Usefulness of Charlson comorbidity index-adjusted mortality prediction tools and factors influencing mortality in intensive care unit patients: a retrospective medical record review-based study.

To study the incidence and prognosis of thrombocytopenia in adult intensive care unit (ICU) patients. Prospective observational cohort study. The medical ICU of a university hospital and the combined medical-surgical ICU of a regional hospital. All patients consecutively admitted during a 5-month period. Patient surveillance and data collection. The primary outcome measure was ICU mortality. Data of 329 patients were analyzed. Overall ICU mortality rate was 19.5%. A total of 136 patients (41.3%) had at least one platelet count <150 x 10(9)/L. These patients had higher Multiple Organ Dysfunction Score (MODS), Simplified Acute Physiology Score (SAPS) II, and Acute Physiology and Chronic Health Evaluation (APACHE) II scores at admission, longer ICU stay (8 [4-16] days vs. 5 [2-9] days) (median [interquartile range]), and higher ICU mortality (crude odds ratio [OR], 5.0; 95% confidence interval [CI], 2.7-9.1) and hospital mortality than patients with daily platelet counts >150 x 10(9)/L (p < .0005 for all comparisons). Bleeding incidence rose from 4.1% in nonthrombocytopenic patients to 21.4% in patients with minimal platelet counts between 101 x 10(9)/L and 149 x 10(9)/L (p = .0002) and to 52.6% in patients with minimal platelet counts <100 x 10(9)/L (p < .0001). In all quartiles of admission APACHE II and SAPS II scores, a nadir platelet count <150 x 10(9)/L was related with a substantially poorer vital prognosis. Similarly, a drop in platelet count to < or =50% of admission was associated with higher death rates (OR, 6.0; 95% CI, 3.0-12.0; p < .0001). In a logistic regression analysis with ICU mortality as the dependent variable, the occurrence of thrombocytopenia had more explanatory power than admission variables, including APACHE II, SAPS II, and MODS scores (adjusted OR, 4.2; 95% CI, 1.8-10.2). Thrombocytopenia is common in ICUs and constitutes a simple and readily available risk marker for mortality, independent of and complementary to established severity of disease indices. Both a low nadir platelet count and a large fall of platelet count predict a poor vital outcome in adult ICU patients.

OBJECTIVE: To compare resource utilization and outcomes between cohorts of pediatric and adult intensive care unit (ICU) patients from a single institution. DESIGN: Prospective, observational cohort study. SETTING: A large, urban, tertiary care medical center. PATIENTS: A total of 780 patients consecutively admitted to the pediatric ICU, adult medical ICU, and adult surgical ICU. MEASUREMENTS AND MAIN RESULTS: ICU, hospital and 6-month survivals and hospital costs from index ICU admission. Predicted mortality by Pediatric Risk of Mortality III and Acute Physiology and Chronic Health Evaluation II. Health status at 6 months from index ICU admission. Pediatric patients had lower ICU (7.8% vs. 13.7%; p =.01), hospital (10.1% vs. 16.9%; p =.009), and 6-month (16.2% vs. 29.2%; p <.001) mortalities compared with adult patients. Adult patients had significantly lower probability of survival for 6 months from initial ICU admission compared with pediatric patients. The difference in survival was primarily accounted for by adult and pediatric medical patients. No differences could be observed between pediatric and adult ICU patients for mean hospital costs ($33,316 +/- $48,467 vs. $32,877 +/- $46,411; p =.92). Pediatric and adult patients incurred increasing costs with increasing risks of mortality. More than 50% of pediatric patients had a risk of mortality <0.5% compared with 1.2% of adult patients, but there was no difference in the mean use of ICU-specific interventions. CONCLUSIONS: Pediatric critical care patients have better short-term and longer-term survival compared with adult patients. The difference in survival is accounted for by the lower survival of adult medical patients. Despite the survival differences, pediatric and adult ICU patients incur similar hospital costs, and the proportions of patients who receive active ICU interventions are similar.

PurposeThis study compared the Charlson comorbidity index (CCI) information derived from chart review and administrative systems to assess the completeness and agreement between scores, evaluate the capacity to predict 30-day and 1-year mortality in intensive care unit (ICU) patients, and compare the predictive capacity with that of the Simplified Acute Physiology Score (SAPS) II model.Patients and methodsUsing data from 959 patients admitted to a general ICU in a Norwegian university hospital from 2007 to 2009, we compared the CCI score derived from chart review and administrative systems. Agreement was assessed using % agreement, kappa, and weighted kappa. The capacity to predict 30-day and 1-year mortality was assessed using logistic regression, model discrimination with the c-statistic, and calibration with a goodness-of-fit statistic.ResultsThe CCI was complete (n=959) when calculated from chart review, but less complete from administrative data (n=839). Agreement was good, with a weighted kappa of 0.667 (95% confidence interval: 0.596–0.714). The c-statistics for categorized CCI scores from charts and administrative data were similar in the model that included age, sex, and type of admission: 0.755 and 0.743 for 30-day mortality, respectively, and 0.783 and 0.775, respectively, for 1-year mortality. Goodness-of-fit statistics supported the model fit.ConclusionThe CCI scores from chart review and administrative data showed good agreement and predicted 30-day and 1-year mortality in ICU patients. CCI combined with age, sex, and type of admission predicted mortality almost as well as the physiology-based SAPS II.

To establish a 180-day mortality predictive score based on frailty syndrome in elderly sepsis patients [elderly sepsis score (ESS)]. A prospective study for sepsis patients aged 60 years and above who were admitted to a medical intensive care unit of the General Hospital of Southern Theatre Command from January 1st, 2018 to December 31st, 2018 was conducted. Univariate analysis was performed on 19 independent variables including gender, age, body mass index (BMI), tumor, charlson comorbidity index (CCI), activity of daily living (ADL), instrumental activity of daily living (IADL), mini-mental state examination (MMSE), geriatric depression scale (GDS), clinical frail scale (CFS), sequential organ failure assessment (SOFA), Glasgow coma scale (GCS), acute physiology and chronic health evaluation (APACHE II, APACHE IV), modified NUTRIC score (MNS), multiple drug resistance (MDR), mechanical ventilation (MV), continuous renal replacement therapy (CRRT) and palliative care. Continuous independent variables were converted into classified variables. Multivariate binary regression analysis of risk factors was conducted to screen independent risk factors which affecting 180-day mortality in elderly sepsis patients. Then a 180-daymortality predictive score was established, and the discrimination of the mortality of patients using CFS, SOFA, GCS, APACHE II, APACHE IV, MNS scores were compared. A total of 257 patients were enrolled, with a 180-day mortality of 60.7%. Univariate analysis showed that age, tumor, CCI, ADL, IADL, MMSE, CFS, SOFA, GCS, APACHE II, APACHE IV, MNS, MDR, MV, CRRT, palliative care were risk factors of 180-day mortality in elderly sepsis patients [age: odds ratio (OR) = 1.027, 95% confidence interval (95%CI) was 1.005-1.050, P = 0.018; tumor: OR =2.001, 95%CI was 1.022-3.920, P = 0.043; CCI: OR = 1.193, 95%CI was 1.064-1.339, P = 0.003; ADL: OR = 0.851, 95%CI was 0.772-0.940, P = 0.001; IADL: OR = 0.894, 95%CI was 0.826-0.967, P = 0.005; MMSE: OR = 0.962, 95%CI was 0.937-0.988, P = 0.004; CFS: OR = 1.303, 95%CI was 1.089-1.558, P = 0.004; SOFA: OR = 1.112, 95%CI was 1.038-1.191, P = 0.003; GCS: OR = 0.918, 95%CI was 0.863-0.977, P = 0.007; APACHE II: OR = 1.098, 95%CI was 1.053-1.145, P < 0.001; APACHE IV: OR = 1.032, 95%CI was 1.020-1.044, P < 0.001; MNS: OR = 1.315, 95%CI was 1.159-1.493, P < 0.001; MDR: OR = 2.029, 95%CI was 1.197-3.437, P = 0.009; MV: OR = 6.408, 95%CI was 3.480-11.798, P < 0.001, CRRT: OR = 2.744, 95%CI was 1.529-4.923, P = 0.001, palliative care: OR = 5.760, 95%CI was 2.177-15.245, P < 0.001]. By binary regression analysis, CFS stratification (OR = 1.934, 95%CI was 1.267-2.953, P = 0.002), MV (OR = 4.531, 95%CI was 2.376-8.644, P < 0.001), CRRT (OR = 2.471, 95%CI was 1.285-4.752, P = 0.007), palliative care (OR = 6.169, 95%CI was 2.173-17.515, P = 0.001) were independent risk factors of 180-day mortality in elderly patients with sepsis. The model of "ESS = 0.660×CFS stratification+1.511×MV+0.905×CRRT+1.820×palliative care" was established. Receiver operating characteristic curve (ROC curve) analysis showed that the area under the ROC curve (AUC) for predicting 180-day mortality by ESS was 0.785 (95%CI was 0.730-0.834, P < 0.001). When the best cut-off value was 2.2 points, its sensitivity was 78.9%, specificity was 70.3%, the positive predictive value was 80.4%, and the negative predictive value was 68.3%. Simplified ESS was defined as "0.5×CFS stratification+1.5×MV+1×CRRT+2×palliative care". ROC curve analysis showed that AUC for predicting 180-day mortality by simplified ESS was 0.784 (95%CI was 0.729-0.833, P < 0.001). When the best cut-off value was 2.0 points, sensitivity was 76.9%, specificity was 70.3%, the positive predictive value was 80.0%, and the negative predictive value was 66.4%. Compared with CFS, SOFA, GCS, APACHE II, APACHE IV and MNS, ESS had a significant difference in discriminating 180-day mortality in elderly patients with sepsis (AUC was 0.785 vs. 0.607, 0.607, 0.600, 0.664, 0.702, 0.657, 95%CI: 0.730-0.734 vs. 0.537-0.678, 0.537-0.677, 0.529-0.671, 0.598-0.730, 0.638-0.766, 0.590-0.725, all P < 0.05). CFS, MV, CRRT, and palliative care are independent risk factors of 180-day mortality in elderly patients with sepsis. We established ESS based on these risk factors. The ESS model has good discrimination and can be used as a reference and assessment tool for prediction and treatment guidance in elderly patients with sepsis.

Objectives: The primary objective was to evaluate the antimicrobial susceptibility of Pseudomonas aeruginosa causing infection in elderly (≥65 years old) patients hospitalized in intensive care units (ICUs) of United States medical centers. Susceptibility results from isolates of elderly patients in ICUs were compared to isolates from elderly patients not in ICUs (elderly non-ICU) and adult ICU patients (18 to 64 years old; adult ICU). Methods: P. aeruginosa isolates were consecutively collected from 74 US medical centers in 2021-2025 and susceptibility tested by reference broth microdilution in the monitoring laboratory (Element Iowa City [JMI Laboratories]). The organism collection included 999 isolates from elderly ICU, 2027 isolates from elderly non-ICU, and 1022 isolates from adult ICU patients. Results: The most active agents against P. aeruginosa from all three patient groups were ceftazidime-avibactam (95.8% to 97.3% susceptible), ceftolozane-tazobactam (96.0% to 98.3% susceptible), imipenem-relebactam (97.6% to 98.7% susceptible), and tobramycin (91.4% to 94.7% susceptible). Susceptibility to piperacillin-tazobactam, ceftazidime, cefepime, meropenem, and imipenem were markedly lower among isolates from elderly and adult ICU patients compared to elderly non-ICU patients. Susceptibility to levofloxacin and tobramycin were lower among isolates from adult ICU patients compared to elderly ICU and non-ICU patients. Moreover, the frequency of multidrug-resistant (MDR) isolates was markedly higher among elderly (18.4%) and adult (22.4%) ICU patients compared to elderly non-ICU (11.0%) patients. An annual analysis of susceptibility to selected β-lactams showed a slight variation in susceptibility rates without a clear trend. Conclusions: Ceftazidime-avibactam, ceftolozane-tazobactam, and imipenem-relebactam were highly active and exhibited similar coverage against a large contemporary collection of P. aeruginosa isolates from ICU elderly, non-ICU elderly, and ICU adult patients. Cross-resistance among these β-lactamase inhibitor combinations (BLICs) varied markedly, indicating that all three should be tested in the clinical laboratory and available for clinical use.

Chronic alcohol consumption is a prevalent issue. Healthcare professionals often discover their patient has an alcohol consumption issue when they are admitted to the hospital and no longer have access to alcohol. The global standard for treating alcohol withdrawal syndrome (AWS) symptoms are benzodiazepines; however this therapy is often inadequate to control symptoms of delirium in adult intensive care unit (ICU) patients due to an imbalance of inhibitory and excitatory neurotransmitters. The objective of the systematic review is to examine the clinical effectiveness of dexmedetomidine as an adjuvant to benzodiazepine-based therapy versus benzodiazepine-based therapy alone in decreasing the severity of delirium associated with AWS in adult ICU patients. This review considered studies that included adult ICU patients over the age of 18 who were experiencing delirium associated with alcohol withdrawal. Patients admitted to the ICU with the diagnosis of AWS were included in the study.This review considered studies that evaluated dexmedetomidine as an adjuvant therapy to benzodiazepine-based therapy, compared to the use of benzodiazepine-based therapy alone in ICU patients experiencing alcohol withdrawal delirium.This review considered randomized controlled trials, non-randomized controlled trials, quasi-experimental, before and after studies, prospective and retrospective cohort studies, case control studies, analytical cross sectional studies, case series, individual case reports and descriptive cross sectional studies for inclusion.The systematic review evaluated dexmedetomidine as an adjuvant to benzodiazepine-based therapy to decrease delirium severity in alcohol withdrawal in ICU patients. The general outcome of delirium severity was measured using the Clinical Institute Withdrawal Assessment Score - Revised (CIWA), the Ramsey scale, the Richmond Agitation Sedation Score (RASS) and the Confusion Assessment Method for the ICU (CAM-ICU). The search strategy aimed to find both published and unpublished studies. A three-step search strategy was utilized in this review and included English language studies published after 1997. A search of Ovid/MEDLINE, EMBASE, Cochrane, Joanna Briggs Institute and nine other databases was conducted. Two independent reviewers using the Joanna Briggs Institute's standardized appraisal tool critically appraised the studies. A third independent reviewer was available to appraise studies if the two original reviewers disagreed in their assessments. There were no disagreements in findings between the two independent reviewers. Data was extracted using the standardized Joanna Briggs Institute's data extraction instruments. Statistical pooling was done using meta-analysis and findings are presented using a forest plot and narrative form. Four studies were included in the review, three retrospective case series and one prospective case series with a total sample size of 55 patients. Three studies used the CIWA score as the outcome measure and one study used the RASS score as the outcome measure. A meta-analysis of the three studies using the CIWA demonstrated that adjuvant use of dexmedetomidine with benzodiazepine-based therapy decreased CIWA scores (Weighted Mean Difference [WMD] -5.2, 95% Confidence Interval [CI] -6.24 to -4.16, p <0.0001). The final study using RASS scores reported improvement with adjuvant treatment with dexmedetomidine compared to benzodiazepine-based therapy alone. The use of dexmedetomidine as an adjuvant to benzodiazepine-based therapy decreased delirium more effectively than benzodiazepine-based therapy alone in adult ICU patients experiencing alcohol withdrawal delirium as evidenced by a decrease in CIWA and RASS scores. In adult ICU patients who are experiencing alcohol withdrawal delirium that is not controlled with benzodiazepine-based therapy alone, healthcare providers should consider dexmedetomidine as an adjuvant to standard benzodiazepine-based therapy. The use of dexmedetomidine in the management of delirium associated with alcohol withdrawal in adult ICU patients should be further studied via large scale randomized controlled trials.

Rotational thromboelastometry to assess hypercoagulability in COVID-19 patients

Platelet transfusions are frequently used in intensive care unit (ICU) patients, but contemporary epidemiological data are sparse. We aim to present contemporary international data on the use of platelet transfusions in adult ICU patients with thrombocytopenia. This is a protocol and statistical analysis plan for a post hoc sub-study of 504 thrombocytopenic patients from the 'Thrombocytopenia and platelet transfusions in ICU patients: an international inception cohort study (PLOT-ICU)'. The primary outcome will be the number of patients receiving platelet transfusion in the ICU reported according to the type of product received (apheresis-derived versus pooled whole-blood-derived transfusions). Secondary platelet transfusion outcomes will include platelet transfusion volumes; timing of platelet transfusion; approach to platelet transfusion dosing (fixed dosing versus weight-based dosing) and platelet count increments for prophylactic transfusions. Secondary clinical outcomes will include the number of patients receiving red blood cell- and plasma transfusions during ICU stay; the number of patients who bled in the ICU, the number of patients who had a new thrombosis in the ICU, and the number of patients who died. The duration of follow-up was 90 days. Baseline characteristics and secondary clinical outcomes will be stratified according to platelet transfusion status in the ICU and severity of thrombocytopenia. Data will be presented descriptively. The outlined study will provide detailed epidemiological data on the use of platelet transfusions in adult ICU patients with thrombocytopenia using data from the large international PLOT-ICU cohort study. The findings will inform the design of future randomised trials evaluating platelet transfusions in ICU patients.

Introduction and aimAcute kidney injury (AKI) is part of the multiple organ dysfunction syndrome in critically ill patients and is a common condition in intensive care units (ICUs). Renal replacement therapy (RRT) is the cornerstone of treatment for AKI in critically ill patients. This patient population has a high mortality rate despite RRT. There are two methods of RRT for patients in ICUs: intermittent hemodialysis (IHD) and continuous renal replacement therapy (CRRT). Both CRRT and IHD similarly provide adequate metabolic control. We aimed to compare these two RRT modalities in terms of ICU stay, mortality, and laboratory recovery in these patients with high mortality.Materials and methodsA total of 120 patients with AKI who needed RRT in the ICU were included in the study (CRRT, n:40; IHD, n:80). Acute Physiology and Chronic Health Evaluation (APACHE) II, Sepsis-related Organ Failure Assessment (SOFA), and Simplified Acute Physiology Score (SAPS)-II scores at the time of admission to the ICU were calculated. Mean arterial pressure, urea, creatinine, sodium, potassium, calcium, pH, lactate, and bicarbonate levels were measured before and after dialysis. Patients were classified as living and deceased. Factors affecting the length of stay in the intensive care unit and 30-day mortality were evaluated. The variability in laboratory parameters between groups before and after dialysis was examined. The groups were compared with these parameters.ResultsSixty-one point seven percent (61.7%, n:74) of the patients were female. The mean age was 62.90±13.64 years. At the time of admission to the ICU, the patients' SAPS II score was 45.05±12.76, APACHE II score was 22.05±6.32, and SOFA score was 8.26±2.48. 66.7% (n:80) of the patients included in the study died, and the length of stay of these patients in the ICU was 12.85±10.23 days. When the groups were compared, SAPS II, APACHE II scores, and SOFA scores were significantly higher in the CRRT group than in the IHD group (p:0.038, p:0.015, p:0.027, respectively). Although the length of stay in the ICU was shorter in the CRRT group, it was not statistically significant (p:0.075). There was no statistically significant difference between the groups in terms of mortality (p: 0.891). SAPS-II, APACHE II, and SOFA score affected 30-day mortality while age, gender, and RRT modalities were not associated with mortality. The improvement in laboratory parameters between the pre and post-RRT groups was statistically more significant in the IHD group (p<0.001). It was determined that there was a statistically greater decrease in mean arterial pressure in the IHD group (p<0.001).ConclusionsIt was determined that there was no difference between the CRRT and IHD modalities applied in patients with AKI admitted to the ICU in terms of mortality and length of stay in the ICU. It was observed that both modalities improved on laboratory parameters, but the improvement was greater in the IHD group. However, it was determined that there was a statistically greater decrease in mean arterial pressure in the IHD group.

To compare 4 general severity classification scoring systems concerning prognosis of outcome in 123 liver transplant recipients. The compared scoring systems were: the mortality prediction model (admission model and 24 h model); the simplified acute physiology score; the acute physiology and chronic health evaluation (Apache II) and the acute organ systems failure score. Retrospective, consecutive sample. Adult intensive care unit in a university hospital. 123 adult liver allograft recipients after admission to the intensive care unit. The scoring systems were calculated as described by the authors to classify the severity of illness after admission of the allograft recipients to the intensive care unit. The mean and median values of survivors and the group of patients, that died during hospital stay were compared. Receiver-operating characteristics were plotted for all scoring systems and the areas under the curves of receiver-operating characteristics were calculated. The predictive value of the 4 scoring systems was tested using a variety of sensitivity analyses. The mortality prediction model (24 h model) was found to have a high significance (p < 0.001) in predicting mortality and showed the greatest area under the curve (0.829). Simplified acute physiology score (p < 0.001) and acute physiology and chronic health evaluation (Apache II) (p < 0.01) had a high significance as well, but did not hit the level of prognosis of mortality prediction model, as shown in the area under the curves. Accordingly, sensitivity was highest in MPM-24 h (83%), followed by SAPS (72%) and Apache II (71%). MPM-24 h had a total misclassification rate of 22% (SAPS = 32%, Apache II = 33%). MPM-admission failed in predicting mortality (sensitivity = 52%). Organ systems failure score seemed not to be useful in liver transplant recipients. General disease classification systems, such as the mortality prediction model, simplified acute physiology score or acute physiology and chronic health evaluation are good mortality prediction models in patients after liver transplantation. We suggest that there is no need for improvement of a special scoring system.

Expert consensus on research priorities for the prevention of delirium in adult ICU patients.

The influence of hypophosphatemia on outcomes of low- and high-intensity continuous renal replacement therapy in critically ill patients with acute kidney injury

The death of a loved one in the intensive care unit may cause prolonged grief, social distress, and even symptoms of posttraumatic stress in family members. End-of-life interventions, such as grief education materials, telephone follow-up, and cards, are an important part of bereavement care for families. However, use of these interventions varies across institutions, and research findings on their effectiveness are conflicting.Takaoka and colleagues examined the use of multiauthored, customized handwritten sympathy cards mailed to families 2 to 8 weeks post mortem. They interviewed family members and clinicians and identified 3 themes: The use of personalized, written condolences from staff was experienced as a meaningful and compassionate intervention for bereaved families.See Article, pp 422-428Pressure injuries occur twice as often in intensive care unit (ICU) patients as in other acute care patients, with estimated costs of care exceeding $26.8 billion in the United States. Although there are known risk factors for hospital-acquired pressure injury (HAPI), such as decreased mobility, having surgery, and age greater than 65 years, little research has examined the relationship between skin status and HAPI in ICU patients.Alderden and colleagues examined data from routine nursing skin assessments to identify risk factors for HAPI in adult surgical ICU patients. They found the following: Although further study is warranted to examine intraoperative risk factors, the authors recommend nurses consider the impact of skin changes on the development of HAPIs and remove causes of skin irritation.See Article, pp e128-e134Catheter-associated urinary tract infections continue to occur in both intensive care unit (ICU) patients and non-ICU patients, despite national prevention initiatives. Use of bladder scanning and ultrasound technology has helped to decrease the number of indwelling urinary catheter days, but the technology yields inaccuracies in patients with acute kidney injury, especially those with abdominal fluid/ascites.Schallom et al compared the accuracy of new bladder scanning technology and 2-dimensional ultrasound in measuring bladder volume in ICU patients unable to void 6 hours after catheter removal or patients receiving dialysis. They found Both bladder scanning and ultrasound can be used to measure bladder volumes accurately, but findings suggest that ultrasound should be used to measure bladder volume in patients with ascites.See Article, pp 458-467Many intensive care unit (ICU) patients require feeding tubes for nutritional support and medication delivery. Traditional practice includes every-4-hour assessment to verify tube placement. However, complications from malplaced feeding tubes continue to occur owing to inaccurate verification methods and lack of knowledge about tube migration.Using an electromagnetic placement device (EMPD) for tube placement, Bourgault and colleagues explored the factors associated with tube migration in adult ICU patients. They found See Article, pp 439-447

Intensive-care units (ICUs) treat the most critically ill patients, which is complicated by the heterogeneity of the diseases that they encounter. Severity scores based mainly on acute physiology measures collected at ICU admission are used to predict mortality, but are non-specific, and predictions for individual patients can be inaccurate. We investigated whether inclusion of long-term disease history before ICU admission improves mortality predictions. Registry data for long-term disease histories for more than 230 000 Danish ICU patients were used in a neural network to develop an ICU mortality prediction model. Long-term disease histories and acute physiology measures were aggregated to predict mortality risk for patients for whom both registry and ICU electronic patient record data were available. We compared mortality predictions with admission scores on the Simplified Acute Physiology Score (SAPS) II, the Acute Physiologic Assessment and Chronic Health Evaluation (APACHE) II, and the best available multimorbidity score, the Multimorbidity Index. An external validation set from an additional hospital was acquired after model construction to confirm the validity of our model. During initial model development data were split into a training set (85%) and an independent test set (15%), and a five-fold cross-validation was done during training to avoid overfitting. Neural networks were trained for datasets with disease history of 1 month, 3 months, 6 months, 1 year, 2·5 years, 5 years, 7·5 years, 10 years, and 23 years before ICU admission. Mortality predictions with a model based solely on disease history outperformed the Multimorbidity Index (Matthews correlation coefficient 0·265 vs 0·065), and performed similarly to SAPS II and APACHE II (Matthews correlation coefficient with disease history, age, and sex 0·326 vs 0·347 and 0·300 for SAPS II and APACHE II, respectively). Diagnoses up to 10 years before ICU admission affected current mortality prediction. Aggregation of previous disease history and acute physiology measures in a neural network yielded the most precise predictions of in-hospital mortality (Matthews correlation coefficient 0·391 for in-hospital mortality compared with 0·347 with SAPS II and 0·300 with APACHE II). These results for the aggregated model were validated in an external independent dataset of 1528 patients (Matthews correlation coefficient for prediction of in-hospital mortality 0·341). Longitudinal disease-spectrum-wide data available before ICU admission are useful for mortality prediction. Disease history can be used to differentiate mortality risk between patients with similar vital signs with more precision than SAPS II and APACHE II scores. Machine learning models can be deconvoluted to generate novel understandings of how ICU patient features from long-term and short-term events interact with each other. Explainable machine learning models are key in clinical settings, and our results emphasise how to progress towards the transformation of advanced models into actionable, transparent, and trustworthy clinical tools. Novo Nordisk Foundation and Innovation Fund Denmark.

BackgroundAs the COVID-19 pandemic continues, the number of patients admitted to the intensive care unit (ICU) is still increasing. The aim of our article is to estimate which of the conventional ICU mortality risk scores is the most accurate at predicting mortality in COVID-19 patients and to determine how these scores can be used in combination with the 4C Mortality Score.MethodsThis was a retrospective study of critically ill COVID-19 patients treated in tertiary reference COVID-19 hospitals during the year 2020. The 4C Mortality Score was calculated upon admission to the hospital. The Simplified Acute Physiology Score (SAPS) II, Acute Physiology and Chronic Health Evaluation (APACHE) II, and Sequential Organ Failure Assessment (SOFA) scores were calculated upon admission to the ICU. Patients were divided into two groups: ICU survivors and ICU non-survivors.ResultsA total of 249 patients were included in the study, of which 63.1% were male. The average age of all patients was 61.32 ± 13.3 years. The all-cause ICU mortality ratio was 41.4% (n = 103). To determine the accuracy of the ICU mortality risk scores a ROC-AUC analysis was performed. The most accurate scale was the APACHE II, with an AUC value of 0.772 (95% CI 0.714–0.830; p < 0.001). All of the ICU risk scores and 4C Mortality Score were significant mortality predictors in the univariate regression analysis. The multivariate regression analysis was completed to elucidate which of the scores can be used in combination with the independent predictive value. In the final model, the APACHE II and 4C Mortality Score prevailed. For each point increase in the APACHE II, mortality risk increased by 1.155 (OR 1.155, 95% CI 1.085–1.229; p < 0.001), and for each point increase in the 4C Mortality Score, mortality risk increased by 1.191 (OR 1.191, 95% CI 1.086–1.306; p < 0.001), demonstrating the best overall calibration of the model.ConclusionsThe study demonstrated that the APACHE II had the best discrimination of mortality in ICU patients. Both the APACHE II and 4C Mortality Score independently predict mortality risk and can be used concomitantly.