PL01: Creation of a risk scoring system for emergency department patients with acute heart failure. (May 2020)

• Record Type: Journal Article
• Authors: Stiell, I.; Perry, J.; Clement, C.; Sibley, S.; McRae, A.; Rowe, B.; Borgundvaag, B.; McLeod, S.; Mielniczuk, L.; Dreyer, J.; Yan, J.; Brown, E.; Brinkhurst, J.; Nemnom, M.; Taljaard, M.
• Is Part Of: CJEM. Volume 22(2020)Supplement 1
• Page Start: S5
• Page End: S5
• Publication Date: 2020-05
• Subjects: heart failure, -- risk scale, -- safety; Emergency Treatment -- Periodicals; Emergency Medicine -- Periodicals; Emergency medical services -- Canada -- Periodicals; Medical emergencies -- Canada -- Periodicals; Emergency medical services; Medical emergencies; Canada; Periodicals; 616.02505
• DOI: 10.1017/cem.2020.46
• Languages: English
• ISSNs: 1481-8035

Abstract

Abstract : Introduction: Acute heart failure (AHF) is a common emergency department (ED) presentation and may be associated with poor outcomes. Conversely, many patients rapidly improve with ED treatment and may not need hospital admission. Because there is little evidence to guide disposition decisions by ED and admitting physicians, we sought to create a risk score for predicting short-term serious outcomes (SSO) in patients with AHF. Methods: We conducted prospective cohort studies at 9 tertiary care hospital EDs from 2007 to 2019, and enrolled adult patients who required treatment for AHF. Each patient was assessed for standardized real-time clinical and laboratory variables, as well as for SSO (defined as death within 30 days or intubation, non-invasive ventilation (NIV), myocardial infarction, coronary bypass surgery, or new hemodialysis after admission). The fully pre-specified, logistic regression model with 13 predictors (age, pCO2, and SaO2 were modeled using spline functions with 3 knots and heart rate and creatinine with 5 knots) was fitted to the 10 multiple imputation datasets. Harrell's fast stepdown procedure reduced the number of variables. We calculated the potential impact on sensitivity (95% CI) for SSO and hospital admissions and estimated a sample size of 170 SSOs. Results: The 2, 246 patients had mean age 77.4 years, male sex 54.5%, EMS arrival 41.1%, IV NTG 3.1%, ED NIV 5.2%, admission on initial visit 48.6%. Overall there were 174 (7.8%) SSOs including 70 deaths (3.1%). The final risk scale is comprised of five variables (points) and had c-statistic of 0.76 (95% CI: 0.73-0.80): 1.Valvular heart disease (1) 2.ED non-invasive ventilation (2) 3.Creatinine 150-300 (1) ≥300 (2) 4.Troponin 2x-4x URL (1) ≥5x URL (2) 5.Walk test failed (2) The probability of SSO ranged from 2.0% for a total score of 0 to 90.2% for a score of 10, showing good calibration. The model was stable over 1, 000 bootstrap samples. Choosing a risk model total point admission threshold of >2 would yield a sensitivity of 80.5% (95% CI 73.9-86.1) for SSO with no change in admissions from current practice (48.6% vs 48.7%). Conclusion: Using a large prospectively collected dataset, we created a concise and sensitive risk scale to assist with admission decisions for patients with AHF in the ED. Implementation of this risk scoring scale should lead to safer and more efficient disposition decisions, with more high-risk patients being admitted and more low-risk patients being discharged.