Utilization of biomarkers for the prognostic prediction of cardiac arrest survivors using a multi-modal approach

International guidelines for post-cardiac arrest care recommend using multi-modal strategies to avoid the withdrawal of life-sustaining therapy (WLST) in patients with the potential for neurological recovery.[1] However, a clear methodology for multi-modal approaches has yet to be developed. Neuron-specific enolase (NSE) is currently the only recommended biomarker, and the European Resuscitation Council (ERC) and the European Society of Intensive Care Medicine (ESICM) have proposed a cutoff value of 60 μg/L at 48 and/or 72 h after the return of spontaneous circulation (ROSC) as a multi-modal prognostic tool for predicting poor neurological outcomes.[2] However, several validation studies have reported that the optimal cutoff value is higher than 60 μg/L when the false-positive rate (FPR) is <2%.[2-4] Therefore, we aimed to investigate the applicability of NSE values dichotomized at 60 μg/L as part of a multi-modal prognostic approach that affects the FPR. In this study, it is divided into four sections: “NSE level”, dichotomous values (“cutoff value of 60 μg/L” or “FPR 0% threshold”) and “dichotomous values including the 0% FPR interval” (Figure 1). We compared the predictive performance of NSE and the sensitivity to a 0% FPR by including the predictors identified in the multi-modal strategy. We hypothesized that combining dichotomized NSE values (with a threshold of 60 μg/L) with predictors of low sensitivity in a multi-modal strategy would significantly reduce sensitivity compared with that observed in other groups at 0% FPR.

Figure 1.

The four sections according to the NSE value. NSE: neuron-specific enolase; ROSC: return of spontaneous circulation; FPR: false-positive rate. IN: interval; PNL: peak NSE level.

METHODS

Study design and population

In this retrospective observational study, we obtained data from adult comatose out-of-hospital cardiac arrest (OHCA) survivors treated with target temperature management (TTM) at the Department of Emergency Medicine, Chungnam National University Hospital, between May 2018 and January 2023. We included patients with OHCA aged ≥18 years who received TTM and serum NSE levels measured at 48 and/or 72 h after ROSC. We excluded patients who experienced traumatic cardiac arrest or interrupted TTM due to hemodynamic instability and those who received extracorporeal membrane oxygenation (ECMO). We assessed neurological outcomes at 6 months post-ROSC using the Glasgow-Pittsburgh cerebral performance category (CPC) scale. We dichotomized the results into good (CPC 1–2) and poor (CPC 3–5). WLST did not occur during TTM; however, some patients died owing to circulatory or neurological complications, despite maximal support.

Data acquisition

We extracted the following data from the registry database: age, sex, Charlson comorbidity index score, witnessed arrest, bystander cardiopulmonary resuscitation (CPR), first monitored rhythm, etiology of cardiac arrest, time from cardiac arrest to the first basic life support (no-flow time), time from CPR to ROSC (low-flow time), serum NSE levels, pupillary light reflex (PLR) (absence vs. presence), electroencephalography (EEG) (highly malignant pattern vs. other patterns), and grey-to-white matter ratio (GWR) values on brain computed tomography (CT) images. Peak serum NSE levels measured at 48 or 72 h post-ROSC were used for analyses (48/72PNL). Additionally, we assessed NSE levels at two and three intervals. The first two intervals were divided based on the recommended cutoff values from the ERC/ESICM international guidelines (≤60 μg/L or >60 μg/L) (2IN). The second two intervals were divided based on the cutoff values derived from the 0% FPR threshold (2INFPR0). Additionally, three intervals (3IN) were determined by extending the 0% FPR threshold from the 2IN. The measurement timing and methods for determining the serum NSE levels, PLR, EEG, and GWR on CT images are presented in supplementary Table 1.

Statistical analysis

Receiver operating characteristic (ROC) curves were constructed to evaluate the prognostic performance of single predictors and their combination models for poor neurological outcomes at 6 months post-ROSC. We combined three predictors (PLR, EEG, and GWR) with 48/72PNL, 2IN, 2INFPR0, and 3IN. The optimal cutoff values were determined using the cutoff value for 100% specificity (i.e., an FPR equal to 0) and the Youden index. Subsequently, the DeLong test was used to determine the differences in the prognostic performance of each test and their combinations. Statistical analyses were performed using SPSS (version 26.0; IBM Corp., USA) and the MedCalc program version 15.2.2 (MedCalc Software, Belgium). The significance level was set at a P-value <0.05.

RESULTS

Patient characteristics

In total, 166 comatose OHCA survivors underwent TTM after ROSC. Of the 166 patients, 14 patients underwent ECMO, 5 had cardiac arrest due to trauma, and 14 had neither 48 h nor 72 h NSE values. Ultimately, 133 patients were included in the analysis. Supplementary Table 2 shows the baseline and OHCA characteristics stratified by neurological outcomes at 6 months. The poor neurological outcome group exhibited consistently higher NSE values throughout the duration than that exhibited by the good outcome group. These patients also had a high incidence of highly malignant patterns and an absence of PLR but a lower GWR.

Prognostic performance of NSE

The 2INFPR0 was divided into ≤144 μg/L or >144 μg/L, representing a 0% FPR of 48/72PNL. The 3IN were 0−60 μg/L, 60−144 μg/L, and >144 μg/L. Table 1 shows the prognostic performance of NSE for poor outcomes at 6 months post-ROSC. At 48 h post-ROSC, NSE levels had the best prognostic performance and sensitivity. According to the ROC analysis, 3IN, with NSE cutoffs of 60 μg/L (Youden index) and 144 μg/L, had sensitivities of 82.8% and 53.1% at 5% and 0% FPR, respectively.

Table 1.

Prognostic performances of NSE value for the prediction of 6-month poor neurological outcome

Prognostic performance of multi-modal approach by applying NSE levels across various intervals

Table 2 shows the prognostic performance of the multi-modal approach using 48/72PNL, 2IN, 2INFPR0, and 3IN for poor neurological outcomes at 6 months post-ROSC. PLR, EEG, and GWR had fair to good (AUC 0.77), fair to excellent (0.84), and failed to reach fair prognostic performance (0.65), respectively.

Table 2.

Comparison of prognostic performance and sensitivity to poor neurological outcome with a 0% false positive rate after 6 months of return of spontaneous circulation using multi-modal combination predictor

Among the two combinations performed respectively with PLR, EEG, and GWR, and four sections of NSE, 48/72PNL, 2IN, 2INFPR0, and 3IN had improved prognostic performance compared with that shown by the single predictors (all P<0.05), except for 2INFPR0 and EEG (P=0.08). Except for the combination of PLR and GWR, the three combinations and all four combinations did not show better prognostic performance. Furthermore, among the four combinations, at 0% FPR, the sensitivity was highest in the combination with 3IN at 83.3% (95% CI: 69.8% –92.5%), and lowest in the combination with 2IN at 79.2% (95% CI: 65.0% –89.5%).

DISCUSSION

Several studies have reported on FPR-based NSE thresholds. A systematic review reported an FPR <5% for predicting poor neurological outcomes with NSE levels ≥58 μg/L at 24 h, ≥50 μg/L at 48 h, and ≥47 μg/L at 72 h.[5] A study validating NSE values for OHCA suggested FPRs <5% and 0% for predicting poor neurological outcomes with NSE levels ≥59.2 μg/L (sensitivity 60%) and ≥85.5 μg/L (sensitivity 48.7%), respectively, at 72 h.[6,7] In this study, the NSE values with an FPR <5% and 0% at 48 and 72 h were 39.7 μg/L, 64.2 μg/L, 57.0 μg/L, and 144.0 μg/L, respectively. These values are similar to those of OHCA validation studies.[7] Differences in the NSE cutoff values among studies may be due to the timing of the variable outcome assessment, specimen handling methods, variability in WLST, and confounders of NSE elevation (NSE-producing tumors, acute brain disease, and hemolysis). In particular, the higher 72 h NSE values in this study compared with those in other studies are presumed to be due to the absence of WLST during TTM.

We observed a shortcoming in 2IN when it was combined with a few rather than many predictors, particularly with low sensitivity at 0% FPR. Compared with 2INFPR0 and 3IN, 2IN demonstrated comparable predictive performance. However, its sensitivity was notably low at an FPR of 0%. Specifically, for the combinations with PLR, the sensitivity was 0% at an FPR of 0%, which was significantly lower than the 50.8% for 2INFPR0 and 53.1% for 3IN. Moreover, the predictive performance did not improve when 48/72PNL, 2IN, 2INFPR0, or 3IN were combined with the PLR, EEG, or GWR. Interestingly, among all the NSE predictors, 3IN exhibited the highest sensitivity at an FPR of 0%, while 2IN showed the lowest sensitivity.

Our findings suggested that applying NSE levels to a multi-modal strategy algorithm will increase sensitivity if divided into three intervals, including the 0% FPR range, rather than employing a binary division for the NSE threshold based solely on the Youden index.

This study had several limitations and strengths. First, this was a single-center retrospective study with a small number of patients; therefore, the generalizability of the findings might be limited. Second, the attending physicians were not blinded to the PLR, EEG, or GWR results, and subjective bias might have occurred. Third, the PLR, EEG, and GWR used in this multi-modal strategy used data from only a single center, and N20 somatosensory evoked potential (SSEP) wave, status myoclonus, and magnetic resonance imaging data were not included in the analysis. Nevertheless, to our knowledge, no prior study has explored the predictive performance by applying NSE levels across various intervals using a multi-modal strategy. This may account for the limitations encountered in explaining our findings.

CONCLUSIONS

When applying NSE levels to a multi-modal strategy across various intervals, dividing the levels into a minimum of three intervals, including the 0% FPR range, instead of dichotomizing them may enhance sensitivity. A multi-center prospective study is needed to confirm our findings.