Data source
The French Hospital Database on HIV (FHDH) is a hospital-based open multicentre cohort in which inclusions have been ongoing since 1989 [7]. It includes data from 70 French general or university hospitals distributed throughout France. Individuals are eligible if they have documented HIV-1 or HIV-2 infection and give their written informed consent to participate. Data are collected prospectively by trained research assistants on standardized forms which include demographic characteristics, biological markers such as the CD4 cell count and plasma HIV RNA level, the date and type of AIDS and non AIDS-defining events, antiretroviral treatments, and the date and causes of death.
Study population
From the FHDH ANRS CO4 cohort, we selected heavily treated HIV-1-infected individuals (prior exposure to at least 2 NRTI, 2 PI and 1 NNRTI), with VL > 50 copies/mL who were starting a new drug (either a PI, or a NNRTI or an integrase inhibitor, or an entry inhibitor or a fusion inhibitor) between 2005 and 2011. The date of inclusion or baseline was the date the first such drug was prescribed during the study period. The study period was chosen in order to assess the risk of hospitalization before and after ETR became available in France. Individuals with less than 6 months of follow-up after the prescription of a first new drug and no available CD4 cell count within 6 months before the date of inclusion were excluded.
Statistical analysis
We compared follow-up with or without access to ETR + PI and a given participant can contribute to only the ETR + PI or to only the no ETR + PI or both, over the course of the study. As a given individual could initially be unexposed and then exposed to ETR + PI during the study period, methods taking into account the time-dependent nature of the variable of interest were used for descriptive analyses comparing characteristics of exposed and unexposed individuals and to analyze the impact of ETR + PI exposure on the hospitalization rate. For each individual, the follow-up period was divided into months. To compare the characteristics of exposed and unexposed individuals, a weighted Chi-square test was used, the weight being the duration of follow-up within each month.
The primary endpoint was the rate of hospitalization. Secondary endpoints were the two individual components of the primary endpoint, namely hospitalization from an AIDS-defining cause and hospitalization from a non-AIDS defining cause. We considered only hospitalizations lasting > 24 h, that were not due to pregnancy or to regular medical follow-up or medical examinations. Using an intention-to-continue-treatment approach, the number of hospitalizations and person-times were calculated for each calendar month for the individuals who received ETR + PI, during the months after initiating ETR + PI (ETR + PI) or for the individuals who received ETR + PI, in the months before ETR + PI initiation and for the individuals who never received ETR + PI (no ETR + PI). Individuals who received ETR + PI could also received another ARV at the same time. Adjusted incidence rates and relative risks (RR) were obtained from Poisson regression models in order to compare hospitalization rates between the exposed and unexposed groups, taking into account the following potential confounders: gender and transmission group (men who have sex with men, injecting drug users, other men, other women), geographic origin (sub-Saharan Africa, others), age, HCV co-infection status, the nadir CD4 cell count (< 50, 50–100, 100–200, ≥ 200 /mm3), the baseline CD4 cell count (< 200, 200–350, 350–500, ≥ 500 /mm3), baseline viral load (50–500, 500–5000, 5000–30,000, 30,000–100,000, ≥ 100,000 copies/mL), AIDS status at baseline, pneumocystis jiroveci prophylaxis (dapsone, cotrimoxazole or pentamidine aerosol) at baseline and number of previous antiretroviral drugs (ARV). Boosted ritonavir was not counted as an additional drug. In addition, as the centre size can influence quality of care, the analyses were adjusted for total follow-up (person-years (P-Y)) of heavily treated patients in each centre, using a four-level variable according to quartiles distribution (< 140 P-Y, 140–240 P-Y, 240–400 P-Y, ≥ 400 P-Y). A second model was further adjusted on exposure to NRTI and to raltegravir (RAL), a new drug in the new integrase inhibitor class, also made available in 2006 for individuals on failing regimens, as a time-dependent variable, to disentangle the respective roles of access to ETR + PI and to RAL. As improvements in individual management during the study period, rather than exposure to ETR + PI, could also explain observed changes, we conducted a sensitivity analysis including the period in the model (2005–2006 / 2007–2008 / 2009–2011). All tests of significance were two-sided, and p values < 0.05 were considered to denote significant differences. All analyses were done with SAS software version 9.3.