Underestimation of relative risks by standardized incidence ratios for AIDS-related cancers

PURPOSE: Registry-based studies provide valuable data regarding cancer risk among people with HIV/AIDS (PWHA). Such studies utilize the standardized incidence ratio (SIR) to estimate the relative risk (RR), an etiologically relevant measure. However, SIR may underestimate RR when HIV/AIDS prevalence in the general population or RR is high. We quantified the extent of this underestimation for 3 AIDS-related cancers: Kaposi sarcoma (KS), central nervous system non-Hodgkin lymphoma (CNS NHL) and cervical cancer. METHODS: We used data on cancer risk among PWHA from the U.S. HIV/AIDS Cancer Match Study. SIRs were compared with RRs estimated using two methods: (1) SIRs calculated using pre-AIDS era (19731979) cancer incidence rates (SIRpre-AIDS) and (2) SIRs calculated after subtraction of cancers known to be among PWHA from general population rates (SIRexclusion). RESULTS: For KS and CNS NHL, SIRs (117.8 and 133.9, respectively) calculated using overall general population rates substantially underestimated both SIRpre-AIDS (19,778 and 3,612, respectively) and SIRexclusion (657.7 and 536.4, respectively). In contrast, the extent of underestimation was negligible for cervical cancer (SIR = 4.9 vs. SIRexclusion = 5.1). For KS and CNS NHL, SIRs were higher in females than in males. However, SIRpre-AlDs and SIRexclusion estimates were more similar, indicating that SIR differences artifactually reflect differences in HIV/AIDS prevalence between males and females. For KS and CNS NHL, trends across calendar time were weaker in SIRs than in SIRpre-AlDs and SIRexclusion. CONCLUSION: For KS and CNS NHL, SIRs substantially underestimate RRs. This underestimation arises from the exceptionally high relative risk of KS and CNS NHL among PWHA. SIRs must be interpreted cautiously when HIV/AIDS prevalence is high or varies across groups of interest.