We read with interest the recent article by Jia et al.1 on familial risk of cancer and the clustering of nasopharyngeal carcinoma (NPC). The authors reported that first-degree relatives of patients with NPC had a dramatically decreased risk of developing malignant disease in general (standardized incidence ratio [SIR], 0.35), and carcinoma of the lung (SIR, 0.24), carcinoma of the esophagus (SIR, 0.23), and carcinoma of the breast (SIR, 0.08) in particular. The results are striking, as, to our knowledge, protective effects of this magnitude have not been observed before among relatives of individuals with cancer. The results of the study were derived from a large group of patients with NPC. However, the method used deserves comment, as the study design suggests limitations not addressed in the article.
First, the authors obtained information on malignant diseases in relatives from Guangdong by interviewing probands. Thus, dates of birth, occurrences of malignancies, dates of diagnosis, and dates and causes of death were based on recall. This information was then compared with data from the Hong Kong Cancer Registry, which includes information from hospitals, laboratory and pathologic departments, death certificates, and discharge summaries.2 It is therefore highly questionable as to whether the two data sources were comparable in terms of their completeness.
Second, the authors estimated SIRs by applying stratum-specific incidence rates from Hong Kong for the period 1988–1992 to the corresponding numbers of person-years in the study cohort. However, the observation time for relatives was calculated from the time of birth to the date of death, diagnosis, or interview, which means that the observation period for relatives could begin a lifetime earlier. Comparing cancer rates across such a time span is difficult, as incidences will have changed significantly over the period in question. For example, over the period 1973–1999, the age-adjusted rate of breast carcinoma in Hong Kong has increased considerably.3 Therefore, instead of the desired comparison between relatives and the background population, a comparison between earlier and current cancer rates is achieved.
Both of these limitations easily could have led to the low estimates of cancer risk that were reported. Thus, the results of the study are inconclusive; exclusion of the possibility of an excess of malignancies other than NPC in these families is not reasonable based on the study design that was chosen. A more accurate estimation of risk among relatives could have been achieved by recruiting and interviewing a matching control group, an option that would have yielded the maximum benefit from this large and valuable data source.