Introduction
Several studies have reported on the poor short-term prognosis associated with a CD4 cell count below 50 × 106/l in HIV-1-infected patients [1-3]. Estimates of the median survival from the first CD4 cell count below 50 × 106/l range from 12 to 18 months [1-3]. In a previous study, we found that the hazard of death increased sixfold in those patients whose CD4 count fell below 50 × 106/l, compared with those who maintained a count above 200 × 106/l [4]. However, even with advanced immunodeficiency, patients vary considerably in their subsequent clinical course, and it is apparent that there exists a subgroup of patients with prolonged survival. To date, only one published study has systematically evaluated the reasons for the relative longevity of these patients [3]. An understanding of the factors that contribute to long-term survival despite advanced HIV disease is important for the following reasons: (i) to gain insights into the natural history and pathogenesis of late-stage HIV infection; (ii) to identify prognostic factors for survival in late-stage disease, relevant to the design and analysis of clinical trials; and (iii) to develop a profile of clinical and laboratory characteristics in those patients at a particularly high short-term risk of death, for whom it may be appropriate to intensify treatment.
We have analysed more than a thousand patients with advanced immunodeficiency (CD4 cell count ≤ 50 × 106/l), in order to identify demographic, clinical and laboratory factors associated with survival. We have also developed a user-friendly prognostic scoring system, based on clinical and laboratory data available routinely to the physician in the clinic setting.
Methods
Study population
Eligible patients were 1284 HIV-1-seropositive subjects attending two hospital-based HIV clinics in West London, UK who had had at least one CD4 count ≤ 50 × 106/l between June 1985 and June 1993. The subjects were further categorized into incident cases (n = 769, patients who had a previous CD4 cell count > 50 × 106/l documented, which subsequently fell below 50 × 106/l during follow-up), and prevalent cases (n = 515, patients whose first documented CD4 measurement was already < 50 × 106/l).
Data collection
A computerized medical record was maintained routinely on all HIV-infected patients registered at the two HIV clinics. Each record contained information on demographic, clinical, and laboratory data at clinic enrolment and during follow-up. Linked pharmacy records provided information on date of initiation of antiretroviral therapy and Pneumocystis carinii pneumonia (PCP) prophylaxis. All patients were reviewed periodically and their T-lymphocyte subsets measured by flow cytometry (FACScan, Becton-Dickinson, Cowley, Oxfordshire, UK) as clinically indicated. The same flow cytometer, monoclonal antibodies, analytical procedures and sample preparation were used throughout the study.
The following data were abstracted for each of the 1284 patients with a CD4 cell count ≤ 50 × 106/l (CD4 ≤ 50): gender, year of HIV-positive diagnosis, transmission risk group, age at and year of first CD4 ≤ 50 × 106/l, history and duration of initial AIDS-defining diagnosis prior to CD4 ≤ 50 × 106/l, history of other AIDS-defining diagnoses, duration of antiretroviral and PCP prophylaxis use prior to CD4 ≤ 50 × 106/l, rate of CD4 decline prior to CD4 ≤ 50 × 106/l, total white blood cell count, neutrophil and lymphocyte counts, haemoglobin level, and absolute CD4 and CD8 cell counts at CD4 ≤ 50 × 106/l. Laboratory values at CD4 ≤ 50 × 106/l were taken as those measurements performed at the same time as the CD4 ≤ 50 measure, or the nearest measurement to CD4 ≤ 50 within the preceding 2 months. The patient's HIV clinical status during follow-up was categorized according to the Centers for Disease Control 1987 classification [5]. Information on deaths was verified by cross-checking with the UK Office of Population Census and Surveys.
Survival analysis
Patients still alive on 30 June 1993 were censored at the date of their last assessment. Patients whose last follow-up had occurred more than 4 months prior to that date were considered lost to follow-up. Survival time was calculated from the date of the first documented CD4 ≤ 50 × 106/l to the date of death or last follow-up alive. Survival curves were plotted using the Kaplan-Meier product-limit method [6] and differences between subgroups of patients were tested using the log-rank test [7]. A proportional hazards regression model was used to estimate the independent effect of the clinical and laboratory variables measured at CD4 ≤ 50 × 106 cells/l [8]. Continuous variables were stratified and analysed according to either tertiles, quartiles or the median value of their distribution. The rate of CD4 decline before CD4 ≤ 50 was estimated in the subgroup of 769 incident cases, by linear regression of the serial CD4 measurements on calendar time. The relative contribution of the different variables to survival outcome was estimated using a multivariate regression model and a forward selection procedure. The proportional hazards assumption was checked using graphical methods by examining log [-log(survival probability)] versus log(time) plots for each of the variables [9].
Prognostic survival score
A prognostic score based on the best multivariate model was derived in the 769 patients with an incident CD4 ≤ 50 value. Numerically, the prognostic score represents a linear combination of the estimated k coefficients (b1, b2, …, bk) obtained in the final model with k variables (x1, x2, …, xk) and is calculated for each patient as follows [9]:
Equation (Uncited)Image Tools
Case-control analysis
We also performed a nested case-control analysis, and compared long-term survivors (LTS, patients who survived 3 years or more from CD4 ≤ 50) with short-term survivors (STS, patients who died within 1 year of CD4 ≤ 50 × 106 cells/l). Each LTS was matched on year of first HIV-positive diagnosis with three STS selected randomly from the whole group of STS (n = 365). Comparisons between LTS and STS were performed using a Student t test for normally distributed continuous variables, and a Mann-Whitney U test for variables with a skew distribution. The distribution of categorical variables were compared with a χ2 test.
Results
A total of 1284 HIV-positive patients who had at least one documented CD4 cell count of ≤ 50 × 106/l were identified from our clinic database. Study participants were predominantly men (n = 1254, 98%) and homosexual (n = 946, 92%). In 831 patients (64%), the year of the first CD4 ≤ 50 was after 1990. Study participants were categorized according to the duration of AIDS prior to CD4 ≤ 50: 528 (41%) were still AIDS-free at CD4 ≤ 50, 272 (21%) had AIDS for less than 3 months prior to CD4 ≤ 50, 249 (19%) had AIDS for 4-11 months, and 235 (19%) had AIDS for more than 12 months prior to CD4 ≤ 50. Overall, 674 (52%) had received zidovudine (ZDV) and 489 (38%) PCP prophylaxis prior to CD4 ≤ 50. The proportion of patients who had received ZDV increased according to year of CD4 ≤ 50, from 18.2% in 1985-1987 to 42.2% in 1988-1990 and 57.0% in 1991-1993. The same trend was observed for the use of PCP prophylaxis, from 6.4% in 1985-1987 to 20.3% in 1988-1990 and 56.6% in 1991-1993. The median duration of ZDV and PCP prophylaxis use prior to CD4 ≤ 50 was 11.3 months (interquartile range, 5.9-19.0 months) and 9.0 months (interquartile range, 4.5-15.6 months), respectively. The mean CD4 cell count at CD4 ≤ 50 was 30.3 × 106/l (SD, 13.4 × 106/l), and the median interval between CD4 count measurements was 2 months (range, 1 week to 6 months).
Differences between patients with an incident versus prevalent CD4 ≤ 50 value
In patients with an incident CD4 ≤ 50 (n = 769), the mean ± SD CD4 count at CD4 ≤ 50 was significantly higher (34.6 ± 11.9 × 106/l) than in patients with a prevalent CD4 ≤ 50 (n = 515; 23.9 ± 13 × 106/l; P = 0.01). The proportion of patients who had received ZDV and PCP prophylaxis prior to CD4 ≤ 50 was also higher in incident (74.3 and 56.8%, respectively) than in prevalent CD4 ≤ 50 patients (20 and 8.9%, respectively; P < 0.001). There were no other statistically significant differences between incident and prevalent patients.
Univariate predictors of survival
As of 30 June 1993, 813 (63.3%) of the 1284 eligible patients had died. After a median follow-up of 13 months, overall survival from CD4 ≤ 50 was 67.1% at 1 year, 29.8% at 2 years, and 12.9% at 3 years. The median survival was 17.1 months (range, 0.1-79.4 months). As expected, incident patients had a marginally better survival than prevalent patients, but this difference was not statistically significant (Fig. 1).
The nature of the initial AIDS diagnosis was an important predictor of survival. Patients with a diagnosis of Kaposi's sarcoma or lymphoma had a shorter survival compared with those whose initial AIDS-defining diagnosis was an opportunistic infection (P = 0.006). Age, actual CD4 and CD8 lymphocyte, and haemoglobin value at CD4 ≤ 50, as well as duration of AIDS prior to CD4 ≤ 50, were also all strong determinants of survival (P < 0.0001; Table 1). Among the subgroup of patients with an incident CD4 ≤ 50 value, a greater prior rate of CD4 decline was associated with a poorer survival (P = 0.001). Factors not associated with survival included year of HIV-positive diagnosis, gender, risk group status, and total white blood cell and neutrophil count.
Multivariate survival analysis
We performed a pooled analysis which included patients with both an incident as well as a prevalent CD4 ≤ 50 × 106 cells/l. Four variables remained independently associated with survival: age, CD4 count, haemoglobin value at CD4 ≤ 50 × 106 cells/l, and duration of AIDS prior to CD4 ≤ 50 × 106/l. The risk of dying increased by 2% for each year of age [95% confidence interval (CI), 1-3], by 10% (95% CI, 3-16) for each 10 × 106/l decrease in CD4 count at CD4 ≤ 50, and by 14% (95% CI, 9-18) for each 1 g/dl decrease in haemoglobin level at CD4 ≤ 50. Similar results were obtained when transformed data, either log(CD4) or log(haemoglobin) were used. When compared with patients who were AIDS-free at CD4 ≤ 50, the risk of dying was 1.5-fold (95% CI, 1.2-1.9) that of patients who had an AIDS diagnosis for less than 3 months prior to CD4 ≤ 50, 1.8-fold that of patients with an AIDS diagnosis for 4-11 months prior to CD4 ≤ 50, and twofold that of patients with AIDS for ≥ 12 months prior to CD4 ≤ 50 × 106 cells/l. After controlling for these four variables, univariate factors that were no longer predictive of survival were an AIDS index diagnosis of Kaposi's sarcoma or lymphoma, and the total lymphocyte and CD8 count at CD4 ≤ 50 × 106 cells/l.
A multivariate hazards analysis was repeated after incorporating the rate of CD4 decline into the regression model in the subgroup of patients with an incident CD4 ≤ 50 × 106/l (Table 2). In this analysis, the actual CD4 count value at CD4 ≤ 50 was no longer predictive of survival. In contrast, the rate of CD4 decline emerged as one of the strongest independent predictors of survival, after adjusting for age and haemoglobin level at CD4 ≤ 50, and duration of AIDS prior to CD4 ≤ 50. The risk of dying for patients with an average 6 monthly CD4 loss of > 40 × 106/l was 1.71 times (95% CI, 1.31-2.27) that of patients with an average loss of < 40 × 106/l.
To determine whether the negative prognostic impact of haemoglobin at CD4 ≤ 50 × 106/l was mediated through use of ZDV, we examined the relationship between these two variables. Although there was a trend towards a lower mean haemoglobin at CD4 ≤ 50 in the 538 patients who received ZDV prior to CD4 ≤ 50 (11.7 ± 1.7 g/dl), compared with the 388 who had not (12.0 ± 1.7 g/dl), this did not reach statistical significance (P = 0.06). Similarly, we found no significant trend between haemoglobin level at CD4 ≤ 50 and duration of ZDV use prior to CD4 ≤ 50.
Prognostic score of survival
A prognostic score for the likelihood of death was developed based on the subgroup of CD4 ≤ 50 incident patients, and derived from the best multivariate model. Variables included in the model were age and haemoglobin at CD4 ≤ 50, and duration of AIDS and rate of CD4 decline prior to CD4 ≤ 50 (Table 3). The coefficient value for each of these variables was divided by the lowest coefficient value observed (b = 0.028 for age), and then rounded in order to simplify the calculations [9]. The sequence of calculations to derive the prognostic score is given below:
We stratified the 769 patients in our cohort into quintiles of the prognostic score, based on their individual profiles of age, haemoglobin values at CD4 ≤ 50, and duration of AIDS and rate of CD4 decline prior to CD4 ≤ 50. Table 3 presents the predicted survival of the subjects in each quintile of the prognostic score, where a low score denotes a better survival. For example, among those patients with scores in the lowest quintile, 89% survived 1 year, 42% survived 3 years, and the median survival was 29.2 months. In contrast, among those patients in the highest score quintile, 80% were predicted to survive 6 months, but only 10% at 2 years, with a median survival of 12.8 months.
These results were used to generate predicted survival curves using the proportional hazards model, corresponding to three hypothetical patient profiles (Fig. 2): an 'average prognosis' patient (median prognostic score), a 'poor prognosis' patient (prognostic score in the highest quintile), and a 'good prognosis' patient (prognostic score in the lowest quintile).
Case-control analysis
A case-control analysis was performed to compare the characteristics of LTS (n = 55) at CD4 ≤ 50 with STS (n = 165). LTS differed significantly from STS on the same clinical and laboratory characteristics identified in the multivariate model of predictors of survival (Table 4). In particular, LTS were more likely to be older (median, 31 versus 37 years; difference in median age, 6 years), to be AIDS-free at CD4 ≤ 50 (65.5 versus 27.9%; difference in frequency, 39%), to have a higher CD4 cell count at CD4 ≤ 50 (37 versus 28 × 106/l; difference in median count, 11 × 106/l), to have a higher haemoglobin level at CD4 ≤ 50 (13.0 versus 11.2 g/dl; difference in median value, 1.8 g/dl), and a lower rate of CD4 decline per 6 months prior to CD4 ≤ 50 (median loss, 33.3 versus 45.2 × 106/l; difference in median loss, 12 × 106/l per 6 months).
Discussion
In this analysis of a large cohort of HIV-infected patients with advanced immunodeficiency (CD4 cell count ≤ 50 × 106/l), we have identified clinical and laboratory characteristics associated with a more favourable survival outcome. These included a younger age and higher haemoglobin value at the first CD4 cell count ≤ 50 × 106/l, the absence of an AIDS diagnosis, and a slower rate of CD4 cell decline prior to CD4 ≤ 50 × 106/l. These findings were consistent between the two analytical approaches employed: a multivariate survival analysis using a proportional hazards model, and a nested case-control comparison. They were also in agreement with the two other studies of predictors of survival from a CD4 count < 100 × 106/l [10], and a CD4 count < 50 × 106/l [3], although rate of CD4 decline was not examined in either of these studies. Table 5 summarizes the prognostic findings in these various studies of advanced immunodeficiency. There was also agreement among these studies on factors not associated with outcome, including year of HIV-positive diagnosis, gender, risk group status, total white blood cell, neutrophil and CD8 count. An insufficient number of our patients had other potentially important prognostic factors such as p24 antigen levels [11] and weight [12] measured to include them in our analysis.
It is noteworthy that the actual CD4 count at CD4 ≤ 50 was no longer significantly associated with survival in incident cases, after adjustment for the prior rate of CD4 cell decline. The important independent contribution of rate of CD4 cell decline has been documented previously in both ZDV-treated [4] and ZDV-naive patients [13]. This implies that in the setting of advanced immunodeficiency, the actual CD4 value becomes less discriminating as a prognostic marker, and is displaced by the rate of CD4 decline. Patients with a sustained high rate of CD4 decline in the setting of advanced immunodeficiency are a subgroup who might benefit from further intensification of their antiretroviral regimen. We now propose to assess and quantify the relative prognostic value of viral load prior to and at CD4 ≤ 50 in a subgroup of these individuals.
The predictive value of haemoglobin on survival even after adjustment for CD4 level and clinical status has been a consistent finding in other studies of patients with advanced HIV infection [10,14,15] and malignant disease [16,17]. The precise pathogenesis of HIV-associated anaemia is unclear, but is likely to be multifactorial. Contributory factors to the dysregulation of haematepoesis observed in advanced HIV disease include an imbalance in cytokine production secondary to chronic immune activation, and a direct effect of HIV on the survival and proliferative capacity of haematopoetic progenitor cells [18]. The picture is further complicated by the frequent occurence of coexistent opportunistic infections and neoplasms, and the cytotoxic effect of antiretroviral therapies. Importantly, the observed association between a low haemoglobin and poor survival in our analysis was not explained by a greater frequency of ZDV use and therefore ZDV-induced anaemia among poor prognosis patients. Whether correction of anaemia with, for example recombinant erythropoietin [19], will have an impact on survival as well as on morbidity warrants further investigation. This finding has practical implications, given that monitoring of haemoglobin is routine and given the magnitude of its predictive value.
We failed to find a significant association between prior use of ZDV or PCP prophylaxis and survival. This is in agreement with an analysis from the Multicenter AIDS Cohort Study by Apolonio et al. [3] who reported that initiation of ZDV after (but not before) the CD4 count fell below 50 × 106/l was predictive of survival. Since the focus in this analysis was on clinical and laboratory data that would be available routinely to the physician in the clinic at the time of a patient presenting with a CD4 count < 50 × 106/l, we did not attempt to assess the value of initiating ZDV after the CD4 count had fallen below this level. One possible explanation for the absence of a significant impact of ZDV on survival is the possibility of a selection bias, whereby those who received ZDV were more likely to have exhibited progressive disease. This is further compounded by the considerable changes in antiretroviral use that have taken place during the 8-year follow-up period.
Limitations to our study include, first, the absence of a standardized schedule of laboratory tests, since haemoglobin level and CD4 counts were only measured when clinically indicated, and second, the potential misclassification of some patients who may have had just a random CD4 count < 50 × 106/l, given the considerable random fluctuation in the measurement of the CD4 cell count [20] and our requirement for only a single count < 50 × 106/l for inclusion in the analysis. However, a preliminary examination of our data has revealed that only a very few of our patients had subsequent CD4 cell counts > 50 × 106/l. Finally, our prognostic score needs further validation using data from other HIV cohorts, to establish the robustness of our findings, especially in other risk groups.
In summary, routinely available clinical and laboratory data including haemoglobin level, rate of CD4 decline and time from initial AIDS diagnosis can be translated into a prognostic score, which can in turn be used to predict the survival experience of an HIV-infected patient with advanced immunodeficiency. This can be performed easily in the outpatient clinic, through the application of a simple formula [21].
References
1. Easterbrook PJ, Emami J, Gazzard B, et al.: Progressive CD4 cell decline and death in zidovudine-treated patients. J Acquir Immune Defic Syndr 1993, 6:927-929.
2. Yarchoan R, Venzon DJ, Pluda JM, et al.: CD4 count and the risk of death in patients infected with HIV receiving antiretroviral therapy. Ann Intern Med 1991, 115:185-189.
3. Apolonio EG, Hoover DR, He Y, et al.: Prognostic factors in human immunodeficiency virus-positive patients with a CD4+ lymphocyte count < 50/ml. J Infect Dis 1995, 171:829-836.
4. Easterbrook PJ, Emami J, Gazzard B: Rate of CD4 cell decline and prediction of survival in zidovudine-treated patients. AIDS 1993, 7:959-967.
5. Centers for Disease Control: Revision of the CDC surveillance case definition for acquired immunodeficiency syndrome. MMWR 1987, 36 (suppl 1S):1S-15S.
6. Kaplan EL, Meier P: Non parametric estimation from incomplete samples. J Am Stat Assoc 1958, 53:457-481.
7. Mantel N: Evaluation of survival data and two new rank order statistics arising in its consideration. Cancer Chemother Rep 1966, 50:163-170.
8. Cox DR: Regression models and life tables. J R Stat Soc [B] 1972, 34:187-200.
9. Collett D: Modelling Survival Data in Medical Research. London: Chapman & Hall; 1994.
10. Colford JM, Ngo L, Tager I: Factors associated with survival in HIV-infected patients with very low CD4 counts. Am J Epidemiol 1994, 139:206-218.
11. Keet IPM, Krol A, Koot M, et al.: Predictors of disease progression in HIV-infected homosexual men with CD4+ cells < 200 × 106/l but free of AIDS-defining clinical disease. AIDS 1994, 8:1577-1583.
12. Kotter DP, Tiernay M, Wang J, et al.: Magnitude of body cell mass depletion and the timing of death from wasting in AIDS. Am J Clin Nutr 1989, 50:444-447.
13. Burcham J, Marmor M, Dubin N, et al.: CD4% is the best predictor of development of AIDS in a cohort of HIV-infected homosexual men. AIDS 1991, 5:365-372.
14. Hoover DR, Rinaldo C, He Y, Phair J, Fahey J, Graham NMH: Long-term survival without clinical AIDS after CD4+ cell counts fall below 200 × 106/l. AIDS 1995, 9:145-152.
15. Saah AJ, Hoover DR, He Y, et al.: Factors influencing survival after AIDS: report from the Multicenter AIDS Cohort Study. J Acquir Immune Defic Syndr 1994, 7:287-295.
16. Graf W, Glimelius B, Pahlman L, et al.: Determinants of prognosis in advanced colorectal cancer. Eur J Cancer 1991, 27:1119-23.
17. Thrasher JB, Frazier HA, Robertson JE, et al.: Clinical variables which serve as predictors of cancer-specific survival among patients treated with radical cystectomy for transitional cell carcinoma of the bladder and prostate. Cancer 1994, 73:1708-1715.
18. Davis BR, Zauli G: Effect of human immunodeficiency virus infection in haematopoeisis. Bailliere Clin Haematol 1995, 8:113-130.
19. Henry DH, Beall GN, Benson CA, et al.: Recombinant human erythropoietin in the treatment of anemia associated with human immunodeficiency virus (HIV) infection and zidovudine therapy. Overview of four clinical trials. Ann Intern Med 1992, 117:739-748.
20. Hoover DR, Graham NMH, Chen B, et al.: Effect of CD4+ cell count measurement variability on staging HIV-1 infection. J Acquir Immune Defic Syndr 1992, 5:794-802.
21. Wyatt JC, Altman DG: Prognostic models: clinically useful or quickly forgotten? BMJ 1995, 311:1539-1540.
© Lippincott-Raven Publishers.