HIV Articles  
Back 
 
 
Association of Serum Lipid Levels With HIV Serostatus, Specific Antiretroviral Agents, and Treatment Regimens in WIHS (Women)
 
 
  JAIDS Journal of Acquired Immune Deficiency Syndromes:Volume 45(1)1 May 2007pp 34-42
 
Anastos, Kathryn MD*; Lu, Dalian PhD; Shi, Qiuhu PhD; Tien, Phyllis C MD; Kaplan, Robert C PhD*; Hessol, Nancy A MSPH; Cole, Steven PhD; Vigen, Cheryl MS#; Cohen, Mardge MD**; Young, Mary MD; Justman, Jessica MD
 
".....In summary, we found that the dyslipidemia associated with HIV infection was not present in women taking non-PI HAART, whereas it was more severe in women on PI-containing treatment, with the exception of atazanavir without ritonavir. This suggests that non-PI regimens may be superior in avoiding cardiovascular risk in women. In addition, the large size of this cohort allowed us to detect differences in lipid levels associated with specific agents, including some of the NRTIs. Specifically, lamivudine, didanosine, tenofovir, atazanavir, and non-PI treatment regimens were associated with less dyslipidemia, potentially mitigating an adverse effect of HIV infection on cardiovascular risk, whereas d4T, nelfinavir, lopinavir, and ritonavir were associated with greater dyslipidemia. Longitudinal studies are needed to define these patterns further and to ascertain the associated cardiovascular risk....."
 
Influence of Individual Drugs

The association of individual antiretroviral agents with each lipoprotein is shown in Figure 3. Reported P values are for comparison of users with nonusers among HIV-positive women, adjusted for any of the other agents that demonstrated an association with that specific lipoprotein at P ≦ 0.05 and for age, race, BMI, menopausal status, smoking, nadir and change in CD4 cell count, and use of lipid-lowering drugs. The observed difference in total-C for the commonly used NNRTIs efavirenz and nevirapine (181 vs. 174 mg/dL; P = 0.027 for efavirenz and 179 vs. 174 mg/dL; P = 0.183 for nevirapine) was entirely attributable to the highly statistically significant increase in HDL-C (see Fig. 3B; 55 and 52 vs. 46 mg/dL, respectively; P < 0.0001 for both). There was no association of the NNRTIs with differences in LDL-C or TGs.
 

Chlo-1.gif

Tri-2.gif

FIGURE 3. Association of individual antiretroviral agents with LDL-C, HDL-C, and triglycerides. ATZ indicates atazanavir; HIV + NoMed, HIV-positive untreated; IDV, indinavir; RTV, ritonavir; SQV, saquinavir. Values for HIV-negative and HIV-positive untreated women are shown for reference. A, Association with LDL-C. In users versus nonusers, P < 0.0001, P = 0.004, P = 0.002, and P = 0.031 for nelfinavir, indinavir plus ritonavir, ritonavir, and lamivudine, respectively. Adjusted for use of pharmacologic agents that are lipid lowering. B, Association with HDL-C. P < 0.0001 in users versus nonusers of lamivudine, nevirapine, and efavirenz; P = 0.0002 in users versus nonusers of didanosine; P = 0.043 in users versus nonusers of saquinavir plus ritonavir. Adjusted for use of pharmacologic agents that are lipid lowering. C, Association with triglycerides. P = 0.002, P = 0.006, and P = 0.009 in users versus nonusers of d4T, lopinavir, and tenofovir, respectively; P = 0.019 in users versus nonusers of abacavir; and P = 0.057 in users versus nonusers of atazanavir without ritonavir. All analyses are adjusted for BMI, race, age, menopausal status, smoking, nadir CD4 cell count, and statin. In addition, each of the NRTIs and nucleotide RTIs is adjusted for every other medication in the graph; the 2 NNRTIS are adjusted for each of the NRTIs and nucleotide RTIs; and each of the PIs not used with ritonavir is adjusted for each of the NRTIs and nucleotide RTIs plus efavirenz, nevirapine, and ritonavir. P values >0.050 are not reported. Among the adjustment parameters, only the main drug considered was treated as a categoric variable. Error bars indicate the 95% confidence interval.
 
Abstract
Background: The effects of HIV infection, highly active antiretroviral therapy (HAART), and specific antiretroviral agents on lipoproteins in women are not well described.
 
Methods: In a cross-sectional substudy of the Women's Interagency HIV Study with 623 HIV-negative and 1556 HIV-positive women (636 untreated, 419 on non-protease inhibitor [PI] HAART, and 501 on PI-containing HAART), we performed multivariate analyses of associations among fasting lipoprotein levels, HIV infection, and HAART.
 
Results: Untreated HIV-positive women had lower high-density lipoprotein cholesterol (HDL-C) and higher triglycerides (TGs) but not lower low-density lipoprotein cholesterol (LDL-C) than HIV-negative women and were the most likely to have unfavorable HDL-C by National Cholesterol Education Program (NCEP) guidelines. PI HAART users had higher LDL-C than untreated HIV-infected women (107 vs. 100 mg/dL, P = 0.0006) and were the most likely to have unfavorable LDL-C and TGs by NCEP guidelines. HIV-negative women and non-PI HAART users had similar HDL-C levels (55 and 53 mg/dL, respectively), which were higher than those in untreated HIV-infected women and PI HAART users (42 and 49 mg/dL, respectively; P < 0.001 for all). Lamivudine, didanosine, nevirapine, and efavirenz were independently associated with higher HDL-C (P < 0.001 for all). Ritonavir, indinavir/ritonavir, and nelfinavir were associated with higher LDL-C (P < 0.01 for all). Stavudine, abacavir, and all ritonavir-containing regimens were associated with higher TGs (P < 0.05 for all), and tenofovir was associated with lower TGs (P = 0.009).
 
Conclusions: A dyslipidemic pattern was associated with HIV infection itself, was more severe in users of PI-containing HAART, but was not present in women taking non-PI HAART.
 
From puberty onward, serum lipoprotein concentrations differ between women and men.1-5 Premenopausal women, on average, have lower total cholesterol (total-C) and low-density lipoprotein cholesterol (LDL-C) than age-matched men, but this relation reverses after natural or surgical menopause.5,6 High-density lipoprotein cholesterol (HDL-C) is, on average, higher in women throughout adulthood7,8 and in American blacks compared with American whites.8
 
Dyslipidemias have been described in HIV-infected individuals since 1989, before the availability of potent antiretroviral therapy (ART), primarily in white men.9-14 These abnormalities included increased levels of triglycerides (TGs)10-14 and decreased levels of total-C,9,13-15 HDL-C,11,14-16 and LDL-C,9,11-15,17 with greater degrees of abnormalities associated with more advanced immune suppression.10-14 Some of the HIV-1 protease inhibitors (PIs) have also been reported to cause lipoprotein abnormalities, especially increases in TGs, total-C, and LDL-C, with variable effects on HDL-C17-25 in HIV-positive and HIV-negative individuals.26-28 One study of HIV-infected women found that PI- and nonnucleoside reverse transcriptase inhibitor (NNRTI)-based therapy was associated with higher TGs and total-C.29 Atazanavir, a PI, and tenofovir, a nucleotide reverse transcriptase inhibitor (NRTI), have been associated with fewer adverse lipid changes, or even improvement.24,30-34 The NNRTIs have been associated with beneficial effects on HDL-C and adverse effects on TGs.35,36 Of the NRTIs, stavudine (d4T) and abacavir have been shown to have an effect on lipids.35-38
 
In the United States and worldwide, women and people of African descent represent most of those infected with and a growing proportion of those treated for HIV infection. We therefore assessed in a cross-sectional analysis the association of HIV infection and treatment with lipoprotein levels in a cohort of American women of diverse ethnic backgrounds.
 
DISCUSSION
This study of 2179 WIHS participants is the largest to date investigating fasting lipoprotein levels in HIV-infected women. HIV infection itself, in the absence of treatment, was not associated with lower LDL-C, unlike the findings in previous reports in men,15-17 but it was associated with modestly higher TGs and markedly lower HDL-C, as has been seen in men.16,17 In population-based studies, differences of this magnitude in lipoprotein levels have been shown to increase the risk of cardiovascular events by 20% to 30%.1,48
 
The associations of lipoprotein levels with ART differed for PI- and non-PI-containing regimens. Of note, the adjusted mean LDL-C was significantly higher in the PI HAART users but not in the non-PI HAART users compared with the HIV-negative women, suggesting that the higher LDL-C seen with PIs may be a true adverse effect and not just the result of renewed health from effective ART. The finding of higher CD4 cell counts but lower LDL-C in users of non-PI HAART also supports this interpretation. PI-containing HAART was also associated with a smaller favorable difference in HDL-C compared with the use of non-PI regimens.
 
Of note, a strength of our study is the diversity of the participants, with a large proportion of African Americans, who are at greater risk than whites of adverse cardiovascular outcomes. We found higher HDL-C and lower TG levels in African American women than in white women, consistent with previous reports in HIV-uninfected persons. Although African American women are at greater risk of adverse cardiovascular outcomes than white women, factors other than dyslipidemia likely contribute to this risk.
 
Differing from findings in men,17,35,49 higher TG levels were independently associated only with d4T and lopinavir/ritonavir but not with other PIs unless they were used in combination with ritonavir, and there were no significant TG differences associated with the use of nevirapine and efavirenz. The demonstrated association of abacavir with higher TGs is consistent with a prior report.50 Our findings are also consistent with those of others35,36,51 who have reported better HDL-C and less deleterious TG responses to treatment with NNRTIs in women than in men. The strong independent associations of higher HDL-C with lamivudine and didanosine have not, to our knowledge, been described previously.
 
In both genders, there is a strong direct relation of LDL-C and a strong inverse relation of HDL-C with risk for coronary artery disease (CAD). The relative importance of each lipoprotein fraction differs by gender, however, with HDL-C more strongly and LDL-C less strongly associated with CAD in women compared with men.52,53 If the higher HDL-C found in users of non-PI HAART is, in fact, cardioprotective, it may carry clinical implications for cardiovascular protection in women treated for HIV infection. PI treatment but not non-PI treatment was associated with a larger proportion of women having unfavorable HDL-C and TG levels, as defined by NCEP guidelines, compared with HIV-negative women. HAART without a PI may therefore be preferable to minimize adverse lipid patterns for HIV-positive women at any increased risk of cardiovascular disease. Further data are needed to assess the incidence of adverse cardiovascular outcomes in association with the demonstrated higher levels of HDL-C, however. In addition, other factors must be considered, including the low genetic threshold for resistance and adverse effects of NNRTIs. Other risk factors for cardiovascular disease, including tobacco smoking, obesity, and diabetes mellitus, have a high prevalence or incidence in our cohort,40,54,55 which is representative of HIV-infected women in the United States.40 African American women, who are more frequently infected with HIV than any other racial/ethnic group in the United States,56 are at higher risk than white women for cardiovascular outcomes in the absence of HIV infection or HAART;1 thus, the attributable risk of increased rates of cardiovascular outcomes, if present, may be substantial.
 
Of note, the higher total-C in women receiving NNRTIs was a result of the higher HDL-C, without an unfavorable change in LDL-C. This suggests that reports of increased total-C must indicate the contribution of HDL-C.
 
TG levels are also more strongly associated with CAD in women than in men.52,53 We found modestly higher TGs in untreated HIV-infected women compared with HIV-negative women and significant ART-related increases only with ritonavir-containing regimens, d4T, and abacavir. This finding of limited antiretroviral agents associated with higher TGs may represent a gender difference in response to HIV infection. Most of our participants are African American, however, whose HDL-C and TGs were favorable compared with those of white women. Thus, race may be contributing some or all of the potentially favorable associations we have described in HDL-C and TGs.
 
Several limitations of our study should be noted. This is a cross-sectional analysis, and better estimates of changes in lipids associated with classes of or specific antiretroviral agents can be obtained from clinical trials and longitudinal observational studies of individuals before and after initiation of therapy. Better estimation of the effects of HIV infection itself requires values obtained before and after HIV seroconversion with an age-matched control group, as has been demonstrated in men.17 Such information in women is not available. The large sample size, the adjustment for known confounding factors, and the high level of statistical significance for these differences suggest that causal relations likely drive the observed associations, however.
 
Another limitation is that many women self-identifying as Hispanic did not identify as white or black, whereas others did. This may have introduced misclassification by race. In addition, comparisons of individual agents are somewhat artificial in that no drug is ever used alone. Although analyses by HIV serostatus and treatment group were adjusted for multiple comparisons, this is not true of the associations of individual drugs with lipoprotein profiles. We thus caution the reader to remain skeptical of differences with modest P values (ie, P ≥ 0.01).
 
Because this is not a controlled trial, there are possible unmeasured confounders, including potential indication bias (ie, women who had high lipid levels before treatment may have been more likely to be prescribed lipid-sparing ART). The WIHS did not collect data on exercise history and diet, and these potential confounders could thus not be included in analyses. In addition, we did not have enough users of (fos)amprenavir to allow assessment of its association with the lipoproteins.
 
In summary, we found that the dyslipidemia associated with HIV infection was not present in women taking non-PI HAART, whereas it was more severe in women on PI-containing treatment, with the exception of atazanavir without ritonavir. This suggests that non-PI regimens may be superior in avoiding cardiovascular risk in women. In addition, the large size of this cohort allowed us to detect differences in lipid levels associated with specific agents, including some of the NRTIs. Specifically, lamivudine, didanosine, tenofovir, atazanavir, and non-PI treatment regimens were associated with less dyslipidemia, potentially mitigating an adverse effect of HIV infection on cardiovascular risk, whereas d4T, nelfinavir, lopinavir, and ritonavir were associated with greater dyslipidemia. Longitudinal studies are needed to define these patterns further and to ascertain the associated cardiovascular risk.
 
RESULTS
Characteristics of the 2179 women who contributed a fasting sample are shown in Table 1. The HIV-negative women were younger, less likely to report postmenopausal status, and had a higher mean BMI than the HIV-positive women on treatment (P < 0.0001 for all). The median BMI was in the overweight range (between 25.0 and 30.0 kg/m2) for all groups. There was little use of lipid-lowering agents. Among the HIV-positive women, the median nadir CD4 cell count was significantly lower (P < 0.0001) and the median current CD4 cell count was higher (P < 0.0001) among the women on HAART. The median current HIV RNA levels were <80 cps/mL in both groups on treatment.
 
The use of each specific antiretroviral agent at the time of the study visit is shown in Table 2. Of the 636 HIV-infected women not on any ART at the time of their study visit, 264 (41%) had previously received HAART, 81 (13%) had received non-HAART ART in the past, and 291 (46%) were ART naive.
 
All data presented for the associations of HIV serostatus and treatment category with the individual lipoprotein levels result from multivariate models that included age, menopausal status, race, BMI, smoking, nadir and change in CD4 cell count, and use of pharmacologic agents that are lipid lowering.
 
Low-Density Lipoprotein Cholesterol
LDL-C was not different in the HIV-negative women compared with the untreated HIV-positive women or the users of non-PI HAART (100 vs. 98 and 102 mg/dL; P = 0.44 and P = 0.61, respectively; Fig. 1). Women using PI HAART had higher LDL-C (107 mg/dL) than women in each of the other groups, but after adjustment for multiple comparisons, the difference was statistically significant only compared with the untreated HIV-positive women (P = 0.0006).
 
High-Density Lipoprotein Cholesterol
The adjusted mean HDL-C was markedly lower in untreated HIV-positive women than in all other groups: 42 vs. 55, 53, and 49 mg/dL for HIV-negative or HIV-positive women using non-PI and PI HAART, respectively (P < 0.0001 for all). Users of non-PI HAART had a mean HDL-C value similar to that of HIV-negative women (53 and 55 mg/dL, respectively; P = 0.163), and HDL-C was higher in users of non-PI HAART compared with PI HAART (P = 0.0002). Of untreated HIV-positive women, 70% had HDL-C <50 mg/dL, the defining threshold of HDL in the American Heart Association definition of metabolic syndrome,47 compared with 39%, 46%, and 58% of the HIV-negative and non-PI and PI HAART-using women, respectively.
 
Triglycerides
The untreated HIV-positive women had higher TG levels than the HIV-negative women (131 vs. 109 mg/dL; P = 0.001) and lower levels than the non-PI (143 mg/dL; P = 0.033) and PI HAART users (151 mg/dL; P = 0.0004).
 
Total Cholesterol
The untreated HIV-positive women had lower total-C than the HIV-negative women (166 vs. 175 mg/dL; P = 0.004), and PI HAART users had higher total-C (186 mg/dL) than all other groups (P < 0.002), except users of non-PI HAART (180 mg/dL; P = 0.061), in multivariate modeling, as described previously.
 
Associations With Demographic Variables
African American compared with white and Hispanic women, had higher HDL-C (51 vs. 47 and 45 mg/dL; P = 0.0002 and P < 0.0001, respectively) and lower TGs (117 vs. 158 and 152 mg/dL; P < 0.0001). Higher LDL-C was associated with higher BMI (P < 0.0001), higher CD4 cell count nadir (P < 0.0001), and greater CD4 cell count change (P = 0.0003). Lower HDL-C was associated with higher BMI (P = 0.0002), lower CD4 cell count nadir (P < 0.0001), and smaller CD4 cell count change (P = 0.001). Higher TG levels were seen with greater age (P < 0.0001), higher BMI (P = 0.029), postmenopausal status (P < 0.0001), lower CD4 cell count nadir (P < 0.0001), and greater CD4 cell count change (P < 0.0001).
 
Use of lipid-lowering agents was associated with higher LDL-C (114 vs. 101 mg/dL; P = 0.0004) and TG levels (226 vs. 131 mg/dL; P < 0.0001). Adherence was similar in the 2 groups using ART: 45% of both groups reported 100% adherence, and 36% of the non-PI and 31% of the PI users (P = 0.22 by χ2) reported 95% to 99% adherence in the previous 3 days, with `20% of participants not having these data available.
 
Proportion With Abnormal Values
Figure 2 displays the proportion of women in each group who had unfavorable lipid levels as defined by the NCEP44: HDL-C <40 mg/dL and TGs >150 mg/dL. We assessed 3 LDL thresholds: >160, >130, and >100 mg/dL, corresponding to the treatment thresholds for increasing numbers of cardiovascular risk factors; the trends were similar in all 3 analyses, and only results for LDL-C >130 mg/dL are reported here.
 
There were significant differences among the groups for all the lipoproteins. PI users were most likely to have elevated LDL-C and TGs, whereas the untreated HIV-infected women were most likely to have low HDL-C. Of note, although the adjusted mean values of HDL-C were comparable in the HIV-negative women and the users of non-PI HAART, as reported previously, the HIV-negative women were significantly less likely to have HDL-C levels less than 40 mg/dL (15% vs. 24%; P = 0.0004). Similarly, although LDL-C levels were comparable in HIV-negative and untreated HIV-positive women, the HIV-negative women were more likely to have values >130 mg/dL (22% vs. 15%; P = 0.001). All P values reported here for pairwise comparisons were adjusted using Bonferroni criteria.
 
METHODS
Study Population

The Women's Interagency HIV Study (WIHS) is a multicenter prospective cohort study of HIV-1 infection and its treatment in US women. WIHS methods have been described previously.40 Briefly, from October 1994 to November 1995 (69.6% of the cohort) and from October 2001 to September 2002 (30.4% of the cohort), 3770 women (2794 HIV-1-seropositive and 976 HIV-1-seronegative) were enrolled. Every 6 months, participants were interviewed, received a physical examination, and provided gynecologic and blood specimens. At each visit, interviewers assessed self-reported ART use in the period since the previous study visit, querying the participants about each antiretroviral agent. The definition of highly active antiretroviral therapy (HAART), whose patterns of use in the WIHS cohort have previously been described,41,42 was taken from the Department of Health and Human Services (DHHS) guidelines as: self-reported use of (1) 2 or more NRTIs in combination with at least 1 PI or 1 NNRTI; (2) 1 NRTI in combination with at least 1 PI and at least 1 NNRTI; (3) a regimen containing ritonavir and saquinavir in combination with 1 NRTI and no NNRTIs; or (4) an abacavir- or tenofovir-containing regimen of 3 or more NRTIs in the absence of PIs and NNRTIs, except for the 3-NRTI regimens consisting of abacavir plus tenofovir plus lamivudine or didanosine plus tenofovir plus lamivudine. Combinations of zidovudine (AZT) and d4T with a PI or NNRTI were not considered HAART. Informed consent was obtained from the participants in accordance with procedures and consent materials reviewed and approved by the committee on human research at each of the collaborating institutions.
 
The WIHS initiated collection of fasting blood specimens in October 2000. Women contributing a fasting blood sample (self-report of not eating for at least 8 hours) from October 2000 through March 2005 are included in this analysis. If a woman contributed more than 1 fasting sample, the most recent specimen was used for this analysis. The 70 women using ART not meeting the definition of HAART were excluded. Serum samples for lipoprotein determinations were frozen and underwent subsequent batch analysis in a central laboratory.
 
Laboratory Methods
Total-C, TG, and HDL-C analysis was performed on the Roche Modular automated system (Roche Diagnostics Corporation, Indianapolis, IN). LDL-C was measured directly for specimens collected in 2004 through 2005 and for all women seen in 2000 through 2003 with TGs >400 mg/dL. For the remaining 592 women, LDL-C was calculated using the Friedewald equation:43 [LDL-C = (total-C) - (HDL-C) - (TG/5.0)]. Simultaneous measurement of direct and calculated LDL in the 2004-2005 WIHS specimens with TGs <400 mg/dL indicated comparable values, and analyses including only the calculated and only the directly measured LDLs did not substantially alter the results.
 
Plasma HIV-1 RNA in plasma was quantified using the isothermal nucleic acid sequence-based amplification (Nuclisens) method (Organon Teknika Corporation, Durham, NC), with a lower limit of quantification of 80 copies (cps)/mL, in laboratories participating in the National Institutes of Health (NIH)/National Institute of Allergy and Infectious Diseases (NIAID) Virology Quality Assurance Laboratory proficiency testing program. Lymphocyte subsets were quantified using standard flow cytometry methods in laboratories participating in the NIH/NIAID Flow Cytometry Quality Assessment Program.44
 
Outcome Variables
The outcome variables were total-C, LDL-C, HDL-C, and TGs, defined (1) as continuous variables and (2) as unfavorable levels, as defined by the National Cholesterol Education Program (NCEP).45
 
Exposure Variables
The primary independent variables were HIV infection, antiretroviral regimens categorized as HAART without a PI and HAART containing a PI, and use of individual antiretroviral agents. Because of the relatively rapid effect on lipoprotein levels with initiation or discontinuation of ART, HIV-positive women not receiving ART at the time of the visit were treated as a single group without regard to prior ART use. Use of a PI with ritonavir at a dose <400 mg/d was analyzed separately from PI use without ritonavir. Variables assessed as confounding factors included body mass index (BMI) calculated as weight in kilograms divided by height in meters, age, self-reported menopausal status, current tobacco smoking (yes vs. no), current use of lipid-lowering agents, and race. Race was based on self-categorization as African American (Hispanic and non-Hispanic), white (Hispanic and non-Hispanic), or Hispanic who did not identify as black or white. For the HIV-positive women, the CD4 cell count (nadir and at most recent WIHS 6-month visit) and HIV-1 RNA level (peak and at most recent WIHS visit) were considered. For women using HAART, the change in CD4 cell count after HAART initiation was considered, calculated as the difference between the nadir and the most recent CD4 cell count. Less than 4% of participants reported use of oral contraceptives, which was not included as a covariate. Dietary factors and exercise history were not measured in the WIHS, and thus could not be included as potential confounders.
 
Statistical Methods
Analysis of variance or χ2 tests was used to compare mean lipid levels and percent of participants with abnormal lipid levels among the 4 groups defined by HIV status and ART. Variables to be included in the multivariate models were chosen by their association with lipoprotein values or by their different representation in the 4 major groups. Thus, potential confounding variables that had a P value <0.10 were age, race, BMI, menopausal status, smoking, nadir and change in CD4 cell count, and use of lipid-lowering drugs, which were included in the multivariable models analyzing the outcome variables among the 4 groups and in analyses of specific antiretroviral medications. HIV RNA level was not included in the multivariate models, because CD4 cell count and HIV RNA level are highly correlated. If the overall group test for differences in lipid levels among HIV treatment groups was significant, a 2-sample test was used to assess pairwise comparisons. If the overall group test was not significant, pairwise comparisons are not reported. Analyses were also performed excluding users of lipid-lowering agents. For analyses of individual drugs, all comparisons are of HIV-positive women taking versus not taking the specific agent, with additional adjustment for other drugs that demonstrated a possible association with specific lipid parameters at a P value ≦0.05 in univariate analyses. Bonferroni adjustment criteria were applied to pairwise comparisons.46 All analyses were performed using the SAS statistical package (version 8; SAS Institute, Cary, NC).
 
 
 
 
  icon paper stack View Older Articles   Back to Top   www.natap.org