TRIPOD: a new reporting baseline for developing and interpreting prediction models

Prediction of esophagogastroduodenoscopy therapeutic usefulness for in-ICU suspected upper gastrointestinal bleeding: the SUGIBI score study

BACKGROUND

Suspected upper gastrointestinal bleeding (SUGIB) is a common issue during ICU stay. In the absence of specific guidelines on the indication and timing of esophagogastroduodenoscopy (EGD), there is substantial variability in EGD indication depending on accessibility and clinical presentation. This study aimed to investigate factors associated with the need for per-EGD hemostatic therapy and to create a score predicting therapeutic benefit of emergency bedside EGD in ICU patients with SUGIB.

METHODS

We conducted a retrospective study in our ICU to identify factors associated with the need for hemostatic procedure during EGD performed for SUGIB. From this observational cohort, we derived a score predicting the need for hemostasis during EGD, the SUGIBI score. This score was subsequently validated in a retrospective multicenter cohort.

RESULTS

Two hundred fifty-five patients not primarily admitted for GI bleeding who underwent a bedside EGD for SUGIB during their ICU stay were analyzed. The preeminent EGD indication were anemia (79%), melena (19%), shock (14%), and hematemesis (13%). EGD was normal in 24.7% of cases, while primary lesions reported were ulcers (23.1%), esophagitis (18.8%), and gastritis (12.5%). Only 12.9% of patients underwent hemostatic endotherapy during EGD. A SUGIBI score < 4 had a negative predictive value of 95% (91-99) for hemostatic endotherapy [AUC of 0.81; 0.75-0.91 (p < 0.0001)]. The SUGIBI score for predicting the need for an EGD-guided hemostatic procedure was next validated in a multicenter cohort with an AUC of 0.75 (0.66-0.85) (p < 0.0001), a score < 4 having a negative predictive value of 95% (92-97).

CONCLUSIONS

Our study shows that the therapeutic usefulness of bedside emergency EGD for SUGIB in critically ill patients is limited to a minority of patients. The SUGIBI score should help clinicians stratify the probability of a therapeutic EGD.

Prediction of Long-Term Physical, Mental, and Cognitive Problems Following Critical Illness: Development and External Validation of the PROSPECT Prediction Model

OBJECTIVES

ICU survivors often suffer from long-lasting physical, mental, and cognitive health problems after hospital discharge. As several interventions that treat or prevent these problems already start during ICU stay, patients at high risk should be identified early. This study aimed to develop a model for early prediction of post-ICU health problems within 48 hours after ICU admission.

DESIGN

Prospective cohort study in seven Dutch ICUs.

SETTING/PATIENTS

ICU patients older than 16 years and admitted for greater than or equal to 12 hours between July 2016 and March 2020.

INTERVENTIONS

None.

MEASUREMENTS AND MAIN RESULTS

Outcomes were physical problems (fatigue or ≥ 3 new physical symptoms), mental problems (anxiety, depression, or post-traumatic stress disorder), and cognitive impairment. Patient record data and questionnaire data were collected at ICU admission, and after 3 and 12 months, of 2,476 patients. Several models predicting physical, mental, or cognitive problems and a composite score at 3 and 12 months were developed using variables collected within 48 hours after ICU admission. Based on performance and clinical feasibility, a model, PROSPECT, predicting post-ICU health problems at 3 months was chosen, including the predictors of chronic obstructive pulmonary disease, admission type, expected length of ICU stay greater than or equal to 2 days, and preadmission anxiety and fatigue. Internal validation using bootstrapping on data of the largest hospital ( n = 1,244) yielded a C -statistic of 0.73 (95% CI, 0.70-0.76). External validation was performed on data ( n = 864) from the other six hospitals with a C -statistic of 0.77 (95% CI, 0.73-0.80).

CONCLUSIONS

The developed and externally validated PROSPECT model can be used within 48 hours after ICU admission for identifying patients with an increased risk of post-ICU problems 3 months after ICU admission. Timely preventive interventions starting during ICU admission and follow-up care can prevent or mitigate post-ICU problems in these high-risk patients.

Prediction of very early major bleeding risk in acute pulmonary embolism: an independent external validation of the Pulmonary Embolism-Syncope, Anemia, and Renal Dysfunction (PE-SARD) bleeding score

BACKGROUND

The Pulmonary Embolism-Syncope, Anemia, and Renal Dysfunction (PE-SARD) bleeding score was derived to predict very early major bleeding (MB) in patients with acute pulmonary embolism (PE). Before adoption into practice, the score requires external validation in different populations.

OBJECTIVES

We independently validated the PE-SARD score in a prospective multicenter Swiss cohort of 687 patients aged ≥65 years with acute PE.

METHODS

The PE-SARD score uses 3 variables (syncope, anemia, and renal dysfunction) to classify patients into 3 categories of increasing bleeding risk. The outcomes were very early MB at 7 days (primary) and MB at later time points (secondary). We calculated the PE-SARD score for each patient and classified the proportion of patients as being at low, intermediate, and high risk. To assess discrimination and calibration, we calculated the area under the receiver operating characteristic curve and the Hosmer-Lemeshow goodness-of-fit test, respectively.

RESULTS

The prevalence of MB was 2.0% (14/687) at 7 days and 14.0% (96/687) after a median follow-up of 30 months. The PE-SARD score classified 40.2%, 42.2%, and 17.6% of patients as low, intermediate, and high risk for MB, respectively. The frequency of observed very early MB at 7 days was 1.8% in low-, 2.1% in intermediate-, and 2.5% in high-risk patients. The area under the receiver operating characteristic curve was 0.52 (95% CI, 0.48-0.56) at 7 days and increased to 0.60 (95% CI, 0.56-0.64) at the end of follow-up. Score calibration was adequate (p > .05) over the entire follow-up.

CONCLUSION

In our independent validation, the PE-SARD score did not accurately predict very early MB and may not be transportable to older patients with PE.

Prediction of in-hospital bleeding in acutely ill medical patients: External validation of the IMPROVE bleeding risk score

INTRODUCTION

Pharmacological thromboprophylaxis slightly increases bleeding risk. The only risk assessment model to predict bleeding in medical inpatients, the IMPROVE bleeding risk score, has never been validated using prospectively collected outcome data.

METHODS

We validated the IMPROVE bleeding risk score in a prospective multicenter cohort of medical inpatients. Primary outcome was in-hospital clinically relevant bleeding (CRB) within 14 days of admission, a secondary outcome was major bleeding (MB). We classified patients according to the score in high or low bleeding risk. We assessed the score's predictive performance by calculating subhazard ratios (sHRs) adjusted for thromboprophylaxis use, positive and negative predictive values (PPV, NPV), and the area under the receiver operating characteristic curves (AUC).

RESULTS

Of 1155 patients, 8 % were classified as high bleeding risk. CRB and MB within 14 days occurred in 0.94 % and 0.47 % of low-risk and in 5.6 % and 3.4 % of high-risk patients, respectively. Adjusted for thromboprophylaxis, classification in the high-risk group was associated with an increased risk of 14-day CRB (sHR 4.7, 95 % confidence interval [CI] 1.5-14.5) and MB (sHR 4.9, 95%CI 1.0-23.4). PPV was 5.6 % and 3.4 %, while NPV was 99.1 % and 99.5 % for CRB and MB, respectively. The AUC was 0.68 (95%CI 0.66-0.71) for CRB and 0.73 (95%CI 0.71-0.76) for MB.

CONCLUSION

The IMPROVE bleeding risk score showed moderate to good discriminatory power to predict bleeding in medical inpatients. The score may help identify patients at high risk of in-hospital bleeding, in whom careful assessment of the risk-benefit ratio of pharmacological thromboprophylaxis is warranted.

Red blood cell distribution width for the prediction of outcomes after cardiac arrest

The red blood cell distribution width (RDW) is a routinely available blood marker that measures the variation of the size/volume of red blood cells. The aim of our study was to investigate the prognostic value of RDW in cardiac arrest patients and to assess whether RDW improves the prognostic value of three cardiac arrest-specific risk scores. Consecutive adult cardiac arrest patients admitted to the ICU of a Swiss university hospital were included. The primary outcome was poor neurological outcome at hospital discharge assessed by Cerebral Performance Category. Of 702 patients admitted to the ICU after cardiac arrest, 400 patients (57.0%) survived, of which 323 (80.8%) had a good neurological outcome. Higher mean RDW values showed an independent association with poor neurological outcomes at hospital discharge (adjusted OR 1.27, 95% CI 1.14 to 1.41; p < 0.001). Adding the maximum RDW value to the OHCA- CAHP- and PROLOGUE cardiac arrest scores improved prognostic performance. Within this cohort of cardiac arrest patients, RDW was an independent outcome predictor and slightly improved three cardiac arrest-specific risk scores. RDW may therefore support clinical decision-making.

External validation of postoperative nausea and vomiting risk scores in patients with liver cancer: A single-centre prospective cohort study

OBJECTIVES

This study aimed to test the external validity of postoperative nausea and vomiting (PONV) risk assessment tools in patients undergoing hepatectomy, and to guide healthcare professionals' assessment of postoperative patients.

BACKGROUND

The identification of PONV risk is particularly important in the context of prevention. However, the predictive performance of the current PONV risk scores has not been confirmed in patients with liver cancer, and its applicability is still unknown. These uncertainties pose difficulties in performing routine risk assessment of PONV for patients with liver cancer in a clinical practice setting.

METHODS

Patients diagnosed with liver cancer and undergoing hepatectomy were prospectively consecutively recruited. All enrolled patients received PONV assessments and PONV risk assessments via the Apfel risk score and the Koivuranta risk score. Receiver operating characteristic curves (ROC curves) and calibration curves were used to assess the external validity. This study was reported according to the TRIPOD Checklist.

RESULTS

Among 214 PONV-assessed patients, 114 patients (53.3%) developed PONV. For the Apfel simplified risk score, the ROC area was 0.612 (95% confidence interval [CI]: 0.543-0.678) in the validation dataset, which demonstrated imperfect discrimination; the calibration curve showed poor calibration with a slope of 0.49. For the Koivuranta score, the ROC area was 0.628 (CI: 0.559-0.693) in the validation dataset, which showed limited discrimination; the calibration curve indicated an unsatisfactory calibration with a slope of 0.71.

CONCLUSIONS

The Apfel risk score and the Koivuranta risk score were not well validated in our study and disease-specific risk factors should be taken into account when updating or developing PONV risk stratification instruments.

Radiomics-based prediction of two-year clinical outcome in locally advanced cervical cancer patients undergoing neoadjuvant chemoradiotherapy

PURPOSE

The aim of this study is to determine if radiomics features extracted from staging magnetic resonance (MR) images could predict 2-year long-term clinical outcome in patients with locally advanced cervical cancer (LACC) after neoadjuvant chemoradiotherapy (NACRT).

MATERIALS AND METHODS

We retrospectively enrolled patients with LACC diagnosis who underwent NACRT followed by radical surgery in two different institutions. Radiomics features were extracted from pre-treatment 1.5 T T2w MR images. The predictive performance of each feature was quantified in terms of Wilcoxon-Mann-Whitney test. Among the significant features, Pearson correlation coefficient (PCC) was calculated to quantify the correlation among the different predictors. A logistic regression model was calculated considering the two most significant features at the univariate analysis showing the lowest PCC value. The predictive performance of the model created was quantified out using the area under the receiver operating characteristic curve (AUC).

RESULTS

A total of 175 patients were retrospectively enrolled (142 for the training cohort and 33 for the validation one). 1896 radiomic feature were extracted, 91 of which showed significance (p < 0.05) at the univariate analysis. The radiomic model showing the highest predictive value combined the features calculated starting from the gray level co-occurrence-based features. This model achieved an AUC of 0.73 in the training set and 0.91 in the validation set.

CONCLUSIONS

The proposed radiomic model showed promising performances in predicting 2-year overall survival before NACRT. Nevertheless, the observed results should be tested in larger studies with consistent external validation cohorts, to confirm their potential clinical use.

Prognostic Research in Traumatic Brain Injury: Markers, Modeling, and Methodological Principles

Prognostic assessment in traumatic brain injury (TBI) is embedded deeply in clinical care. Considering the limitations of current prognostic indicators, there is increasing interest in understanding the role of new biomarkers, and in finding other prognostic indicators of long-term outcomes following TBI. New prognostic indicators may result in the development of more accurate prediction models that could be useful for both risk stratification and clinical decision making. We aimed to review methodological issues and provide tentative guidelines for prognostic research in TBI. Prognostic factor research focuses on the role of a specific patient or disease-related characteristic in relation to outcome. Typically, univariable relations of the prognostic factor are studied, followed by analyses adjusting for other variables related to the outcome. Following existing guidelines, we emphasize the importance of transparent reporting of patient and specimen characteristics, study design, clinical end-points, and statistical analysis. Prognostic model research considers combinations of predictors, with challenges for model specification, estimation, evaluation, validation, and presentation. We highlight modern approaches and opportunities related to missing values, exploration of non-linear effects, and assessing between-study heterogeneity. Prognostic research in TBI can be improved if key methodological principles are adhered to and when research is performed in collaboration among multiple centers to ensure generalizability.

Screening the risk factors for methamphetamine use in pregnant women not receiving prenatal care

OBJECTIVE

To develop a screening tool for the risk factors potentially indicating methamphetamine use in pregnant women who are not receiving prenatal care.

METHOD

This prospective cohort, Institutional Review Board-approved study was performed at a university hospital in Thailand between January 2017 and January 2019. A screening tool was developed using data from 125 pregnant women not receiving prenatal care upon their first admission for childbearing at the hospital delivery room. Potential factors obtained from the patient's history, physical examination, and methamphetamine use in pregnancy or had a urine amphetamine test positive were entered into a logistic regression analysis. The discriminative ability of the screening tool was expressed by the area under the receiver operating characteristic curve (AUROC) sensitivity and specificity, while bootstrapping was used for internal validation.

RESULTS

The screening covered four factors: smoking (odds ratio 7.73, score = 2), drinking (3.81, score = 1), living with a spouse or friend who uses methamphetamine (17.28, score = 3), BP ≥ 130/90 mmHg (2.47, score = 1). The AUROC for the model was 0.87, 95% CI, 0.81-0.93 (SE: 0.03). A total points score ≥3 represented the best cut-off value, with a sensitivity of 81% and specificity of 82%. Across the bootstrapping, the C-statistic for the full screening was 0.86, 95% CI, 0.81-0.93 (SE: 0.03).

CONCLUSION

A screening tool was developed with an excellent ability to discriminate the risk factors potentially indicating methamphetamine use in pregnant women not receiving prenatal care. Validation in pregnant women receiving prenatal care still needs to be performed.

Artificial Intelligence and Liability in Medicine: Balancing Safety and Innovation

Policy Points With increasing integration of artificial intelligence and machine learning in medicine, there are concerns that algorithm inaccuracy could lead to patient injury and medical liability. While prior work has focused on medical malpractice, the artificial intelligence ecosystem consists of multiple stakeholders beyond clinicians. Current liability frameworks are inadequate to encourage both safe clinical implementation and disruptive innovation of artificial intelligence. Several policy options could ensure a more balanced liability system, including altering the standard of care, insurance, indemnification, special/no-fault adjudication systems, and regulation. Such liability frameworks could facilitate safe and expedient implementation of artificial intelligence and machine learning in clinical care.

Development and Validation of a Hospital Indicator of Activity-Based Costs for Injury Admissions

OBJECTIVES

To develop a hospital indicator of resource use for injury admissions.

METHODS

We focused on resource use for acute injury care and therefore adopted a hospital perspective. We included patients ≥16 years old with an Injury Severity Score >9 admitted to any of the 57 trauma centers of an inclusive Canadian trauma system from 2014 to 2018. We extracted data from the trauma registry and hospital financial reports and estimated resource use with activity-based costing. We developed risk-adjustment models by trauma center designation level (I/II and III/IV) for the whole sample, traumatic brain injuries, thoraco-abdominal injuries, orthopedic injuries, and patients ≥65 years old. Candidate variables were selected using bootstrap resampling. We performed benchmarking by comparing the adjusted mean cost in each center, obtained using shrinkage estimates, to the provincial mean.

RESULTS

We included 38 713 patients. The models explained between 12% and 36% (optimism-corrected r²) of the variation in resource use. In the whole sample and in all subgroups, we identified centers with higher- or lower-than-expected resource use across level I/II and III/IV centers.

CONCLUSIONS

We propose an algorithm to produce the indicator using data routinely collected in trauma registries to prompt targeted exploration of potential areas for improvement in resource use for injury admissions. The r² of our models suggest that between 64% and 88% of the variation in resource use for injury care is dictated by factors other than patient baseline risk.

Functional Outcome Prediction in Ischemic Stroke: A Comparison of Machine Learning Algorithms and Regression Models

Background and Purpose: Stroke-related functional risk scores are used to predict patients' functional outcomes following a stroke event. We evaluate the predictive accuracy of machine-learning algorithms for predicting functional outcomes in acute ischemic stroke patients after endovascular treatment.

Methods: Data were from the Precise and Rapid Assessment of Collaterals with Multi-phase CT Angiography (PROVE-IT), an observational study of 614 ischemic stroke patients. Regression and machine learning models, including random forest (RF), classification and regression tree (CART), C5.0 decision tree (DT), support vector machine (SVM), adaptive boost machine (ABM), least absolute shrinkage and selection operator (LASSO) logistic regression, and logistic regression models were used to train and predict the 90-day functional impairment risk, which is measured by the modified Rankin scale (mRS) score > 2. The models were internally validated using split-sample cross-validation and externally validated in the INTERRSeCT cohort study. The accuracy of these models was evaluated using the area under the receiver operating characteristic curve (AUC), Matthews Correlation Coefficient (MCC), and Brier score.

Results: Of the 614 patients included in the training data, 249 (40.5%) had 90-day functional impairment (i.e., mRS > 2). The median and interquartile range (IQR) of age and baseline NIHSS scores were 77 years (IQR = 69–83) and 17 (IQR = 11–22), respectively. Both logistic regression and machine learning models had comparable predictive accuracy when validated internally (AUC range = [0.65–0.72]; MCC range = [0.29–0.42]) and externally (AUC range = [0.66–0.71]; MCC range = [0.34–0.42]).

Conclusions: Machine learning algorithms and logistic regression had comparable predictive accuracy for predicting stroke-related functional impairment in stroke patients.

Oncotype DX Predictive Nomogram for Recurrence Score Output: The Novel System ADAPTED01 Based on Quantitative Immunochemistry Analysis

PURPOSE

Oncotype DX (ODX) predicts breast cancer recurrence risk, guiding the choice of adjuvant treatment. In many countries, access to the test is not always available. We used correlation between phenotypical tumor characteristics, quantitative classical immunohistochemistry (IHC), and recurrence score (RS) assessed by ODX to develop a decision supporting system for clinical use.

PATIENTS AND METHODS

Breast cancer patients who underwent ODX testing between 2014 and 2018 were retrospectively included in the study. The data selected for analysis were age, menopausal status, and pathologic and IHC features. IHC was performed with standardized quantitative methods. The data set was split into two subsets: 70% for the training set and 30% for the internal validation set. Statistically significant features were included in logistic models to predict RS ≤ 25 or ≤ 20. Another set was used for external validation to test reproducibility of prediction models.

RESULTS

The internal set included 407 patients. Mean (range) age was 53.7 (31-80) years, and 222 patients (54.55%) were > 50 years old. ODX results showed 67 patients (16.6%) had RS between 0 and 10, 272 patients between 11 and 25 (66.8%), and 68 patients > 26 (16.6%). Logistic regression analysis showed that RS score (for threshold ≤ 25) was significantly associated with estrogen receptor (P = .004), progesterone receptor (P < .0001), and Ki-67 (P < .0001). Generalized linear regression resulted in a model that had an area under the receiver operating characteristic curve (AUC) of 92.2 (sensitivity 84.2%, specificity 80.1%) and that was well calibrated. The external validation set (183 patients) analysis confirmed the model performance, with an AUC of 82.3 and a positive predictive value of 91%. A nomogram was generated for further prospective evaluation to predict RS ≤ 25.

CONCLUSION

RS was related to quantitative IHC in patients with RS ≤ 25, with a good performance of the statistical model in both internal and external validation. A nomogram for enhancing clinical approach in a cost-effective manner was developed. Prospective studies must test this application in clinical practice.

External Validation of Two Models to Predict Delirium in Critically Ill Adults Using Either the Confusion Assessment Method-ICU or the Intensive Care Delirium Screening Checklist for Delirium Assessment

OBJECTIVES

To externally validate two delirium prediction models (early prediction model for ICU delirium and recalibrated prediction model for ICU delirium) using either the Confusion Assessment Method-ICU or the Intensive Care Delirium Screening Checklist for delirium assessment.

DESIGN

Prospective, multinational cohort study.

SETTING

Eleven ICUs from seven countries in three continents.

PATIENTS

Consecutive, delirium-free adults admitted to the ICU for greater than or equal to 6 hours in whom delirium could be reliably assessed.

INTERVENTIONS

None.

MEASUREMENTS AND MAIN RESULTS

The predictors included in each model were collected at the time of ICU admission (early prediction model for ICU delirium) or within 24 hours of ICU admission (recalibrated prediction model for ICU delirium). Delirium was assessed using the Confusion Assessment Method-ICU or the Intensive Care Delirium Screening Checklist. Discrimination was determined using the area under the receiver operating characteristic curve. The predictive performance was determined for the Confusion Assessment Method-ICU and Intensive Care Delirium Screening Checklist cohort, and compared with both prediction models' original reported performance. A total of 1,286 Confusion Assessment Method-ICU-assessed patients and 892 Intensive Care Delirium Screening Checklist-assessed patients were included. Compared with the area under the receiver operating characteristic curve of 0.75 (95% CI, 0.71-0.79) in the original study, the area under the receiver operating characteristic curve of the early prediction model for ICU delirium was 0.67 (95% CI, 0.64-0.71) for delirium as assessed using the Confusion Assessment Method-ICU and 0.70 (95% CI, 0.66-0.74) using the Intensive Care Delirium Screening Checklist. Compared with the original area under the receiver operating characteristic curve of 0.77 (95% CI, 0.74-0.79), the area under the receiver operating characteristic curve of the recalibrated prediction model for ICU delirium was 0.75 (95% CI, 0.72-0.78) for assessing delirium using the Confusion Assessment Method-ICU and 0.71 (95% CI, 0.67-0.75) using the Intensive Care Delirium Screening Checklist.

CONCLUSIONS

Both the early prediction model for ICU delirium and recalibrated prediction model for ICU delirium are externally validated using either the Confusion Assessment Method-ICU or the Intensive Care Delirium Screening Checklist for delirium assessment. Per delirium prediction model, both assessment tools showed a similar moderate-to-good statistical performance. These results support the use of either the early prediction model for ICU delirium or recalibrated prediction model for ICU delirium in ICUs around the world regardless of whether delirium is evaluated with the Confusion Assessment Method-ICU or Intensive Care Delirium Screening Checklist.

Radiomics for radiation oncologists: are we ready to go?

Radiomics have emerged as an exciting field of research over the past few years, with very wide potential applications in personalised and precision medicine of the future. Radiomics-based approaches are still however limited in daily clinical practice in oncology. This review focus on how radiomics could be incorporated into the radiation therapy pipeline, and globally help the radiation oncologist, from the tumour diagnosis to follow-up after treatment. Radiomics could impact on all steps of the treatment pipeline, once the limitations in terms of robustness and reproducibility are overcome. Major ongoing efforts should be made to collect and share data in the most standardised manner possible.

Multinational development and validation of an early prediction model for delirium in ICU patients

Rationale

Delirium incidence in intensive care unit (ICU) patients is high and associated with poor outcome. Identification of high-risk patients may facilitate its prevention.

Purpose

To develop and validate a model based on data available at ICU admission to predict delirium development during a patient’s complete ICU stay and to determine the predictive value of this model in relation to the time of delirium development.

Methods

Prospective cohort study in 13 ICUs from seven countries. Multiple logistic regression analysis was used to develop the early prediction (E-PRE-DELIRIC) model on data of the first two-thirds and validated on data of the last one-third of the patients from every participating ICU.

Results

In total, 2914 patients were included. Delirium incidence was 23.6 %. The E-PRE-DELIRIC model consists of nine predictors assessed at ICU admission: age, history of cognitive impairment, history of alcohol abuse, blood urea nitrogen, admission category, urgent admission, mean arterial blood pressure, use of corticosteroids, and respiratory failure. The area under the receiver operating characteristic curve (AUROC) was 0.76 [95 % confidence interval (CI) 0.73–0.77] in the development dataset and 0.75 (95 % CI 0.71–0.79) in the validation dataset. The model was well calibrated. AUROC increased from 0.70 (95 % CI 0.67–0.74), for delirium that developed <2 days, to 0.81 (95 % CI 0.78–0.84), for delirium that developed >6 days.

Conclusion

Patients’ delirium risk for the complete ICU length of stay can be predicted at admission using the E-PRE-DELIRIC model, allowing early preventive interventions aimed to reduce incidence and severity of ICU delirium.

Electronic supplementary material

The online version of this article (doi:10.1007/s00134-015-3777-2) contains supplementary material, which is available to authorized users.