Post-surveillance cohort studies and long-term follow-up studies of clinical trials, like the Merck 035 study, continue to suggest the ongoing durability of highly active antiretroviral therapy (HAART) [¹?³]. Mortality among persons on these three- or four-drug regimens continues to remain low, and may be no higher than for other chronic diseases [⁴,⁵]. As one large multinational study indicates, depending upon the baseline characteristics of the persons initiating HAART, subsequent disease progression and death can be as low as 2.4% at 3.5 y [⁴].

Emerging resistance poses a growing threat to the ongoing success and durability of HAART regimens. Nearly 80% of antiretroviral-treated patients with detectable viraemia in one large US-based clinic study had evidence of phenotypic drug resistance [⁶]. The frequency of resistance in this population-based study was significantly higher in patients with a history of antiretroviral use, advanced HIV disease, higher plasma HIV viral load, and the lowest CD4 cell counts. Continuing viral replication during therapy leads to the accumulation of drug-resistance mutations, resulting in increased viral load.

The current impact of the emergence of drug-resistance mutations on mortality, however, is not well characterized in antiretroviral-na?ve patients first starting HAART. Patients may have sustained immunologic function with resistant virus [⁷], and there is limited evidence that reduced sensitivity to antiretrovirals leads to rapid disease progression or death [⁸]. Previously, we documented the selection of resistance in an initially drug-na?ve cohort starting HAART [⁹]. We therefore undertook the present analysis to characterize the determinants of mortality in this prospective cohort study with a median of nearly 5 y of follow-up. The aim of this study was to determine the impact of the emergence of drug-resistance mutations on survival among persons initiating HAART.

Between August 1, 1996, and September 30, 1999, a total of 1,312 antiretroviral-na?ve participants aged 18 y and over initiated triple combination therapy consisting of two nucleosides plus a protease inhibitor or a non-nucleoside reverse transcriptase inhibitor. Of these, 121 (9.2%) were excluded from this analysis for not having both baseline CD4 and plasma HIV-1 RNA level measures available within 6 mo prior to the start of antiretroviral therapy or for having initiated therapy as part of a clinical trial. Among the remaining 1,191 study participants, there were 679 patients with HIV-drug-resistance genotyping done on samples of plasma HIV-1 RNA before initiating HAART at the time of this analysis. HIV-drug resistance to any class was observed in 53 (7.8%) of the patients. Among these patients, 39 (73.6%) exhibited resistance to one class, ten (18.9%) to two classes, and four (7.6%) to three classes. A total of 14 (26.4%) exhibited mutations conferring resistance to lamivudine, 29 (54.7%) to other nucleoside reverse transcriptase inhibitors, eight (15.1%) to non-nucleoside reverse transcriptase inhibitors, and 19 (35.8%) to protease inhibitors. These patients exhibiting baseline resistance were excluded from the analysis, and the total study sample was based on the remaining 1,138 (86.7%) participants. Study participants had HIV-1 RNA levels higher than those of individuals excluded from the analysis (median viral load 87 versus 120 copies/ml; p < 0.001). No other significant associations were noted between the two groups.

Among the remaining HOMER participants (n = 1,138), a total of 229 (20.1%) patients achieved durable viral suppression (HIV-1 RNA levels < 1,000 copies/ml) during the entire follow-up period. There were 776 (68.2%) patients with at least one HIV-1 RNA level measurement at 1,000 copies/ml or above that showed no detectable resistance during the follow-up period, and we assumed these individuals had no drug resistance. Although we systematically genotyped all samples with HIV-1 RNA levels of 1,000 copies/ml or more during follow-up, there were three (0.3%) patients for whom the resistance test failed at least once for technical reasons, and for these particular tests we assumed that no mutation was detected.

Study participants were first prescribed 26 different triple combination antiretroviral regimens. Table 1 shows that over half of these participants (848; 74.5%) initiated therapy with a protease inhibitor, while the rest of the study participants (290; 25.5%) had a regimen that included a non-nucleoside reverse transcriptase inhibitor. A total of 123 (10.8%) commenced therapy in 1996, 402 (35.3%) in 1997, 340 (29.9%) in 1998, and 273 (24.0%) in 1999. The median age was 37 y (inter-quartile range: 32, 44 y), CD4 cell count was 280 cells/mm³ (inter-quartile range: 130, 430 cells/mm³), plasma HIV-1 RNA level was 120,000 copies/ml (inter-quartile range: 43,000, 310,500 copies/ml), physician experience was 44 patients per physician (inter-quartile range: 5, 131 patients per physician). There were 317 (27.9%) participants who had a history of injection drug use and 149 (13.1%) study participants with a prior diagnosis of AIDS. Among all participants, 956 were males (84.0%).

The overall median time of follow-up was 56.4 mo (inter-quartile range: 45.8, 69.2 mo). The median number of CD4 count measurements was 3.59 per year (inter-quartile range: 2.24, 4.59 per year), and the median number of plasma HIV-1 RNA measurements was 3.44 per year (inter-quartile range: 2.56, 4.25 per year). A total of 17,632 CD4 cell counts and 17,358 HIV-1 RNA determinations were collected for participants over the study period. The median numbers of CD4 cell counts and HIV-1 RNA determinations per participant were 15 (inter-quartile range: 7, 23) and 15 (inter-quartile range: 9, 21), respectively. In this study, a switch in drug regimen was recorded when any modification (addition or removal) in the original drug regimen occurred. Men and women switched one or more drugs in their regimen a median three times (inter-quartile range: 1, 5 times) over the follow-up period. Protease inhibitors were added to or removed from a regimen a median of one time (inter-quartile range: 1, 3 times), non-nucleoside reverse transcriptase inhibitors one time (inter-quartile range: 0, 2 times), and nucleoside reverse transcriptase inhibitors three times (inter-quartile range: 1, 5 times). The median time to first switch was 7.8 mo (inter-quartile range: 2.1, 23.6 mo) in this study population.

Drug-resistance testing was completed on a total 3,099 samples over the first 30 mo of follow-up. A median of three (inter-quartile range: 1, 5) samples were tested per participant. HIV-drug resistance to any class was observed in 302 (28.5%) participants. Of these, 19 (6.3%) exhibited resistance to all three classes, 137 (45.4%) to two classes, and 146 (48.3%) to one class. A total of 221 (73.2%) exhibited mutations conferring resistance to lamivudine, 98 (32.5%) to other nucleoside reverse transcriptase inhibitors, 157 (52.0%) to non-nucleoside reverse transcriptase inhibitors, and 66 (21.9%) to protease inhibitors. The median time to first emergence of resistance to any class of drug under study was 16.5 mo (inter-quartile range: 8, 27 mo).

Table 2 presents the association between baseline characteristics, adherence during first yea,r and emergence of resistance. Persons who developed any mutations were more likely to be injection drug users (p = 0.007), to have lower CD4 counts (p < 0.001), and to have higher plasma viral load (p < 0.001) than persons who did not develop any resistance over the first 30 mo.

As of June 30, 2003, a total of 207 deaths were identified in the study population over the follow-up period, with an overall crude mortality rate of 18.2%. The product limit estimate of the cumulative mortality rate at 12 and 24 mo was 4.4% (?0.6%) and 8.7% (?0.8%), respectively. Of these deaths, 30 (14.5%) were to patients who exhibited resistance to one class, 26 (12.6%) to two classes, and one (0.5%) to three classes. A total of 41 (19.8%) deaths were to patients who exhibited resistance to lamivudine, 15 (7.3%) to other nucleosides, 32 (15.5%) to non-nucleoside reverse transcriptase inhibitors, and six (2.9%) to protease inhibitors. Nearly, three-quarters of the deaths involved no resistance to any class of drug (150; 72.5%).

The univariable and multivariable analyses of the baseline and time-dependent factors associated with all cause mortality are presented in Table 3 through four different models. In the univariable analysis for baseline characteristics (Model 1), only age, a prior AIDS diagnosis, protease inhibitor use, physician experience, CD4 count, and HIV-1 RNA level were associated with mortality. When we considered time-dependent factors, the univariable analysis (Model 2) showed us that age, baseline drug combination, physician experience, a prior AIDS diagnosis, protease inhibitor use, adherence, CD4 cell count, HIV-1 RNA level, and resistance to any antiretroviral were associated with mortality. We also observed that resistance to some classes of drug, more specifically to lamivudine and non-nucleoside reverse transcriptase inhibitors, were also associated with mortality. After controlling for other prognostic explanatory variables, Model 3 shows that there is a strong association between emergence of any resistance and risk of death (hazard ratio: 1.58 [95% confidence interval (CI): 1.14, 2.20]). The heterogeneity in the association between resistance and mortality also depends on the type of resistance that was acquired (Model 4), with persons who exhibited resistance to non-nucleoside reverse transcriptase inhibitors being 2.57 times (95% CI: 1.74, 3.78) more likely to die.

The same analyses were repeated with non-accidental deaths as the outcome of interest (n = 160; 77.3%). Individuals dying from non-accidental deaths in the study had less experienced physicians (median: 16 patients per physician [inter-quartile range: 2, 81 patients per physician]), worse CD4 cell counts (median: 150 cells/mm³ [inter-quartile range: 45, 310 cells/mm³]), and higher viral load levels (median: 175,000 copies/ml [inter-quartile range: 88,000, 405,000 copies/ml]) than the individuals included in the all cause mortality (see Table 1). When we assessed the association between survival and resistance in this population we observed that persons who exhibited reduced sensitivity to any antiretroviral had death rates that were 1.51 times (95% CI: 1.04, 2.20) higher than those who did not exhibit resistance. We also looked at the individuals dying via accidental death (n = 47; 22.7%). These individuals had more experienced physicians (median: 53 patients per physician [inter-quartile range: 12, 131 patients per physician]), better CD4 cell counts (median: 300 cells/mm³ [inter-quartile range: 210, 470 cells/mm³]), and lower viral load levels (median: 129,000 copies/ml [inter-quartile range: 39,000, 210,000 copies/ml]) than the individuals included in the all cause mortality (see Table 1). When we assessed the association between survival and emergence of resistance in this population we observed that the hazard ratio was 1.83 (95% CI: 0.92, 3.63).

Table 4 presents the final marginal structural models of the causal effect of drug resistance on mortality. In all models, the time-dependent confounders were protease inhibitor/non-nucleoside experience, AIDS diagnosis, CD4 cell count, viral load, and HAART regimen. The exposures of main interest were emergence of any resistance, and emergence of resistance to each separate class of mutation (lamivudine, other nucleosides, non-nucleosides, and protease inhibitors). Model 1 considers the emergence of any resistance as the exposure of main interest. The results show that persons who developed any resistance had death rates that were 1.75 times (95% CI: 1.27, 2.43) higher than those who did not exhibit any resistance. When we considered each separate class of mutation as the exposure of interest (Models 2?5), while controlling for the presence/absence of other mutations, we observed that those persons who exhibited reduced sensitivity to non-nucleoside reverse transcriptase inhibitors had the highest risk: they were 3.02 times (95% CI: 1.99, 4.57) more likely to die than those who did not exhibit this type of resistance.

Our results demonstrate that emergence of resistance to non-nucleoside reverse transcriptase inhibitors was associated with a greater risk of subsequent death than was emergence of resistance to any other class of drug. Our finding that resistance to non-nucleoside reverse transcriptase inhibitors was significantly associated with mortality even after adjustment for plasma HIV-1 RNA levels, adherence, and CD4 cell counts suggests resistant viruses may have differences in fitness [¹⁹]. Resistance to non-nucleoside reverse transcriptase inhibitor therapy has also previously been shown to occur at low to moderate levels of adherence, indicating that non-nucleoside reverse transcriptase resistance may be an independent marker for poor adherence in our population [²⁰]. Mortality was not associated with viruses resistant to protease inhibitors, lamivudine, or other nucleoside reverse transcriptase inhibitors.

The emergence of drug resistance to antiretrovirals in this drug-na?ve cohort remains low [⁸]. The median time to first emergence of resistance to any class of drug was just over a year. HIV-drug resistance to any class was observed in just over a quarter of participants over 30 mo, with less than 10% of patients exhibiting resistance to all three classes. In contrast to other studies that included both therapy-na?ve and experienced patients [⁶], our results are likely much closer to true levels of multiple drug resistance in a population initiating HAART.

Most patients in our study died with wild-type virus. Nearly, three-quarters of the deaths were of patients who exhibited no resistance to any class of drug, and less than 2% exhibited resistance to all three classes. Such patients may have had insufficient drug exposure because of poor adherence, thus selecting for resistant virus. We have previously shown significant differences in survival among adherent and non-adherent patients at all levels of CD4 count, with a particularly dramatic impact on survival in patients with highly advanced disease [⁹]. These findings are independent of whether the measurement of adherence is based on prescription refills or antiretroviral concentrations in untimed plasma samples [⁹,²¹?²³]. Similarly, in an indigent cohort in San Francisco, the level of adherence was closely associated with longitudinal viral suppression and the rate of subsequent disease progression [²⁴].

Two recent studies have reported on the association between drug resistance and risk of death [⁸,²⁵]. Lucas et al. [²⁵], based on a cohort of non-na?ve individuals starting HAART, found that emergence of resistance was not linked to increased mortality rates in a multivariable analysis, adjusting for prognostic factors such as HIV-1 exposure group, CD4 cell count at resistance testing, mean log₁₀ HIV-1 RNA level during prior treatment with HAART, and whether or not HAART was used during the 6 mo following resistance testing. There are two important differences between this study and ours. First, Lucas et al. failed to control for several other important fixed and time-varying confounder factors that could have influenced their results, such as adherence, HAART regimen, and injection drug use, to cite a few. Second, they did not exclude from the analysis individuals who had received HAART previously, and who had already presented some type of mutation, which could have potentially biased their results. Another study, by Recsky et al. [⁸], based on a subpopulation of individuals from our study that died between July 1997 and December 2001, also failed to demonstrate an association between drug resistance and mortality. The authors of this latter paper used univariable analyses to compare the presence or absence of resistance at the time of death across several characteristics: age (at initiation of HAART and at death), sex, injection drug use status, CD4 cell count and plasma HIV-1 RNA level at the time right before the genotype test, adherence, and time undergoing treatment. The only difference they found between these two groups was due to the duration of treatment, where individuals who presented any drug resistance had longer median duration of therapy than those without any resistance (32 versus 15 mo; p < 0.001). We believe that the differences among the results of these two studies and ours are due to differences between study populations (e.g., comparison groups and definitions of cohort) and differences in methodology used to assess the relationship between outcome and exposures.

Although we concluded that emergence of resistance was strongly associated with elevated risk of mortality, there are two important potential sources of bias that could explain our results. First, HOMER patients with samples of HIV-1 RNA of less than 1,000 copies/ml during the entire follow-up period (n = 229; 20.1%) were assumed to have no drug-resistance mutations. We therefore conducted a sensitivity analysis restricted to patients with plasma viral load levels of 1,000 copies/ml or more to assess the impact of this assumption on our results and the robustness of the marginal structural models in dealing with this possible source of bias. We observed that patients who showed the emergence of any resistance were 1.68 times (95%CI: 1.19, 2.38) more likely to die than those who did not, which is consistent with the results found in our original analysis. Second, in the present study we did not exclude patients who achieved durable viral elevation (HIV-1 RNA levels ? 1,000 copies/ml) during the entire follow-up period (n = 130; 11.4%). To assess whether this group drove our mortality trends, and not emergence of resistance per se, we conducted a second sensitivity analysis restricted to patients with no durable viral elevation. We observed that those patients that developed any type of resistance were 2.09 times (95%CI: 1.41, 3.10) more likely to die than those who did not, which is also consistent with our previous results.

There are several features of our study that should be highlighted. First, our study was carried out within a province-wide treatment program, 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, a factor that has often compromised the interpretation of similar population- and cohort-based studies. Second, this study was based on treatment-na?ve individuals, thus our results are not confounded by previous therapy use. Third, delayed reporting was not likely a factor; as the vast majority of deaths are reported within 3 mo of death through active follow-up with physicians and hospitals and regular linkages. Finally, although we adjusted our analyses for pertinent demographic and clinical characteristics, like all studies of patients treated in observational cohorts, unmeasured differences may exist among study populations, and for this reason caution is warranted.

In summary, we demonstrated that emergence of resistance to non-nucleoside reverse transcriptase inhibitors was associated with a greater risk of subsequent death than was emergence of resistance to other drug classes. Future research is needed to identify the particular subpopulations of men and women at greatest risk. These efforts also need to elucidate the impact of resistance over a longer follow-up period.

n102Competing Interests: RSH, DRB, CA, EW, JSGM, and PRH have received honorariums, travel grants to attend conferences, and research grants from pharmaceutical companies working in the area of HIV/AIDS.

ack1Author contributions. RSH, DRB, VDL, CA, SB, EW, JSGM, and PRH designed the study. RSH, BY, WWYD, and PRH participated in data gathering. VDL and SB participated in statistical analysis. RSH, DRB, VDL, CA, SB, BY, EW, WWYD, JSGM, and PRH contributed to writing the paper.

n101Funding: This work was supported by the Michael Smith Foundation for Health Research through a Senior Scholar Award to RSH and by the Canadian Institutes of Health Research through a peer-reviewed grant. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

We thank Jennifer Adachi, Bonnie Devlin, Elizabeth Ferris, Nada Gataric, Kelly Hsu, Myrna Reginaldo, Peter Vann, Beheroze Sattha, Justin Woodward, Theresa Mo, Shawna Langridge, and Lillian Ting for their research and administrative assistance.