Lima, Viviane D^a; Hogg, Robert S^a,b; Harrigan, P Richard^a,c; Moore, David^a; Yip, Benita^a; Wood, Evan^a; Montaner, Julio SG^a,c

Introduction

Over 40 million people worldwide are infected with HIV, with the majority living in sub-Saharan Africa ^[1]. Cumulative worldwide mortality associated with HIV and AIDS has been estimated at 3.1 million (range 2.8–3.6) in 2005 ^[1], with Canada having accumulated death estimates attributable to HIV/AIDS ranging between 13 293 and 15 148 ^[2].

The introduction of combination antiretroviral therapies has led to improvements in HIV-related mortality in North America and western Europe ^[3–13]. HIV-related deaths, however, continue to occur across the developed world, and the emergence of drug-resistant HIV variants and drug toxicities are still major barriers to successful long-term antiretroviral efficacy ^[14–18].

In the absence of large-scale randomized trials directly comparing different HAART regimens, observational studies can provide important insights into potential regimen differences in the rates of disease progression or death over time. A number of studies have already been undertaken ^[6,19–25], but none so far has been large enough to examine longitudinal differences in mortality rates by regimen type. The objective of this study was to examine whether the mortality rates differed longitudinally according to the initial HAART regimen, which was broken down into three different classes of therapy: mono, dual or therapies containing three or more antiretroviral drugs.

Methods

The HIV/AIDS drug treatment program

The distribution of antiretroviral therapy in the province of British Columbia (BC) has been described in detail elsewhere ^[8,26]. Briefly, antiretroviral drugs have been centrally distributed at no cost to eligible HIV-infected individuals since 1986. In October 1992, the HIV/AIDS Drug Treatment Program became the responsibility of the BC Centre for Excellence in HIV/AIDS (the Centre). Since then, the Centre has been prospectively collecting information on antiretroviral use, sociodemographic characteristics, and the clinical and health status of all participants enrolled in the program.

Data collection

Our study was restricted to antiretroviral-naive HIV-positive men and women aged 18 years or older who started antiretroviral therapy between January 1993 and September 2004, which is the cutoff date of our last data linkage with the BC Vital Statistics Agency. Physicians enrolling an HIV-positive individual into the Centre's HIV/AIDS Drug Treatment Program must complete a drug request enrollment form, which acts as a legal prescription and compiles information on the HIV-positive applicant's address and enrolling physician, past HIV-specific drug history, CD4 cell counts, and current drug requests. Each request is reviewed by a qualified practitioner to ensure that it meets the Centre's established therapeutic guidelines ^[27]. The guidelines used at the Centre have been updated and are consistent with those recommended by the International AIDS Society – USA ^[28,29]. Approved prescriptions are renewed every 2 months. At the time of the initial refill each participant is asked to complete an enrollment survey and programme consent form, whereas the physician is asked to complete a clinical staging form. Participant surveys and clinical staging forms are completed annually. The clinical staging form records participant-specific information on HIV/AIDS-related conditions according to the World Health Organization clinical staging system ^[30].

Outcome measure and independent variables

The primary endpoint in this analysis was all-cause mortality. Deaths during the follow-up period were identified on a continuous basis from physician reports and through annual record linkages carried out with the BC Division of Vital Statistics.

The following independent variables were investigated: age (in years), sex (male versus female), CD4 cell count (per 100 decrement), baseline AIDS diagnosis (yes versus no), history of injection drug use (yes versus no), physician experience (per 100 patients followed), adherence (< 95% versus ≥ 95%), year of initial therapy (1993–1995, 1996–1998, 1999–2001, 2002–2004), and initial drug regimen. Physician experience was defined as the number of HIV-positive patients the physician had previously treated at the time the study subject was enrolled into the HIV/AIDS Drug Treatment Program. Estimates of adherence to antiretroviral therapy were based on medications dispensed, not prescribed. Our measure of adherence was limited to the first year of therapy and estimated by dividing the number of months of medications dispensed by the number of months of follow-up. This measure of adherence has been found to be independently associated with HIV viral suppression and survival among HIV-infected individuals enrolled in the HIV/AIDS Drug Treatment Program ^[16,31]. Patients were defined a priori as non-adherent if they received antiretroviral medications for less than 95% of the follow-up period during the first year of therapy, as in previously published work ^[32,33]. The regimen type was defined as: (i) mono: initial monotherapy; (ii) dual: initial dual nucleoside combination; (iii) three or more antiretroviral drugs, which included any of the following therapies: dual nucleoside plus a single protease inhibitor, dual nucleoside plus a non-nucleoside reverse transcriptase inhibitor, dual nucleoside plus a protease inhibitor and less than 800 mg ritonavir, or three or more antiretroviral drugs containing a nucleotide reverse transcriptase inhibitor. In order to be consistent with clinical practice across all study periods, the study was restricted to those programme participants who initiated antiretroviral therapy with baseline CD4 cell counts of 200 cells/μl or less.

Life tables construction

Abridged life tables were constructed to compare age-specific mortality rates and life expectancies at the age of 20 years in four different time periods, 1993–1995, 1996–1998, 1999–2001, 2002–2004 ^[34–36]. Large populations are needed to overcome systematic and random variations in mortality when building complete life tables, therefore abridged life tables were used in this study. These tables describe the mortality experience that hypothetical cohorts of HIV-positive individuals would have had if they were subjected to the mortality rates in the observed four time periods. The life expectancy at an exact age measures the average number of additional years that will be lived by a person after that age, according to mortality rates for all causes combined during the study period. Potential years of life lost (PYLL) were calculated at 75 years and expressed as an absolute number and per 1000 person-years.

Statistical analysis

In the primary analysis, time zero was the initial start date of antiretroviral therapy whereas the event date was the date of death. Event-free subjects were right censored as of 30 September 2004. Those subjects with the latest contact before 30 September 2004 were censored at the date of last known contact.

For the purposes of analysis, we followed the intent-to-treat principle, with subjects retained in their initial treatment groups irrespective of whether participants subsequently switched to regimens available later. This approach provides the most conservative estimate of the true treatment effect. Cumulative mortality rates were estimated using Kaplan–Meier (KM) methods ^[37]. Survival functions were compared using the log-rank test ^[37].

The distribution of patients according to several independent variables was examined in each of the four different time periods (1993–1995, 1996–1998, 1999–2001, 2002–2004) to determine whether the characteristics of the study population has changed over time. Categorical variables were compared using the Cochran–Armitage trend test ^[38]. Continuous variables were compared using linear models analysis ^[39]. Cox proportional hazard models were used to estimate the hazard of death with associated 95% confidence intervals ^[40]. A forward stepwise technique was used in the selection of covariates. All reported P values are two-sided. All analyses were performed using SAS 9.1.3 service pack 3 (SAS Institute Inc., Cary, North Carolina, USA).

Results

A total of 2474 participants aged 18 years or older with baseline CD4 cell counts of 200 cells/μl or less started naive on one of the six types of combination antiretroviral regimen between 1 January 1993 and 30 September 2004. Among them, 236 (10%) were excluded from the analyses because there were no data on baseline CD4 cell counts taken within 6 months before the antiretroviral start date. Excluded individuals were more likely to be female (P < 0.001), less likely to have an AIDS diagnosis at baseline (P = 0.017), and were younger (P < 0.001) than those who were included in the analysis.

The remaining 2238 individuals were mainly men (N = 1889; 84.4%), and 649 participants (29.0%) had a previous diagnosis of AIDS at baseline. The median age was 38.5 years [interquartile range (IQR) 33.3–45.0], and the median CD4 cell count was 90 cells/μl (IQR 30–150). Of these, 518 (23.2%) started therapy in 1993–1995, 709 (31.7%) in 1996–1998, 489 (21.9%) in 1999–2001, and 522 (23.3%) in 2002–2004. The overall median follow-up time was 35.5 months (IQR 14.1–75.3). In total, 327 (14.6%) started on monotherapy, 547 (24.4%) on dual nucleoside combination, 467 (20.9%) on dual nucleoside plus a non-nucleoside reverse transcriptase inhibitor, 485 (21.7%) on dual nucleoside plus a single protease inhibitor, 355 (15.9%) on dual nucleoside plus a protease inhibitor and less than 800 mg ritonavir regimens, and 57 (2.6%) started on therapies consisting of a nucleotide reverse transcriptase inhibitor. The number of antiretroviral drugs available for the initial therapy increased substantially from 1993–1995 to 2002–2004. Patients in 1993–1995 were assigned to eight different drug regimens, and 90.0% of these regimens contained zidovudine. In contrast, in 2002–2004, 53 drug combination regimens were assigned to different patients, and 95.0% of these regimens contained lamivudine.

Table 1 compares the baseline demographic and clinical characteristics of individuals across the different time periods (1993–1995, 1996–1998, 1999–2001, and 2002–2004). As the study group was restricted to those with baseline CD4 cell counts of 200 cells/μl or less, the median baseline CD4 cell count was comparable and not statistically different across all time periods (P = 0.2182). A history of injection drug use was also not associated with the period of antiretroviral therapy initiation (P = 0.4113). Differences were observed at baseline between all time periods with respect to all other variables. As noted here, towards the 2002–2004 period, subjects were more likely to be older (P < 0.0001), were more likely to be women (P < 0.0001), were less likely to have a diagnosis of AIDS at baseline (P < 0.0001), were more likely to adhere to therapy, and were more likely to be treated by more experienced physicians (P < 0.0001). There was also a significant increase in the median number of drugs in the initial combination (P < 0.0001). The number starting on monotherapy regimens was the highest in 1993–1995 (N = 282; 54.4%). The number of patients who started on dual therapy was high in both 1993–1995 and 1996–1998 (N = 236; 45.6% and N = 304; 42.9%, respectively). The number of participants starting on three or more antiretroviral drugs has been increasing steadly since 1996–1998, achieving its highest in the period 2002–2004 (N = 516; 98.9%).

Table 1

As of 30 September 2004, a total of 683 deaths were identified, yielding an overall all-cause mortality rate of 30.5%. The product limit estimates of the cumulative mortality rate at 12 months were 15.8% (± 1.6%) in 1993–1995, 9.4% (± 1.1%) in 1996–1998, 9.1% (± 1.3%) in 1999–2001, and 6.1% (± 1.1%) in 2002–2004. The univariate analysis showed that subjects in 2002–2004 were less likely to die in the first 12 months after the start of therapy than those who started therapy in 1993–1995 [hazard ratio (HR) 0.29, 95% CI 0.21–0.41].

Table 2 presents mortality rates, life expectancy and PYLL for the study participants, stratified by the time periods 1993–1995, 1996–1998, 1999–2001 and 2002–2004. There has been a decline in the mortality rate and PYLL and an increase in life expectancy since 1993–1995. The survival probability from the age of 20 up to 44 years has increased substantially from 1.2% in 1993–1995 to 36.4% in 2002–2004. The PYLL (at age 75 years) has decreased almost five times, from 6005 per 1000 population in 1993–1995 to 1234 per 1000 population in 2002–2004. The life expectancy at the age of 20 years has tripled since 1993–1995, ranging from 9.1 years (± 2.3 years) in 1993–1995 to 23.6 years (± 4.4 years) in 2002–2004.

Table 2

The gain in life expectancy since the period 1993–1995 occurred in all age groups, and was highest for the age group 35–40 years (from 5.4 ± 0.5 years to 21.5 ± 1.1 years), and lowest for the age group 50–55 years (from 6.4 ± 1.3 years to 16.9 ± 1.0 years). The biggest differences in life expectancy for the younger age groups were seen between those initiating therapy before and after 1998, whereas for the older age groups the biggest differences were seen among those initiating therapy before and after 2001. During 1993–1995 there was no age-dependent trend in life expectancy; however, since then we have seen a large variability in life expectancy across different age groups (data not shown).

Table 3 shows the results for the univariate and multivariate associations between the year of antiretroviral therapy initiation and mortality, after adjusting for age, sex, AIDS diagnosis, history of injection drug use, adherence, CD4 cell count, and physician experience. The multivariate model shows a decreasing gradient in the mortality risk according to the year of antiretroviral therapy initiation. Patients who initiated therapy in 2002–2004 were 2.78 (HR 0.36, 95% CI 0.26–0.52) times less likely to die than those who started therapy in 1993–1995. A similar gradient in mortality is observed in Figure 1a, in which we provided the crude KM estimates of survival by the year of antiretroviral therapy initiation. For the first 12 months of therapy, there was a relative increase of 11.5% between the probability of survival among those who initated therapy in 2002–2004 (KM 94.0 ± 1.1%) as opposed to 1993–1995 (KM 84.2 ± 1.6%). If we now look at the first 24 months of therapy, however, the relative increase in the survival probability was more substantial (41.2%), with the crude KM estimates ranging from 64.4% (± 2.2%) in 1993–1995 to 91.0% (± 1.5%) in 2002–2004.

Table 3

Fig. 1

Table 4 shows the results for the univariate and multivariate associations between initial drug regimen and mortality, after adjusting for age, sex, baseline AIDS diagnosis, history of injection drug use, adherence, CD4 cell count, and physician experience. Note that patients who initiated on dual therapy or therapies containing three or more antiretroviral drugs were, respectively, 1.49 (HR 0.67, 95% CI 0.55–0.81) and 2.56 (HR 0.39, 95% CI 0.32–0.47) times less likely to die than those who started on monotherapy. Figure 1b shows the crude KM estimates of the cumulative survival rate per initial drug regimen. The crude KM estimates at 12 months ranged from 81.5% (± 2.2%) for monotherapy to 92.6% (± 0.7%) for therapies containing three or more antiretroviral drugs, reflecting a percentage increase in the survival probability of 13.6%. The increase in the survival probability is more evident at 24 months, with survival probabilities ranging from 63.5% (± 2.7%) for monotherapy to 87.7% (± 1.0%) for therapies containing three or more antiretroviral drugs, reflecting a percentage increase in survival probability of 38.1%.

Table 4

In Table 3 and Table 4 we also observed that independent predictors such as adherence to therapy and physician experience played an important role in the associations of the risk of mortality with the year of antiretroviral therapy initiation and with the first regimen. In both analyses, individuals with poor adherence (< 95%) were 2.50 (HR 0.40, 95% CI 0.34–0.48; and HR 0.40, 95% CI 0.34–0.47, respectively) more likely to die than those with higher adherence levels (≥ 95%). We observed that individuals with less experienced physicians were also at a higher risk, because in these analyses the HR were, respectively, 1.14 (HR 0.88, 95% CI 0.78–0.98) and 1.18 (HR 0.85, 95% CI 0.76–0.95).

Discussion

In the case of a low initial CD4 cell count (≤ 200 cells/μl) at the start of antiretroviral therapy, our results demonstrate a statistically significant decrease in mortality and increase in life expectancy over the 12 years of study. Individuals who started therapy in the 2002–2004 time period were 2.78 times less likely to die than those in the 1993–1995 period. Life expectancy at the age of 20 years has increased by 14.5 years from 9.1 years in 1993–1995 to 23.6 years in 2002–2004. The decrease in mortality, and thus the increase in life expectancy, was directly associated with regimens containing a combination of three or more antiretroviral drugs. Individuals who initiated therapy based on three or more antiretroviral agents were 2.56 times less likely to die than those who initiated monotherapy regimens.

Other studies conducted in Canada ^[5], the United States ^[13,41–44], France ^[22], and Switzerland ^[45] have reported on similar results soon after the introduction of triple therapy. Pallela et al. ^[43], using data on 1255 patients enrolled in the HIV Outpatient Study from nine clinics across different states in the USA, showed a decline in mortality from 29.4 per 100 person-years in 1995 to 8.8 per 100 person-years in 1997. The authors from this study also observed that more intense antiretroviral therapies were linked to lower mortality rates, with the risk of death being 3.0 times (95% CI 2.1–4.1 times) higher for those receiving monotherapy in comparison with those receiving a protease inhibitor in a triple combination regimen. Crum et al. ^[41], using data on 4241 HIV-infected beneficiaries from the Department of Defense and receiving treatment from several clinics across the United States, looked at the impact of HIV disease on mortality according to three time periods: pre (1990–1996), early (1997–1999) and late (2000–2003) HAART eras. The authors showed that the annual mortality rate since the pre-HAART era has been declining (P = 0.003), with an overall 80% decrease in the mortality rates from 1990 to 2003. Crum et al. ^[41] also observed that the median time of survival after HIV seroconversion was 8 years in 1994–1996, 10 years during the earlier HAART era, and 12 years during the late HAART era. This manuscript confirms these observations and in addition, it demonstrates that life expectancy continues to improve with the advent of modern therapies. In a recent study by Walensky et al. ^[44], the authors also observed significant temporal benefits in survival among AIDS patients receiving newer and more potent combination antiretroviral therapies. The authors divided the HIV disease treatment history into six periods between 1989 and 2003, and estimated the per-person survival benefit among those patients receiving treatment in comparison with the absence of treatment. If we focus on the periods starting from 1993, the authors estimated a per-person survival gain ranging from 2.0 years (1993–1995) to 13.3 years (2003). This represents a 6.7-fold increase in survival in their study population. Although this new study and ours used conceptually different cohorts to assess the benefit of disease treatment over time, it is important to note that both studies agree on the clinical benefits of newer therapies.

There are several features of our study that should be highlighted. Our study was carried out within a province-wide treatment programme in which all individuals had access to medical attention, combination antiretroviral therapy, and laboratory monitoring free of charge. We are confident, therefore, that our results are not influenced by access to therapy or incomplete follow-up issues that have often compromised the interpretation of similar population or cohort-based studies. Second, this study was based on treatment-naive individuals, thus our results are not confounded by previous therapy use. Finally, as we have a unique database with 12 years' worth of data, we were able to demonstrate the decrease in mortality and the gains in life expectancy among patients in the early HAART era (1995–1996), when mainly mono antiretroviral therapies were available, to the late HAART era (2002–2004), when mainly triple antiretrovrial combinations were available.

A few limitations associated with our study should be noted. The results from our multivariate analysis rely on information about prognostic variables collected at baseline. All our analyses were conducted on the basis of an intent-to-continue initial regimen category; however, many individuals switched therapy [i.e. any modification (addition or removal) in the original drug regimen] during the course of their treatment history. This is particularly the case for those starting mono or dual therapy and then changing to triple therapy, or for those starting the less tolerable triple therapies initially available, then changing to improved regimens that became available over time. These effects would contribute to an underestimate of the real differences in the effect of the initial regimen or calendar year, respectively. Unfortunately, estimating the effect of the initial regimen on mortality, accounting for regimen change, would be methodologically challenging, because regimen-specific factors such as toxicities will themselves be associated with regimen change. Second, we applied Cox proportional hazard survival models to estimate the effect of the calendar year and initial HAART regimen on the hazards of survival. We adjusted these models for several important variables; however, unmeasured differences may exist among our study population, and for this reason caution is warranted. Finally, our results could have been influenced by residual confounding originating from factors influencing the decision to start treatment and the initial choice of drug regimen. As our HR for the time starting antiretroviral therapy and the initial drug regimen did not change significantly after adjusting for several prognostic factors in the multivariate models (the differences in coefficients were less than 0.10), it may indicate that residual confounding, if present, may not play a significant role in explaining our results.

In conclusion, our results demonstrate a significant and sustained improvement in mortality among HIV-infected men and women over the 12-year time period and among those who received therapy with initial regimens including three or more antiretroviral agents. Our results remained statistically significant even after adjusting for the CD4 cell count, AIDS diagnosis, sex, and age at initiation of antiretroviral therapy. Future research is needed to examine whether further decreases in mortality can continue to be sustained into the future.

Acknowledgements

The authors would like to thank Fern Ragnier, Bonnie Devlin, Elizabeth Ferris, Nada Gataric, Kelly Hsu, Myrna Reginaldo, and Peter Vann for their administrative assistance.

Sponsorship: This work was supported by the Michael Smith Foundation for Health Research through a senior scholar award to R.S.H.

References