Reduced levels of serum albumin, even within the ‘normal range', are associated with increased mortality in a number of populations, including healthy individuals and those with a variety of acute and chronic conditions . This inverse relationship between serum albumin levels and mortality appears to be continuous over a broad range of serum albumin values. For each 25 g/l decrement in serum albumin concentration the odds of death have been reported to increase by 24–56%. The association predicts overall and cause-specific mortality.
The serum albumin concentration, therefore, appears to be a sensitive predictor of preclinical disease and disease severity. The protective effect of higher levels of serum albumin persists after adjustment for other known risk factors, pre-existing illness and the exclusion of early mortality. However, little data exist that address the utility of serum albumin as a prognostic factor in HIV-infected individuals. Several small studies [2–4] with limited follow-up have reported an association between very low albumin levels and death in men. In the present study, we examined the baseline serum albumin level as a prognostic factor for up to 3 years for HIV-1-infected women enrolled in the Women's Interagency HIV Study (WIHS) , a cohort study of the natural history of HIV infection.
The WIHS is a multicenter prospective study of the natural history of HIV-1 infection in women conducted in five United States cities: New York (two sites), Washington DC, Chicago, Los Angeles and San Francisco. The WIHS methods and baseline cohort characteristics have been described previously . From October 1994 to November 1995, 2056 women with confirmed HIV-1 serostatus and 563 seronegative women were enrolled in the WIHS. Six women who had positive HIV-1 serostatus confirmed at the 6 month follow-up visit were considered positive at ‘baseline’ for the purposes of this analysis. Women were recruited through HIV primary care sites, drug treatment facilities and community-based outreach facilities.
Every 6 months, WIHS participants were interviewed using a structured questionnaire, and received a physical examination. Multiple gynecological and blood specimens were collected at each visit. Notification of the death of a participant was continuously obtained from friends, relatives and medical providers of the participants, and at some sites, through local death registries. Death certificates were requested for all women known to have died, and the date of death was ascertained from the following sources in descending order of priority: death certificate, medical records, medical provider, and family/friends. Follow-up visits to 3 April 1998 were analysed. Data were censored at the date of last interview for those not known to have died, or on 3 April 1998 for those who died after this date.
Whole peripheral blood was collected at baseline and every 6 months thereafter. Tubes were centrifuged locally if plasma could not be separated within 6 h and shipped to a central laboratory where final separation was performed. Plasma was aliquoted and stored at −80°C. Quantification of HIV-1 RNA in plasma was performed using the isothermal nucleic acid sequence-based amplification method (Organon Teknika Corp., Durham, NC, USA). The HIV-1-RNA determinations were performed in laboratories that participated in the National Institutes of Health, Virology Quality Assurance Laboratory Proficiency Testing Program. The lower limit of quantification was 4000 copies/ml. T cell subsets were determined using standard flow cytometry performed in laboratories certified by the AIDS Clinical Trials Groups.
Except where noted, analysis included all women who had at least one follow-up visit or were known to have died after the baseline visit, and who had baseline HIV-1 plasma RNA determination and CD4 cell counts performed (N = 1818). The numbers of patients differ in several analyses because complete information for each patient was not always available. The survival time was measured from the date of baseline WIHS visit. All-cause mortality by baseline albumin levels was calculated by the Kaplan–Meier method and summarized as the probability of dying within 36 months.
Cox's proportional hazard models were used to assess the relative risk of mortality among the various subgroups considered. Albumin was examined as a continuous and grouped variable. The grouping was based on standard definitions of clinical normality and the absence of a difference in mortality among groups with albumin levels above 42 (g/l). As the results were similar for albumin as a continuous and grouped variable, only data on grouped albumin are presented. The covariates age, body mass index (BMI), CD4 cell group, viral load group, and hematocrit group were included in most models. Groupings of CD4 cell counts, viral load, BMI and hematocrit were based on accepted standards and previous reports . Interactions between serum albumin and the covariates were evaluated as was the goodness of fit of the models. Because the level of serum albumin is influenced by liver and kidney function, separate models were examined for subgroups of patients with normal organ functioning as measured by creatinine, blood urea nitrogen (BUN), alanine aminotransferase (ALT) and aspartate aminotransferase (AST) levels. This allowed estimation of the ability of serum albumin to predict mortality independent of kidney and liver function. The proportionality assumption [i.e. constant relative hazard (RH) throughout the study] was tested in the various models and was found to be valid for baseline albumin levels. Nonetheless, a stratified analysis of the RH by baseline serum albumin was performed for the 20 categories formed by combining the four CD4 cell count and five HIV-1-RNA groupings. Proportional hazard models were stratified by the site of enrolment in order to adjust for differences in the ascertainment of deaths as well as any other sources of heterogeneity.
The general health status of 79% of the women in the WIHS was considered to be healthy by the examining clinician at enrolment. The most common chronic condition (other than HIV-1 infection) in the WIHS cohort was hypertension, with a prevalence of 18%. The distributions of CD4 cell count, hematocrit, albumin, BMI and HIV-1-RNA groupings are shown in Table 1. Median values were 324/mm3, 37.1%, 42 g/l, 25.5 kg/m2 and 22 230 copies/ml, respectively. At the time of analysis, 160 patients had not returned even for a single follow-up visit. These patients did not differ from patients with follow-up either in age (P = 0.37), race/ethnicity (P = 0.87), CD4 cell count at baseline (P = 0.37) or HIV-1-RNA grouping (P = 0.88) (data not shown).
HIV-infected women had a significantly lower mean value of serum albumin than did uninfected women [4.16 g/l (0.48) versus 4.43 g/l (0.42);P < 0.001]. Among HIV-infected women the distribution of baseline serum albumin was associated with the HIV-1 transmission category, groupings of CD4 cell counts, HIV-1-RNA l levels and hematocrit levels (P < 0.001) (Table 2). There were weak but statistically significant correlation coefficients between the serum albumin value and CD4 cell count (Spearman rank correlation 0.10), viral load (−0.14) and hematocrit (0.28).
At the time of this analysis, 1728 women (85%) were still in follow-up (mean, 28.9; median, 30.6 months); there were 268 deaths. The cumulative 3 year mortality was 16.7% (standard error 1.0%). The baseline serum albumin level was significantly associated with mortality (P < 0.001). Women with the lowest albumin levels (< 35 g/l) had a cumulative hazard of death of 48%, compared with 21, 16 and 11% for women with levels of 35–40, 40–42 and 42 g/l or greater, respectively (Fig. 1). As shown in Table 3, the association of serum albumin level with mortality persisted after adjustment for CD4 cell count, HIV-1-RNA level, hematocrit level, BMI and age. The adjusted RH of death by albumin level was 3.1 [95% confidence interval (CI) 2.1–4.7] in those with albumin levels of less than 35 g/l and declined to 1.5 (95% CI 1.0–2.1) in those with levels of 40–42 compared with the reference group with levels of 42 or greater. The adjusted risk of death varied consistently by albumin level at each of the six centers in the five different geographical areas (data not shown).
Significant associations with mortality were also seen for the CD4 cell count, HIV-1-RNA level, hematocrit and BMI. The RH for women with CD4 cell counts of less than 50/mm3 compared with women with CD4 cell counts greater than 350/mm3 was 11.1 (95% CI 7.1–17.4) but for patients with CD4 cell counts of 50–199/mm3 the RH of 3.2 (2.1–5.1) was almost the same as for an albumin level of less than 35 g/l. The RH for women with over 500 000 copies/ml compared with women with a viral load of less than 4000 copies/ml was 6.3 (95% CI 3.5–11.3). At levels between 4000 and 500 000 copies/ml the RH was approximately double that of patients with levels of less than 4000 copies/ml. Patients with abnormally low hematocrit (< 35%) levels had double the RH of patients with levels of 40% or greater.
The relationship between the serum albumin level and mortality persisted when the analysis was limited to the 1451 women who were not acutely ill at enrolment and had normal baseline creatinine, BUN and ALT and AST of less than 80 IU/l. The RH for patients with serum albumin levels below 35 g/l compared with women with albumin levels of 42 g/l or higher was 3.3-fold (95% CI 2.0–5.4), a ratio similar to that seen when all patients were included. The RH of 1.7 (95% CI 1.2–2.4) for patients with albumin levels of 35–40 g/l and 1.3 (95% CI 0.9–2.0) for patients with levels of 40–42 g/l were also similar. The relative risk of death in this subgroup by CD4 cell category, hematocrit level and viral load also remained similar to that of the entire group (data not shown).
When the above subset was restricted to those patients with normal baseline creatinine, BUN and ALT and AST of less than 80 IU/l, and CD4 cell counts over 50 at enrolment, serum albumin of less than 35 g/l remained a powerful predictor of mortality after adjustment for the other covariates (RH, 3.4 95% CI 1.4–8.2). When this subset was further limited to patients with CD4 cell counts of over 200 (N = 918; deaths 34) the RH of death for a serum albumin level less than 35 g/l increased to 8.3-fold (95% CI 2.3–30.2). In addition to serum albumin level, an age of 45 years or over (RH 3.8; 95% CI 1.1–14.1), HIV-1 RNA 500 000 copies/ml or higher (RH = 5.7; 95% CI 1.5–21.2), CD4 cell count 200–349 (RH 2.6; 95% CI 1.1–6.5) and BMI of less than 20/kg/m2 (RH 5.3; 95% CI 1.9–15.2) were also significant predictors of excess mortality in this subset.
The prognostic value of serum albumin was examined in patients with varying levels of ALT and AST (Table 4). The RH for a patient with a serum albumin level of less than 35 g/l ranged from 5 to 13-fold greater than for a patient whose albumin level was over 42 g/l, depending on the range of ALT and AST and whether the other prognostic factors were included as covariates.
Finally, the impact of the serum albumin level on mortality was estimated by allowing the baseline hazards to vary within each of the 20 strata formed by combining the four CD4 count groupings and the five HIV-1-RNA groups. In this analysis the RH of death across strata for patients with serum albumin levels of less than 35, 35–40 and 40–42 g/l versus greater than 42 g/l was 2.9 (95% CI 2.0–4.2), 1.8 (95% CI 1.4–2.5) and 1.4 (95% CI 1.0–2.1). The RH by serum albumin levels across strata remained similar when the stratified analysis was restricted to patients with normal kidney and liver function as defined previously.
In this large prospective study of HIV-1-infected women in North America, baseline serum albumin was a major independent predictor of survival. The correlation of serum albumin with other markers of HIV disease progression reinforces its validity as an indicator of disease progression. Even after adjusting for these factors and other important covariates, the risk of death for women with baseline serum albumin levels of less than 35 g/l was similar or greater than that for women with a very low hematocrit, low BMI, moderately low CD4 cell count or moderately high HIV-1-RNA level. Only women with very low CD4 cell levels (< 50/mm3) or very high HIV-1-RNA levels (500 000 or over copies/ml) had higher adjusted RH of death. Compared with women with the highest baseline albumin levels of 42 g/l or greater even a low normal (35–40 g/l) or mid-normal (40–42 g/l) level was associated with a significant increase in the 3 year cumulative mortality.
The WIHS cohort is broadly representative of HIV-infected women in the USA. Among women in the United States with AIDS reported to the Centers for Disease Control and Prevention there was a greater proportion of injection drug users (47 versus 34%) and fewer unknown risk category (12 versus 20%) but race, ethnicity, and age were similar . The median age of 36.3 years in the WIHS was almost the same as the median age of 36.1 in the national series, although there was a higher percentage of WIHS patients 40–49 years of age (27 versus 19%). It does not seem likely that these differences would impact on serum albumin as a predictor of mortality. In any event, serum albumin predicted survival for a variety of different patient subgroups and at each of six different centers in five geographical locations, as noted earlier.
The magnitude of the adjusted risk of mortality for women with albumin levels of less than 35 g/l was similar to that for patients with CD4 cell counts of 50–199/mm3 and viral loads of 20 000–500 000 copies/ml. Women with albumin levels of 35–40 g/l (which falls in the supposed ‘normal range') were almost twice as likely to die as women with levels of 42 g/l or higher, other factors being equal. Serum albumin was predictive of mortality in patients who were not acutely ill and had normal kidney and liver function as measured by levels of creatinine, BUN and transaminase. Serum albumin remained a powerful predictor of mortality when severely immunosuppressed patients (CD4 cell count < 50) were excluded. When women with advanced immunosuppression (CD4 cell count < 200) were excluded, albumin less than 35 g/l and viral load of over 500 000 copies/ml were the dominant predictors of survival in this important subgroup. Even when limiting patients to those with transaminase levels of 40 IU/l or less the serum albumin level was an important predictor of mortality.
Serum albumin was a better predictor of mortality in women with CD4 cell counts over 200 than under 200, suggesting that albumin may be associated with causes of mortality other than those related directly to AIDS. Analyses that adjusted or stratified for laboratory markers of renal and liver dysfunction, the most common medical illnesses in HIV-1-infected persons resulting in low serum albumin, actually increased the magnitude of the association of serum albumin and mortality. It is possible that these laboratory markers are not sensitive markers of these conditions. A limitation of our study is the absence of cause-specific mortality. However, ascertainment of the specific cause of death in HIV-1-infected patients without autopsy confirmation is problematical. Nonetheless, it is clear that the vast majority of patients who died had AIDS-related conditions, because half the women who died had CD4 cell counts of less than 50 at baseline, and another 26% had counts between 50 and 200. Simply based on CD4 cell counts, over three-quarters of the patients who expired would be classified as having had AIDS before death.
The data suggest that serum albumin may be a particularly good predictor in the earlier stages of HIV before patients have progressed to AIDS. It is well established that hypoalbuminemia is a marker of malnutrition . Because albumin is affected by protein–energy malnutrition induced by the stress of illness or injury, a low albumin level may reflect poor nutritional status at early stages of disease before changes in body weight or other clinical markers . This is a time when prognostic indicators may be especially useful in planning future therapy.
Treatment for HIV infection was not considered in the present analysis. Less than 1% of the study women reported the use of highly active antiretroviral therapy (HAART) before the baseline visit, and only a small proportion of the measured follow-up time could have been influenced by the receipt of such therapy . If HAART disproportionately prolonged the survival of women with low albumin levels, then the predictive power of serum albumin would be diminished by the widespread use of HAART, although there are presently no data to suggest this.
It is possible that the association between serum albumin concentration and mortality is caused by albumin's association with other variables that may influence immunosuppression or mortality. Diminished CD4 T cell function has been associated with falling serum albumin and rising serum immunoglobulin levels reversing the albumin : globulin ratio. This has been reported for cirrhosis as well as AIDS and occasionally diabetes . Low albumin levels may be indicative of other conditions such as hepatitis C, which is associated with injection drug use and may hasten mortality. However, the HIV-1 transmission category was not related to survival in the present analysis (P = 0.30), and including it as a covariate in the model did not alter the relationship of serum albumin to mortality. The association of serum albumin with mortality in the current study actually increased when study subjects were limited to immunocompetent patients with normal kidney and liver function.
A more subtle confounding may result from the inverse association of albumin with markers of inflammation such as C-reactive protein or cytokines such as TNF or IL-1. Several of these markers have been associated with the pathogenesis of various diseases and can also depress serum albumin levels [9,10]. A recent review  found convincing evidence of an association of fibrinogen, C-reactive protein, albumin and leukocyte count with the risk of coronary heart disease. Another study  found that markers of inflammation including low serum albumin were associated with an increased risk of diabetes mellitus in middle-aged adults. Serum albumin has been inversely associated with IL-6 levels in a number of clinical conditions, and IL-6, a pro-inflammatory cytokine, has been shown to upregulate HIV-1 replication strongly in vitro[13,14]. Serum albumin may thus be a marker for cytokine activity, which may be only partly accounted for by the CD4 cell count and HIV-1-RNA level.
Another possibility is that serum albumin has a direct protective effect. Several possible mechanisms have been proposed including albumin's effect as a vasodilator, its antihemostatic and platelet-lowering properties, its antioxidant activity and its antiviral activity [1,13,14]. Even if serum albumin does have a direct protective effect, it is not clear that raising serum albumin levels would be beneficial.
Serum albumin was a major independent predictor of 3 year mortality in a large North American cohort of HIV-1-infected women. It remains to be determined whether baseline serum albumin will predict survival beyond 3 years. The potential influence of HAART and other therapies on this relationship require further study. Nonetheless, the serum albumin level is an inexpensive and widely available measurement, and appears to provide important prognostic information for HIV-1-infected patients that complements the CD4 cell count and HIV-1-RNA levels. It may be particularly useful in developing countries where these more expensive assays are often not available.
1. Goldwasser P, Feldman JG. Association of serum albumin and mortality risk.
J Clin Epidemiol 1997, 50: 693 –703.
2. Justice AC, Feinstein AR, Wells CK. A new prognostic staging system for the acquired immunodeficiency syndrome.
N Engl J Med 1989, 320: 1388 –1393.
3. Guenter P, Muurahainen N, Simons G, Kosok A, Cohan GR, Rudenstein R, Turner JL. Relations among nutritional status, disease progression and survival in HIV infection.
J Acquired Immune Defic Syndr 1993, 6: 1130 –1138.
4. Suttmann U, Ockenga J, Selberg O, Hoogestraat L, Deicher H, Muller MJ. Incidence and prognostic value of malnutrition and wasting in human immunodeficiency virus infected outpatients.
J Acquired Immune Defic Syndr 1995, 8: 239 –246.
5. Barkan SE, Melnick SL, Preston-Martin S, Weber K, Kalish LA, Miotti P. et al. The womens interagency study (WIHS)-design, methods, cohort characteristics and comparison with reported AIDS cases in US women.
Epidemiology 1998, 9: 117 –125.
6. Anastos K, Kalish L, Hessol N. et al. The relative value of CD4 cell count and quantification HIV-1 RNA in predicting survival in HIV-1 infected women: results of the women's interagency HIV study.
AIDS 1999, 13: 1717 –1726.
7. Gibbs J, Cull W, Henderson W, Daley J, Hur K, Khuri SF. Preoperative serum albumin level as a predictor of operative mortality and morbidity.
:Results from the national VA surgical risk study.
Arch Surg 1999, 134: 36 –42.
8. Henry JB. Clinical diagnosis and management by laboratory methods
, Ch 2, 19th ed. Philadelphia, PA: W.B. Saunders Co.; 1996.
9. Tracey KJ, Cerami A. Tumor necrosis factor, other cytokines and disease.
Annu Rev Cell Biol 1993, 9: 317 –343.
10. Bellomo R. The cytokine network in the critically ill.
Anesth Intens Care 1992, 20: 288 –302.
11. Danesh J, Collins R, Appleby P, Peto R. Association of fibrinogen, C-reactive protein, albumin, or leukocyte count with coronary heart disease.
:Meta analysis of prospective studies.
JAMA 1998, 279: 1477 –1482.
12. Schmidt MI, Duncan BB, Sharrett AR. et al. Markers of inflammation and prediction of diabetes mellitus in adults (Atherosclerosis Risk in Communities Study): a cohort study.
Lancet 1999, 353: 1649 –1652.
13. Ward T, Powell RM, Evans DJ, Almond JW. Serum albumin inhibits echovirus 7 uncoating.
J Gen Virol 1999, 80: 283 –290.
14. Kuipers ME, Berg M, Swart PJ, Laman JD, Meijer DK, Koppelman MH, Huisman H. Mechanism of anti-HIV activity of succinylated human serum albumin.
Biochem Pharmacol 1999, 57: 889 –898.
Data in this manuscript were collected by the WIHS Collaborative Study Group with centers (principal investigators) at New York City/Bronx Consortium (Kathryn Anostos); Brooklyn, NY (Howard Minkoff); Washington DC Metropolitan Consortium (Mary Young); The Connie Wofsy Study Consortium (Ruth Greenblatt, Herminia Palacio); Los Angeles County/Southern California Consortium (Alexandra Levine); Chicago Consortium (Mardge Cohen); Data Coordinating Center (Alvaro Munoz, Stephen J. Gange).
Keywords:© 2000 Lippincott Williams & Wilkins, Inc.
proportional hazards; serum albumin; survival