The UK Collaborative Group on HIV Drug Resistance, UK CHIC Study Group

Introduction

While viral load failure of antiretroviral therapy with detected presence of resistance mutations is relatively common ^[1–3], much of this arises in the subgroup of people who initially started ART between 1987–1996, in whom nucleoside mono- or dual- therapy were used initially. However, the major issue for the long term future prospects for people starting therapy today is to understand the rate of development of drug resistance in those who started therapy with 3 or more drugs, as this is now the standard of care. While some randomized trials have produced estimates over up to 3 years ^[4,5], clinic-based cohorts have the potential to provide estimates which are over a longer period of follow-up and better reflect the situation in routine practice. In this paper we present estimates of the cumulative risk of resistance development over up to 6 years in patients from six clinics who started ART with 3 or more drugs.

Patients and Methods

The UK Collaborative HIV Cohort Study (UK CHIC) contains data on all patients seen at one or more of six large HIV clinics in London and Brighton (see appendix) since January 1996. The study has been described in detail ^[6]. In brief, data collected include all information used as part of routine clinical care, including demographic information, all start and stop dates of antiretrovirals, CD4 cell counts, viral loads, AIDS diseases and dates of death. Data on resistance were obtained from a linked database, the UK HIV Drug Resistance Database, which contains information on resistance tests performed on behalf of most HIV clinics in the UK.

Analyses Performed / Statistical Methods

Patients eligible for inclusion in the analysis were all those in UK CHIC who used one of the following regimens as their first ART regimen: two nucleosides plus either a single PI, a PI with ritonavir, abacavir, or an NNRTI. Dose of drugs was not recorded so it was not possible to know if ritonavir used with another PI was always used as a pharmacological booster, but knowledge of the regimens used in the clinics would suggest that this was the case in over 90% of instances in which ritonavir was used with another PI. The other eligibility criteria were, first, that if a pre-ART viral load was measured it must be above 400 copies/mL and, second, at least one viral load measure had to be available more than 24 weeks from the start of ART.

Viral load failure was defined as two consecutive viral load values > 1000 copies/mL mL at least 6 months after starting therapy, or one value followed by the starting of at least one new drug. The exception to this was when the first viral load value above 1000 copies/mL was obtained while the patient was off ART, which was not included as an incidence of viral load failure. The date of failure was the date of the first viral load > 1000 copies/mL. While this was the definition used in our main analyses (definition 1), we also assessed the effect of varying the criteria as follows; definition 2: the threshold level of 1000 copies/mL was changed to 500 copies/mL; definition 3: the requirement for two consecutive values was dropped – one being sufficient; definition 4: viral load values > 1000 copies/mL while the patients was off ART were counted; definition 5: those lost to follow-up and those who died were counted as having experienced viral load failure.

For assessing the occurrence of resistance mutations, the following mutations which have been associated with reduced drug susceptibility were considered: Nucleoside reverse transcriptase inhibitors (nucleosides): M41L, E44D, A62V, K65R, D67N, T69N/D, K70R, L74V, V75I, F77L, Y115F, F116Y, V118I, Q151M, M184V/I, L210W, K219Q, K219E, T215any, any 69 insertion (of which TAMS: M41L, D67N, K70R, L210W, K219Q/E, T215any); NNRTIs: L100I, K103N, V106M/A, V108I, Y181C/I, Y188C/L/H, P225H, M230L, P236L, G190any); PIs: D30N, L33I/F/V, M46I, M46L, G48V, I50V/L, V82A/F/T/S/L, I84V/A/C, L90M. This list is very similar to the International AIDS Society – USA (IAS –USA) list of major mutations, with small modifications ^[7].

Kaplan-Meier estimation was used to assess the proportion of patients who experience viral load failure by a given time. The same approach was used to estimate the proportion of patients for whom one or more resistance mutations had been detected by a given time. Follow-up was right censored at the last viral load for the time to viral load failure and at the date of the last resistance test or viral load measure for the time to resistance mutation detection. Logistic regression was used to assess factors associated with having a resistance test result around the time of viral load failure (between 6 months before and 1 year afterwards) ^[8]. Cox regression models were used to assess factors associated with the hazard of detecting resistance mutations ^[8]. Models were stratified by clinic and calendar year of starting ART, so patients were compared with others in the same clinic starting ART in the same calendar year.

Results

A total of 4306 patients were included and their characteristics are shown in Table 1. Most (80%) were male, most were infected through sex and the median age was 35 years. At start of ART the median CD4 count was 183 /mm³, the median viral load was 5.0 log copies/mL and 21% had previously experienced an AIDS disease. The most commonly used regimens were those including two nucleosides plus an NNRTI; 1436 (33%) started with a single PI, 279 (6%) with a PI plus ritonavir, 156 (4%) with triple nucleosides and 2435 (57%) with an NNRTI. The most common nucleosides in the starting regimen were lamivudine (used in 78% of patients), zidovudine (59%), stavudine (38%) and didanosine (21%). The NNRTI's, nevirapine and efavirenz were used by 29% and 28%, respectively. The most commonly used PI's were nelfinavir (13%) and indinavir (11% alone and 2% with ritonavir).

Table 1

The median follow-up to time of last viral load was 160 weeks (3.1 years). 790 (19%) patients were followed for more than 5 years. The percentage of patients lost to follow-up (defined as last viral load before 1 Jan 2002, in patients who were not known to have died) was 5% at 2 years, 10% at 4 years and 11% at 6 years. A median of 3.9 viral load values per patient per year were obtained. Changes to the original drug regimen were frequent. Kaplan-Meier estimates were 48% starting at least one new drug by 2 years, 69% by 4 years and 80% by 6 years. Equivalent values for starting a new class of drugs were 19% by 2 years, 35% by 4 years and 47% by 6 years. For those who started with an NNRTI regimen, the probability of switching to a PI was 11% by 2 years, 18% by 4 years and 26% by 6 years. For those who started with a regimen containing a single PI, the probability of switching to an NNRTI was 27% by 2 years, 51% by 4 years and 61% by 6 years. For those who started with a regimen containing a PI with ritonavir, the probability of switching to an NNRTI was 27% by 2 years and 42% by 4 years. Finally, for those who started with abacavir with two other nucleosides the probability of switching to an NNRTI or a PI was 29% by 2 years and 35% by 4 years.

A total of 1057 (25%) patients experienced viral load failure. Fig. 1 shows the Kaplan-Meier estimates of the probability of viral load failure according to the time from starting ART; 21% (95% CI 22%-23%) experience viral load failure by 2 years, 30% (95% CI 28%-32%) by 4 years and 38% (95% CI 35% – 41%) by 6 years. Table 2 shows the Kaplan-Meier estimates of the percent with viral load failure obtained if the definition is modified in various ways from our main approach (definition 1 in Table 2), changing one criterion at a time.

Fig. 1

Table 2

913 (21%) of the patients had a resistance test performed on a sample taken prior to the start of ART that was successful (ie a sequence obtained – in some cases the sample was tested retrospectively). Of these, 139 (15%) had a resistance mutation (of those listed in Methods) detected. During the follow-up, after the start of ART, a total of 933 (22%) patients had a resistance test performed; 808 (19%) had a test performed which successfully generated a result (ie a sequence) was obtained. Kaplan-Meier estimates of the percentage of patients with at least one resistance test result were 13% (95% CI 12% – 14%) by 2 years, 24% (95% CI 22%-26%) by 4 years and 32% (95% CI 29% – 35%) by 6 years. Of those 1057 patients with viral load failure (definition 1), 461 (44%) had a resistance test result available between 6 months before and 1 year after the date of viral load failure. 603 (57%) had at least one successfully performed resistance test at some time after the start of ART, which contrasts with only 205 (6%) of those who did not experience viral load failure. In a multiple logistic regression model the key factors associated with a greater chance of a resistance test being available around the time of viral load failure were calendar year of failure in or after 1999 (due to more recent widespread use of resistance testing) and lower viral load at time of failure (although by definition, this viral load was > 1000 copies/mL). The type of regimen showed no significant independent association with the probability of a test being available.

For 632 (15%) of the patients, at least one resistance mutation was detected after the start of ART. Fig. 1 shows the Kaplan-Meier estimates of the percentage of patients with at least one mutation detected according to the time from the start of ART; 9% (95% CI 8% – 10%) by 2 years, 19% (95% CI 17%-21%) by 4 years and 27% (95% CI 24% – 30%) by 6 years. When we restricted to patients who had been seen at the clinic for at least 6 months before start of ART the 6 year risk of a mutation being detected became 24%. We also looked at the risk of a resistance mutation being detected by two years after the start of ART, according to calendar year of starting ART. This was 5%, 7%, 11%, 9%, 10% and 9%, for patients starting ART in 1996, 1997, 1998, 1999, 2000 and 2001, respectively, suggesting no change over calendar time since 1998 when resistance testing became used widely.

A Cox model was fitted to examine factors associated with more rapid detection of at least one resistance mutation (Table 3). The main factors independently associated with a greater risk of resistance mutations were younger age, HIV exposure which was unknown or other than sexual / IDU, viral load above 100,000 copies/mL at start of ART, CD4 count missing or below 200 /mm³ at start of ART, previous AIDS diagnosis before ART and use of a triple nucleoside regimen with abacavir or a single PI regimen.

Table 3

We also assessed for each patient the cumulative risk of having detected mutations from at least two of the three main drug classes. Kaplan-Meier estimates for 2, 4 and 6 years were 6% (95% CI 5% – 7%), 14% (95% CI 12% – 16%) and 20% (95% CI 18% – 22%), respectively. The percent accruing resistance mutations from all three main drug classes were 1.0% (95% CI 0.7% – 1.3%) by 2 years, 2.7% (95% CI 2.0% – 3.4%) by 4 years and 4.1% (95% CI 3.0% -5.2%) by 6 years. When we restricted the analysis to those who had had at least one mutation detected, 21% had accrued resistance mutations for all three main drug classes by 4 years from having had the first mutation detected.

Table 4 shows the Kaplan-Meier estimates of the percentage of patients who have mutations associated with specific drug classes, or specific drugs, according to the specific drugs or drug classes included in the initial regimen. Overall, 24% of patients are estimated to have at least one nucleoside analogue-related mutation by 6 years; 13% with at least one TAM and 18% with M184V. For major PI mutations, the 6 year estimate is 10%, which becomes 15% when restricted to those who started ART with a single PI in the regimen. When restricted to those initiating a PI plus ritonavir, the estimate at 4 years is 7%, compared with 12% for those starting on single PI regimens. For NNRTI mutations, 16% overall have a mutation by 6 years; 21% when restricting to those patients starting ART with a NNRTI in the regimen.

Table 4

We wished to study further the differences observed between patients starting different types of regimen. To make such comparisons better reflect types of regimens used more commonly in clinical practice nowadays, we restricted this additional analysis to those starting ART in 1998 or after. Here the breakdown of regimens used was different from that overall (which was shown in Table 1); 2274 used an NNRTI regimen (67%; of which 48% nevirapine, 52% efavirenz), 693 used a single PI regimen (21%; of which 27% indinavir, 58% nelfinavir,7% saquinavir sgc, 8% ritonavir), 151 used abacavir (4%) and 263 used a PI plus ritonavir (8%; of which 32% indinavir, 24% saquinavir, 44% lopinavir). Table 5 shows adjusted relative hazards (RH) comparing the risk of certain mutations being detected, according to the initial regimen. The groups of mutations are as follows; (i) any nucleoside mutation, (ii) a PI mutation in those on a PI or a NNRTI mutation in those on an NNRTI (this outcome was not considered for the abacavir triple nucleoside group), (iii) mutations to at least two drug classes, and (iv) mutations to at least three drug classes.

Table 5

Considering first the comparisons between those on single PI and those on NNRTI regimens, (first row of Table 5) there was a 1.59-fold increased hazard of a nucleoside mutation for those on a PI regimen (p = 0.0008), but there was no significantly raised hazard of a mutation to the PI/NNRTI component, nor of mutations to two or three classes. When comparing those on regimens including a PI plus ritonavir with those on NNRTI regimens, there was a similar hazard of nucleoside mutations (RH 0.74 (95% CI; 0.45-1.24); p = 0.26), but a markedly lower hazard of mutations to the PI/NNRTI component (RH 0.31; 95% CI 0.15 –0.61; p = 0.0008), which remained similar when follow-up was censored if and when a person stopped the PI/NNRTI component (RH 0.34; 95% CI 0.14-0.79; p = 0.01). The relative hazard of mutations to two drug classes being detected was also lower in those on a PI plus ritonavir (0.43; 95% CI 0.22-0.83; p = 0.01) but the relative hazard of mutations to three classes being detected was close to one (0.84; 95% CI 0.24-2.97; p = 0.79). Lastly, comparing those starting ART with triple nucleoside regimens including abacavir and those starting with NNRTI regimens, there was a significantly raised hazard of nucleoside mutations being detected (RH 2.24; 95% CI 1.42-3.52; p = 0.0005). However, the relative hazards of detection of mutations to two and three classes were not significantly above one (1.55; 95% CI 0.85 – 2.66, and 1.4395% CI 0.32 – 6.29, respectively).

Comment

To our knowledge, this study provides the first estimates of the long term risk of acquiring resistance mutations in patients who started ART with three or four drugs in routine clinical practice. Resistance mutations were relatively common, with 27%, 20% and 4% estimated to have at least one mutation to one, two and three drug classes, respectively, by 6 years from start of ART. These are likely to be under-estimates the true percentage of patients who have acquired resistance mutations both because resistance testing was not performed on all patients at the time of viral load failure and the fact that resistance tests are not sufficiently sensitive to discern the presence of minority variants with resistance. The incidence of viral load failure over this time was dependent on the definition adopted (Table 2), but the lowest estimate obtained for the percentage of patients experiencing failure by 6 years was still relatively high, at 38%. The rate of viral load failure was approximately constant after 2 years from the start of HAART. We did not detect a significant decline in viral load failure rate, in contrast to a previous observation ^[9].

We observed high rates of switching to new drugs during follow-up. This is likely to reflect the relatively poor tolerability and acceptability of some drugs and regimens, especially those used early in the HAART era, as well as the relatively high viral load failure rate. It was interesting to observe that resistance tests were more likely to be performed in those that experienced viral load failure with lower viral load levels, albeit by definition still above 1000 copies/mL. This could reflect the fact that in cases where viral rebound is to a very high level the clinician will either suspect or know that the patient has interrupted ART and will therefore be less inclined to look for resistance as a possible cause underlying the viral rebound.

We found a greater likelihood of detection of resistance mutations in younger than older patients. This is consistent with the greater rate of viral load failure seen in younger individuals and presumably this relates to a tendency for lower levels of adherence in younger people. We also found a greater chance of resistance being identified in those with baseline viral load above 100,000 copies/mL (compared with lower viral load), those with an AIDS diagnosis and those starting with a single PI or abacavir triple nucleoside regimen were more likely to acquire resistance mutations, compared with those on NNRTI or PI's with ritonavir. A CD4 cell count of below 200/mm³ was also associated with a raised risk, as was having an unknown exposure group. The explanation for this latter finding is not clear.

We further compared the detection of different groups of resistance mutations among patients starting different regimens. For this analysis we restricted to those starting ART in 1998 or later, as regimens used before this date are not generally now used and we also wished for the follow-up time on each of the regimens not to differ too greatly. Risk of PI mutations being detected in people using regimens containing a PI with ritonavir was markedly lower than the risk of NNRTI mutations being detected in those using NNRTI regimens (relative hazard 0.31 (0.15-0.61); p = 0.0008). This did not appear to be explained by a tendency for those starting PI's with ritonavir being more likely to stop the PI, because the result remained similar when we censored follow-up on those stopping the PI/NNRTI component. The relative hazard was also similar when follow-up was restricted to 2 years only (data not shown). While there has been evidence suggesting that resistance mutations have a relatively low probability of emerging in patients who use regimens including ritonavir boosted PI's ^[10,11], we are not aware of any study which has compared such regimens with NNRTI containing regimens on substantive numbers of patients. However, this is not a randomized comparison so, despite the fact that we adjusted for potential confounding variables, there could be a residual bias relating to choice of who is treated with which type of regimen. We also found that those using abacavir triple nucleoside regimens had a highly significantly increased risk of acquiring nucleoside mutations, compared with those on NNRTI regimens, consistent with trial results ^[12]. Interestingly, however, there was not statistically significant evidence that those starting therapy with abacavir triple nucleoside regimens acquire mutations to two or three drug classes (as a result of resistance development on regimens used subsequent to the triple nucleoside regimen) any more rapidly than those who started on NNRTI regimens, although there was a trend towards such an effect.

Two recent trials have produced short-term estimates of the probability of resistance being detected within the first 1-3 years from the start of ART ^[4,5] – with findings generally consistent with our own: 8%-10% with resistance by 2 years from start of ART. In addition, Harrigan et al performed resistance testing on all samples with viral load > 1000 copies/mL in the first 2.5 years after the start of HAART in the HOMER Cohort and found 30% with a resistance mutation by this time ^[13]. This emphasises that our estimates may be under-estimates due to the lower frequency of resistance testing compared with the HOMER cohort. We not aware of any comparable estimates over the longer follow-up time covered by our study, either from trials or routine clinic cohorts.

In summary, we have found that patients treated in a routine clinical setting experience relatively high levels of antiretroviral drug resistance over the first 6 years of therapy, despite having received potent drug regimens from the outset. Thus, while the prospects for maintaining viral suppression in people starting ART remain good over the first 5-10 years, the longer term prospects for continued viral suppression may increasingly depend on development of new antiretroviral drugs.

Acknowledgement

Funding Support has been received the Medical Research Council (Grant No. 0000199) and the Department of Health in the UK.

References

Appendix

Analysis and Writing Committee

Andrew N Phillips, PhD¹, David Dunn, PhD², Caroline Sabin, PhD¹, Anton Pozniak³, Ryanne Matthias², Anna Maria Geretti⁴, John Clarke ⁵, Duncan Churchill⁶, Ian Williams^1,7, Teresa Hill¹, Hannah Green, MSc², Kholoud Porter, PhD², George Scullard⁵, Margaret Johnson, MD⁴, Philippa Easterbrook ⁸, Richard Gilson^1,7, Martin Fisher⁶, Clive Loveday PhD⁹, Brian Gazzard³, Deenan Pillay, FRCPath^1,10

1 Departments of Primary Care and Population Sciences and Virology, Royal Free and University College Medical School, London; 2 Medical Research Council Clinical Trials Unit, London; 3 Chelsea and Westminster Hospital; 4 Royal Free Hospital NHS Trust; 5 St Mary's Hospital and Imperial College School of Medicine; 6 Brighton and Sussex University Hospitals; 7 Mortimer Market Centre, Camden Primary Care NHS Trust; 8 King's College Hospital, London; 9 ICVC, Bucks, UK; 10 Specialist and Reference Microbiology Division, Health Protection Agency.

UK CHIC Study Group

Steering Committee

Abdel Babiker, David Dunn, Philippa Easterbrook, Martin Fisher, Richard Gilson, Margaret Johnson, Janet Mortimer, Barry Peters, Andrew Phillips, Kholoud Porter, Caroline Sabin, George Scullard, Brian Gazzard (Chair), Ryanne Matthias, Teresa Hill.

Data management group

Loveleen Bansi, David Dunn, Philippa Easterbrook, Richard Gilson, Teresa Hill, Ryanne Matthias, Janet Mortimer, Andrew Phillips, Kholoud Porter, Caroline Sabin.

Central co-ordination

Medical Research Council Clinical Trials Unit (MRC CTU), London (Abdel Babiker, David Dunn, Ryanne Matthias, Kholoud Porter).

HIV Epidemiology Group, Department of Primary Care and Population Sciences, Royal Free and UC Medical School, London (Loveleen Bansi, Teresa Hill, Andrew Phillips, Caroline Sabin, Colette Smith)

Participating clinical centres

King's College Hospital, London (Philippa Easterbrook, Anele Waters, Dorian Crates, Natasha Morgan).

Brighton and Sussex University Hospitals NHS Trust (Martin Fisher, Nicky Perry, Anthony Pullin, Duncan Churchill, Wendy Harris).

Chelsea and Westminster NHS Trust, London (Brian Gazzard, Steve Bulbeck, Sundhiya Mandalia).

Mortimer Market Centre, Camden Primary Care NHS Trust and Royal Free and University College Medical School (RFUCMS) (Richard Gilson, Julie Dodds, Nina Fudge, Andy Rider, Ian Williams).

Health Protection Agency -Communicable Disease Surveillance Centre (HPA-CDSC), London (Janet Mortimer, Katy Sinka, Barry Evans, Valerie Delpech).

St. Thomas’ Hospital, London (Barry Peters, Nick Larbalestier, Kimberly Gray)

Royal Free NHS Trust, London (Margaret Johnson, Mike Youle, Fiona Lampe, Colette Smith, Helen Gumley, Clinton Chaloner, Jani Puradiredja)

St. Mary's Hospital, London (George Scullard, Jonathan Weber, John Clarke, Christine Owens)

UK Collaborative Group on HIV Drug Resistance Steering Committee

Jodi Blackham, Health Protection Agency, London; Sheila Burns, City Hospital, Edinburgh; Sheila Cameron, Gartnavel General Hospital, Glasgow; Pat Cane, University of Birmingham;

Ian Chrystie, St. Thomas’ Hospital, London; Duncan Churchill, Brighton and Sussex University Hospitals NHS Trust; John Clarke, St. Marys Hospital, London; Philippa Easterbrook, Mark Zuckerman, King's College Hospital, London; David Goldberg, Scottish Centre For Infection and Environmental Health; Mark Gompels, Southmead Hospital, Bristol; Tony Hale, PHLS, Leeds; Linda Lazarus, Deparment of Health, London; Andrew Leigh-Brown, University of Edinburgh; Clive Loveday, ICVC, High Wycombe; Anna Maria Geretti, Royal Free NHS Trust, London; Deenan Pillay, Andrew Phillips, Caroline Sabin, Teresa Hill, Paul Kellam, Royal Free and University College Medical School, London; David Dunn, Kholoud Porter, Hannah Green, Ryanne Matthias, MRC Clinical Trials Unit, London; Anton Pozniak, Chelsea & Westminster Hospital, London; Peter Tilston, Manchester Royal Infirmary; Ian Williams, Mortimer Market Centre.