Sow, Papa Salif MD, MS*; Hawes, Stephen E PHD†§; Critchlow, Cathy W PHD†; McIntosh, Martin W PHD‡¶; Diop, Aissatou MD*; Diouf, Mame B MD*; Gottlieb, Geoffrey S MD, PHD∥; Starling, Alison K BA§; Coll-Seck, Awa M MD*; Kiviat, Nancy B MD§
While sub-Saharan Africa is characterized by the highest HIV-1 and HIV-2 incidence and prevalence rates in the world, resources for HIV screening and treatment of HIV-associated diseases are extremely limited.1,2 The systematic screening, identification, counseling, and treatment of HIV-infected persons are crucial to achieve eventual control of HIV transmission.3 However, in many sub-Saharan countries, to best use limited available resources, priority is given to screening pregnant women, to prevent vertical transmission,4-8 and to men and women engaging in high-risk sexual behaviors.9-12 In areas of endemic HIV infection, screening of groups at potentially higher risk of HIV infection as compared with the general population, eg, patients presenting to certain outpatient medical clinics, is also clearly desirable. Unfortunately, systematic screening of such individuals is often limited by fiscal constraints and the fact that many patients, unaware of their infection risk, do not agree to HIV screening. Often, medical personnel first identify patients who appear to be at high risk of HIV infection and then attempt to convince them of the value of HIV testing. Decisions for HIV testing have generally been based on presence of symptoms and demographic or behavioral characteristics previously found associated with HIV-1 infection among hospitalized patients. While this selection strategy is widely used throughout Africa, there has been no systematic approach to choosing the specific demographic or behavioral characteristics or symptoms upon which to base patient selection. Further, the utility of such an approach in Africa has never been examined, either for identifying HIV-infected subjects among general outpatient medical clinic populations or in populations where HIV-2 infection is also present.
The present study was undertaken to assess, in an outpatient medical clinic in a resource-poor setting (Senegal, West Africa), the utility of selecting patients for HIV testing according to criteria based on presenting complaint and demographic and behavioral characteristics. Our approach was to first determine specific demographic and behavioral characteristics, as well as presenting complaints, that were associated with HIV-1 or HIV-2 infection among outpatients presenting to the University of Dakar Infectious Disease Clinic. To decide which of these factors would best predict which patients would most benefit from HIV testing, we used logistic regression analyses to estimate a composite screening rule that maximized sensitivity (the proportion of patients who would be tested among those who are truly infected) for any arbitrarily selected specificity (the proportion of patients who would not be tested among those who are truly not infected).
Study Population and Study Design
Between October 1994 and January 1998, as part of screening for research studies regarding oral manifestations of HIV infection, all patients between 18-65 years of age presenting to the University of Dakar Fann Hospital Infectious Disease Clinic were offered free HIV screening and a dental/oral examination. In addition, women were offered free Papanicolaou screening as part of a study assessing interrelationships between HIV infection, detection of human papillomavirus in the cervix, and cervical neoplasia, as previously described.13 To increase study enrollment, recruitment for screening was expanded to other hospital services. Participants provided oral informed consent according to procedures approved by the University of Washington Human Subjects Review Committee and the Senegalese National HIV Committee. Blood was drawn from all subjects for HIV serology testing, and cervical samples were obtained from women for cytology screening. All patients were asked to return in 2 weeks, and those found to be HIV seropositive were counseled according to recommendations of the Senegalese National HIV Committee. All women in whom cervical intraepithelial neoplasia grade 2-3 or invasive cervical cancer was diagnosed, and all men and women with acute dental problems, were immediately treated by the study clinician (or dentist) or referred for treatment.
At the time of screening, all study participants completed a short standardized interview, including an inquiry as to the reason for visit to the clinic as well as questions regarding general demographic information and medical, gynecologic, and sexual history. In addition, subjects underwent physical, dental, and gynecologic examination. Specific reasons for the clinic visit were abstracted from data obtained at the screening visit, and the most prevalent responses (≥20 occurrences) were noted. For this analysis, other less frequently noted responses (<20 occurrences) were categorized as “other specified reasons.” General responses, such as “check-up,” were categorized as “unspecified.” If reason for visit was unavailable (which occurred for <5% of those screened), the response was categorized as “no answer.”
All specimens were first screened for anti-HIV antibodies using an HIV-1/HIV-2 enzyme immunoassay (HIV 1/2 EIA, Sanofi Diagnostics Pasteur, Redmond, WA). Positive samples were confirmed using a rapid, HIV peptide-based membrane immunoassay that distinguishes between antibodies to HIV-1 and HIV-2 (Multispot, Genetic Systems, Redmond, WA).
Pearson χ2 tests were performed to compare patients with and without HIV infection with respect to categorical variables, such as marital status, while Mantel-Haenszel tests for trend assessed differences between groups in ordinal variables, such as categorized age. For analyses performed to create the screening rules, mutually exclusive dichotomous dummy variables (coded as 0 = “no” and 1 = “yes”) were created for each level of categorical variables, such as marital status and reason for study visit. Multivariable logistic regression analyses were then performed to evaluate the independent effects of specific demographic and behavioral factors on risk of HIV infection and to construct screening rules to identify subjects for HIV screening.14 Six separate logistic regression models (3 models for each gender) were constructed using forward stepwise selection procedures. First, a model using only the reason for clinic visit was determined. Second, a model using only predictive demographic and behavioral information was created. Lastly, a model combining the significant demographic and clinic visit covariates to the reason for clinic visit was determined.
For each of the 6 logistic models, a composite measure (CM) of HIV risk was computed for each subject,
Equation (Uncited)Image Tools
where the βj are the regression coefficients (from the logistic regression model) corresponding to the Xj, which represent values of the demographic and clinic visit variables included in the regression. In the 2 models including only reason for clinic visit (one for each gender), this CM is based simply on the probability of HIV infection among all subjects with that specific reason for clinic visit, since the reasons are mutually exclusive and each subject can report only one reason. This results in a step-function when sensitivities and specificities are plotted. However, in the models including important demographic and behavioral variables, the CM is computed based upon the values of all variables (eg, age, birthplace, marital status) in the model.
Because subjects with higher CM values are more likely to be HIV infected, the CM may be used to define a screening rule. Given an inability to test everyone, one would choose to perform HIV testing for patients whose CM exceeds some value c. A screening rule that has an arbitrary specificity y can be defined by selecting c such that c = the yth percentile among HIV-negative subjects. Similarly, a screening rule that has an arbitrary sensitivity x can be defined by selecting c such that c = the xth percentile among HIV-positive subjects. Each different choice of c yields a unique sensitivity and specificity, which when plotted, yields a receiver operating characteristic (ROC) curve. Sensitivities and specificities for HIV positivity were plotted for the CMs established for each of the 6 regression models. A good screening rule maximizes both specificity and sensitivity. Furthermore, it has been shown that screening rules formed by logistic (or other binary) regression methods maximize sensitivity for all specificities simultaneously.14 All analyses were conducted using SAS 8.0 (SAS Institute, Inc., Cary, NC).
Presenting Complaints Among Attendees of the Infectious Disease Outpatient Clinic
Overall, 4348 women and 1368 men were tested at the University of Dakar Infectious Disease Outpatient Clinic; <5% refused HIV testing. Stated reasons for the clinic visit among women included family planning (26%), general gynecologic problems (17%), perceived infertility (8%), vaginal discharge or pruritis (9%), genital infection (8%), abnormal bleeding (2%), referral for suspected tuberculosis (3%), known HIV disease in the patient (4%) or spouse (2%), and other specified (12%) or unspecified (9%) complaints. Men presented because of fever (9%), fever with productive cough (11%), diarrhea (11%) or other nondiarrheal intestinal problem (3%), urethral discharge (5%), referral for suspected tuberculosis (5%), symptoms of malaria such as acute onset of fever and headache (4%), known HIV disease either in the patient (15%) or spouse (2%), other nongenital infection (2%), mental problems such as anxiety (3%), unspecified pain (3%), and other specified (22%) or unspecified (9%) symptoms. Gender differences in presenting complaints are consistent with the populations selected for HIV testing; men and women presenting to the outpatient infectious disease clinic were tested, while women presenting to other services for non-infectious disease-related reasons, such as family planning and gynecologic problems, were also tested.
Presenting Complaints Associated With HIV Infection Among Patients Who Had Not Been Previously Tested for HIV Infection
HIV antibody was detected in 275 (6.6%) of 4157 women (4.9% HIV-1, 1.2% HIV-2, 0.5% dual infection) and 350 (30.2%) of 1161 men (23.9% HIV-1, 3.0% HIV-2, 3.2% dual infection) who had not been previously tested for HIV infection. Among women, antibody was detected in only 1.2% presenting for family planning, 2.5% of women presenting for gynecologic examinations, 3.9% of women complaining of vaginal discharge or pruritis, 4.6% of women with pelvic or abdominal pain, 1.2% of women presenting for other possible genital infection, and 2.7% of those seeking help for infertility (Table 1). Although only 5.7% of women presented with classic AIDS-associated symptoms, HIV was detected most often among such women, being detected in 13% of those referred for suspected tuberculosis; 39% of those presenting with complaints suggestive of pneumonia, bronchitis, or cough; 33% of women presenting for diarrhea; 46% of those presenting with fever; and 36% of women presenting because of weight loss. While women with these traditionally AIDS-associated symptoms were likely to HIV infected, only 22% of all HIV-positive women presented with these complaints. While only 2.1% of women presented because of an HIV-infected spouse, 67% of such women were found to be HIV positive; this presenting complaint accounted for 21.5% of all HIV-infected women.
In contrast to women, many men (32%) presented with signs and symptoms of AIDS-associated illnesses. These complaints were also highly associated with HIV infection, with HIV detected among 56% of men presenting with suspected tuberculosis; 32% of men with pneumonia, bronchitis, or cough; 59% of men with diarrhea; and 35% of men with fever. In contrast, men presenting for check-ups for an unspecified reason, malaria, or abdominal pain had low HIV rates (3.7, 2.0, and 0.0%, respectively). Only 4.3% of men found to be HIV positive presented because of an HIV diagnosis in a spouse.
Demographic Characteristics Associated With HIV-1 or HIV-2 Infection in Patients Who Had Not Been Previously Tested for HIV
HIV-negative and HIV-positive women were similar with respect to age, gravidity, use of tobacco or alcohol, religion, and ethnicity (Table 2). Further, HIV prevalence was similar between women in a monogamous as compared with polygamous marriage. However, HIV infection was strongly associated with marital status, commercial sex work, lack of contraception use, low education level, and birth place or travel outside of Senegal. In multivariable logistic regression analysis, we found HIV to be independently associated with being divorced or separated or widowed (compared with having a monogamous marriage), commercial sex work, and birth outside of Senegal (data not shown). Compared with women not practicing contraception, women using oral, injected, or “traditional” contraceptive methods had a lower HIV infection rate. Women with a secondary or higher level of education had lower HIV rates than did women without any formal education.
Among men, HIV infection was correlated with older age, marital status, and level of education (Table 2). We found, using multivariable logistic regression analysis, men aged 30 years or older were more likely to be HIV infected than were men younger than 30 (data not shown). Compared with monogamously married men, single men had a lower HIV rate and widowed men had a somewhat higher HIV rate. Similar to findings among women, higher educational attainment was associated with reduced rates of HIV infection, as men with a secondary or higher level of education had lower rates of HIV infection compared with men without any formal education. Men traveling outside of Senegal were more likely to be HIV positive than men who had not traveled outside of Senegal.
Estimation of a Composite Screening Rule for Selection of Patients for HIV Testing
Three logistic regression analyses were then performed separately for men and women, using reason for clinic visit and demographic characteristics separately, and then combined, as potential independent variables. In each model, forward stepwise selection procedures were used to identify demographic factors and presenting complaints that best predicted infection with HIV. The regression coefficients from the logistic regression analyses were used to estimate the likelihood of HIV infection for each subject based upon the values of individual covariates. Each model, ie, each combination of covariates (and threshold value c, see “Methods”), defined a unique screening rule. The sensitivity and specificity of each screening rule were calculated by comparing the predicted to the actual number of HIV-positive and negative subjects. The performances of all 3 screening rules in each gender were summarized in ROC curves plotting the false-positive rate (1 − specificity) vs. sensitivity. Screening rules were first computed using “reason for visit” and demographic characteristics separately (solid lines and dashed curves, Fig. 1A and B). Subsequently, we assessed a rule that combined specific patient demographic characteristics with presenting complaint (dotted curves, Fig. 1A and B).
The sensitivity and specificity associated with using a single categorized variable, “reason for visit,” as the basis for selecting women and men for HIV testing are presented in Table 3A and B, respectively. The resulting composite rule is based solely on the prevalence of HIV among patients presenting for each specific reason, since the reasons are mutually exclusive. Not surprisingly, for each gender, the screening strategies based upon reason for visit begin first with screening those who we know have the highest HIV prevalence (those who present because their spouse is HIV positive; strategy B), and then add groups in order of decreasing CM values (predicted probability of being HIV positive; Table 3A and B). For example, the approach designated “L” (Table 3A), which tests all women except those presenting for genital infection, family planning, or gynecologic or fertility problems, would identify 83% of all HIV-infected women by testing only 35% of women presenting to the clinic.
In the second model, we used the demographic and behavioral factors that were significantly associated with HIV status (from a stepwise forward logistic regression) to form the basis for selection for HIV testing. In women, these factors included age, marital status, commercial sex work, birth control method, birth outside of Senegal, and the number of children. In men, age, marital status, smoking, and alcohol use were the factors used to create the HIV testing criteria. Based upon these 2 models (one each for women and men), the probability of HIV infection for each individual can be calculated from the logistic regression model coefficients, with individuals then ranked from highest to lowest probability of HIV infection. For example, from Table 2, we can see that a widowed commercial sex worker born outside of Senegal would be at high risk for HIV infection, while an educated, monogamously married, non-commercial sex worker who used contraception would be at low risk for HIV infection. With this strategy, as with that based on reason for visit, selection for HIV testing starts with individuals with the highest predicted probability of infection, then proceeding to other subjects in order of decreasing predicted HIV risk. In women, across a wide range of specificities, testing based upon demographic information alone (dashed line) results in a significantly lower number of HIV infections identified compared with a testing strategy that includes the specific reason for clinic visit (Fig. 1A). For example, the testing of 35% of women identified by a strategy utilizing demographic information alone potentially identifies 68% of HIV-positive women, while using the single specific reason for visit and testing 35% of women identifies 83% of HIV-positive women. However, in men, testing based upon demographic information alone (dashed line) identifies similar numbers of HIV-infected subjects compared with testing based upon reason for clinic visit (Fig. 1B). For example, testing 40% of men using demographic information alone potentially identifies 74% of men with HIV infection, compared with 78% when a strategy using the single specific reason for visit is used.
In the final models, demographic information and the reason for clinic visit were both included to form the basis for the HIV testing strategy (dotted lines in Fig. 1A and B). In women, the similarity between the dotted and solid curves suggests that selection for HIV testing in female patients based upon reason for visit provides nearly equivalent performance to a rule adding information regarding demographic characteristics. Thus, among women presenting to our clinics, factors such as age, marital status, birth control method, number of children, place of birth, and commercial sex work added little, in terms of identifying additional women that would benefit from HIV testing, to a testing strategy based only upon reason for visit. Among men, however, across all levels of specificity, a screening scenario that includes both reason for visit and demographic information (age, marital status, smoking, and alcohol use) provides additional useful information compared with strategies employing either demographic characteristics or reason for visit alone, as the numbers of HIV-infected subjects identified using the former strategy (dotted curve, Fig. 1B) are substantially higher than those identified by rules using reason for visit or demographic information alone (solid and dashed curves, Fig. 1B). Using the combined information, testing 40% of men potentially increases the sensitivity of identifying HIV-infected men to 84%.
Selecting a subset of patients for HIV testing, from among all patients presenting for care, is a common practice in sub-Saharan Africa. While it is widely accepted that HIV screening should be offered to all at-risk persons living in high HIV prevalence areas, this is not yet feasible in settings with limited resources (materials, infrastructure, personnel) or in populations where patients do not readily agree to HIV testing. Among men and women presenting to a large West African outpatient clinic who were not known to be HIV positive, we developed an optimal rule for HIV testing by determining factors that were highly associated with HIV-1 or HIV-2 infection and then using a simple statistical approach to identify the specific combination of factors providing the most sensitive and specific screening approach.
In our study population, demographic characteristics associated with HIV infection included divorced or separated marital status, birthplace outside of Senegal, commercial sex work, and lower educational attainment among women, and older age, lack of education, and travel outside of Senegal among men. Similar findings with respect to age and marital status have been described in Zimbabwe.15,16 Among pregnant women in Senegal, Abbott et al5 found age and marital status to be associated with HIV infection, in addition to parity and birthplace in Guinea-Bissau.
At our clinic, the primary reasons that HIV-infected persons sought care included having a spouse with HIV and presence of symptoms classically associated with HIV infection, such as weight loss, diarrhea, and referral for suspected tuberculosis. Previous studies conducted in Africa evaluating presenting complaints among HIV-1-infected patients, have, for the most part, been limited to hospitalized rather than nonhospitalized patients. In such patients, the most frequent reasons for hospitalization were HIV-associated symptoms such as fever, diarrhea, wasting syndrome, cough, and nonhealing genital ulcers.17-21 Common diagnoses among these patients included septicemia, meningitis, herpes zoster, and generalized lymphadenopathy. In a Senegalese study of hospitalized patients, those with HIV-1 compared with HIV-2 had similar signs and symptoms, although patients with HIV-2 were more likely to have diarrhea, and those with HIV-1 were more likely to have fever or oral candidiasis.18 In one of the few published studies among outpatients, Baboo et al22 reported that 52% of adults presenting to a Zambian community-based health center with acute diarrhea were HIV-1 positive.
We used logistic regression analyses to estimate, for each presenting reason for visit and various demographic and behavioral characteristics, an independent probability of testing HIV seropositive; thus, each subject could be assigned a risk score based upon these characteristics. Various screening rules for HIV testing of subjects can be established by varying the risk score cutpoint (ie, only test subjects with scores greater than the cutpoint or, equivalently, those with the highest predicted risk of HIV) were evaluated and compared with testing all subjects, and ROC curves summarizing the predicted sensitivity and specificity of the various risk-based strategies for selecting patients for HIV testing were generated. Screening algorithms based upon presenting symptoms or demographic and behavioral characteristics in hospitalized African patients have been described, with sensitivities ranging between 66-78%, and specificities ranging between 46-82%; however, these studies have generally been small (<400 subjects).23-25 This study is the first to describe the use of CMs to define screening rules for selecting subsets of patients for HIV testing. In our clinic, among women, the most sensitive and specific algorithm for HIV testing would be based upon the stated reason for clinic visit, as the addition of demographic and behavioral characteristics did not appreciably increase sensitivity. Although HIV prevalence among women was 6.6%, we could, by simply using presenting complaint, test only 35% of the population and identify 83% of all HIV-infected women. Thus, choosing to not test the 2560 women with presenting complaints associated with very low HIV rates (family planning, genital infection, gynecologic or fertility problems), and limiting testing to the remaining 1597 women, would have allowed us to identify 229 women with HIV, while missing 46 HIV-infected women. Among men (of whom 30.2% were HIV seropositive), by using primary complaint and demographic characteristics, our CM identified 84% of HIV-infected men by screening approximately 40% of men presenting to the clinic. Thus, using our strategy, we could identify at least 80% of both men and women infected with HIV by testing <40% of those presenting to the clinic, even though HIV seroprevalence was low among women and high among men.
Senegal has historically been characterized by a low prevalence of HIV infection in comparison to other nations in sub-Saharan Africa, with an estimated population of <2%, and until the 1990s, HIV-2 was the predominant HIV type.26,27 During the study period, antiretrovial medications for HIV were not available to study participants. However, treatment of opportunistic infections associated with HIV infections was given to study participants, and nutritional support and HIV counseling were offered to all subjects at the clinic, regardless of HIV status. At the present time, identification of HIV-infected individuals has become even more important, since infected individuals are eligible for enrollment into the Senegalese government’s Antiretroviral Drug Access Initiative, which was launched in 1998 and provides a subsidized, comprehensive treatment program for HIV-infected individuals.28,29
In summary, we have described the development and utility of a screening rule to identify, from among patients presenting to a general outpatient infectious disease clinic, those who would be most likely to test HIV seropositive. In the absence of universal HIV screening, use of a simple screening rule results in identifying a maximum number of HIV-infected patients while minimizing the number of patients tested. A similar screening strategy could be developed and applied in other settings where health care resources are limited. Clearly, the specific demographic or behavioral characteristics or presenting complaints associated with HIV infection, as well as the sensitivities and specificities achieved from various screening rules, may vary according to the prevalence of HIV in the population of patients presenting to a specific clinic and the types of illnesses with which patients typically present. However, our study demonstrates how construction of a screening rule for selection of patients for HIV serologic testing can provide a rational approach to resource allocation.
The authors thank Deana Rich, Elise Reay-Ellers, and Macoumba Touré for their invaluable coordination of study procedures in Senegal; Mame Dieumbe Mbengue-Ly, Marie Pierre Sy, and Dr. Pierre Ndiaye for patient care; Diouana Ba and Haby Agne for their leadership in the laboratory; and Fatou Faye-Diop for data entry.
1. Report on the global HIV/AIDS epidemic, June 2000. Joint United Nations Programme on HIV/AIDS (UNAIDS, Geneva, Switzerland); 2000.
2. Grant AD, De Cock KM. ABC of AIDS: HIV infection and AIDS in the developing world. BMJ. 2001;322:1475-1478.
3. Plummer FA, Nagelkerke NJ, Moses S, et al. The importance of core groups in the epidemiology and control of HIV-1 infection. AIDS. 1991;5(Suppl 1):S169-S176.
4. Dunn DT, Newell ML, Ades AE, et al. Risk of human immunodeficiency virus type 1 transmission through breastfeeding. Lancet. 1992;340:585-588.
5. Abbott RC, NDour-Sarr A, Diouf A, et al. Risk factors for HIV-1 and HIV-2 infection in pregnant women in Dakar, Senegal. J Acquir Immune Defic Syndr. 1994;7:711-717.
6. Shaffer N, Chuachoowong R, Mock PA, et al. Short-course zidovudine for perinatal HIV-1 transmission in Bangkok, Thailand: a randomised controlled trial. Bangkok Collaborative Perinatal HIV Transmission Study Group. Lancet. 1999;353:773-780.
7. Wiktor SZ, Ekpini E, Karon JM, et al. Short-course oral zidovudine for prevention of mother-to-child transmission of HIV-1 in Abidjan, Cote d’Ivoire: a randomised trial. Lancet. 1999;353:781-785.
8. Ayisi JG, van Eijk AM, ter Kuile FO, et al. Risk factors for HIV infection among asymptomatic pregnant women attending an antenatal clinic in western Kenya. Int J STD AIDS. 2000;11:393-401.
9. Quinn TC, Mann JM, Curran JW, et al. AIDS in Africa: an epidemiologic paradigm. Science. 1986;234:955-963.
10. Kreiss JK, Koech D, Plummer FA, et al. AIDS virus infection in Nairobi prostitutes: spread of the epidemic to East Africa. N Engl J Med. 1986;314:414-418.
11. Laga M, Alary M, Nzila N, et al. Condom promotion, sexually transmitted diseases treatment, and declining incidence of HIV-1 infection in female Zairian sex workers. Lancet. 1994;344:246-248.
12. Ghys PD, Diallo MO, Ettiegne-Traore V, et al. Effect of interventions to control sexually transmitted disease on the incidence of HIV infection in female sex workers. AIDS. 2001;15:1421-1431.
13. Hawes SE, Critchlow CW, Faye-Niang MA, et al. Increased risk of high-grade cervical squamous intraepithelial lesions and invasive cervical cancer among African women with human immunodeficiency virus type 1 and 2 infections. J Infect Dis. 2003;188:555-563.
14. McIntosh MW, Pepe MS. Combining diagnostic and screening markers: optimality of the risk score. Biometrics. 2002;58:657-664.
15. Mbizvo MT, Latif AS, Machekano R, et al. HIV seroconversion among factory workers in Harare: who is getting newly infected? Cent Afr J Med. 1997;43:135-139.
16. Gregson S, Mason PR, Garnett GP, et al. A rural HIV epidemic in Zimbabwe? Findings from a population-based survey. Int J STD AIDS. 2001;12:189-196.
17. Kwesigabo G, Killewo JZ, Sandstrom A, et al. Prevalence of HIV infection among hospital patients in north west Tanzania. AIDS Care. 1999;11:87-93.
18. Ndour M, Sow PS, Coll-Seck AM, et al. AIDS caused by HIV1 and HIV2 infection: are there clinical differences? Results of AIDS surveillance 1986-1997 at Fann Hospital in Dakar, Senegal. Trop Med Int Health. 2000;5:687-691.
19. Grant AD, Djomand G, Smets P, et al. Profound immunosuppression across the spectrum of opportunistic disease among hospitalized HIV-infected adults in Abidjan, Cote d’Ivoire. AIDS. 1997;11:1357-1364.
20. Edhonu-Elyetu Y. Significance of herpes zoster in HIV/AIDS in Kweneng district, Botswana. East Afr Med J. 1998;75:379-381.
21. Chintu C, Dupont HL, Kaile T, et al. Human immunodeficiency virus-associated diarrhea and wasting in Zambia: selected risk factors and clinical associations. Am J Trop Med Hyg. 1998;59:38-41.
22. Baboo KS, Luo NP, Murphy JR, et al. HIV-1 seroprevalence in Zambian patients with acute diarrhea: a community-based study. J Acquir Immune Defic Syndr Hum Retrovirol. 1999;20:160-163.
23. Miller WC, Thielman NM, Swai N, et al. Diagnosis and screening of HIV/AIDS using clinical criteria in Tanzanian adults. J Acquir Immune Defic Syndr Hum Retrovirol. 1995;9:408-414.
24. Palmer DL, Mason PR, Pasi C, et al. Value of mandatory testing for human immunodeficiency virus in a sub-Saharan hospital population. Clin Infect Dis. 2000;31:1258-1265.
25. Willoughby VR, Sahr F, Russell JB, et al. The usefulness of defined clinical features in the diagnosis of HIV/AIDS infection in Sierra Leone. Cell Mol Biol. 2001;47:1163-1167.
26. Meda N, Ndoye I, M’Boup S, et al. Low and stable HIV infection rates in Senegal: natural course of the epidemic or evidence for success of prevention? AIDS. 1999;13:1397-1405.
27. Senegal Ministry of Health. Senegal: UNAIDS/WHO Epidemiological Fact Sheet on HIV/AIDS and Sexually Transmitted Infections: 2002 Update.
28. Laurent C, Diakhate N, Ngom-Gyeye NF, et al. The Senegalese government’s highly active antiretroviral therapy initiative: an 18-month follow-up study. AIDS. 2002;16:1363-1370.
29. Desclaux A, Ciss M, Tavern B, et al. Access to antiretroviral drugs and AIDS management in Senegal. AIDS. 2003;17:S95-S101.
© 2004 Lippincott Williams & Wilkins, Inc.