An association of increased weight with a slower progression of human immunodeficiency virus (HIV) disease has been reported in studies that have not included large numbers of women. We evaluated the association of HIV disease progression with body mass index (BMI) in 871 women and present cross-sectional, survival, and longitudinal analyses. A higher baseline BMI was associated with a lower rate of occurrence of the first CD4 cell count <200 cells/mm3. In analyses that incorporated time-varying BMI, underweight and normal women had an increased risk of clinical acquired immune deficiency syndrome, and underweight women had increased risk of HIV-related death, compared with obese women. The association between change in BMI and CD4 cell count was estimated; increases in BMI were associated with slight increases in CD4 cell counts, even after controlling for prior values of CD4 cell count, viral load, and treatment. Higher BMI and increases in BMI are associated with a decreased risk of HIV progression.
The adverse effects of being overweight or obese in the general population in terms of mortality and risks of specific diseases are well known. In some situations, excess weight may be beneficial. Several published studies have shown improved survival with overweight or obesity in chronic renal failure  and HIV infection , 2 conditions characterized by immune system compromise. The Miami Intravenous Drug Abuse Study (MIDAS)  found that an increased body mass index (BMI) was associated with decreased mortality in a mixed-sex cohort of HIV-positive illicit drug users. It also found a decreased risk of a 25% decline in CD4 lymphocyte counts among the obese participants, despite no significant difference in baseline CD4 cell count or length of follow-up. Shuter et al.  reported a decreased rate of HIV progression in overweight men and women followed at an inner-city HIV clinic, despite similar baseline CD4 cell counts and a similar time to the initiation of HAART, compared with patients who were not overweight. They also found a trend toward lower HIV RNA load in overweight patients compared with those who were not over the course of follow-up. The Nutrition for Healthy Living Study (NFHL), which used bioelectrical impedance to measure body composition, found that BMI correlated very highly with fat mass in HIV-infected women (Spearman's ρ = .93). The NFHL study found a protective association of fat mass with HIV progression in women (predominantly CD4 cell decline) in multivariate analysis, but no mortality advantage was found (C.Y.J., E. Tchetgen, D. Jacobson, et al., unpublished data).
We used longitudinal data from a large, multicenter, prospective cohort study of HIV infection in women to examine the association of BMI with HIV disease outcomes in 2 types of analyses. The first analysis looked at event rates of clinically important HIV disease outcomes (time to first CD4 cell count <200 cells/mm3, first CD4 cell count <100 cells/mm3, opportunistic infection [OI], and HIV-related death) to determine whether there is a prospective association of BMI with these rates. The second analysis looked at the temporal covariation of BMI and CD4 cell count. Our objective was to discover whether the protective association seen in prior studies would be replicated in this large cohort of HIV-infected women.
SUBJECTS, MATERIALS, AND METHODS
Participants. The HIV Epidemiology Research (HER) Study enrolled 1310 female participants between 1993 and 1996, of whom 871 were HIV positive at study entry. HIV-negative participants were recruited from the same populations as the HIV- positive participants. Participants were aged 16–55 years and had a history of injection drug use (IDU) or high-risk sexual contact. Potential participants with prior AIDS-defining clinical illnesses were excluded. At the time of study enrollment, CD4 cell counts <200 cells/mm3 or CD4 cell percentages <14% were not considered to be AIDS-defining conditions; 17% of the cohort had a CD4 cell count <200 cells/mm3 at baseline. Participants were followed at 6-month intervals; height was measured at baseline, and weight was measured at each study visit.
The present analysis focuses on the HIV-positive women. At every visit, participants were administered a questionnaire by interview that included questions about the occurrence of OIs or malignancy since the last visit and current medications. Blood samples were obtained for immunological assays, including CD4 and CD8 cell counts and HIV load. A complete summary of study methods can be found in Smith et al. .
Outcomes. Five different outcomes were evaluated: CD4 cell count, time to first occurrence of CD4 cell count <200 cells/mm3, first occurrence of CD4 cell count <100 cells/mm3, first occurrence of an OI/malignancy, and HIV-related death. Times to event were recorded from enrollment to the date of the visit at which the event was first recorded. If an outcome was present at baseline (i.e., CD4 cell count <200 or <100 cells/mm3), the participant was excluded from that particular outcome analysis. If a woman had ⩾3 successive visits with missing values for the outcome measure, her results were censored from the time of the first missing value. If a death certificate was available, all causes listed were examined to determine the most likely underlying cause of death. If no death certificate was available, the National Death Index, medical record abstraction, or both were assessed for the cause of death. HIV/AIDS-associated deaths had either 1 of 24 AIDS-defining conditions  or simply “HIV/AIDS” as the underlying cause. For women who did not experience the event of interest or HIV-related death (HIV-related death was recorded as an event in all analyses), follow-up time was treated as a right-censoring time.
Variables/covariates. The following variables were examined in univariate and multivariate analysis. BMI was calculated as weight (lbs)/height (inches)2 × 703 . BMI groups were defined as follows: group 1, underweight (BMI <20); group 2, normal weight (BMI ⩾20 and <25); group 3, overweight (BMI ⩾25 and <30); and group 4, obese (BMI ⩾30). These weight-group cutoffs are those used by C.Y.J., E. Tchetgen, D. Jacobson, et al. (unpublished data) and are similar to those in the NIH Clinical Guidelines on the Identification, Evaluation and Treatment of Overweight and Obesity in Adults .
Viral load was measured by the Quantiplex branched DNA assay (Chiron) and was entered into analyses categorically (<500, ⩾500 and <1000, ⩾1000 and <10,000, and ⩾10,000). The use of antiretroviral therapy (ART) was recorded at each visit and entered into regression models as a 3-part variable (none, ART, or HAART). At baseline, no participant was receiving HAART (which was not available until midway through the study). Each risk cohort was defined as mutually exclusive categories of IDU or sexual exposure; participants reporting injecting any drugs since 1985 were classified in the IDU risk cohort. Visit number (1–12) and study center were entered into analysis, to control for time in study and possible systematic differences by geographic area.
CD4 cell count was analyzed by flow cytometry and complete blood count with differential. In the time-to-event analysis, CD4 cell count was treated as a categorical variable (<200, 200–500, and ⩾500 cells/mm3). CD4 cell count was used as a continuous variable in the analysis of temporal association of BMI and CD4 cell count.
Income was categorized as 1, $0–$500/month; 2, $501–$1000/month; 3, $1001–$2000/month; 4, $2001–$2500/month; 5, $2501–$3000/month; and 6, >$3000/month. Educational level was coded 1–16, to account for each year of schooling through college or vocational training. Depression at baseline was assessed using the Centers for Epidemiologic Studies Depression scale ; participants scoring ⩾16 were classified as depressed.
Statistical analysis. The first part of our analysis summarized the effect of baseline BMI on the outcomes, and the second part examined the effects of BMI varying through time.
Effect of baseline BMI on progression and CD4 cell count. BMI was stratified into 4 BMI weight groups, as described above. Distributions of time-to-event variables were summarized using Kaplan-Meier curves; because times to event are highly correlated with viral load and CD4 cell count, we also did stratified comparisons (stratifying on CD4 cell counts <500 or ⩾500 cells/mm3 and on HIV load greater than or less than 500 RNA copies/mL). Differences in time-to-event distributions between BMI categories were further assessed using discrete-time hazard regression models , to simultaneously adjust for baseline values of CD4 cell count, viral load, and treatment. The effect of baseline BMI group on longitudinal CD4 cell counts was summarized graphically in terms of mean ± SD of CD4 cell count by visit and by use of a repeated-measurements regression model with robust SEs to account for within-subject correlation , using SAS Proc Mixed in SAS (version 8; SAS). These comparisons also were stratified on baseline CD4 cell count and viral load.
Association between changes in BMI and time to discrete events. The second part of our analysis examined the effect of the most recent BMI on HIV progression. For time-to-event outcomes, we used logistic regression to model the discrete-time hazard function λt, defined as the probability of event occurrence at visit t, given that the subject is still at risk for the event at t. The discrete-time hazard naturally incorporates time-varying covariates such as CD4 cell count and viral load, and, when this is used in conjunction with logistic regression, the coefficients are approximately equal to those from Cox's proportional-hazards model . Specifically, our event-time model is where BMIt denotes the set of BMI weight group indicators (dummy variables) corresponding to the BMI most recently observed prior to visit t, Xt is the collection of time-varying covariates, and Z is the time-invariant (baseline) covariate. Like BMI, the time-varying covariates represent the most recently observed value prior to visit t. In addition to BMI, the event-time regressions included the time-varying covariates viral load category, presence of a clinical AIDS-defining condition (except in the clinical AIDS model), treatment status (none, ART, or HAART) and the baseline covariates study site, baseline ART status (yes/no), risk cohort, depression, and years HIV-positive at enrollment (self-reported). In addition, the CD4 cell count <200 and <100 models were adjusted for baseline CD4 cell count; the clinical AIDS and HIV death models were adjusted for the most recent CD4 cell count. The parameter αt captured underlying temporal variations in the visit-specific hazard rate (i.e., the baseline hazard function). The coefficient of primary interest is β, or the change in the log odds of the event corresponding to BMI weight group at any given visit. These models were fitted using Proc Genmod in SAS (version 8; SAS).
Temporal association between BMI and CD4 cell count. Our analysis of the association between CD4 cell count and BMI treated each as a longitudinal process. Combining results from 2 regression models, our analysis was designed to capture the covariate-adjusted correlation between the processes as a function of time. Specifically, for a set of time-varying covariates Vt and baseline covariates W, we fitted 2 separate models: These models regressed separately the per-visit change in CD4 cell count and in BMI on V and W, where μt and γt are time-specific intercepts, η1 and ϕ1 are coefficients of V, η2 and ϕ2 are coefficients of W, and the error terms are ɛt and δt. Both CD4 cell count and BMI were continuous variables in these models. The error terms represented, respectively, for CD4 cell count and BMI, residual variation at time t, after removing any variation attributable to covariates in V and W. The covariates in V included viral load group indicators, the presence of clinical AIDS, treatment status, and visit number. In the ΔCD4/Δt model, V also included prior CD4 cell count but not prior BMI; in the ΔBMI/Δt model, V included prior BMI but not prior CD4 cell count. Covariates in W included site, risk cohort, baseline depression status (yes/no), and years HIV positive at baseline. It is important to include prior BMI (in the BMI model) and prior CD4 cell count (in the CD4 model) to handle “regression to the mean” effects.
One possible summary of a covariate-adjusted association between BMI and CD4 cell count is the visit-specific Pearson correlation between the error terms ɛt and δt. A more interpretable measure is the coefficient from a regression of the CD4 cell count residual ɛt on the BMI residual δt, which is the mean change at visit t in CD4 cell count per-unit change over the same period in BMI, after adjusting out covariate effects. An advantage to fitting separate models, as opposed to a single model in which the change in CD4 cell count is regressed on change in BMI and the adjustment variables, is that the adjustments can be made separately for both change in CD4 cell count and change in BMI—that is, our approach acknowledges that covariates may have different explanatory effects on the 2 processes.
The regression coefficient is computed by first fitting the 2 regression models listed above via weighted least squares, computing the residuals as the difference between observed and fitted values at each time point, and computing the time-specific regression coefficient by regressing residuals ɛt on δt separately at each visit. The models were fitted using Proc GLM in SAS (version 8; SAS), and an estimate of a common slope parameter for BMI was obtained using an unstructured covariance matrix structure using Proc Mixed in SAS (version 8; SAS).
Summary of participant characteristics. Of 871 HIV-positive women, 62 were pregnant during the study and were excluded, and 10 additional women were excluded because of missing baseline values for height or weight. Baseline characteristics of the remaining 799 women are shown in table 1. Overall, the median baseline BMI was 24.6. The median BMI in the 4 BMI weight groups (underweight, normal weight, overweight, and obese) was 18.5, 22.5, 26.9, and 33.6 kg/m2, respectively. There were significant differences by BMI group in risk cohort, CD4 cell count, viral load, race, and number of study visits. Overweight or obese women were less likely to have a history of IDU than normal weight and underweight women. The mean CD4 cell count was progressively higher and the mean log viral load progressively lower as BMI increased. There was no significant difference in age, duration of HIV positivity, income, or education by BMI weight group. The median baseline income level in each BMI group was $501–$1000/month (level 2). Although the number of study visits was significantly different, the mean was >5 in all groups. No participant was receiving HAART at baseline. The median BMI at baseline among the HIV-negative participants was 25.6 (mean, 26.7).
Analysis of time-to-event outcomes. Kaplan-Meier curves for the time to first event, stratified by baseline BMI category, are shown in figure 1 (log-rank P values were all statistically significant). The most prominent effect of baseline BMI was on time to first CD4 cell count <200 cells/mm3, an immunological end point that corresponds with a diagnosis of AIDS ; those with a BMI <20 were most likely to have disease that progressed sooner. Figure 2 shows separate Kaplan-Meier curves for the outcome of first CD4 cell count <200/mm3 for those with a nondetectable baseline viral load, a detectable baseline viral load, a baseline CD4 cell count between 200 and 500/mm3, and a baseline CD4 cell count >500 cells/mm3. In each instance, those with a baseline BMI <20 had disease that progressed more quickly.
Table 2 lists hazard ratios and associated 95% CIs for the outcome rate in each BMI category, relative to the BMI ⩾30 category, after adjusting for baseline values of CD4 cell count, viral load, and ART status. Those with a baseline BMI <20 were more likely to achieve a CD4 cell count <200 cells/mm3 than those in the BMI >30 category (adjusted hazard ratio, 1.59 relative to those with BMI ⩾30; 95% CI, 0.97–2.58); this was borderline significant. With the exception of time to CD4 cell count <200 cells/mm3, there was apparently little association between baseline BMI category and time to event.
We used a discrete-time hazard model (as described above) to examine the effect of time-varying BMI on event rate (table 3). Each column in the table shows the regression results for 1 of 4 outcomes (CD4 cell count <200 or <100 cells/mm3, clinical AIDS–related death, or HIV-related death). For example, for the outcome CD4 cell count <200 cells/mm3 at time t, the model included the most recently recorded prior BMI (BMIt-1), years HIV positive at baseline, the baseline CD4 group, the most recent viral load prior to the current visit (viral loadt-1), the presence of an AIDS-defining clinical condition at the last visit, and baseline depression status. Thus, the BMI effect was estimated conditionally on several very strong markers for the outcomes. Because the first 2 events were defined in terms of CD4 cell count, we adjusted only for baseline CD4 cell count; for the other events, we adjusted for the most recent CD4 cell count. In each case, an increased BMI was associated with slower event rates.
For all time-to-event outcomes, our estimated ORs indicated that women with a higher BMI were less likely to experience HIV-related events, whether clinical, immunological, or mortality-related; for clinical events, the effect was ordered. The most pronounced effect was for HIV-related death, where women in the lowest BMI category had estimated odds of death 3.1 times greater than those in the highest category (95% CI, 1.2–8.1). A similar finding held for clinical AIDS/HIV-related death, where those in the lowest BMI category had odds of progression 3.0 times greater than those in the highest category (95% CI, 1.3–6.7). For the immunological events, the effect of BMI was smaller in magnitude but still strongly positive. The odds of CD4 cell count <100 cells/mm3 was nearly 2 times greater for women in all categories than in those in the highest BMI category, with ORs ranging 1.68–1.87. The association of BMI on CD4 cell count <200 cells/mm3 ranged from 1.2 to 1.5, which indicated a positive trend, but the CIs included 1 for all pairwise comparisons.
Analysis of CD4 cell count and BMI. Figure 3 shows the mean CD4 (±SD) cell count stratified by baseline BMI category, which indicates an ordered effect of BMI with the CD4 cell count at each visit. Those with a greater BMI had a correspondingly greater CD4 cell count across visits. When stratified by baseline viral load and CD4 cell count categories (figure 4), a similar, if slightly attenuated, association was seen. Again, the BMI <20 category appears to distinguish itself from the others. Table 4 summarizes differences in mean CD4 cell count across baseline BMI categories, adjusted for baseline values for CD4 group, viral load group, and ART, and indicates that variation across the categories was statistically significant. Those with BMI <25 at baseline had lower CD4 cell count than those who had a BMI ⩾25.
Finally, to examine the dynamic association between BMI and CD4 cell count, we used the bivariate modeling of residuals described above. Figure 5 depicts the adjusted residual effect of change in BMI on change in CD4 cell count at each visit, after adjustment for covariates. Eight of 11 estimates were positive, and 3 were significantly positive (1 was significantly negative). The range of visit-specific 95% CIs suggests the existence of a common effect, which we found by combining residuals across visits into a single regression model and using robust SEs to adjust for temporal correlations. This gives an overall average change in CD4 cell count of 1.65 cells/mm3 per unit change in BMI (95% CI, 0.05–3.25), which indicates that, at any particular visit, positive changes in BMI tended to be associated with positive changes in CD4 cell count, even controlling for previous CD4 cell count and BMI.
Women with a baseline BMI <25 consistently had lower mean CD4 cell counts over time than women with a BMI >25. Even among the women who entered the cohort with an undetectable baseline viral load or with a baseline CD4 cell count >500 cells/mm3, those who were overweight or obese had higher CD4 cell counts. Regression analysis showed an association between BMI and a risk of HIV progression, whether immunological, clinical, or mortality related.
Increases in fat, lean body mass, or both can result in an increased BMI. The literature suggests that a decreased lean body mass is associated with poorer survival . Specific attempts to increase lean body mass through exercise, androgen, or growth-hormone therapy may result in a net loss of fat in the setting of a maintained or increased BMI [11–13] and could bias an association of increased BMI with improved outcomes. However, HERS participants with increased BMIs likely had increased fat rather than lean body mass. Increases in body weight are predominantly increases in fat mass in HIV-infected persons receiving nutritional supplementation . Although exercise history was not specifically obtained, it is unlikely that a substantial proportion of this largely poor cohort of inner-city women was engaged in systematic rigorous exercise or weight training. Androgen or growth-hormone therapy have not been widely used in women outside of clinical trials, particularly during the years covered by our study (1993–2000). Although 139 HER participants did enroll in other clinical trials during the course of this study, none were enrolled in trials of human growth hormone, and only 1 was enrolled in a trial of the testosterone patch. In addition, 7 women were in clinical trials of megestrol acetate (Bristol-Meyers Squibb), and others may have used marijuana for weight-gain purposes. The results of studies have consistently shown that the weight gained with these latter medications is predominantly fat . Thus, it is likely that the improved outcomes associated with an increased BMI were associated with increased fat mass. Pregnancy did not confound our results, because women with pregnancies during the course of the study were excluded. (However, an analysis that included pregnant women did not show any substantially different results.)
Our findings are similar to those from the NFHL study of HIV-infected women (C.Y.J., E. Tchetgen, D. Jacobson, et al., unpublished data), which found a protective association of fat mass with HIV progression (predominantly CD4 cell decline) in multivariate analysis controlling for viral load, HAART use, and years known to be HIV positive. Lean body mass was not a significant predictor of HIV disease progression in models that also included fat mass. Our results are also consistent with the MIDAS cohort  of HIV-infected persons, which included 43 women, and found both a trend for persons with higher BMI to have slower decline in CD4 cell counts (a decreased percentage of overweight patients had a 25% decline in CD4 cell count) and a significant association of increased BMI with decreased mortality. The findings are consistent with those of Shuter et al. , who found slower progression of HIV disease among the overweight patients. Last, in the results of preliminary studies, an association of fat mass with CD4 cell counts has also been seen in HIV-negative people followed with serial CD4 cell measurements (HIV-negative participants in the Tufts School of Medicine Bienestar cohorts, C.Y.J., J. Forrester, unpublished data).
Although we have explicitly examined longitudinal associations, our findings must be interpreted with caution. It is tempting to conclude that an increased BMI has a protective effect with respect to HIV progression, but our analyses have not demonstrated a biological cause-and-effect relationship. In fact, 2 possible biological mechanisms could give rise to the same observed association, namely, (1) women with lower CD4 cell counts are at greater risk to lose weight or (2) women with higher weight are able to maintain their CD4 cell count to a greater degree. Without having a specific knowledge of the biological mechanism, the choice of dependent variable (CD4 cell count or BMI) is actually an arbitrary modeling decision, even with longitudinal data , which is why they were modeled contemporaneously in the longitudinal regressions.
Nevertheless, there is reason to believe that the second explanation is more likely, which suggests the possibility of a protective effect of obesity. Fat cells are the principal source for the hormone leptin, which circulates in the body in direct proportion to fat cell mass . Leptin levels tend to be higher in women than in men (both in vitro tissue cultures [16, 17] and in vivo), even at the same level of body fat . Leptin levels tend to increase as acute-phase reactants in the case of infection . CD4 cells (as well as other hematologic cells) have leptin receptors, and in vitro leptin supports CD4 cell proliferation in response to other stimuli . Leptin levels fall acutely with starvation (which is associated with an immunocompromised state) in both humans and mice [21, 22]. In a mouse model, leptin supplementation was able to completely reverse starvation-induced defects in delayed hypersensitivity . Leptin supplementation protected normal mice against starvation-induced thymic lymphoid atrophy . In mice, Faggioni et al.  found that starvation-associated increased susceptibility to endotoxic shock caused by the administration of lipopolysaccharide could be ameliorated by the administration of leptin. In an excellent review, Matarese  summarized research on leptin and the immune system. Leptin is a long-chain helical cytokine with structural similarities to IL-6, IL-11, IL-12, leukemia inhibitory factor, granulocyte colony-stimulating factor, and others, and there is structural homology between the leptin receptor and cytokine receptors. It is theoretically possible that leptin might inhibit the deleterious action of some of the cytokines seen at increased levels in HIV infection.
Could the women labeled as “normal weight” have already experienced weight loss compared with their usual weight before entering the study? The question as to whether the “normal weight” women were actually subclinically ill is important, because the association of weight and CD4 outcomes might be confounded by the process responsible for the decreased weight. This concern arose from the observation that the overall prevalence of being overweight or obese (BMI ⩾25) in black women is estimated at 65.8% and in Mexican American women at 65.9% (National Health and Examination Survey III, 1988–1994)  and that women with lower incomes or lower educational levels are more likely to be overweight or obese. When age-adjusted population estimates are examined, an estimated 49% of black women are overweight (BMI >27.5) . The majority of the women included in the HER analysis were black (60.4%) or latina (17.6%), the mean income was $6000–12,000/year, and the prevalence of being overweight or obese (BMI ⩾25) at baseline was 49%. Many of our participants had a history of IDU, which is itself often associated with weight loss . The HER questionnaire did not include a question about usual adult weight. However, the HIV-negative and -positive participants were recruited from the same populations, and the median BMI in the HIV-negative and the HIV-positive participants was quite similar (26.7 and 25.7, respectively).
Although the percentage of women with a history of IDU at baseline was different in the BMI groups, it seems unlikely that this explains the association. Risk cohort (IDU vs. high-risk sexual contact) was not a significant predictor of outcomes in our study. Ickovics et al. , in a study of the HER cohort that specifically looked at the effect of depression, did not find a significant association of IDU with a decline in the CD4 cell count or HIV-related mortality.
Schuman et al.  found that depression was associated with poorer medication adherence in a joint study of very immunocompromised HIV-infected women in the Women's Interagency HIV Study and the HER cohorts, and Ickovics et al.  found depression to be related to increased risk of mortality. However, depression at baseline was not associated with adverse outcomes in our analysis.
The association of fat mass with CD4 cell counts could represent a direct effect of fat or fat-cell products, a correlation of increased fat with better overall nutrition, or an immune-system effect of some specific micronutrient that fatter people have in greater abundance. These possible mechanisms should be clarified in future studies.
There is growing evidence that increased BMI is associated with an increased CD4 cell counts and with lower rates of the events that characterize the progression of HIV disease. Whether this is a protective effect cannot be determined using data from existing epidemiological studies. Given the known adverse consequences of excess fat in the general population, the accumulating body of epidemiological evidence suggests the need for further study to identify the mechanism of this association.
HIV EPIDEMIOLOGY RESEARCH STUDY GROUP MEMBERS
Robert S. Klein, Ellie Schoenbaum, Julia Arnsten, Robert D. Burk, Chee Jen Chang, Penelope Demas, and Andrea Howard (Montefiore Medical Center and Albert Einstein College of Medicine, Bronx, NY); Paula Schuman and Jack Sobel (Wayne State University School of Medicine, Detroit, MI); Anne Rompalo, David Vlahov, and David Celentano (Johns Hopkins University School of Medicine, Baltimore, MD); Charles C. J. Carpenter and Kenneth H. Mayer (Brown Medical School, Providence, RI); and Ann Duerr, Lytt I. Gardner, Scott D. Holmberg, Denise J. Jamieson, Janet S. Moore, Ruby M. Phelps, Dawn K. Smith, and Dora Warren (Centers for Disease Control and Prevention, Atlanta, GA).
We acknowledge Dr. Donna Spiegelman, Dr. Camille Jones, and Dr. Sherwood Gorbach for helpful discussions and manuscript review. We are especially indebted to Dr. Lytt Gardner for his critical review of the manuscript.