Natural language processing (NLP) is a branch of artificial intelligence focused on enabling computers to interpret and analyse text-based data. The intensive care specialty is known to generate large volumes of data, including free-text, however, NLP applications are not commonly used either in critical care clinical research or quality improvement projects. This review aims to provide an overview of how NLP has been used in the intensive care specialty and promote an understanding of NLP's potential future clinical applications.
Scoping review.
A systematic search was developed with an information specialist and deployed on the PubMed electronic journal database. Results were restricted to the last 10 years to ensure currency.
Screening and data extraction were undertaken by two independent reviewers, with any disagreements resolved by a third. Given the heterogeneity of the eligible articles, a narrative synthesis was conducted.
Eighty-seven eligible articles were included in the review. The most common type (n = 24) were studies that used NLP-derived features to predict clinical outcomes, most commonly mortality (n = 16). Next were articles that used NLP to identify a specific concept (n = 23), including sepsis, family visitation and mental health disorders. Most studies only described the development and internal validation of their algorithm (n = 79), and only one reported the implementation of an algorithm in a clinical setting.
Natural language processing has been used for a variety of purposes in the ICU context. Increasing awareness of these techniques amongst clinicians may lead to more clinically relevant algorithms being developed and implemented.
Acute respiratory distress syndrome (ARDS) is associated with significant mortality, morbidity, and cost. We aimed to describe characteristics and management of adult patients admitted to intensive care units (ICUs) in Australia and New Zealand with moderate-severe ARDS, to better understand contemporary practice.
Bi-national, prospective, observational, multi-centre study.
19 ICUs in Australia and New Zealand.
Mechanically ventilated patients with moderate-severe ARDS.
Baseline demographic characteristics, ventilation characteristics, use of adjunctive support therapy and all-cause mortality to day 28. Data were summarised using descriptive statistics.
200 participants were enrolled, mean (±SD) age 55.5 (±15.9) years, 40% (n = 80) female. Around half (51.5%) had no baseline comorbidities and 45 (31%) tested positive for COVID-19. On day 1, mean SOFA score was 9 ± 3; median (IQR) PaO2/FiO2 ratio 119 (89, 142), median (IQR) FiO2 70% (50%, 99%) and mean (±SD) positive end expiratory pressure (PEEP) 11 (±3) cmH2O. On day one, 10.5% (n = 21) received lung protective ventilation (LPV) (tidal volume ≤6.5 mL/kg predicted body weight and plateau pressure or peak pressure ≤30 cm H2O). Adjunctive therapies were received by 86% (n = 172) of patients at some stage from enrolment to day 28. Systemic steroids were most used (n = 127) followed by neuromuscular blockers (n = 122) and prone positioning (n = 27). Median ventilator-free days (IQR) to day 28 was 5 (0, 20). In-hospital mortality, censored at day 28, was 30.5% (n = 61).
In Australia and New Zealand, compliance with evidence-based practices including LPV and prone positioning was low in this cohort. Therapies with proven benefit in the treatment of patients with moderate-severe ARDS, such as lung protective ventilation and prone positioning, were not routinely employed.
Modern intensive care for moderate-to-severe traumatic brain injury (msTBI) focuses on managing intracranial pressure (ICP) and cerebral perfusion pressure (CPP). This approach lacks robust clinical evidence and often overlooks the impact of hypoxic injuries. Emerging monitoring modalities, particularly those capable of measuring brain tissue oxygen, represent a promising avenue for advanced neuromonitoring. Among these, brain tissue oxygen tension (PbtO2) shows the most promising results. However, there is still a lack of consensus regarding the interpretation of PbtO2 in clinical practice. This review aims to provide an overview of the pathophysiological rationales, monitoring technology, physiological determinants, and recent clinical trial evidence for PbtO2 monitoring in the management of msTBI.
To describe the proportion of patients admitted to intensive care who have anaphylaxis as a principal diagnosis and their subsequent outcomes in Australia and New Zealand.
Retrospective observational study of ICU admissions for severe anaphylaxis.
ICU admissions recorded in the Australian and New Zealand Intensive Care Society Adult Patient Database between 2012 and 2022.
Adults 16 years or older with severe anaphylaxis admitted to the ICU.
None.
Proportion of patients admitted to ICU who have anaphylaxis as a principal diagnosis, mortality rate, ICU and hospital length of stay.
7189 of the 7270 ICU admissions for severe anaphylaxis recorded between 2012 and 2022, were included in the analysis. This represented a proportion from 0.25% in 2012 to 0.43% in 2022. ICU and hospital mortality were 0.4% and 0.8%, respectively. The proportion of ICUs reporting at least one severe anaphylaxis each year increased from 61.7% in 2012 to 83.0% in 2022. Most of the patients were discharged home (92.6%, n = 6660). Increasing age (OR = 1.055; 95%CI: 1.008–1.105) and SOFA scores (OR = 1.616; 95%CI: 1.265–2.065), an immunosuppressive chronic condition (OR = 16.572; 95%CI: 3.006–91.349) and an increasing respiratory rate above 16 breaths/min (OR = 1.116; 95%CI: 1.057–1.178) predicted in-hospital mortality in patients with anaphylaxis, while higher GCS decreased in-hospital mortality (OR = 0.827; 95%CI: 0.705–0.969).
The overall proportion of patients admitted to ICU who have anaphylaxis as a principal diagnosis has increased. In-hospital mortality remains low despite the need for vital organ support. Further studies should investigate these identified factors that may predict in-hospital mortality among these patients.
Not applicable.
Frequent measurement of creatinine by point-of-care testing (POCT) may facilitate the earlier detection of acute kidney injury (AKI) in critically ill patients. However, no robust data exist to confirm its equivalence to central laboratory testing. We aimed to conduct a multicenter study to compare POCT with central laboratory creatinine (CrC) measurement.
Retrospective observational study, using hospital electronic medical records. Obtained paired point-of-care creatinine (CrP) from arterial blood gas machines and CrC.
Four intensive care units in Queensland, Australia.
Critically ill patients, where greater than 50% of POCT contained creatinine.
Mean difference, bias, and limits of agreement between two methods, and biochemical confounders.
We studied 79,767 paired measurements in 19,118 patients, with a median Acute Physiology and Chronic Health Evaluation 3 score of 51. The mean CrC was 115.5 μmol/L (standard deviation: 100.2) compared to a CrP mean of 115 μmol/L (standard deviation: 100.7) (Pearson coefficient of 0.99). The mean difference between CrP and CrC was 0.49 μmol/L with 95% limits of agreement of −27 μmol/L and +28 μmol/L. Several biochemical variables were independently associated with the difference between tests (e.g., pH, potassium, lactate, glucose, and bilirubin), but their impact was small.
In critically ill patients, measurement of creatinine by POCT yields clinically equivalent values to those obtained by central laboratory measurement and can be easily used for more frequent monitoring of kidney function in such patients. These findings open the door to the use of POCT for the earlier detection of acute kidney injury in critically ill patients.