With this new technology, CDC began consulting with internal and external experts about the best approaches for implementing the STARHS technology to estimate the number of new HIV infections in the United States each year. Between 2001 and 2005, CDC held five consultations with experts to solicit guidance in epidemiology, surveillance, biostatistics, laboratory science, ethics, policy, and state/local public health practice on a variety of topics associated with developing and implementing HIV incidence surveillance. These topics included surveillance methods, statistical issues, policy and ethical considerations, implementation procedures, and laboratory and specimen transport issues. The recommendations from these consultations helped shape the development and implementation of the National HIV Incidence Surveillance System described here. Currently, 34 state and local moderate-to-high morbidity jurisdictions are funded to conduct HIV incidence surveillance (), covering approximately 85% of the epidemic. All funded systems began collecting data in 2005 and are expected to achieve full implementation in both public and private sectors in 2006.
34 U.S. jurisdictions implementing HIV incidence surveillance, August 2005
Recommendations from the first expert consultation on surveillance methods were fundamental in creating the incidence surveillance system. The panel considered several methods for estimating HIV incidence, including national household surveys, unlinked serosurveys, cohort record review studies, and expanding the current case surveillance system by applying STARHS to all new HIV diagnoses in the United States. After reviewing each model and considering their individual strengths and weaknesses, the panel recommended that HIV incidence be estimated by building on the existing national HIV case surveillance system. This system had collected complete, accurate, and timely data on AIDS cases for more than 15 years. Many states had integrated confidential, name-based HIV diagnosis reporting into their AIDS systems using the same validated methods as used for AIDS case ascertainment. Given the strength and reliability of this infrastructure, the panel recommended expanding it to include STARHS testing of all new HIV diagnoses reported to the system.
Expanding existing case surveillance infrastructure
Expanding the existing HIV case surveillance system to incorporate HIV incidence as a sentinel event also enables data use at the state level. CDC provides epidemiologic technical assistance to surveillance areas to meet program obligations for data collection and dissemination, and calculation of incidence will be included in this assistance.
The existing HIV case surveillance infrastructure has several limitations that affect the incorporation of incidence surveillance activities. First, implementation of HIV case surveillance has been staggered over the past two decades, with the last states adopting HIV reporting in 2004. Incidence estimates will not be stable until all states establish HIV case reporting systems that enumerate all HIV cases that were diagnosed prior to the year reporting was implemented. Identifying all diagnosed patients, and differentiating newly diagnosed individuals from prevalent ones, often requires that a reporting system operate for several years.8
Second, only a portion of the actual new infections can be observed using this system (i.e., infected individuals who chose to be tested for HIV). Anonymous testers—those not reportable to the confidential system until they seek care and receive their first confidential HIV test—or individuals who are not tested at all cannot be directly ascertained. Estimating the total number of new infections requires the use of statistical methods to account for the bias associated with using a system that directly measures only individuals who test voluntarily.
The HIV incidence surveillance system, developed as an extension of national HIV case reporting, has received a determination of non-research by the human subjects protection process at CDC and is exempt from review by the Institutional Review Board (IRB). Informed consent is not required to collect and analyze cases of notifiable diseases, so consent for STARHS is not required and most case surveillance data are derived from clinical records. Because HIV testing is a medical procedure, the privacy of HIV information that is maintained in clinical records is protected by federal and state laws,9
which specifically allow public health data to be collected from these clinical records. Therefore, data contained in the HIV incidence and case surveillance system are maintained according to the U.S. Department of Health and Human Services manual on security and confidentiality.10
Policies and procedures based on these guidelines and local laws are in place at state and local health departments. Information maintained in hard copy and electronic formats are secured and protected using these rules to assure the privacy and confidentiality of individuals reported as having HIV infection. These measures extend to protect the HIV incidence surveillance information held locally. Access by HIV incidence surveillance staff to information in the case record, HIV testing history, and STARHS data is governed by the same security and confidentiality requirements. Under these guidelines, information that could identify an individual (i.e., name, address, zip code) is not included in the dataset that is transmitted from local surveillance areas to CDC.
Laboratory technology for HIV incidence surveillance
The assay currently used in STARHS is the BED HIV-1 Capture Enzyme Immunoassay (EIA), named for the branched peptide sequences for HIV-1 subtypes B, E, and D used in the assay.11
CDC developed the assay for the express purpose of estimating population-based HIV incidence, not for clinical or diagnostic purposes. Because its individual-level clinical and diagnostic usefulness has not been determined or approved by the Food and Drug Administration, it is at this time considered a tool for public health surveillance only. Therefore, CDC and its surveillance partners are prohibited from providing individual test results to patients or their health-care providers. CDC has recommended to state and local surveillance programs that health-care and partner counseling and referral providers use other readily available epidemiologic, laboratory, and clinical information (such as history of a negative HIV test, history of primary HIV infection illness, CD4 count, and HIV viral load results) to choose the best course of action for treatment and partner notification.
The principle of the BED HIV-1 Capture EIA is based on the observation that the ratio of HIV-specific immunoglobulin G (IgG) to total IgG increases with time after HIV infection. The time lapse between an individual's HIV-positive test on the standard EIA to the time when the individual's serum—if tested with the BED HIV-1 Capture EIA—reaches an optical density (OD) level predetermined to distinguish recent from long-standing infections, is defined as the mean STARHS window period (). Although the mean STARHS window period may vary slightly by HIV subtype, the mean window period used to calculate population-based incidence estimates is 153 days.
In 2006, results from an incidence estimation using the BED HIV-1 Capture EIA in South Africa indicated concern that the assay, along with the statistical estimator, had overestimated the incidence of HIV infection compared with incidence estimated using other methods, including modeling and prospective studies.12
The inability to exclude individuals with AIDS in serosurveys, the use in the developing world of diagnostic algorithms with lower specificity, and the high levels of chronic co-infections (and thus IgG) among people living in the developing world may contribute to overestimation. In the U.S., however, linked case information is available to exclude individuals with AIDS who may falsely test “recent” on STARHS. Additionally, STARHS is done in the U.S. only in cases in which the specimen is confirmed HIV-1 antibody positive using a Western blot or immunofluorescence assay (IFA) confirmatory test. Finally, levels of chronic co-infection are low in the U.S. As such, false-recent results due to high IgG do not impact the assay's performance. A key factor for successful use of the BED assay in the U.S. is incorporating its use into the existing case-based surveillance system so that STARHS results can be interpreted in the context of all case information, as opposed to a single, cross-sectional assay result.
HIV case surveillance
National HIV case surveillance on which the incidence system is based is described in detail elsewhere.8,13
Briefly, all states and territories have laws or regulations requiring that HIV cases be reported to state or local public health agencies. As of July 2006, 50 states and territories use confidential, name-based reporting for HIV cases that they use for AIDS case reporting, resulting in an integrated HIV and AIDS disease reporting system. Information is reported on a standard case report form and includes data on patient demographics, HIV risk behaviors, laboratory and clinical events, and virologic and immunologic status. These data are submitted to the state or local public health authority and entered into the HIV/AIDS Reporting System, a standard software data management system. The reports are then forwarded without personal identifiers to CDC, where duplicate cases reported from more than one state are taken out through a process that identifies cases as potential duplicates (i.e., have the same alphanumeric soundex code,14
date of birth, and sex) of cases reported by other states. Through interstate communication procedures endorsed by the Council of State and Territorial Epidemiologists (CSTE),15
case identification numbers are sent back to state surveillance coordinators, who then use the identifying and epidemiologic information they retain to resolve whether the potential duplicates are the same or different people. This in turn is reported back to the CDC via the case identification numbers, and potential duplicates are resolved in the national dataset. This de-duplicated national dataset is used to develop the national epidemiologic profile of HIV/AIDS in the United States.
Data for HIV incidence
Two additional components are necessary for HIV incidence estimation: testing history information, including the number and timing of previous tests and a remnant blood sample from all newly diagnosed cases for STARHS testing. The acquisition of testing history information occurs as part of routine case surveillance through an expanded or supplemental case report form. The acquisition of the remnant blood sample occurs after an HIV test is confirmed positive. Remnant blood is aliquoted by the diagnostic laboratory and sent to the public health laboratory at the same time that the confirmation test is reported to the surveillance unit in the health department. The HIV incidence surveillance coordinator in each state then determines whether the report represents a new HIV diagnosis and, if so, notifies the public health lab to send the specimen for STARHS testing. STARHS results are returned to the reporting state's HIV incidence surveillance coordinator, where they are combined with the other case data.
The source of cases categorized as recent by STARHS consists of the population of people who volunteer for HIV testing and are diagnosed with HIV infection. However, the population of interest is all newly infected people, including those who were not tested for HIV soon after their infection. To extrapolate to the population as a whole from the observed sample, each observed recent infection must be statistically weighted based on the probability of testing shortly after infection. The details of the calculations used to estimate population-based incidence are presented in greater detail elsewhere.16
To calculate population-based incidence, all newly diagnosed cases classified as recently infected by STARHS or other information (i.e., a recent negative HIV test) during a calendar year are regarded as a sample of all newly infected individuals. The weights assigned to each identified case are based on an estimate of the probability of detecting a recent HIV infection in such a cross-sectional sample. This probability of detection is the probability that the diagnostic HIV test is undertaken during the window period of the BED assay. Thus, this probability is dependent on an individual's frequency of testing. The weight for each case is the inverse of the probability of detection. To estimate the probability of selection and assign a corresponding weight, members of the sample are assigned to mutually exclusive groups based on information about previous testing behaviors. These groups are used to assign weights based on whether the person was tested before the first positive result, type of HIV diagnosis (HIV diagnosed with or without AIDS), result of the BED test, and availability of necessary information.
Individuals in the observed sample are assigned one of three types of weight: zero weight, calculated weight based on detection probability, or derived weight using a proportional method for cases with incomplete information ().
Clinical and testing history information for people who are newly diagnosed with HIV used to assign weights to calculate the number of incident HIV infections in the entire population
A zero weight is assigned to all cases diagnosed as AIDS at the time of the first HIV-positive test. The incubation period for untreated HIV infection (i.e., time from HIV infection to AIDS symptoms) is approximately 10 years.17
By definition, an individual with AIDS at the time of HIV diagnosis has been infected for greater than the six-month STARHS window period. Cases diagnosed with AIDS when first diagnosed with HIV can be incorrectly categorized as a recent infection by STARHS, however, because of a fall in HIV-antibody levels. Cases with concomitant HIV and AIDS diagnoses are assigned a zero weight regardless of the STARHS result. A zero weight is also assigned to individuals who did not have AIDS at HIV diagnosis, did not take antiretroviral medications within the six months before testing positive, and who were deemed “long-standing” by STARHS.
Based on detection probabilities, calculated weights are assigned to cases for which it is known whether or not there was an HIV test before the first diagnostic test; that did not have concomitant HIV and AIDS diagnoses; that did not take antiretroviral medications within the six months before testing positive; and that were deemed “recent” by STARHS. Cases assigned a calculated weight include individuals whose previous testing history is known, including both those who report having had no HIV test before their first positive test and those who report having had a negative test prior to their first positive test.
Derived weights are assigned using proportional allocation to people for whom the existence of a test prior to the first positive HIV test is unknown.
There are two types of calculated weights: one for people who report having no previous test before the first positive test and the other for those who report having tested negative before the first positive test. The probability of selection and the corresponding weight for cases with no previous test can be obtained using survival functions that account for the distribution of the AIDS incubation period and the distribution of the BED window period. This probability is parameterized based on the proportion of individuals diagnosed with AIDS at the time of their HIV diagnosis.18
The calculated weight for people who report having had a negative test before the first positive test is given by the time interval (T) from the last negative HIV test to the first positive HIV test divided by the average time spent in the window period during the interval T among those who seroconverted at time zero. The average time in the window period is proportional to the size of T, because people who test very frequently will have less opportunity to spend long periods of time in the window period. However, for values of T greater than 24 months, this period essentially equals the overall mean window period, μ = 153 days, for the BED assay.
For all other cases for which the existence of an HIV test prior to the first diagnostic test is unknown, weights are derived from the calculated weights using proportional allocation—where the proportion of such cases among all cases with complete information is extrapolated to the population without STARHS data based on the entire number of newly diagnosed cases in the population.
Finally, in the incidence estimation step, cases are assigned an incidence weight, and the incidence count for a specific population group is estimated by summing the weights of respondents in the population considered:
where I = incidence, W = statistical weight, and i = case.
Computation of the associated variance is complex and has been outlined elsewhere.16,17