BACKGROUND:While clinical benefits of guideline-directed medical therapy in patients with heart failure (HF) are well established, acute hemodynamic changes after initiation of these agents are not well described. Wireless pulmonary artery pressure (PAP) monitoring using implantable sensors is ideally suited to determine changes in pressure following medication titration. METHODS:We compared PAP from 7 days pre-initiation to the 7-day period following 30 days of newly initiated treatment with core HF therapies: angiotensin receptor-neprilysin inhibitors, mineralocorticoid receptor antagonists, sodium-glucose co-transporter-2 inhibitors, and β-blockers in patients implanted with a PAP sensor (CardioMEMS, Abbott) as part of the GUIDE-HF study (Hemodynamic-Guided Management of HF). Patients with changes in loop diuretics or other core HF therapies during the 30-day interval were excluded from the analysis. RESULTS:Of 2358 patients in the study, 50 patients had new angiotensin receptor-neprilysin inhibitor initiation, 97 patients had new sodium-glucose co-transporter-2 inhibitor initiation, 112 patients had new mineralocorticoid receptor antagonist initiation, and 33 had new β-blocker initiation at least 30 days post-implant, continued to take the medication for at least 30 days, and were included in this analysis. Following angiotensin receptor-neprilysin inhibitor or mineralocorticoid receptor antagonist initiation, diastolic PAP decreased by -1.56 (95% CI, -3.54 to -0.40) and -1.15 (95% CI, -1.82 to -0.47) mm Hg, respectively, in the absence of changes in loop diuretics or other core HF therapies. By contrast, there was no detectable change in PAP following the new initiation of the sodium-glucose co-transporter-2 inhibitor. A trend to higher PAP was observed after β-blocker initiation though differences were not significant. CONCLUSIONS:These results may inform the selection of pharmacological therapy to manage elevated filling pressures during hemodynamic-guided therapy. REGISTRATION:URL: https://www.clinicaltrials.gov; Unique identifier: NCT03387813.

BACKGROUND AND HYPOTHESIS:Serum creatinine frequently increases during hospitalizations for acute heart failure (AHF) meeting criteria for acute kidney injury (AKI), but the significance of this increase is variable. We evaluated whether admission levels of kidney tubule and cardiac biomarkers are associated with increases in serum creatinine and adverse in-hospital events. METHODS:In a nested case-control, we identified 214 cases with at least stage 1 AKI defined by the 2012 Kidney Disease Improving Global Outcomes criteria within seven days of admission in the Acute Kidney Injury Neutrophil Gelatinase-Associated Lipocalin Evaluation of Symptomatic Heart Failure Study and matched with 214 controls who did not experience AKI. The primary outcome was the development of AKI within the first seven days; secondary outcomes were the development of severe AKI (stages 2 or 3) and a composite of adverse in-hospital events. Associations of 6 blood and 14 urine biomarkers were assessed with logistic regression and Receiver Operating Characteristic Area Under the Curve (ROC-AUC) curve analysis. RESULTS:Individuals were 71 ± 13 years; 64% were men; with median admission creatinine 1.3 [IQR 1.0-1.8] mg/dL. Among AKI cases, 181 (85%), 16 (8%), and 17 (8%) were stages 1, 2, and 3, respectively. Compared to the lowest tertile, only the highest tertiles of fractional excretion of sodium (FeNa ≥ 3.65; OR 1.7, 95% CI 1.0-2.9) and urinary MCP-1 (≥ 433 ng/g; OR 1.7, 95% CI 1.0-2.8) were associated with AKI risk, though discrimination was poor (ROC-AUC < 0.60). Neither was associated with severe AKI. Only elevated B-type natriuretic peptide (≥ 932 pg/mL) predicted the composite adverse in-hospital event (OR 2.3, 95% CI 1.2-4.2). CONCLUSION:Kidney tubule and cardiac biomarkers at admission for AHF are not associated with risk for increases in creatinine or adverse in-hospital events reaffirming that creatinine changes in AHF are largely functional in nature.

Background Loop diuretics require functioning kidney tubules to increase the excretion of sodium into urine thereby relieving congestion in patients with acute heart failure (AHF). Whether biomarkers of kidney tubular health are associated with the natriuretic response of loop diuretics in AHF is unknown. Methods Spot urine sodium and 14 urine biomarkers of tubular health were measured after intravenous loop diuretic administration on the day of enrollment in 418 patients with AHF from AKINESIS (Acute Kidney Injury Neutrophil Gelatinase–Associated Lipocalin [NGAL] Evaluation of Symptomatic Heart Failure Study). The association of each biomarker with urine sodium concentration was assessed with multivariable linear regression. Results The mean age was 71±12 years, 64% were men, average estimated glomerular filtration rate was 56±24 mL/min per 1.73 m2, and urine sodium was 94±48 mEq/L. Higher urine concentrations of α‐1‐microglobulin (−5.0 mEq/L [95% CI, −8.3 to −1.6 mEq/L]), insulin‐like growth factor–binding protein‐7 (IGFBP‐7 mEq/L; −4.1 [95% CI, −8.1 to 0.0 mEq/L]), tissue inhibitor of metalloproteinases 2 (TIMP‐2; −3.6 mEq/L [95% CI, −6.5 to −0.7 mEq/L]), IGFBP‐7*TIMP‐2 product (−5.1 mEq/L [95% CI, −6.6 to −3.7 mEq/L]), interleukin 18 (−8.2 mEq/L [95% CI, −13.2 to −3.1 mEq/L]), kidney injury molecule 1 (−2.4 mEq/L [95% CI, −4.4 to −0.3 mEq/L]), and chemokine ligand 14 (−2.3 mEq/L [95% CI, −4.0 to −0.5 mEq/L]) were associated with lower urine sodium concentration after adjustment for confounders including estimated glomerular filtration rate and urine albumin. Conclusions Urine biomarkers of tubular health are associated with lower natriuretic response to loop diuretics in patients with AHF. Kidney tubule health may be an underappreciated contributor to diuretic responsiveness in AHF.

Background Whether variability of B-type natriuretic peptide (BNP) values between races affects its clinical integration as a marker for congestion and predictor of prognosis in acute heart failure (AHF) remains unknown. We aimed to investigate the relationship between diuretic therapy, change in BNP value, and prognosis in AHF in relation to racial differences. Methods This analysis combined data from the AKINESIS and REALITY-AHF studies. We included White, Black, and Asian individuals admitted with AHF requiring intravenous diuretic therapy. We examined the relative change in BNP values at 48 h post hospital admission, and its association with diuretic therapy and one-year mortality. Results Of 1380 participants, 29 % were White, 12 % were Black, and 58 % were Asian. Admission BNP values were highest in Black, followed by Asian and White individuals. After adjusting for confounding factors, Black individuals had significantly higher admission BNP values compared to White individuals. During the first 48 h of hospitalization, Asian individuals received the lowest diuretic dose but demonstrated the greatest diuretic response and BNP decrease. After adjustment for confounding factors, Asian individuals were more likely to have a BNP decrease compared to White individuals. Higher admission BNP values predicted higher one-year mortality in White and Asian but not in Black individuals (p for interaction = 0.021). BNP decrease was associated with a lower one-year mortality without a significant interaction by race. Conclusions In AHF patients, admission BNP was higher in Black, and its decrease after diuretic therapy was greater in Asian individuals. A BNP decrease predicted a better prognosis, regardless of race.

BACKGROUND:Kidney dysfunction, defined by measures of glomerular health, in patients hospitalized with acute heart failure (HF) is associated with death and HF readmission. We aimed to determine if kidney tubule damage and dysfunction are associated with these outcomes in acute HF. METHODS:In AKINESIS (Acute Kidney Injury Neutrophil Gelatinase-Associated Lipocalin [NGAL] Evaluation of Symptomatic Heart Failure Study), 218 individuals admitted with acute HF experiencing acute kidney injury were matched with 218 individuals without acute kidney injury. Fourteen urine tubular damage and dysfunction biomarkers were measured at hospital admission in this case-control cohort. Associations between biomarkers and the composite outcome of death or HF readmission, death alone, and HF readmission alone were evaluated adjusting for confounders including kidney and cardiac biomarkers. RESULTS:The mean age was 71±12 years, 64% were men, and mean admission estimated glomerular filtration rate was 55±23 mL/min per 1.73 m2. There were 156 deaths or HF readmissions, 87 deaths, and 92 HF readmissions over 1 year. Each 2-fold higher level of IGFBP-7 (insulin-like growth factor binding protein-7), IGFBP-7*TIMP-2 (tissue inhibitor of metalloproteinases-2) product, KIM-1 (kidney injury molecule-1), and MCP-1 (monocyte chemoattractant protein-1) was associated with a 1.36 (95% CI, 1.13-1.64), 1.08 (95% CI, 1.01-1.14), 1.12 (95% CI, 1.02-1.24), and 1.18 (95% CI, 1.03-1.34) higher risk of death or HF readmission, respectively. IGFBP-7 was associated with death alone, whereas MCP-1 and CCL-14 (C-C motif chemokine ligand-14) were associated with HF readmission alone. CONCLUSIONS:Biomarkers of kidney tubular health are associated with risk of death and HF readmission among people admitted with acute HF independent of measures of glomerular function and cardiac risk.

BACKGROUND:Worsening renal function (WRF) is common in hospitalized patients being treated for acute heart failure. However, discriminating clinically significant WRF remains challenging. In patients hospitalized with acute heart failure, we evaluated if blood and urine biomarkers of cardiac and kidney dysfunction were associated with adverse outcomes. METHODS:We identified 175 of 927 participants in the AKINESIS study (Acute Kidney Neutrophil Gelatinase-Associated Lipocalin Evaluation of Symptomatic Heart Failure Study) who met criteria for stage 1 or 2 Kidney Disease: Improvement Global Outcomes acute kidney injury during the first 3 days of hospitalization. We measured 24 blood and urine biomarkers from specimens collected within 24 hours of meeting acute kidney injury criteria. The primary composite outcome consisted of worsening WRF (higher acute kidney injury stage), need for dialysis, or death at 30 days. Biomarkers' association with the composite outcome was assessed with logistic regression by tertiles and area under the curve (AUC). RESULTS:Of the 175 participants, 32 (18%) developed the primary composite outcome. Only history of chronic kidney disease was significantly different between those with and without the composite outcome. The highest tertile of plasma Gal-3 (galectin-3) and urine epidermal growth factor were associated with increased odds of the composite outcome compared with the lowest tertile in unadjusted analyses. After adjusting for serum creatinine, systolic blood pressure, and blood urea nitrogen, only the highest tertile of Gal-3 was associated with greater odds of the composite outcome (odds ratio, 4.6 [95% CI, 1.4-16.0). Gal-3 had the highest AUC (0.70 [95% CI, 0.58-0.82]), while epidermal growth factor had a lower AUC (0.63 [95% CI, 0.53-0.74]). Notably, urine biomarkers of kidney tubule injury were not associated with the composite outcome. CONCLUSIONS:Tubular injury does not occur in most patients with acute heart failure experiencing WRF, consistent with the functional mechanisms of WRF in this patient population. REGISTRATION:URL: https://www.clinicaltrials.gov/study/NCT01291836?term=NCT01291836&rank=1; Unique identifier: NCT01291836.

BACKGROUND Trials evaluating implantable hemodynamic monitors to manage patients with heart failure (HF) have shown reductions in HF hospitalizations but not mortality. Prior meta -analyses assessing mortality have been limited in construct because of an absence of patient-level data, short -term follow-up duration, and evaluation across the combined spectrum of ejection fractions. OBJECTIVES The purpose of this meta -analysis was to determine whether management with implantable hemodynamic monitors reduces mortality in patients with heart failure and reduced ejection fraction (HFrEF) and to confirm the effect of hemodynamic-monitoring guided management on HF hospitalization reduction reported in previous studies. METHODS The patient-level pooled meta -analysis used 3 randomized studies (GUIDE -HF [Hemodynamic-Guided Management of Heart Failure], CHAMPION [CardioMEMS Heart Sensor Allows Monitoring of Pressure to Improve Outcomes in NYHA Class III Heart Failure Patients], and LAPTOP-HF [Left Atrial Pressure Monitoring to Optimize Heart Failure Therapy]) of implantable hemodynamic monitors (2 measuring pulmonary artery pressures and 1 measuring left atrial pressure) to assess the effect on all -cause mortality and HF hospitalizations. RESULTS A total of 1,350 patients with HFrEF were included. Hemodynamic-monitoring guided management significantly reduced overall mortality with an HR of 0.75 (95% CI: 0.57-0.99); P 1/4 0.043. HF hospitalizations were significantly reduced with an HR of 0.64 (95% CI: 0.55-0.76); P < 0.0001. CONCLUSIONS Management of patients with HFrEF using an implantable hemodynamic monitor significantly reduces both mortality and HF hospitalizations. The reduction in HF hospitalizations is seen early in the first year of monitoring and mortality benefits occur after the first year. (J Am Coll Cardiol 2024;83:682-694) (c) 2024 Published by Elsevier on behalf of the American College of Cardiology Foundation.

Background Higher cardiac troponin is associated with worse outcomes in patients with acute heart failure. The significance of repeat measurements over hours remains unclear. We assessed whether a repeat measurement and the Δ between measurements of high‐sensitivity cardiac troponin I (hs‐cTnI) were associated with outcomes in hypervolemic patients with acute heart failure without acute coronary syndrome. Methods and Results We analyzed 582 individuals from AKINESIS (Acute Kidney Injury Neutrophil Gelatinase‐Associated Lipocalin Evaluation of Symptomatic Heart Failure Study) with hs‐cTnI measured ≤12 hours from admission and repeated ≤6 hours thereafter. Associations between hs‐cTnI levels and their Δ with short‐term (death, intensive care unit admission, receipt of inotropes, or positive pressure ventilation during hospitalization) and long‐term (death or heart failure readmission within 1 year) outcomes were assessed. The average age was 69±13 years, 62% were men, 65% were White, 46% had coronary artery disease, and 22% had chest pain. Median hs‐cTnI levels were 27 (interquartile range [IQR], 13–62) ng/L initially and 28 (IQR, 14–68) ng/L subsequently, with a Δ of 0 [IQR, −2 to 4] ng/L over 3.4±1 hours. Only the second measurement was associated with short‐term outcomes (odds ratio, 1.14 per 2‐fold higher [95% CI, 1.02–1.28]). Both individual measurements and the Δ were associated with long‐term outcomes (hazard ratios, 1.09, 1.12, and 1.16 for first, second, and Δ, respectively). Associated risk for the first and second measurements were not constant over the year but highest early after being measured and decreased over 1 year. Conclusions Repeat measurements of hs‐cTnI over hours can identify individuals with acute heart failure without acute coronary syndrome at risk for short‐ and long‐term outcomes.

BACKGROUND:Body mass index (BMI) is a known confounder for natriuretic peptides, but its influence on other biomarkers is less well described. We investigated whether BMI interacts with biomarkers' association with prognosis in patients with acute heart failure (AHF). METHODS AND RESULTS:B-type natriuretic peptide (BNP), high-sensitivity cardiac troponin I (hs-cTnI), galectin-3, serum neutrophil gelatinase-associated lipocalin (sNGAL), and urine NGAL were measured serially in patients with AHF during hospitalization in the AKINESIS (Acute Kidney Injury Neutrophil gelatinase-associated lipocalin Evaluation of Symptomatic Heart Failure) study. Cox regression analysis was used to determine the association of biomarkers and their interaction with BMI for 30-day, 90-day and 1-year composite outcomes of death or HF readmission. Among 866 patients, 21.2%, 29.7% and 46.8% had normal (18.5-24.9 kg/m2), overweight (25-29.9 kg/m2) or obese (≥ 30 kg/m2) BMIs on admission, respectively. Admission values of BNP and hs-cTnI were negatively associated with BMI, whereas galectin-3 and sNGAL were positively associated with BMI. Admission BNP and hs-cTnI levels were associated with the composite outcome within 30 days, 90 days and 1 year. Only BNP had a significant interaction with BMI. When BNP was analyzed by BMI category, its association with the composite outcome attenuated at higher BMIs and was no longer significant in obese individuals. Findings were similar when evaluated by the last-measured biomarkers and BMIs. CONCLUSIONS:In patients with AHF, only BNP had a significant interaction with BMI for the outcomes, with its association attenuating as BMI increased; hs-cTnI was prognostic, regardless of BMI.

BACKGROUND In patients with symptomatic heart failure (HF) and previous heart failure hospitalization (HFH), hemodynamic-guided HF management using a wireless pulmonary artery pressure (PAP) sensor reduces HFH, but it is unclear whether these benefits extend to patients who have not been recently hospitalized but remain at risk because of elevated natriuretic peptides (NPs). OBJECTIVES This study assessed the efficacy and safety of hemodynamic-guided HF management in patients with elevated NPs but no recent HFH. METHODS In the GUIDE-HF (Hemodynamic-Guided Management of Heart Failure) trial, 1,000 patients with New York Heart Association (NYHA) functional class II to IV HF and either previous HFH or elevated NP levels were randomly assigned to hemodynamic-guided HF management or usual care. The authors evaluated the primary study composite of all-cause mortality and total HF events at 12 months according to treatment assignment and enrollment stratum (HFH vs elevated NPs) by using Cox proportional hazards models. RESULTS Of 999 evaluable patients, 557 were enrolled on the basis of a previous HFH and 442 on the basis of elevated NPs alone. Those patients enrolled by NP criteria were older and more commonly White persons with lower body mass index, lower NYHA class, less diabetes, more atrial fibrillation, and lower baseline PAP. Event rates were lower among those patients in the NP group for both the full follow-up (40.9 per 100 patient-years vs 82.0 per 100 patient-years) and the pre-COVID-19 analysis (43.6 per 100 patient-years vs 88.0 per 100 patient-years). The effects of hemodynamic monitoring were consistent across enrollment strata for the primary endpoint over the full study duration (interaction P = 0.71) and the pre-COVID-19 analysis (interaction P = 0.58). CONCLUSIONS Consistent effects of hemodynamic-guided HF management across enrollment strata in GUIDE-HF support consideration of hemodynamic monitoring in the expanded group of patients with chronic HF and elevated NPs without recent HFH. (Hemodynamic-Guided Management of Heart Failure [GUIDE-HF]; NCT03387813) (c) 2023 Published by Elsevier on behalf of the American College of Cardiology Foundation.

BACKGROUND: Several different B-type natriuretic peptide (BNP) assays are used clinically for diagnostic and prognostic evaluation of heart failure (HF). BNP binds weakly to neprilysin and is cleaved in multiple areas adjacent to the binding sites for the antibodies used in these immunoassays. We assessed the changes in BNP following neprilysin inhibition as measured by 3 immunoassays that recognize different epitopes. METHODS: Among 130 participants with HF with reduced ejection fraction, blood was collected prior to treatment with sacubitril/valsartan (sac/val) and then repeatedly measured through 52 weeks of treatment. BNP concentrations were measured with 3 widely used BNP assays (Siemens, Abbott, and Quidel). RESULTS: Study participants had a mean age of 65 +/- 13 years and 76% were men. The median BNP concentration at baseline was 133 ng/L by the Siemens assay, 127 ng/L by the Abbott assay, and 141 ng/L by the Quidel assay. Following initiation of sac/val, there were significantly greater declines in BNP measured by Quidel and Abbott (P = 0.009 and P < 0.001), respectively (both with N-terminal capture antibodies), compared to Siemens (with C-terminal capture antibodies). The difference from baseline was not statistically significant until after week 12 (mean -10.1% for Quidel and -14.3% for Abbott) compared to non-significant differences before 12 weeks (mean -4.5% for Quidel and -6.0% for Abbott). CONCLUSIONS: Following initiation of sac/val, BNP measurements may modestly differ depending on the assay method used, particularly after a few months of treatment. Whether these differences relate to neprilysin-mediated degradation of antibody binding sites deserves further study. STUDY REGISTRATION: PROVE-HF Identifier: NCT02887183.

BACKGROUND:Cardiac troponin (cTn) can be elevated in many patients presenting to the emergency department (ED) with chest pain but without a diagnosis of acute coronary syndrome (ACS). We compared the prognostic significance of cTn in these different populations.METHODS:We retrospectively analyzed the CHOPIN study, which enrolled patients who presented to the ED with chest pain. Patients were grouped as ACS, non-ACS cardiovascular disease, noncardiac chest pain and chest pain not otherwise specified (NOS). We examined the prognostic ability of cTnI for the clinical endpoints of mortality and major adverse cardiovascular event (MACE; a composite of acute myocardial infarction, unstable angina, revascularization, reinfarction, and congestive heart failure and stroke) at 180-day follow-up.RESULTS:Among 1982 patients analyzed, 14% had ACS, 21% had non-ACS cardiovascular disease, 31% had a noncardiac diagnosis and 34% had chest pain NOS. cTnI elevation above the 99th percentile was observed in 52, 18, 6 and 7% in these groups, respectively. cTnI elevation was associated with mortality and MACE, and their relationships were more prominent in noncardiac diagnosis and chest pain NOS than in ACS and non-ACS cardiovascular diagnoses for mortality, and in non-ACS patients than in ACS patients for MACE (hazard ratio for doubling of cTnI 1.85, 2.05, 8.26 and 4.14, respectively; P for interaction 0.011 for mortality; 1.04, 1.23, 1.54 and 1.42, respectively; P for interaction <0.001 for MACE).CONCLUSION:In patients presenting to the ED with chest pain, cTnI elevation was associated with a worse prognosis in non-ACS patients than in ACS patients.

BACKGROUND Sacubitril/valsartan (Sac/Val) improves outcomes in patients with heart failure with reduced ejection fraction (HFrEF).OBJECTIVES In this study, the authors sought to explore age differences in effects of Sac/Val on biomarkers, Kansas City Cardiomyopathy Questionnaire (KCCQ)-23 scores and cardiac remodeling. METHODS After initiation and titration of Sac/Val, concentrations of N-terminal pro-B-type natriuretic peptide (NT-proBNP), high-sensitivity troponin T (hs-cTnT), and soluble suppressor of tumorigenicity 2 (sST2) were measured and KCCQ-23 scores obtained from baseline to 12 months. Left ventricular ejection fraction (LVEF), and indexed left ventricular end-systolic (LVESVi) and indexed left ventricular end-diastolic (LVEDVi) and left atrial volume index (LAVi) volumes were measured with the use of echocardiography. Safety end points were assessed. Age-stratified analysis was performed for groups aged <65, 65-74, and >= 75 years.RESULTS Among 794 participants with HFrEF (mean age 65.1 years, 28.5% women), compared with patients aged <65 years (n = 369), 65-74 years (n = 237), and those aged >= 75 years (n =188), had similar reductions in hs-cTnT and sST2, but less NT-proBNP reduction (-45.6% vs -40.2% vs -30.5%, respectively; P = 0.02). Gains in KCCQ-23 were smaller (+11.8 vs +11.4 vs +6.0 points; P = 0.03) in patients aged >= 75 years, although similar proportions of each age group achieved >= 10-point and >= 20-point increases in KCCQ-23 by month 12. Improvements in LVEF, LVEDVi, LVESVi, and LAVi were similar among age groups. Incidence of safety end points was also similar.CONCLUSIONS Sac/Val resulted in significant improvements in prognostic biomarkers and measures of cardiac remodeling and health status from baseline to month 12 across age categories. Older study participants showed somewhat blunted reduction in NT-proBNP and less improvement in KCCQ-23 overall summary scores. (Effects of Sacubitril/Valsartan Therapy on Biomarkers, Myocardial Remodeling, and Outcomes [PROVE-HF]; NCT02887183) (J Am Coll Cardiol HF 2022;10:976-988)(c) 2022 The Authors. Published by Elsevier on behalf of the American College of Cardiology Foundation. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/ licenses/by-nc-nd/4.0/).

OBJECTIVE:To determine the utility of a highly sensitive troponin assay when utilized in the emergency department.METHODS:The FAST-TRAC study prospectively enrolled >1,500 emergency department patients with suspected acute coronary syndrome within 6 hours of symptom onset and 2 hours of emergency department presentation. It has several unique features that are not found in the majority of studies evaluating troponin. These include a very early presenting population in whom prospective data collection of risk score parameters and the physician's clinical impression of the probability of acute coronary syndrome before any troponin data were available. Furthermore, two gold standard diagnostic definitions were determined by a pair of cardiologists reviewing two separate data sets; one that included all local troponin testing results and a second that excluded troponin testing so that diagnosis was based solely on clinical grounds. By this method, a statistically valid head-to-head comparison of contemporary and high sensitivity troponin testing is obtainable. Finally, because of a significant delay in sample processing, a unique ability to define the molecular stability of various troponin assays is possible.TRIAL REGISTRATION:ClinicalTrials.gov Identifier NCT00880802.

AIM:To develop a machine learning model to predict the diagnosis of pulmonary embolism (PE).METHODS AND RESULTS:We undertook a derivation and internal validation study to develop a risk prediction model for use in patients being investigated for possible PE. The machine learning technique, generalized logistic regression using elastic net, was chosen following an assessment of seven machine learning techniques and on the basis that it optimized the area under the receiver operator characteristic curve (AUC) and Brier score. Models were developed both with and without the addition of D-dimer. A total of 3347 patients were included in the study of whom, 219 (6.5%) had PE. Four clinical variables (O2 saturation, previous deep venous thrombosis or PE, immobilization or surgery, and alternative diagnosis equal or more likely than PE) plus D-dimer contributed to the machine learning models. The addition of D-dimer improved the AUC by 0.16 (95% confidence interval 0.13-0.19), from 0.73 to 0.89 (0.87-0.91) and decreased the Brier score by 14% (10-18%). More could be ruled out with a higher positive likelihood ratio than by the Wells score combined with D-dimer, revised Geneva score combined with D-dimer, or the Pulmonary Embolism Rule-out Criteria score. Machine learning with D-dimer maintained a low-false-negative rate at a true-negative rate of nearly 53%, which was better performance than any of the other alternatives.CONCLUSION:A machine learning model outperformed traditional risk scores for the risk stratification of PE in the emergency department. However, external validation is needed.