Study design

A multicentric observational prospective cohort study will be conducted in around 15 hospitals participating in the IMPACTO MR platform,14 representing all Brazilian regions. Eligible centres must have active Infections Control Committees (ICC) and regular reporting of HAI to ANVISA (figure 1). Both public and private hospitals were included, as well as heterogeneous ICUs profiles. Paediatric or neonatal ICUs were not included.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

IMPACTO MR-PAV study flowchart. ICU, intensive care unit; IVAC, infection-related ventilator-associated complication; PVAP, possible ventilator-associated pneumonia; VAC, ventilator-associated condition; VAE, ventilator-associated events; VAP, ventilator-associated pneumonia.

Eligibility

All patients aged 18 years or older who are hospitalised in public or private ICUs using mechanical ventilation will be eligible for inclusion. Centres participating in other HAI studies funded by the Brazilian Ministry of Health (ie, ‘Saúde em Nossas Mãos’) will not be eligible to participate in IMPACTO MR-PAV because we recognised the potential for data quality bias from the participating ICUs—‘Saúde em Nossas Mãos’ provides training in ways to prevent hospital infections, which could be a bias for this present study. Currently, 83% of participating centres are public or philanthropic facilities and 17% are university-affiliated ICUs.

Data collection and source or information

Data will be collected for a period of 6 months in each centre. Seasonal sampling will not be necessary considering the duration of study accrual period (>1 year), differences in initiation dates and Brazilian macroregions coverage.

The patients will be monitored during the period they remain on mechanical ventilation in the participating ICU of the study. Trained health professionals from ICC of each participating centre will collect data. Demographic and clinical profile data will be captured through the Epimed database, an administrative database used for quality improvement in Brazil (https://secure.epimedmonitor.pt/Login.aspx)20 at the time of patient admission to the study ICU. Mechanical ventilation data will be entered daily into the institutional database system from the moment the patient is submitted to mechanical ventilation until the first day after the end of mechanical ventilation. VAPs notified to ANVISA, according to its own criteria, will be reported monthly in the database system for adjudication. The adjudication will be carried out in an independently fashion (ie, one adjudicator is unaware of other’s judgement—masked) by two professionals (doctor and nurse) with experience in HAIs surveillance. They are professionals who work in hospital infection control. They were properly trained by videoconferencing and received a manual with instructions. In case of disagreement, a third adjudicator will be consulted. The CDC criteria15 will be applied to define VAPs. The VAE definition will be performed automatically in the database system by a specific algorithm. The algorithm’s diagnosis will be manually validated by a nurse with experience in HAI surveillance. The demographic profile of patients who develop VAP will be described, including age, sex, comorbidities, Simplified Acute Physiology Score - 3 (SAPS 3),21 time on mechanical ventilation until the occurrence of the event and outcomes such as time on mechanical ventilation, ICU length of stay, hospital length of stay and 28-day mortality. The overall incidence of VAP per 1000 ventilator-days, ventilator-associated condition (VAC) per 1000 ventilator-days, infection-related ventilator-associated complication (IVAC) per 1000 ventilator-days and possible ventilator-associated pneumonia (PVAP) per 1000 ventilator-days will be calculated. Outcome data will be compared between different VAP groups. The microbiological profiles of isolates in VAP cultures by CDC criteria and ANVISA criteria will be described. The concordance (kappa) and the accuracy results for identifying VAP (sensitivity, specificity, positive predictive value and negative predictive value) between the two criteria will be evaluated through VAPs confirmed by adjudication and PVAP validated in automated surveillance. As detailed in ‘Statistical analysis’ section, three complementary approaches will be performed considering we are facing imperfect standard tests. Comparative scenarios will consider: (1) both ANVISA criteria and VAE criteria will be compared with adjudication by a specialist as a standard test and (2) VAE criteria will be compared with ANVISA criteria as the standard test considering some assumptions in a Bayesian analysis. The main characteristics of both criteria are described in table 1.

To validate our electronic case report form, we conducted two preliminary pivotal analyses. The first involved an artificial simulation of events (VAC, IVAC and PVAP), while the second used formal patient data collected at the coordinating centre. These tests demonstrated that all events were correctly identified, with no ambiguous classifications. The system’s diagnosis will be manually validated by a nurse with experience in HAI surveillance as a double-check. The primary and secondary outcomes of this study were defined based on their clinical relevance and direct impact on patients’ well-being, including mortality and potential effects on antimicrobial stewardship programmes and on antimicrobial resistance.

Outcomes

Primary and secondary outcomes are defined as follows. More detailed information regarding outcomes definition is available on online supplemental material 1.

Statistical analysis

Initially, we planned that a sample size of 226 individuals would be able to detect a kappa coefficient of 0.8 (or greater) with a statistical power of 90%, assuming a kappa of 0.5 under the null hypothesis, a two-sided hypothesis test with a significance level of 5% and a proportion of positive evaluations (positive diagnosis) across a wide range of possible values (0.10; 0.30; 0.50; 0.7; 0.90).22 In this context, the kappa coefficient with a 95% CI would be used to evaluate the degree of agreement between two diagnostic methods defined by binary responses (positive or negative).23 The Landis and Koch24 criteria would be used for interpreting the agreement, as follows: (a) almost perfect, for values from 0.81 to 1.00; (b) substantial, for values from 0.61 to 0.80; (c) moderate, for values between 0.41 and 0.60; (d) fair, for values between 0.21 and 0.40; and (e) slight, for values from 0 to 0.20; and poor, for negative values. Finally, we considered that if the sample size is less than 30, the CI for kappa will be estimated using the bootstrap method.

However, when the estimated sample size was achieved in November 2022, there were only few reported events (VAP). Therefore, no reliable conclusion could be assumed from such a small event rate. Thus, the IMPACTO MR-PAV steering committee decided to extend the recruitment and to recalculate the sample size based on VAP incidence. Typically, the literature reports VAP incidence rates between 5% and 40%.2 25 We considered an anticipated frequency of VAP infection of 3%, margin of error of 1% and CI of 95%, according to the equation:

where

EDFF=design effect (value equal to 1 was considered)

d=confidence limits

p=hypothesised% frequency of outcome factor in the population

N=population size (value equal to 1 000 000 was adopted).

Thus, the required sample size is 1117 participants. This will be sufficient even for the assessment of diagnostic accuracy, considering that an area under the curve of 0.7 between tests, with an alpha of 5%, statistical power of 90% and the same VAP rate of 3%, would lead to a sample of 22 cases and 718 control.26 As we are facing imperfect standard tests, diagnostic accuracy will be described: (1) narratively (ie, discussing the event rate and the qualitative result with both tests); (2) considering the medical adjudication as the standard to be compared with both VAP and VAE criteria and (3) a Bayesian estimate of diagnostic accuracy parameters assuming the ANVISA PAV criterion as the perfect standard. In this last case, we will use the Modelling of Infectious Disease Centre tool,27 assuming a prior distribution of 5% for prevalence of VAP, and 5% for sensitivity and specificity for each test. We will assume 1000 burn-in iterations and 10 000 Monte Carlo iterations. To evaluate the diagnostic performance of the NHSN VAE method compared with the current PAV method for detecting PAV, sensitivity, specificity, positive predictive value, negative predictive value, positive and negative likelihood ratios will be calculated with a 95% CI.28 Data analysis will be performed using SAS 9.4 (SAS Institute, Cary, NC). All hypothesis tests will be two-sided, and a p value<0.05 will be considered statistically significant.

Patient and public involvement

Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research. As we are planning to recruit ICU patients who require mechanical ventilation, their active participation in the study will be limited, due to severity of their condition. However, we have designed this study based on an observational study of VAP adjudication using internal data from the Brazilian Ministry of Health. This study has revealed a potential discrepancy between what is formally reported to ANVISA and clinical adjudication. As a result, our current research question focuses on patient outcomes.