Predicting patient outcomes, futility, and resource utilization in the intensive care unit: the role of severity scoring systems and general outcome prediction models

Serum osmolarity does not predict mortality in patients with respiratory failure

We aimed to determine the parameters that affect mortality in pulmonary intensive care units that are faster and inexpensive to determine than existing scoring systems. The relationship between serum osmolarity and prognosis was demonstrated for predialysis patients, in acute pulmonary embolism, heart failure, acute coronary syndrome, myocardial infarction, and acute spontaneous intracerebral hemorrhage in the literature. We hypothesized that serum osmolarity, which is routinely evaluated, may have prognostic significance in patients with respiratory failure.This study comprised 449 patients treated in the Pulmonary Intensive Care Clinic (PICU) of our hospital between January 1, 2020, and December 31, 2020. The modified Charlson Comorbidity Index (mCCI), Acute Physiology and Chronic Health Assessment (APACHE II), Sequential Organ Failure Evaluation Score (SOFA), Nutrition Risk Screening 2002 (NRS-2002), and hospitalization serum osmolarity levels were measured.Of the 449 patients included in the study, 65% (n = 292) were female and the mean age of all patients was 69.86 ± 1.72 years. About 83.1% (n = 373) of the patients included in the study were discharged with good recovery. About 4.9% (n = 22) were transferred to the ward because their intensive care needs were over. About 6.9% (n = 31) were transferred to the tertiary intensive care unit after their status deteriorated. About 5.1% (n = 23) died in the PICU. In the mortality group, APACHE II (P = .005), mCCI (P < .001), NRS-2002 total score (P < .001), and SOFA score (P < .001) were significantly higher. There was no statistically significant difference between the groups in terms of serum osmolarity levels.Although we could not determine serum osmolarity as a practical method to predict patient prognosis in this study, we assume that our results will guide future studies on this subject.

Comparison of prognosis predictive value of 4 disease severity scoring systems in patients with acute respiratory failure in intensive care unit: A STROBE report

Various disease severity scoring systems were currently used in critically ill patients with acute respiratory failure, while their performances were not well investigated.The study aimed to investigate the difference in prognosis predictive value of 4 different disease severity scoring systems in patients with acute respiratory failure.With a retrospective cohort study design, adult patients admitted to intensive care unit (ICU) with acute respiratory failure were screened and relevant data were extracted from an open-access American intensive care database to calculate the following disease severity scores on ICU admission: acute physiology score (APS) III, Sequential Organ Failure Assessment score (SOFA), quick SOFA (qSOFA), and Oxford Acute Severity of Illness Score (OASIS). Hospital mortality was chosen as the primary outcome. Multivariable logistic regression analyses were performed to analyze the association of each scoring system with the outcome. Receiver operating characteristic curve analyses were conducted to evaluate the prognosis predictive performance of each scoring system.A total of 4828 patients with acute respiratory failure were enrolled with a hospital mortality rate of 16.78%. APS III (odds ratio [OR] 1.03, 95% confidence interval [CI] 1.02-1.03), SOFA (OR 1.15, 95% CI 1.12-1.18), qSOFA (OR 1.26, 95% CI 1.11-1.42), and OASIS (OR 1.06, 95% CI 1.05-1.08) were all significantly associated with hospital mortality after adjustment for age and comorbidities. Receiver operating characteristic analyses showed that APS III had the highest area under the curve (AUC) (0.703, 95% CI 0.683-0.722), and SOFA and OASIS shared similar predictive performance (area under the curve 0.653 [95% CI 0.631-0.675] and 0.664 [95% CI 0.644-0.685], respectively), while qSOFA had the worst predictive performance for predicting hospital mortality (0.553, 95% CI 0.535-0.572).These results suggested the prognosis predictive value varied among the 4 different disease severity scores for patients admitted to ICU with acute respiratory failure.

Survival and predictors of mortality among patients admitted to the intensive care units in southern Ethiopia: A multi-center cohort study

BACKGROUND

The burden of life-threatening conditions requiring intensive care units has grown substantially in low-income countries related to an emerging pandemic, urbanization, and hospital expansion. The rate of ICU mortality varied from region to region in Ethiopia. However, the body of evidence on ICU mortality and its predictors is uncertain. This study was designed to investigate the pattern of disease and predictors of mortality in Southern Ethiopia.

METHODS

After obtaining ethical clearance from the Institutional Review Board (IRB), a multi-center cohort study was conducted among three teaching referral hospital ICUs in Ethiopia from June 2018 to May 2020. Five hundred and seventeen Adult ICU patients were selected. Data were entered in Statistical Package for Social Sciences version 22 and STATA version 16 for analysis. Descriptive statistics were run to see the overall distribution of the variables. Chi-square test and odds ratio were determined to identify the association between independent and dependent variables. Multivariate analysis was conducted to control possible confounders and identify independent predictors of ICU mortality.

RESULTS

The mean (±SD) of the patients admitted in ICU was 34.25(±5.25). The overall ICU mortality rate was 46.8%. The study identified different independent predictors of mortality. Patients with cardiac arrest were approximately 12 times more likely to die as compared to those who didn't, AOR = 11.9(95% CI:6.1 to 23.2).

CONCLUSION

The overall mortality rate in ICU was very high as compared to other studies in Ethiopia as well as globally which entails a rigorous activity from different stakeholders.

Development and validation of high definition phenotype-based mortality prediction in critical care units

Objectives

The objectives of this study are to construct the high definition phenotype (HDP), a novel time-series data structure composed of both primary and derived parameters, using heterogeneous clinical sources and to determine whether different predictive models can utilize the HDP in the neonatal intensive care unit (NICU) to improve neonatal mortality prediction in clinical settings.

Materials and Methods

A total of 49 primary data parameters were collected from July 2018 to May 2020 from eight level-III NICUs. From a total of 1546 patients, 757 patients were found to contain sufficient fixed, intermittent, and continuous data to create HDPs. Two different predictive models utilizing the HDP, one a logistic regression model (LRM) and the other a deep learning long–short-term memory (LSTM) model, were constructed to predict neonatal mortality at multiple time points during the patient hospitalization. The results were compared with previous illness severity scores, including SNAPPE, SNAPPE-II, CRIB, and CRIB-II.

Results

A HDP matrix, including 12 221 536 minutes of patient stay in NICU, was constructed. The LRM model and the LSTM model performed better than existing neonatal illness severity scores in predicting mortality using the area under the receiver operating characteristic curve (AUC) metric. An ablation study showed that utilizing continuous parameters alone results in an AUC score of >80% for both LRM and LSTM, but combining fixed, intermittent, and continuous parameters in the HDP results in scores >85%. The probability of mortality predictive score has recall and precision of 0.88 and 0.77 for the LRM and 0.97 and 0.85 for the LSTM.

Conclusions and Relevance

The HDP data structure supports multiple analytic techniques, including the statistical LRM approach and the machine learning LSTM approach used in this study. LRM and LSTM predictive models of neonatal mortality utilizing the HDP performed better than existing neonatal illness severity scores. Further research is necessary to create HDP–based clinical decision tools to detect the early onset of neonatal morbidities.

Characterizing Early Inpatient Death After Trauma

BACKGROUND

There is a paucity of data to predict early death or futility after trauma. The objective of this study was to characterize the laboratory values, blood product administration, and hospital disposition for patients with trauma who died within 72 h of admission.

METHODS

All deaths within 72 h of admission over a 5-y period at a level I trauma center were reviewed. Blood transfusion within the first 4 h of arrival and patient disposition from the emergency department to the operating room (OR), surgical intensive care unit, or the neuroscience intensive care unit (NSICU) were analyzed. Kaplan-Meier curves were generated to determine time to death.

RESULTS

A total of 622 subjects were identified; 39.5% died in the emergency department, 10.6% went directly to the OR, 13.6% were admitted to the surgical intensive care unit, and 29.7% admitted to the NSICU. Of these subjects, 201 (32.2%) patients received blood within the first 4 h. By 24 h, early blood transfusion was associated with more rapid death for patients who were admitted to the NSICU (80% versus 60% mortality, P = 0.01) but not for patients taken directly to the OR (80% versus 70% mortality, P = 0.2). Admission coagulopathy by international normalized ratio (P < 0.01), but not anemia (P = 0.64) or acidosis (P = 0.45), correlated with a shorter time to death. In contrast, laboratory values obtained at 4 h after admission did not correlate with time to death.

CONCLUSIONS

Our data demonstrate that admission coagulation derangement and need for early blood product transfusion are the two factors most associated with early death after injury, particularly in those patients with traumatic brain injury. These data will help construct future models for futility of continued care in patients with trauma.

Use of a laboratory only score system to define trajectories and outcomes of older people admitted to the acute hospital as medical emergencies

AIM

Increasing numbers of older people are admitted to hospital as medical emergencies. They are a heterogeneous population with uncertain trajectories and outcomes. Our aim was to retrospectively characterize subgroups of older inpatients based on their acuity trajectories.

METHODS

This was a single-center patient series from St James's Hospital Dublin, Ireland (2002-2010). The Medical Admissions Risk System (MARS) score was used to classify a sample of 14,607 patients aged ≥ 65 years, from admission to end of episode, into four trajectory groups: (i) static high acuity (group 1); (ii) static low acuity (group 2); (iii) inpatient deterioration (group 3); and (iv) inpatient improvement (group 4). K-means cluster analysis was used for the classification.

RESULTS

Group 1 (4.1%): median length of stay (LOS) 7.4 days, 23.6% used intensive care, mortality rate 79.2%; sepsis and renal failure were the dominant features. Group 2 (76.6%): median LOS 8.0 days, 5.2% used intensive care, mortality rate 9.5%; younger age, low comorbidity and diseases of non-vital organs were predominant. Group 3 (7.6%): median LOS 17.2 days, 17.4% used intensive care, mortality rate 76.1%; high comorbidity and sepsis/respiratory disease featured. Group 4 (11.7%): median LOS 12.1 days, 12.8% used intensive care, mortality rate 22.7%; sepsis and renal/metabolic disease were frequent, and comorbidity levels were intermediate.

CONCLUSIONS

In older acute medical inpatients, the outcome seemed more driven by specific diagnoses (such as sepsis and renal failure) and comorbidity, than by age. Using the MARS score to retrospectively categorize older inpatients might help to understand their heterogeneity and promote the design of appropriate care pathways.

Development of inpatient risk stratification models of acute kidney injury for use in electronic health records

OBJECTIVE

Patients with hospital-acquired acute kidney injury (AKI) are at risk for increased mortality and further medical complications. Evaluating these patients with a prediction tool easily implemented within an electronic health record (EHR) would identify high-risk patients prior to the development of AKI and could prevent iatrogenically induced episodes of AKI and improve clinical management.

METHODS

The authors used structured clinical data acquired from an EHR to identify patients with normal kidney function for admissions from 1 August 1999 to 31 July 2003. Using administrative, computerized provider order entry and laboratory test data, they developed a 3-level risk stratification model to predict each of 2 severity levels of in-hospital AKI as defined by RIFLE criteria. The severity levels were defined as 150% or 200% of baseline serum creatinine. Model discrimination and calibration were evaluated using 10-fold cross-validation.

RESULTS

Cross-validation of the models resulted in area under the receiver operating characteristic (AUC) curves of 0.75 (150% elevation) and 0.78 (200% elevation). Both models were adequately calibrated as measured by the Hosmer-Lemeshow goodness-of-fit test chi-squared values of 9.7 (P = 0.29) and 12.7 (P = 0.12), respectively.

CONCLUSIONS

The authors generated risk prediction models for hospital-acquired AKI using only commonly available electronic data. The models identify patients at high risk for AKI who might benefit from early intervention or increased monitoring.

A revised method to assess intensive care unit clinical performance and resource utilization

OBJECTIVE

In 1994, Rapoport et al. published a two-dimensional graphical tool for benchmarking intensive care units (ICUs) using a Mortality Probability Model (MPM0-II) to assess clinical performance and a Weighted Hospital Days scale (WHD-94) to assess resource utilization. MPM0-II and WHD-94 do not calibrate on contemporary data, giving users of the graph an inflated assessment of their ICU's performance. MPM0-II was recently updated (MPM0-III) but not the model for predicting resource utilization. The objective was to develop a new WHD model and revised Rapoport-Teres graph.

DESIGN

Multicenter cohort study.

SETTING

One hundred thirty-five ICUs in 98 hospitals participating in Project IMPACT.

PATIENTS

Patients were 124,855 MPM0-II eligible Project IMPACT patients treated between March 2001 and June 2004.

INTERVENTIONS

None.

MEASUREMENTS AND MAIN RESULTS

WHD was redefined as 4 units for the first day of each ICU stay, 2.5 units for each additional ICU day, and 1 unit for each non-ICU day after the first ICU discharge. Stepwise linear regression was used to construct a model to predict ICU-specific log average WHD from 39 candidate variables available in Project IMPACT. The updated WHD model has four independent variables: percent of patients dying in the hospital, percent of unscheduled surgical patients, percent of patients on mechanical ventilation within 1 hr of ICU admission, and percent discharged from the ICU to an external post-acute care facility. The first three variables increase average WHD and the last decreases it. The new model has good performance (R = 0.47) and, when combined with MPM0-II, provides a well-calibrated Rapoport-Teres graph.

CONCLUSIONS

A new WHD model has been derived from a large, contemporary critical care database and, when used with MPM0-III, updates a popular method for benchmarking ICUs. Project IMPACT participants will likely perceive a decline in their ICU performance coordinates due to the recalibrated graph and should instead focus on their unit's performance relative to their peers.

Effects of SVM parameter optimization on discrimination and calibration for post-procedural PCI mortality

Support vector machines (SVM) have become popular among machine learning researchers, but their applications in biomedicine have been somewhat limited. A number of methods, such as grid search and evolutionary algorithms, have been utilized to optimize model parameters of SVMs. The sensitivity of the results to changes in optimization methods has not been investigated in the context of medical applications. In this study, radial-basis kernel SVM and polynomial kernel SVM mortality prediction models for percutaneous coronary interventions were optimized using (a) mean-squared error, (b) mean cross-entropy error, (c) the area under the receiver operating characteristic, and (d) the Hosmer-Lemeshow goodness-of-fit test (HL chi(2)). A threefold cross-validation inner and outer loop method was used to select the best models using the training data, and evaluations were based on previously unseen test data. The results were compared to those produced by logistic regression models optimized using the same indices. The choice of optimization parameters had a significant impact on performance in both SVM kernel types.

Prognosis in critical care

Prognostic risk prediction models have been employed in the intensive care unit (ICU) setting since the 1980s and provide health care providers with important information to help inform decisions related to treatment and prognosis, as well as to compare outcomes across institutions. Prognostic models for critical care are among the most widely utilized and tested predictive models in healthcare. In this article, we review and compare mortality prediction models, including the APACHE (1981), SAPS (1984), APACHE-II (1985), MPM (1987), APACHE-III (1991), SAPS-II (1993), and MPM-II (1993). We emphasize the importance of model calibration in this domain. In addition, we present a brief review of the statistical methodology, multiple logistic regression, which underlies most of the models currently used in critical care.

Predictors of intensive care unit admission and related outcome for patients after pancreaticoduodenectomy

High-volume centers have low morbidity and mortality after pancreaticoduodenectomy (PD). Less is known about treatment pathways and their influence on intensive care unit (ICU) utilization. Patients who underwent PD at a tertiary cancer center during the five-year period between January 1998 and December 2003 were identified from a prospective database. Preoperative and intraoperative factors relating to ICU admission and outcome were analyzed. Five hundred ninety-one pancreaticoduodenectomies were performed during the study period. Of these, 536 patients had complete records for analysis. Of the 536 patients, 51 (10%) were admitted to the ICU after surgery. Admission to the ICU was associated with decreased overall survival (P < .0001). Of the preoperative predictors of ICU admission, serum creatinine, albumin, and increased body mass index (BMI) were associated with ICU admission (P = .02, .05, and .002, respectively). Age, blood glucose, diagnosis of diabetes mellitus, and chronic obstructive pulmonary disease were not predictive of ICU admission on univariate analysis. Of the intraoperative factors, longer operative time and estimated blood loss (EBL) correlated with ICU admission (P = .003 and .0001, respectively). On multivariate analysis, only preoperative BMI and intraoperative EBL were independent predictors of ICU admission (P = .03 and .003, respectively). Patients with a preoperative BMI greater than 30 had a substantially higher risk of ICU admission (relative risk 2.4). The majority of patients who undergo PD do not require admission to the ICU. Factors most associated with ICU admission after PD are increased preoperative BMI and intraoperative blood loss.