Advancing the science of ventilator-associated pneumonia surveillance

The landmark Study on the Efficacy of Nosocomial Infection Control definitively demonstrated that infection surveillance and control programs prevent hospital-acquired infections. The rise of public reporting, benchmarking, and pay for performance movements, however, has considerably changed the infection surveillance landscape in the 27 years since this study was published. Clinically nuanced surveillance definitions that served the profession well for many years have fallen into disfavor because their complexity and subjectivity allow for conscious and subconscious gaming. These limitations make it very difficult to determine whether changes in surveillance rates represent true changes in disease incidence or artifacts of definition subjectivity, external reporting pressures, and internal biases. Surveillance definitions need to be revised to enhance objectivity and to ensure that they detect clinically meaningful events associated with compromised outcomes. The US Centers for Disease Control and Prevention recently released modified definitions for ventilator-associated events that have the potential to make safety surveillance for ventilated patients more credible and useful once again.

Th e report by Bénet and colleagues in the previous issue of Critical Care is at once a welcome reminder of the pivotal importance of infection surveillance programs and a window into current weaknesses in surveillance science that threaten the credibility and utility of infection control programs [1]. Bénet and colleagues assessed the impact of a one-year lapse in surveillance in a medical-surgical ICU (surveillance stopped because the infection preventionist charged with supporting the unit became burned out -evidence, perhaps, of the underresourcing of infection control programs). Ventilatorassociated pneumonia (VAP) rates rose from 13.4 VAP cases/1,000 ventilator-days in the 3 years prior to surveillance interruption to 22.9 VAP cases/1,000 ventilatordays in the 3 years following surveillance interruption. In addition, the average duration of mechanical ventilation rose from 7.7 days to 11.3 days (P = 0.007) and hospital mortality rose from 13.5% to 18.8% (P = 0.028). VAP rates and mortality rates were stable, by contrast, in a parallel medical unit that enjoyed continuous surveillance and infection control support throughout the study period.
Bénet and colleagues' fi ndings echo the seminal conclusions of the Study of the Effi cacy of Nosocomial Infection Control (SENIC) published by the Centers for Disease Control and Prevention in 1985 [2]. Th e SENIC project took advantage of a natural experiment: US hospitals began establishing infection control programs of varying intensities starting in the early 1970s. SENIC investigators reviewed approximately 338,000 patient charts from 338 representative US hospitals and compared hospital-acquired infection rates in 1970 versus those in 1976. Th e investigators also characterized the intensity of surveillance and infection prevention activities in each hospital and then measured the association of each with changes in infection rates [3]. After adjusting for changes in population risk profi les, average lengths of stay, nurse-to-patient ratios, and other factors, SENIC investigators found that high-intensity infection surveillance and prevention programs reduced hospital infection rates by 32% while infection rates rose by 18% in hospitals without infection control programs [2].
Th e SENIC project cost $27 million and took over a decade to complete but it defi nitively established the importance of infection surveillance and control pro grams. Th e infection surveillance and reporting land scape has changed considerably, however, in the 27 years since SENIC was published. Public reporting, benchmarking, and pay for performance movements have focused considerable attention upon hospitals' VAP rates and thereby generated powerful incentives for hospitals to make their

Abstract
The landmark Study on the Effi cacy of Nosocomial Infection Control defi nitively demonstrated that infection surveillance and control programs prevent hospital-acquired infections. The rise of public reporting, benchmarking, and pay for performance movements, however, has considerably changed the infection surveillance landscape in the 27 years since this study was published. Clinically nuanced surveillance defi nitions that served the profession well for many years have fallen into disfavor because their complexity and subjectivity allow for conscious and subconscious gaming. These limitations make it very diffi cult to determine whether changes in surveillance rates represent true changes in disease incidence or artifacts of defi nition subjectivity, external reporting pressures, and internal biases. Surveillance defi nitions need to be revised to enhance objectivity and to ensure that they detect clinically meaningful events associated with compromised outcomes. The US Centers for Disease Control and Prevention recently released modifi ed defi nitions for ventilator-associated events that have the potential to make safety surveillance for ventilated patients more credible and useful once again.
rates as low as possible [4]. Close evaluation of Bénet and colleagues' study helps illuminate current weaknesses in VAP surveillance science that are begin ning to be addressed by the Centers for Disease Control and Prevention.
Bénet and colleagues made a valiant attempt to adjust for changes in patient mix and severity of illness over time, including factors such as age, gender, immunosuppression, antibiotics on ICU admission, and SAPS II score on ICU admission. Nonetheless, the dramatic rise in patients' average duration of mechanical ventilation and mortality following surveillance interruption raises the possibility of important residual diff erences in patient characteristics that may better account for the observed jump in VAP rates than surveillance interruption alone. Higher VAP rates alone are probably not suffi cient to account for the 3.6-day increase in average duration of mechanical ventilation and the 39% increase in hospital mortality following surveillance interruption since the relative rarity of VAP tends to limit its impact on population-level outcomes. Bénet and colleagues observed a net increase of 9.3 VAP cases/100 patients: the attributable duration of mechanical ventilation for VAP would have to be about 40 days to increase the average duration of mechanical ventilation for the entire population by the observed 3.6 days. Th is prediction is almost eight times the additive duration of mechanical ventilation typically associated with VAP [5]. Alternatively or additionally, surveil lance interruption may have led to a drop in the fre quency of preventive care procedures aff ecting all patients, such as daily sedative interruptions and spon taneous breathing trials that could account for the observed changes in population-level outcomes as well as the increase in VAP rates [6]. Unfortunately, however, Bénet and colleagues did not present data on preventive care patterns so it is not possible to disentangle the relative contributions of changes in case mix, care patterns, and complication rates towards their patients' ultimate outcomes.
Bénet and colleagues' observed rise in VAP rates following surveillance interruption is also complicated by the strong possibility of observer bias. VAP surveillance is highly subjective. Th e study VAP defi nition included criteria such as 'lung infi ltrates' , 'sputum modifi cation' , 'suggestive auscultation' , 'low oxyhaemoglobin saturation' , and 'increased pulmonary oxygen consumption' . VAP defi nitions typically do not include concrete guidance on interpreting these criteria; rather, surveyors apply the criteria using their own discretion. Given that this was a retrospective study and infection surveyors were not blinded to unit identities, surveyors' conscious or subconscious expecta tions regarding VAP rates in a unit that foreswore infec tion surveillance activities for a year may have infl uenced their interpretations. Furthermore, the surveyor who resumed surveillance in the target unit following inter ruption was not the same person res ponsible for surveil lance prior to interruption. Diff erent observers' VAP assessments in a common population can vary by a factor of two or more [7]. We cannot know how much the increase in VAP following surveillance interruption was due to the change in surveillance personnel and diff erent subjective assess ments (perhaps infl uenced by sub con scious expectations) versus a bona fi de rise in invasive pneumonias.
As Bénet and colleagues' study demonstrates, the problems of patient-mix adjustment and observer bias threaten to undermine the credibility of infection surveillance reports [8,9]. Cognizant of these concerns, the Centers for Disease Control and Prevention recently proposed a new surveillance paradigm for ventilatorasso ciated events that is designed to make surveillance objec tive, more reproducible, and potentially auto matable in hospitals with comprehensive electronic health records [10,11]. Th e new paradigm shifts the emphasis of surveillance from pneumonia in particular to ventilatorassociated complications (VACs) in general. VACs are defi ned quantitatively on the basis of sustained increases in daily minimum positive end-expiratory pressure or fraction of inspired oxygen after at least two calendar days of stable or improving settings. Early work has shown VAC to be an independent predictor of increased length of stay and increased hospital mortality [12,13]. Subdefi nitions seek to classify VACs into infectionrelated events and possible and probable pneumonias on the basis of temper ature changes, white blood cell counts, antibiotic prescribing, quantity of neutrophils on pulmonary speci men Gram stains, and culture results. Th e new surveil lance defi nitions do not yet incorporate criteria for case-mix adjustments but this has been identifi ed as an area for active research in the future. For the present, however, these defi nitions have the potential to make surveillance more objective and effi cient.
Bénet and colleagues' study is a valuable reminder of the potential risk associated with under-resourcing infection surveillance and control programs. Better surveillance defi nitions and better case-mix adjustment protocols will help to enhance the credibility and impact of their important message.