Ecological Effects of Decolonisation Strategies in Intensive Care

Introduction

The "R-GNOSIS: Ecological Effects of Decolonization Strategies in Intensive Care" study assesses three decolonization interventions against standard care to evaluate unit wide ecological effects and compare effectiveness.

Previous studies have demonstrated that decontamination interventions were beneficial to individual patients but also influence ICU ecology, affecting patients who do not receive the intervention. Decolonization with antibiotics have been shown to reduce the prevalence of resistant bacteria during treatment [de Smet et al. NEJM 2009]. Reducing the presence of these bacteria in some patients (that are decolonized), reduces cross transmission and is therefore beneficial to all patients in the unit. The decolonization strategies therefore represent an intensive care unit population rather than an individual patient intervention. In this respect the study represents a cluster-cluster randomized clinical trial which requires the intervention is undertaken on the whole ICU population [Edwards et al. BMJ 1999].

As decolonization strategies represent an ecological intervention on the whole critical care population, all patients meeting inclusion/exclusion criteria will be entered into the study according to ethics approval in each participating country. Each participating ICU will use three decolonization strategies in a randomised order. The interventions are administered four times daily to ventilated patients until extubation. The interventions will be compared to a 6 month baseline period consisting of standard care only.

Standard operating procedures

Standard care

The baseline period is the first 6-month period and will be used to implement universal "standard care":

• Chlorhexidine 2% body washings (CHX-BW) for all ICU patients. CHX-BW ensures state of the art standard of care to prevent carriage and transmission of (resistant) gram positive bacteria commonly residing on body surface, such as Staphylococcus aureus and Enterococci.
• A hand hygiene improvement program (HHIP) based on the program designed by the World Health organisation (WHO). Implementation of the hand hygiene program derived from the WHO hand hygiene program ensures state of the art standard of care for transmission prevention of all relevant pathogens.
• Standard oropharyngeal care consists of oral washing with sterile water (3-4 times daily) and tooth brush twice daily.

"Standard care" will be the only protocolised intervention in the baseline period and will be used throughout the entire study.

Intervention periods

After the baseline period, the first decolonization regimen will be implemented. The order of regimens per ICU is decided by randomization. The three regimens are:

• Chlorhexidine oral care (CHX-Oro) with chlorhexidine 1% oromucosal gel.
• Selective oropharyngeal decontamination (SOD) with antibiotics. SOD consists of application of a paste containing colistin, tobramycin in a 2% concentration and nystatin 1 x 10^5 units.
• Selective digestive decontamination (SDD), in which a 10 ml suspension via the nasogastric tube containing 100 mg colistin, 80 mg tobramycin and nystatin 2 x 10^6 i.u. will be added to application of SOD paste.

All intervention periods last 6 months and all regimens are applied four times daily.

In contrast to some other SDD studies, systemic prophylaxis with Cefotaxime (or other broad spectrum cephalosporins) will not be implemented as part of SDD.

Patient recruitment

All patients in the ICU receive standard care and minor anonymized personal data are collected from them. Once monthly point prevalence cultures are collected from patients in the ICU on that day.

Eligible patients in the ICU - in addition to the above mentioned - will receive one of the interventions and will undergo surveillance sampling. More anonymized personal data are collected (including some clinical culture results).

A waiver of informed consent is in place, but patients can opt out for data collection.

Culture sampling

Point prevalence cultures are taken once monthly from each patient present in the ICU at that moment. These include a rectal swab and a respiratory sample and serve the purpose of monitoring and evaluating ecological changes during all regimens.

Surveillance cultures are taken twice weekly from included patients and also include a rectal swab and a respiratory sample. The samples are collected to measure the treatment effect.

Finally, results from regularly obtained cultures for clinical purposes (blood and respiratory samples) will be recorded to measure the treatment effect.

Data-collection

Data-collection can be performed by two methods. All data will be "anonymised" by recoding the patient identifier (ID) to a study patient ID and by removing their personal identifiers.

1. A web based electronic case report form (ECRF) has been designed within "Research Online". This system meets all requirements according to International Conference on Harmonisation Good Clinical Practice (ICH-GCP)standards for electronic data entry with respect to safeguarding data integrity and data security regulations.
2. Automatic extraction of data from electronic patient data dossiers may be performed if technically possible without harming the patients' privacy.

Data dictionary

• Data from all patients admitted to the ICU include sex, age, disease severity score, admission/discharge date, mechanical ventilation (yes/no) and duration, ICU-survival.
• Data from included patients additionally include: hospital admission date, place before admission, reason for ICU admission, acute illness (yes/no), sites of organ failure, antibiotic use on ICU-admission (yes/no), comorbidity, ventilation data, disposition at 28-days after ICU- admission, at ICU-discharge and at hospital-discharge, isolation precautions.
• Culture data include results from clinical respiratory and blood samples and surveillance cultures from included patients and point prevalence cultures (monthly) from all patients.
• Ward-level antibiotic use will be recorded per study period.

Sample size calculation

In a Dutch SDD ICU trial the day-28 mortality rate during the baseline period was 27.5% (3). Assuming a low level of cluster-effects, 2016 patients are needed in each phase to demonstrate a 10% relative reduction in day-28 mortality as compared to Standard Care (alpha=0.05; beta=0.8). We intend to include 2700 patients per arm. The margin of 600 patients per arm is included to allow for adjustment for differences in baseline characteristics in a random-effects logistic regression model if needed, or to include cluster-effects. Of note, assuming day-28 mortality in standard care to be 27%, the absolute reduction that can be demonstrated is 2.7%. Day-28 mortality data will be derived from clinical data obtained as part of routine standard care.

STATISTICAL ANALYSIS PLAN (updated October 2017)

INTRODUCTION.

Analysis will determine the effect of each intervention in the occurrence of bacteremia, patient survival, colonization rates, and the use of antibiotics. Statistical analysis of the primary objective and secondary objectives regarding mortality, bacteremia and ward-level colonization with antibiotic resistant bacteria will account for ICU-level clustering and the statistical methods used are described in detail in below. The use of antibiotics will be a descriptive statistic.

All available data on patient colonization with MDR-GNB (both from screening and clinical cultures) will be used to determine, as carefully as possible, the extended-spectrum beta-lactamase (ESBL) colonization status of each patient on every study day. Nosocomial transmission capacities (RA-values) for different species of MDR-GNB during study regimens will be quantified. As a secondary aim species-specific RA values will be compared between wards. Available data will be used to quantify incidences of cross-transmission in both study periods, using sophisticated modeling approaches.

Investigators and the R-GNOSIS staff will make every attempt to collect complete data from all subjects enrolled in the study. Where possible, automatic extraction of data from hospital information systems will be used (without disclosing patient identifiers). There will be regular contact between the study coordinating centre and study sites to track and retrieve missing data. Should culture results be inadvertently lost, those data will be treated as missing at random. All inferential analyses will be based on available data.

The detailed statistical analysis plan has been established prior to database lock and is divided into two parts.

PART 1: Clinical outcomes (patient data)

The data

The data analysis will be performed on all patients included during the baseline period, the last 2 weeks of the wash-out/in periods and those included during one of the three intervention periods. Two ICU admissions of the same patient with less than 3 days in between will be merged and analyzed as one ICU admission.

The following cohorts will be made for analysis of the following clinical outcomes:

1. Cohort "ICU-admissions": ICU-acquired bacteremia, ICU survival.
2. Cohort "Hospital-admissions": Hospital survival
3. Cohort "first ICU-admissions", excluding re-admissions to the ICU within 30 days after prior ICU-discharge: 28-day survival

Missing data.

Missing data will be retrieved where possible, after which a complete case analysis will be performed.

Statistical models.

To adjust for potential selection bias in this cluster randomized trials with crossover (without blinding), the statistical analysis will be performed using doubly robust estimation. [Funk MJ et al. Am J Epidemiol 2011]

Propensity score model.

The propensity score model will include the following a priori selected confounders:

• Age
• Gender
• Disease severity (either APACHE II or SAPS II score)
• Use of antibiotics upon ICU-admission
• Prior location before ICU-admission
• Admission type (medical/surgical/trauma)
• Charlson comorbidity score
• Hospital of recruitment (hospital) As two different scoring methods , APACHE II or SAPS II, have been used to determine disease severity by different hospitals, two separate propensity score models will be fitted (one for hospitals that recorded APACHE II and one for hospitals that recorded SAPS II).

These propensity score models will be fitted in the cohort "ICU-admissions" using the R-package 'twang'. [McCaffrey Stat Med. 2013] This package uses generalized boosted models machine learning techniques to calculate weights for each patient. The resulting weights represent the inverse probability for a patient to be included in the baseline, CHX, SOD or SDD arm and will be used to weigh the data in the outcome models , creating pseudo-populations with an equal distribution of the specified covariates over treatment groups.

Outcome models.

Separate models will be fitted per endpoint, as specified in table 1. All models will include the inverse probability weights, the confounders included in the propensity score model and the mean hand hygiene compliance per study period per hospital (hand hygiene compliance might differ per study period and act as a confounder on all outcomes) to obtain doubly robust estimators. A dummy variable indicating the measure of disease severity (APACHE II or SAPS II) will be included as an interaction with the standardized disease severity to overcome different hospital having registered different measures. In addition, two levels of clustering will be taken into account, as follows:

• Hospital of recruitment: fixed effect, acknowledging that the risk of the outcome differs per hospital
• Cluster period (i.e. periods 1-4 per hospital): random intercept, acknowledging that patients recruited in different periods within hospitals may be more alike with regard to the risk of the outcome.

Table 1. Outcome models per endpoint

• ICU-acquired bacteremia Model: Cox proportional hazard (hazard ratio) Competing endpoints: ICU discharge, death in ICU
• ICU survival Model: Cox proportional hazard (hazard ratio) Competing endpoint: ICU-discharge
• ICU survival Model: Cox proportional hazard (hazard ratio) Competing endpoint: hospital-discharge
• 28-day survival Model: generalized linear model (odds ratio) Family: (quasi)binominal Link: logit

Results will be presented as hazard ratios or odds ratios with 95%-CI. R and STATA will be used to perform the analyses specified above.

Sensitivity analysis.

As eligibility in this open cluster-randomized study was defined as "Expected length of MV >24h", selection bias may have occurred (as discussed under 'confounding adjustment'). To quantify this potential bias a sensitivity analysis will be performed in which patients who left the ICU within 2 days after study inclusion are excluded, as these patients could not reach the ICU-acquired bacteremia primary endpoint (which requires at least three days in ICU).

Exploratory analysis.

As an exploratory analysis the treatment effect on 28-day survival and ICU-acquired bacteremia per ICU will be visualized in Forest plots, in which ICUs are ranked on the prevalence of bacteremia with highly resistant micro-organisms (HRMO) during the baseline period.

PART 2: Antibiotic resistance (ward level data)

The data.

The monthly point prevalence screenings on both included and non-included patients will be analyzed on two levels, each including outcomes for individual HRMO (e.g. carbapenem resistant GNB, MRSA, etc.) and the aggregate "any HRMO".

1. Cohort "respiratory tract"
2. Cohort "digestive tract"

Missing data.

If a patient or tractus was not sampled, it will be excluded from the analysis (and the denominator). If an antibiotic susceptibility result was missing, the highest susceptibility result from the same species in the same tractus 7 days prior or after the point prevalence date was "imputed", if available. Completeness of susceptibility testing will be reported as a descriptive statistic.

Since monthly point prevalence measurements are taken on fixed days (i.e. first Monday, occasional exceptions accepted) on all patients present in the ward (both included and non-included) we do not expect bias due to selective inclusion per study period (as in the analysis of the clinical outcomes).

Final models.

We will perform logistic regression analyses with a log-link for each endpoint and include terms for underlying time-trend per hospital (months since study start * hospital) and time-trend per intervention (months since start study period * study period) and correct for repeated measurements on the same patient (corrected standard errors with sandwich estimator). Results will be presented as risk ratio's with 95%-CI.

Quality and safety assurance plan

The quality of the study will be assured by two methods.

1. An external safety committee (SCom) consisting of three experts has the objective to guard the ecological safety during the study. During the intervention periods (CHX-Oro, SOD, SDD), the SCom will issue recommendations to continue or stop the study on a quarterly basis (three monthly), based on the results of monthly point prevalence cultures and input by participating ICUs. The primary safety measure to detect an increase of anti-microbial drug resistant bacteria is performing point prevalence cultures in all patients in the participating ICUs. The purpose of these cultures is to evaluate and ensure ecologic safety. Analysis will detect any multi-drug resistant gram negative bacteria (MDR-GNB), vancomycin resistant enterococci (VRE) or methicillin-resistant Staphylococcus aureus (MRSA) isolates. Also, susceptibility of /MDR-GNB to colistin, both used in SDD and SOD, will be tested. This way, any increase in the number of resistant isolates will be detected early.
2. Guidelines for reporting suspected unexpected serious adverse reactions (SUSAR's) have been developed. In ICU patients, co-morbidity and the natural history of the underlying critical illness can cause events which would meet the definition of (serious) adverse events. Given the natural occurrence of these events and the low risk for adverse drug reactions based on the broad experience with the current study medication, only the following adverse events will be recorded:
3. Adverse events possibly related to the medication (as judged by medical and scientific judgement) AND
4. Deemed serious by medical or scientific judgement (as judged by either the investigator of treating physician) AND
5. Not part of the natural history of the underlying critical illness Adverse events meeting these criteria should be reported by the local investigator within the following time limits:

• to the coordinating investigator within 24 hours
• to the accredited Institutional Review Boar (IRB) that has approved the protocol in that country (within 7 days if the event is life-threatening or fatal, within 15 days if the event is not life-threatening or fatal)
• to the competent authority of that country (within 7 days if the event is life-threatening or fatal, within 15 days if the event is not life-threatening or fatal)

1. A monitoring plan has been developed, in which:
2. The SPONSOR completes a central monitoring form for each participating hospital every three 3 months. This form registers for example the quality of transferred data, recruitment rate and the occurrence of site specific problems.
3. Each participating hospital completes a site self-monitoring form every three months. This form registers the completeness of the Investigator Site File, completeness of drug accountability and the correct reporting of suspected unexpected serious adverse reactions (SUSAR's).