BACKGROUND: Improved risk stratification of acute heart failure in the emergency department may inform physicians' decisions regarding patient admission or early discharge disposition. We aimed to validate the previously-derived Emergency Heart failure Mortality Risk Grade for 7-day (EHMRG7) and 30-day (EHMRG30-ST) mortality.
METHODS: We conducted a multicenter, prospective validation study of patients with acute heart failure at 9 hospitals. We surveyed physicians for their estimates of 7-day mortality risk, obtained for each patient before knowledge of the model predictions, and compared these with EHMRG7 for discrimination and net reclassification improvement. We also prospectively examined discrimination of the EHMRG30-ST model, which incorporates all components of EHMRG7 as well as the presence of ST-depression on the 12-lead ECG.
RESULTS: We recruited 1983 patients seeking emergency department care for acute heart failure. Mortality rates at 7 days in the 5 risk groups (very low, low, intermediate, high, and very high risk) were 0%, 0%, 0.6%, 1.9%, and 3.9%, respectively. At 30 days, the corresponding mortality rates were 0%, 1.9%, 3.9%, 5.9%, and 14.3%. Compared with physician-estimated risk of 7-day mortality (PER7; c-statistic, 0.71; 95% CI, 0.64-0.78) there was improved discrimination with EHMRG7 (c-statistic, 0.81; 95% CI, 0.75-0.87; P=0.022 versus PER7) and with EHMRG7 combined with physicians' estimates (c-statistic, 0.82; 95% CI, 0.76-0.88; P=0.003 versus PER7). Model discrimination increased nonsignificantly by 0.014 (95% CI, -0.009-0.037) when physicians' estimates combined with EHMRG7 were compared with EHMRG7 alone ( P=0.242). The c-statistic for EHMRG30-ST alone was 0.77 (95% CI, 0.73-0.81) and 30-day model discrimination increased nonsignificantly by addition of physician-estimated risk to 0.78 (95% CI, 0.73-0.82; P=0.187). Net reclassification improvement with EHMRG7 was 0.763 (95% CI, 0.465-1.062) when assessed continuously and 0.820 (0.560-1.080) using risk categories compared with PER7.
CONCLUSIONS: A clinical model allowing simultaneous prediction of mortality at both 7 and 30 days identified acute heart failure patients with a low risk of events. Compared with physicians' estimates, our multivariable model was better able to predict 7-day mortality and may guide clinical decisions.
CLINICAL TRIAL REGISTRATION: URL: https://www.clinicaltrials.gov . Unique identifier: NCT02634762.
The article is interesting in providing a tool that is better than physician gestalt in determining risk for CHF patients presenting to the ED. However, the score doesn't give a final disposition, which in CHF is also governed by other factors such as improvement in symptoms and follow up (ie it's only mortality). Also, the scoring is complex and not easy to use, probably requiring a separate calculator to score, thus decreasing its usefulness.
I still will rely on my clinical gestalt. Practicing in the US is different than Canada where malpractice is less of an issue. We admit most of our cases of CHF; whereas, in Canada they admit a lot less.
This is a very interesting study on comparison of algorithm vs human judgement. IT's disappoint that the addition of physician judgment did not increase accuracy of risk stratification.