Development and Validation of a Machine Learning COVID-19 Veteran (COVet) Deterioration Risk Score.

BACKGROUND AND OBJECTIVE: To develop the COVid Veteran (COVet) score for clinical deterioration in Veterans hospitalized with COVID-19 and further validate this model in both Veteran and non-Veteran samples. No such score has been derived and validated while incorporating a Veteran sample. DERIVATION COHORT: Adults (age ≥ 18 yr) hospitalized outside the ICU with a diagnosis of COVID-19 for model development to the Veterans Health Administration (VHA) (n = 80 hospitals). VALIDATION COHORT: External validation occurred in a VHA cohort of 34 hospitals, as well as six non-Veteran health systems for further external validation (n = 21 hospitals) between 2020 and 2023. PREDICTION MODEL: eXtreme Gradient Boosting machine learning methods were used, and performance was assessed using the area under the receiver operating characteristic curve and compared with the National Early Warning Score (NEWS). The primary outcome was transfer to the ICU or death within 24 hours of each new variable observation. Model predictor variables included demographics, vital signs, structured flowsheet data, and laboratory values. RESULTS: A total of 96,908 admissions occurred during the study period, of which 59,897 were in the Veteran sample and 37,011 were in the non-Veteran sample. During external validation in the Veteran sample, the model demonstrated excellent discrimination, with an area under the receiver operating characteristic curve of 0.88. This was significantly higher than NEWS (0.79; p < 0.01). In the non-Veteran sample, the model also demonstrated excellent discrimination (0.86 vs. 0.79 for NEWS; p < 0.01). The top three variables of importance were eosinophil percentage, mean oxygen saturation in the prior 24-hour period, and worst mental status in the prior 24-hour period. CONCLUSIONS: We used machine learning methods to develop and validate a highly accurate early warning score in both Veterans and non-Veterans hospitalized with COVID-19. The model could lead to earlier identification and therapy, which may improve outcomes.

Multicenter Development and Prospective Validation of eCARTv5: A Gradient Boosted Machine Learning Early Warning Score.

Rationale: Early detection of clinical deterioration using early warning scores may improve outcomes. However, most implemented scores were developed using logistic regression, only underwent retrospective internal validation, and were not tested in important patient subgroups. Objectives: To develop a gradient boosted machine model (eCARTv5) for identifying clinical deterioration and then validate externally, test prospectively, and evaluate across patient subgroups. Methods: All adult patients hospitalized on the wards in seven hospitals from 2008- 2022 were used to develop eCARTv5, with demographics, vital signs, clinician documentation, and laboratory values utilized to predict intensive care unit transfer or death in the next 24 hours. The model was externally validated retrospectively in 21 hospitals from 2009-2023 and prospectively in 10 hospitals from February to May 2023. eCARTv5 was compared to the Modified Early Warning Score (MEWS) and the National Early Warning Score (NEWS) using the area under the receiver operating characteristic curve (AUROC). Measurements and Main Results: The development cohort included 901,491 admissions, the retrospective validation cohort included 1,769,461 admissions, and the prospective validation cohort included 46,330 admissions. In retrospective validation, eCART had the highest AUROC (0.835; 95%CI 0.834, 0.835), followed by NEWS (0.766 (95%CI 0.766, 0.767)), and MEWS (0.704 (95%CI 0.703, 0.704)). eCART's performance remained high (AUROC ≥0.80) across a range of patient demographics, clinical conditions, and during prospective validation. Conclusions: We developed eCARTv5, which accurately identifies early clinical deterioration in hospitalized ward patients. Our model performed better than the NEWS and MEWS retrospectively, prospectively, and across a range of subgroups.

Combination of the National Early Warning Score (NEWS) and inflammatory biomarkers for early risk stratification in emergency department patients: results of a multinational, observational study.

OBJECTIVES: The National Early Warning Score (NEWS) helps to estimate mortality risk in emergency department (ED) patients. This study aimed to investigate whether the prognostic value of the NEWS at ED admission could be further improved by adding inflammatory blood markers (ie, white cell count (WCC), procalcitonin (PCT) and midregional-proadrenomedullin (MR-proADM). DESIGN: Secondary analysis of a multinational, observational study (TRIAGE study, March 2013-October 2014). SETTING: Three tertiary care centres in France, Switzerland and the USA. PARTICIPANTS: A total of 1303 adult medical patients with complete NEWS data seeking ED care were included in the final analysis. NEWS was calculated retrospectively based on admission data. MAIN OUTCOME MEASURES: The primary outcome was all-cause 30-day mortality. Secondary outcome was intensive care unit (ICU) admission. We used multivariate regression analyses to investigate associations of NEWS and blood markers with outcomes and area under the receiver operating curve (AUC) as a measure of discrimination. RESULTS: Of the 1303 included patients, 54 (4.1%) died within 30 days. The NEWS alone showed fair prognostic accuracy for all-cause 30-day mortality (AUC 0.73), with a multivariate adjusted OR of 1.26 (95% CI 1.13 to 1.40, p<0.001). The AUCs for the prediction of mortality using the inflammatory markers WCC, PCT and MR-proADM were 0.64, 0.71 and 0.78, respectively. Combining NEWS with all three blood markers or only with MR-proADM clearly improved discrimination with an AUC of 0.82 (p=0.002). Combining the three inflammatory markers with NEWS improved prediction of ICU admission (AUC 0.70vs0.65 when using NEWS alone, p=0.006). CONCLUSION: NEWS is helpful in risk stratification of ED patients and can be further improved by the addition of inflammatory blood markers. Future studies should investigate whether risk stratification by NEWS in addition to biomarkers improve site-of-care decision in this patient population. TRIAL REGISTRATION NUMBER: NCT01768494; Post-results.

Are admission procalcitonin levels universal mortality predictors across different medical emergency patient populations? Results from the multi-national, prospective, observational TRIAGE study.

BACKGROUND: Procalcitonin (PCT), an inflammatory blood biomarker, is well studied in infectious diseases. Its prognostic value in unselected emergency department (ED) patients remains yet undefined. Herein, we investigated association of admission PCT levels and mortality in a large, international-multicenter ED patient cohort. METHODS: We prospectively enrolled 6970 unselected, consecutive, adult, medical patients seeking ED care in three tertiary-care hospitals in Switzerland, France and the USA. We used multivariable logistic regression models to examine association of admission PCT levels (as a continuous predictor and across cut-offs) and 30-day mortality. We also investigated subgroup effects by main diagnosis, comorbidities and clinical features at presentation. RESULTS: During the 30-day follow-up, 328 (4.7%) participants died. Mortality increased stepwise within higher PCT cut-offs (0.05, 0.1, 0.25, 0.5 ng/mL) from 1%, 3%, 7%, 13% to 15%, respectively. This association was also confirmed in a fully-adjusted model including age, gender, main symptom, main diagnosis and vital parameters on admission. Receiver operating characteristic (ROC) curve analysis showed that PCT differentiated well between survivors and non-survivors in the overall cohort (area under ROC curve [AUC] 0.75) with best results for patient with metabolic (AUC: 0.85) and cardiovascular disease (AUC: 0.82). Addition of PCT also improved the prognostic accuracy of the quick sequential organ failure assessment (qSOFA) score from an AUC of from 0.61 to 0.76 (p<0.001). Results were similar for other secondary endpoints including intensive care unit (ICU) admission and hospital readmission. CONCLUSIONS: In this large and heterogenous medical ED patient cohort, admission PCT was a strong and independent outcome predictor for 30-day mortality across different medical diagnoses independent of underlying infection. PCT may help to improve risk stratification in unselected medical ED patients.

The TRIAGE-ProADM Score for an Early Risk Stratification of Medical Patients in the Emergency Department - Development Based on a Multi-National, Prospective, Observational Study.

INTRODUCTION: The inflammatory biomarker pro-adrenomedullin (ProADM) provides additional prognostic information for the risk stratification of general medical emergency department (ED) patients. The aim of this analysis was to develop a triage algorithm for improved prognostication and later use in an interventional trial. METHODS: We used data from the multi-national, prospective, observational TRIAGE trial including consecutive medical ED patients from Switzerland, France and the United States. We investigated triage effects when adding ProADM at two established cut-offs to a five-level ED triage score with respect to adverse clinical outcome. RESULTS: Mortality in the 6586 ED patients showed a step-wise, 25-fold increase from 0.6% to 4.5% and 15.4%, respectively, at the two ProADM cut-offs (≤0.75nmol/L, >0.75-1.5nmol/L, >1.5nmol/L, p ANOVA <0.0001). Risk stratification by combining ProADM within cut-off groups and the triage score resulted in the identification of 1662 patients (25.2% of the population) at a very low risk of mortality (0.3%, n = 5) and 425 patients (6.5% of the population) at very high risk of mortality (19.3%, n = 82). Risk estimation by using ProADM and the triage score from a logistic regression model allowed for a more accurate risk estimation in the whole population with a classification of 3255 patients (49.4% of the population) in the low risk group (0.3% mortality, n = 9) and 1673 (25.4% of the population) in the high-risk group (15.1% mortality, n = 252). CONCLUSIONS: Within this large international multicenter study, a combined triage score based on ProADM and established triage scores allowed a more accurate mortality risk discrimination. The TRIAGE-ProADM score improved identification of both patients at the highest risk of mortality who may benefit from early therapeutic interventions (rule in), and low risk patients where deferred treatment without negatively affecting outcome may be possible (rule out).

Predictors for Delayed Emergency Department Care in Medical Patients with Acute Infections - An International Prospective Observational Study.

INTRODUCTION: In overcrowded emergency department (ED) care, short time to start effective antibiotic treatment has been evidenced to improve infection-related clinical outcomes. Our objective was to study factors associated with delays in initial ED care within an international prospective medical ED patient population presenting with acute infections. METHODS: We report data from an international prospective observational cohort study including patients with a main diagnosis of infection from three tertiary care hospitals in Switzerland, France and the United States (US). We studied predictors for delays in starting antibiotic treatment by using multivariate regression analyses. RESULTS: Overall, 544 medical ED patients with a main diagnosis of acute infection and antibiotic treatment were included, mainly pneumonia (n = 218; 40.1%), urinary tract (n = 141; 25.9%), and gastrointestinal infections (n = 58; 10.7%). The overall median time to start antibiotic therapy was 214 minutes (95% CI: 199, 228), with a median length of ED stay (ED LOS) of 322 minutes (95% CI: 308, 335). We found large variations of time to start antibiotic treatment depending on hospital centre and type of infection. The diagnosis of a gastrointestinal infection was the most significant predictor for delay in antibiotic treatment (+119 minutes compared to patients with pneumonia; 95% CI: 58, 181; p<0.001). CONCLUSIONS: We found high variations in hospital ED performance in regard to start antibiotic treatment. The implementation of measures to reduce treatment times has the potential to improve patient care.

Pre-morbid glycemic control modifies the interaction between acute hypoglycemia and mortality.

PURPOSE: To study the impact of pre-morbid glycemic control on the association between acute hypoglycemia in intensive care unit (ICU) patients and subsequent hospital mortality in critically ill patients. METHODS: We performed a multicenter, multinational, retrospective observational study of patients with available HbA1c levels within the 3-month period preceding ICU admission. We separated patients into three cohorts according to pre-admission HbA1c levels (<6.5, 6.5-7.9, ≥8.0%, respectively). Based on published data, we defined a glucose concentration of 40-69 mg/dL (2.2-3.8 mmol/L) as moderate hypoglycemia and <40 mg/dL (<2.2 mmol/L) as severe hypoglycemia. We applied logistic regression analysis to study the impact of pre-morbid glycemic control on the relationship between acute hypoglycemia and mortality. RESULTS: A total of 3084 critically ill patients were enrolled in the study. Among these patients, with increasing HbA1c levels from <6.5, to 6.5-7.9, and to ≥8.0%, the incidence of both moderate (3.8, 11.1, and 16.4%, respectively; p < 0.001) and severe (0.9, 2.5, and 4.3%, respectively; p < 0.001) hypoglycemia progressively and significantly increased. The relationship between the occurrence of hypoglycemic episodes in the ICU and in-hospital mortality was independently and significantly affected by pre-morbid glucose control, as assessed by adjusted odds ratio (OR) and 95 % confidence interval (CI) for hospital mortality: (1) moderate hypoglycemia: in patients with <6.5, 6.5-7.9, and ≥8.0 % of HbA1c level-OR 0.54, 95% CI 0.25-1.16; OR 0.82, 95 % CI 0.33-2.05; OR 3.42, 95 % CI 1.29-9.06, respectively; (2) severe hypoglycemia: OR 1.50, 95% CI 0.42-5.33; OR 1.59, 95% CI 0.36-7.10; OR 23.46, 95% CI 5.13-107.28, respectively (interaction with pre-morbid glucose control, p = 0.009). We found that the higher the glucose level before admission to the ICU, the higher the mortality risk when patients experienced hypoglycemia. CONCLUSIONS: In critically ill patients, chronic pre-morbid hyperglycemia increases the risk of hypoglycemia and modifies the association between acute hypoglycemia and mortality.

Biomarkers from distinct biological pathways improve early risk stratification in medical emergency patients: the multinational, prospective, observational TRIAGE study.

INTRODUCTION: Early risk stratification in the emergency department (ED) is vital to reduce time to effective treatment in high-risk patients and to improve patient flow. Yet, there is a lack of investigations evaluating the incremental usefulness of multiple biomarkers measured upon admission from distinct biological pathways for predicting fatal outcome and high initial treatment urgency in unselected ED patients in a multicenter and multinational setting. METHOD: We included consecutive, adult, medical patients seeking ED care into this observational, cohort study in Switzerland, France and the USA. We recorded initial clinical parameters and batch-measured prognostic biomarkers of inflammation (pro-adrenomedullin [ProADM]), stress (copeptin) and infection (procalcitonin). RESULTS: During a 30-day follow-up, 331 of 7132 (4.6 %) participants reached the primary endpoint of death within 30 days. In logistic regression models adjusted for conventional risk factors available at ED admission, all three biomarkers strongly predicted the risk of death (AUC 0.83, 0.78 and 0.75), ICU admission (AUC 0.67, 0.69 and 0.62) and high initial triage priority (0.67, 0.66 and 0.58). For the prediction of death, ProADM significantly improved regression models including (a) clinical information available at ED admission (AUC increase from 0.79 to 0.84), (b) full clinical information at ED discharge (AUC increase from 0.85 to 0.88), and (c) triage information (AUC increase from 0.67 to 0.83) (p <0.01 for each comparison). Similarly, ProADM also improved clinical models for prediction of ICU admission and high initial treatment urgency. Results were robust in regard to predefined patient subgroups by center, main diagnosis, presenting symptoms, age and gender. CONCLUSIONS: Combination of clinical information with results of blood biomarkers measured upon ED admission allows early and more adequate risk stratification in individual unselected medical ED patients. A randomized trial is needed to answer the question whether biomarker-guided initial patient triage reduces time to initial treatment of high-risk patients in the ED and thereby improves patient flow and clinical outcomes. TRIAL REGISTRATION: ClinicalTrials.gov NCT01768494 . Registered January 9, 2013.

Optimizing triage and hospitalization in adult general medical emergency patients: the triage project.

BACKGROUND: Patients presenting to the emergency department (ED) currently face inacceptable delays in initial treatment, and long, costly hospital stays due to suboptimal initial triage and site-of-care decisions. Accurate ED triage should focus not only on initial treatment priority, but also on prediction of medical risk and nursing needs to improve site-of-care decisions and to simplify early discharge management. Different triage scores have been proposed, such as the Manchester triage system (MTS). Yet, these scores focus only on treatment priority, have suboptimal performance and lack validation in the Swiss health care system. Because the MTS will be introduced into clinical routine at the Kantonsspital Aarau, we propose a large prospective cohort study to optimize initial patient triage. Specifically, the aim of this trial is to derive a three-part triage algorithm to better predict (a) treatment priority; (b) medical risk and thus need for in-hospital treatment; (c) post-acute care needs of patients at the most proximal time point of ED admission. METHODS/DESIGN: Prospective, observational, multicenter, multi-national cohort study. We will include all consecutive medical patients seeking ED care into this observational registry. There will be no exclusions except for non-adult and non-medical patients. Vital signs will be recorded and left over blood samples will be stored for later batch analysis of blood markers. Upon ED admission, the post-acute care discharge score (PACD) will be recorded. Attending ED physicians will adjudicate triage priority based on all available results at the time of ED discharge to the medical ward. Patients will be reassessed daily during the hospital course for medical stability and readiness for discharge from the nurses and if involved social workers perspective. To assess outcomes, data from electronic medical records will be used and all patients will be contacted 30 days after hospital admission to assess vital and functional status, re-hospitalization, satisfaction with care and quality of life measures. We aim to include between 5000 and 7000 patients over one year of recruitment to derive the three-part triage algorithm. The respective main endpoints were defined as (a) initial triage priority (high vs. low priority) adjudicated by the attending ED physician at ED discharge, (b) adverse 30 day outcome (death or intensive care unit admission) within 30 days following ED admission to assess patients risk and thus need for in-hospital treatment and (c) post acute care needs after hospital discharge, defined as transfer of patients to a post-acute care institution, for early recognition and planning of post-acute care needs. Other outcomes are time to first physician contact, time to initiation of adequate medical therapy, time to social worker involvement, length of hospital stay, reasons for discharge delays, patient's satisfaction with care, overall hospital costs and patients care needs after returning home. DISCUSSION: Using a reliable initial triage system for estimating initial treatment priority, need for in-hospital treatment and post-acute care needs is an innovative and persuasive approach for a more targeted and efficient management of medical patients in the ED. The proposed interdisciplinary , multi-national project has unprecedented potential to improve initial triage decisions and optimize resource allocation to the sickest patients from admission to discharge. The algorithms derived in this study will be compared in a later randomized controlled trial against a usual care control group in terms of resource use, length of hospital stay, overall costs and patient's outcomes in terms of mortality, re-hospitalization, quality of life and satisfaction with care. TRIAL REGISTRATION: ClinicalTrials.gov Identifier, NCT01768494.

Using procalcitonin-guided algorithms to improve antimicrobial therapy in ICU patients with respiratory infections and sepsis.

PURPOSE OF REVIEW: In patients with systemic bacterial infections hospitalized in ICUs, the inflammatory biomarker procalcitonin (PCT) has been shown to aid diagnosis, antibiotic stewardship, and risk stratification. Our aim is to summarize recent evidence about the utility of PCT in the critical care setting and discuss the potential benefits and limitations of PCT when used for clinical decision-making. RECENT FINDINGS: A growing body of evidence supports PCT use to differentiate bacterial from viral respiratory infections (including influenza), to help risk stratify patients, and to guide decisions about optimal duration of antibiotic therapy. Different PCT protocols were evaluated for these and similar purposes in randomized controlled trials in patients with varying severities of predominantly respiratory tract infection and sepsis. These trials demonstrated effectiveness of monitoring PCT to de-escalate antibiotic treatment earlier without increasing rates of relapsing infections or other adverse outcomes. Although serial PCT measurement has shown value in risk stratification of ICU patients, PCT-guided antibiotic escalation protocols have not yet shown benefit for patients. SUMMARY: Inclusion of PCT data in clinical algorithms improves individualized decision-making regarding antibiotic treatment in patients in critical care for respiratory infections or sepsis. Future research should focus on use of repeated PCT measurements to risk-stratify patients and guide treatment to improve their outcomes.

Procalcitonin decrease over 72 hours in US critical care units predicts fatal outcome in sepsis patients.

INTRODUCTION: Close monitoring and repeated risk assessment of sepsis patients in the intensive care unit (ICU) is important for decisions regarding care intensification or early discharge to the ward. We studied whether considering plasma kinetics of procalcitonin, a biomarker of systemic bacterial infection, over the first 72 critical care hours improved mortality prognostication of septic patients from two US settings. METHODS: This retrospective analysis included consecutively treated eligible adults with a diagnosis of sepsis from critical care units in two independent institutions in Clearwater, FL and Chicago, IL. Cohorts were used for derivation or validation to study the association between procalcitonin change over the first 72 critical care hours and mortality. RESULTS: ICU/in-hospital mortality rates were 29.2%/31.8% in the derivation cohort (n=154) and 17.6%/29.4% in the validation cohort (n=102). In logistic regression analysis of both cohorts, procalcitonin change was strongly associated with ICU and in-hospital mortality independent of clinical risk scores (Acute Physiology, Age and Chronic Health Evaluation IV or Simplified Acute Physiology Score II), with area under the curve (AUC) from 0.67 to 0.71. When procalcitonin decreased by at least 80%, the negative predictive value for ICU/in-hospital mortality was 90%/90% in the derivation cohort, and 91%/79% in the validation cohort. When procalcitonin showed no decrease or increased, the respective positive predictive values were 48%/48% and 36%/52%. DISCUSSION: In septic patients, procalcitonin kinetics over the first 72 critical care hours provide prognostic information beyond that available from clinical risk scores. If these observations are confirmed, procalcitonin monitoring may assist physician decision-making regarding care intensification or early transfer from the ICU to the floor.

Effectiveness and safety of procalcitonin-guided antibiotic therapy in lower respiratory tract infections in "real life": an international, multicenter poststudy survey (ProREAL).

BACKGROUND: In controlled studies, procalcitonin (PCT) has safely and effectively reduced antibiotic drug use for lower respiratory tract infections (LRTIs). However, controlled trial data may not reflect real life. METHODS: We performed an observational quality surveillance in 14 centers in Switzerland, France, and the United States. Consecutive adults with LRTI presenting to emergency departments or outpatient offices were enrolled and registered on a website, which provided a previously published PCT algorithm for antibiotic guidance. The primary end point was duration of antibiotic therapy within 30 days. RESULTS: Of 1759 patients, 86.4% had a final diagnosis of LRTI (community-acquired pneumonia, 53.7%; acute exacerbation of chronic obstructive pulmonary disease, 17.1%; and bronchitis, 14.4%). Algorithm compliance overall was 68.2%, with differences between diagnoses (bronchitis, 81.0%; AECOPD, 70.1%; and community-acquired pneumonia, 63.7%; P < .001), outpatients (86.1%) and inpatients (65.9%) (P < .001), algorithm-experienced (82.5%) and algorithm-naive (60.1%) centers (P < .001), and countries (Switzerland, 75.8%; France, 73.5%; and the United States, 33.5%; P < .001). After multivariate adjustment, antibiotic therapy duration was significantly shorter if the PCT algorithm was followed compared with when it was overruled (5.9 vs 7.4 days; difference, -1.51 days; 95% CI, -2.04 to -0.98; P < .001). No increase was noted in the risk of the combined adverse outcome end point within 30 days of follow-up when the PCT algorithm was followed regarding withholding antibiotics on hospital admission (adjusted odds ratio, 0.83; 95% CI, 0.44 to 1.55; P = .56) and regarding early cessation of antibiotics (adjusted odds ratio, 0.61; 95% CI, 0.36 to 1.04; P = .07). CONCLUSIONS: This study validates previous results from controlled trials in real-life conditions and demonstrates that following a PCT algorithm effectively reduces antibiotic use without increasing the risk of complications. Preexisting differences in antibiotic prescribing affect compliance with antibiotic stewardship efforts. TRIAL REGISTRATION: isrctn.org Identifier: ISRCTN40854211.

Role of procalcitonin in managing adult patients with respiratory tract infections.

Respiratory infections remain the most common reason why patients seek medical care in ambulatory and hospital settings, and they are the most frequent precursor of sepsis. In light of the limitations of clinical signs and symptoms and traditional microbiologic diagnostics for respiratory infections, blood biomarkers that correlate with the presence and extent of bacterial infections may provide additional useful information to improve diagnostic and prognostic efforts and help with therapeutic decisions in individual patients. A growing body of evidence supports the use of procalcitonin (PCT) to differentiate bacterial from viral respiratory diagnoses, to help risk stratify patients, and to guide antibiotic therapy decisions about initial need for, and optimal duration of, therapy. Although still relatively new on the clinical frontier, a series of randomized controlled trials have evaluated PCT protocols for antibiotic-related decision making and have included patients from different clinical settings and with different severities of respiratory infection. In these trials, initial PCT levels were effective in guiding decisions about the initiation of antibiotic therapy in lower-acuity patients, and subsequent measurements were effective for guiding duration of therapy in higher-acuity patients, without apparent harmful effects. Recent European respiratory infection guidelines now also recognize this concept. As with any other laboratory test, PCT should not be used on a stand-alone basis. Rather, it must be integrated into clinical protocols, together with clinical, microbiologic data and with results from clinical risk scores. The aim of this review is to summarize recent evidence about the usefulness of PCT in patients with lower respiratory tract infections and to discuss the potential benefits and limitations of this marker when used for clinical decision making.