The neutrophil–lymphocyte ratio and incident atherosclerotic events: analyses from five contemporary randomized trials

Abstract

Aims

The neutrophil–lymphocyte ratio (NLR) is a readily available inflammatory biomarker that may associate with atherosclerosis and predict cardiovascular (CV) events. The aims of this study are to determine whether the NLR predicts incident major adverse cardiovascular events (MACE) and is modified by anti-inflammatory therapy.

Methods and results

Baseline and on-treatment NLRs were calculated from complete blood counts among 60 087 participants randomized in the CANTOS, JUPITER, SPIRE-1, SPIRE-2, and CIRT trials to receive placebo or canakinumab, rosuvastatin, bococizumab, or methotrexate, respectively, and followed up for MACE. All analyses were performed first in CANTOS, and then externally validated in the other four trials. For the five trials, hazard ratios for major CV events and mortality comparing NLR quartiles were computed using Cox proportional hazards models, and the effect of each randomized intervention on the NLR was evaluated in comparison to placebo. The NLR modestly correlated with interleukin-6, C-reactive protein, and fibrinogen levels but minimally with lipids. In all five randomized trials, baseline NLR predicted incident CV events and death; the per-quartile increase in risk of MACE was 20% in CANTOS [95% confidence interval (CI) 14–25%, P < 0.0001], 31% in SPIRE-1 (95% CI 14–49%, P = 0.00007), 27% in SPIRE-2 (95% CI 12–43%, P = 0.0002), 9% in CIRT (95% CI 0.2–20%, P = 0.045), and 11% in JUPITER (95% CI 1–22%, P = 0.03). While lipid-lowering agents had no significant impact on the NLR, anti-inflammatory therapy with canakinumab lowered the NLR (P < 0.0001).

Conclusion

The NLR, an easily obtained inflammatory biomarker, independently predicts CV risk and all-cause mortality, and is reduced by interleukin-1β blockade with canakinumab.

See page 904 for the editorial comment on this article (doi: 10.1093/eurheartj/ehaa1104)

Introduction

Four recent clinical trials demonstrate that targeting inflammation improves clinical outcomes for patients with atherosclerosis. In 2017, the Canakinumab Thrombosis Outcomes Study (CANTOS) reported that interleukin (IL)-1β inhibition with canakinumab significantly reduces cardiovascular (CV) event rates without lowering cholesterol or blood pressure.¹ In 2019 and 2020, the Colchicine Cardiovascular Outcomes Trial (COLCOT)² and the second Low Dose Colchicine trial (LoDoCo-2)³ provided independent affirmation of the inflammation hypothesis of atherosclerosis using colchicine, a drug that interferes, in part, with IL-1 activation.⁴ In CANTOS, the greatest clinical benefit accrued among those with the greatest reductions in IL-6 and C-reactive protein (CRP).^{5  ,  6} By contrast, the Cardiovascular Inflammation Reduction Trial (CIRT) failed to find a significant CV benefit from methotrexate.⁷ This intervention also did not reduce CRP or IL-6, thus providing an informative neutral study. These four contemporary trials not only increase interest in traditional biomarkers of inflammation such as IL-1, IL-6, and CRP, but also in finding a simple biomarker widely available to the clinical community.

One such biomarker is the neutrophil–lymphocyte ratio (NLR) which is derived from the complete blood count. The NLR leverages early haematologic observations that the absolute neutrophil count (ANC) is positively associated with CV events, while the absolute lymphocyte count (ALC) is negatively associated.^{8  ,  9} Moreover, while total white blood cell count has been shown to predict CV risk, the NLR appears to be a superior predictor.⁹ Small studies have evaluated the NLR as a predictor of total mortality in the setting of acute coronary interventions,^10–12 hypertension,^{13  ,  14} and heart failure.¹⁵ However, few data evaluate the NLR in the context of contemporary large-scale CV clinical trials.

We revisited the potential clinical utility of the NLR to predict incident CV events and all-cause mortality in five major contemporary clinical trials, which together included more than 60 000 patients. Then, leveraging the unique biologic context of each trial, we evaluated whether random allocation to four different drug classes altered the NLR over time compared to placebo.

Methods

Clinical trials and participants

CANTOS randomized 10 061 patients with prior myocardial infarction (MI) and a baseline high‐sensitivity CRP (hsCRP) ≥2mg/L to placebo or canakinumab in doses of 50 mg, 150 mg, or 300 mg subcutaneously once every 3 months.¹ Subjects were followed up for a primary CV endpoint of three-point major adverse cardiovascular events (MACE), defined as the composite of non-fatal MI, non-fatal stroke, or CV death (median follow-up 3.7 years). A secondary endpoint was defined as the primary CV endpoint plus hospitalization for unstable angina requiring urgent revascularization (MACE+).

Justification for the Use of Statins in Prevention: an Intervention Trial Evaluating Rosuvastatin (JUPITER) randomized 17 802 healthy patients with LDL-cholesterol (LDL-C) of less than 130 mg/dL and hsCRP ≥2.0 mg/L to 20 mg of rosuvastatin or placebo.¹⁶ The primary endpoint was the composite of MI, stroke, arterial revascularization, hospitalization for unstable angina, or CV death (median follow-up: 1.9 years).

Studies of PCSK9 Inhibition and the Reduction of Vascular Events (SPIRE) in two separate trials (SPIRE-1 and SPIRE-2) collectively randomized 27 438 patients (16 817 and 10 621 respectively) with prior CV events or at high risk to 150 mg of bococizumab subcutaneously every two weeks or placebo.¹⁷ The LDL-C cut-offs were greater than or equal to 70 mg/dL in SPIRE-1 and greater than or equal to 100 mg/dL in SPIRE-2. Patients were followed up for a primary endpoint of MACE+ (median follow-up 7 months in SPIRE-1 and 12 months in SPIRE-2).

CIRT randomized 4786 patients with previous MI or multi-vessel coronary disease and either type 2 diabetes or metabolic syndrome to low-dose methotrexate (15–20 mg weekly) or placebo.⁷ Patients were followed up for a primary endpoint of MACE+ (median follow-up 2.3 years).

Measurement of the ANC, ALC, and NLR

In all trials, the baseline NLR was computed from the ANC and ALC from complete blood count data obtained at randomization. In CANTOS, follow-up complete blood counts were measured at 3, 6, 9, 12, 24, 36, and 48 months, allowing us to address stability over time. In JUPITER, follow-up samples were collected at the final visit, the timing of which varied for each participant. In SPIRE-1 and SPIRE-2, complete blood counts are available from the 14, 26, 40, and 52-week follow-up visits; the largest sample of follow-up measurements was obtained at 52 weeks. In CIRT, follow-up measurements were acquired at four and eight months.

Randomized trial analyses and incident cardiovascular events

We first divided CANTOS subjects into quartiles according to baseline ANC, ALC, and NLR. Cox proportional-hazard models stratified by time since index infarction estimated the relative hazards for MACE, MACE+, all-cause mortality, and CV death across quartiles. P-values for the test of trend were calculated across these four groups. Multivariable-adjusted models included treatment group, age, sex, body mass index, hypertension, diabetes, smoking, LDL-C, and CRP. To address the ANC and ALC together, sixteen groups were formed from all combinations of ANC and ALC quartiles and hazard ratios were computed comparing each group to those in the lowest quartile of both ANC and ALC. Within CANTOS, Spearman correlation coefficients assessed the magnitude of association between the NLR, traditional CV risk factors, and inflammatory markers.

We sought to externally validate any observations made in CANTOS in the independent JUPITER, SPIRE-1, SPIRE-2, and CIRT trials. Using parallel methods, participants in each trial were divided into quartiles of baseline NLR. Hazard ratios comparing baseline NLR quartiles for the same outcomes were computed using cox regressions. To allow for broader clinical application, this analysis was repeated dividing participants into four NLR groups (<1.5, 1.5–<2.5, 2.5–<3.5, and >3.5). These cut-off points were determined by selecting values similar to the NLR quartiles from CANTOS, but rounded for ease of use and generalizability across trials. Multivariable analyses adjusted for treatment group, age, sex, hypertension, diabetes, LDL-C, and CRP. Multivariable analyses from JUPITER did not include LDL-C or diabetes (both exclusion criteria) as covariates. The CIRT models were not adjusted for diabetes, as the cox proportional-hazards model was stratified by diagnosis of diabetes vs. metabolic syndrome.

The effect of each randomized drug (canakinumab in CANTOS, rosuvastatin in JUPITER, bococizumab in SPIRE-1 and SPIRE-2, and methotrexate in CIRT) on the NLR was determined by computing the median change in NLR from randomization to follow-up.

Wilcoxon two-sample t-tests were used for comparison of two groups of numerical data, Kruskal–Wallis tests for three or more groups of numerical data, and χ² tests for comparison of categorical data. All P-values are two-sided and all confidence intervals (CIs) computed at the 95% level. All statistical analysis was performed using SAS version 9.4.

Results

Initial CANTOS analyses: relationships of baseline cell counts to usual risk factors, stability over time, and to future clinical outcomes

The baseline distribution of the NLR from CANTOS is presented in Figure 1A. Supplementary material online, Tables S1–S3 show clinical characteristics of the CANTOS participants according to increasing baseline quartiles of ANC, ALC, and NLR, respectively. CANTOS showed modest positive correlations between baseline NLR and CRP, IL-6, and fibrinogen, but minimal correlation with lipid levels (Supplementary material online, Table S4). Within the CANTOS placebo group, the ANC, ALC, and NLR remained stable over time (Supplementary material online, Figure S1).

Figure 1.

Baseline NLR distribution. Histogram plots for baseline NLR from CANTOS (A), JUPITER (B), SPIRE-1 (C), SPIRE-2 (D), CIRT (E), and all trials combined (F).

The baseline ANC in CANTOS positively associated with incident CV events and all-cause mortality (Figure 2A and Supplementary material online, Table S5). An inverse association for these endpoints was observed for the ALC (Figure 2B and Supplementary material online, Table S6). The greatest risk for MACE was observed among participants in the highest ANC and lowest ALC quartiles (hazard ratio 1.61, 95% CI 1.21–2.13, P = 0.0009) (Supplementary material online, Table S7).

Figure 2.

NLR and incident atherosclerotic events. Left: hazard ratios for cardiovascular events and mortality across quartiles of baseline ANC (A), ALC (B), and NLR (C) from CANTOS. Right: hazard ratios for MACE+ across NLR quartiles (D) and across clinical cut-off points of NLR (E) in CANTOS, JUPITER (including all revascularizations and hospitalizations for unstable angina), SPIRE-1, SPIRE-2, and CIRT.

Given these differential associations for the ANC and ALC, the NLR emerged as the strongest overall blood count-based predictor of incident CV risk (Figure 2C and Table 1). Within CANTOS, each increasing NLR quartile increased the hazard of MACE by 22% (95% CI 16–28%, P < 0.0001), MACE+ by 20% (95% CI 14–25%, P < 0.0001), CV death by 36% (95% CI 27–46%, P < 0.0001), and all-cause mortality by 38% (95% CI 30–46%, P < 0.0001). These effects were modestly attenuated but remained statistically significant in multivariable-adjusted analyses (all P-values across quartiles < 0.0001) (Table 1). Per-quartile hazard ratios for various sub-groups are presented in Supplementary material online, Figure S2.

Table 1.

NLR and incident atherosclerotic events in CANTOS

CANTOS	NLR 1st quartile	NLR 2nd quartile	NLR 3rd quartile	NLR 4th quartile	Effect across groups
NLR (baseline) range	NLR ≤ 1.81	1.81 < NLR ≤ 2.36	2.37 < NLR ≤ 3.08	3.09 < NLR
n	2507	2510	2504	2514
MACE
IR	3.21 (295)	3.54 (322)	4.14 (373)	5.72 (495)
HR (95% CI); P	1 (ref)	1.11 (0.94–1.29); 0.21	1.29 (1.11–1.50); 0.001	1.78 (1.54–2.06); <0.0001	1.22 (1.16–1.28); <0.0001
HRadj (95% CI); P	1 (ref)	1.07 (0.91–1.26); 0.39	1.23 (1.05–1.43); 0.01	1.51 (1.30–1.75); <0.0001	1.15 (1.10– 1.21); <0.0001
MACE+
IR	3.71 (337)	3.97 (358)	4.59 (409)	6.27 (536)
HR (95% CI); P	1 (ref)	1.07 (0.92–1.24); 0.36	1.23 (1.07–1.43); 0.004	1.69 (1.47–1.93); <0.0001	1.20 (1.14–1.25); <0.0001
HRadj (95% CI); P	1 (ref)	1.05 (0.90–1.21); 0.57	1.19 (1.02–1.37); 0.02	1.46 (1.26–1.68); <0.0001	1.14 (1.09–1.19); <0.0001
CV death
IR	1.21 (116)	1.36 (130)	1.61 (154)	2.90 (265)
HR (95% CI); P	1 (ref)	1.13 (0.88–1.45); 0.34	1.34 (1.05–1.70); 0.02	2.41 (1.94–2.99); <0.0001	1.36 (1.27–1.46); <0.0001
HRadj (95% CI); P	1 (ref)	1.07 (0.83–1.37); 0.62	1.21 (0.95–1.55); 0.13	1.76 (1.41–2.21); <0.0001	1.22 (1.14–1.31); <0.0001
All-cause mortality
IR	1.97 (189)	2.15 (206)	2.51 (240)	4.83 (442)
HR (95% CI); P	1 (ref)	1.10 (0.90–1.34); 0.34	1.28 (1.06–1.55); 0.01	2.46 (2.08–2.92); <0.0001	1.38 (1.30–1.46); <0.0001
HRadj (95% CI); P	1 (ref)	1.02 (0.84–1.24); 0.85	1.12 (0.92–1.35); 0.26	1.74 (1.46–2.08); <0.0001	1.22 (1.15–1.29); <0.0001

Incidence rates, hazard ratios, and multivariable adjusted hazard ratios for MACE, MACE+, CV death, and all-cause mortality comparing NLR quartiles from CANTOS. Crude model adjusted for treatment group. Multivariable model adjusted for randomized treatment group, age, sex, baseline LDL-C, smoking, hypertension, diabetes, BMI, and hsCRP.

CI, confidence interval; HR, hazard ratio; HRadj, multivariable adjusted hazard ratio; IR, incidence rate (per 100 person-years).

External validation: JUPITER, SPIRE-1, SPIRE-2, and CIRT

The distribution of the NLR at baseline was consistent across trials (Figure 1A–F) and quartile-based analyses of baseline NLR as a predictor of future adverse CV events in the JUPITER, SPIRE-1, SPIRE-2, and CIRT trials were similar to those observed in CANTOS (Figure 2D, Table 2). Moreover, effects of the NLR on all-cause mortality were also consistent across CANTOS, JUPITER, SPIRE-1, SPIRE-2, and CIRT (Supplementary material online, Tables S8–S11). Baseline NLR predicted death from CV causes in all trials except for JUPITER, a primary prevention trial. When using clinical cut-off points of <1.5, 1.5–<2.5, 2.5–<3.5, and ≥3.5 rather than trial-specific quartiles, all five trials showed a common pattern of increasing risk with increasing NLR (Figure 2E, Supplementary material online, Tables S12–S17).

Table 2.

NLR and incident MACE+ in CANTOS, JUPITER, SPIRE-1, SPIRE-2, and CIRT

MACE+	NLR 1st quartile	NLR 2nd quartile	NLR 3rd quartile	NLR 4th quartile	Effect across groups
CANTOS
NLR (baseline) range	NLR ≤ 1.81	1.81 < NLR ≤ 2.36	2.36 < NLR ≤ 3.08	3.09 < NLR
n	2507	2510	2504	2514
IR	3.71 (337)	3.97 (358)	4.59 (409)	6.27 (536)
HR (95% CI); P	1 (ref)	1.07 (0.92–1.24); 0.36	1.23 (1.07–1.43); 0.004	1.69 (1.47–1.93); <0.0001	1.20 (1.14–1.25); <0.0001
HRadj (95% CI); P	1 (ref)	1.05 (0.90–1.21); 0.57	1.19 (1.02–1.37); 0.02	1.46 (1.26–1.68); <0.0001	1.14 (1.09–1.19); <0.0001
JUPITER
NLR (baseline) range	NLR ≤ 1.58	1.58 < NLR ≤ 2.10	2.10 < NLR ≤ 2.75	2.75 < NLR
n	4160	4197	4147	4183
IR	0.96 (80)	0.90 (79)	1.03 (90)	1.28 (113)
HR (95% CI); P	1 (ref)	0.95 (0.69–1.29); 0.73	1.08 (0.80–1.46); 0.62	1.33 (1.00–1.77); 0.053	1.11 (1.01–1.22); 0.03
HRadj (95%CI); P	1 (ref)	0.91 (0.66–1.24); 0.53	0.94 (0.70–1.28); 0.71	1.03 (0.77–1.38); 0.85	1.02 (0.93–1.12); 0.73
SPIRE-1
NLR (baseline) range	NLR ≤ 1.72	1.72 < NLR ≤ 2.24	2.24 < NLR ≤ 2.93	2.93 < NLR
n	4025	4104	4244	4336
IR	2.86 (78)	2.23 (73)	2.56 (73)	4.28 (131)
HR (95% CI); P	1 (ref)	0.79 (0.56–1.10); 0.161	0.91 (0.66–1.25); 0.55	1.53 (1.16–2.04); 0.003	1.31 (1.14–1.49); 0.00007
HRadj (95%CI); P	1 (ref)	0.74 (0.53–1.04); 0.081	0.78 (0.56–1.08); 0.13	1.25 (0.93–1.67); 0.14	1.19 (1.03–1.36); 0.01
SPIRE-2
NLR (baseline) range	NLR ≤ 1.67	1.67 < NLR ≤ 2.16	2.16 < NLR ≤ 2.83	2.83 < NLR
n	2785	2706	2567	2473
IR	2.67 (76)	3.52 (97)	4.33 (113)	4.68 (116)
HR (95% CI); P	1 (ref)	1.32 (0.98–1.78); 0.07	1.62 (1.21–2.17); 0.001	1.73 (1.29–2.31); 0.0002	1.27 (1.12–1.43); 0.0002
HRadj (95% CI); P	1 (ref)	1.27 (0.94–1.72); 0.12	1.50 (1.12–2.02); 0.007	1.54 (1.14–2.08); 0.005	1.20 (1.06–1.36); 0.005
CIRT
NLR (baseline) range	NLR ≤ 1.73	1.73 < NLR ≤ 2.27	2.27 < NLR ≤ 2.96	2.96 < NLR
n	1192	1193	1191	1192
IR	3.85 (92)	3.79 (92)	4.01 (98)	5.27 (123)
HR (95% CI); P	1 (ref)	0.99 (0.74–1.32); 0.92	1.02 (0.77–1.35); 0.90	1.31 (1.00–1.72); 0.05	1.09 (1.00–1.20); 0.045
HRadj (95% CI); P	1 (ref)	0.99 (0.73–1.35); 0.94	1.03 (0.76–1.39); 0.88	1.35 (1.01–1.80); 0.04	1.10 (1.01–1.21); 0.04

Incidence rates, hazard ratios, and multivariable adjusted hazard ratios for MACE+ (Jupiter inclusive of all revascularizations and all hospitalizations for unstable angina) comparing NLR quartiles in five randomized controlled trials. Crude model adjusted for treatment group. Multivariable model adjusted for randomized treatment group, age, sex, LDL-C, smoking, hypertension, BMI, diabetes, and hsCRP (JUPITER model not adjusted for LDL-C or diabetes. CIRT not adjusted for diabetes).

CI, confidence interval; HR, hazard ratio; HRadj, multivariable adjusted hazard ratio; IR, incidence rate (per 100 person-years).

Effects of pharmacotherapy on ANC, ALC, and NLR

In JUPITER, SPIRE-1, and SPIRE-2, lipid-lowering therapy with rosuvastatin and bococizumab, respectively, had no significant effect on the NLR (Supplementary material online, Tables S18–S20). In CIRT, methotrexate increased the NLR by 15.3% compared to 0% in the placebo group at eight months (P < 0.0001) due to a decrease in the ALC (Supplementary material online, Table S21). By contrast, canakinumab significantly reduced the NLR in a dose-dependent manner; the median change in NLR at three months was −1.62%, −11.3% (P < 0.0001), −16.7% (P < 0.0001), and −23.6% (P < 0.0001) in the placebo and canakinumab 50, 150, and 300 mg groups, respectively (Figure 3, top, Supplementary material online, Table S22). These changes were dependent on reductions in the ANC of 0%, −9.26%, −15%, and −21.4%. Canakinumab did not significantly influence the ALC. The dose-dependent reductions in the NLR with canakinumab mimic that for CRP and IL-6 (Figure 3, bottom).

Figure 3.

Change in NLR in CANTOS. Top: median change (%) in NLR across randomized treatment groups. Bottom: median change in hsCRP, IL-6, and NLR at 3 months.

Discussion

Among 60 087 participants in five contemporary randomized trials, the NLR consistently predicted future CV events and all-cause mortality. In secondary prevention settings, these effects were only modestly attenuated after adjustment for usual risk factors and CRP. The observed effects were weaker in the primary prevention JUPITER cohort, possibly due to lower event rates, but baseline NLR still significantly predicted the trial’s primary outcome and all-cause mortality in unadjusted analyses. Although the NLR may be a weaker predictor of MACE than hsCRP¹⁸ or IL-6,¹⁹ it has the advantage of coming from a test that is obtained ubiquitously in routine blood draws. Among those allocated to placebo, NLR levels were stable over time, raising the potential for use as a clinical biomarker. Finally, our data demonstrate that random allocation to lipid-lowering therapies had no significant effect on the NLR, while methotrexate increased and canakinumab decreased the NLR.

The current data make significant contributions to the literature of inflammation and atherosclerosis. First, the extent of data available from five large contemporary trials not only allowed us to greatly expand evidence from prior smaller studies relating the ANC, ALC, and NLR to incident vascular events,^{8  ,  9} but also allowed us to demonstrate the consistency of NLR measures over time and to create simple clinical cut-off points that appear effective across diverse trial populations.

Second, previously, the effects of CV pharmacotherapy on the NLR had only been studied in samples of less than one hundred participants and delivered uncertain results.^{13  ,  14  ,  20} The current data demonstrate with clarity that neither rosuvastatin nor bococizumab has a significant effect on the NLR. It is unsurprising that the PCSK9 inhibitor would not have an effect on the NLR.²¹ However, statins consistently lower CRP independent of their effects on LDL-C.²² Thus, while we observed a modest correlation between hsCRP and NLR, the current data indicate that these variables do not measure an identical process. The NLR may therefore be a marker of an inflammatory pathway orthogonal to the effect of statins, but still relevant to atherothrombosis, and a potential therapeutic target.

Third, our results raise interesting hypotheses regarding therapeutic monitoring of potential anti-inflammatory therapies. On the one hand, methotrexate, which failed to lower CV event rates in CIRT, increased the NLR almost exclusively through a reduction in the lymphocyte count. By contrast, IL-1β inhibition with canakinumab, which significantly lowered CV event rates in CANTOS, lowered the NLR in a dose-dependent manner. Whether these changes will associate with coronary plaque development and rupture will require longer follow-up. The NLR may therefore serve as a marker for response to anti-inflammatory therapy, allowing for more targeted use of these agents, which are often costly.

The current data cannot address whether an elevated NLR is part of a causal pathway for atherosclerotic events. On the one hand, the NLR is non-specific and predicts outcomes in other settings, such as in patients with cancer,²³ appendicitis,²⁴ and sepsis.²⁵ The NLR may also rise with physiologic stress, which is supported by the fact that glucocorticoids cause relative neutrophilia and lymphopenia.

On the other hand, the myeloid cell compartment has long been known to be a crucial player in the pathogenesis of atherothrombosis. The canonical model for atherosclerotic plaque development begins with circulating monocytes adhering to the endothelium and migrating into the sub-endothelial space. These cells transform into monocyte-derived macrophages, proliferate, phagocytose oxidized LDL-C, further transform into foam cells, ultimately forming much of the lipid core of unstable plaques. A similarly robust description of neutrophils has been lacking until recently. Emerging research increasingly indicates an integral role for neutrophils in atherothrombosis. Neutrophils localize near plaques, especially those with rupture-prone features.²⁶ Neutrophils further enhance monocyte adhesion and transmigration into atherosclerotic plaques and contribute to endothelial cell dysfunction and oxidative stress by releasing myeloperoxidase, NADPH oxidase, lipoxygenases, and neutrophil extracellular traps (NETs).^{27  ,  28} Neutrophil-derived mediators may also influence lesion fibrous cap rupture or contribute to superficial erosion, two mechanisms that promote atherothrombosis. In this regard, neutrophils promote fibrous cap rupture through release of proteases, which degrade matrix elements.²⁶ Neutrophils can participate in superficial plaque erosion by promoting endothelial cell death and desquamation.²⁹ NETs can propagate atherothrombosis and present tissue factor at sites of infarction.²⁸ Finally, early massive neutrophil influx into ischaemically injured myocardium can cause collateral damage and impair myocardial healing.²⁷ How much altered granulocyte production in the bone marrow and altered margination of these cells affect NLR remains unknown. Moreover, changes in particular lymphocyte populations that could influence NLR merit consideration.

The NLR outperformed the ANC as a predictor of CV risk due, in part, to the atheroprotective contribution of higher lymphocyte counts. Mechanistic research suggests that lymphocytes can be both atheroprotective and atherogenic. T helper (T_H)-1 cells, which secrete interferon gamma, and T_H17 cells are atherogenic. Conversely, T_reg cells, which produce anti-inflammatory mediators IL-10 and transforming growth factor-β, and T_H2 cells are atheroprotective. B cells can also shift the balance in either direction. While B1 cells produce IgM that marks lipids for Fc receptor-mediated removal, B2 cells secrete atherogenic inflammatory cytokines.²⁷ It is also possible that the observed atherogenic effect of lymphopenia reflects a more generalized process, such as frailty.³⁰

Limitations

Our study has limitations. First, analyses of the NLR were not pre-specified when these clinical trials were designed and thus these data are hypothesis-generating. Second, despite the consistency of our findings, there is heterogeneity in the inclusion and exclusion criteria, end points, and follow-up periods between trials, making comparison imprecise. This heterogeneity, however, could be viewed as an advantage, as it allowed us to examine the NLR in multiple epidemiologic and biologic contexts. Third, pre-existing anti-inflammatory therapy or neoplastic diseases could have direct effects on the NLR, altering its utility as a clinical biomarker. These conditions were excluded in all five trials, so they are unlikely to act as confounders. Moreover, while haematologic conditions may confound the NLR in such a way that it is not useful for prediction of CV risk, auto-inflammatory conditions may directly influence the NLR in such a way that CV risk is increased. As such, it would be useful to include these conditions in future analyses when possible. Finally, CANTOS and JUPITER enrolled only individuals with CRP levels greater than or equal to 2 mg/L, which could influence the NLR distribution given the modest correlation between NLR and CRP. However, the relative similarity of the NLR distributions across all trials, including those that did not screen for CRP, suggests that the magnitude of any such effect is likely to be small.

Future directions

Out data should influence future research and guidelines related to inflammation and atherosclerosis. First, incorporation of NLR monitoring into clinical practice guidelines for management of stable coronary artery disease may identify many patients with residual inflammatory risk not explained by traditional risk factors or CRP. Second, given the clinical significance of the association between NLR and MACE, this interaction should be further explored mechanistically. Third, our data promote monitoring of the NLR at baseline and over time in future clinical trials, especially those involving anti-inflammatory agents. Doing so could determine if any beneficial effect is mediated through changes in this biomarker. Fourth, incorporation of the NLR into a reclassification analysis using an accepted risk score would clarify how the biomarker might be used for risk stratification. Finally, our data justify the assessment of CV efficacy of drugs that are known to lower the NLR.

Conclusion

These contemporary data from 60 087 participants in five clinical trials identify the NLR as an inexpensive and easily obtained biomarker that independently predicts CV risk in both primary and secondary prevention settings. Moreover, our drug response data for canakinumab raises the possibility of using the NLR to monitor emerging anti-inflammatory strategies for atherothrombosis. These data should motivate future research into the pathophysiology of inflammation and atherosclerosis, and clinical trials on the CV efficacy of anti-inflammatory agents.

Supplementary material

Supplementary material is available at European Heart Journal online.

Data availability

The data underlying this article were used by permission of the individual trial sponsors. Data will be shared on reasonable request to the corresponding author with permission from the individual sponsors, if required.

Funding

The CANTOS, JUPITER, SPIRE, and CIRT trials were funded by Novartis, AstraZeneca, Pfizer, and the National Heart Lung and Blood Institute, respectively.

Conflict of interest: J.G.M. reports grants from Novartis and AstraZeneca, during the conduct of the CANTOS and Jupiter Trials. R.J.G. reports receiving salary support from grants (AstraZeneca, Kowa, Novartis, and Pfizer) to the Brigham & Women’s Hospital. P.L. is an unpaid consultant to, or involved in clinical trials for Amgen, AstraZeneca, Baim Institute, Beren Therapeutics, Esperion, Therapeutics, Genentech, Kancera, Kowa Pharmaceuticals, Medimmune, Merck, Norvo Nordisk, Merck, Novartis, Pfizer, Sanofi-Regeneron. P.L. is a member of scientific advisory board for Amgen, Corvidia Therapeutics, DalCor Pharmaceuticals, Kowa Pharmaceuticals, Olatec Therapeutics, Medimmune, Novartis, and XBiotech, Inc. P.L.’s laboratory has received research funding in the last 2 years from Novartis. P.L. is on the Board of Directors of XBiotech, Inc. P.L. has a financial interest in Xbiotech, a company developing therapeutic human antibodies. P.L.’s interests were reviewed and are managed by Brigham and Women's Hospital and Partners HealthCare in accordance with their conflict of interest policies. P.L. receives funding support from the National Heart, Lung, and Blood Institute (1R01HL134892), the American Heart Association (18CSA34080399), the RRM Charitable Fund, and the Simard Fund. P.M.R. served as the Principle Investigator for the CANTOS, CIRT, JUPITER, SPIRE-1, and SPIRE-2 trials which were supported by investigator-initiated research grants to the Brigham and Women’s Hospital by Novartis, the National Heart Lung and Blood Institute, AstraZeneca, and Pfizer, respectively. P.M.R. has served as a consultant to Novartis, Inflazome, Flame, Agepha, Uppton, and Corvidia and was listed as a co-inventor on patents related to inflammatory biomarkers in cardiovascular disease that, now expired, were licensed in the past to Siemens and AstraZeneca.