Should We Focus on the “Who” When Identifying Candidates for Extracorporeal Cardiopulmonary Resuscitation?
論文ID: CJ-21-0910
この記事には本公開記事があります。
詳細
論文ID: CJ-21-0910
Evidence to support the use of extracorporeal cardiopulmonary resuscitation (ECPR) in patients with out-of-hospital cardiac arrest (OHCA) is accumulating.1 Several studies have reported factors that predict a favorable outcome such as shorter time to ECPR initiation, initial documented cardiac rhythm, and cardiac rhythm conversion.2–4 However, it can be difficult to determine whether ECPR should be implemented based on single factors because patients with OHCA are heterogeneous. Identifying “subphenotypes” that suggest ECPR would have a beneficial effect on survival with favorable neurological outcomes might be required (Figure). In medicine, the recent incorporation of machine learning (ML), which integrates multiple quantitative variables, is expected to improve predictive accuracy.5–7
Identification of candidates for extracorporeal cardiopulmonary resuscitation (ECPR) based on single factors (A) vs. multiple factors (B). Asys, asystole; OHCA, out-of-hospital cardiac arrest; PEA, pulseless electrical activity; VF, ventricular fibrillation; VT, ventricular tachycardia.
Article p ????
Based on such a concept of “subphenotypes”, in this issue of the Journal Okada et al8 describe an interesting approach to identifying candidates for ECPR among patients with OHCA. Using an ML-based clustering approach, they retrospectively analyzed data from the Comprehensive Registry of Intensive Care for OHCA Survival (CRITICAL) study, which was a multicenter prospective observational study of the prehospital and in-hospital experiences of patients with OHCA in Osaka Prefecture, Japan. Between 2012 and 2017, a total of 920 patients with OHCA and a shockable rhythm were included in the latent class analysis. An ML-based unsupervised clustering approach identified 3 groups based on 16 clinically meaningful prehospital and in-hospital variables. Group 1 had the lowest pO2 and highest pCO2 values. Group 2 had higher pO2 and lower pCO2 values than Group 1. Group 3 had pO2 values similar to those of Group 2, as well as the lowest pCO2 values, highest pH values, highest proportion of patients with return of spontaneous circulation (ROSC) at hospital arrival, and highest estimated glomerular filtration rate (eGFR). In Group 1, 30-day survival with favorable neurological outcomes was 7.3%, whereas it was 18.9% in Group 2 and 75.3% in Group 3. In the development dataset, the odds ratio (95% confidence interval) for ECPR with 30-day survival as the outcome was 1.87 (1.08–3.2) in Group 1, 1.01 (0.56–1.84) in Group 2, and 0.1 (0.04–0.24) in Group 3. Analysis of the validation dataset yielded similar results. Thus, patients who have higher pO2 values, lower pCO2 values, higher pH values, higher eGFR values, and ROSC might not be suitable for ECPR.
These results might not come as a surprise, because many previous studies have excluded patients achieving ROSC.1,9,10 In addition, higher pO2, lower pCO2, and higher pH values might reflect the result of high-quality CPR. However, it is worthwhile to apply the concept of “subphenotypes” to ECPR candidates because, as the authors describe, the effectiveness of ECPR in a systematic review of observational studies might be controversial due to the heterogeneity of patients with OHCA who were treated with ECPR.11 In resuscitation science, concepts and approaches based on ML might become increasingly common as hypothesis-generating studies.
T.N. declares no conflicts of interest with regards to the submitted work. T.N. received a Grant-in-Aid for Young Scientists (A) (20K17914) from the Japan Society for the Promotion of Science and the Uehara Memorial Foundation Overseas Research Fellowship unrelated to the submitted work.
