| |
Short-term clinical disease progression in HIV-1-positive patients taking combination antiretroviral therapy: the EuroSIDA risk-score
|
| |
| |
AIDS:Volume 21(14)September 2007p 1867-1875
Mocroft, Amandaa; Ledergerber, Brunob; Zilmer, Kaie; Kirk, Olef; Hirschel, Bernardc; Viard, Jean-Paulg; Reiss, Peterh; Francioli, Patrickd; Lazzarin, Adrianoi; Machala, Ladislavj; Phillips, Andrew Na; Lundgren, Jens Df; for the EuroSIDA study group and the Swiss HIV Cohort Study
From the aDepartment of Primary Care and Population Sciences, Royal Free and University College Medical School, London
bDivision of Infectious Diseases and Hospital Epidemiology, University Hospital, Zurich
cDivision of Infectious Diseases, Geneva University Hospital, Geneva
dService of Infectious Diseases, University Hospital, Lausanne, Switzerland
eWest-Tallinn Central Hospital, Tallinn, Estonia
fCopenhagen HIV Program, University of Copenhagen, Panum Institute, Denmark
gHopital Necker-Enfants Malades, Paris, France
hAcademisch Medisch Centrum bij de Universiteit van Amsterdam, Amsterdam, the Netherlands
iOspedale San Raffaele, Milan, Italy
jFaculty Hospital Bulovka, Prague, Czech Republic.
Abstract
Objectives: To derive and validate a clinically applicable prognostic score for predicting short-term disease progression in HIV-infected patients taking combination antiretroviral therapy (cART).
Design and methods: Poisson regression was used to identify prognostic markers for new AIDS/death in patients taking cART. A score was derived for 4169 patients from EuroSIDA and validated on 5150 patients from the Swiss HIV Cohort Study (SHCS).
Results: In EuroSIDA, 658 events occurred during 22 321 person-years of follow-up: an incidence rate of 3.0/100 person-years of follow-up [95% confidence interval (CI), 2.7-3.3]. Current levels of viral load, CD4 cell count, CD4 cell slope, anaemia, and body mass index all independently predicted new AIDS/death, as did age, exposure group, a prior AIDS diagnosis, prior antiretroviral treatment and stopping all antiretroviral drugs. The EuroSIDA risk-score was divided into four strata; a patient in the lowest strata would have predicted chance of new AIDS/death of 1 in 801, 1 in 401 and 1 in 201 within the next 3, 6 or 12 months, respectively. The corresponding figures for the highest strata were 1 in 17, 1 in 9 and 1 in 5, respectively. A single-unit increase in the risk-score was associated with a 2.70 times higher incidence of clinical progression (95% CI, 2.56-2.84) in EuroSIDA and 2.88 (95% CI, 2.75-3.02) in SHCS.
Conclusions: A clinically relevant prognostic score was derived in EuroSIDA and validated within the SHCS, with good agreement. The EuroSIDA risk-score will be made available publicly via an interface that will perform all calculations for the individual.
Introduction
Following the introduction of combination antiretroviral therapy (cART) in 1996 and the decline in the incidence of new AIDS-defining illnesses and deaths among HIV-infected patients [1-3], attention focused on the immunological or virological response to cART regimens, particularly in the clinical trials setting [4,5], or on the long-term risk of clinical disease [6-8]. A score for predicting clinical progression that utilized the most recent values of CD4 cell count, viral load and haemoglobin was published in 2002 [9] and successfully validated on two patient groups. With longer follow-up, and changes in response to cART over time [10], the predictors of clinical progression may have changed. In addition, in the pre-cART era, the rate of change in CD4 cell count (CD4 cell slope) was demonstrated to have prognostic power over and above the information provided by the most recent CD4 cell count [11]. In that study, patients whose CD4 cell count was decreasing more rapidly had a higher risk of progression, even after adjustment for the current level of immunodeficiency. This relationship, or the extent to which it contributes to differences in clinical progression, has not been evaluated in patients exposed to cART.
Other groups have also developed prognostic scores [6], but these have not utilized all routinely measured laboratory markers, and generally they have concentrated on predicting long-term clinical progression based on information known at the date of starting cART. Such scores identify groups of patients with elevated risks of disease progression at the time of starting cART. Markers of disease progression change, however, after a patient starts cART. If a patient returns to the clinic at 12 months after starting cART, it is more clinically relevant to use the information at that time, after the first year of treatment, to predict the risk of new AIDS or death occurring before the next scheduled patient visit, say in 3 months. If this risk is sufficiently low, both the clinician and patient may agree that it would be more appropriate to return in 6, rather than 3 months. Therefore, the objectives of this study were to develop a EuroSIDA risk-score to predict short-term clinical progression; to design this score so that it could be implemented into clinical practice by using online calculation tools, and to validate it on an independent cohort of patients.
Discussion
EuroSIDA and the SHCS are two of the largest European observational studies of HIV-infected patients, with over 20 000 patients recruited to these studies combined. In patients on cART in EuroSIDA, current values of CD4 cell count, viral load, level of anaemia, BMI and rate of CD4 cell change were highly prognostic for clinical progression. A risk-score was derived that was validated on HIV-infected patients from the SHCS, with good agreement, both in terms of discrimination and validation. A patient with a EuroSIDA risk-score of < 1.5 had a 1 in 801 chance of disease progression within the next 3 months, compared with a chance of 1 in 17 for a patient whose EuroSIDA risk-score was > 4.5.
Previous studies have demonstrated the prognostic role of CD4 cell count, viral load, haemoglobin, and weight or BMI in patients with HIV receiving cART [9,17-19]. Haemoglobin and BMI may be markers of other disease processes not captured by the factors known to be related to HIV disease progression. Some of the patients in EuroSIDA were excluded because of missing data for BMI and haemoglobin. These data may also be missing in other clinic settings, and further work will identify the prognostic power of the risk-score in the absence of these markers. Consistent with previous published research from EuroSIDA, we found no increased risk of new AIDS/death associated with coinfection with either hepatitis B or C [20,21]. There was an increased risk of new AIDS/death among patients reporting infection via IDU, as previously reported, and this may be partly attributable to the higher risk of non-HIV-related deaths observed in this patient group [22].
This study is the first, to our knowledge, that has also demonstrated that the rate of change of CD4 cell count was important for calculating the short-term risk of clinical disease progression. Patients with a current CD4 cell slope that was decreasing by > 25 cells/μl per 3-month period had a significantly increased risk of disease progression, over and above that measured by the other prognostic variables in the EuroSIDA risk-score. Work from the pre-cART era demonstrated that the CD4 cell decline was a determinant of progression to AIDS independent of the most recent CD4 cell count [10], and that the rate of CD4 cell decline was independently associated with a poorer survival in patients with low CD4 cell counts [23]. Patients with the most rapidly increasing CD4 cell counts had an increased risk of new AIDS/death, although this was not statistically significant. Additional work is needed to describe these patients; one possible explanation is that some of the effect is from the immune reconstitution syndrome [24].
The ART-Cohort Collaboration has also derived a prognostic model for new AIDS/death [6]. For example, a 35-year-old ART-naive patient, thought to have been infected via heterosexual sex, starting cART without prior AIDS, no history of intravenous drug use and with a viral load of 150 000 copies/ml and a CD4 cell count of 220 cells/μl would have a 6.1% probability of progression to AIDS or death at 3 years after starting cART, using the ART-Cohort Collaboration score. However, it is well known that patients experience different immunological and virological responses after starting cART [25-27], and that these responses affect the risk of clinical progression [28-30]. If the patient described above responded well to cART, and at 3 years after starting has a viral load < 50 copies/ml, a CD4 cell count of 400 cells/μl, a CD4 cell slope which was increasing by 10 cells/μl per 3-month period, no anaemia, a BMI of 23 and currently taking treatment, they would have a EuroSIDA risk-score of 0.42, and a 1 in 801 (or 0.12% probability) chance of new AIDS/death within the next 3 months. The EuroSIDA risk-score will be publicly available at www.cphiv.dk , and it will be possible for clinicians or patients to input laboratory values for the risk-score to be calculated online. Initially this will be possible on an individual basis, but it will be developed further to allow the calculation of the score for many patients, by transferring the data in the HICDEP format [31].
The discrimination of any prognostic model tends to decrease when the model is applied to another dataset [32]. However, the EuroSIDA risk-score was validated on patients from the SHCS who started cART. The EuroSIDA risk-score was highly prognostic in the SHCS, with a 1 unit increase in the EuroSIDA risk-score giving an almost identical relative incidence rate of disease progression as that in the EuroSIDA patients on whom the score was derived. Validation of the EuroSIDA risk-score in other large groups of patients, particularly non-European, would be beneficial.
The EuroSIDA risk-score could be used in several ways. Patients recruited to clinical endpoint trials could be stratified at randomization by the EuroSIDA risk-score to ensure patients with comparable risks of clinical disease progression are well balanced between treatment groups. It could also be used as a surrogate marker of immediate clinical benefit in the comparison of various antiretroviral drugs in trials designed to assess differences in laboratory rather than clinical endpoints, or it could be used after initiation of the trial, for example, to mimic the design of a trial within the setting of an observational study. This allows the trial assumptions and expected duration of the trial to be assessed [33]. It could also be used by the individual and their clinician in future planning, for example to assess if the risk of clinical disease progression was sufficiently low that the patient could be seen in 6 months rather than 3 (while recognizing that other factors, such as adverse events, also need to be closely monitored).
There are several limitations to this study that should be considered. We are not able to develop the score separately for ART-naive and ART-experienced patients. We have included an adjustment to estimate the increased risk among patients who stop all ART. This group of patients is likely to be a mix of patients having a structured treatment interruption, who are known to have an increased risk of clinical progression [11,34-36], and patients who stop all ART prior to death. However, censoring EuroSIDA patients at the date of stopping ART led to a remarkably similar score to predict clinical progression. We could have derived the EuroSIDA risk-score including all continuous variables without categorization, which would provide a better statistical solution. Work is ongoing in this area, and also in the predictive power of the score if a key component of the score, such as haemoglobin, is missing. We have not developed a score separately for AIDS or death, and it is possible there may be small differences in the prognostic factors for each. Although developing a separate score has its merits, it also has disadvantages, making the score more difficult to implement and interpret in the routine clinical setting. Further, randomized clinical trials, one of the settings where the EuroSIDA risk-score could be implemented, continue to use a combined clinical endpoint [33,34].
To conclude, we have derived and validated a highly prognostic risk-score for short-term clinical progression in HIV-infected patients. The EuroSIDA risk-score will be made publicly available via an interface that will perform all calculations for the individual. The risk-score is highly clinically relevant for patient management, or it could be used as a surrogate endpoint in clinical trial design.
Results
Of 14 274 patients within EuroSIDA: 9049 started cART; 5402 had a CD4 cell count and viral load measured in the 6 months prior to starting cART; 5302 had the potential to calculate the CD4 cell slope prior to starting cART; and 4169 had haemoglobin and BMI measured during follow-up. Patients who had started cART but were excluded from analyses because they did not have a CD4 cell count or viral load measured before starting cART tended to have started cART some time before recruitment to EuroSIDA and, therefore, information on laboratory measurements was not available. The remainder of the patients who were excluded tended to have more advanced disease, as indicated by lower CD4 cell counts at starting cART, a higher proportion with AIDS, and starting cART during 1996 and 1997. The characteristics of the 4169 included patients are described in Table 1.
During a median follow-up of 6.7 years [interquartile range (IQR), 3.6-8.7], or 22 321 person-years there were 658 events; 388 of the events were new AIDS-defining illnesses (59.0%) and 270 patients (41.0%) died. The incidence rate was 3.0/100 person-years of follow-up [95% confidence interval (CI), 2.7-3.3]. Figure 1 illustrates the distribution of events and person-years of follow-up throughout this study, stratified by current values of CD4 cell count, CD4 cell slope, viral load, anaemia and BMI. For example, among patients with a current CD4 cell count of ≦ 50 cells/μl, there was 436 person-years of follow-up (2.0% of the total person-years of follow-up), and 133 events (20.2% of the total events). Further, 91.2% of the follow-up time was spent on treatment, and 66.9% of the events occurred among patients taking ART. Table 2 shows the significant univariate and multivariate incidence rate ratios of new AIDS/death. The EuroSIDA risk-score was derived using the logarithm of the incidence rate ratios shown in Table 2, to give scores given in Table 3. For example, a 30-year-old patient infected via intravenous drug use (IDU), who started cART before a diagnosis of AIDS and was ART experienced attending clinic 3 years after starting cART with a current CD4 cell count of 400 cells/μl, viral load 50 copies/ml, mild anaemia, BMI of 22, a CD4 cell slope that had increased by 15 cells/μl over the past 3 months, and currently taking cART would have a risk-score of 1.74. This is derived from: 0 (CD4 cell component) + 0 (viral load component) + 0.68 (mild anaemia) + 0 (BMI component) + (0.027 X 30) (age) + 0 (CD4 cell slope component) + 0 (ART experienced) + 0 (on ART) + 0.25 (infected via IDU) + 0 (no prior AIDS).
Immediately prior to clinical progression, the median EuroSIDA risk-score was 3.44 (IQR, 2.48-4.39), measured a median time of 2 months (IQR, 1-4) before new AIDS/death. A single unit increase in the current EuroSIDA risk-score was associated with a 2.70 times higher incidence of clinical progression (95% CI, 2.56-2.84; P < 0.0001). The EuroSIDA risk-scores were grouped into four strata; < 1.5, 1.5-2.99, 3-4.49 and ≥ 4.5. Table 4 illustrates the chance of clinical progression within the next 3, 6 or 12 months after a clinic visit, within these four strata. For example, regardless of the time since starting cART, a patient with a score of 1.74, as in the above example, would have a 1 in 276 chance of new AIDS/death within 3 months (95% CI, 1 in 240 to 1 in 326), while a patient with a score of 4.8 would have a 1 in 17 chance (95% CI, 1 in 15 to 1 in 19).
Validation in the Swiss HIV Cohort Study
In the SHCS, 5105 patients satisfied the inclusion criteria and were included as the validation cohort. The patients were generally quite similar to those from the EuroSIDA study: 3585 were male (69.6%), 1753 belonged to the homosexual exposure group (34.0%) and 1260 were IDU (24.5%). Over half, 3126 (60.7%) were ART naive at starting cART. The median age at baseline was 37 years (IQR, 32-43); CD4 cell count was 209 cells/μl (IQR, 98-347) and viral load was 4.68 log10 copies/ml (IQR, 3.85-5.23). During 27 987 person-years of follow-up, 897 patients progressed to a new AIDS-defining illness or death (17.4%), of which 494 were new AIDS-defining events (55.1%), giving a clinical progression rate of 3.2/100 person-years of follow-up (95% CI, 3.0-3.4). The median EuroSIDA risk-score at starting cART was 2.46 (IQR, 1.72-3.33). Figure 2 illustrates the incidence rates of new AIDS/death after stratification by current EuroSIDA risk-score in both EuroSIDA and the SHCS. The event rates within the strata were consistent, with rates per 100 person-years of follow-up in the four strata of 0.50, 1.45, 6.37 and 23.59 in EuroSIDA and 0.33, 1.69, 8.20 and 42.28 in the SHCS. Further, in the SHCS patients, a 1 unit increase in the current EuroSIDA risk-score was associated with a 2.88 times higher incidence rate of clinical progression (95% CI, 2.75-3.02; P < 0.0001).
Patients and methods
EuroSIDA
EuroSIDA is a prospective, European clinic-based cohort study of 14 262 patients with HIV-1 infection in 94 centres from 31 countries across Europe, plus Israel and Argentina. Details of the study have been published [12]. At recruitment, in addition to demographic and clinical information, a complete antiretroviral therapy (ART) history was collected, together with the eight most recent CD4 cell count and viral load measurements. Data were extracted at 6-monthly intervals from patient clinical charts on to follow-up forms. This analysis includes data to May 2006 and includes all CD4 cell counts, viral load, haemoglobin, weight, the date of starting and stopping each antiretroviral drug, the use of drugs for prophylaxis against opportunistic infections and all AIDS-defining illnesses using the clinical criteria from 1993 [13]. CD4 cell count and viral load were measured, on average, at 3-monthly intervals, and weight and haemoglobin at 6-monthly intervals. Anaemia was based on measurements of haemoglobin and defined as previously described within the EuroSIDA study [14]: severe, ≦ 80 g/l; mild, 80.1-1200 g/l (females) and 80.1-140 g/l (males); none, > 120 g/l (females) and > 140 g/l (males).
Members of the coordinating office visited all centres to ensure correct patient selection and that accurate data were provided by checking the information provided against case notes for all reported clinical events and a random sample of 10% of all other patients.
Swiss HIV Cohort Study
The Swiss HIV Cohort Study (SHCS) is a prospective population based cohort study (www.shcs.ch ; [14]). Any HIV-infected person aged > 16 years is eligible to participate. Data collection and study procedures are standardized. Data quality and protocol monitoring are conducted at the coordination and data centre in Lausanne. Similar data items to those in EuroSIDA are collected at recruitment and at 6-monthly intervals. Follow-up within the SHCS for this analysis is to February 2007.
Statistical methods
Patients initiating cART using one of the recommended first-line cART regimens between 1997 and 2005 [15] were included if they had a CD4 cell count and viral load measured in the 6 months prior to starting cART. A recommended regimen comprised two nucleoside analogues, plus one of the following: a single protease inhibitor, a ritonavir-boosted PI, a non-nucleoside reverse transcriptase inhibitor or abacavir. Patients were also required to have at least one haemoglobin and weight measured during follow-up, and for their height to be recorded. Where patients have more than one CD4 cell count measured within a 28 day window, the mean CD4 cell count and date was used to assign a single CD4 cell count value and date for that period. Least squares regression was used to determine the slope of CD4 cell counts based on each set of three (or averaged) CD4 cell count values. Thus each time a patient had an additional CD4 cell count measured, the slope was recalculated based on this new value and the previous two values. This slope was standardized per 3 months of follow-up.
Poisson regression was used to determine factors associated with the short-term risk of clinical progression. Patient follow-up began at the date of starting cART (baseline) and ended at the first new AIDS-defining illness or death (from any cause), or at last follow-up for patients without clinical progression. Patient follow-up was left-truncated until the date when all prognostic variables were first recorded. Explanatory variables were first included in univariate analyses. Baseline variables included gender, HIV-exposure group, CD4 cell nadir, ethnic origin, region of Europe, details of prior ART, date of starting cART, cART regimen started and prior AIDS diagnoses. Age, hepatitis B and C virus status, time since starting cART, CD4 cell count, CD4 cell slope, viral load, anaemia, body mass index (BMI) and whether or not patients were currently receiving any ART were all modelled as time-updated values, which means that they were used to describe the risk of new AIDS/death over the short term. This is in contrast to using fixed values at starting cART, which typically predicts the long-term risk of new AIDS/death. Continuous variables, such as CD4 cell count, were categorized a priori, using common cut-offs or to ensure sufficient events within each category to allow calculation of event rates. Variables that were significant in univariate analyses (P < 0.1) were included in multivariate models; categories were combined where appropriate, and a backwards stepwise procedure was used to remove variables not significant in this model. Finally, all excluded variables were added, in turn, to the final model. None of the excluded variables either improved the fit of the model (as measured by the change in deviance of the model) or were confounding variables (significantly altering the estimate of the incidence rate ratio), and, therefore, they were not included in any subsequent analyses. The natural logarithms of the incidence rate ratios were used to construct a prognostic score, which assessed the risk of clinical disease progression over the next 3-6 months (i.e., the short-term risk).
Simple categories of the score were created and the incidence of new AIDS/death within each of these categories, together with the probability of clinical progression, was calculated. This score was validated on patients from the Swiss HIV Cohort Study [16]. SHCS patients were required to satisfy the same inclusion criteria and 672 patients included in both the SHCS and EuroSIDA were excluded.
Statistical analyses were performed in EuroSIDA using SAS version 9.1 (SAS institute, Cary North Carolina, USA), and in SHCS using STATA (Version 9.1, StataCorp, College Station, Texas, USA).
|
|
| |
| |
|
|
|
