Dynamic clinical prediction models for discrete time-to-event data with competing risks-A case study on the OUTCOMEREA database.

The development of clinical prediction models requires the selection of suitable predictor variables. Techniques to perform objective Bayesian variable selection in the linear model are well developed and have been extended to the generalized linear model setting as well as to the Cox proportional hazards model. Here, we consider discrete time-to-event data with competing risks and propose methodology to develop a clinical prediction model for the daily risk of acquiring a ventilator-associated pneumonia (VAP) attributed to P. aeruginosa (PA) in intensive care units. The competing events for a PA VAP are extubation, death, and VAP due to other bacteria. Baseline variables are potentially important to predict the outcome at the start of ventilation, but may lose some of their predictive power after a certain time. Therefore, we use a landmark approach for dynamic Bayesian variable selection where the set of relevant predictors depends on the time already spent at risk. We finally determine the direct impact of a variable on each competing event through cause-specific variable selection.

Treatment of severe hospital-acquired and ventilator-associated pneumonia: a systematic review of inclusion and judgment criteria used in randomized controlled trials.

BACKGROUND: Hospital-acquired and ventilator-associated pneumonia (HAP/VAP) are often selected for randomized clinical trials (RCTs) aiming at new drug approval. Guidelines for the design of such RCTs have been repeatedly updated by regulatory agencies. We hypothesized that large variability in the enrolled populations, the endpoints assessed and the HAP/VAP definition criteria may impact the results of these studies, and addressed this through a systematic review of HAP/VAP RCTs. METHODS: A search (Pubmed-Embase-ICAAC-ECCMID) of all RCTs published between 1994 and 2016 comparing antimicrobial treatment for HAP/VAP in the intensive care unit was conducted. The populations enrolled, inclusion/exclusion criteria, statistical design and endpoints assessed were recorded. All unpublished RCTs recorded on the ClinicalTrials.gov registry were also screened. RESULTS: From the 93 abstracts reviewed, 39 potentially relevant studies were inspected, leading to 27 studies being included. As expected, illness severity or the proportion with VAP (27-100%) differed greatly among the enrolled populations. The HAP/VAP definition used various clinical and biological criteria, and only 55% of studies required a microbiological sample. The mandatory duration of prior hospital stay was variable; the mechanical ventilation duration was an inclusion criterion in only 41% of VAP studies. Nine studies had non-inferiority design, but nine studies (33%) did not have a pre-specified statistical hypothesis. Clinical cure was the primary endpoint in 24 studies, but was recorded in several populations or as the co-primary endpoint in 13 studies. The definition of clinical cure and the timing of its assessment greatly differed. This variability slightly improved over time but remained significant in the 13 registered but currently unpublished RCTs that we screened. CONCLUSION: Our study provides a description of populations and endpoints of RCTs evaluating antimicrobials for treatment of HAP/VAP in the ICU. There was significant heterogeneity in enrollment criteria, endpoints and statistical design, which may influence the ability of studies to demonstrate differences between studied drugs.

Attributable mortality of ICU-acquired bloodstream infections: Impact of the source, causative micro-organism, resistance profile and antimicrobial therapy.

OBJECTIVES: ICU-acquired bloodstream infection (ICUBSI) in Intensive Care unit (ICU) is still associated with a high mortality rate. The increase of antimicrobial drug resistance makes its treatment increasingly challenging. METHODS: We analyzed 571 ICU-BSI occurring amongst 10,734 patients who were prospectively included in the Outcomerea Database and who stayed at least 4 days in ICU. The hazard ratio of death associated with ICU-BSI was estimated using a multivariate Cox model adjusted on case mix, patient severity and daily SOFA. RESULTS: ICU-BSI was associated with increased mortality (HR, 1.40; 95% CI, 1.16-1.69; p = 0.0004). The relative increase in the risk of death was 130% (HR, 2.3; 95% CI, 1.8-3.0) when initial antimicrobial agents within a day of ICU-BSI onset were not adequate, versus only 20% (HR, 1.2; 95% CI, 0.9-1.5) when an adequate therapy was started within a day. The adjusted hazard ratio of death was significant overall, and even higher when the ICU-BSI source was pneumonia or unknown origin. When treated with appropriate antimicrobial agents, the death risk increase was similar for ICU-BSI due to multidrug resistant pathogens or susceptible ones. Interestingly, combination therapy with a fluoroquinolone was associated with more favorable outcome than monotherapy, whereas combination with aminoglycoside was associated with similar mortality than monotherapy. CONCLUSIONS: ICU-BSI was associated with a 40% increase in the risk of 30-day mortality, particularly if the early antimicrobial therapy was not adequate. Adequacy of antimicrobial therapy, but not pathogen resistance pattern, impacted attributable mortality.

Colonization and infection with extended-spectrum ß-lactamase-producing Enterobacteriaceae in ICU patients: what impact on outcomes and carbapenem exposure?

OBJECTIVES: It remains uncertain whether colonization and infection with ESBL-producing Enterobacteriaceae (ESBL-PE) affect the outcomes for ICU patients. Our objectives were to measure the effects of ESBL-PE carriage and infection on mortality, ICU length of stay (LOS) and carbapenem exposure in this population. METHODS: A cause-specific hazard model based on prospectively collected data was built to assess the impact of ESBL-PE colonization and infection on competing risks of death and ICU discharge at day 28 in a multicentre cohort of ICU patients. Carbapenem exposure during the ICU stay was compared between infected carriers, uninfected carriers and non-carriers. RESULTS: Among the 16,734 included patients, 594 (3.5%) were ESBL-PE carriers, including 98 (16.4%) with one or more ESBL-PE infections during the ICU stay. After adjustment for baseline and time-dependent confounders, ESBL-PE infections increased the probability of death at day 28 [adjusted cause-specific hazard ratio (aCSHR), 1.825, 95% CI 1.235-2.699, P = 0.0026] and the ICU LOS (aCSHR for discharge alive at day 28, 0.563, 95% CI 0.432-0.733, P < 0.0001). ESBL-PE carriage without infection extended the LOS (aCSHR, 0.623, 95% CI, 0.553-0.702, P < 0.0001), without affecting mortality (aCSHR, 0.906, 95% CI, 0.722-1.136, P = 0.3916). Carbapenem exposure increased in both infected and uninfected carriers when compared with non-carriers (627, 241 and 69 carbapenem days per 1000 patient days, respectively, P < 0.001). CONCLUSIONS: ESBL-PE infections increased carbapenem consumption, LOS and day 28 mortality. ESBL-PE infections were rather infrequent in carriers; however, even ESBL-PE carriage without infection increased carbapenem exposure and delayed discharge, thereby amplifying the selective pressure and the colonization pressure in the ICU.