Crude proportions of subjects with virological failure by year of starting triple-combination antiretroviral therapy (CART). Virological failure is defined as viral load (VL) greater than 500 copies/mL based on the first VL from 6 to 12 months. Strategy A includes all patients, with missing VL measurements counted as virological failure; strategy B, all patients with VL measurements; and strategy C, all patients receiving any antiretroviral therapy (ART) at VL measurement. VL indicates viral load.
Unadjusted and adjusted risk ratios (with 95% confidence intervals) of virological failure by year of starting triple-combination antiretroviral therapy (CART). The reference year is 1999. Virological failure is defined as viral load (VL) greater than 500 copies/mL based on the first VL from 6 to 12 months. Strategy A (A) includes all patients, with missing VL measurements counted as virological failure; strategy B (B), all patients with VL measurements; and strategy C (C), all patients receiving any antiretroviral therapy at VL measurement. Unadjusted models were adjusted for cohort only. Adjusted 1 models included cohort, risk group (incorporating sex and exposure group), previous AIDS diagnosis, pre-CART CD4 count, age at starting CART, and pre-CART log10 VL. Adjusted 2 models included all factors in adjusted 1 models plus the starting regimen defined by the nucleoside combination and the third drug.
Lampe FC, Gatell JM, Staszewski S, Johnson MA, Pradier C, Gill MJ, de Lazzari E, Dauer B, Youle M, Fontas E, Krentz HB, Phillips AN. Changes Over Time in Risk of Initial Virological Failure of Combination Antiretroviral TherapyA Multicohort Analysis, 1996 to 2002. Arch Intern Med. 2006;166(5):521-528. doi:10.1001/archinte.166.5.521
Triple-combination antiretroviral therapy (CART) for human immunodeficiency virus infection has been in use for almost a decade, but the extent to which treatment success has changed is uncertain. We examined risk of initial virological failure of CART according to the year of starting therapy.
We included subjects from 5 complete clinic cohorts in Europe and Canada who started CART without previous antiretroviral therapy from 1996 to 2002 with 1 or more pre-CART viral load (VL) measurement and CD4 count. Based on the first VL measurement from 6 to 12 months after CART initiation, virological failure was defined as a VL of more than 500 copies/mL. We used the following 3 inclusion strategies: (1) including all subjects, with missing VL measurement counted as virological failure (n = 3825; strategy A); (2) including all subjects with VL measurement (n = 3120; strategy B); and (3) including all subjects receiving antiretroviral therapy at VL measurement (n = 2890; strategy C).
From 1996 to 2002, risk of virological failure fell from 38.9% to 24.8% for strategy A, 28.4% to 12.0% for strategy B, and 22.8% to 8.2% for strategy C. Estimated relative reductions in risk (95% confidence interval) over the 7-year period, adjusted for cohort, demographic factors, pre-CART VL and CD4 count, and previous AIDS, were 48% (39%-56%), 64% (53%-73%), and 79% (69%-85%) for strategies A, B, and C, respectively. Reductions in risk were greatest from 1996 to 1999, with weaker trends subsequently. Trends remained but were attenuated after further adjustment for the starting regimen.
Over a 7-year period of CART use in clinical practice, risk of initial virological failure of treatment has halved at least. These data suggest the trend is due to improvements in CART regimens and greater effectiveness of their use.
Triple-combination antiretroviral treatment (CART) for human immunodeficiency virus (HIV) infection has been in use for almost a decade.1,2 The primary initial goal of treatment is to suppress viral replication, as measured by plasma viral load (VL). Viral suppression enables recovery of CD4 lymphocyte numbers3 and hence CD4-mediated immune responses, and thereby reduces risk of opportunistic infections and death.4- 6 Initial response to CART is important, as inadequate viral suppression may result in the development of drug resistance,7,8 compromising the effect of the current CART regimen and reducing future treatment options. However, the likelihood of initial CART success may have changed over time, owing to changing drug options, accumulating clinical experience, and increasing understanding of treatment adherence issues by patients and clinicians. We combined data from 5 HIV clinic cohorts to examine trends from 1996 to 2002 in the risk of initial virological and immunological failure of CART according to the year of starting therapy, among previously untreated subjects.
We used observational databases from 5 HIV clinics in Barcelona, Spain; Frankfurt, Germany; London, England; Calgary, Alberta; and Nice, Italy9- 13 (Table 1). Each database contains prospectively collected demographic, clinical, and laboratory information for all patients with HIV seen in each clinic, including start and stop dates for each individual antiretroviral therapy (ART) drug. Information on adherence was not routinely collected. Subjects were selected who (1) started CART (≥3 drugs, among which was included a protease inhibitor, a nonnucleoside reverse transcriptase inhibitor, or abacavir succinate) from 1996 to 2002 (inclusive) at 16 years or older and had received no previous ART; (2) started CART at least 1 year before the date of last cohort follow-up; and (3) had 1 or more pre-CART VL measurements and CD4 counts (≤1 year before and ≤5 days after starting CART). All VL assays used had limits of detection of no more than 500 copies/mL.
Based on the first VL measurement from 6 to 12 months (days 183-365) after CART initiation, virological failure was defined as a VL of more than 500 copies/mL according to the following 3 inclusion strategies: (1) including all subjects starting CART, those with missing VL measurements from 6 to 12 months were counted as having virological failure (strategy A); (2) including all subjects with 1 or more VL measurement from 6 to 12 months (strategy B); and (3) including all subjects receiving (any) ART at the time of the first VL measurement from 6 to 12 months (strategy C).
Immunological response was defined according to the first CD4 count from 6 to 12 months after CART initiation, and was counted as failure if the increase in CD4 count from the pre-CART value was less than 50 cells/μL. We used the 3 inclusion strategies (A, B, and C) described in the “Virological Failure” subsection. We also examined the risk of a new clinical AIDS event or death (due to any cause) within the first year of starting CART. The AIDS events were defined according to clinical diagnoses without including CD4 count criteria.
We examined trends in virological and immunological failure from 1996 to 2002 using Poisson regression models with robust standard errors to generate risk ratios and 95% confidence intervals (CIs).14 We used PROC GENMOD in SAS release 8.02 (SAS Institute Inc, Cary, NC). We examined trends in 2 ways. First, calendar year of starting CART was fitted as a categorical variable to assess the pattern of trends. We used 1999 as the reference category because we hypothesized that trends may differ in the earlier and more recent periods. Second, we used calendar year as a continuous variable in models to estimate average trends from 1996 to 2002 overall and during the time periods from 1996 to 1999 (inclusive) and 1999 to 2002 (inclusive). Risk ratios from these models are referred to as risk ratios for trend per calendar year and represent the average risk ratio for each increase in the year of starting CART. Unadjusted models were adjusted for cohort only. Adjusted models included cohort, risk group (incorporating sex and exposure group), age at starting CART, previous AIDS diagnosis, pre-CART CD4 count, and pre-CART log10 VL. In further models, we additionally adjusted for the starting regimen defined by the nucleoside combination and the third drug. Variations in trend according to cohort and other factors were assessed using tests of interaction with calendar year as a continuous variable. We performed separate analyses for inclusion strategies A, B, and C throughout.
Overall, 3825 subjects from the 5 cohorts fulfilled the inclusion criteria. The numbers starting CART according to calendar year are given in Table 2. There were changes over time in the characteristics of subjects starting CART (Table 2). Compared with the early period, in recent years there were higher proportions of men and women with heterosexual risk and lower proportions of men with homosexual risk and people exposed via intravenous drug use. Over time there was a small increase in median age at starting CART and a decrease in median pre-CART CD4 count. Median pre-CART VL and the prevalence of previous AIDS varied little over time. The most frequent type of starting regimen changed over time from a nonboosted protease-inhibitor–containing combination in the early period to a boosted protease inhibitor–containing combination or a nonnucleoside reverse transcriptase inhibitor–containing combination in recent years. Throughout the period, the most common nucleoside combination in the starting regimen was zidovudine with lamivudine. Use of stavudine with lamivudine or stavudine with didanosine became less common, whereas use of other nucleoside combinations tended to increase.
Of all 3825 subjects who started CART, 3120 (81.6%) had a VL measurement from 6 to 12 months after CART initiation (median, 31.3 weeks), and 2890 (75.6%) were receiving any ART at this time. The percentages of subjects with virological failure according to inclusion strategies A, B, and C were 34.1% (1305/3825), 19.2% (600/3120), and 14.0% (406/2890), respectively. Figure 1 shows crude percentages of subjects with virological failure by year of starting CART. From 1996 to 2002, risk fell from 38.9% to 24.8% for strategy A, 28.4% to 12.0% for strategy B, and 22.8% to 8.2% for strategy C. Figure 2 shows risk ratios adjusted for cohort only (labeled unadjusted) and for cohort plus age, risk group, pre-CART CD4 count, pre-CART VL, and previous AIDS (labeled adjusted 1). Adjustment for these factors had little effect on the pattern of trends; risk fell greatly from 1996 to 1999, with evidence of a further decrease by 2002. Table 3 presents risk ratios for trend per calendar year during the 7-year period, which were 0.90, 0.84, and 0.77 for strategies A, B, and C, respectively, from adjusted models (P<.001). This represents a 48% (95% CI, 39%-56%) reduction in risk over the entire 7-year period for strategy A, a 64% (95% CI, 53%-73%) reduction for strategy B, and a 79% (95% CI, 69%-85%) reduction for strategy C. The trend was greatest when we considered only subjects who continued to receive ART (strategy C), because reductions over time in the probability of having a missing VL and of discontinuing ART were modest (Figure 1), and therefore including these subjects diluted the trend. Table 3 also presents trends according to time period. Reduction in risk of virological failure was most marked from 1996 to 1999, with weaker trends from 1999 to 2002.
We examined whether reductions in risk of virological failure could be explained by changes over time in starting regimens. Figure 2 and Table 3 also present risk ratios after further adjustment for the starting regimen (labeled adjusted 2), defined by the third drug and nucleoside combination. Trends were attenuated after adjustment for starting regimen, but remained highly significant across the 7-year period. There remained a substantial reduction in risk from 1996 to 1999, but there was no clear trend subsequently.
Table 4 provides the risk ratios of virological failure for demographic factors, pre-CART characteristics, and starting regimen from models including all factors plus cohort and calendar year. Risk group was strongly associated with virological failure; heterosexual men and women, intravenous drug users, and subjects with other exposure had greater risk of failure than did homosexual men. Risk of failure was lower among older people and was higher among those with previous AIDS. These associations were broadly consistent for strategies A, B, and C. A higher pre-CART VL was associated with greater risk of failure for strategies B and C. Risk of virological failure also varied according to starting regimen. Compared with regimens including other single nonboosted protease inhibitors, regimens containing saquinavir mesylate hard-gel capsules tended to be associated with higher risk of failure, whereas boosted protease inhibitor– and efavirenz-containing regimens tended to be associated with lower risk. There was less variation in risk according to nucleoside combination, although there was some suggestion that lamivudine combined with zidovudine was associated with lower risk than other combinations.
Trends in virological failure did not differ significantly by age group (<30 vs ≥30 years), previous AIDS, baseline CD4 count (<200 vs ≥200 cells/μL), or baseline VL (<5 vs ≥5 log10 copies/mL) (P>.05 for interaction between each factor and calendar year, from adjusted models). However, there was evidence that trends varied according to risk group. Reductions over time in failure risk tended to be greater among homosexual men than other risk groups combined. Risk ratios for trend per calendar year over the 7-year period for homosexual men vs other groups were 0.83 vs 0.93 for strategy A; 0.77 vs 0.87 for strategy B; and 0.74 vs 0.79 for strategy C (P<.001, P = .02, and P = .28, respectively, for interaction). In addition, trends varied according to clinic cohort. Risk ratios for trend per calendar year over the 7-year period for each of the 5 cohorts were 0.81, 0.91, 0.92, 0.93, and 1.06 for strategy A; 0.77, 0.83, 0.83, 0.89, and 0.95 for strategy B; and 0.70, 0.78, 0.57, 0.82, and 0.95 for strategy C (P<.001, P = .16, and P = .02, respectively, for interaction).
The overall proportions of subjects with immunological failure were 38.5% (1473/3825), 24.5% (765/3117), and 21.3% (613/2884) for strategies A, B, and C, respectively. Table 5 presents crude percentages with immunological failure, unadjusted and adjusted risk ratios, and median changes in CD4 count for each calendar year of starting CART. Trends in immunological failure were highly significant in unadjusted and adjusted analyses, but were less marked than those seen for virological failure, and similar for each inclusion strategy. Risk ratios (95% CIs) for trend per calendar year were 0.94 (0.92-0.97), 0.95 (0.92-0.99), and 0.94 (0.91-0.99) for strategies A, B, and C, respectively (P<.01 for all, adjusted for cohort, age, risk group, pre-CART viral load and CD4 count, and previous AIDS. This corresponds to relative reductions in risk (95% CIs) over the entire 7-year period of 30% (19%-39%), 26% (7%-41%), and 29% (8%-44%) for strategies A, B, and C, respectively. The magnitude of trends in immunological failure was similar in the 2 time periods (1996-1999 and 1999-2002), and trends were not attenuated by further adjustment for starting regimen (data not shown). Overall, 290 patients had a new clinical AIDS event or died within the first year of starting CART. The percentages in order from 1996 to 2002 were 7.4%, 6.7%, 6.6%, 10.1%, 9.5%, 10.0%, and 7.4%, with no significant trend over time (adjusted risk ratio [95% CI] for trend per calendar year, 1.03 [0.97-1.10]; P = .30].
These results demonstrate dramatic improvements in initial treatment success of first CART among previously untreated individuals in routine clinical practice. After accounting for changes in patient characteristics, estimated relative reductions in risk of virological failure from 1996 to 2002 were 48%, 64%, and 79% for our 3 inclusion strategies. Risk of immunological failure also fell substantially during the period, although trends were weaker than those for virological failure.
We chose definitions of virological and immunological failure based on a single measurement to minimize bias due to changes in measurement frequency over time, and used 3 inclusion strategies. Substantial reductions over time in risk of failure were apparent in each case. We excluded 1417 otherwise eligible subjects who had no pretreatment VL measurement and/or CD4 count within the specified window. Inclusion of these subjects in an unadjusted analysis resulted in trend estimates identical to those of the unadjusted trends presented in this report, suggesting that their exclusion did not bias trends. Our analysis could not account for factors such as coinfection, comorbidity, sociodemographic factors, and participation in clinical trials. However, adjustment for known demographic and pre-CART characteristics did not attenuate trends. There was evidence of heterogeneity in virological failure trends between clinics, with the magnitude of trends differing according to cohort, although no single cohort was driving the trends. These results are generalizable to the extent that they represent an average across a group of representative HIV clinics in industrialized countries. In Europe and Canada, health care is free at the point of use, and most patients with a diagnosis of HIV receive care at specialist clinics. It is possible that virological failure rates and trends may differ in industrialized countries with other systems of HIV care. This study was concerned with initial response to CART; trends in long-term failure risk may differ.
Calendar time is recognized as a potentially important correlate of virological response to CART,15- 18 but few studies have quantified trends in initial treatment success since CART introduction. In the Swiss cohort study, subjects starting protease-inhibitor–containing CART in 1997 through 1998 were 30% more likely to achieve viral suppression than those starting in 1995 through 1996 (adjusted for previous ART).19 Subsequent information comes predominantly from the EuroSIDA study. Significant improvements over time in virological response to CART were found for ART-naïve subjects in the EuroSIDA study.20 As with this present study, the greatest improvements occurred in the initial years, from 1996-1997 to 1998-1999, with smaller changes subsequently. The effect on trend estimates of accounting for the starting regimen was not presented. In an analysis of treatment-experienced individuals in the EuroSIDA study who started nevirapine- or efavirenz-containing CART from mid-1997 to 2000, the strong inverse associations of calendar year with virological failure and clinical progression were no longer apparent after adjustment for regimen characteristics, previous ART, and other factors.21 This appears consistent with our results among ART-naïve subjects, which showed no clear evidence of improvement in virological response since 1999, once the starting regimen was taken into account. The Danish HIV cohort found a substantial reduction in risk of triple-class virological failure among patients starting CART from 1996 to 2003 (including ART-naïve subjects and those who had received previous ART).22 In this case, however, the trend appeared at least as strong from 1999 onward as in the early years, which may also reflect trends in response to second- or third-line as well as initial CART. Further evidence from the EuroSIDA study also suggests an improvement over time in virological and immunological responses to salvage regimens, although response patterns in specific time periods were not examined.23
When CART was first used, the most common starting regimen was a single nonboosted protease inhibitor in combination with 2 nucleosides. This combination tended to be associated with poorer virological outcome than 2 nucleosides combined with a boosted protease inhibitor or efavirenz—the most common regimens in recent years. Because of the strong correlation between starting regimen and calendar year, it is difficult to disentangle the effects of each factor. Our analysis suggests that changes in starting regimens partially accounted for the reduction in risk of virological failure, but did not fully explain the trend, particularly during the early period from 1996 to 1999. In addition, trends were not explained by decreasing rates of ART discontinuation or increasing rates of early treatment change, because reduction in risk of virological failure was strongest in the analysis including only subjects who continued to receive ART (strategy C), and a similarly marked trend was apparent among the subgroup who continued their starting regimen (data not shown).
The early improvement in virological response that was independent of the starting regimen could reflect a general increase in adherence to treatment, resulting from accumulating clinical experience, more effective clinical management, and perhaps an increase in patients' knowledge about treatment. The existence of a learning curve in the implementation of a new treatment has been recognized for medical and surgical interventions,24 and may be particularly applicable to complex drug combinations such as those used for HIV. Over time, physicians may learn how best to advise patients regarding treatment adherence, how frequently to monitor patients, and how to manage adverse effects. Additional services (such as pharmacy services and counseling) may also expand and develop over time. Other studies give some support to the role of clinical experience in HIV treatment success in the CART25 and pre-CART24,26 eras.
The marked reduction in risk of virological failure from 1996 to 1999 may also reflect a change in emphasis from achieving a reduction in VL to attaining maximal viral suppression,27- 29 particularly in view of the introduction of increasingly sensitive VL assays. This may be one reason why trends in virological failure were stronger than those for immunological failure in the initial years. Other possible reasons are that immunological response may be less sensitive to factors such as small differences in adherence or regimen potency, and that CD4 count is a more variable measure than viral load. There was no trend over time in risk of clinical AIDS or death during the first year after CART. Initial treatment failure may have an impact on these end points during a longer period, particularly among subjects who start CART with moderate or high CD4 counts.
During a 7-year period of CART use in clinical practice, the risk of initial virological failure of treatment among these clinic populations has fallen by at least half. This trend is partly explained by improvements in starting regimens, but other factors such as increasing clinical experience probably played an important role, particularly in the first few years. For subjects starting CART who continue to receive treatment, initial failure risk is now very low and may have fallen as low as is realistically possible. The small remaining risk of initial failure despite continued ART may be due at least in part to factors such as suboptimal adherence or infection with a resistant virus. Initial failure due to complete treatment discontinuation or loss to follow-up remains more common, suggesting potential for continued improvement in future years with simpler, less toxic regimens and increasingly effective clinical care. Although clinical circumstances and drugs options are likely to be quite different in resource-poor areas, these results may also have implications for CART introduction in new settings, suggesting that notable improvements in initial treatment success may occur over time, especially during the first few years of CART use.
Correspondence: Fiona C. Lampe, PhD, Department of Primary Care and Population Sciences, Royal Free and University College Medical School, Royal Free Campus, Rowland Hill Street, London NW3 2PF, England (firstname.lastname@example.org).
Accepted for Publication: October 21, 2005.
Author Contributions: Dr Lampe had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Financial Disclosure: None.
Acknowledgment: We thank the following for their contributions to the study: J. M. Miro, J. Mallolas, F. Garcia, E. Martinez, and J. L. Blanco for clinical participation, J. A. Arnaiz for data management, and E. de Lazzari for biostatistics (Hospital Provincial HIV cohort, Barcelona, Spain); all of the clinic staff and B. Jennings for database management (Goethe University HIV cohort, Frankfurt, Germany); J. Ballinger, S. Bhagani, R. Breen, P. Byrne, A. Carroll, I. Cropley, Z. Cuthbertson, T. Drinkwater, T. Fernandez, A. Geretti, G. Murphy, D. Ivens, M. Johnson, S. Kinloch-de Loes, M. Lipman, S. Madge, B. Prinz, D. Robertson Bell, S. Shah, L. Swaden, M. Tyrer, and M. Youle for clinical participation, C. Chaloner, H. Gumley, J. Holloway, J. Puradiredja, J. Sweeney, and R. Tsintas for data management, W. Bannister, L. Bansi, A. Cozzi-Lepri, Z. Fox, T. Hill, F. Lampe, A. Mocroft, A. Phillips, C. Sabin, and C. Smith for epidemiology and biostatistics, E. Amoah, G. Clewley, L. Dann, B. Gregory, I. Jani, G. Janossy, M. Kahan, C. Loveday, and M. Thomas for laboratory support (Royal Free Centre for HIV Medicine, London, England); M. Henry and B. Beckthold for data management (Southern Alberta HIV cohort); and C. Pradier, E. Fontas, and C. Caissotti for central coordination, P. Dellamonica, L. Bentz, E. Bernard, S. Chaillou, E. Cua, F. de Salvador-Guillouet, J. Durant, R. Guttman, L. Héripret, V. Mondain-Miton, I. Perbost, B. Prouvost-Keller, P. Pugliese, V. Rahelinirina, P. M. Roger, and F. Vandenbos for clinical participation (Nice HIV cohort, Italy).