraction of the medical records, three clinicians (a medical oncologist and two urologists), all of whom were blinded to the COD assigned by the death certificate, independently reviewed the abstraction forms and assigned an underlying COD to one of three prospectively defined categories: 1) related to prostate cancer, 2) unrelated to prostate cancer, or 3) uncertain. Disagreements in assignments were resolved by consensus among the clinicians. The assignment by the clinicians of underlying COD was then compared with that of the death certificate, using the National Center for Health Statistics algorithm for International Classification of Diseases coding of underlying COD (3). Agreement between the clinicians’ impressions and death certificate determination of COD was measured by the statistic (10). All three clinicians agreed on the underlying COD in 113 (88%) of 128 case subjects. After consensus was reached for the remaining 15 case subjects, all underlying CODs were compared with those from the death certificates (Table 1). After two case subjects with uncertain CODs were excluded, the statistic was 0.91, with agreement between the death certificate COD and clinicianassigned COD (related to or unrelated to prostate cancer) in 122 (97%) of 126 case subjects. Of the two case subjects in which the clinicians were uncertain of COD, one had metastases detected by a bone scan 1 year before death. This case subject also had clinically significant cardiovascular disease and increasing respiratory difficulty on final hospital admission. Because the clinicians were unable to determine the etiology of these respiratory problems, they coded the COD as uncertain, whereas the death certificateassigned COD was prostate cancer. The second case subject was diabetic and had metastatic disease detected by a bone scan 1 year before death. He also had increasing renal failure, poor mental status, and general debilitation. The death certificate-assigned COD was not prostate cancer. In the single case where the clinicians thought that the COD was unrelated to prostate cancer and the death certificateassigned COD was prostate cancer, the subject had respiratory failure as a result of staphylococcal pneumonia. The only reference to prostate cancer was a brief mention in the admitting history. For the three case subjects where the clinicians believed that the COD was related to prostate cancer and the death certificate-assigned COD was not prostate cancer, the medical records clearly stated that the treating providers thought that the deaths were directly related to prostate cancer. For one case subject, the immediate COD was respiratory failure