To identify any indicator of questionable reliability, we examined the pairwise distributions of positive status for each individual indicator of HIV-positive NHL. Most indicators shared high probabilities of positive status with each other, with the exception of presumptive death certificate evidence. The HIV/AIDS flag was positive in 95% of NHL cases with other positive registry indicators, including a definitive registry abstract mention or a registry report of KS, and also was positive in at least 91% of cases with positive cause of death (91%), definitive death certificate mention (92%), or AIDS registry match (91%) indicators. However, the HIV/AIDS flag was unlikely to be positive (15%) in the 87 patients with presumptive death certificate mention of HIV. The definitive abstract mention indicator was positive in at least 87% of NHL cases with other positive indicators (data not shown) except presumptive death certificate mention (25%). Although KS was recorded in the cancer registry for <20% of the NHL patients, the majority of these 378 patients also had a positive HIV/AIDS flag (95%), definitive registry abstract mention (99%), presumptive abstract mention (94%), underlying cause of death (90%), definitive death certificate mention (98%), or matched with the AIDS registry (88%). Among all deceased NHL patients for whom cause of death was known and a death certificate could be obtained, 99% of 1262 with an HIV/AIDS underlying cause of death also had a definitive death certificate mention; however, the converse was true for only 88% of 1424 patients with definitive death certificate mention, suggesting some discrepancy in the coding of cause of death. In addition, a positive cause of death or definitive death certificate mention was found in at least 84% of deceased NHL patients with positive registry or match indicators (data not shown), with the exception of presumptive abstract mention (29%). The AIDS registry match indicator was positive in at least 78% of patients with positive registry or death indicators; however, only 67% of patients with a presumptive abstract mention and 22% with a presumptive death certificate mention had a positive AIDS match. Eighty-eight percent of NHL patients with KS and 90% of NHL patients with death certificate mention of HIV/AIDS matched with the AIDS registry.
Completeness and Comparisons of HIV Indicators for HL Patients
For HL, which is not an AIDS-defining cancer and for which proportionately fewer cases were deceased than for NHL, completeness and results of comparisons between HIV indicators were different. Table 2 shows the completeness of each indicator for HL patients. Only 8% of all patients had any positive HIV evidence. Nearly twice as many had positive evidence from a cancer registry source (n = 105) as from a death source (n = 53) or from the linkage (n = 56), and the HIV/AIDS flag was the most informative indicator. Table 3 shows that 35% of HL patients with any HIV-positive evidence were identified by all 3 sources, 18% by 2 sources, and nearly half (48%) on the basis of 1 source only, overwhelmingly cancer registry information. Only 1 HL patient (<1%) was identified exclusively by the linkage.
Comparisons of individual indicators in HL patients showed that the HIV/AIDS flag, definitive abstract mention, and definitive death certificate mention indicators generally demonstrated high probabilities of positive status with each other. Deceased patients with positive registry indicators (HIV/AIDS flag, definitive abstract mention) were only moderately likely (51–58%) to have positive status for the cause of death indicator, although they were very likely (85–88%) to have definitive mention on the death certificate. The presumptive death certificate mention indicator was positive for only 1 patient who did not have positive evidence from any other indicator. The 56 HL patients who matched with the AIDS registry were likely to have a positive HIV/AIDS flag (77%), definitive abstract mention (74%), cause of death (81%), and definitive death certificate mention (92%). Conversely, the proportions of HL patients with positive evidence from these indicators who also matched with the AIDS registry were lower (44–52%), reflecting the middling proportion (49%) of patients with any positive indicator who matched with the AIDS registry.
Completeness and Comparisons of HIV Indicators for Anal Cancer Patients
Completeness of HIV indicators for anal cancer patients was generally similar to those for HL (Table 2). Overall, 168 patients (18%) with anal cancer, a non-AIDS-defining cancer, had evidence of HIV-positive status on the basis of at least 1 of the 6 indicators, and the single most informative indicator was the registry abstract. Consequently, although the HIV/AIDS flag is not collected for anal cancer patients, more patients had HIV/AIDS evidence from a cancer registry source (n = 138) than from a death source (n = 75) or from the linkage (n = 85). Table 3 shows that about half of all anal cancer patients with any HIV evidence were identified by multiple sources, 21% by 2, and 29% by all 3 sources. Fifty-two percent of patients were identified by 1 source only, mostly registry abstract information. However, in comparison to NHL and HL cases, larger proportions of anal cancers were identified as HIV positive solely on the basis of death information (9%) or the linkage (5%).
Interindicator concordance was wider in anal cancer patients than in NHL or HL patients. About half of the 110 anal cancer patients with definitive abstract mention of HIV/AIDS had positive status for match (52%) and cause of death (56%) indicators, although they were very likely to have definitive mention on the death certificate (90%). Patients with a presumptive abstract mention showed variable levels of positive status for other indicators. Thirty-five percent of patients matched with the AIDS registry; deceased patients were likely to have definitive mention on the death certificate (86%) but less likely to have an HIV/AIDS cause of death (50%). Anal cancer patients with the KS indicator were likely to have other positive registry and death indicators (data not shown), but only 62% of these 21 matched with the AIDS registry. In deceased patients with other registry and match evidence, most (86–100%) had definitive mention of HIV/AIDS on the death certificate. One hundred percent of patients with an HIV/AIDS underlying cause of death had a definitive death certificate mention, but the converse was true for only 67% of patients, providing more evidence for nonuniformity in the coding of underlying cause of death. The presumptive death certificate mention indicator was positive for only 1 patient and thereby could not be compared effectively. Patients who matched with the AIDS registry were likely to have a definitive death certificate mention (91%) but less likely to have definitive abstract mention (70%) or an HIV/AIDS cause of death (63%).
Development of a Classification Strategy
The majority of presumed HIV-positive NHL patients (83%) and at least half of HL and anal cancer patients were identified by multiple sources of HIV indicators (Table 3). Acceptable levels of agreement were demonstrated between most pairs of individual indicators, except for presumptive abstract mention, which shared middling levels of positive status across all sites, and presumptive death certificate mention, which was uncommon for HL and anal cancer patients but was unlikely to be positive in NHL patients with other positive evidence. Thus, it seemed prudent to exclude presumptive indicators of HIV-positive status from a registry-based classification scheme for HIV-associated NHL, HL, and anal cancer.
As both the manual review of registry abstracts and death certificates, and record linkages with AIDS registries, are time-consuming processes, we also evaluated the mode of identification (electronic, manual review, linkage) for patients with evidence of HIV-positive status, but excluding patients with presumptive evidence only (Table 4). For presumed HIV-positive NHL patients, 98% were ascertained by at least 1 electronic, registry-based indicator. For HIV-positive HL patients, 92% were identified by at least 1 electronic indicator, while manual review ascertained 8% of patients. However, for anal cancer, the manual review identified the majority of patients (54%), electronic indicators identified a smaller proportion (38%), and linkage alone identified 7%.
These results suggest that reasonable strategies for the cancer registry–based classification of HIV-associated cancers differ for NHL, HL, and anal cancer. For NHL and HL patients, evidence from any electronic indicators (HIV/AIDS flag, KS report, underlying cause of death) should identify the great majority of HIV-positive cases with accuracy and efficiency. However, for anal cancer, manual review should be conducted to identify HIV-associated patients and used in addition to electronic information; AIDS registry linkage information should be considered optional.
This study of HIV status information routinely available for NHL, HL, and anal cancer patients in our population-based cancer registry has demonstrated that available registry and death indicators are useful in classifying HIV status in cancer registry patients. Our results show that the combination of 5 cancer registry–available indicators of HIV status identified 25% more NHL patients and nearly 50% more HL and anal cancer patients than were detected by linkage with the state AIDS registry. Furthermore, patients identified as HIV/AIDS positive by registry-available indicators included nearly all of the patients identified through AIDS registry linkage (99% of NHL and HL and 93% of anal cancer). These observations suggest that, compared with the linkage-based mode of classification, cancer registry–available indicators are of sufficient completeness to classify the HIV status of NHL, HL, and anal cancer patients. While the validity of cancer-registry indicators could only be approximated in these data, we did not uncover any contraindication to using the HIV/AIDS flag, HIV/AIDS cause of death, record of KS, or definitive mention of HIV/AIDS from registry abstracts or death certificates. However, as outlined above, our recommendations for classifying HIV-associated NHL, HL, and anal cancer differed by cancer site.
The accuracy of an HIV/AIDS classification system based on cancer registry and death information ultimately depends on the validity of the medical records and death certificates from which this information is derived. The few studies addressing the quality of these sources of information suggest acceptable sensitivity. Cancer registry abstracts were found to mention HIV in 82% of San Francisco NHL patients known to be HIV positive through recruitment for a case-control study, 21 and death certificates recorded HIV infection or AIDS for 91% of deceased AIDS patients in a San Francisco HIV cohort study. 22 Thus, our method, particularly the classification of persons with only 1 source or indicator of HIV-positive status, remains susceptible to misclassification, particularly false classification of truly HIV-positive persons as HIV negative.
Our results indicate that linkage with the AIDS registry identified substantially fewer HIV/AIDS-associated cancer patients than could be detected with cancer registry–available information, even among those with AIDS-defining conditions (NHL, KS), or death-certificate indication of AIDS. It is uncertain what proportion of HL and anal cancer patients did not link with the AIDS registry because the patient had not developed AIDS. in addition, although the linkage protocol included manual review of all possible matches, differences in patient-identifying information between the AIDS and cancer registries or faulty linkage methods could explain some or all of the discrepancy for NHL and KS patients. However, because study patients with evidence of AIDS who did not match with the AIDS registry were substantially more likely than matched patients to be female, over age 55, or alive at the study end date, but were equally likely to have been diagnosed in recent years or in San Francisco county, their failure to match may relate more to AIDS registry incompleteness. Such incompleteness could involve the entire patient record, as may occur when patients migrate interstate between initial AIDS registration and later cancer diagnosis, or be limited to patient identifiers (e.g, name, social security number, address) that would preclude confirmation of a possible match with cancer registry identifiers.
Few reports have addressed why AIDS and cancer registry records might not match appropriately. Grulich et al. 23 found that 86% of Australian HIV-associated NHL patients first identified as such as part of a case-control study later matched with the AIDS registry and that no false-positive matches occurred. In a Southern California population, Diamond et al. 24 reported that several NHL patients who did not match with the local AIDS registry had an HIV/AIDS underlying cause of death, consistent with our observation of non-linkage for persons with positive death evidence, and suggesting that linkage identified a maximum of 88% of true HIV-associated NHL patients. Our observation that only 75% of presumed HIV-associated NHL patients linked with the state AIDS registry may not necessarily bear on the results of AIDS-cancer registry matches conducted elsewhere. However, our results do underscore the possibility of linkage inadequacy and suggest that future AIDS-cancer registry matches should confirm the absence of cancer registry–available indicators of positive status in unmatched cases to ensure complete ascertainment among dually registered patients.
The generalizability of our results to other cancer registries should be explored further. Our methodology should be reasonable for use by other SEER registries or by any other US or international cancer registries with similar data resources and medical record abstraction protocols and without confidentiality restrictions on recording the HIV/AIDS status of cancer patients. On the other hand, the generalizability of our methodology may be impacted by certain features of the Greater San Francisco Bay Area population, which differs from others in its extremely high HIV prevalence (approximately 3.5% of the San Francisco County population in 2001) 25 and the predominance of gay and bisexual men in its HIV-infected population. Thus, availability of HIV information in medical or cancer registry records may be different in this than in other regions. For example, as KS is more common in gay and bisexual men, owing to sexual transmission of the presumed etiologic virus, 26 it may have lower prevalence in populations with lower proportions of this HIV risk group. Similarly, the higher prevalence of HIV disease in San Francisco might correspond with improved medical surveillance and disease reporting than in low-incidence areas.
The introduction of HAART has intensified the need for timely population-based surveillance of cancer in HIV-infected persons, but, paradoxically, it has also made it more methodologically complex. Because HAART slows progression from HIV infection to the development of AIDS and from AIDS to death, 27,28 fewer HIV-infected persons will be identifiable as such on the basis of AIDS registry linkages and death records. Thus, going forward, our methodology for identifying HIV-associated cancers in cancer registry data likely will be based on higher proportions of patients identified by only 1 source of evidence. Indeed, the proportions of NHL, HL, and anal cancer patients identified by a single source were higher in 1996–998 (33, 68, and 71%, respectively) after the introduction of HAART, than in 1990–1995 (13, 36, and 37%, respectively). Furthermore, availability of HIV indicators may differ by factors influencing access to HAART. In San Francisco County, African-Americans, injection drug users, and persons without health insurance were less likely to have used HAART before AIDS diagnosis. 29 As demonstrated in these data, there is a need for all future activities for the surveillance of HIV-associated cancers, including AIDS-cancer registry matches, to incorporate as many diverse sources of information as possible to optimize detection and accurate classification of cases.
In the Greater San Francisco Bay Area, population-based cancer registry surveillance of HIV-associated cancers identified using registry-available data offers reasonable completeness and involves less administrative challenge and, thus, the opportunity for improved timeliness, compared with surveillance limited to AIDS registry-linked data. Using the methodologies outlined here, some measures of the burden of HIV-associated cancer, such as case counts, patient characteristics, and survival time, are readily calculated, while others, specifically cancer rates, are more difficult to estimate, due to challenges in estimating the underlying population at risk. While AIDS registry linkage-based studies circumvent this issue by using all AIDS-registry patients as the population at risk, analyses of cancer registry data can estimate the burden of HIV-associated cancers as the difference between overall rates and rates in presumed HIV-negative persons based on census-based population denominators. 30 However, improved methods are needed for estimating age-, sex-, and race-specific numbers of HIV-infected persons at risk. Production of these estimates may be facilitated with the advent of HIV surveillance programs, initiated recently in California. Even as we await such methodologic refinements, cancer registry data should be investigated more fully in more regions as an important resource for the population-based surveillance of HIV-associated cancer, especially in light of HAART-related reductions in AIDS registrations.
The authors thank Dr. Scarlett Lin Gomez, Patricia Harasty, Anne-Marie France, and Mary Schroeder for their contributions to this study; Marie Jungkeit, Jim Creeger, and Dr. Juan Ruiz of the California Department of Health Services Office of AIDS for conducting the data linkage; and Julie Durr, Dr. Maryjean Schenk, and Dr. Rick Severson of the Karmanos Cancer Institute, Detroit, MI, for their collaboration on development of the criteria for presumptive evidence.
1. Centers for Disease Control and Prevention. A cluster of Kaposi’s sarcoma and Pneumocystis carinii
pneumonia among homosexual male residents of Los Angeles and Orange Counties, California. MMWR
2. Ziegler JL, Beckstead JA, Volberding PA, et al. Non-Hodgkin’s lymphoma in 90 homosexual men: relation to generalized lymphadenopathy and the acquired immunodeficiency syndrome. N Engl J Med
3. Frisch M, Biggar RJ, Engels EA, et al. Association of cancer with AIDS-related immunosuppression in adults. JAMA
4. Grulich AE, Wan X, Law MG, et al. Risk of cancer in people with AIDS. AIDS
5. Serraino D, Boschini A, Carrieri P, et al. Cancer risk among men with, or at risk of, HIV infection in southern Europe. AIDS
6. International Collaboration on HIV and Cancer. Highly active antiretroviral therapy and incidence of cancer in human immunodeficiency virus-infected adults. J Natl Cancer Inst
7. Jones JL, Hanson DL, Dworkin MS, et al. Effect of antiretroviral therapy on recent trends in selected cancers among HIV-infected persons. Adult/Adolescent Spectrum of HIV Disease Project Group. J Acquir Immune Defic Syndr
. 1999;21(suppl 1):S11–S17.
8. Grulich AE, Li Y, McDonald AM, et al. Decreasing rates of Kaposi’s sarcoma and non-Hodgkin’s lymphoma in the era of potent combination antiretroviral therapy. AIDS
9. Grulich AE, Wan X, Law MG, et al. B-cell stimulation and prolonged immune deficiency are risk factors for non-Hodgkin’s lymphoma in people with AIDS. AIDS
10. Clarke CA, Glaser SL. Changing incidence of non-Hodgkin’s lymphomas in the United States. Cancer
11. Borges T, Watkins J, Stafford R, et al. Collection and verification of matched data from selected United states cancer and HIV/AIDS registries. Abstract. Paper presented at: 2nd National AIDS Malignancy Conference; April 6–8, 1998; Bethesda, MD.
12. Cote TR, O’Brien TR, Ward JW, et al. AIDS and cancer registry linkage: measurement and enhancement of registry completeness. The National AIDS/Cancer Match Study Group. Prev Med
13. Biggar RJ, Horm J, Goedert JJ, et al. Cancer in a group at risk of acquired immunodeficiency syndrome (AIDS) through 1984. Am J Epidemiol
14. Rabkin CS, Yellin F. Cancer incidence in a population with a high prevalence of infection with human immunodeficiency virus type 1. J Natl Cancer Inst
15. Clarke CA. Changing incidence of Kaposi’s sarcoma and non-Hodgkin’s lymphoma among young men in San Francisco. AIDS
16. Eltom MA, Jemal A, Mbulaiteye SM, et al. Trends in Kaposi’s sarcoma and non-Hodgkin’s lymphoma incidence in the United States from 1973 through 1998. J Natl Cancer Inst
17. Franceschi S, Dal Maso L, Arniani S, et al. Risk of cancer other than Kaposi’s sarcoma and non-Hodgkin’s lymphoma in persons with AIDS in Italy. Cancer and AIDS Registry Linkage Study. Br J Cancer
18. Li Y, Law M, McDonald A, et al. Estimation of risk of cancers before occurrence of acquired immunodeficiency syndrome in persons infected with human immunodeficiency virus. Am J Epidemiol
19. Engels EA, Frisch M, Biggar RJ, et al. Re: Estimation of risk of cancers before occurrence of acquired immunodeficiency syndrome in persons infected with human immunodeficiency virus. Am J Epidemiol
20. Grulich AE, Li Y, McDonald A, et al. Rates of non-AIDS-defining cancers in people with HIV infection before and after AIDS diagnosis. AIDS
21. Holly EA, Gautam M, Bracci PM. Comparison of interviewed and non-interviewed non-Hodgkin’s lymphoma (NHL) patients in the San Francisco Bay Area. Ann Epidemiol
22. Hessol NA, Buchbinder SP, Colbert D, et al. Impact of HIV infection on mortality and accuracy of AIDS reporting on death certificates. Am J Public Health
23. Grulich AE, Wan X, Coates M, et al. Validation of a non-personally identifying method of linking cancer and acquired immunedeficiency syndrome register data. J Epidemiol Biostat
24. Diamond C, Taylor TH, Anton-Culver H. Characteristics and survival of patients with non-Hodgkin’s lymphoma with and without acquired immunodeficiency syndrome. Hematol Oncol
25. San Francisco Department of Public Health AIDS Surveillance Unit. Annual AIDS Surveillance Report
26. Rezza G, Andreoni M, Dorrucci M, et al. Human herpesvirus 8 seropositivity and risk of Kaposi’s sarcoma and other acquired immunodeficiency syndrome-related diseases. J Natl Cancer Inst
27. Palella FJ, Delaney KM, Moorman AC, et al. Declining morbidity and mortality among patients with advanced human immunodeficiency virus infection. HIV Outpatient Study Investigators. N Engl J Med
28. McNaghten AD, Hanson DL, Jones JL, et al. Effects of antiretroviral therapy and opportunistic illness primary chemoprophylaxis on survival after AIDS diagnosis. Adult/Adolescent Spectrum of Disease Group. AIDS
29. Hsu LC, Vittinghoff E, Katz MH, et al. Predictors of use of highly active antiretroviral therapy (HAART) among persons with AIDS in San Francisco, 1996–1999. J Acquir Immune Defic Syndr
30. Glaser SL, Clarke CA, Gulley ML, et al. Population-based patterns of HIV-related Hodgkin lymphoma in the Greater San Francisco Bay Area: 1988–98. Cancer
Keywords:© 2004 Lippincott Williams & Wilkins, Inc.
non-Hodgkin lymphoma; Hodgkin lymphoma; anal cancer; HIV; cancer registries; methodologic