Themes for May 2011 Grant Application

Themes for May 2011 grant application

12 April 2011

Prognosis of HIV-infected patients treated with ART: comparative analyses and causal modelling of treatment strategies

Projects relating to prognosis for clinical events (deaths from all causes and specific causes,AIDS and specific AIDS-defining conditions) in the era of long term antiretroviral therapy:

We will develop “on treatment” models for prognosis up to five years in the future, based on current and previous values of prognostic factors (including both conventional HIV risk factors such as CD4 count, HIV-1 RNA, and prior AIDS defining illnesses, other factors such as haemoglobin, platelets, aspartate and alanine transaminases, and creatinine that have been shown to be independently prognostic for all-cause mortality, and information on co-infections such as Hepatitis B and C).

In an era when patients are living with HIV-infection for several decades, they experience morbidity and mortality from non-AIDS conditions that are associated with systemic inflammation and immune system activation, which may result from exposure to ongoing, uncontrolled, intermittently controlled or low-level viral replication. We will therefore focus on the role of HIV virus burden in predicting outcomes and in modifying effects of other factors, with a particular focus on prognosis for cause-specific mortality. We will examine prognosis separately in patients who can maintain complete or near-complete virologic suppression, and those with periods of virologic failure. Cumulative virus burden (copy-years viremia) from seroconversion has been shown to predict AIDS and death in untreated patients, but the extent to which this measure is prognostic for AIDS and death in treated patientsremains unclear. We will investigate the prognostic value of copy-years viremia in conjunction with the viral load measurement immediately prior to starting ART, which is correlated with cumulative virus from infection to start of treatment. We will also examine the implications of isolated short-term virologic failure, because this is the outcome in trials of regimen simplification such as that comparing maintenance treatment with a single PI after viral suppression with continuation on the standard triple drug regimen.

We will examine the extent and causes of heterogeneity in prognosis between cohorts, regions and settings. Estimates of prognosis will be corrected for the extent of under-ascertainment of death in individual cohorts. We will examine heterogeneity in effects of prognostic factors: for example we have already demonstrated striking differences in the prognostic value of ethnic group, comparing Europe and North America, and will now examine sex differences in response to ART and cause-specific mortality across regions.

We will examine whether inclusion of new prognostic markers improves overall predictions and risk stratification.

Recent work has demonstrated the external validity of a prognostic index that includes both conventional HIV and non-HIV risk factors (the VACS Index) and that this index performs better in predicting mortality from 12 months after start of ART than an index restricted to conventional HIV risk factors. In collaboration with VACS investigators, we will further develop the index for use in clinical practice and as a tool for monitoring response to treatment, and will investigate whether the inclusion of new non-HIV prognostic factors further improves its performance. We will examine the geographic transportability and generalisability of the VACS index by examining its performance in new patient groups of particular interest, for example immigrants from sub-Saharan Africa participating in European cohorts.

We will examine prognosis for clinical events and death according to HIV-1 subtype, and investigate whether virologic response to initial ART regimen explains all or part of any association. Globally, the majority of HIV infections are with non-B subtype, but most research has been carried out on those with B subtype in high-income countries. Previous studies have grouped non-B subtypes together when comparing with B subtype because of lack of sufficient numbers of patients and clinical events in individual cohorts, but this may have obscured differences between diverse subtypes. A large collaboration of Canadian and European cohorts is required to provide sufficient data to investigate differences in disease progression and rates of viral failure by sub-type.

Both % CD4 cells among lymphocytes and CD4/CD8 ratio have been shown to be associated with progression to AIDS and death. We will examine whether inclusion of these variables improves prediction and risk stratification, and whether these associations vary with cause of death (for example, CD4/CD8 ratio may be associated with death from cardiovascular disease).

Patients, physicians and policy makers are particularly interested in prognosis expressed as life expectancy (LE), defined as the average number of additional years a person will live from a particular age. Very large datasets are required in order to accurately estimate rates in each age/sex group and to estimate LE for particular patient groups, particularly because estimates of life expectancy are driven by mortality rates in older age groups, in which few patients were treated until recently. Based on our updated prognostic models, we will estimate LE for treated HIV-infected patients. We will update previous estimates taking into account under-ascertainment of death and calculate LE for different patient groups, for example by CD4 count at initiation of ART, risk transmission group.We will examine trends in LE over time, and LE in successfully treated patients (for example those who maintain virologic suppression from 6 months after treatment initiation. We will quantify the effect of AIDS deaths in reducing LE, by calculating cause-deleted life expectancy.
Collaborative work on prognosis in low income settings.Networks of HIV clinic cohorts in different African regions and other low and middle income settings are now well established, with the framework of IeDEA. We will collaborate with these networks within topic areas where we can provide methodological support (including provision of training) and where analyses comparing prognosis between regions and settings are of interest. Such analyses will be conducted primarily by investigators from the IeDEA networks, with support from the Bristol team and other applicants with relevant expertise.

We will compare prognosis for all-cause mortality between South Africa, Europe and North America, allowing for incompleteness of ascertainment of death (subject to further discussion – not confirmed). High mortality rates among patients lost to programme are a well-known feature of data from low income settings, and IeDEA Southern Africa have developed a nomogram to adjust for such under-ascertainment of death. We have surveyed cohorts in Europe and North America in order to estimate completeness of death ascertainment, and have shown that accounting for under-ascertainment reduces between-cohort heterogeneity in mortality rates. We will compare mortality during the first 2 years after initiation of ART between these settings, based on methods that adjust for under-ascertainment of death.

Prognostic model for HIV-2 infection (subject to further discussion – not confirmed). We will collaborate with IeDEA West Africa to develop models for prognosis among individuals infected with HIV-2 and compare estimates with those from prognostic models for treated individuals infected with HIV-1 infection. Such work could potentially include both West African cohorts and European cohorts that include HIV-2 infected individuals.

Projects relating to treatment strategies, with emphasis on causal modelling

We will investigate strategies for regimen modification (switching) in patients with virologic failure on first ART regimen. We will compare the relative effects of different treatment strategies in patients who experience virological failure while on their first ART regimen. We will use dynamic marginal structural models (Cain et al. International Journal of Biostatistics 2010) to mimic a randomized controlled trial in which such patients are randomized to switch regimen within 3 months of confirmed virologic failure with HIV-RNA > 200 copies/ml) (control group) with a strategy of switching within 3 months of confirmed failure with HIV-RNA > 10,000 copies/ml. We will also consider thresholds that are intermediate between 200 and 10,000 copies/ml.We will also provide methodological support for parallel analyses on switching strategies in low income settings, based on data from South African sites with viral load monitoring.
We will establish a collaboration of ART-CC cohorts that collect adherence data, in order to explore the relationship between adherence to ART and clinical outcomes. We will explore standardization of adherence definitions across cohorts using the same method of data collection. We examine prognosis in patients grouped according to level of adherence. Analyses will first be done separately according to type of adherence information (prescription refill or self-report), and then based on combinations of different types of adherence data. We will use marginal structural modelsto estimate the causal impact of non-adherence to ART on treatment failure, progression to AIDS and death, and will evaluate the role of dosing frequency and class of ART in mediating or modifying the effect of adherence.
We will use dynamic marginal structural models in order to when to start ART in ART-naïve patients who present with specific AIDS-defining opportunistic infections such as Cryptococcus or Toxoplasmosis (collaboration with J Miro, H Furrer and the COHERE OI group). Such analyses will be restricted to cohorts who collect data on ART-naïve patients who are hospitalized with an opportunistic infection at presentation, even if they die during the admission. Very large collaborative datasets are required for such analyses, because relevant outcomes occur within the first year of ART.

Projects that may be of lower priority, or better addressed by other groups, or may not be achievable within the resources requested for the application

Modelling work combining information on progression rates and regimen durability to examine adverse-benefit ratio curves for common currently used initial ART regimens, in the era when off-patent generic medications are becoming available.
Further work on social and behavioural determinants of response to and prognosis on ART, in particular alcohol, drug use, SES and education. Would need to clarify which cohorts have good information and are interested in collaboration, eg North American cohorts that participate in CNICS. Examine whether the effects of these adverse factors vary between cohorts and across regions. (Would be interesting to get data on alcohol from South Africa – is this possible?)
Work on treatment strategies for HCV coinfected individuals. The COHERE HCV working group has nearly finished a paper using marginal structural models to estimate the effect of HCV treatment, based on cohorts with appropriate data on HIV-HCV coinfected patients. We have had initial discussion with Giotta Touloumi about analyses examining timing of treatment (“when to start”) in such patients. We could also conduct analyses of when to start ART in such patients, if sufficiently large numbers were available.
Extend existing models to estimate prognosis up to 10 years after starting ART. Prognostic models are a trailing indicator of what a patient starting ART today would experience. Results from models predicting long-term survival are likely to reflect the survival of those starting ART on older drugs that would not be prescribed to a new patient starting ART today. Since the highest risk of death, treatment toxicities and switching regimens occurs in the first year of ART, a method of adjusting for the lower mortality in patients that start more recently is to model survival in the first year of ART for each calendar year and assume that subsequent survival from the second year onwards is dependent more duration of ART and age rather than calendar time of starting ART. A related subject is to investigate competing risks of cause –specific mortality by calendar year of starting ART and/or duration of ART as well as associations with age and other risk factors.

Update on discussions with partners for 2011 application

COHERE/EuroCoord: broad agreement in place. Data management costs are for 1 month per year from each data centre. Issues include getting agreement in advance of the EC on 27 May, what to say about “opt in” for cohorts not traditionally part of ART-CC, and arrangements with the COHERE prognosis working group.

NA-ACCORD – MM and JS spoke to Richard Moore, Stephen Gange and Mike Saag on 4 April. The conversation was generally constructive and the conclusion was that we should approach NA-ACCORD with specific projects, not go for an overarching collaboration agreement.

During the Prague meeting we discussed comparative analyses with Matthias Egger, Andrew Boulle and Francois Dabis. We reached a tentative agreement that we would work in collaboration with IeDEA South and West Africa, by offering methodological support for comparative analyses, including “when to switch” for IeDEA-SA and perhaps a prognostic model for HIV-2, as well as offering training in South Africa.

We have agreed to include Miguel Hernán as a coapplicant with funded time (2 or 4 hours per week). We plan to coordinate all causal analyses with Miguel and his colleagues. We are also discussing a separate application to the MRC/NIHR methodology panel.

We had a useful discussion with David Haerry about better engagement with patient groups. We are updating the web site to include a “Patients” section with lay summaries of recent papers.

Summary of feedback from 2010 application

This application was led by an excellent team with world leading profile in the statistical modelling for longitudinal epidemiological cohorts. The cohort has been highly productive and established a strong profile.

Board members were disappointed that the applicants had not been more visionary and innovative. In particular the investigators had not made a strong case for the uniqueness of the ARTCC in an increasingly crowded field. The Board would have liked greater detail on:

What were the key aspects that ART–CC could tackle that others could not?
Were there competing or complementary approaches to the work of others?
What data (combinations) did ART-CC hold that others did not – is it more complete or do they include populations the others do not - the application lacked information for example on gender distribution; injecting drug users; geographic representation; how the contributing cohorts were selected, how many patients across cohorts etc.
The proposal was short on demographic and clinical detail eg the analysis of viral subtype as a potential prognostic marker. How many sites have information on subtype and subtype distribution, methodology to derive subtype, was this standardized across sites?

There might have been better engagement with other cohorts, including those in resource poor settings. Many of the anticipated outputs would not be not be applicable to resource limited countries - this was a potential missed opportunity.

IIB was aware that there was a lot of horizon scanning going on internationally in the HIV field, with The Futures Institute and UN agencies doing modeling work and the recent AIDS 2030 report. Members would have liked greater demonstration of early engagement with the likely users of the work.