Neutrophil to lymphocyte ratio for predicting delayed neurologic sequelae caused by carbon monoxide poisoning

Abstract

Background:

This study summarized and analyzed the prognostic value of the neutrophil to lymphocyte ratio (NLR) for delayed neurologic sequelae (DNS) caused by carbon monoxide (CO) poisoning.

Methods:

PubMed, Embase, Cochrane Library, Web of Science, ScienceDirect, and China National Knowledge Infrastructure Database were searched to identify relevant articles from their inception to February 2026. The odds ratio and its 95% confidence interval (CI), area under the curve, meta-regression analysis, sensitivity analysis, and publication bias for NLR were estimated to predict DNS.

Results:

Eight studies were included, involving 1757 patients with acute CO poisoning, among whom 382 cases (21.7%) developed DNS. The pooled results show that NLR was related to DNS (odds ratio = 14.67, 95% CI: 8.26–26.04, P < .001) and had high heterogeneity (I² = 65.1%, P = .005). The area under the curve of NLR for predicting DNS was 0.88 (95% CI: 0.85–0.91), with a sensitivity of 86% and a specificity of 72%. Subgroup analysis indicated that study design may be the source of heterogeneity. Sensitivity analysis suggested the robustness of the results. Deek’s test did not detect significant publication bias (P = .24).

Conclusion:

Current evidence indicates that an elevated NLR is associated with an increased risk of DNS after acute CO poisoning, supporting its potential utility as a prognostic marker. However, because the included studies showed substantial heterogeneity and were predominantly retrospective, these findings should be interpreted with caution and require validation in high-quality prospective studies. NLR should therefore be considered an auxiliary tool for risk stratification rather than an independent diagnostic method. Its principal clinical value may lie in resource-limited settings where advanced monitoring is not feasible.

1. Introduction

Acute carbon monoxide (CO) poisoning accounts for one of the highest poisoning events worldwide.^[1] By binding to hemoglobin and other heme-containing proteins, CO reduces oxygen delivery and induces tissue damage and complications.^[2] Delayed neurologic sequelae (DNS) is the most common and serious complication of acute CO poisoning, and it results in a high disability rate and fatality rate, thus seriously affecting the quality of life of patients.^[3] Therefore, early and precise identification of DNS is crucial to delay disease progression and decrease mortality.

The neutrophil to lymphocyte ratio (NLR), a biomarker derived from complete blood cell counts, reflects the balance between immune activation and systemic inflammation.^[4,5] Although several studies have reported that NLR may help predict delayed neurologic sequelae (DNS) after acute CO poisoning, important uncertainties remain. First, the predictive performance of NLR for DNS varies widely across studies, and no standardized optimal cutoff value has been established. Second, most previous studies used single-center retrospective designs with limited sample sizes, making it difficult to distinguish the prognostic from the diagnostic value of NLR. Furthermore, no meta-analysis has systematically synthesized the overall evidence on the ability of NLR to predict DNS. Therefore, this meta-analysis aimed to systematically evaluate the prognostic value of NLR for DNS after acute CO poisoning, explore sources of heterogeneity, supplement the existing evidence, and clarify its practical utility for early detection and risk stratification.

2. Materials and methods

2.1. Study design

This meta-analysis was performed in accordance with the Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA).^[6] Ethical approval was not required as this meta-analysis involved secondary data analysis from published studies without direct patient involvement. This review was registered in PROSPERO (CRD420261292948).

2.2. Literature search

PubMed, Embase, the Cochrane Library, Web of Science, ScienceDirect, and the China National Knowledge Infrastructure database were searched from inception to February 2026 to identify studies evaluating NLR for the prediction of DNS in patients with acute CO poisoning by 2 independent reviewers (XTL and YYH). The search strategy included the terms “carbon monoxide” and “neutrophil to lymphocyte ratio.” The full search strategy is provided in Table S1, Supplemental Digital Content 1. No language restrictions were applied. To minimize publication bias, gray literature, including dissertations and conference proceedings, was also searched. In addition, the reference lists of relevant reviews and included articles were searched manually.

2.3. Inclusion and exclusion criteria

Studies were included if they met the following criteria: involved patients with acute CO poisoning, evaluated NLR ratio and DNS, and had sufficient data available for calculating pooled sensitivity and specificity.

The exclusion criteria comprised the following: duplicates, meeting abstracts, letters, comments, corrections, and responses; animal experiments or cellular experiments.

2.4. Definition of DNS

DNS refers to neuropsychiatric symptoms that emerge or recur after a period of apparent recovery from acute CO poisoning. Delayed neurological sequelae were diagnosed according to the following criteria: definite acute CO exposure; a clear pseudo-recovery phase during the disease course; and the development of new neurological signs, including cognitive decline, extrapyramidal symptoms, and psychomotor retardation.^[3]

2.5. Data extraction and quality evaluation

Data were extracted from the included studies using a predefined data collection form. Two reviewers (XTL and XLM) independently performed the extraction, and any discrepancies were resolved through discussion with a third reviewer (WZW). The extracted contents included the following: the first author, publication year, age, gender, study location, study design, sample size, NLR level, and DNS rate.

The quality of the included studies was evaluated using the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) tool by 2 reviewers (QMX and WPX) independently.^[7] QUADAS-2 comprises 4 domains: patient selection, index test, reference standard, and flow and timing. Each domain was assigned an overall judgment of “low,” “high,” or “unclear” risk by both authors, and all discrepancies were resolved by a third reviewer (WZW).

2.6. Statistical analysis

Meta-analysis was conducted using Stata 14.0 software. The odds ratio (OR) and its 95% confidence interval (CI) were used to represent the relationship between NLR and DNS. The I² statistic and Cochran Q test were used to assess heterogeneity. I² < 25% indicated low heterogeneity; 25% to 50%, moderate heterogeneity; >50%, high heterogeneity. If high heterogeneity was observed, a random effects model was used to reduce the effect of heterogeneity on results; otherwise, a fixed-effect model was considered. Sensitivity analysis was performed to assess the effect of individual studies on the combined effect size; that is, meta-analysis was conducted again after eliminating each study 1 by 1. Meta-regression analysis based on sample size, study location, study design, follow-up duration, NLR cutoff value, and DNS rate was used to explore the sources of heterogeneity. Deek’s test was conducted to evaluate publication bias. P < .05 indicated statistical significance.

3. Results

3.1. Literature screening process

The literature screening process is shown in Figure 1 and Table S1, Supplemental Digital Content 1. The initial search yielded 165 relevant articles. After eliminating 23 duplicate articles, 142 articles remained. After reading the titles and abstracts, 117 articles were excluded, which left 25 articles for full-text reading. Eventually, 8 studies involving 1757 patients with acute CO poisoning were included.^[8-15]

Figure 1.

Flowchart of the study selection process.

3.2. Basic characteristics of the included studies

Table 1 shows the basic characteristics of studies included in this work. The publication dates were from 2017 to 2024, and the sample size ranged from 60 to 607 cases. Among the 8 included studies, 5 were retrospective research,^{[8,10,11,13,14]} 1 was prospective work,^[9] and 2 were retrospective combined with prospective studies.^[12,15] Six studies were from China,^{[8,10-12,14,15]} 1 was from Egypt,^[9] and 1 was from Turkey.^[13] The incidence of DNS ranged from 5.4% to 38.3%.

Table 1.

Characteristics of the included studies.

First author	Mean age (yr)	Male (%)	Study location	Study design	Sample size	NLR level		DNS rate (%)
						DNS	Non-DNS
Gao 2021	53.4	52.6	China	Retro	171	10.60 (9.69–15.34)	7.53 (5.86–8.56	28.7
Ghanem 2022	32.0	48.3	Egypt	Pro	60	11.71 ± 6.23	4.05 ± 2.28	26.7
Han 2017	61.6	42.5	China	Retro	181	10.08 (6.23–14.17)	5.73 (3.04–9.21)	36.5
Wang 2023	57.2	48.7	China	Retro	607	11.3 (9.2–16.4)	7.9 (4.4–12.7)	5.4
Xu 2023	53.5	40.3	China	Retro and pro	288	6.84 (4.22–12.43)	3.23 (1.91–5.60)	29.2
Yalçin 2023	11.0	48.2	Turkey	Retro	137	6.5 (4.8–9.3)	2.5 (1.3–4.3)	33.6
Zhao 2024	60.9	55.6	China	Retro	180	14.22 ± 3.56	6.92 ± 2.33	20.5
Zhou 2022	73.1	54.1	China	Retro and pro	133	11.67 ± 2.52	6.58 ± 1.92	38.3

DNS = delayed neurologic sequelae, NLR = neutrophil to lymphocyte ratio, Pro = prospective, Retro = retrospective.

3.3. Quality evaluation

Figure 2 presents the quality assessment outcomes. For patient selection, 87.5% of the studies had an unclear risk of bias because they did not mention consecutive patients. For the index test, 100% had a high or unclear risk of bias because they did not mention blinding or did not use a prespecified cutoff. For the reference standard, 100% were classified as unclear risk status because they did not mention the implementation of blinding. For flow and timing, 87.5% of the studies had an unclear risk or high risk of bias because they did not describe whether all patients were included in the analysis. In addition, applicability concerns were low in all studies across the domains of patient selection, index test, and reference standard.

Figure 2.

Quality of the included studies scored against the QUADAS-2 criteria. QUADAS-2 = Quality Assessment of Diagnostic Accuracy Studies.

3.4. Meta-analysis of NLR predicting DNS

A high heterogeneity was observed among the included studies (I² = 65.1%, P = .005), and a random effects model was adopted (Fig. 3). The pooled results indicate that NLR was related to DNS (OR = 14.67, 95% CI: 8.26–26.04, P < .001), with high heterogeneity. Given the limited number of studies and the majority being retrospective, the OR values should be interpreted with caution, and further verification through high-quality prospective studies is required. The area under the curve (AUC) of NLR for predicting DNS was 0.88 (95% CI: 0.85–0.91), with a sensitivity of 86% and specificity of 72% (Fig. 4). Since the NLR thresholds used in each study are different, it is recommended that a unified optimal cutoff value be adopted in future studies.

Figure 3.

Association between NLR and DNS in patients of CO poisoning. CI = confidence interval, CO = carbon monoxide, DNS = delayed neurologic sequelae, NLR = neutrophil to lymphocyte ratio, OR = odds ratio.

Figure 4.

Summary of receiver operating characteristic curve of NLR for predicting DNS in patients CO poisoning. AUC = area under the curve, CO = carbon monoxide, DNS = delayed neurologic sequelae, NLR = neutrophil to lymphocyte ratio.

3.5. Meta-regression analysis and subgroup analysis of NLR predicting DNS

Meta-regression analysis and subgroup analysis based on the mean sample size (>220 vs ≤220), study location (China vs other country), study design (prospective, retrospective, vs both), follow-up duration (≥12 months vs <12 months/unclear), NLR cutoff value (≥4 vs <4), and mean DNS rate (>27.4% vs ≤27.4%) were performed to explore the sources of heterogeneity. The meta-regression analysis indicated that sample size, study location, study design, and DNS rate were not sources of heterogeneity (Table S2, Supplemental Digital Content 2). The subgroup analysis showed that study design might be a source of heterogeneity (Table S3, Supplemental Digital Content 3).

3.6. Sensitivity analysis and publication bias

Sensitivity analysis suggested the robustness of the results (Fig. 5). Deek’s test did not detect significant publication bias (P = .24, Fig. 6).

Figure 5.

Sensitivity analysis of the association between NLR and DNS in patients CO poisoning. CI = confidence interval, CO = carbon monoxide, DNS = delayed neurologic sequelae, NLR = neutrophil to lymphocyte ratio.

Figure 6.

Deek’s asymmetry plot showed no publication bias among the included studies.

4. Discussion

CO poisoning is among the most common forms of accidental poisoning. In 2021, the global incidence of unintentional CO poisoning was 7.87 per 1,00,000, corresponding to 2,94,173 incident cases.^[16] Long-term sequelae among survivors include persistent neurological deficits and cognitive impairment. DNS has been reported in 6.8% to 52% of patients with acute CO poisoning,^[17-19] which is consistent with the 21.7% observed in our analysis. In current clinical practice, conventional assessment tools for acute CO poisoning have important limitations. Although the Glasgow Coma Scale is widely used, its prognostic value for DNS is not well established, and the quality of evidence has been rated as low to moderate.^[20] Carboxyhemoglobin, the standard biomarker for confirming CO exposure, does not correlate closely with the severity of the clinical presentation; therefore, its use for guiding clinical management is currently discouraged.^[21] CT is not suitable for detecting brain injury in the acute phase of CO poisoning, and the prognostic value of acute-phase CT findings remains unclear. Magnetic resonance imaging is more sensitive but is time-consuming and costly, limiting its accessibility in emergency and resource-limited settings.^[22] Furthermore, diffusion-weighted imaging may yield false-negative results when performed in the very early phase.

NLR reflects the dynamic balance between inflammation and immunity, but it is only used as a rapid and low-cost supplementary indicator, facilitating its use in emergency environments. This study is the first to systematically evaluate the predictive value of the NLR for delayed encephalopathy after acute CO poisoning. The combined AUC of NLR for predicting delayed encephalopathy was 0.88, which indicates a high diagnostic performance. However, this estimate warrants cautious interpretation. The QUADAS‑2 assessment revealed that most included studies were at high or unclear risk of bias across 3 domains: the index test (owing to lack of blinding and absence of pre-specified cut-offs), the reference standard (blinding not reported), and flow and timing (incomplete inclusion of patients in the analysis). These methodological shortcomings could have inflated both the pooled OR and AUC. Consequently, the seemingly high discriminatory power of NLR should not be interpreted as definitive evidence of clinical utility until it has been validated in well‑designed prospective studies employing standardized thresholds.

The mechanism by which elevated NLR predicts the occurrence of DNS can be explained from the following aspects: First, after CO poisoning, neutrophils adhere and aggregate on the surface of damaged endothelial cells, which results in the release of myeloperoxidase, reactive oxygen species, and proteases and subsequent lipid peroxidation, blood–brain barrier disruption, and demyelination.^[21,23] The increase in neutrophils directly reflects the intensity of this acute inflammatory response.^[24,25] Second, lymphopenia reflects immunosuppression.^[26] CO poisoning activates the sympathetic nervous system and the hypothalamic–pituitary–adrenal axis, which leads to elevated cortisol levels that induce lymphocyte apoptosis and redistribution. Lymphopenia reflects the body’s stress-induced immunosuppressive state, which may weaken neural repair and anti-inflammatory capabilities.^[27]

Substantial heterogeneity persists among the included studies (I² = 65.1%). Although subgroup analyses (sample size, study region, study design, follow-up duration, NLR cutoff value, and DNS rate) and meta-regression were added, the sources of heterogeneity are still only partially explained. Study design appears influential, yet residual heterogeneity remains. The heterogeneity may be attributed to the following: numerous inadequately documented confounding factors – including history of coma, Glasgow Coma Scale score, carboxyhemoglobin level, time interval from exposure to blood draw, duration of coma, hyperbaric oxygen protocols, and diagnostic criteria for DNS – contributed to the between-study variability. Moreover, the baseline incidence of DNS itself varied widely, ranging from 5.4% to 38.3%, which further amplified the overall heterogeneity. NLR measurement time: although all were measured upon admission, the time intervals from poisoning to hospital visit might vary among the studies, which could have affected the absolute value of NLR. Differences in the selection methods of population: Pediatric research showed that the AUC of NLR in predicting DNS was 0.828, with a cutoff value of >4;^[11] this finding suggests that NLR also has predictive value in the pediatric population. However, differences in immune responses may exist between children and adults. Since the NLR thresholds used in each study are different, it is recommended that a unified optimal cutoff value be adopted in future studies.

This study encountered the following limitations: first, most of the included studies were of retrospective designs, which might have introduced selection and information bias. Some research did not adequately control for confounding factors (such as age, underlying diseases, and severity of poisoning). Second, significant heterogeneity was noted among the studies, which limited the generalizability of the results. Although subgroup analyses and meta-regression were conducted, the sources of heterogeneity could not be fully explained. Third, most of the studies did not report the dynamic changes in NLR; a single measurement may not reflect the dynamic process of the inflammatory response. Fourth, a large proportion of the studies originated from a single geographical region (mainly China), which may limit the generalizability of our findings to other populations with different genetic backgrounds, healthcare systems, and environmental exposures. This underscores the need for multi-center prospective validation across diverse populations.

Given NLR’s limited predictive efficacy as a single marker, future research should pursue 2 main directions. First, multi-marker models combining NLR with brain injury biomarkers (e.g., neuron specific enolase, Recombinant Human Central Nervous System Specific Protein [S100β]) and oxidative stress indicators (e.g., ischemia‑modified albumin) could improve DNS prediction sensitivity and specificity. Second, the biomarker pool for the inflammation-oxidative stress network should be expanded. Recent studies show that phenolic-rich extracts from Glycyrrhiza glabra, Prunus cerasus, and Myrtus communis possess anti-inflammatory, antioxidant, and immunomodulatory properties. Although not CO-related, these findings reveal inflammatory-oxidative crosstalk and suggest that similar approaches could identify circulating bioactive molecules as supplementary markers to NLR, enabling more comprehensive pathophysiological assessment.

In summary, current evidence indicates that elevated NLR levels following acute CO poisoning are associated with an increased risk of DNS, supporting its potential as a prognostic marker. However, given the substantial heterogeneity, potential biases, limited number of included studies, absence of standardized thresholds, and lack of prospective validation, the reported AUC of 0.88 may be overestimated. Moreover, positive and negative predictive values are expected to vary with local DNS prevalence. Therefore, NLR cannot currently be considered an independent basis for clinical decision-making. Its role should be viewed as an auxiliary, low-cost risk-stratification tool, particularly in resource-limited emergency settings where advanced imaging or continuous monitoring is unavailable. Future research should evaluate NLR in combination with other biomarkers (e.g., S100 and neuron specific enolase), assess its integration into machine-learning models, and examine its cost-effectiveness. Before NLR is incorporated into early assessment strategies, high-quality prospective studies are needed to establish the optimal cutoff value, confirm its incremental prognostic utility, and define its specific role within clinical decision pathways.

Acknowledgments

The authors thank all colleagues who participated in this study.

Author contributions

Conceptualization: Wen Pin Xu.

Data curation: Xiao Ting Liu, Yong Yan Han, Xiao Lu Mi, Qing Mian Xiao, Wei Zhan Wang, Wen Pin Xu.

Formal analysis: Xiao Ting Liu.

Funding acquisition: Xiao Ting Liu, Wei Zhan Wang.

Writing – original draft: Xiao Ting Liu.

Writing – review & editing: Xiao Ting Liu, Wei Zhan Wang, Wen Pin Xu.