Tool For Predicting Mortality Research Articles

Promoting early goals of care conversations in the CICU with a surprise question-based EHR workflow

BackgroundThe Surprise Question (SQ) - Would you be surprised if this patient died within the next 6 months? - is a validated tool for mortality prediction. The Mount Sinai Cardiac Intensive Care Unit (CICU) incorporated the SQ into a novel EHR workflow to identify patients who would benefit from early initiation of Palliative Care (PC).MethodsImplementation of the SQ proceeded in two steps. During the feasibility pilot (December 2021-March 2022), providers answered the SQ using an EXCEL spreadsheet for all CICU patients, without changing other workflows. In April 2022, the CICU launched a new workflow-column built into the Epic patient-list dashboard with the SQ as the backbone. For patients with SQ answers of “NO,” providers were prompted to facilitate and document a goals of care (GOC) conversation. We conducted a retrospective, observational, quasi-experimental study of all admissions to the CICU with SQ = NO between December 2021-September 2022. Clinical data was obtained via EHR query and chart review. We compared the frequency and timing of GOC conversations and the likelihood of redirected GOC (defined as code status change and/or hospice discharge) during the 3-month pilot versus the 6-month implementation period.Results195 admissions were included: median [IQR] age 72.0 [61.0, 84.0] years; LOS > 5 days 43.6%; CICU mortality 17.9%. These clinical characteristics were comparable between the pilot (N = 57) and implementation (N = 138) periods. However, ICU interventions (i.e. mechanical ventilation, renal replacement therapy) were more common among the pilot cohort (52.6% vs. 33.3%, p = .015). For the primary outcomes, compared to the pilot period, there was a significantly higher frequency of GOC conversations (61.4% vs. 81.2%, p = .004) and GOC conversations < 2 days from CICU admission (40.4% vs. 61.6%, p = .007) in the intervention period. There was no difference in the likelihood of redirected GOC towards comfort or no escalation (28.1% vs. 21.0%, p = .288).ConclusionWe facilitated earlier GOC conversations directed to critically ill patients with high mortality risk by integrating the SQ into the EHR.

Early risk stratification of patients with acute myocardial infarction complicated by severe cardiogenic shock - The ECLS-SHOCK Risk Score

Abstract Background Prognosis after acute myocardial infarction complicated by cardiogenic shock (AMI-CS) remains dismal. Timely risk stratification is crucial to make adequate treatment decisions. Existing CS scores are limited to either registry data, included not readily available parameters (such as coronary flow after revascularisation, cystatin C or interleukin-6), or are not tailored to patients with AMI-CS. Purpose The aim of this study was to develop a readily available and easy-to-use risk prediction tool for 30-day mortality in patients with severe AMI-CS derived from the ECLS-SHOCK (Extracorporeal Life Support in Infarct-Related Cardiogenic Shock) trial. Methods Backward stepwise regression analysis was used to develop the risk score across 43 centres in two countries. Results Five variables emerged as independent predictors for 30-day mortality and were used as risk score parameters: age ≥69 years, female sex, ≥3 known cardiovascular risk factors (i.e., hypertension, dyslipidaemia, diabetes, smoking or known vascular disease), baseline arterial lactate ≥7.1 mmol/L and performed resuscitation. Between 1 and 7 points were attributed to each variable, resulting in a score with three risk categories: low (0 to 1), intermediate (2 to 3) and high (4 to 7) (Figure). The observed 30-day mortality rates were 23%, 50% and 80% in each risk category (p&amp;lt;0.001) with a good discrimination at an area under the curve of 0.75. A stepwise increase in mortality between the different risk score categories was observed in Kaplan-Meier analyses (p&amp;lt;0.001 for low vs. intermediate, intermediate vs. high and low vs. high; see Figure). Conclusion The ECLS-SHOCK risk score can be easily and promptly calculated in daily clinical practice and correlated with mortality in patients with severe AMI-CS. It may help stratify patient risk for short-term mortality and facilitate early clinical decision making.

Predictive values and correlation of CRIB Score II and perfusion index in assessment of severity of illness in sick preterm neonates: An observational study.

Perfusion index (PI) and CRIB (Clinical risk index of babies) Score II both are minimally invasive tools for prediction of mortality and morbidity in sick neonates. This study aims to know the predictive values of both PI and CRIB Score II in assessment of severity of illness in preterm neonates of 28 to 32 weeks and their correlation. PI and CRIB II Score of 125 sick preterm neonates admitted in NICU were assessed within 12 hours of admission. Severity noted in form of outcome. Correlation between PI and CRIB Score II were observed. Receiver operating characteristic (ROC) curve were plotted for PI and CIRB Score II with outcome (discharged vs expired). Area under curve for perfusion index was 0.776 at 95% confidence interval. Optimum cutoff point based on ROC curve was 1.65 with sensitivity of 84%., specificity of 37.2% with positive predictive value of 70% and negative predictive value of 57%. Area under curve for CRIB Score II was 0.622 (p value < 0.028), optimum cut off point based on ROC curve for CRIB II score was 3.5 with sensitivity 79.10% of specificity of 40% positive predictive value is 43.03% and negative predictive value of 76.9%. Spearman's correlation coefficient between PI and CRIB II score was -0.272 and p value is 0.002. This study concluded that perfusion index (p < 0.001) is better for predicting the severity compared to CRIB II score (p < 0.028). A weak negative correlation exists between PI and CRIB II score (r = -0.272).

Impact of a mortality prediction tool on end-of-life (EOL) quality measures.

404 Background: Determining treatment options for patients with cancer can be difficult for providers, especially when it comes to deciding the appropriate time to stop therapy. Many factors come into play when making important treatment and quality of life decisions including cancer responsiveness to treatment, patient preferences, labs, and diagnostic studies. Providers use their experience, instinct, and patient input in making these decisions, and can still have a difficult time knowing the best option for the patient. To support providers in identifying patients at high risk for poor outcomes, we developed a 90-day mortality risk machine-learning model (MRM) for patients with metastatic cancer. The tool was designed to facilitate earlier advance care planning (ACP) discussions and prognostic awareness leading to alignment of treatment with patient goals and values. The US Oncology Network (The Network) implemented the MRM beginning in early 2021 through late 2022. We assessed the impact of the MRM using performance on 5 EOL Quality Measures included in Medicare’s Merit-based Incentive Payment System (MIPS) Program. Methods: MRM predictions were made every two weeks and provided to practices through a secure web-based dashboard. Practices customized their use of the MRM based on their unique workflow. MRM utilization was collected July – December 2023 and determined how frequently practices accessed the dashboard. Practices were separated into two groups based on MRM utilization: High Utilizers (HU) used the tool more than once a week and Low Utilizers (LU) did not use the tool or used it less than or equal to once a week. EOL MIPS measure performance was collected and evaluated from the 2023 calendar year. For each of the EOL MIPS measures, we compared the average score between the two practice groups. A 2-sample unequal variance t-test with 1-tailed distribution was used to determine if there was a significant difference between the two practice groups for each EOL MIPS measure. Results: A total of 25 practices had available EOL MIPS scores. Five practices were classified as a HU group and 20 were in the LU group. EOL MIPS measure scores were higher among HU. All results trended towards improvement with a statistically significant difference observed for PIMSH 1 and MIPS 457 for HU compared to LU. The table shows the average MIPS scores across the two practice groups. Conclusions: Practices with the highest reported utilization of the MRM also reported improved MIPS EOL quality scores for the 2023 calendar year. While each practice implemented MRM independently, those using the tool more than once per week had improved performance in their MIPS EOL quality scores. EOL MIPS Measure HU LU Delta P MIPS 047 ACP 80.7 59.8 20.9 0.066 PIMSH 1 ACP Stage 4 disease 67.7 40.2 27.5 0.002 MIPS 453 Chemo Last 14 days (inv) 10.7 11.7 -1.0 0.279 MIPS 457 Hospice &lt;3 days (inv) 11.7 14.8 -3.1 0.038 PIMSH 9 Supportive RX last 14 days (inv) 5.3 6.6 -1.3 0.119

Comparing NEWS2, TRISS, RTS, SI, GAP, and MGAP in predicting early and total mortality rates in trauma patients based on emergency department data set: A diagnostic study

IntroductionIn the global context, trauma remains a significant cause of mortality, particularly among individuals under 50 years old with mortality and morbidity. Timely assessment, efficient triage, and appropriate post-trauma care can help mitigate long-term mortality and morbidity risks. While the trauma and injury severity score (TRISS) is considered accurate, its complexity limits clinical utility, prompting the development of new tools like the National Early Warning Score 2 (NEWS2) and others. Thus, this study aims to evaluate the predictive abilities of NEWS2, RTS (revised trauma score), TRISS, SI (shock index), GAP (Glasgow coma score, age, and systolic blood pressure), and MGAP (mechanism of injury, Glasgow coma score, age, and systolic blood pressure) scoring systems in trauma patient care to enhance outcomes and reduce mortality and morbidity rates. MethodsThis retrospective diagnostic study included a total of 6894 trauma patients with data collected from the emergency department. Patient information, including vital signs and demographic data, was recorded on datasheets completed by ED staff. Ethical guidelines were followed, and patient consent forms were not required due to the study's retrospective nature. Scoring systems such as NEWS2, RTS, TRISS, SI, GAP, and MGAP were calculated and interpreted based on specific formulas and scoring charts. ResultsOf 6894 patients, 2.83% died during hospitalization, and 0.72% within the first 24 hours. Factors associated with mortality included age, cause of injury, duration of hospitalization, and various physiological parameters. The majority of scoring systems are reliable in predicting mortality, except for the sock index (SI), with NEWS2 demonstrating the highest sensitivity and accuracy for predicting mortality rates, while TRISS and RTS were the most specific. GAP was also the most competent predictor of overall mortality. ConclusionMost scoring systems effectively predict mortality, with TRISS and NEWS2 demonstrating high sensitivity and specificity together in forecasting in-hospital mortality. NEWS2 stands out as a precise tool for mortality prediction in the emergency department due to its simplicity and quick application. However, further research is needed to identify the most effective predictive system in trauma patients.

SP2.9 - National Emergency Laparotomy Audit (NELA) and American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) risk prediction calculators are not accurate in predicting 30-day mortality in patients aged 80 and above

Abstract Background 30- to 50000 Emergency Laparotomies (EL) are performed annually in the UK. More than half of patients are aged 70 or over and have a high post-operative complication rate along with a 16-20% risk of 30-day mortality (30-dm). To date, the literature lacks a dedicated risk prediction tool for mortality and morbidity following EL in the elderly population. Methods A multi-center, retrospective study of patients aged ≥80 undergoing EL examined the correlation between predicted 30-dm using NELA and ACS-NSQIP and actual 30-dm. This occurred in two UK centers over a 2-year period. The primary outcome was defined as the correlation between actual 30-dm and NELA or ACS-NSQIP predicted 30-dm. GraphPad Prism 10.1.1. was used for statistical analysis. Results 141 cases were included in the final analysis (Table 1). The actual 30-dm was 17.73%, NELA predicted 30-dm 13.07% (7.21;27.84), and ACS-NSQIP 9.4% (3.8;20.6). NELA correlated positively with ACS-NSQIP predicted 30-dm (R²=0.79; p&amp;lt;0.0001). No association was found between actual 30-dm and predicted NELA 30-dm, R²=0.01; p&amp;lt;0.0001 or ACS-NSQIP, R²=0.01; p&amp;lt;0.0001. All data is expressed as Median + IQR. Conclusion In octo- and nonagenarians, following an emergency laparotomy, both NELA and ACS-NSQIP are not accurate in predicting 30-day mortality.Table 1Male/female ratio0.63Age (years)84 (81;87)BMI (kg/m²)24.27 (21.43;27)Lactate (most recent preoperative value) mmol/L1.7 mmol/L (1.14;3.01)ASA grade3 (2;4)Clinical frailty score3 (3;5)Charlson Comorbidity Index4 (2;5)

Machine Learning-Based Predictive Model for Mortality in Female Breast Cancer Patients Considering Lifestyle Factors.

To construct a free and accurate breast cancer mortality prediction tool by incorporating lifestyle factors, aiming to assist healthcare professionals in making informed decisions. In this retrospective study, we utilized a ten-year follow-up dataset of female breast cancer patients from a major Chinese hospital and included 1,390 female breast cancer patients with a 7% (96) mortality rate. We employed six machine learning algorithms (ridge regression, k-nearest neighbors, neural network, random forest, support vector machine, and extreme gradient boosting) to construct a mortality prediction model for breast cancer. This model incorporated significant lifestyle factors, such as postsurgery sexual activity, use of totally implantable venous access ports, and prosthetic breast wear, which were identified as independent protective factors. Meanwhile, ten-fold cross-validation demonstrated the superiority of the random forest model (average AUC = 0.918; 1-year AUC = 0.914, 2-year AUC = 0.867, 3-year AUC = 0.883). External validation further supported the model's robustness (average AUC = 0.782; 1-year AUC = 0.809, 2-year AUC = 0.785, 3-year AUC = 0.893). Additionally, a free and user-friendly web tool was developed using the Shiny framework to facilitate easy access to the model. Our breast cancer mortality prediction model is free and accurate, providing healthcare professionals with valuable information to support their clinical decisions and potentially promoting healthier lifestyles for breast cancer patients.

BAUSSS biomarker improves melanoma survival risk assessment.

The American Joint Committee on Cancer (AJCC) method of staging melanoma is dated and inaccurate. It ignores important prognostic melanoma features, especially the patient's age. BAUSSS is more accurate in determining survival risk for primary cutaneous melanoma patients who have no clinical or imaging evidence of nodal or distant metastases. BAUSSS is an algorithm incorporating analysis of Breslow thickness, Age, Ulceration, Subtype of melanoma, Sex and Site. These are the six features from the patient history along with the details from the melanoma pathology report that are most predictive of mortality outcome. To develop a single-page document that allows the clinician to determine BAUSSS biomarker-predicted prognosis in consultation with the patient. From various data sources, we developed an algorithm to predict melanoma mortality using the BAUSSS biomarker system. The single-page algorithm was made available to download at https://globalmelanoma.net/bausss-survival-chart, thus being readily available without charge to all clinicians and their patients. BAUSSS method of determining melanoma prognosis is more accurate and less costly than the AJCC staging system. The only surgery the patient requires is wide local excision of the primary tumour. This method of ascertaining melanoma risk does not require added surgery, costs, hospitalization, tests and anaesthesia, such as would be required if sentinel lymph node biopsy was undertaken. BAUSSS can be a useful tool in determining which primary melanoma patients are at sufficiently high risk to be considered for adjuvant drug therapy. We encourage clinicians to download and print in colour this single-page BAUSSS mortality prediction tool, laminate it, and use it face to face with the patient in consultations. Not only will the patient be able to recognize his/her long-term prognosis but will also be able to see how their tumour severity compares with others.

Can peripheral perfusion index predict disease mortality in COVID-19 patients in the emergency department

BackgroundCoronavirus disease 2019 (COVID-19) pneumonia remains a major public health concern. The prognostic efficacy of Peripheral Perfusion Index (PPI) has been researched in different pathologies such as trauma and sepsis. We hypothesized that PPI may serve as predictor of mortality in hospitalized patients with COVID-19 infection. This study aimed to describe the association between PPI at admission and COVID-19 mortality, a new mortality prediction tool. MethodsThis retrospective, observational study was conducted at a tertiary care center in Turkey. Adult patients diagnosed with COVID-19 infection were enrolled in this study between Februrary 15, 2022 to April 15, 2023. Patient demographic and clinical data including vital signs, laboratory parameters and PPI on admission were collected from an electronic database. PPI was measured using Philips G30E patient monitor system. The primary outcome was in-hospital mortality. ResultsIn total, 200 patients with COVID-19 infection were included and 42 (21 %) in-hospital deaths were identified. For all parameters of study, age, oxygen saturation, respiratory rate, PPI, urea, creatinine, White Blood Cell (WBC), and High-sensitive cardiac Troponin T (hs-cTnT) values were significantly different between survivors vs non-survivors. hs-cTnT >21,25 pg/mL[HR:2.823 (95 % CI:1.211–6583)], PPI <2,15 [HR:2485 (95 % CI:1.194–5.175)], Oxygen saturation <87 % [HR:2258 (95 % CI:1.191–4.282)], and WBC >9680 x103/ml [HR:2.124 (95 % CI:1.083–4.163)] were independent predictors of in-hospital mortality. ConclusionsThis study identified the factors affecting in-hospital mortality among COVID-19 patients. Importantly, besides many parameter, PPI at admission was significantly associated with COVID-19 mortality and could be a feasible marker in emergency department to identify high risk patients.

Clinical Characteristics, Predictors for Mortality and Comparison of the Birmingham Vasculitis Activity Score and the Five-Factor Score on Survival in ANCA-Associated Vasculitis in Hong Kong

Objective: To describe the clinical profile and predictors of mortality of antineutrophil cytoplasmic antibody (ANCA)-associated vasculitis (AAV) patients in Hong Kong. To compare the accuracy of the latest Five-Factor Score (FFS-2009) and the Birmingham Vasculitis Activity Score (BVAS) in prediction of survival with this local cohort. Methods: A retrospective observational study on newly diagnosed AAV patients, from January 1, 2011 to March 31, 2022, managed in the Kowloon West Cluster (KWC) hospitals in Hong Kong. Demographic and baseline characteristics, clinical profile, and treatment profile were reviewed. Factors associated with mortality were analyzed with the Cox proportional hazards model. The performances of FFS and BVAS in mortality prediction were analyzed by receiver operating characteristic (ROC) curves. Results: A total of 83 AAV patients were included in the study. The median age was 70.5 years at diagnosis. Microscopic polyangiitis (MPA; 69.9%) was the most common AAV subtype. The median FFS and BVAS were 2 and 20, respectively. The overall mortality was 45.6% across the study period. Multivariate Cox regression identified age at diagnosis (HR 1.043, [Formula: see text]), stabilized peak serum creatinine (HR 1.002, [Formula: see text]), hemoglobin level (HR 0.754, [Formula: see text]), cardiac involvement (HR 3.862, [Formula: see text]), and use of maintenance therapy (HR 0.261, [Formula: see text]) as independent predictors of overall survival. Both FFS and BVAS were significant predictors of overall survival. The areas under the curve (AUC) of ROC curves suggested FFS was a good prediction tool for early mortality in 1 year, with an AUC value of 0.874. Conclusion: Despite the advances in treatment, AAV still carried significant morbidities with high mortality. Clinical predictors and existing scoring systems showed good predictive power on mortality.

External validation of a 2-year all-cause mortality prediction tool developed using machine learning in patients with stage 4-5 chronic kidney disease.

Chronic kidney disease (CKD) is associated with increased mortality. Individual mortality prediction could be of interest to improve individual clinical outcomes. Using an independent regional dataset, the aim of the present study was to externally validate the recently published 2-year all-cause mortality prediction tool developed using machine learning. A validation dataset of stage 4 or 5 CKD outpatients was used. External validation performance of the prediction tool at the optimal cutoff-point was assessed by the area under the receiver operating characteristic curve (AUC-ROC), accuracy, sensitivity, and specificity. A survival analysis was then performed using the Kaplan-Meier method. Data of 527 outpatients with stage 4 or 5 CKD were analyzed. During the 2years of follow-up, 91 patients died and 436 survived. Compared to the learning dataset, patients in the validation dataset were significantly younger, and the ratio of deceased patients in the validation dataset was significantly lower. The performance of the prediction tool at the optimal cutoff-point was: AUC-ROC = 0.72, accuracy = 63.6%, sensitivity = 72.5%, and specificity = 61.7%. The survival curves of the predicted survived and the predicted deceased groups were significantly different (p < 0.001). The 2-year all-cause mortality prediction tool for patients with stage 4 or 5 CKD showed satisfactory discriminatory capacity with emphasis on sensitivity. The proposed prediction tool appears to be of clinical interest for further development.

670 Investigating Delay During Presentation of Trauma in Young Adults: A UK Major Trauma Centre Retrospective Audit and a Systematic Review of the Literature

Abstract Introduction The current evidence in major trauma highlights that delays at multiple points along the patient care pathway, from incident to surgery, are associated with decreased survival. The aim of this audit was to assess performance across select delay parameters in juveniles aged 16-18 presenting to an adult MTC service following a systematic review of the literature. Method Patient records were interrogated between 1/10/2022-1/3/2023 and audited against 3 national standards: (1) Ambulance Response and Transfer (7 and 45mins respectively) (2) Tranexamic Acid infusion (&amp;lt;3hrs) and (3) CT scan (&amp;lt;60mins). A systematic search was also performed using EMBASE and OVID databases (01/01/2005-/01/11/2023). All studies assessing impact of time to theatre in patients with acute trauma were considered. Results 32 consecutive patients (3F:29M) (median 18 years[16-18]) were identified within the audit period. 40%(n=13) presented following penetrating injury and 35%(n=11) following a road traffic accident. Mean time from incident to admission was 83mins[±36], from incident to tranexamic acid infusion 50mins[±35], and from admission to CT Trauma Series 61mins[+/-100]. Median length of stay was 4 days [IQR=5]. The systematic search yielded 25 studies which were included in the analysis. An Incident2Surgery framework is presented which highlights key time-critical points for diagnosis and intervention. Conclusions Local compliance to national standards for tranexamic acid infusion in appropriate patients and performance of trauma CT was excellent with poor compliance to ambulance arrival and transfer time. The proposed Incident2Surgery framework could be incorporated into mortality prediction and audit tools and assist in the assessment of interventions.

Predicting mortality within 1 year of ART initiation in children and adolescents living with HIV in sub-Saharan Africa: a retrospective observational cohort study

Differentiated service delivery (DSD) for children and adolescents living with HIV can improve targeted resource use. We derived a mortality prediction score to guide clinical decision making for children and adolescents living with HIV. Data for this retrospective observational cohort study were evaluated for all children and adolescents living with HIV and initiating antiretroviral therapy (ART); aged 0-19 years; and enrolled at Baylor clinics in Eswatini, Malawi, Lesotho, Tanzania, and Uganda between 2005 and 2020. Data for clinical prediction, including anthropometric values, physical examination, ART, WHO stage, and laboratory tests were captured at ART initiation. Backward stepwise variable selection and logistic regression were performed to develop predictive models for mortality within 1 year of ART initiation. Probabilities of mortality were generated, compared with true outcomes, internally validated, and evaluated against WHO advanced HIV criteria. The study population included 16 958 children and adolescents living with HIV and initiated on ART between May 18, 2005, and Dec 18, 2020. Predictive variables for the most accurate model included: age, CD4 percentage, white blood cell count, haemoglobin concentration, platelet count, and BMI Z score as continuous variables, and WHO clinical stage and oedema, abnormal muscle tone and respiratory distress on examination as categorical variables. The area under the curve (AUC) of the predictive model was 0·851 (95% CI 0·839-0·863) in the training set and 0·822 (0·800-0·845) in the test set, compared with 0·606 (0·595-0·617) for the WHO advanced HIV criteria (p<0·0001). This study evaluated a large, multinational population to derive a mortality prediction tool for children and adolescents living with HIV. The model more accurately predicted clinical outcomes than the WHO advanced HIV criteria and has the potential to improve DSD for children and adolescents living with HIV in high-burden settings. National Institute of Health Fogarty International Center.

Proof-of-concept for an automatable mortality prediction scoring in hospitalised older adults

IntroductionIt is challenging to prognosticate hospitalised older adults. Delayed recognition of end-of-life leads to failure in delivering appropriate palliative care and increases healthcare utilisation. Most mortality prediction tools specific for older adults require additional manual input, resulting in poor uptake. By leveraging on electronic health records, we aim to create an automatable mortality prediction tool for hospitalised older adults.MethodsWe retrospectively reviewed electronic records of general medicine patients ≥75 years at a tertiary hospital between April–September 2021. Demographics, comorbidities, ICD-codes, age-adjusted Charlson Comorbidity Index (CCI), Hospital Frailty Risk Score, mortality and resource utilization were collected. We defined early deaths, late deaths and survivors as patients who died within 30 days, 1 year, and lived beyond 1 year of admission, respectively. Multivariate logistic regression analyses were adjusted for age, gender, race, frailty, and CCI. The final prediction model was created using a stepwise logistic regression.ResultsOf 1,224 patients, 168 (13.7%) died early and 370 (30.2%) died late. From adjusted multivariate regression, risk of early death was significantly associated with ≥85 years, intermediate or high frail risk, CCI &amp;gt; 6, cardiovascular risk factors, AMI and pneumonia. For late death, risk factors included ≥85 years, intermediate frail risk, CCI &amp;gt;6, delirium, diabetes, AMI and pneumonia. Our mortality prediction tool which scores 1 point each for age, pneumonia and AMI had an AUC of 0.752 for early death and 0.691 for late death.ConclusionOur mortality prediction model is a proof-of-concept demonstrating the potential for automated medical alerts to guide physicians towards personalised care for hospitalised older adults.

Looking Beyond Mortality Prediction: Primary Care Physician Views of Patients' Palliative Care Needs Predicted by a Machine Learning Tool.

To assess primary care physicians' (PCPs) perception of the need for serious illness conversations (SIC) or other palliative care interventions in patients flagged by a machine learning tool for high 1-year mortality risk. We surveyed PCPs from four Brigham and Women's Hospital primary care practice sites. Multiple mortality prediction algorithms were ensembled to assess adult patients of these PCPs who were either enrolled in the hospital's integrated care management program or had one of several chronic conditions. The patients were classified as high or low risk of 1-year mortality. A blinded survey had PCPs evaluate these patients for palliative care needs. We measured PCP and machine learning tool agreement regarding patients' need for an SIC/elevated risk of mortality. Of 66 PCPs, 20 (30.3%) participated in the survey. Out of 312 patients evaluated, 60.6% were female, with a mean (standard deviation [SD]) age of 69.3 (17.5) years, and a mean (SD) Charlson Comorbidity Index of 2.80 (2.89). The machine learning tool identified 162 (51.9%) patients as high risk. Excluding deceased or unfamiliar patients, PCPs felt that an SIC was appropriate for 179 patients; the machine learning tool flagged 123 of these patients as high risk (68.7% concordance). For 105 patients whom PCPs deemed SIC unnecessary, the tool classified 83 as low risk (79.1% concordance). There was substantial agreement between PCPs and the tool (Gwet's agreement coefficient of 0.640). A machine learning mortality prediction tool offers promise as a clinical decision aid, helping clinicians pinpoint patients needing palliative care interventions.

Immediate Operative Trauma Assessment Score: A Simple and Reliable Predictor of Mortality in Trauma Patients Undergoing Urgent/Emergent Surgery.

Objective: Many current trauma mortality prediction tools are either too intricate or rely on data not readily available during a trauma patient's initial evaluation. Moreover, none are tailored to those necessitating urgent or emergent surgery. Our objective was to design a practical, user-friendly scoring tool using immediately available variables, and then compare its efficacy to the widely-known Revised Trauma Score (RTS). Methods: The adult 2017-2021 Trauma Quality Improvement Program (TQIP) database was queried to identify patients ≥18 years old undergoing any urgent/emergent operation (direct from Emergency Department to operating room). Patients were divided into derivation and validation groups. A three-step methodology was used. First, multiple logistic regression models were created to determine risk of death using only variables available upon arrival. Second, the weighted average and relative impact of each independent predictor was used to derive an easily calculated Immediate Operative Trauma Assessment Score (IOTAS). We then validated IOTAS using AUROC and compared it to RTS. Results: From 249208 patients in the derivation-set, 14635 (5.9%) died. Age ≥65, Glasgow Coma Scale score <9, hypotension (SBP <90mmHg), and tachycardia (>120/min) on arrival were identified as independent predictors for mortality. Using these, the IOTAS was structured, offering scores between 0-8. The AUROC for this was .88. A clear escalation in mortality was observed across scores: from 4.4% at score 1 to 60.5% at score 8. For the validation set (250182 patients; mortality rate 5.8%), the AUROC remained consistent at .87, surpassing RTS's AUROC of .83. Conclusion: IOTAS is a novel, accurate, and now validated tool that is intuitive and efficient in predicting mortality for trauma patients requiring urgent or emergent surgeries. It outperforms RTS, and thereby may help guide clinicians when determining the best course of action in patient management as well as counseling patients and their families.

External validation of the 2-year mortality prediction tool in hemodialysis patients developed using a Bayesian network.

In recent years, a number of predictive models have appeared to predict the risk of medium-term mortality in hemodialysis patients, but only one, limited to patients aged over 70 years, has undergone sufficiently powerful external validation. Recently, using a national learning database and an innovative approach based on Bayesian networks and 14 carefully selected predictors, we have developed a clinical prediction tool to predict all-cause mortality at 2 years in all incident hemodialysis patients. In order to generalize the results of this tool and propose its use in routine clinical practice, we carried out an external validation using an independent external validation database. A regional, multicenter, observational, retrospective cohort study was conducted to externally validate the tool for predicting 2-year all-cause mortality in incident and prevalent hemodialysis patients. This study recruited a total of 142 incident and 697 prevalent adult hemodialysis patients followed up in one of the eight Association pour l'Utilisation du Rein Artificiel dans la région Lyonnaise (AURAL) Alsace dialysis centers. In incident patients, the 2-year all-cause mortality prediction tool had an area under the receiver curve (AUC-ROC) of 0.73, an accuracy of 65%, a sensitivity of 71% and a specificity of 63%. In prevalent patients, the performance for the external validation were similar in terms of AUC-ROC, accuracy and specificity, but was lower in term of sensitivity. The tool for predicting all-cause mortality at 2 years, developed using a Bayesian network and 14 routinely available explanatory variables, obtained satisfactory external validation in incident patients, but sensitivity was insufficient in prevalent patients.

Modified blood cell GAP model as a prognostic biomarker in idiopathic pulmonary fibrosis.

The Gender, Age and Physiology (GAP) model is a simple mortality prediction tool in patients with idiopathic pulmonary fibrosis that uses demographic and physiological variables available at initial evaluation. White blood cell variables may have associations with idiopathic pulmonary fibrosis outcomes. We evaluated whether incorporating blood cell counts in modified GAP (cGAP) models would improve outcome prediction in patients with idiopathic pulmonary fibrosis. This retrospective analysis included pooled data from phase 3 randomised trials of pirfenidone in idiopathic pulmonary fibrosis (ASCEND, CAPACITY 004, CAPACITY 006). Study outcomes (disease progression, all-cause mortality, all-cause hospitalisation, respiratory-related hospitalisation) were evaluated during the initial 1-year period. Shared frailty models were used to evaluate associations between continuous and categorical baseline white and red blood cell parameters and study outcomes in a bivariate context, and to evaluate the impact of adding continuous monocyte count (cGAP1) or white and red blood cell parameters (cGAP2) to traditional GAP variables in a multivariable context based on C-statistics changes. Data were pooled from 1247 patients (pirfenidone, n=623; placebo, n=624). Significant associations (bivariate analyses) were idiopathic pulmonary fibrosis progression with neutrophil and eosinophil counts; all-cause mortality with monocyte and neutrophil counts; all-cause hospitalisation with monocyte count, neutrophil count and haemoglobin level; and respiratory-related hospitalisation with monocyte count, neutrophil count and haemoglobin level. In multivariate analyses, C-statistics were highest for the cGAP2 model for each of the outcomes. Modified GAP models incorporating monocyte counts alone or plus other white and red blood cell variables may be useful to improve prediction of outcomes in patients with idiopathic pulmonary fibrosis.

A novel prediction tool for mortality in patients with acute lower gastrointestinal bleeding requiring emergency hospitalization: a large multicenter study

The study aimed to identify prognostic factors for patients with acute lower gastrointestinal bleeding and to develop a high-accuracy prediction tool. The analysis included 8254 cases of acute hematochezia patients who were admitted urgently based on the judgment of emergency physicians or gastroenterology consultants (from the CODE BLUE J-study). Patients were randomly assigned to a derivation cohort and a validation cohort in a 2:1 ratio using a random number table. Assuming that factors present at the time of admission are involved in mortality within 30 days of admission, and adding management factors during hospitalization to the factors at the time of admission for mortality within 1 year, prognostic factors were established. Multivariate analysis was conducted, and scores were assigned to each factor using regression coefficients, summing these to measure the score. The newly created score (CACHEXIA score) became a tool capable of measuring both mortality within 30 days (ROC-AUC 0.93) and within 1 year (C-index, 0.88). The 1-year mortality rates for patients classified as low, medium, and high risk by the CACHEXIA score were 1.0%, 13.4%, and 54.3% respectively (all P < 0.001). After discharge, patients identified as high risk using our unique predictive score require ongoing observation.

A machine learning approach to predict mortality due to immune-mediated thrombotic thrombocytopenic purpura

BackgroundMortality due to immune-mediated thrombotic thrombocytopenic purpura (iTTP) remains significant. Predicting mortality risk may potentially help individualize treatment. The French Thrombotic Microangiopathy (TMA) Reference Score has not been externally validated in the United States. Recent advances in machine learning technology can help analyze large numbers of variables with complex interactions for the development of prediction models. ObjectivesTo validate the French TMA Reference Score in the United States Thrombotic Microangiopathy (USTMA) iTTP database and subsequently develop a novel mortality prediction tool, the USTMA TTP Mortality Index. MethodsWe analyzed variables available at the time of initial presentation, including demographics, symptoms, and laboratory findings. We developed our model using gradient boosting machine, a machine learning ensemble method based on classification trees, implemented in the R package gbm. ResultsIn our cohort (n = 419), the French score predicted mortality with an area under the receiver operating characteristic curve of 0.63 (95% CI: 0.50-0.77), sensitivity of 0.35, and specificity of 0.84. Our gradient boosting machine model selected 8 variables to predict acute mortality with a cross-validated area under the receiver operating characteristic curve of 0.77 (95% CI: 0.71-0.82). The 2 cutoffs corresponded to sensitivities of 0.64 and 0.50 and specificities of 0.76 and 0.87, respectively. ConclusionThe USTMA Mortality Index was acceptable for predicting mortality due to acute iTTP in the USTMA registry, but not sensitive enough to rule out death. Identifying patients at high risk of iTTP-related mortality may help individualize care and ultimately improve iTTP survival outcomes. Further studies are needed to provide external validation. Our model is one of many recent examples where machine learning models may show promise in clinical prediction tools in healthcare.