It is well documented that the introduction of highly active antiretroviral therapy (HAART) has slowed the progression of clinical HIV disease and improved overall
mortality rates. Non-AIDS-related deaths such as viral hepatitis B or C, non-HIV-associated cancers, and complications of antiretroviral therapy may now represent a much larger proportion of 1-9 mortality among HIV-infected individuals, however. Notably, all-cause 10-13 mortality in HIV-infected individuals remains higher than would be expected in the general population. Certain groups of HIV-infected individuals may be at particularly high risk, such as injection drug users (IDUs). 14,15 14,16-18
Precise quantification of the risk of non-AIDS-related death that accounts for the competing risk of AIDS-related death needs to be further quantified. If individuals who experience AIDS-related death are treated in analyses as censored individuals with standard Kaplan-Meier methods, it is well known that the non-AIDS-related (eg, cause-specific)
mortality rate is overestimated. To quantify these rates more precisely and provide a more relevant picture of non-AIDS-related 19,20 mortality among HIV-infected individuals, we used data from the Johns Hopkins HIV Clinical Cohort (JHHCC) with competing-risk methods that properly account for competing events that preclude an individual from having the event of interest. METHODS
Study Design and Population
The Johns Hopkins AIDS Service provides primary and subspecialty care for a large proportion of HIV-infected individuals in the Baltimore metropolitan area. In 1990, the JHHCC was established to understand and quantify the processes and outcomes of care for HIV-infected patients seen in clinical practice. The details of the cohort design have been described previously.
The population for this study consisted of HIV-infected patients receiving care at the Johns Hopkins AIDS Service who enrolled in the cohort before January 1, 2004. This research was done in accordance with the ethical standards of the Johns Hopkins Institutional Review Board and with the Helsinki Declaration of 1975. 21
Information on death was obtained from a death registry maintained by the clinic that receives reports from families, funeral homes, other medical institutions, and local coroners. In addition, death records of the Maryland Bureau of Vital Records and the Social Security death index were regularly searched. The cause of death was determined through medical records and from reports to the clinic. A death was coded as AIDS related if the individual died of an opportunistic infection or malignancy as defined by Centers for Disease Control and Prevention
or had no listed cause of death and had a CD4 22 + count ≤150 cells/mm 3 obtained within 6 months of death. Deaths were coded as non-AIDS related if a specific alternative cause of death was identified or if the individual had a CD4 + count >150 cells/mm 3 in the past 6 months. This cutoff was chosen by examining a subset of individuals with a coded cause of death and a CD4 + count within 6 months of death. A cutoff of 150 cells/mm 3 minimized the amount of misclassification. Most individuals reported to have died but who had indeterminate causes of death were no longer in care in our HIV practice. The rate of individuals lost to follow-up (defined as >18 months since last contact) in the JHHCC is 11.9%. Their information was censored at the time they last received care and was not included as deaths in our analyses. We found no reason to believe that the outcome for those who left care and died would have been different if they continued under care through the Johns Hopkins AIDS Service, because such individuals did not differ on demographic information (gender, race, and transmission risk factors) or on last CD4 + counts compared with individuals for whom a cause of death could be ascertained. Statistical Methods
Our analyses used data collected through December 31, 2003 and consisted of 2 components. First, we evaluated the incidence of death by calendar year in our cohort. To calculate the annual incidence rates, person-time contributed by each individual to each calendar year was determined as the denominator. The number of events (all-cause
mortality, AIDS-related mortality, and non-AIDS-related mortality) that occurred each year was calculated as the numerator for the determination of the incidence. Incidence rates were reported per 1000 person-years, and 95% confidence intervals (95% CIs) were calculated under Poisson distribution assumptions. Calendar year incidences were stratified by HIV transmission risk group (men who have sex with men [MSM], heterosexual contact, IDU, and other risk factors, which included those with multiple risk factors that were not IDU, transfusion, hemophilia, other accidental exposure, and no identifiable risk). The P value for trend was determined using Poisson regression analysis with calendar year as the predictor.
Second, we examined the cause-specific
mortality rates (eg, AIDS-related and non-AIDS-related) in 3 eras and by 4 CD4 + count categories. The eras were defined as before January 1, 1996, 1996 through 1998, and 1999 through 2003, corresponding to time before the widespread use of HAART, early HAART use in clinical practice, and the most recent HAART era. The CD4 + count at enrollment was categorized into individuals with ≤200, 201 to 350, 351 to 500, and >500 cells/mm 3. For each of these eras, we used nonparametric competing-risk methods (see the article by Gaynor et al for a review) to estimate the cause-specific 20 mortality rate. Often, the probability of an event occurring is estimated as 1 minus the Kaplan-Meier estimator, in which individuals who fail because of causes other than the event of interest are treated as censored observations. This censoring is informative, because this changes the probability of having the event of interest (eg, death from AIDS precludes death from other causes). This is a competing-risk framework, because to be at risk for the event of interest, one must survive all other causes, which is not accounted for in the 1 minus the Kaplan-Meier estimator. Therefore, the standard Kaplan-Meier estimator overestimates the probability of being at risk of failure at a particular time. Competing-risk methods properly account for the 2 competing broad classifications of 19,20 mortality (AIDS-related and non-AIDS-related mortality). The cause-specific risk of mortality at time t is the product of the overall survival function multiplied by the estimated cause-specific hazard at time t. The sum of the risk between time 0 and time t is the cumulative mortality risk. Because individuals could contribute to more than 1 era, we used staggered entry methods to prevent survival bias. Contributed person-time of an individual was divided according to when he or she crossed into different time periods. Time started from enrollment into the clinic, and for each era, time ended when the individual died, was lost from clinical follow-up (which would preclude our assessment of cause of death), or was censored at the end of the era. Individuals who survived to the next era were carried forward. These individuals contributed time at risk starting from the amount of time that had previously elapsed in the prior eras. Thus, an individual who enrolled on January 1, 1993 and survived until January 1, 2002 would contribute 3 years to each of the eras with starting times of 0, 3, and 6 years in the first, second, and third eras, respectively. Because the cause-specific 20 mortality curves were not independent of each other and current methods for competing-risk analyses do not compare the cumulative incidences from 2 competing events, point-wise 95% CIs for the differences between curves were determined through bootstrap methods with 1000 replicates. All analyses were conducted using SAS (version 9.1; SAS Institute, Cary, NC) or S-PLUS (version 6.2; Insightful Corporation, Seattle, WA). 23 RESULTS
The study population was composed of 5460 patients enrolled in the cohort before January 1, 2004. The cohort consisted primarily of men (69%) and was predominantly African American (76%), with a median age of 38 years at enrollment. Individuals were self-classified as MSM (n = 1193), heterosexual (n = 1250), IDU (n = 2498), and other transmission risk factors (n = 519). The median CD4
+ count at enrollment was 259 (interquartile range [IQR]: 72-462) cells/mm 3 (median [IQR] by transmission group: MSM = 222 [60-428] cells/mm 3; heterosexual = 279 [65-477] cells/mm 3; IDU = 279 [96-480] cells/mm 3; and other = 207 [47-433] cells/mm 3). Overall, there were 21,417 person-years of follow-up, with 1337 deaths before January 1, 2004. Of these deaths, 971 could be attributed to AIDS-related causes and 366 to non-AIDS-related causes ( Table 1). TABLE 1:
Causes of Death Sorted by AIDS-Related and Non-AIDS-Related Categories
There was a significant decrease in all-cause
mortality beginning in 1996 and continuing through 1997 (a high of 112 deaths per 1000 person-years in 1995 and 39 deaths per 1000 person-years in 1997). This decrease was concurrent with an increase in HAART use ( Fig. 1A). Since 1997, the all-cause mortality rate has remained stable at approximately 40 deaths per 1000 person-years ( P = 0.58 for trend). In contrast, AIDS-related mortality continued to decline after 1996, with a zenith of 40.9 deaths (95% CI: 32.3 to 51.2) in 1998 to a low of 20.2 deaths (95% CI: 14.9 to 26.9) in 2003 ( P = 0.008). The drop in AIDS-related mortality after 1996 has been offset by an increase in non-AIDS-related mortality ranging from 10.7 deaths (95% CI: 6.2 to 17.1) in 1997 to a high of 22.7 deaths (95% CI: 17.2 to 29.5) in 2003 ( P < 0.001). FIGURE 1:
A, Calendar year
mortality rate for all-cause mortality (solid line), AIDS-related mortality (dash-dot-dash line), and non-AIDS-related mortality (long dash line) plotted with the proportion receiving therapy (monotherapy, no shading; combination therapy, area indicated by vertical line shading; and HAART, area indicated by diagonal line shading). Calendar year indices for all-cause mortality (B), AIDS-related mortality (C), and non-AIDS-related mortality (D) by transmission risk group (MSM, solid line; heterosexual, long dashed line; IDU, dash-dot-dash line; and other, short dashed line). Mortality rates are plotted at the midpoint of each calendar year to reflect deaths occurring throughout the year.
Mortality rates for each era were examined by HIV risk group (see Fig. 1B-D). Although there was no apparent trend in each of the risk groups for all-cause mortality (see Fig. 1B; P > 0.09 for trend), MSM had a decline in AIDS-related mortality ( P = 0.016 for trend; see Fig. 1C). When combined together to represent a non-MSM group, there was a borderline significant trend for decreasing AIDS-related mortality ( P = 0.06), suggesting that the decline in AIDS-related mortality in the overall cohort since 1997 was not driven solely by the MSM group. Only among IDUs was there a significant increasing trend for non-AIDS-related mortality ( P < 0.001; see Fig. 1D). The combined non-IDU risk group had a significant increasing non-AIDS-related trend in mortality ( P = 0.03), suggesting that the increasing trend in non-AIDS-related mortality in the overall cohort was not attributable solely to the IDU group. The overall non-IDU cohort composition changed during this period as the proportion with a heterosexual transmission risk began to exceed the proportion with an MSM transmission risk beginning in 1998. Thus, the significant trend among the non-IDU population may reflect this shift.
Competing-risk analyses were conducted stratified by era and category of CD4
+ counts (≤200, 201-350, 351-500, and >500 cells/mm 3). The curves quantify the risk of dying ( Figs. 2-5) attributable to an AIDS-related cause or a non-AIDS-related cause among those individuals who remained alive up to that point in time. Among those with CD4 + counts ≤200 cells/mm 3, the cumulative incidence for AIDS-related mortality exceeded that for non-AIDS-related mortality for all eras (see Fig. 2A-C). AIDS-related mortality risk was highest before 1996 (see Fig. 2A) and has declined subsequently, indicating the effectiveness of therapy among those with the highest risk of dying attributable to an AIDS-related cause. The proportions dying within 3 years of clinic enrollment before 1996 were 57.5% and 5.06% for AIDS-related and non-AIDS-related mortality, respectively, with a 95% CI for the difference between the cumulative mortality rates of 48% and 58% at 3 years. The risks of dying within 3 years of clinic enrollment attributable to AIDS-related causes were 21.4% and 15.7% for the early and most recent therapy eras, respectively. The non-AIDS-related mortality risk at 3 years was 3.0% in 1996 through 1999 and 4.2% after 1999. FIGURE 2:
Cumulative risk for AIDS-related
mortality (solid line) and non-AIDS-related mortality (dashed line) calculated using nonparametric competing-risk methods for those with CD4 + counts <200 cells/mm 3 in the period 1990 through 1995 (A), 1996 through 1998 (B), and 1999 through 2003 (C). The upper portion of each panel corresponds to the difference between mortality curves (solid line) and 95% CIs (dashed line). FIGURE 3:
Cumulative risk for AIDS-related
mortality (solid line) and non-AIDS-related mortality (dashed line) calculated using nonparametric competing-risk methods for those with CD4 + counts between 201 and 350 cells/mm 3 in the period 1990 through 1995 (A), 1996 through 1998 (B), and 1999 through 2003 (C). The upper portion of each panel corresponds to the difference between mortality curves (solid line) and 95% CIs (dashed line). FIGURE 4:
Cumulative risk for AIDS-related
mortality (solid line) and non-AIDS-related mortality (dashed line) calculated using nonparametric competing-risk methods for those with CD4 + counts between 351 and 500 cells/mm 3 in the period 1990 through 1995 (A), 1996 through 1998 (B), and 1999 through 2003 (C). The upper portion of each panel corresponds to the difference between mortality curves (solid line) and 95% CIs (dashed line). FIGURE 5:
Cumulative risk for AIDS-related
mortality (solid line) and non-AIDS-related mortality (dashed line) calculated using nonparametric competing-risk methods for those with CD4 + counts >500 cells/mm 3 in the period 1990 through 1995 (A), 1996 through 1998 (B), and 1999 through 2003 (C). The upper portion of each panel corresponds to the difference between mortality curves (solid line) and 95% CIs (dashed line).
mortality and non-AIDS-related mortality are shown for those with CD4 + counts of 201 to 350, 351 to 500, and >500 cells/mm 3 in Figures 3, 4, and 5, respectively, by therapy era. Before 1996 and from 1996 through 1999 and in each CD4 + category >200 cells/mm 3, AIDS-related mortality was initially lower but later exceeded non-AIDS-related mortality (see Figs. 3A, 4A, 5A and Figs. 3B, 4B, 5B, respectively), although the difference between AIDS-related and non-AIDS-related mortality in the era from 1996 through 1999 was not significant. In contrast, after 1999, AIDS-related mortality is uniformly lower than that of non-AIDS-related mortality in the CD4 + categories >200 cells/mm 3 (see Figs. 3C, 4C, 5C). The higher risk of non-AIDS-related mortality is supported by the estimated 95% CI for the difference in mortality risk (see upper panels of Figs. 3C, 4C, 5C) because a large proportion of time does not include 0.
The 5-year cumulative AIDS-related and non-AIDS-related
mortality after 1999 stratified by HIV risk group and estimated using competing-risk methods is presented in Table 2. AIDS-related cumulative mortality was significantly higher than non-AIDS-related cumulative mortality for those with CD4 + counts ≤200 cells/mm 3 for all risk groups. Individuals who enrolled with CD4 + counts >200 cells/mm 3 rarely died as the result of an AIDS-related cause because the 5-year cumulative mortality rate ranged between 0.0% and 6.7%, whereas the non-AIDS-related 5-year cumulative mortality rate ranged between 0.0% and 12.1%. Furthermore, significantly higher non-AIDS-related 5-year cumulative mortality as compared to AIDS-related cumulative mortality was observed only among IDUs and those with CD4 + counts >200 cells/mm 3 (borderline significant among those with CD4 + counts between 351 and 500 cells/mm 3; P = 0.074; otherwise, P ≤ 0.016). Ten of the 12 categories (defined by risk group and CD4 category >200 cells/mm 3) had higher estimated non-AIDS-related 5-year cumulative mortality than AIDS-related cumulative mortality, however. TABLE 2:
Percent of Individuals Dying Attributable to AIDS-Related
Mortality and Non-AIDS-Related Mortality Within 5 Years of Enrollment in the Clinic With 95% CIs for the Period From 1999 Through 2003 Stratified by Transmission Risk Group and CD4 + Count at Enrollment
There was an increase in the median age at enrollment with calendar time (range: 34 years in 1991 to 41 years in 2002). In separate analyses, we evaluated the risk of AIDS-related and non-AIDS-related death using age as the time scale in lieu of time from enrollment in the cohort, continuing to stratify by CD4
+ cell count and calendar time. These analyses did not alter our results significantly and demonstrated the same trends as seen previously (data not shown).
In contrast with the competing-risk methods, standard Kaplan-Meier estimates of the cumulative
mortality for AIDS-related and non-AIDS-related mortality were determined for the total cohort. The impact of different analytic strategies is substantial: standard Kaplan-Meier methods overestimated AIDS-related mortality by up to 5%, 6%, and 13% for the pre-1996, 1996 through 1999, and post-1999 eras, respectively. For non-AIDS-related mortality, the standard Kaplan-Meier methods overestimated the risk by 21%, 21%, and 19% for the pre-1996, 1996 through 1999, and post-1999 eras, respectively. DISCUSSION
mortality has decreased in the JHHCC since the introduction of HAART in 1996. This decline is consistent with trends that we and others have reported. Despite the continued decline in AIDS-related 1-7 mortality, overall mortality has not changed since 1997 because the decline in AIDS-related mortality has been offset by an increase in non-AIDS-related mortality. Since 1999, individuals who enrolled in the cohort with CD4 + counts >200 cells/mm 3 were much more likely to die from non-AIDS-related causes. Specifically, the non-AIDS-related mortality risk was significantly higher than that for AIDS-related mortality among those enrolling with a CD4 + count >200 cells/mm 3 for the IDU risk group. Seven of 9 of the other categories (by risk group and CD4 + count category >200 cells/mm 3) had higher estimated non-AIDS-related mortality than AIDS-related mortality, suggesting that non-AIDS-related mortality is a major reason for death even among the non-IDU groups. An analysis using updated CD4 + counts for individuals surviving into the next HAART era was also conducted. This analysis also supported the conclusion that the risk of non-AIDS-related mortality either exceeds or equals that of AIDS-related mortality for those with CD4 + counts >200 cells/mm 3. This trend started in the era from 1996 through 1999, however.
It has previously been reported that the causes of death have changed since the introduction of HAART.
A shift in the proportion of deaths attributable to non-HIV-related illness has been previously reported, with a portion occurring among patients with good control of HIV replication and increasing CD4 10,11,13 + counts. Furthermore, an equivalent incidence of non-HIV-related and HIV-related 13 mortality was reported among individuals who had initiated HAART. Among the Women's Interagency HIV Study (WIHS) cohort, however, only 20% of deaths were attributable to non-AIDS-related causes. 24 The competing-risk approach used in this study properly accounted for the fact that one type of death precludes the other from occurring and has shown that standard survival methods overestimated the risk of 25 mortality by up to 21% in our cohort. Furthermore, it has been previously been suggested that the number of non-HIV-related deaths was higher among IDUs compared with MSM, but the authors did not report whether these deaths exceeded the number HIV-related deaths. Recently the Concerted Action on SeroConversion to AIDS and Death in Europe (CASCADE) Collaboration has concluded in a competing-risk framework that opportunistic infections remain a large cause of death in the current HAART era. 18 This does not contradict our findings, because AIDS-related death remains significant among those with low CD4 26 + cell counts. The CASCADE Collaboration presents comparisons of the cumulative risk of death across HAART eras and across risk transmission groups. Thus, whether the risk of AIDS-related mortality exceeds that of non-AIDS-related mortality within the HAART era, transmission risk group, and CD4 + count category cannot be determined and directly compared. 26
These results further emphasize the remarkable effectiveness of HAART in reducing
mortality from AIDS. In addition, they highlight the growing importance of competing non-AIDS-related causes of death, because HIV infection has been transformed from a terminal illness into a chronic illness. Improvement in overall mortality in HIV-infected patients is likely to be hampered unless an increasing focus is devoted to managing other comorbidities that lead to death, especially in those patients whose transmission risk group is IDU. Injection drug use is a relapsing illness, and continued injection of drugs is common. Hepatitis C virus is especially common in IDU patients and can be associated with a relatively high 27 mortality rate. Other comorbidities, including endocarditis, bacteremia, drug overdose, and renal failure, occur more frequently in IDUs. Yet, the non-AIDS-related mortality rate is also increasing in the combined non-IDU risk groups, emphasizing the importance of identifying and treating competing comorbidities that can lead to death. These results also have important epidemiologic implications for HIV research. With the potential variety in the causes of death, standard Kaplan-Meier methods are not adequate to assess the risk of different outcomes and competing-risk methods are more appropriate. The large difference we found between standard Kaplan-Meier and competing-risk estimates in our data emphasizes the inaccuracies using standard Kaplan-Meier methods. One potential easily applied method for competing risk analyses is the application of a Cox proportional hazards model to augmented data. These methods even allow for direct comparison of different failure types if there is a constant hazard ratio between failure types, which did not occur in our data. 28
The Johns Hopkins AIDS Service treats a large proportion of HIV-infected patients in the Baltimore metropolitan area,
and this population may be most representative of the HIV epidemic in a large US city. Although the all-cause 21 mortality trends observed in the cohort may not be as representative of other populations, the stratified analyses show the trends for each of the HIV transmission risk groups, and these subgroup analyses may increase the generalizability of our results to other populations. Even though coding of deaths was based on a review of available clinical data, the cause of death could not always be known specifically (eg, when an AIDS-defining illness and an end-stage comorbidity occurred concomitantly) and the potential for misclassification exists. Individuals whose death was classified by their last CD4 + count are the most likely to be misclassified. Examining the last CD4 + counts within 6 months for individuals whose death was classified by the threshold cutoff of 150 cells/mm 3 resulted in different probabilities of misclassification for those with CD4 + counts greater than and less than this threshold. Examining individuals with known causes of death, the categorization of death by CD4 + count <150 cells/mm 3 within 6 months of death as AIDS related resulted in a positive predictive value and negative predictive value of 79.7 and 64.6, respectively. Given the number of individuals classified by CD4 + cell counts, the expected number of individuals misclassified would be 77 currently classified as AIDS related and 60 currently classified as non-AIDS related. Thus, the risk of non-AIDS-related causes of death was likely to have been underestimated.
Some deaths could not be classified because some individuals have left medical care. These deaths were censored at the time of last care, however, when individuals were still at risk for having a cause of death classified. Nevertheless, on examining some of the demographic and prognostic data, we found no reason to believe that the outcome in those who left care would have been different if they continued under care through the Johns Hopkins AIDS Service.
In summary, although there has been a reduction in
mortality since the introduction of HAART, the continued decline in AIDS-related mortality has been offset by an increase in non-AIDS-related mortality. In the current era of HAART, the cumulative risk of dying as the result of a non-AIDS-related cause exceeds that of an AIDS-related cause for individuals with a CD4 + count >200 cells/mm 3 among IDU populations and is equivalent in non-IDU transmission risk groups. It is therefore of increasing clinical and public health importance that attention be directed toward conditions that may not traditionally be considered HIV related. REFERENCES
1. Egger M, Hirschel B, Francioli P, et al. Impact of new antiretroviral combination therapies in HIV infected patients in Switzerland: prospective multicentre study.
2. Palella FJ Jr, Delaney KM, Moorman AC, et al. Declining morbidity and
among patients with advanced human immunodeficiency virus infection.
N Engl J Med
3. Detels R, Munoz A, McFarlane G, et al. Effectiveness of potent antiretroviral therapy on time to AIDS and death in men with known HIV infection duration.
4. McNaghten AD, Hanson DL, Jones JL, et al. Effects of antiretroviral therapy and opportunistic illness primary chemoprophylaxis on survival after AIDS diagnosis.
5. Mocroft A, Madge S, Johnson AM, et al. A comparison of exposure groups in the EuroSIDA study: starting highly active antiretroviral therapy (HAART), response to HAART, and survival.
J Acquir Immune Defic Syndr
6. Mocroft A, Ledergerber B, Katlama C, et al. Decline in the AIDS and death rates in the EuroSIDA study: an observational study.
7. Moore RD, Chaisson RE. Natural history of HIV infection in the era of combination antiretroviral therapy.
8. Cohen C, Revicki DA, Nabulsi A, et al. A randomized trial of the effect of ritonavir in maintaining quality of life in advanced HIV disease.
9. Revicki DA, Moyle G, Stellbrink HJ, et al. Quality of life outcomes of combination zalcitabine-zidovudine, saquinavir-zidovudine, and saquinavir-zalcitabine-zidovudine therapy for HIV-infected adults with CD4 cell counts between 50 and 350 per cubic millimeter.
10. Mocroft A, Brettle R, Kirk O, et al. Changes in the cause of death among HIV positive subjects across Europe: results from the EuroSIDA study.
11. Selik RM, Byers RH Jr, Dworkin MS. Trends in diseases reported on US death certificates that mentioned HIV infection, 1987-1999.
J Acquir Immune Defic Syndr
12. Lewden C, Salmon D, Morlat P, et al. Causes of death among human immunodeficiency virus (HIV)-infected adults in the era of potent antiretroviral therapy: emerging role of hepatitis and cancers, persistent role of AIDS.
Int J Epidemiol
13. Valdez H, Chowdhry TK, Asaad R, et al. Changing spectrum of
due to human immunodeficiency virus: analysis of 260 deaths during 1995-1999.
Clin Infect Dis
14. Keiser O, Taffé P, Zwahlen M, et al. All cause
in the Swiss HIV cohort study from 1990 to 2001 in comparison with the Swiss population.
15. Wang C, Vlahov D, Galai N, et al.
in HIV-seropositive versus-seronegative persons in the era of highly active antiretroviral therapy: implications for when to initiate therapy.
J Infect Dis
16. Egger M, May M, Chêne G, et al. Prognosis of HIV-1-infected patients starting highly active antiretroviral therapy: a collaborative analysis of prospective studies.
17. Chêne G, Sterne JA, May M, et al. Prognostic importance of initial response in HIV-1 infected patients starting potent antiretroviral therapy: analysis of prospective studies.
18. Mocroft A, Gatell J, Reiss P, et al. Causes of death in HIV infection: the key determinant to define the clinical response to anti-HIV therapy.
19. Cox DR. The analysis of exponentially distributed life-times with two types of failure.
J R Stat Soc [Ser B]
20. Gaynor JJ, Feuer EJ, Tan CC, et al. On the use of cause-specific failure and conditional failure probabilities: examples from clinical oncology data.
J Am Stat Assoc
21. Moore RD. Understanding the clinical and economic outcomes of HIV therapy: the Johns Hopkins HIV Clinical Practice Cohort.
J Acquir Immune Defic Syndr Hum Retrovirol
. 1998;17(Suppl 1):S38-S41.
22. Centers for Disease Control. 1993 Revised classification system for HIV infection and expanded surveillance case definition for AIDS among adolescents and adults.
MMWR Morb Mortal Wkly Rep
23. Efron B.
The Jackknife, the Bootstrap, and Other Resampling Plans
. Philadelphia: Society for Industrial and Applied Mathematics; 1982.
24. Van Sighem AI, van de Wiel MA, Ghani AC, et al.
and progression to AIDS after starting highly active antiretroviral therapy.
25. Cohen MH, French AL, Benning L, et al. Causes of death among women with human immunodeficiency virus infection in the era of combination antiretroviral therapy.
Am J Med
26. Smit C, Geskus R, Walker S, et al. Effective therapy has altered the spectrum of cause-specific
following HIV seroconversion.
27. Galai N, Safaeian M, Vlahov D, et al. Longitudinal patterns of drug injection behavior in the ALIVE Study cohort, 1988-2000: description and determinants.
Am J Epidemiol
28. Lunn M, McNeil D. Applying Cox regression to competing risks.