Trends in ICU mortality and underlying risk over three decades among mechanically ventilated patients. A group level analysis of cohorts from infection prevention studies.

BACKGROUND: Has either the underlying risk or the mortality incidence among ICU patients receiving mechanical ventilation (MV) in the literature changed in recent decades? Interpreting ICU mortality trends requires an adjusted analysis accounting for changes in underlying patient risk. METHODS: Control and intervention groups from 147 randomized concurrent control trials (RCCT) of various VAP prevention interventions, as listed primarily within 13 Cochrane reviews and 63 observational studies listed primarily within four systematic reviews. Eligible studies were those including ICU patients with > 50% of patients receiving > 24 h of MV with mortality data available. ICU mortality (censored day 21 or before) or late (after day 21) mortality together with group-mean age, and group-mean APACHE II scores were extracted from all groups. These incidences were summarized in five meta-regression models versus publication year being variously adjusted for age, APACHE II scores, type of study intervention and other group level parameters. RESULTS: Among 210 studies published between 1985 and 2021, 169 being found in systematic reviews, the increase per decade in mean mortality incidence, group-mean APACHE II scores, and group-mean age, were < 1 percentage point (p = 0.43), 1.83 (95% CI; 0.51-3.15) points, and 3.9 (95% CI; 1.1-6.7) years, respectively. Only in the model with risk adjustment for both group-mean age and group-mean APACHE II score was a significant decline in mortality apparent. In all models, the mortality incidence among concurrent control groups of decontamination studies was paradoxically five percentage points higher than benchmark and showed greater dispersion. CONCLUSION: Mortality incidence has changed little over 35 years among ICU infection prevention studies whilst the patient age and underlying disease severity, measured as APACHE II, have both increased. The paradoxically high mortality among concurrent control groups within studies of decontamination methods of infection prevention remains unaccounted for.

Establishing the safety of selective digestive decontamination within the ICU population: a bridge too far?

BACKGROUND: Infection prevention interventions within the intensive care unit (ICU) setting, whether studied within quality improvement projects or cluster randomized trials (CRT), are seen as low risk and grounded in an ethical imperative. Selective digestive decontamination (SDD) appears highly effective at preventing ICU infections within randomized concurrent control trials (RCCTs) prompting mega-CRTs with mortality as the primary endpoint. FINDINGS: Surprisingly, the summary results of RCCTs versus CRTs differ strikingly, being respectively, a 15-percentage-point versus a zero-percentage-point ICU mortality difference between control versus SDD intervention groups. Multiple other discrepancies are equally puzzling and contrary to both prior expectations and the experience within population-based studies of infection prevention interventions using vaccines. Could spillover effects from SDD conflate the RCCT control group event rate differences and represent population harm? Evidence that SDD is fundamentally safe to concurrent non-recipients in ICU populations is absent. A postulated CRT to realize this, the SDD Herd Effects Estimation Trial (SHEET), would require > 100 ICUs to achieve sufficient statistical power to find a two-percentage-point mortality spillover effect. Moreover, as a potentially harmful population-based intervention, SHEET would pose novel and insurmountable ethical issues including who is the research subject; whether informed consent is required and from whom; whether there is equipoise; the benefit versus the risk; considerations of vulnerable groups; and who should be the gatekeeper? CONCLUSION: The basis for the mortality difference between control and intervention groups of SDD studies remains unclear. Several paradoxical results are consistent with a spillover effect that would conflate the inference of benefit originating from RCCTs. Moreover, this spillover effect would constitute to herd peril.

Staphylococcus aureus hitchhiking from colonization to bacteremia via Candida within ICU infection prevention studies: a proof of concept modelling.

Whether Candida within the patient microbiome drives the pathogenesis of Staphylococcus aureus bacteremia, described as microbial hitchhiking, cannot be directly studied. Group-level observations from studies of various decontamination and non-decontamination-based ICU infection prevention interventions and studies without study interventions (observational groups) collectively enable tests of this interaction within causal models. Candidate models of the propensity for Staphylococcus aureus bacteremia to arise with versus without various antibiotic, anti-septic, and antifungal exposures, each identified as singleton exposures, were tested using generalized structural equation modelling (GSEM) techniques with Candida and Staphylococcus aureus colonization appearing as latent variables within the models. Each model was tested by confrontation against blood and respiratory isolate data, obtained from 467 groups within 284 infection prevention studies. Introducing an interaction term between Candida colonization and Staphylococcus aureus colonization substantially improved GSEM model fit. Model-derived coefficients for singular exposure to anti-septic agents (- 1.28; 95% confidence interval; - 2.05 to - 0.5), amphotericin (- 1.49; - 2.3 to - 0.67), and topical antibiotic prophylaxis (TAP; + 0.93; + 0.15 to + 1.71) as direct effects versus Candida colonization were similar in magnitude but contrary in direction. By contrast, the coefficients for singleton exposure to TAP, as with anti-septic agents, versus Staphylococcus colonization were weaker or non-significant. Topical amphotericin would be predicted to halve both candidemia and Staphylococcus aureus bacteremia incidences versus literature derived benchmarks for absolute differences of < 1 percentage point. Using ICU infection prevention data, GSEM modelling validates the postulated interaction between Candida and Staphylococcus colonization facilitating bacteremia.

Structural equation modelling the impact of antimicrobials on the human microbiome. Colonization resistance versus colonization susceptibility as case studies.

The impact of antimicrobials on the human microbiome and its relationship to human health are of great interest. How antimicrobial exposure might drive change within specific constituents of the microbiome to effect clinically relevant endpoints is difficult to study. Clinical investigation of each step within a network of causation would be challenging if done 'step-by-step'. An analytic tool of great potential to clinical microbiome research is structural equation modelling (SEM), which has a long history of applications to research questions arising within subject areas as diverse as psychology and econometrics. SEM enables postulated models based on a network of causation to be tested en bloc by confrontation with data derived from the literature. Case studies for the potential application of SEM techniques are colonization resistance (CR) and its counterpart, colonization susceptibility (CS), wherein specific microbes within the microbiome are postulated to either impede (CR) or facilitate (CS) invasive infection with pathogenic bacteria. These postulated networks have three causation steps: exposure to specific antimicrobials are key drivers, clinically relevant infection endpoints are the measurable observables and the activity of key microbiome constituents mediating CR or CS, which may be unobservable, appear as latent variables in the model. SEM methods have potential application towards evaluating the activity of specific antimicrobial agents within postulated networks of causation using clinically derived data.

Candida and the Gram-positive trio: testing the vibe in the ICU patient microbiome using structural equation modelling of literature derived data.

BACKGROUND: Whether Candida interacts with Gram-positive bacteria, such as Staphylococcus aureus, coagulase negative Staphylococci (CNS) and Enterococci, to enhance their invasive potential from the microbiome of ICU patients remains unclear. Several effective anti-septic, antibiotic, anti-fungal, and non-decontamination based interventions studied for prevention of ventilator associated pneumonia (VAP) and other ICU acquired infections among patients receiving prolonged mechanical ventilation (MV) are known to variably impact Candida colonization. The collective observations within control and intervention groups from numerous ICU infection prevention studies enables tests of these postulated microbial interactions in the clinical context. METHODS: Four candidate generalized structural equation models (GSEM), each with Staphylococcus aureus, CNS and Enterococci colonization, defined as latent variables, were confronted with blood culture and respiratory tract isolate data derived from 460 groups of ICU patients receiving prolonged MV from 283 infection prevention studies. RESULTS: Introducing interaction terms between Candida colonization and each of S aureus (coefficient + 0.40; 95% confidence interval + 0.24 to + 0.55), CNS (+ 0.68; + 0.34 to + 1.0) and Enterococcal (+ 0.56; + 0.33 to + 0.79) colonization (all as latent variables) improved the fit for each model. The magnitude and significance level of the interaction terms were similar to the positive associations between exposure to topical antibiotic prophylaxis (TAP) on Enterococcal (+ 0.51; + 0.12 to + 0.89) and Candida colonization (+ 0.98; + 0.35 to + 1.61) versus the negative association of TAP with S aureus (- 0.45; - 0.70 to - 0.20) colonization and the negative association of anti-fungal exposure and Candida colonization (- 1.41; - 1.6 to - 0.72). CONCLUSIONS: GSEM modelling of published ICU infection prevention data enables the postulated interactions between Candida and Gram-positive bacteria to be tested using clinically derived data. The optimal model implies interactions occurring in the human microbiome facilitating bacterial invasion and infection. This interaction might also account for the paradoxically high bacteremia incidences among studies of TAP in ICU patients.

How to apply structural equation modelling to infectious diseases concepts.

BACKGROUND: Structural equation modelling (SEM) can address causation questions of great interest to infectious disease physicians and infection control practitioners that would elude techniques based on tests of association. These models address questions such as the size of intervention effects mediated on entities that cannot be easily measured, questions that cannot be studied in randomized controlled trials and question arising from 'big data.' OBJECTIVES: To outline the computational and, moreover, conceptual differences between SEM methods versus the traditional tests of association. SOURCES: Google scholar search for "structural equation modelling" and "infection." CONTENT: Several examples of SEM applications to infectious diseases topics are used to illustrate. The SEM technique enables postulated causation models to be confronted with data. With this, the candidate models emerge as either 'importantly wrong' or potentially useful for enabling empiric predictions from the one identified as optimal. IMPLICATIONS: Applications of SEM techniques and related modelling techniques to infectious diseases research will likely continue to emerge, especially so with the availability of 'big data.'

Structural equation modelling the relationship between anti-fungal prophylaxis and Pseudomonas bacteremia in ICU patients.

PURPOSE: Animal models implicate candida colonization facilitating invasive bacterial infections. The clinical relevance of this microbial interaction remains undefined and difficult to study directly. Observations from studies of anti-septic, antibiotic, anti-fungal, and non-decontamination-based interventions to prevent ICU acquired infection collectively serve as a natural experiment. METHODS: Three candidate generalized structural equation models (GSEM), with Candida and Pseudomonas colonization as latent variables, were confronted with blood culture and respiratory tract isolate data derived from 464 groups from 279 studies including studies of combined antibiotic and antifungal exposures within selective digestive decontamination (SDD) interventions. RESULTS: Introducing an interaction term between Candida colonization and Pseudomonas colonization substantially improved GSEM model fit. Model derived coefficients for singular exposure to anti-septic agents (- 1.23; - 2.1 to - 0.32), amphotericin (- 1.78; - 2.79 to - 0.78) and topical antibiotic prophylaxis (TAP; + 1.02; + 0.11 to + 1.93) versus Candida colonization were similar in magnitude but contrary in direction. By contrast, the model-derived coefficients for singular exposure to TAP, as with anti-septic agents, versus Pseudomonas colonization were weaker or non-significant. Singular exposure to amphotericin would be predicted to more than halve candidemia and Pseudomonas bacteremia incidences versus literature benchmarks for absolute differences of approximately one percentage point or less. CONCLUSION: GSEM modelling of published data supports the postulated interaction between Candida and Pseudomonas colonization towards promoting bacteremia among ICU patients. This would be difficult to detect without GSEM modelling. The model indicates that anti-fungal agents have greater impact in preventing Pseudomonas bacteremia than TAP, which has no impact.

Selective digestive decontamination, a seemingly effective regimen with individual benefit or a flawed concept with population harm?

Selective digestive decontamination (SDD) regimens, variously constituted with topical antibiotic prophylaxis (TAP) and protocolized parenteral antibiotic prophylaxis (PPAP), appear highly effective for preventing ICU-acquired infections but only within randomized concurrent control trials (RCCT's). Confusingly, SDD is also a concept which, if true, implies population benefit. The SDD concept can finally be reified  in humans using the broad accumulated evidence base, including studies of TAP and PPAP that used non-concurrent controls (NCC), as a natural experiment. However, this test implicates overall population harm with higher event rates associated with SDD use within the ICU context.

Could simulation methods solve the curse of sparse data within clinical studies of antibiotic resistance?

Infectious disease (ID) physicians and ID pharmacists commonly confront therapeutic questions relating to antibiotic resistance. Randomized controlled trial data are few and meta-analytic-based approaches to develop the evidence-base from several small studies that might relate to an antibiotic resistance question are not simple. The overriding challenge is the sparsity of data which is problematic for traditional frequentist methods, being the paradigm underlying the derivation of 'P value' inferential statistics. In other sparse data contexts, simulation methods enable answers to key questions that are meaningful, quantitative and potentially relevant. How these simulation methods 'work' and how Bayesian-based methods, being not 'P value based', can facilitate simulation are reviewed. These methods are becoming increasingly accessible. This review highlights why sparse data is less of an issue within Bayesian versus frequentist paradigms. A fictional pharmacokinetic study with sparse data illustrates a simplistic application of Bayesian and simulation methods to antibiotic dosing. Whether within epidemiological projections or clinical studies, simulation methods are likely to play an increasing role in antimicrobial resistance research within both hospital and community studies of either rare infectious disease or infections within specific population groups.

Candida-Acinetobacter-Pseudomonas Interaction Modelled within 286 ICU Infection Prevention Studies.

BACKGROUND: Whether Candida interacts to enhance the invasive potential of Acinetobacter and Pseudomonas bacteria cannot be resolved within individual studies. There are several anti-septic, antibiotic, anti-fungal, and non-decontamination-based interventions to prevent ICU acquired infection. These effective prevention interventions would be expected to variably impact Candida colonization. The collective observations within control and intervention groups from numerous ICU infection prevention studies simulates a multi-centre natural experiment with which to evaluate Candida, Acinetobacter and Pseudomonas interaction (CAPI). METHODS: Eight Candidate-generalized structural equation models (GSEM), with Candida, Pseudomonas and Acinetobacter colonization as latent variables, were confronted with blood culture and respiratory tract isolate data derived from >400 groups derived from 286 infection prevention studies. RESULTS: Introducing an interaction term between Candida colonization and each of Pseudomonas and Acinetobacter colonization improved model fit in each case. The size of the coefficients (and 95% confidence intervals) for these interaction terms in the optimal Pseudomonas (+0.33; 0.22 to 0.45) and Acinetobacter models (+0.32; 0.01 to 0.5) were similar to each other and similar in magnitude, but contrary in direction, to the coefficient for exposure to topical antibiotic prophylaxis (TAP) on Pseudomonas colonization (-0.45; -0.71 to -0.2). The coefficient for exposure to topical antibiotic prophylaxis on Acinetobacter colonization was not significant. CONCLUSIONS: GSEM modelling of published ICU infection prevention data supports the CAPI concept. The CAPI model could account for some paradoxically high Acinetobacter and Pseudomonas infection incidences, most apparent among the concurrent control groups of TAP studies.

Estimated Treatment Effects of Tight Glycaemic Targets in Mild Gestational Diabetes Mellitus: A Multiple Cut-Off Regression Discontinuity Study Design.

Background: We investigated the treatment effects of tight glycaemic targets in a population universally screened according to the International Association of Diabetes and Pregnant Study Groups (IADPSG)/World Health Organisation (WHO) gestational diabetes mellitus (GDM) guidelines. As yet there, have been no randomized control trials evaluating the effectiveness of treatment of mild GDM diagnosed under the IADPSG/WHO diagnostic thresholds. We hypothesize that tight glycaemic control in pregnant women diagnosed with GDM will result in similar clinical outcomes to women just below the diagnostic thresholds. Methods: A multiple cut-off regression discontinuity study design in a retrospective observational cohort undergoing oral glucose tolerance tests (OGTT) (n = 1178). Treatment targets for women with GDM were: fasting capillary blood glucose (CBG) of ≤5.0 mmol/L and the 2-h post-prandial CBG of ≤6.7 mmol/L. Regression discontinuity study designs estimate treatment effects by comparing outcomes between a treated group to a counterfactual group just below the diagnostic thresholds with the assumption that covariates are similar. The counterfactual group was selected based on a composite score based on OGTT plasma glucose categories. Results: Women treated for GDM had lower rates of newborns large for gestational age (LGA), 4.6% versus those just below diagnostic thresholds 12.6%, relative risk 0.37 (95% CI, 0.16-0.85); and reduced caesarean section rates, 32.2% versus 43.0%, relative risk 0.75 (95% CI, 0.56-1.01). This was at the expense of increases in induced deliveries, 61.8% versus 39.3%, relative risk 1.57 (95% CI, 1.18-1.9); notations of neonatal hypoglycaemia, 15.8% versus 5.9%, relative risk 2.66 (95% CI, 1.23-5.73); and high insulin usage 61.1%. The subgroup analysis suggested that treatment of women with GDM with BMI ≥30 kg/m2 drove the reduction in caesarean section rates: 32.9% versus 55.9%, relative risk 0.59 (95%CI, 0.4-0.87). Linear regression interaction term effects between non-GDM and treated GDM were significant for LGA newborns (p = 0.001) and caesarean sections (p = 0.015). Conclusions: Tight glycaemic targets reduced rates of LGA newborns and caesarean sections compared to a counterfactual group just below the diagnostic thresholds albeit at the expense of increased rates of neonatal hypoglycaemia, induced deliveries, and high insulin usage.

Structural equation modeling the "control of gut overgrowth" in the prevention of ICU-acquired Gram-negative infection.

BACKGROUND: Conceptually, the "control of gut overgrowth" (COGO) is key in mediating prevention against infection with Gram-negative bacilli by topical antibiotic prophylaxis, a common constituent of selective digestive decontamination (SDD) regimens. However, the relative importance of the other SDD components, enteral and protocolized parenteral antibiotic prophylaxis, versus other methods of infection prevention and versus other contextual exposures cannot be resolved within individual studies. METHODS: Seven candidate generalized structural equation models founded on COGO concepts were confronted with Pseudomonas and Acinetobacter bacteremia as well as ventilator-associated pneumonia data derived from > 200 infection prevention studies. The following group-level exposures were included in the models: use and mode of antibiotic prophylaxis, anti-septic and non-decontamination methods of infection prevention; proportion receiving mechanical ventilation; trauma ICU; mean length of ICU stay; and concurrency versus non-concurrency of topical antibiotic prophylaxis study control groups. RESULTS: In modeling Pseudomonas and Acinetobacter gut overgrowth as latent variables, anti-septic interventions had the strongest negative effect against Pseudomonas gut overgrowth but no intervention was significantly negative against Acinetobacter gut overgrowth. Strikingly, protocolized parenteral antibiotic prophylaxis and concurrency each have positive effects in the model, enteral antibiotic prophylaxis is neutral, and Acinetobacter bacteremia incidences are high within topical antibiotic prophylaxis studies, moreso with protocolized parenteral antibiotic prophylaxis exposure. Paradoxically, topical antibiotic prophylaxis (moreso with protocolized parenteral antibiotic prophylaxis) appears to provide the strongest summary prevention effects against overall bacteremia and overall VAP. CONCLUSIONS: Structural equation modeling of published Gram-negative bacillus infection data enables a test of the COGO concept. Paradoxically, Acinetobacter and Pseudomonas bacteremia incidences are unusually high among studies of topical antibiotic prophylaxis.

Discrepancies in Control Group Mortality Rates Within Studies Assessing Topical Antibiotic Strategies to Prevent Ventilator-Associated Pneumonia: An Umbrella Review.

OBJECTIVES: To test the postulate that concurrent control patients within ICUs studying topical oropharyngeal antibiotics to prevent ventilator-associated pneumonia and mortality would experience spillover effects from the intervention. DATA SOURCES: Studies cited in 15 systematic reviews of various topical antibiotic and other infection prevention interventions among ICU patients. STUDY SELECTION: Studies of topical antibiotics, stratified into concurrent control versus nonconcurrent control designs. Studies of nondecontamination-based infection prevention interventions provide additional points of reference. Studies with no infection prevention intervention provide the mortality benchmark. Data from additional studies and data reported as intention to treat were used within sensitivity tests. DATA EXTRACTION: Mortality incidence proportion data, mortality census, study characteristics, group mean age, ICU type, and study publication year. DATA SYNTHESIS: Two-hundred six studies were included. The summary effect sizes for ventilator-associated pneumonia and mortality prevention derived in the 15 systematic reviews were replicated. The mean ICU mortality incidence for concurrent control groups of topical antibiotic studies (28.5%; 95% CI, 25.0-32.3; n = 41) is higher versus the benchmark (23.7%; 19.2-28.5%; n = 34), versus nonconcurrent control groups (23.5%; 19.3-28.3; n = 14), and versus intervention groups (24.4%; 22.1-26.9; n = 62) of topical antibiotic studies. In meta-regression models adjusted for group-level characteristics such as group mean age and publication year, concurrent control group membership within a topical antibiotic study remains associated with higher mortality (p = 0.027), whereas other group memberships, including membership within an antiseptic study, are each neutral (p = not significant). CONCLUSIONS: Within topical antibiotic studies, the concurrent control group mortality incidence proportions are inexplicably high, whereas the intervention group mortality proportions are paradoxically similar to a literature-derived benchmark. The unexplained ventilator-associated pneumonia and mortality excess in the concurrent control groups implicates spillover effects within studies of topical antibiotics. The apparent ventilator-associated pneumonia and mortality prevention effects require cautious interpretation.

How the Cluster-randomized Trial "Works".

Cluster-randomized trials (CRTs) are able to address research questions that randomized controlled trials (RCTs) of individual patients cannot answer. Of great interest for infectious disease physicians and infection control practitioners are research questions relating to the impact of interventions on infectious disease dynamics at the whole-of-population level. However, there are important conceptual differences between CRTs and RCTs relating to design, analysis, and inference. These differences can be illustrated by the adage "peas in a pod." Does the question of interest relate to the "peas" (the individual patients) or the "pods" (the clusters)? Several examples of recent CRTs of community and intensive care unit infection prevention interventions are used to illustrate these key concepts. Examples of differences between the results of RCTs and CRTs on the same topic are given.