Part Seven:
My HIV Risk Assessment Software: A $406,000 Project Sabotaged by Waste, Mismanagement, and Fraud in the AIDS Program by Michael Phillip Wright
Norman, Oklahoma
Copyright 2002, by the Author
Succinct SummaryFrom the federal Small Business Innovation Research program I was in the early 90s awarded approximately $406,000 in grant funds to assist in the development and testing of expert system software to provide personal anonymous assessment of current risk for HIV infection. These awards were the outcome of a four-year application effort. I was also funded to apply the computer risk assessment concept to chlamydia and hepatitis B. 1,2
My grant applications required me to submit an experimental plan for scientific review and to defend the expectation that the software would be marketable. I suggested that anticipated clients would include public health agencies in the United States. The agreement of federal grant review teams with this expectation was clearly implied by the fact that I was funded.
Only .4 percent of the U.S. population was estimated to be HIV-infected at that time. My concept was based upon limited testing focusing on individuals likely to be infected. It was intended to be cost-effective and efficient by eliminating wasteful testing and counseling of persons not at risk and by maximizing the potential benefits of current technology. Another likely benefit was reduction in risk for false positive HIV test results.
In the time since my grant application process began, the Centers for Disease Control undertook the deceitful course of spreading the fear of universality of risk for HIV infection. If all sexually active people were indeed at risk, it would follow that all should be routinely tested. This of course undermined the marketability of my software, developed to utilize a refined probability computation to make discriminating judgements about who is likely to be infected.
This unjustified scare campaign, while succeeding in significantly augmenting the AIDS budget for CDC and other agencies, has accomplished nothing else other than waste of resources and unnecessary alarm in the general population. Their campaign has likely inflicted thousands with the burden of false positive HIV test results, and it has obviously obstructed the possibility of the American public enjoying the benefit of the software for which their tax dollars supported development and extensive testing.
Origins of the Software Project
I began my professional endeavor in developing HIV risk assessment software in the summer of 1987. My background had been mostly in contracted research and consulting. For the year ending in April 1987 I had been a health planner for a federally-funded clinic serving mostly low-income clients in northeast Oklahoma County. In that time I began to read professional literature about AIDS.
After my one-year employment contract expired I conceived of the idea of an anonymous computer interview and intelligent risk assessment system to convince people at risk of current HIV infection to seek an antibody test and receive appropriate counseling and treatment. Since such an interview would require detailed questioning about sexual history and drug abuse, I believed that the non-judgemental encounter with a computer would be a comfortable setting for users.
At this time my assets consisted of a 1981 Mustang, a microcomputer and printer, a desk, two file cabinets, and about $15,000 in cash. Instead of pursuing other employment prospects I decided to commit myself to the risky task of trying to obtain funding for my software concept.
I created a somewhat simple first-stage interview and risk assessment system within a couple of months and included a sample diskette with a December 1987 grant application submitted to the highly competitive Small Business Innovation Research (SBIR) program of the US Public Healh Service. This application was not funded.
In 1988 I created a much more elaborate version of the software. The 1988 effort began with the discovery of this equation for computing the cumulative probability of infection from an individual's sexual history:
This equation was published in a 1988 Science article by Harvey Fineberg, Dean of the Harvard School of Public Health.3 The incorporated terms are estimated HIV prevalence (p) among potential sexual partners, risk of infection (r) from a single unprotected sexual event, estimated number (NC) of events per partner without using a condom, estimated number of events (C) per partner using a condom, and number (m) or estimated number of partners within a specified time interval.i = 1- [p(1 - r) NC (1 - [1 - c]r)C + (1 - p)]mI rebuilt the software around this equation. The task required extensive review of the epidemiologic literature to look for estimates of per-event HIV transmission risk for vaginal and anal intercourse. It also required me to research literature necessary for creating the software's database of HIV prevalence estimates for different metropolitan areas, risk groups, and ethnic groups.
Extremely Low Risk of Vaginal HIV Transmission
A milestone event in this research was my discovery of an article in the Journal of the American Medical Association (JAMA) 4 which referenced a study of HIV vaginal transmission risk presented at the Third International AIDS Conference. The researchers reported that the per-event, male-to-female HIV transmission risk was .001 (one chance in a thousand). 5 The authors of the JAMA article used this estimate to demonstrate that the odds were 5 million to one against a new HIV infection taking place in a single act of unprotected vaginal sex between two people who are members of that massive population which the CDC recognized and labeled as "heterosexuals without specific identified risk." 6
This and numerous other articles in the medical press convinced me that there was absolutely no reason to fear an HIV epidemic sustained by vaginal sex in this country.7 Although the CDC in the late 80s began to fan the flames of fear with such statements as "heterosexual women are now the fastest-growing group of new AIDS cases," they were deliberately misleading the public.8 Careful analysis would reveal that AIDS cases among women have been largely confined to those who have either used injecting drugs or submitted to anal penetration.
In February 1995, The New York Times published a CDC finding that only about one fourth of the estimated 40,000 new HIV infections in the US during the earlier year were among gay men. 9 The remaining three quarters, mostly residents of inner city minority populations, were counted as either injecting drug users or their sexual partners.
1991 Grant Awards
The SBIR application process was an excrutiating ordeal which drained my energy and funds. Over a four-year period I wrote eleven applications and resubmissions before finally being funded in September 1991. I submitted applications also to apply the computer risk assessment concept to chlamydia and hepatitis B.
In one month I was awarded three grants simultaneously for the HIV, chlamydia, and hepatitis B projects. This was quite a surprise. I immediately had to request extensions in order to perform all three satisfactorily. The funding was not a moment too soon. My personal savings were almost exhausted and I was close to destitution.
During fall 1991 I also enjoyed being published in the health science press. In 1988 I had begun an unfunded research project to assess the validity of my software. This involved recruiting interview subjects from the Monday Night Clinic, an anonymous HIV testing facility in Oklahoma City. Using anonymous IDs entered into the software by subjects answering questions about their personal histories, the computer's risk assessment was compared to the clinical record of antibody status for each. The results of this study were published in the fall 1991 issue of the journal AIDS Education and Prevention.10 This article was reviewed and cited by the authors of a study of medical diagnostic decision support (MDDS) systems for the premier issue of the Journal of the American Medical Informatics Association. 11
High Accuracy of Software
The results of my first SBIR HIV software project were presented at the Proceedings of the Oklahoma Symposium on Artificial Intelligence at Oklahoma State University in November 1993.12 In a paper co-signed by me and evaluators Larry Prater, MD and E. Laurette Taylor, PhD, it was reported that the software gave predictions of negative HIV status consistent with subjects' antibody status in 234 of 235 trials. Neither Prater nor Taylor had any financial interest in the software. They were compensated from grant funds as scientific consultants.
Accepting antibody status as a gold standard would yield a negative predictive value of .996 (235/234). After this project, new information reached me about the large number of factors which can trigger false positive HIV antibody test results.13 For this reason, the software may well have been accurate in predicting negative status for the one case for which an error was recorded. Further, this case was associated with a subject who self-reported antibody status without clinical confirmation.
The main task of the Phase II project, initiated in September 1993, was to express the software as a telephone interview in which users could make a toll-free call from Dallas and enter data with the telephone keypad in response to a voice interview. Over a year's time the software successfully completed 2813 telephone interviews and risk assessments. Of these, 811 callers were assessed to be at risk, and 166 (20.5%) were confirmed to have presented themselves for antibody-testing at a cooperating facility.14
Refusal of Oklahoma Health Department to Cooperate
I began my effort to enlist cooperation from the Oklahoma Department of Health (ODH) in 1988. At my request, Gayle Townley, former director of the Oklahoma Department of Mines, wrote to Health Commissioner Joan Leavitt and called her attention to my project. In a letter to Townley and dated February 23, 1988, Leavitt replied:
While claiming that their "pamphlets" provide "essentially the same information," Dr. Leavitt, did not explain how pamphlets could receive data from individual respondents and then provide a refined estimate of cumulative probability of infection after computation incorporating Dr. Fineberg's equation. Further, she did not offer any empirical evidence to contradict the findings of research indicating patients' preference for computer interviews when sensitive personal behaviors are at issue. 15,16 She also ignored numerous published findings indicating that many people at risk for AIDS fear ostracism and betrayal of confidence and thus are reluctant to seek counseling and antibody testing in conventional settings. 17,18,19There is certainly merit to this; however, we choose to provide the information in a different manner. We have developed several pamphlets which essentially provide the same information and can be widely distributed. Additionally, there are toll free hotlines at the national and local levels and we are currently training a number of individuals to run a statewide toll free hotline, where persons can receive information anonymously. We feel that persons who think that they are at risk of HIV infection want to and should talk to an individual rather than a machine and thus do not feel that this proposal is consistent with our needs.No ODH Cooperation for Phase II Grant
In September 1993 I was awarded a Phase II SBIR grant in the amount of $357,000 from the National Cancer Institute. This was to support continued development and testing of the HIV risk assessment software. The main task was to express the software as a telephone interview in which users could make a toll-free call from Dallas and enter data with the telephone keypad in response to a voice interview.
I also wanted participation from Oklahoma City area residents, and for this reason asked the Oklahoma Department of Health to provide the phone number for the automated risk assessment interview to individuals calling their AIDS hotline. The ODH refused to do this.
Hot Pursuit of Federal Ryan-White Funds
My request for this cooperation was made in a February 1992 meeting with Dr. Paul Zenker of the AIDS Division. At that time I was writing the proposal for the Phase II SBIR grant. Refusing to cooperate, Dr. Zenker said that my project presented a problem because it refers subjects to anonymous sites for antibody-testing. The research plan approved by federal agencies included this requirement.
Dr. Zenker said that the problem related to his agency's federal funding considerations, and mentioned the Ryan-White Act. In a letter to my project officer at the National Cancer Institute (3/24/92) he stated that federal Ryan-White funds for HIV are directly proportional to the number of AIDS cases reported by the states. Objecting to the fact that my software referred people to clinics providing anonymous HIV-testing, he pointed out that subjects who are tested consequent to my software interview will not count towards CDC statistics. The names of anonymous subjects testing positive obviously are not entered into state records.
Zenker also wanted to change the software's advice screens to declare subjects having less than a 1% chance of current infection to be "at some risk." In earlier testing, subjects were advised by the software that they were at risk only if they had a probability greater than or equal to .01. Further, even if they were less than .01 the software still advised them to be tested at a clinic to see if its prediction of negative antibody status would be confirmed. Not interested in the fact that in its Phase I funded experimental phase the software achieved a negative predictive value measure of .996, Zenker appeared to be afraid of missing a few outlying HIV-positive cases at the margins of statistical likelihood.
It was clear that Zenker was concerned that his bureaucracy might not enjoy a sufficiently satisfying degree of revenue enhancement from Ryan-White funds if he cooperated with my software project. He did not want to miss any opportunities to increase the AIDS case count for Oklahoma.
The obvious error in Zenker's stated objections was the fact that my software project did not even concern itself with the question of whether respondents were AIDS cases. The issue was whether they were likely to be HIV positive. The distribution of HIV positives across states does not affect Ryan-White funds. As Zenker himself pointed out, the pertinent measure is AIDS cases
How do we account for his paradoxical behavior? Official wisdom holds that the median time elapsed from the moment of HIV infection to the development of AIDS is ten years. Was Zenker making plans for ten years into the future? He clearly did not want to miss the opportunity to bring any prospective HIV positives into his records. On behalf of the ODH's revenue enhancement aspirations, each HIV case was seen as a seed which, with proper cultivation with toxic anti-viral drugs, could grow into an AIDS case which would bear financial fruit in the form of Ryan-White dollars for the AIDS division.
ODH Hostility to Anonymous Testing Facility
The AIDS Division of ODH has a history of antagonism to Larry Prater, MD, who was an important collaborator in my software development effort. Dr. Prater was the director of the Monday Night Clinic (MNC), an anonymous testing clinic which was Oklahoma's most effective HIV antibody-testing site in the late 80s and early 90s.
Data originating with the ODH confirms that from 1985 through 1990, the MNC yielded 59% of all HIV positives tested in Oklahoma.20 In 1988 the MNC accounted for 70% of all HIV positive tests with a total of 228. The average HIV positive yield for all other facilities operated by or contracted with the ODH that year was only 7. Thus in that year the MNC was 33 times more effective than the average of the other facilities. The other facilities were funded by ODH. MNC was not.
In spite of its excellent performance, the MNC received no contract support from the ODH until FY 1991. For the years 1988 through 1990, the ODH paid contract sums totaling $342,943 to a group of HIV antibody-testing facilities. These units performed a total of 2845 tests yielding only 37 positive results. 20 Thus the per capita cost of positives yielded by this program was $9268. At this rate, estimating 8,000 Oklahomans who were positive but not tested at that time, the predictable cost for yielding the state's remaining HIV positives would have been 74 million dollars. This does not include the lab cost of the test -- only the preliminary counseling and drawing of the blood samples.
The per capita cost was so high because the clinics tested 66 negatives for every positive yielded by the program. One of the intended functions of my software was to provide an inexpensive preliminary screen to sort out people not at risk who do not need to test. This could then prioritize human effort into testing and counseling persons who are really at risk for HIV and likely to be positive. The ODH was not interested.
ODH Provided No Incentive for Testing HIV-Positives
In a memo dated 1/10/92, I described my software project to Jay Moskowitz, who was involved with drafting the NIH Strategic Plan. Moskowitz was the Associate Director for Science Policy and Legislation. I addressed the inefficient and wasteful practices I had observed in conventional HIV testing and counseling operations:
Waste, Inefficiency, and False PositivesThe goal of many HIV testing facilities is obviously to test as many people as possible without regard to the issue of the number of positives yielded. An HIV counselor employed by Drug Recovery Inc. (DRI), a facility contracted to utilize my software under the SBIR research project, expressed dismay at the fact that the software encourages their clients to follow up with antibody testing at anonymous sites in the area. This was because she was paid under funds provided by the Oklahoma Department of Health, which requires most of its contracted sites to report names of persons testing positive. She explained that she receives an "effort point" under her contract for every client she tests, and a point for every three refusals. She was afraid she would lose her effort points if the clients followed the software's suggestion to follow up at other locations. She received no additional points if anyone tests positive.In the 1993 paper presented at the Proceedings of the 7th Oklahoma Symposium on Artificial Intelligence,12 my research collaborators and I reported that, according to the National Health Interview Survey, 36 million Americans had been tested for HIV as of 1989.21 At that time the CDC was estimating that there were one million positives. 22
Researchers involved with the NHIS study concluded that the majority of persons at increased risk had not yet been tested. According to CDC publications, publicly-funded HIV testing facilities conducted 4.6 million tests from 1985 through 1991 and yielded 207,518 positives over the seven-year period.23,24 The average annual positive yield from the funded facilities was about 30,000 -- three percent of the one million believed at that time to be positive. Since 1988 the CDC has been estimating 40,000 new annual infections. 9 Thus the testing and counseling system in place appeared unlikely ever to achieve the success of having informed all HIV positive individuals of their status.
The Danger of False Positives
A likely consequence of the campaign of fear and indiscriminate HIV-testing of persons not at risk has been to inflict thousands of Americans with false positive HIV test results. A 1988 article in the AIDS Bulletin,, a U.S. Department of Justice publication, reported that the HIV Western Blot test is susceptible to quality control problems 25. The Bulletin continued:
Commenting on "numerous proposals to extend HIV antibody mass screening to various new populations," the Bulletin concluded with the cautionary statement that "HIV antibody screening should not be considered a panacea for the problem of AIDS in our society."...several researchers contend that the number of of false positives will be unacceptably high in populations where the actual incidence of infection is very low...These researchers have calculated that the percentage of positive results that will be false after the entire test sequence (including the Western Blot) in very low-risk populations could be in the range of 28 to 90 percent.In October 1987, Dr. Stephen Pauker of the New England Medical Center in Boston gave Congressional testimony about the risk of false positive HIV test results. 26 He testified:
Dr. Pauker used the example of Peoria, a community he expected to have a true HIV prevalence of only 1 infection per 10,000 persons. After studying various laboratory proficiency studies, he and his colleagues concluded that the false positive rate for the ELISA/Western Blot testing procedure was 1 in 300 and 1 in 98 for samples in participating laboratories. Another survey found the joint false positive rate to be 1 in 25. He concluded that widespread HIV antibody-testing in a community like Peoria would result in 400 false positives for every truly infected person identified.We come before you today to argue against the widespread use of serologic tests for evidence of HIV infection in so-called low risk populations. We firmly believe that widespread screening for such infection, as has been proposed for immigrants, marriage license applicants, and, at least in one state, members of the general population, is unwise; it may in fact lead to a social catastrophe.In 1998 the Journal of the American Medical Association reported a study of HIV antibody test result which found that 4.8% of the HIV antibody positive results among blood donors were false. They concluded that from 48 to 56 blood donors are annually victimized by false positives. 27For blood donor victims, this indicates a total number between 816 and 952 since the HIV screening program began in 1985.
CDC Denial of False Positive Hazard
The blindness of the CDC to the false positive problem was illustrated in a FAX dated 1/5/95 and sent to me by Wanda Jones, the agency's Acting Associate Director for Women's Health. In response to my concerns about false positives, she wrote:
In plain language, Jones' use of the term 0.9999999 indicates that she believes only one in 10 million positive test results will be false. The false positive rate of 4.8% reported in JAMA for blood donors was 480,000 times greater. In her FAX, Jones stated: "HIV antibody testing, when done according to published algorithms, is virtually perfect."The predictive value of a positive HIV test (EIA and Western blot) is 0.9999999, or virtually 100%. If 38 million people were tested, and assuming no duplicate samples, then 31 people may have received false positive results.Jones' estimate that 31 people "may have received false positive results" is nonsensical. As commonly used by epidemiologists, the phrase positive predictive value ordinarily means the proportion of positive test results which are true positives. Since she made no claim about the number of positive results obtained from 38 million tests, her estimate of 31 possible false positives has no basis.
Jones' statement also was in conflict with a CDC claim quoted by USA Today during the month before her FAX. 28 The article stated that under circumstances the CDC considers to be "ideal," the odds of a false positive HIV antibody test are less than 1 in 20,000. The same claim was made by the CDC in a March 1995 Associated Press article about my project.29 A rate of 1 in 20,000 would be 500 times greater than Jones' estimate.
In the AP story, Jones was reported to have "agreed that false positives are a concern." Then she stated that the HIV antibody testing procedure is "probably the closest thing to perfect of any medical diagnostic test on the market." She also claimed that "virtually 100 percent" of the positive results are true positives.
More evidence of the CDC's inability to keep its story straight about the false positive hazard emerges in one of its booklet Voluntary HIV Counseling and Testing. 30 In this publication the agency claims that the testing procedure is "more than 99.9 percent accurate." The differing claims made by CDC spokesmen and publications are tabulated below:
The comparisons indicated by the table might be problematic, due to vagueness about the meaning of the measures. For precision, it is necessary to specify whether the claim of 1 in 20,000 refers to 20,000 positive results or 20,000 tests performed. The CDC source quoted by USA Today did not clarify this point. If the number 20,000 is for number of tests, assuming a three percent positivity rate would lead to the still questionable claim that only 1 in 600 positive results would be false.Differing CDC Statements About False Positive HazardSource____________________________Estimate__
CDC booklet...........................................less than 1 in 1,000
CDC claim reported by USA Today.........less than 1 in 20,000
CDC claim made by Wanda Jones...........1 in 10,000,000A more realistic picture is offered by the false positive rate of 4.8% (1 in 21) reported in JAMA for blood donors. There is reason to doubt that the true rate is that low. Christine Johnson has compiled a bibliography of sixty-four scientific articles documenting dozens of factors implicated for triggering false positive HIV test reactions. Following a careful review of scientific literature, Johnson concluded that the validity of the HIV antibody-testing procedure has never been assessed by research involving confirmation of the presence of HIV by means of viral isolation.31 Her finding was consistent with a 1986 statement by the editors of the Annals of Internal Medicine, who wrote: 32
Neither the Western Blot nor any other assay to confirm HIV infection has been rigorously evaluated for sensitivity, specificity, or predictive value.Go To Part Eight
References
1. Go here for a governnment website (NCI/NIH) which lists me as a grant recipient for multimedia technology and health communication. (The email address there is no longer current).
2. Richard Havel and Michael Wright, "Automated Interviewing for Hepatitis B Risk Assessment and Vaccination Referral," American Journal of Preventive Medicine, 13, no. 5 (1997): 392-395
3. Fineberg, H. "Education to Prevent AIDS: Prospects and Obstacles," Science 1988; 239:592.
4. Norman Hearst and Steven Hulley, "Preventing the Heterosexual Spread of AIDS: Are We Giving Our Patients the Best Advice?" Journal of the American Medical Association, 259, no. 16 (1988): 2428-2432.
5. Abstract THP.3-48:171, presented at the Third International AIDS Conference, Washington, DC, June 1987.
6. CDC, Morbidity and Mortality Weekly Report, December 18, 1987, Vol 36/No. S-6, Table 14.
7. For more references to articles about the extremely low risk of vaginal HIV transmission, see my website article, "Manufacturing the AIDS Scare."
8. An excellent journalistic expose of the CDC's campaign of deceit was written by Amanda Bennet and Anita Sharpe. See "AIDS Fight Is Skewed by Federal Campaign Exaggerating Risks," Wall Street Journal, May 1, 1996. The writers were awarded a Pulitzer Prize .
9. Gina Kolata, "New Picture of Who Will Get AIDS Is Crammed With Addicts," The New York Times (February 28, 1995), p. B6.
10. Daniel Schneider, E. Laurette Taylor, Larry Prater, and Michael Wright, "Risk Assessment for HIV Infection: Validation Study of a Computer-Assisted Preliminary Screen," AIDS Education and Prevention, 3, no. 3 (1991): 215-229
11. Randolph Miller, "Medical Diagnostic Decision Support Systems -- Past, Present, and Future," Journal of the American Medical Informatics Association 1, no. 1 (1994): 15. See also reference note 299.
12. Michael Wright, E. Laurette Taylor, Larry Prater, "Accuracy and Yield of Expert System Software for Personal HIV Risk Assessment," Proceedings of the 7th Oklahoma Symposium on Artificial Intelligence, November 18-19, Oklahoma State University (Stillwater, Oklahoma).
13. Christine Johnson, Whose Antibodies Are They Anyway", Continuum, Sept/Oct 1996.
14. Michael Wright, Larry Prater, E. Laurette Taylor, Carrie Zeman, and Deborah Trott, "Automated Telephone Screening for Risk of Current HIV Infection" (Unpublished paper provided to the National Cancer Institute)
15. H. Angle, "Computer Interview Support for Clinicians," Professional Psychology 10 (1979): 49.
16. Seventh Special Report to the U.S. Congress on Alcohol and Health (National Institute on Alcohol Abuse and Alcoholism, Department of Health and Human Services), p. 300.
17. DE Beauchamp, "Morality and the Health of the Body Politic," Hastings Center Report 16, no. 6 (1986): 30.
18. CL Chng and W Roddy, "Ethical Implications: Screening for and Treatment of AIDS," Health Education 18, no. 4 (1987): 4.
19. 19. J Joseph et al., "Coping with the Threat of AIDS," American Psychologist 39 (1984): 1297.
20. The ODH was not eager to release the data about cost and yield of its testing program. Dr. Prater had to threaten a lawsuit under the Open Records Act in order to obtain it.
21. JE Anderson, et al., ""HIV Antibody Testing and Post-Test Counseling in the United States: Data from the 1989 National Health Interview Survey," American Journal of Public Health, 82 (1992): 1533-1535.
22. CDC, "HIV Prevalence Estimates and AIDS Case Projections for the United States," Morbidity and Mortality Weekly Report , 39, RR16 (1990):5.
23. CDC, "Publicly Funded HIV Counseling and Testing United States, 1985-1989," Morbidity and Mortality Weekly Report , 39 (1990): 137-140.
24. CDC, "Publicly Funded HIV Counseling and Testing -- United States, 1991," Morbidity and Mortality Weekly Report , 41 (1992): 613-617.
25. Theodore Hammett, "HIV Antibody Testing: Procedures, Interpretation, and Reliability of Results," AIDS Bulletin, October 1988 (National Institute of Justice/U.S. Department of Justice)
26. Quality AIDS Testing, Congressional Hearing No. 100-32, First Session, 100th Congress, October 19, 1987, p. 28-29.
27. Steven Kleinman, et al., "False-Positive HIV-1 Test Results in a Low-Risk Screening Setting of Voluntary Blood Donation," Journal of the American Medical Association 280 (1998):1080-1085.
28. Doug Levy, "The Lingering Pain of False Positive Diagnosis," USA Today, December 12, 1994.
29. Rochelle Hines, "Computer Assesses AIDS Risk," Associated Press, March 19, 1995.
30. CDC, Voluntary HIV Counseling and Testing: Facts, Issues, and Answers [NAIEP/10-90/11]
31. Christine Johnson, "Understanding the HIV Antibody Test," Journal of International Health Research, no. 1 (May 1994):6-13.
32. "Testing for Human Immunodeficiency Virus," [editorial], Annals of Internal Medicine, 105, no. 4 (1986): 609.