Causal inference using observational intensive care unit data: a scoping review and recommendations for future practice.

This scoping review focuses on the essential role of models for causal inference in shaping actionable artificial intelligence (AI) designed to aid clinicians in decision-making. The objective was to identify and evaluate the reporting quality of studies introducing models for causal inference in intensive care units (ICUs), and to provide recommendations to improve the future landscape of research practices in this domain. To achieve this, we searched various databases including Embase, MEDLINE ALL, Web of Science Core Collection, Google Scholar, medRxiv, bioRxiv, arXiv, and the ACM Digital Library. Studies involving models for causal inference addressing time-varying treatments in the adult ICU were reviewed. Data extraction encompassed the study settings and methodologies applied. Furthermore, we assessed reporting quality of target trial components (i.e., eligibility criteria, treatment strategies, follow-up period, outcome, and analysis plan) and main causal assumptions (i.e., conditional exchangeability, positivity, and consistency). Among the 2184 titles screened, 79 studies met the inclusion criteria. The methodologies used were G methods (61%) and reinforcement learning methods (39%). Studies considered both static (51%) and dynamic treatment regimes (49%). Only 30 (38%) of the studies reported all five target trial components, and only seven (9%) studies mentioned all three causal assumptions. To achieve actionable AI in the ICU, we advocate careful consideration of the causal question of interest, describing this research question as a target trial emulation, usage of appropriate causal inference methods, and acknowledgement (and examination of potential violations of) the causal assumptions.

Return to work and health-related quality of life up to 1 year in patients hospitalized for COVID-19: the CO-FLOW study.

BACKGROUND: Currently, evidence about the long-term consequences of COVID-19 on return to work and health-related quality of life (HRQoL) is limited. We evaluated return to work and its associations with baseline characteristics and physical and mental recovery over time in patients up to 1 year after hospitalization for COVID-19. Secondly, we aimed to evaluate the association between return to work and health-related quality of life (HRQoL). METHODS: CO-FLOW, a multicenter prospective cohort study, enrolled adult participants hospitalized for COVID-19, aged ≥ 18 years within 6 months after hospital discharge. Return to work and HRQoL were collected at 3, 6, and 12 months after hospital discharge using the iMTA Productivity Cost Questionnaire and the 36-Item Short Form Health Survey, respectively. Data were collected between July 1, 2020, and September 1, 2022. Generalized estimating equations with repeated measurements were used to assess outcomes over time. RESULTS: In the CO-FLOW study, 371 participants were employed pre-hospitalization. At 3, 6, and 12 months post-discharge, 50% (170/342), 29% (92/317), and 15% (44/295) of participants had not returned to work, and 21% (71/342), 21% (65/317), and 16% (48/295) only partially, respectively. ICU admission (adjusted odds ratio (95% confidence interval): 0.17 (0.10 to 0.30), p < 0.001), persistent fatigue (0.93 (0.90 to 0.97), p < 0.001), female sex (0.57 (0.36 to 0.90), p = 0.017), and older age (0.96 (0.93 to 0.98), p < 0.001) were independently associated with no return to work. ICU patients required a longer time to return to work than non-ICU patients. Patients who did not return or partially returned to work reported lower scores on all domains of HRQoL than those who fully returned. CONCLUSIONS: One year after hospitalization for COVID-19, only 69% of patients fully returned to work, whereas 15% did not return and 16% partially returned to work. No or partial return to work was associated with reduced HRQoL. This study suggests that long-term vocational support might be needed to facilitate return to work. TRIAL REGISTRATION: World Health Organization International Clinical Trials Registry Platform NL8710.

Dynamic prediction of mortality in COVID-19 patients in the intensive care unit: A retrospective multi-center cohort study.

Background: The COVID-19 pandemic continues to overwhelm intensive care units (ICUs) worldwide, and improved prediction of mortality among COVID-19 patients could assist decision making in the ICU setting. In this work, we report on the development and validation of a dynamic mortality model specifically for critically ill COVID-19 patients and discuss its potential utility in the ICU. Methods: We collected electronic medical record (EMR) data from 3222 ICU admissions with a COVID-19 infection from 25 different ICUs in the Netherlands. We extracted daily observations of each patient and fitted both a linear (logistic regression) and non-linear (random forest) model to predict mortality within 24 h from the moment of prediction. Isotonic regression was used to re-calibrate the predictions of the fitted models. We evaluated the models in a leave-one-ICU-out (LOIO) cross-validation procedure. Results: The logistic regression and random forest model yielded an area under the receiver operating characteristic curve of 0.87 [0.85; 0.88] and 0.86 [0.84; 0.88], respectively. The recalibrated model predictions showed a calibration intercept of -0.04 [-0.12; 0.04] and slope of 0.90 [0.85; 0.95] for logistic regression model and a calibration intercept of -0.19 [-0.27; -0.10] and slope of 0.89 [0.84; 0.94] for the random forest model. Discussion: We presented a model for dynamic mortality prediction, specifically for critically ill COVID-19 patients, which predicts near-term mortality rather than in-ICU mortality. The potential clinical utility of dynamic mortality models such as benchmarking, improving resource allocation and informing family members, as well as the development of models with more causal structure, should be topics for future research.

Nutritional intake and gastro-intestinal symptoms in critically ill COVID-19 patients.

BACKGROUND & AIMS: Critically ill COVID-19 patients seem hypermetabolic and difficult to feed enterally, due to gastro-intestinal (GI) symptoms such as high gastric residual volumes (GRV) and diarrhea. Our aim was to describe the association of nutritional intake and GI symptoms during first 14 days of ICU admission. METHODS: Observational study including critically ill adult COVID-19 patients. Data on nutritional intake [enteral nutrition (EN) or parenteral nutrition] and GI symptoms were collected during 14 days after ICU admission. Target energy and protein feeding goals were calculated conform ESPEN guidelines. GI symptoms included GRV (ml/d), vomiting, abdominal distension, and faeces (ml/d). High GRV's were classified as ≥2 times ≥150 ml/d and diarrhea as Bristol stool chart ≥6. GI symptoms were defined as mild if at least one symptom occurred and as moderate when ≥2 symptoms occurred. Acute gastrointestinal injury (AGI) grades of III were classified as GI dysfunction and grades of IV were considered as GI failure with severe impact on distant organs. Linear mixed model analysis was performed to explore the development of nutritional intake and GI symptoms over time at day (D) 0, 4, 10, and 14. RESULTS: One hundred and fifty patients were included [75% male; median age 64 years (IQR 54-70)]. BMI upon admission was 28 kg/m2 (IQR 25-33), of which 43% obese (BMI > 30 kg/m2). Most patients received EN during admission (98% D4; 96% D10-14). Mean energy goals increased from 87% at D4 to 93% D10-14 and protein goals (g/kg) were increasingly achieved during admission (84% D4; 93% D10-14). Presence of moderate GI symptoms decreased (10% D0; 6% D4-10; 5% D14), reversely mild GI symptoms increased. Occurrence of GI dysfunction fluctuated (1% D0; 18% D4; 12% D10; 8% D14) and none of patients developed grade IV GI failure. Development of high GRV fluctuated (5% D0; 23% D4; 14% D10; 8% D14) and occurrence of diarrhea slightly increased during admission (5% D0; 22% D4; 25% D10; 27% D14). Linear mixed models showed only an association between AGI grades III and lower protein intake at day 10 (p = 0.020). CONCLUSION: Occurrence of GI symptoms was limited and seems no major barrier for EN in our group of critically COVID-19 patients. Nutritional intake was just below requirements during the first 14 days of ICU admission. The effect on nutritional status remains to be studied.

Energy expenditure and feeding practices and tolerance during the acute and late phase of critically ill COVID-19 patients.

BACKGROUND & AIMS: Different metabolic phases can be distinguished in critical illness, which influences nutritional treatment. Achieving optimal nutritional treatment during these phases in critically ill patients is challenging. COVID-19 patients seem particularly difficult to feed due to gastrointestinal problems. Our aim was to describe measured resting energy expenditure (mREE) and feeding practices and tolerance during the acute and late phases of critical illness in COVID-19 patients. METHODS: Observational study including critically ill mechanically ventilated adult COVID-19 patients. Indirect calorimetry (Q-NRG+, Cosmed) was used to determine mREE during the acute (day 0-7) and late phase (>day 7) of critical illness. Data on nutritional intake, feeding tolerance and urinary nitrogen loss were collected simultaneously. A paired sample t-test was performed for mREE in both phases. RESULTS: We enrolled 21 patients with a median age of 59 years [44-66], 67% male and median BMI of 31.5 kg/m2 [25.7-37.8]. Patients were predominantly fed with EN in both phases. No significant difference in mREE was observed between phases (p = 0.529). Sixty-five percent of the patients were hypermetabolic in both phases. Median delivery of energy as percentage of mREE was higher in the late phase (94%) compared to the acute phase (70%) (p = 0.001). Urinary nitrogen losses were significant higher in the late phase (p = 0.003). CONCLUSION: In both the acute and late phase, the majority of the patients were hypermetabolic and fed enterally. In the acute phase patients were fed hypocaloric whereas in the late phase this was almost normocaloric, conform ESPEN guidelines. No significant difference in mREE was observed between phases. Hypermetabolism in both phases in conjunction with an increasing loss of urinary nitrogen may indicate that COVID-19 patients remain in a prolonged acute, catabolic phase.

Pharmacist interventions during patient rounds in two intensive care units: Clinical and financial impact.

INTRODUCTION: The risk of prescribing errors and related adverse drug events (ADE) on the intensive care unit (ICU) is high. Based on studies carried out in North America or the UK, a clinical pharmacy service can reduce ADEs and lower overall costs. This study looks into the clinical and financial impact of interventions made by pharmacists during patient rounds in two ICU settings in the Netherlands. MATERIALS AND METHODS: A quality improvement study was performed in a general teaching hospital (GTH) and a university hospital (UH) in the Netherlands. The improvement consisted of a review of medication orders and participation in patient rounds by an ICU-trained pharmacist. The main outcome measure was the proportion of accepted pharmacist interventions. Secondary outcome measures were the clinical relevance of the accepted interventions, the proportion of prevented potential ADEs (pADE) and a cost-benefit ratio. RESULTS: In the GTH 160 patients and in the UH 174 patients were included. A total of 332 and 280 interventions were analysed. Acceptance of the interventions was 67.3% in the GTH and 61.8% in the UH. The accepted interventions were mostly scored as clinically relevant, resulting in 0.16 and 0.11 prevented pADEs per patient. The cost benefit was €119 (GTH) and €136 (UH) per accepted intervention. CONCLUSION: This clinical pharmacy service in two ICUs resulted in high numbers of accepted and clinically relevant interventions. Our model appeared to be cost-effective in both ICU settings.

The value of arterial pressure waveform cardiac output measurements in the radial and femoral artery in major cardiac surgery patients.

BACKGROUND: A relatively new uncalibrated arterial pressure waveform cardiac output (CO) measurement technique is the Pulsioflex-ProAQT® system. Aim of this study was to validate this system in cardiac surgery patients with a specific focus on the evaluation of a difference in the radial versus the femoral arterial access, the value of the auto-calibration modus and the ability to show fluid-induced changes. METHODS: In twenty-five patients scheduled for ascending aorta, aortic arch replacement, or both we measured CO simultaneously by transpulmonary thermodilution (COtd) and by using the ProAQT® system connected to the radial (COpR), as well as the femoral artery catheter (COpF). Hemodynamic data were assessed at predefined time points; from incision until 16 h after ICU admission. RESULTS: In total 175 (radial) and 179 (femoral) pairs of CO measurement were collected. The accuracy of COpR/COpF was evaluated showing a mean bias of -0.31 L/min (±2.9 L/min) and -0.57 L/min (± 2.8 L/min) with percentage errors of 49 and 46% respectively. Trending ability of the ProAQT® device was evaluated; the four quadrant concordance rates in the radial and femoral artery were 74 and 75% and improved to 77 and 85% after auto-calibration. The mean angular biases in the radial and femoral artery were 6.4° and 6.0° and improved to 5° and 3.3° after auto-calibration. The polar concordance rates in the radial and femoral artery were 65 and 70% and improved to 76 and 84% after auto-calibration. Considering the fluid-induced changes in stroke volume(SV), the coefficient of correlation between the changes in SVtd and SVp was 0.57 (p < 0.01) in the radial artery and 0.60 (p < 0.01) in the femoral artery. CONCLUSIONS: The ProAQT® system can be of additional value if the clinician wants to determine fluid responsiveness in cardiac surgery patients. However, the ProAQT® system provided inaccurate CO measurements compared to transpulmonary thermodilution. The trending ability was poor for COpR but moderate for COpF. Auto-calibration of the system did not improve accuracy of CO measurements nor did it improve the prediction of fluid responsiveness. However, the trending ability was improved by auto-calibration, possibly by correcting a drift over a longer time period.

Experimental validation of frequently-used echocardiographic right-ventricular impedance parameters.

BACKGROUND: Aim of the study was to validate commonly used bedside right-ventricular (RV) impedance parameters, which are utilized in determining heart-lung interactions during mechanical ventilation. METHODS: Fifteen pigs were equally assigned to either an open or a closed pericardium group. In all animals, an inflatable vascular occluder and a flow probe were placed around the main pulmonary artery, which allowed for a gradual increase in pulmonary vascular impedance with banding of the pulmonary artery. A median sternotomy was performed for the open pericardium group, and a lateral thoracotomy was performed for the closed pericardium group. RESULTS: In the open pericardium group, mean acceleration time (ACmean) and the slope of the pulmonary artery flow correlated significantly with Poiseuille resistance over the banding (r=0.67 and r=0.65, respectively). In the closed pericardium group, the ratio of the right to left ventricular area, eccentricity index, and tricuspid annular plane systolic excursion did not correlate with resistance over the banding, only the ACmean showed a significant correlation with resistance over the banding (r=0.88). CONCLUSION: ACmean is a reliable parameter of RV impedance that can be used to study the heart-lung interactions during mechanical ventilation.

Propofol injection pain: is it still an issue? The effect of premedication.

BACKGROUND: The use of propofol often results in pain upon injection. Various strategies can be used to reduce this pain, ranging from the administration of analgesics to modification of the propofol emulsion. However, basic premedication protocol aimed at peri- and postoperative pain reduction could also sufficiently reduce propofol injection pain, rendering other special interventions redundant. METHODS: With the approval of the Local Ethics Committee of the Erasmus Medical Centre Rotterdam, and after obtaining written informed consent from each patient included, 209 surgical patients (ASA physical status I-III) were randomized in a double-blind manner to receive premedication consisting of 7.5 mg midazolam, 50 mg diclofenac and 1000 mg acetaminophen (all orally administered) and fentanyl (intravenously administered) or placebo medication the hour before surgery. In both groups a mixture of 40 mL propofol 1% with 2 mL lidocaine 1% was used to induce anesthesia. Pain scores were assessed using a verbal analog scale (VAS) ranging from 0-10. RESULTS: The premedication group was found to have significantly less pain upon injection of propofol (median VAS 0+/-0-2) (median+/-interquartile range) when compared to the control group (median VAS 1.5+/-0-4; P=0.001). In addition, more patients in the premedication group experienced no pain at all. This effect was still present one hour after extubation. CONCLUSION: Even when injection pain is reduced, the use of a premedication regimen clearly has additional value with respect to the patients' experience. Considering the low VAS scores observed overall, it might be worthwhile to reserve additional injection pain-reducing interventions for individual patients rather than adding them as a component of standard practice.

Pulmonary morbidity following esophagectomy is decreased after introduction of a multimodal anesthetic regimen.

Respiratory morbidity is the most frequent complication after esophagectomy, which can occur in 50% of the patients treated for esophageal cancer. We tested the hypothesis whether an anesthetic regimen, emphasizing intraoperative fluid restriction and early extubation could, positively influence postoperative morbidity, without affecting the gastric tube reconstruction. We introduced an anesthetic regimen, based on early extubation and a controlled intraoperative fluid management (net fluid balance < 4 L) in combination with the use of norepinephrine to maintain mean arterial blood pressure > 65 mmHg. Postoperative morbidity and mortality were compared with a similar group of patients operated one year before. From June 2005 till September 2006, 83 patients were treated according to the new regimen (NR) and compared to a similar number of patients from the same period in 2003-2005 (standard regimen: SR). Applying the NR resulted in significantly less fluid administration (balance of 3.5 +/- 0.2 L NR vs. 5.1 +/- 0.2 L SR, p < 0.05) resulting in fewer patients developing pneumonia (26% in the NR group vs. 42% in the SR group, p < 0.05). Similar per operative blood loss and urine output and occurrence of leakage or ischemia of the gastric tube anastomosis occurred in both groups. Respiratory morbidity is significantly reduced with the introduction of a new anesthetic regimen directed at intraoperative fluid restriction and early extubation, without increasing anastomotic leakage of the gastric tube reconstruction.

The NO donor SIN-1 improves intestinal-arterial P(CO(2)) gap in experimental endotoxemia: an animal study.

BACKGROUND: Dysfunction of the microcirculation is a prominent feature of sepsis and endotoxemia. Recently, it has been shown that microcirculatory alterations are completely reversed by local or systemic application of vasodilators in severely septic patients. Therefore, we investigated the influence of vasodilator therapy on microcirculatory dysfunction of the ileum during endotoxic shock in a prospective, controlled animal study. METHODS: After baseline measurements, shock was induced in 12 domestic pigs by lipopolysaccharide via the mesenteric vein until the mean arterial pressure fell below 60 mmHg. After 30 min in shock, six animals were resuscitated with either fluid alone (control) or fluid and 2 microg/kg/min of the vasodilator 3-morpholino-sydnonimine (SIN-1). The systemic and regional hemodynamics and oxygenation parameters, tonometric ileal P(CO(2)) and microvascular oxygen pressures (muP(O(2))) (by oxygen-dependent Pd-porphyrin phosphorescence) were measured simultaneously. RESULTS: The ileal-arterial P(CO(2)) gap increased during shock and the ileal mucosal and serosal muP(O(2)) decreased concurrently. SIN-1 in addition to fluid resuscitation significantly improved the ileal-arterial P(CO(2)), whereas fluid alone failed to decrease the P(CO(2)) gap. The SIN-1-induced improvement in the P(CO(2)) gap was accompanied by an increase in serosal muP(O(2)) above shock levels. Mucosal muP(O(2)) was resuscitated to baseline levels in both groups. CONCLUSION: The application of the vasodilator SIN-1 in addition to fluid resuscitation improves the ileal-arterial P(CO(2)) gap and mucosal muP(O(2)), together with a moderate increase in serosal muP(O(2)), after endotoxic shock. This finding is consistent with the concept that vasodilators may correct pathologic flow distribution within the intestinal wall.

The effect of the transfusion of stored RBCs on intestinal microvascular oxygenation in the rat.

BACKGROUND: Although it is known that the transfusion of stored RBCs does not always improve tissue O(2) consumption under conditions of limited tissue oxygenation, the efficiency of O(2) delivery to the microcirculation by stored RBCs has never been determined. STUDY DESIGN AND METHODS: In a rat hemorrhagic shock model, the effects of resuscitation with fresh or 28-day-old RBCs stored in CPD plasma, saline-adenine-glucose-mannitol, and CPDA-1 plasma were investigated. Systemic hemodynamic and intestinal oxygenation measures were monitored. Intestinal microvascular PO(2) was determined with the O(2)-dependent quenching of palladium-porphyrin phosphorescence, and the RBC deformability was measured with a Laser-assisted optic rotational cell analyzer. RESULTS: Hemodynamic and oxygenation measures were significantly decreased during hemorrhagic shock. Intestinal oxygen consumption and mesenteric venous pO(2) were restored with the transfusion of both fresh and stored RBCs, except for CPD-stored RBCs. The intestinal microvascular pO(2) improved only with the transfusion of fresh RBCs. Deformability of the stored RBCs was significantly decreased. CONCLUSION: In contrast to that of fresh RBCs, the transfusion of stored RBCs did not restore the microcirculatory oxygenation, possibly because of impaired O(2) unloading, but, except for CPD-stored RBCs, the storage-induced changes were not enough to impair intestinal VO(2) and mesenteric venous pO(2).

Preservation of intestinal microvascular Po2 during normovolemic hemodilution in a rat model.

The effect of hemodilution on the intestinal microcirculatory oxygenation is not clear. The aim of this study was to determine the effect of moderate normovolemic hemodilution on intestinal microvascular partial oxygen pressure (Po2) and its relation to the mesenteric venous Po2 (Pmvo2). Normovolemic hemodilution was performed in 13 anesthetized male Wistar rats. Systemic hemodynamic and intestinal oxygenation parameters were monitored. Intestinal microvascular Po2 was measured by using the oxygen-dependent quenching of palladium-porphyrin phosphorescence. Hemodilution decreased systemic hematocrit from 45.0% +/- 0.1% (average +/- SEM) to 24.6% +/- 1.6%. The mesenteric blood flow did not change from baseline values, resulting in a linear decrease in intestinal oxygen delivery (from 2.77 +/- 0.15 to 1.42 +/- 0.11 mLxkg(-1)xmin(-1)). The intestinal oxygen extraction ratio increased significantly from 24% +/- 1% to 42% +/- 4%. Pmvo2 decreased significantly (from 57 +/- 2 to 41 +/- 2 mm Hg), but intestinal oxygen consumption and microvascular Po2 remained unaffected. As a result, the difference between microvascular Po2 and Pmvo2 increased significantly during hemodilution. Intestinal microvascular Po2 and oxygen consumption were well preserved during moderate normovolemic hemodilution. These results might be explained by the notion of others that hemodilution induces recruitment of capillaries, resulting in redistribution of the intestinal blood flow in favor of the microcirculation, which allows a more efficient extraction of oxygen. These findings further indicate that the use of venous Po2 values as indicators of microvascular oxygenation may be misleading.