The Association Between Absolute Lymphocyte Count and Long-Term Mortality in Critically Ill Medical Patients: Propensity Score-Based Analyses

Background

Advances in critical care over the past two decades have led to improved survival and a steady increase in the number of patients discharged from the intensive care unit (ICU); therefore, studies exploring long-term outcomes are currently a research priority in critical care.^1,2 Accumulating studies have found that critical illness may lead to prolonged physical, metabolic, and immunological impairment, a condition known collectively as post-intensive care syndrome (PICS).^3,4 Hence, the identification of early predictors for long-term outcomes is an essential issue in critical care.

Absolute lymphocyte count (ALC) is a frequently assessed laboratory measurement varied with age and sex.^5,6 Accumulating evidence have shown that low ALC was associated with poor outcomes, mainly short-term mortality, in a wide range of diseases, including sepsis, pneumonia, malignancy, heart failure, and coronavirus disease (COVID) infection.^7–11 Recent studies have found that reduced ALC may correlate with impaired adaptive immunity, particularly T-cell immunity, and impaired adaptive immunity may further lead to an increased risk of infection after acute illness and poor long-term outcomes.^12–14 Our recently published study also found that impaired T-cell immunity on day-8 correlated with high hospital mortality in critically ill elderly patients with sepsis.¹⁵ However, the clinical evidence on the association between early ALC and long-term mortality in critically ill patients remains unclear. In the present study, we linked the critical care database at Taichung Veterans General Hospital (TCVGH) with Taiwan’s nationwide death-registry file and used a propensity score-based approach to investigate the association between week-one ALC and long-term all-cause mortality in patients who were admitted to the medical ICUs.

Methods

Ethical Approval

The Institutional Review Board at the TCVGH approved this study (SE20249B-2) with a waiver of informed consent due to all of the data used were de-identified data. The study complies with the Declaration of Helsinki.

Patient Population and Definition of the Main Exposure

This study utilized a retrospective cohort design and included consecutively admitted critically ill patients at TCVGH, a 1624-bed referral centre in central Taiwan, during the period from 2015 to 2020.

The inclusion criteria were consecutive patients who were admitted to ICUs at TCVGH between 2015 and 2020. Given that we focused on the long-term outcome of patients who were admitted to medical ICUs in this study, the exclusion criteria were compromised of 1) patients who were admitted to surgical ICUs, 2) patients who were admitted to cardiac ICU, and 3) patients whose ICU admission was less than 24 hours. The index ICU admission for patients with multiple admissions was defined as their first admission. We used the average level of ALC in patients with more than two sets of ALC within the first week.

Primary Outcome

The outcome of interest was the time to all-cause mortality that was retrieved from the death registration profile of Taiwan’s National Health Insurance Database (NHID), with the censored date was defined as the date of death in the NHID or March 31, 2021. In brief, Taiwan has mandated a compulsory National Health Insurance (NHI) program since 1995, which encompasses the vast majority of the population; therefore, the date of death data utilized in this study should be accurate.¹⁶

Covariates

The critical care database at TCVGH was utilized to obtain essential clinical data, including demographic details, laboratory results, comorbidities evaluated using the modified Charlson Comorbidity Index (CCI),¹⁷ Acute Physiology and Chronic Health Evaluation (APACHE) II score, presence of septic shock that was defined by diagnosis of sepsis, serum lactate level greater than 18.0 mg/dL as well as usage of vasopressors, receiving mechanical ventilation for more than 3 days, underwent renal replacement therapy, receiving blood transfusion during ICU admission, and cumulative fluid balance status during day 1–3, which have been identified as mortality-relevant factors in our previous investigation and other studies.^18,19 We also included the main organ dysfunction/failure as the covariate.

Statistical Analyses

We used the Kolmogorov–Smirnov test to examine the distribution and used the Mann–Whitney U-test and Fisher’s exact test, given that the distribution of few variables was skewed. Continuous data were represented as median (interquartile range, IQR), while categorical data were presented as percentages. Variables were included in the multivariable model if the associated univariable p value was < 0.20 and the variance inflation factor was < 10.²⁰ We used a Cox proportional hazards model to determine the Hazard ratios (HRs) for all-cause mortality and 95% confidence intervals (CIs) after adjusting for potential confounders such as age, sex, comorbidities, and other factors. Statistical analyses were performed using R (version 3.6.0), and the significance level was set at 0.05.

Subgroup Analysis

To measure the significance of the modification effect of covariates, we employed the Wald test to examine whether the association between week-one ALC and long-term mortality might differ across the covariates in this study.

Propensity Score-Based Analyses

In this study, we used propensity score-matching (PSM) along with two other propensity score weighting methods, namely the inverse probability of treatment weight (IPTW) and covariate balancing propensity score (CBPS), to confirm the association between ALC and long-term all-cause mortality among the enrolled patients.^21–23 We employed the optimal nearest neighbour matching algorithm, with a calliper distance of 0.15 in PSM, given that calliper distance <0.20 was considered acceptable in PSM studies with independent covariates, continuous covariates, and the binary outcome.²⁴ However, PSM may exclude some patients from the analysis due to a lack of matched control subjects, and the exclusion of patients may lead to concerns regarding the representativeness of the study population.^21,25 To address this limitation, we also used propensity score weighting methods, including IPTW and CBPS, which aim to include the entire population and adjust for potential confounders.^22,23,26

Sensitivity Analyses

The sensitivity analysis was conducted to test the robustness of the association between a low ALC and mortality using distinct cut-off values to define the low ALC in both original and propensity score-matched populations. Given that we focused on the long-term outcome, we further attempted to verify the finding by excluding patients whose mortality was within 14 days after the ICU admission.

Results

Baseline Characteristics of the Included Critically Ill Patients

Figure 1 shows the enrollment of subjects in this study (Figure 1). A total of 5722 patients were included for analyses, and the one-year mortality was 44.8% (2568/5722). The non-survivor group was older (68, 57–80 vs 71, 60–82 years, p<0.01), more likely to be male (66.6% vs 60.9%, p<0.01), and had a lower body mass index (BMI) (22.7, 20.1–25.9 vs 23.6, 20.8–26.7, p<0.01), and higher Charlson Comorbidity Index (CCI) compared with those in the survivor group (Table 1). The critical illness-relevant variables, including APACHE II score, presence of septic shock, use of mechanical ventilation for more than three days, positive cumulative fluid balance during day 1–3, leukocytosis, anaemia, and thrombocytopenia, were more severe in the non-survivor group than in the survivor group. Notably, we identified a lower absolute lymphocyte count (1549, 1027–2388 vs 1948, 1373–2743 counts/μL, p<0.01) in the non-survivor group compared with that in the survivor group.

Table 1 Characteristics of Critically Ill Patients Categorized by One-Year Mortality

Figure 1 Flowchart of subject enrollment. Abbreviations: ICUs, intensive care units; TCVGH, Taichung Veterans General Hospital.

Mortality Association of ALC and the Subgroup Analysis for Modification Effect

We used Cox regression to investigate the independent risk factors for long-term mortality among the 5722 critically ill patients and found that a lower ALC was independently associated with increased long-term mortality (adjHR 1.091, 95% CI 1.050–1.134) after adjustment for covariates (Table 2). We further examined the modification effects of covariates and found that the strength of association between ALC and long-term mortality appeared to be stronger among patients with younger age, fewer comorbidities, and lower severities of critical illness (Table 3). In detail, the strength of association between low ALC and long-term mortality were slightly stronger among critically ill patient with lower APACHE II, without mechanical ventilation, negative cumulative fluid balance during day 1–3, high level of haemoglobin and low level of serum creatinine, compared with the corresponding values in the comparison groups.

Table 2 Cox Proportional Hazard Regression Analysis for Long-Term Mortality Among the 5722 Critically Ill Patients

Table 3 Stratified Analyses of Modification Effect on the Association Between the Absolute Lymphocyte Count Cut by 1500 Counts/μL and Risk of Mortality

Propensity Score-Matching and Weighting Analyses

We further used propensity score-matching and weighting methods to address the association between the low week-one ALC (<1500 counts/μL) and long-term mortality (Figure 2). A total of 3916 patients were eligible for the PSM analysis (Table 4), and the Supplemental Table 1, Supplemental Table 2 and Supplemental Figure 1 illustrate the high quality of matching demonstrated by the measurement of the standardized mean difference (SMD) and significance of variables between the two groups in the four distinct cohorts. Table 4 summarises the consistent strength of the association between low ALC and risk for long-term mortality in the four populations. The adjusted HRs in the original PSM, IPTW and CBPS populations were 1.327 (95% CI 1.224–1.438), 1.301 (95% CI 1.188–1.424), 1.292 (95% CI 1.186–1.407), and 1.297 (95% CI 1.191–1.412), respectively.

Table 4 Cox Proportional Hazard Regressions for Estimation of the Association Between the Level of Week-One Absolute Lymphocyte Count and Long-Term All-Cause Mortality in Distinct Patient Populations

Figure 2 Flowchart of propensity score matching. Abbreviations: ICUs, intensive care units; TCVGH, Taichung Veterans General Hospital.

Sensitivity Analyses Using Distinct Cut-Off Values for Low ALC

We conducted sensitivity analyses using distinct cut-off values to define the low ALC in both original and propensity score-matched populations (Figure 3). We found that the association between low ALC and increased mortality existed in a dose-response manner in both the original and propensity score-matched populations.

Figure 3 The strength of association between long-term mortality and week-one absolute lymphocyte count with distinct cut-off values. (A) Original population. (B) Propensity-score matched population.

Sensitivity Analyses by Excluding Patients Whose Mortality Was Within 14 Days

We have excluded patients whose ICU admission was less than 24 hours. To further verify the lasting effect of ALC on mortality, we further excluded patients who died within 1–3, 1–5, 1–7, 1–10 and 1–14 days after ICU admission and found the robust association between ALC and mortality after excluding patient died within two weeks (Table 5).

Table 5 Association Between Absolute Lymphocyte Count Lower Than 1500 per μL and Long-Term Mortality in Critically Ill Patients After Excluding Those Who Died Within Particular Durations

Discussion

Long-term outcome in critical care is currently a growing area of research, and in this investigation we aimed to explore the association between week-one ALC and long-term mortality. We analyzed data from 5722 critically ill patients and found that a lower ALC was an independent risk factor for long-term mortality in critically ill patients. Furthermore, the strength of association between ALC and long-term mortality was stronger among patients with younger ages, fewer comorbidities, and lower severities of critical illness. The propensity score-matching and weighting analyses demonstrated a consistent association between ALC and morality. Our findings provided clinical evidence of the crucial role in the long-term outcome of ALC, which is a frequently measured but somehow ignored parameter in critical care.

Critically ill patients are characterized by a complex morbidity and disease course, and they remain vulnerable after discharge from the ICU, with a greater risk of developing complications, a phenomenon known as PICS.⁴ Advances in critical care management in the past two decades have resulted in an increasing number of patients who survive from critical illness; therefore, the long-term outcomes of critically ill patients are a vital area of research that deserves greater attention.^1,27 Of particular importance is the identification of early determinants for long-term outcomes, which consists of not only risk stratification but also guidance for the development of the early therapeutic approach. For example, the long-term mortality impact of ALC may serve as an immunological indicator, and those with high risk might be benefited from early therapeutic approaches, such as vigilance and prevention of opportunistic infection.²⁸

The low ALC may reflect an altered immunological and metabolic status in critical illness leading to a deleterious long-term outcome. In the present study, we focused on patients who were admitted to medical ICUs, with the majority of patients having sepsis, although the exclusion of patients who were admitted to cardiac and surgical ICUs may limit the generalizability of our findings. Similarly, Adrie et al investigated associations between short-term outcomes, including ICU-associated infection and 28-day mortality, and low ALC, classified by 500, 1000 and 1500 counts/μL, among 753 patients admitted to 4 French ICUs, with the majority (79%, 596/753) were Medical admission.⁷ Adrie et al found that no increase of ALC on day-3 after ICU admission was associated with ICU-associated infection (sub-distribution HR 1.37, 95% CI 1.12–1.67) and 28-day mortality (sub-distribution HR 1.67, 95% CI 1.37–2.03), and the present study with high number of enrolled patients and long follow-up period further identified the independent association between ALC and long-term outcome after adjustment of covariates in patients who were admitted to the medical ICUs.⁷

A number of studies, including studies on sepsis and COVID, have shown that elevated cytokine levels resulting from monocyte/macrophage may lead to increased apoptosis and depletion of lymphocytes.^12,29,30 Cheng et al explored transcriptional and metabolic profiling among patients with sepsis and found that the shift from oxidative phosphorylation to aerobic glycolysis and impaired oxidative phosphorylation in leukocytes were implicated with post-septic immunoparalysis, which may lead to an increased susceptibility to secondary infections and a poor outcome.³¹ Edwards et al used lymphocyte respirometry to explore mitochondrial bioenergetic in 93 surgical patients and found a phenotype with hypometabolic lymphocytes sampled from lymphopenic patients.³² Our recently published study, using whole transcriptome to address immune-metabolism features on day-8 in immunocompromised patients with sepsis, also found an impaired T cell-associated pathway, as well as a dysregulated respiratory electron transport chain and cellular respiration pathway among immunocompromised patients with sepsis.³³ Collectively, the aforementioned evidence shows that altered immunological and metabolic status may lead to impaired lymphocytic immunity in critical illness, and the findings in this study further provided clinical data supporting the role of low ALC.

Intriguingly, the analyses of the interaction effect found that the strength of the association between ALC and long-term mortality was higher in patients with younger age, fewer comorbidities, and lower severities including absence of septic shock, no need for mechanical ventilation, high level of haemoglobin, and low serum creatinine (Table 3). We postulated that critically ill patients with older age, CCI, and severities could have a tendency to have high short-term mortality; therefore, the long-term mortality impact of ALC might be mitigated by the mortality compete effect in patients with older age, CCI, and severities. Our previous study found that anaemia was associated with long-term mortality in critically ill patients, and more studies are needed to elucidate the biological mechanism underlying the interaction effect between low ALC and anaemia with respect to long-term mortality.³⁴ We think the aforementioned analysis further validated the association between week-one ALC and long-term mortality in critically ill patients, particularly those without anaemia.

Randomized Controlled Trials (RCTs) are considered the gold standard in clinical research but they are expensive, time-consuming, and may not be feasible for every research question. Real-world data, on the other hand, can provide insights into a wide range of patient populations and can reveal previously unknown associations in actual clinical practice.^35,36 However, real-world data are subject to various biases and confounders that may affect their validity, and therefore, careful analysis is required to interpret findings. We used PSM, IPTW, and CBPS to reduce the potential confounding effects in this study using RWD. PSM is widely used to balance confounding factors, but may inevitably involve discarding a certain portion of the samples without appropriate matching, particularly with unbalanced samples. For example, the number of matched patients of using ALC 1500 and 1000 counts/μL to divide the ALC were 1958 and 926, respectively (Supplemental Tables 1 and 3). Therefore, the targeted population might be distinct from the originally claimed population, and the selected patient population could raise the concern of generalizability of the results.²⁵ In contrast, PS weighting methods, including IPTW and CBPS, use the data from the original population. In brief, IPTW focuses on reweighting the data to create a pseudo-population with balance in covariate distributions, while CBPS adjusts for confounding by modelling the propensity score and using it as a covariate in a regression model.^22,23 The number of subjects in the present study was high, and the standardized mean differences of variables between the two groups using ALC 1500 as a cut-off value were apparently low (Supplemental Figure 1). Therefore, we identified a highly consistent strength of association between ALC and mortality among the three propensity score-based analyses.

There were several limitations in this study. First, the study enrolled patients who were admitted to medical ICUs at the study hospital, and more studies are warranted for the generalizability of the findings to other settings and specific populations, such as patients with stroke/sepsis. Second, given this study was not a randomized study, the adjusted association observed cannot be taken to indicate causal effect. Thus, the findings should be taken as exploratory and hypothesis-generating. Third, the potential existence of unmeasured confounders, such as solid/haematological transplant recipients.

Conclusions

In conclusion, our study identified a lower week-one ALC as an independent risk factor for long-term mortality in patients who were admitted to the medical ICUs. These findings indicate that ALC may serve as a potential early predictor of long-term outcomes in critically ill patients. Further research is needed to explore the mechanisms underlying the association between ALC and long-term mortality and to determine whether interventions targeting ALC could improve outcomes in critically ill patients.