The methodology of this study is reported in accordance with the Standards for Reporting of Diagnostic Accuracy Studies (STARD) 2015 guidelines [14]. This study was approved by the ethics committee of Kyoto University (R-1045). The need for informed consent was waived in view of the study design.
Study design and settings
We undertook this study to identify the diagnostic accuracy of specific clinical parameters through a retrospective analysis of the JAAM-OHCA Registry [15], a multicenter, prospective nationwide database that includes pre-hospital information, in-hospital information, and outcome among OHCA patients transported to emergency departments in Japan. Details about this registry have been previously reported [15, 16]. The JAAM-OHCA Registry was established in 2014 by the organizing committee of the registry to improve therapeutic strategy, emergency medical systems, and patient outcome. Presently, the registry includes 87 institutions, and 66 of the included hospitals are university hospitals and/or critical care centers. These critical care centers were certified by the Ministry of Health, Labor, and Welfare in Japan, and they are equipped to provide highly specialized treatment, such as ECLS, percutaneous coronary intervention, or targeted 24-hour temperature management. The other 21 hospitals were not certified as critical care centers, but provided emergency medical service to the community. A total of 34,754 OHCA patients were registered in the JAAM-OHCA Registry from June 2014 to December 2017.
Prehospital information was collected by paramedics based on the standardized Utstein format [17], and verified by the Fire and Disaster Management Agency in Japan. In-hospital information was registered by clinicians or clinical data administrators at each institution, using a standardized online form. The in-hospital information has a fundamental and supplemental variables section. Fundamental variables (e.g., basic characteristics, blood gas assessment data, and outcome) were mandatorily registered in all cases if available, while supplemental variables (e.g., blood chemistry data) were recorded if the institutions applied additional protocols and recorded them. The JAAM-OHCA registry committee combined the in-hospital and prehospital information and logically evaluated the data quality. Finally, de-identified data were provided to the researchers by the registry’s committee.
Participants
We included all adult (≥18 years old) OHCA patients transported to emergency departments with moderate-to-severe hypothermia and registered in the database from June 2014 to December 2017. Moderate-to-severe hypothermia was defined as BT 32 °C or lower on hospital arrival, based on the Swiss grading system[18] and the related guidelines [1-3]. In general, it is challenging to differentiate whether the cardiac arrest was primarily caused by hypothermia precisely, or the cardiac arrest patient became hypothermic on hospital arrival; thus, we did not attempt to distinguish these two states. We excluded patients who received no resuscitation attempts in the hospital. This is because those patients were obviously dead, as evidenced by the existence of postpartum changes such as rigor mortis, or the patients had already documented a “do not resuscitate” order. Moreover, we excluded patients who opted out from participation in the study, and patients who were cases of obvious traumatic cardiac arrest or hanging, who had no pre-hospital data, no BT, and had no blood tests conducted. Furthermore, as explained earlier, age, BT, lactate, and pH values were fundamental variables; however, since potassium was a supplemental variable, it was only available in the data from institutions that applied additional protocols. Thus, to analyze the predictive value of potassium, we undertook additional analysis to exclude the patients who were transferred to institutions that did not apply the additional research protocol to record the serum potassium values.
Index test
Based on reports in the previous literature [1, 2, 7-9, 13, 19-21], we selected three potential predictors: serum pH, lactate, and potassium values. These values were defined as the measurements from the initial blood test or blood biochemistry tests conducted on hospital arrival in the emergency department. Moreover, we selected age and BT as a reference, and BT was defined as the body temperature measured initially on hospital arrival.
Target conditionThe primary target condition to be predicted in this study was the 1-month survival.
Statistical analysis
Patient and hospital characteristics
We described the patients and hospital characteristics as follows: sex age, season (Spring: March–June; Summer: July–August; Autumn: September–November; and Winter: December–February), and regions of Japan (Northern, Eastern, Western, and Southern). The season and area were defined by the definition of the Japan Meteorological Agency (details of the area are described in the supplementary materials) [22]. In addition, we described pre-hospital and in-hospital patient data as follows: bystander witness, bystander CPR, shockable on initial rhythm, advanced airway inserted by paramedics, cardiac rhythm on hospital arrival [return of spontaneous circulation (ROSC), shockable, pulseless electrical activity (PEA), asystole], BT on arrival, and ECMO implementation. We also included blood test results on arrival, time course (from emergency call to hospital arrival to blood test to ROSC and/or to ECMO), and the disposition (admit to intensive care unit (ICU)/ward, or death in the emergency department]. A shockable rhythm was defined as ventricular fibrillation (VF) or pulseless ventricular tachycardia (VT). ROSC was defined as the presence of a palpable pulse for more than 30 seconds despite circulatory support by ECMO [23]. Furthermore, we indicated the hospital's basic information as follows: type of hospitals (whether a tertiary-care center) and the number of beds. A tertiary center was defined as university hospitals and/or critical care centers certified by the government, as explained earlier. Data were presented as median and interquartile range (IQR) for continuous variables, and as number and percentages for categorical variables; missing values are shown as “Missing" or "Unknown."
Predictive accuracy
We calculated the predictive accuracy of pH, and lactate on target conditions in the study population. We also calculated the accuracy of age and BT as a reference. We showed the discriminatory ability of each predictor for 1-month survival by using the receiver-operating characteristic curve (ROC) and area under the curve (AUC) with 95% confidence interval (CI). Moreover, we set the cutoff value and created 2×2 tables to calculate sensitivity (Se), specificity (Sp), positive and negative likelihood ratio (LR+ and LR−, respectively), and positive and negative predictive value (PPV and NPV, respectively). In general, a pre-specified cutoff value was recommended to estimate the diagnostic accuracy [14]. The cutoff values were suggested as a serum potassium level of 8 or 10 mmol/L and pH 6.5 in the published literature [2, 3, 8, 9, 13], although these have not been specifically established. Therefore, we specified several rounded-off values in the range of interest as the cutoff points to avoid optimistic interpretation and facilitate ease of clinical use [14]. Furthermore, we suggested cutoff values in each predictor on the basis of the LR+ and LR− (LR+, >5; LR−, <0.2), which are commonly thought to be useful for rule-in or rule-out decision making [24]. We did not undertake a sample-size estimation, because it was the secondary usage of an already available database and a retrospective analysis. Missing data were handled by the exclusion of that specific patient to calculate the predictive accuracy. All statistical results were considered significant at a two-sided P-value of <0.05. All statistical analyses were undertaken in JMP Pro® 14 software (SAS Institute Inc., Cary, NC, USA).
Additional analysis
As mentioned earlier, a substantial number of institutions did not apply the additional protocol to record potassium values. Thus, we conducted additional analyses to calculate the predictive accuracy of potassium and other predictors after excluding the patients who were transferred to the institutions without an additional protocol to record the potassium data.