Combined Antiretroviral Therapy (cART) Reduces AIDS-Related and Non- AIDS-Related Mortality: A Temporal Analysis from Time of Seroconversion (SC)

Maria Dorrucci1, Luca Colarusso2, Vincenza Regine3*, Simona Di Giambenedetto4, Giovanni Di Perri5, Francesco Castelli6, Laura Camoni3, Angela Calamo7, Massimo Giuliani8, Rino Bellocco2,9, Francesca Farchi1, Mauro Zaccarelli10 and Barbara Suligoi3 1Department of Infectious, Parasitic and Immune-mediated Diseases, Istituto Superiore di Sanità, Rome, Italy 2Department of Statistics and Quantitative Methods, University of Milano-Bicocca, Milan, Italy 3National AIDS-Unit, Istituto Superiore di Sanità, Rome, Italy 4Università Cattolica S. Cuore, Rome, Italy 5Infectious Diseases, University of Turin, Italy 6Spedali Civili, Brescia, Italy 7 Malattie Infettive, Policlinico di Bari, Italy 8San Gallicano Dermatological Institute, Rome, Italy 9Department of Medical Epidemiology and Biostatistics, Karolinska Institutet, Stockholm, Sweden 10National Institute for Infectious Diseases “Lazzaro Spallanzani”, Rome, Italy


Introduction
In recent years, mortality for people living with HIV has dramatically decreased. The reason for such decreasing trend is inextricably related to the introduction in the middle of 1996 of the combination antiretroviral therapy (cART) that allowed life expectancy of HIV-infected individuals to approach that of the general population when successfully treated [1].
Recent studies investigated trends in causes of death in HIVpositive populations showing that overall mortality, primarily AIDSrelated mortality, continuously decreased over time [7][8][9]. However, no study has evaluated the trend in mortality taking into consideration the duration of HIV infection, i.e., since the estimated date of seroconversion. This allows to perform estimates among those who have had their HIV-diagnosis early, so have the best chances of access to treatment, while the prevalent cohorts could include patients who start therapy later because of late HIV-diagnosis.
We investigated the change over time of AIDS and non-AIDS related deaths and the association with some demographic and clinical characteristics comparing two post-cART periods (1996-2003 and 2004-2010) in an Italian cohort [Italian HIV-seroconversion study (ISS)] composed of individuals followed since the estimated date of SC.

Study population
Study population includes individuals enrolled in the ISS cohort that is an open, multicenter (18 infectious disease clinics throughout Italy), prospective, incident cohort established in Italy in 1987. Details of this cohort have been described elsewhere [10,11].
Briefly, individual data are collected at clinical sites, extracted electronically from clinical and laboratory data files and sent to the coordinating center (Istituto Superiore di Sanità, the Italian National Institute of Health) approximately every 6 months. Prerequisite to be enrolled in the cohort was the availability of an HIV-negative test followed by an HIV-positive test. The date of seroconversion (SC), estimated as the midpoint of the dates of the two tests, was considered as the starting point in time for all survival-related analyses.
Reference clinicians participating in the cohort study provided anonymous individual data on a predefined set of demographic, laboratory, and clinical variables, including age, sex, exposure-category HCV serostatus, cART regimen, CD4+ and CD8+ cell counts, HIV-1 viral load; HCV positivity was defined as the detection of anti-HCV antibodies and/or HCV-RNA.
The study included individuals enrolled between 1987 and 2010. Participants died for unknown causes or died in the pre-1996 period were excluded from analysis. The patients with unknown cause were excluded after verifying that main results did not change by including them as an additional end-point.
For sensitivity analyses we analyzed subsets of the study population who fulfilled specific criteria, as described in detail later.

Follow-up
Participants were followed from the estimated date of SC until either the date of death, or loss to follow-up, or established censoring date for ascertaining events (i.e., December 31, 2010), whichever occurred first.
To assess trends in overall and cause-specific mortality, the follow-up period was divided into two calendar periods, 1996-2003 and 2004-2010, defined as early and recent cART period, respectively. We chose these two calendar periods when observing that from 1996 to 2010 a decreasing trend over 2 calendar yrs intervals of both AIDS and non AIDSmortalities rates were observed. After 2004 two new classes of ART drugs were introduced in Western countries, i.e., fusion and entry inhibitors that could have had an impact on increased survival. Total person-years, therefore, (i.e., the number of years each individual contributes during follow-up) were estimated for the two calendar periods.
Information on causes of death were obtained from reference ISS cohort clinicians. In particular, clinicians were requested to state in the death certificate the immediate cause, the contributing cause, and the underlying cause of death [10]. Further, the validity of causes reported by cohort clinicians were checked both in the National AIDS registry [12] and/or in the official-local death registries and coded (by clinicians of the study cohort) according to ICD-9 or the ICD-10 [13] depending on when death occurred.
For the analyses deaths were grouped as AIDS and non-AIDSrelated deaths. And non-AIDS deaths were classified as cardiovascular disease-related, hepatitis-related, non-AIDS cancer and other.

Statistical analysis
We applied a competing risks approach, i.e., one of the statistical methods used when an individual is at risk of more than one type of event [14]. In the present study, enrolled individuals were at risk of dying from various diseases and deaths were grouped, for simplicity, as AIDS or non-AIDS-related deaths. In the above-mentioned context, we estimated the cumulative incidence function (CIF) as a measure of competing-risk death, and Cox-cause-specific models. We chose to apply Cox-cause-specific models assuming that the underlying cause of death was accurately recorded by clinicians.
CIF of AIDS and non-AIDS related deaths was estimated and plotted for the two selected calendar periods, controlling for staggered entries in both periods from SC date. We estimated also relative hazards (RH) and 95% Confidence Interval (CI) for AIDS and non-AIDS mortality by Cox-cause-specific models [14]. The variable of interest in these Cox models was calendar period entered as a dichotomous-timedependent covariate (i.e., 2004-2010 vs. 1996-2003); the time zero of all Cox models was the estimate date of SC. Cox models were stratified by gender, HCV serostatus (positive vs. negative), exposure-category: injecting drug users (IDUs), men who have sex with men (MSM), heterosexuals, and not determined. Due to the importance of ageing, the variable age was considered as the time scale instead of time from SC. The latter, in brief, was an alternative way of controlling for age in order to explicit the ageing effect [15,16].
High CD4 count (>500 cells/mm 3 ) or low viral load (VL) (< 200 copies/ml) observed after cART initiation were considered as indicators of effective antiviral treatment. These variables were analyzed to assess associations between effective cART and survival. To this aim, lefttruncation was applied from SC at selected visits with last CD4 ≥ 500, cells/mm 3 or last HIV-load < 200 after start of cART. In these cases only follow-up after the above mentioned dates were considered, i.e., only the survival experiences after these dates (last CD4 ≥ 500, cells/ mm 3 or last HIV-load < 200) were included in the analysis. Of note, to account for possible immortal time-bias, all estimates (CIF graphs and RHs) were performed considering for each individual delayed entries in each calendar period from SC, in other terms, considering the different survival times from SC to the beginning of each calendar period.

Ethics
The study was approved by the "Comitato Etico dell'Istituto Superiore di Sanità" (National committee for ethics of bio-medical studies in Italy).

Recruited patients in the ISS cohort
A total of 3,031 adults were enrolled and followed in the ISS study  Table 1: 1,779 patients were followed in early cART period with 8,768 person-years estimated for the early cART period. Of these, 560 (31%) were new SC (seroconversions) diagnosed in that period. During more recent cART period, 1,715 patient were followed with 8,462 person-years of follow-up, of these 370 (22%) were new SC. Compared with patients in the early cART period, seroconversion in the recent cART period occurred at little more old age (30 vs. 28 years).
from 1987 to 2010. The median interval between last-HIV negative and first HIV-positive dates was 0.7 yrs with IQR: 0.4-1.1 yrs. We excluded 180 (6%) patients with an interval of SC > 3 yrs (median 5 yrs, IQR: 4-6 yrs, range: 3-22 yrs), and we further excluded 541 and 61 patients (because of unknown causes of death, as mentioned in the inclusion criteria in study population paragraph [the majority (18%) because died before 1996, Fluxogram 1].
The selected population of 2,249 individuals was followed for 24,043 person-years of follow-up, during whom a total of 43,394 visits were performed (median number of visits: 14; IQR: 7-28). Of note, the period of interest included a total of 17,230 person-years, i.e. from 1996 to 2010 (Fluxogram 2).
The proportion of males tended slightly to increase from early to recent cART period (48% vs. 86%, respectively) as well as the proportion of men who have sex with men (MSM) resulted higher from early to recent cART period, whereas the proportion of injecting drug users (IDU) decreased. The proportion of heterosexuals slightly increased. The proportion of HCV-HIV co-infected was higher in the first period than in that after (54% vs. 41%).

Causes of death
During follow-up, 278 deaths were identified (197 (71% of total occurred deaths; and 81 (29%) in more recent cART period).
In Table 2 are shown specific causes of AIDS-related death and non-AIDS-related deaths.
Incidence Rates of more common mentioned causes of death observed in the both periods are shown in Table 3. Comparing the two calendar periods the IR of all grouped causes of death was lower in the more recent period, whereas the IR of non-AIDS cancers was stable.

CD4 and VL one year before death
CD4 cell count within one year before death for both early and more recent cART period was graphically summarized in Figure 1: median CD4 cell count before death resulted for all causes higher in more recent period vs. early cART period (medians and interqurtile ranges (

Cumulative incidences functions of non-AIDS and AIDS death
Cumulative incidences (CIF) for the two grouped causes of death in the early cART and in more recent cART period are shown in Figure 2. Both CIF of AIDS and non-AIDS death decreased over time from SC according to the two calendar periods; in the last period CIF of non-AIDS related death tended to be slightly higher than AIDS related death (for instance, at 15 yrs from SC cumulative incidence was 0.03 (95% CI: 0.02-0.04) for non-AIDS and 0.02 (95% CI: 0.02-0.04) for AIDS-related death). In Table 4 are shown RH of death relative to (2004-2010) vs. early calendar period (1996-2003) applying proportional-cause-specific regression models; all hazards of death resulted lower for recent cART period compared to the early cART period.
In particular, starting from SC the non-AIDS mortality decreased of nearly 62% (95% CI: 44%-74%), whilst AIDS mortality was reduced by 70%. Starting from a same age attained, the reduction of mortality was again less for non-AIDS than AIDS mortality: for instance, for those 45 yrs old we observed a reduction of 59% in non-AIDS mortality, whilst for AIDS mortality we found a reduction of 43%.
Even when considered the achievement of CD4 > 500 cells/mm 3 or of a HIV-viral load (VL) <200 copies/mL), the reduction of non-AIDS death resulted not statistically significant and lower than that of AIDS death comparing the two calendar periods.

Discussion
Our findings suggest that both AIDS and non-AIDS mortalities in HIV-individuals with known date of seroconversion (SC) for HIV have further decreased from the early cART (1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003) to the more recent cART therapy period (2004)(2005)(2006)(2007)(2008)(2009)(2010). At the same time, we found that non-AIDS-related mortality was slightly higher than AIDSrelated-mortality. These results were confirmed after controlling for similar duration of HIV-infection and effective cART. The reduction of mortality seemed to be observable for most causes of deaths, except for non-AIDS cancer, the rate of which did not change during the more recent cART years.  deaths over time, including into most recent time [7][8][9].
We observed a RH of AIDS-mortality equal to 0.30 and equal to 0.38 of non-AIDS mortality for the period 2004-2010 vs. 1996-2003 from the estimated HIV-SC. A direct comparison of this estimate with other studies could not be made because, to our knowledge, all other studies were not performed since the date of SC.
The observed decrease in mortality can be mainly explained with the increase of time (i.e., person-years) spent by individuals on cART therapy, given that it almost doubled from the earlier to the more recent cART period (from 48% to 86% of the time on follow-up, respectively). This is further confirmed by the finding that the RH estimated by Cox models decreased when the individual follow-up was considered since start of cART for both AIDS and non AIDS-related mortalities.
Other studies have been performed since the date of cART Smith et al. [7], for instance, estimated a significant decrease in the incidence ratio (IR) [IR = 0.63 (95 % CI: 95 % CI: 0.48-0.84, relative to the 2009-2011 vs. the 1999-2000 period]. The different estimates in our study may be due to the fact that in our seroincidence cohort the patients had most likely to start their therapy earlier than in seroprevalent studies [1,8,9], where patients receive care when they are already HIV-positive. Another reason for the pronounced decrease of mortality in our study could be the increase of CD4 cell counts in the studied population, as evidenced by the higher CD4 medians observed during the period 2003-2010, as well as the lower median HIV-load. This finding was also confirmed among those who reached a CD4 cell count greater than 500 cells/mm 3 ; in particular, the RH of AIDS-related deaths was 0.38 (95% CI: 0.18-0.80) from the last observed count > 500, whereas the RH of non-AIDS-related deaths turned out to be almost close to unity.
In our study the cumulative incidence (CIF) of AIDS-related deaths resulted substantially higher than the CIF of non-AIDS-related deaths in the early cART period, whilst there was a reversal in the 2004-2010 periods: the CIF of non-AIDS-related deaths resulted slightly higher than that of AIDS-related deaths. This finding was confirmed in multiple analyses given that the adjusted IR of non AIDS related-deaths resulted higher than that of AIDS-related deaths. Again, this result was not directly comparable with other studies because of the different design and methodology.
The result that the non-AIDS death rate, as a whole, overlaps with the AIDS death rate in our study was possibly due to the increase observed in non-AIDS-related cancers when comparing the early cART to the more recent cART period. In fact, we observed a decreasing tendency of all other non-AIDS-specific causes of death, such as cardio-vascular or hepatitis-related or due to any other infections except for non-AIDS cancers. Other studies [7,8,[17][18][19][20][21][22][23] observed a similar tendency from the start of cART. Smith et al., [7] in particular, found a statistically significant decrease of adjusted IR of both liver-related deaths and cardio-vascular deaths, but the adjusted IR of non-AIDS-cancers relative to recent cART years vs. early cART years always resulted nearly equal to unity, i.e., non-AIDS-related cancer rates have remained stable over time. These types of cancer seem now be the most common cause of non-AIDS death [8,17].
An important caveat to this study is that, being observational in nature, we cannot rule out the possibility of unmeasured confounding. Although deaths were accurately classified, autopsy results were not available.
In conclusion, we observed a dramatic decrease in both AIDS-and non-AIDS-related mortalities when comparing the early with the more recent cART era. This seems to be mainly due to an improvement in the benefits of cART. At the same time, in more recent cART years the magnitude of non-AIDS mortality, considered as a whole, was close to that of AIDS mortality, even after controlling for duration of HIV infection and effective cART. A possible explanation is that non-AIDS cancers-related deaths tend not to decrease over time, suggesting that this field of research requires further study.
Luca Colarusso (fully) and Rino Bellocco (partially) were funded by the Italian Ministry of Education (PRIN 2009X8YCBN).