Thirty-eight million people are estimated to be living with HIV/AIDS; over 20 million have died of AIDS since the beginning of the epidemic. Accurately estimating the number of people infected with HIV is crucial for the purposes of advocacy, program planning, and evaluation, as is determining the factors associated with being HIV-positive. The need for such efforts is critical in sub-Saharan Africa, which bears two thirds of the global burden of HIV infection.1
One of the first African countries affected by the HIV/AIDS epidemic, Kenya's first case of AIDS was officially recognized in 1984. Kenya is understood to have a severe, generalized epidemic with approximately 1.4 million people infected, 100,000 of whom are children.2 This article focuses on the results of the 2003 Kenya Demographic and Health Survey (KDHS), the first national HIV seroprevalence survey that maintains the link between HIV test results and a rich set of demographic and health data. We report on national HIV seroprevalence in Kenya and assess key variables for their association with HIV serostatus at the individual level. Because the data are cross-sectional, the results presented here reflect only associations, and not causal relationships.
For the 2003 Kenya DHS, a representative probability sample of 9865 households was selected using a 2-stage sample design. The first stage selected sample points from a national master sample maintained by Kenya's Central Bureau of Statistics; the second stage involved the systematic sampling of a household list of the Central Bureau of Statistics updated in June 2003. Ninety-six percent of eligible households responded to the KDHS. Ninety-four percent of all eligible women had a completed interview. A 50% subsample of households was selected in which men were interviewed; the men's response rate was 86%. In that same 50% subsample, all eligible men and women in the selected households were asked to give their informed consent to be anonymously tested for HIV. Samples for testing were obtained by collecting blood drops from a sterile fingerstick onto a filter paper card.* Among the women eligible for HIV testing (n = 4041), 14% refused to give a sample of blood, 3% were absent, and a blood sample was not available for 2% for other reasons. Among the men eligible for testing (n = 3578), corresponding nonresponse proportions were 12%, 4%, and 2%.†
This analysis looks at men's and women's risk factors separately because the biological and social circumstances associated with transmission of HIV differ by sex. The prevalence of HIV is comparatively low among those reporting that they have never had sex (1-2%), and many of the risk factors included in the models are related to sexual behavior; therefore, respondents who have never had sex are omitted from the analysis.
χ2 tests of independence are conducted for the bivariate analysis. Multivariate logistic regression is then used to discern key risk factors associated with being HIV-positive. Four regression models are presented for each analysis, with variables being added to each subsequent model in conceptual groups. As variables of interest lost significance upon the addition of new variables, additional models were run to determine the cause of the change in significance; these results are reported but not shown in tables.
Three key variables (use of condom in the past year, exchange of sex for goods or money, and number of partners in the past year) had a common response category (did not have sex in the past year), which is problematic for multivariate modeling. Although it is preferable to incorporate variables that reflect current-status behaviors, we modified the time period reflected by 2 of the 3 variables to keep the conceptual substance of those variables, rather than dropping them from the analysis. Each variable was independently tested for its relationship to HIV using both bivariate and multivariate approaches. Based on the significance of the associations and changes in model fit, the variables reflecting the number of partners in the past year and ever-use of condoms were included, and the variable reflecting exchange of sex for goods or money was omitted. Variables that were likely to be correlated (such as wealth and education) were assessed using Kendall's Tau-b; no relationships were highly correlated (> 0.500).
There were variables in the multivariate analyses for which there were no HIV-positive cases: the North Eastern province and men's ages 15 to 19 years. To keep the respondents in these categories in the analysis, we created 2 dummy HIV-positive cases: one dummy HIV-positive woman from North Eastern region and one 15- to 19-year-old dummy HIV-positive man from North Eastern region; all remaining characteristics for these cases were assigned the reference category. All statistical analysis was conducted using SPSS 11.5 (SPSS Inc, Chicago, IL).
The dependent variable in the analysis is a dichotomous indicator of HIV serostatus. Factors associated with HIV infection fall roughly into 5 conceptual categories: demographic, residential, social, biological, and behavioral. Variables representative of each conceptual category are included in the analysis and discussed here in brief.
Demographic and Residential Characteristics
Demographic and residential characteristics include current age (expressed as a grouped variable), region, urban or rural residence, and number of own children that have died.
Social characteristics incorporated here are education, household wealth, marital status, religion, and the respondent's perceived risk of contracting HIV. It can be theorized that information on how to avoid infection with HIV is more easily acquired by those who have greater education. Marital status is an important risk factor for HIV because it is associated with number and type of sexual partners as well as with HIV serostatus.4-6 Religious affiliation is included to control for its potential confounding effects.7 An early age at first sex may be associated with the number of lifetime sexual partners, which is considered a key risk factor for contracting HIV.8,9 A variable indicating the respondent's perceived risk for contracting HIV is included to control for unobserved high-risk contexts or behaviors. No indicator of ethnic group has been included, as ethnic group membership is highly correlated with region of residence. Household wealth status is likely to have an important independent relationship to HIV status; details on the methodology used to construct the asset-based household wealth index used here are provided by Rutstein and Johnson.10
Biological characteristics included here are circumcision status (for men only), recent birth status and use of depot medroxyprogesterone acetate (DMPA)/Norplant contraceptives (for women only), and recent experience of a sexually transmitted infection (STI) or a symptom of a STI. Many researchers have pointed to a greater risk of contracting HIV for men who are not circumcised (eg, Gray et al,7 Agot et al,11 Auvert et al,12 Weiss et al,13 and Rakwar et al14). Accumulated scientific evidence (c.f., Lewis et al15 and Zaba and Gregson16) indicates that fertility is reduced in HIV-infected women; therefore, a variable indicating whether the respondent has given birth in the 5 years preceding the survey has been included. A variable reflecting whether the respondent reported diagnosis with STI or experienced symptoms of an STI in the past year was incorporated; the literature is clear on the positive association between non-HIV STIs-particularly those that are ulcerative-and coinfection with HIV. Progesterone-based contraceptives have been hypothesized to increase susceptibility to infection with HIV (eg, Lavreys et al,17 Criniti et al,18 and Kiddugavu et al19); this relationship is also investigated.
Behavioral characteristics include risk factors such as condom use, the exchange of sex for gifts or money, the number of sexual partners the respondent had in the past year, frequency of alcohol use in the past month, and the number of times the respondent slept away from home in the past year (for men only). The ideal analysis would incorporate a variable that reflects the consistent and correct use of condoms for a specified time period. For previously discussed reasons, a variable reflecting ever-use of a condom is included: it is expected that such a variable simultaneously proxies both the availability of condoms to the respondent and the propensity of the respondent to use a condom. Because knowledge of one's own HIV serostatus is unusual in Kenya, it is unlikely that the relationship between condom use and serostatus would suffer from problems of bidirectionality. Alcohol use has been found to be associated with HIV seropositivity.6 The frequency with which a respondent has slept away from home during the past year is an indicator of the mobility of male respondents; some studies find that mobility confers an increased risk for HIV,14 whereas other studies present equivocal findings on the topic.20
Results of the 2003 KDHS indicate a national HIV seroprevalence of 6.7%, with prevalence among men at 4.6% and among women at 8.4% (tables not shown). In the present analysis, the sample is restricted to respondents who have had sex; for this group, prevalence among men is 5.3%, and among women, it is 10.1%. The large difference in prevalence between men and women requires further investigation but is beyond the scope of this article. Bivariate results are presented in Table 1 but are not discussed here.
Four logistic regression models were run for both the women's and men's analyses; the first model contains only the demographic/residential characteristics, the second model adds the social characteristics, the third model adds biological factors, and the fourth and final model adds behavioral factors. Women's regression results are given in Table 2, and those of men are given in Table 3. Reference categories are designated by an ®.
The first model shows that women aged 25 to 29 years are at a significantly higher risk of being HIV-positive than women in the reference category aging from 15 to 19 years. Those living in Nyanza are more than twice as likely as those living in Nairobi to be HIV-positive, whereas the risk for contracting HIV is much lower in North Eastern Province than in Nairobi. Those living in rural areas are half as likely to be positive as those in urban areas. The number of one's children that has died is clearly associated with being HIV-positive: compared with those who have never had any children, those who have experienced the death of 1 child are twice as likely to be HIV-positive, whereas those who have experienced the loss of 2 or more of their children are more than 3 times as likely to be infected.
In the second model, the addition of a variable indicating age at first sex, although not significant itself, clarifies the age-related risks for being HIV-positive: results indicate an inverted U-shaped relationship between age and infection, with the probability of being infected with HIV peaking at age 25 to 29 years. The risk for those living in Nyanza remains double that of Nairobi, whereas rural residence has lost its significant protective effect as a result of controlling for wealth. Those with primary education are nearly twice as likely to be HIV-positive as those with no education; although the odds for being HIV-positive are higher for the secondary and higher education categories, they are not significantly different than those for people with no education.
Wealth is positively and monotonically related to being HIV-positive, with those in the wealthiest quintile being 3 times more likely to be infected than those in the poorest quintile. Regarding marital status, widowed women are nearly 10½ times more likely to be HIV-positive, women who are 1 of 3 or more wives in a polygynous marriage are over 3 times more likely to be HIV-positive, and women who are divorced or separated are about 2½ times more likely to be infected compared with women who are the only wife in a marital or cohabitating union. Never-married women are no more or less likely to be infected than the reference group. Muslim women are 70% less likely than women of other religions to be HIV-positive. One's perceived risk of contracting HIV remains related to HIV serostatus; however, the results differ from the bivariate once demographic factors are controlled for: only women who believe they have a small risk of contracting the virus are more likely, by approximately 50%, to be HIV-positive than those that believe they have no risk at all for contracting the virus.
The third model incorporates 3 biological variables: whether a woman has given birth in the 5 years preceding the survey, whether a woman reports having an STI in the past year, and whether a woman is currently using either DMPA or Norplant as a contraceptive method. Women who have had a birth in the 5 years preceding the survey are approximately 30% less likely to be HIV-positive than those who did not have a birth in the past 5 years (P = 0.033); note that the bivariate relationship did not show significance, and adjustment for other factors makes the relationship stronger. Women who report having had an STI in the year preceding the survey are 80% more likely to be HIV-positive than women who did not. Use of DMPA/Norplant was not a significant contributing factor.
The final model incorporates key risk behaviors into the analysis. Due to failure to improve model fit and lack of significance, the variables reflecting ever-use of condoms and ever-exchange of sex for goods or money have been excluded from the final model. The risk of being HIV-positive does not vary according to reported number of partners in the past year. There are higher risks for women who have ever consumed alcohol compared with women who report that they never have: although the risks for being HIV-positive are 50% higher for women who have ever drunk alcohol but have not done so in the past month (P = 0.055), they are 2½ times higher for women who have drunk alcohol on 1 to 2 days in the past month (P = 0.006). Surprisingly, women who have drunk alcohol on 3 or more days in the past month are not significantly more likely than women who never drink to be HIV-positive. Relationships among the other variables in the model vary little with the addition of these 2 behavioral factors, with 1 exception: upon controlling for the number of partners a woman has had in the past year, the odds that a widow is HIV-positive compared with women in a monogamous union increase from approximately 10 times the risk to nearly 11 times the risk. No other category of marital status is affected by the control for number of partners.
In the first model for the men's analysis, we note the inverted U-shaped relationship between age and infection with HIV, such that those in the ages ranging from 25 to 44 years are the most likely to be infected, with risk peaking for those in the ages 35 to 39 years. All age groups are more likely to be infected than the reference group (ages 15-19 years). As discussed for the women's analysis, those living in Nyanza have a risk of being HIV-positive that is 3½ times greater than that for those living in Nairobi, the reference region. Men living in rural areas had a 57% reduced risk for being HIV-positive; unlike in the women's analysis, the protective effect of rural residence remains in the final model. The probability of being HIV-positive is positively, but not significantly, related to the number of one's children that has died.
The second model incorporates a number of social characteristics, most of which are not significant. Age at first sex does not have a significant relationship with HIV-serostatus in the multivariate analysis. The addition of the wealth variable, although not significant, results in the loss of the protective effect of rural residence. Religion is also a significant factor, with those reporting that they are not affiliated with a particular religion being nearly 2½ times as likely to be HIV-positive as Roman Catholics. As noted in the women's models, those who believe themselves to have only a small risk of contracting HIV are, in fact, more likely to be infected than those who believe they have no risk of infection at all.
In the third model, we find that reporting an STI in the past year is not significantly related to HIV serostatus for men, in contrast to what was found for women. However, the variable reflecting circumcision status is among the strongest in the model: men who are circumcised have one quarter the risk of those who are not circumcised to be HIV-positive. The circumcision variable absorbs most of the influence of STIs on the likelihood of being HIV-positive; number of partners in the past year also absorbs some of the influence of STIs.
The fourth model incorporates behavioral risk factors. The number of sexual partners in the past year does not have a significant relationship to HIV serostatus. Those who drank alcohol on 11 to 19 days in the past month were more than 2½ times as likely to be HIV-positive as those who have never consumed alcohol; otherwise, HIV status varies little by alcohol consumption. Those who did travel away from home were no more or less likely to be HIV-positive than those who did not, with the exception of those who reported staying away from home 11 or more times; these respondents were 78% more likely to be HIV-positive (P = 0.076).
In an effort to improve our understanding of the HIV/AIDS epidemic in sub-Saharan Africa, in general, and in Kenya, in particular, this study has reported on national HIV seroprevalence in Kenya and assessed key variables for their association with HIV serostatus at the individual level. The most important demographic, social, biological, and behavioral factors and their programmatic implications are discussed below.
The key demographic factor in this analysis was region: both men and women from Nyanza Province had double the risk for infection with HIV as compared with the respondents from Nairobi, the most densely-populated area in Kenya. Rural residence did not exert a protective effect on the risk of contracting HIV among women, and did so only weakly among men (P = 0.074), echoing findings of other researchers (eg, Voeten et al21) who show that sexual behavior can be significantly riskier in rural areas. Although many researchers regard urban residence as a key risk marker, the potential for HIV spread in rural areas exists; HIV/AIDS education and VCT services must reach rural residents.
Wealth was positively related to risk for HIV for both men and women, yet education did not show the same relationship to the outcome variable. Because economic status and educational status typically correlate for many outcomes, this finding is intriguing and would benefit from further exploration. Respondents who think they have only a small risk of contracting HIV are, in fact, at highest risk of being HIV-positive, compared with those who think they have no risk. Such findings highlight the crucial importance of getting tested to know one's status: those who believe themselves to have only a small risk are unlikely to take steps to prevent further transmission of the virus.
Marital status proved to be a significant risk factor for women, in particular widowed and divorced statuses. Wife inheritance (the remarriage of a widow to the brother or other male relative of the deceased husband) is a custom that is widespread in western Kenya, particularly among the Luo,22 who are concentrated in Nyanza Province. Given that widows are at extremely high risk of being HIV-positive, should they remarry, their new spouse takes on that increased risk for acquiring the virus. The fact that wife inheritance is widely practiced among the Luo people, in accordance with the fact that Luo men are the least likely among all Kenyans to have been circumcised,23 seems to be a lethal combination in the context of HIV/AIDS.
Women with HIV were 30% less likely to have recently given birth, results that are congruent with the findings of previous studies.16 Because age and current marital status are controlled for, the association between subfertility and HIV serostatus is most likely to arise from the effect of the virus on a woman's biological susceptibility to becoming pregnant. As expected, women who reported a probable STI were significantly more likely to be HIV-positive than women who reported no STIs in the past year. This finding highlights opportunities for interventions at treating health care facilities: those who test positive for an STI should be screened for HIV, and counseled and treated for their condition(s).
The outstanding biological factor associated with being HIV-positive for men was circumcision status: uncircumcised men were 4 times more likely to be HIV-positive than circumcised men. Other studies conducted in the region indicate that circumcised status may be overreported by as much as 10%,12 which leaves the possibility open that our results underestimate the effect of circumcision to some degree. Our findings add to the large body of research indicating that circumcision has a protective effect against HIV infection among men. However, because circumcision is very closely correlated with ethnicity and culture, because not all studies find a protective effect of circumcision, and because there could be a confounding relationship between circumcision and ulcerative STIs such as HSV-2, well-controlled evidence from other national and disciplinary contexts should be considered before widespread planning and implementation of circumcision-focused interventions.
Alcohol consumption proved to have a significant relationship to HIV serostatus for both women and men. For women, any reported consumption of alcohol increases risk; it may be that the consumption of alcohol by a Kenyan woman serves as a broader indicator of lifestyle because it is not common practice for Kenyan women to consume alcoholic beverages. Only men who report having consumed alcohol fairly frequently in the past month are significantly more likely to be HIV-positive.
Sex-related behavioral factors did not have as great an impact in the analyses as expected, although other studies have also failed to find the expected relationships between sexual behavior and HIV serostatus (eg, Lagarde et al24 and Morison et al25). The failure to find the expected relationships may be explained in part by data quality issues (respondents may not report their actual number of recent sexual partners) or imprecision of the variable used in the analysis (ever-exchange of sex for money or goods is a vague indicator of behavioral risk). However, it also must be acknowledged that "risky" sexual behavior while increasing individual risk for contracting HIV may not in fact be the primary driving force behind the epidemic in Kenya. Other factors, such as those that enhance the biological transmission of HIV, may simply matter more.
This analysis demonstrates that HIV is a multidimensional epidemic, with demographic, residential, social, biological, and behavioral factors exerting influence on individual probability of becoming infected with HIV. Although all of these factors contribute to the risk profile for a given individual, ultimately, the results suggest that differences in biological factors may be more important in assessing risk for HIV than differences in sexual behavior. The ways in which these intersecting factors affect risk differ by sex, which implies that program interventions may require gender-specific approaches. Above all, the findings reiterate that the situation of Nyanza Province is egregious within Kenya and is clearly in need of a broad-spectrum approach to HIV prevention, testing and counseling, and treatment.
The authors gratefully acknowledge Jeffrey Mewbourn's assistance with library services. This research was funded in part by USAID under its MEASURE DHS project.
1. UNAIDS. 2004 Report on the Global AIDS Epidemic. Geneva, Switzerland: Joint United Nations Program on HIV/AIDS; 2004.
2. CDC. HHS/CDC Global AIDS Program (GAP) in Kenya-FY2003. Atlanta, GA: Centers for Disease Control and Prevention; 2004.
3. Central Bureau of Statistics (CBS) [Kenya], Ministry of Health (MOH) [Kenya] and ORC Macro. Kenya Demographic and Health Survey 2003. Calverton, MD: CBS, MOH and ORC Macro; 2004.
4. Nunn AJ, Kenyega-Kayondo JF, Malamba SS, et al. Risk factors for HIV-1 infection in adults in a rural Ugandan community: a population study. AIDS. 1994;8:81-86.
5. Glynn JR, Caraël M, Auvert B, et al. Why do young women have a much higher prevalence of HIV than young men? A study in Kisumu, Kenya and Ndola, Zambia. AIDS. 2001;15(suppl 4):S51-S60.
6. Auvert B, Buvé A, Ferry B, et al, for the Study Group on the Heterogeneity of HIV Epidemics in African Cities. Ecological and individual level analysis of risk factors for HIV infection in four urban populations in sub-Saharan Africa with different levels of HIV infection. AIDS. 2001;15(S4):S31-S40.
7. Gray RH, Kiwanuka N, Quinn TC, et al, for the Rakai Project Team. Male circumcision and HIV acquisition and transmission: cohort studies in Rakai, Uganda. AIDS. 2000;14(15):2371-2381.
8. Clark S. Early marriage and HIV risks in sub-Saharan Africa. Stud Fam Plann. 2004;35(3):149-160.
9. Pettifor A, van der Straten A, Dunbar MS, et al. Early age at first sex: a risk factor for HIV infection among women in Zimbabwe. AIDS. 2004;18:1435-1442.
10. Rutstein SO, Johnson K. The DHS wealth index. DHS Comparative Report #6. Calverton, MD: ORC Macro International; 2004.
11. Agot KE, Ndinya-Achola JO, Kreiss JK, et al. Risk of HIV-1 in rural Kenya: a comparison of circumcised and uncircumcised men. Epidemiology. 2004;15(2):157-163.
12. Auvert B, Buvé A, Lagarde E, et al, for the Study Group on the Heterogeneity of HIV Epidemics in African Cities. Male circumcision and HIV infection in four cities in sub-Saharan Africa. AIDS. 2001;15(S4):S31-S40.
13. Weiss HA, Quigley M, Hayes RJ. Male circumcision and risk of HIV infection in sub-Saharan Africa: a systematic review and meta-analysis. AIDS. 2000;14:2361-2370.
14. Rakwar J, Ludo L, Thompson ML, et al. Cofactors for the acquisition of HIV-1 among heterosexual men: prospective cohort study of trucking company workers in Kenya. AIDS. 1999;13:607-614.
15. Lewis JJ, Ronsmans C, Ezeh A, et al. The population impact of HIV on fertility in sub-Saharan Africa. AIDS. 2004;18(S2):S35-S43.
16. Zaba B, Gregson S. Measuring the impact of HIV on fertility in sub-Saharan Africa. AIDS. 1998;12(S1):S41-S50.
17. Lavreys L, Baeten JM, Martin HL Jr, et al. Hormonal contraception and risk of HIV-1 acquisition: results of a 10-year prospective study. AIDS. 2004;18(4):695-697.
18. Criniti A, Mwachari CW, Meier AS, et al. Association of hormonal contraception and HIV-seroprevalence in Nairobi, Kenya. AIDS. 2003;18:2667-2669.
19. Kiddugavu M, Makumbi F, Wawer MJ, et al, Rakai Project Study Group. Hormonal contraceptive use and HIV-1 infection in a population-based cohort in Rakai, Uganda. AIDS. 2003;17(2):233-240.
20. Coffee MP, Garnet G, Mlilo M, et al. Patterns of movement and risk of HIV infection in rural Zimbabwe. J Infect Dis. 2005;191:S159-S167.
21. Voeten HA, Egesah OB, Habbema JD. Sexual behavior is more risky in rural than urban areas among young women in Nyanza Province, Kenya. Sex Transm Dis. 2004;31(8):481-487.
22. Luginaah I, Elkins D, Maticka-Tyndale E, et al. Challenges of a pandemic: HIV/AIDS-related problems affecting Kenyan widows. Soc Sci Med. 2005;60(6):1219-1228.
23. Bailey RC, Muga R, Poulussen R, et al. The acceptability of male circumcision to reduce HIV infections in Nyanza Province, Kenya. AIDS Care. 2002;14(1):27-40.
24. Lagarde E, Auvert B, Caraël M, et al, the Study Group on the Heterogeneity of HIV Epidemics in African Cities. Concurrent sexual partnerships and HIV prevalence in five urban communities of sub-Saharan Africa. AIDS. 2001;15:877-884.
25. Morison L, Weiss HA, Buvé A, et al, for the Study Group on the Heterogeneity of HIV Epidemics in African Cities. Commercial sex and the spread of HIV in four cities in sub-Saharan Africa. AIDS. 2001;15(S4):S61-S69.
Gwatkin DR, Rutstein SO, Johnson K, et al. Socio-economic Differences in Health, Nutrition and Poverty
. Washington, DC: HNP/Poverty Thematic Group of The World Bank; 2000.
*For details on response rate calculation, on the collection of dried blood spots for HIV testing and on ethical protocols, refer to, respectively, Appendix A, Chapter 13, and Chapter 1.10 in the 2003 Kenya DHS Final Report.3 Cited Here...
†An investigation into the effect of nonresponse on the representativeness of the Kenya DHS HIV data has been undertaken and is available from the authors. In short, it was found that nonresponse to the survey did not significantly bias the prevalence estimates. Cited Here...
This article has been cited 9 time(s).
AIDS and BehaviorMultilevel Stigma as a Barrier to HIV Testing in Central Asia: A Context QuantifiedAIDS and Behavior
Is poverty or wealth driving HIV transmission?
American Journal of Mens HealthBehavior, Knowledge, Attitude, and Other Characteristics of Men Who Had Sex With Female Commercial Sex Workers in KenyaAmerican Journal of Mens Health
International Journal of EpidemiologySocioeconomic status and HIV seroprevalence in Tanzania: a counterintuitive relationshipInternational Journal of Epidemiology
Southern African Journal of Hiv Medicine
'Differential Poverty Rates Are Responsible for the Racial Differentials in Hiv Prevalence in South Africa': An Enduring and Dangerous Epidemiological Urban Legend?
Southern African Journal of Hiv Medicine, ():
Household and community income, economic shocks and risky sexual behavior of young adults: evidence from the Cape Area Panel Study 2002 and 2005
International Journal of Health GeographicsSpatial distribution and cluster analysis of sexual risk behaviors reported by young men in Kisumu, KenyaInternational Journal of Health Geographics
Current Opinion in Infectious DiseasesCircumcision and HIV transmissionCurrent Opinion in Infectious Diseases
Kenya; HIV/AIDS; seroprevalence; circumcision
© 2006 Lippincott Williams & Wilkins, Inc.