How infection affects the relationship between leukocyte count and mortality risk in intensive care Authors: Deirdre M. Horan, Tess Evans, Michael Bailey, Edward Litton, Manoj K. Saxena, David V. Pilcher Published: 2026 ABSTRACT PurposeLeukocyte count is widely available and included in Intensive Care Unit prognostic systems. We hypothesised that the relationship between leukocytosis and mortality risk might differ in critically ill patients admitted with infection, where leukocytosis may be an adaptive response.MethodsWe performed a registry-based study using the Australian and New Zealand Intensive Care Society Adult Patient Database between 2010 and 2023, across 212 Intensive Care Units. Using descriptive statistics and mixed-effects multivariable logistic regression, we evaluated the association between early peak leukocyte count and mortality, according to whether infection was the primary diagnosis.ResultsWe examined 2,016,578 patients, of whom 1,742,195 had non-infective illnesses (86.4%). In this group, there were 115,613 (6.6%) deaths, and a progressive increase in the adjusted odds of mortality as leukocyte counts increased above 8 × 109/L, peaking at 40 × 109/L (adjusted odds ratio 2.40 (99% CI 2.16–2.66)). This was equivalent to a rise in risk-adjusted mortality from 5.4% (99% CI 5.1–5.6) for leukocyte counts of 8–12 × 109/L, to 9.4% (99% CI 8.6–10.1) for leukocyte counts at 40 × 109/L. Amongst 274,383 patients with infection, there were 37,350 (13.6%) deaths. Leukocytosis had no effect on the odds of mortality. This group had a risk-adjusted mortality of 12.8% (99% CI 12.2–13.3) for leukocyte counts of 8–12 × 109/L and 13.0% (99% CI 11.8–14.1) for leukocyte counts at 40 × 109/L. Lower leukocyte counts, even within the normal range, were associated with mortality in both groups.ConclusionThe presence of infection determines the prognostic significance of leukocytosis in the critically ill.Supplementary InformationThe online version contains supplementary material available at 10.1007/s15010-026-02804-z. Introduction Leukocyte count is typically in the range of 4–11 × 109/L in healthy, non-pregnant adults [1, 2]. Leukocytosis and leukopenia occur when the leukocyte count lies above or below the expected range, respectively [2, 3]. Dramatic changes in the circulating leukocyte count can occur within hours because there is a large reserve of mature leukocytes in the bone marrow which can be rapidly mobilised, and because leukocytes can quickly migrate to inflamed tissues in response to local and systemic cytokines [4, 5]. These changes may occur in response to multiple insults including infection, trauma, surgery, acute inflammation, ischaemia, reperfusion, and malignancy [6]. Abnormal leukocyte count has been associated with increased risk of mortality in critically ill patients and as such is a component of multiple prognostic scoring systems used in intensive care including Acute Physiology and Chronic Health Evaluation (APACHE), Simplified Acute Physiology Score (SAPS) 3, and Physiologic and Operative Severity Score for the Study of Mortality and Morbidity (POSSUM) scoring systems [7–9]. Notably, the recently published Sequential Organ Failure Assessment score (SOFA-2) did not include measures of immune function (including leukocyte count) [10]. Whilst fever has been associated with poor outcomes in patients admitted to ICU with non-infective illnesses, higher temperatures have been associated with reduced mortality in critically ill patients with infection [11]. This suggests that fever may be an adaptive response aimed at defending against invading pathogens. Although previous studies have found a link between leukocytosis and mortality in intensive care, we hypothesised that this relationship might differ in critically ill patients admitted with infection, where similar to fever, leukocytosis may be an adaptive response. Methods We conducted a retrospective observational cohort study of all patients admitted to Intensive Care Units (ICUs) in Australia and New Zealand between January 2010 and December 2023, who were reported to the Australia and New Zealand Intensive Care Society (ANZICS) Adult Patient Database (APD). Hypothesis We hypothesised that the previously elucidated relationship between leukocytosis and mortality risk in critically ill patients might differ in those admitted with infection, where similar to fever, leukocytosis may be an adaptive response. Participants Patients aged over 16 years of age admitted during the study period to participating ICUs were eligible for inclusion. Those admitted for palliative care and for organ donation were excluded. Readmission episodes to ICU within the same hospitalisation and those without documentation of leukocyte count or survival status were also excluded. Data source The ANZICS APD is a binational registry dataset collected for benchmarking and quality assurance purposes by the ANZICS Centre for Outcome and Resource Evaluation. It includes data from 98% of ICUs in Australia and 68% of ICUs in New Zealand. Data are deidentified and include patient demographics, admission diagnosis, highest and lowest physiological and biochemical values from the first 24 h in ICU (including leukocyte count), chronic health status, invasive treatments, illness severity assessed using APACHE, Australian and New Zealand Risk of Death (ANZROD) and organ dysfunction (SOFA-2) scores, and mortality. We studied the relationship between the peak recorded leukocyte count within the first 24 h of ICU admission (henceforth abbreviated to ‘peak leukocyte count’) and in-hospital mortality. Ethics The research project was approved as a low-risk study, with waiver of individual patient informed consent, by the Human Research Ethics Committee of the Alfred Hospital (project 97/24) with use of data approved by the AMZICS CORE Management Committee. Definitions Patients were divided into two groups: those admitted with non-infective illness (non-infective group) and those admitted with infection (infection group) (Supplemental Table 1). The diagnosis was allocated locally by clinicians and represented the primary reason for admission to ICU. The only additional diagnostic codes available were those introduced to indicate where SARS-CoV-2 infection was a contributing factor to an admission (e.g. asthma with SARS-CoV-2). These were included in the infection group [12]. All other diagnoses (e.g. ‘burns’, ’drug overdose’) were allocated to the non-infective group. No diagnostic information was available after ICU admission. Patient outcomes reported for descriptive purposes included in-ICU mortality as well as length of stay in ICU and in hospital. Primary outcome The primary outcome studied was in-hospital mortality. Statistical methods Data were presented as number/proportion (percentage), mean (± standard deviation) or median (interquartile range) depending on type and distribution of data. Chi square, t-tests, and Wilcoxon rank sum tests were used to compare groups depending on type and distribution of data. Observed mortality with 95% confidence intervals (95% CI) was reported across different levels of peak leukocyte count using increments of 4 × 109/L for tables and increments of 1 × 109/L from 0 to 40 for figures which were then displayed with locally weighted scatterplot smoothing (LOWESS) curves. Leukocyte counts greater than 45 × 109/L (0.5% of total dataset) were truncated, with values between this and 40 × 109/L collapsed into the uppermost category. To adjust for potential confounders, mixed-effects hierarchical logistic regression was used (after assessing co-linearity) to investigate the association between leukocyte count and mortality, with patients clustered by site, and site treated as a random effect. Covariates included sex, acute illness severity in the first 24 h of ICU admission, chronic comorbidities, treatment limitations present on admission to ICU, and diagnosis. Acute illness severity was assessed using the acute physiology component of the APACHE III/IV score with points for leukocyte count removed. Additional covariates included hospital type and year of admission to ICU to account for changes over time. Risk-adjusted mortality estimates representing the model-adjusted probabilities for the ‘typical patient in the typical ICU’ within each leukocyte category were derived using the post-estimation margins command. For reporting in tables, models were run with the peak leukocyte count considered as a categorical ordinal variable in increments of 4 × 109/L from 0 to 40 × 109/L, with 8 × 109/L considered as the reference value. To determine whether the relationship between leukocyte count and mortality differed according to infection status, an interaction term between leukocyte count and infection was included in the multivariable model. For figures, results derived from logistic regression models were displayed as odds ratios per unit increase in peak leukocyte count (109/L) from 0 to 40. To test the robustness of our findings, models were also run using restricted cubic splines, alternative risk adjustment strategies initially with the SOFA-2 score and then ANZROD, using the lowest leukocyte count in the first 24 h and then the average of lowest and peak leukocyte counts. Subgroup analyses were undertaken first after excluding pregnant patients, second after excluding patients with an ICU admission diagnosis related to a primary haematological condition or history of leukaemia or lymphoma, third after exclusion of patients immunosuppressed by disease or therapy. Further subgroups examined patients with infection who met or did not meet Sepsis-3 criteria [11], and only those with immunosuppression. Given the large sample size, to enhance the robustness of inference, a two-sided p-value of 0.01 was used to indicate statistical significance. Data were analysed using Stata version 19.0 College Station, Texas. Results Between January 2010 and December 2023, there were 2,265,050 admissions reported to the ANZICS APD. After exclusions, the final study dataset comprised 2,016,578 patients from 212 ICUs in 211 hospitals (Fig. 1). Comparison of those excluded due to missing information about mortality or leukocyte count is provided in Supplemental Table 2. Patients excluded because of missing leukocyte count or outcome data (4.6% of all admissions) differed modestly from those included, being younger, less often admitted after surgery, and more frequently admitted from emergency departments or wards. Fig. 1Flow chart of inclusions and exclusions Flow chart of inclusions and exclusions Patient characteristics Of those included in our study, 1,742,195 (86.4%) were admitted with non-infective illnesses, whilst 274,383 (13.6%) were admitted with infection. The baseline characteristics of both groups are summarised in Table 1. Table 1Demographics of patients in non-infective and infection groupsPatients’ demographicsNon-infective group (n = 1,742,195)Infection group (n = 274,383)Peak leukocyte count in first 24 h in ICU [median n × 109/L (IQR)]12.0 (9.1–15.8)13.1 (8.7–19.0)Age (years) [mean (SD)]62 (17.7)63.4 (17.1)Men [number of patients (%)]995,805 (57.2)153,186 (55.8)Pregnancy status [number of patients (%)] Pregnant3684 (0.2)1398 (0.5) Post-partum17,296 (1.0)1235 (0.5) Not pregnant1,721,215 (98.8)271,750 (99.0)Source of admission to ICU [number of patients (%)] Operating theatre1,055,577 (60.6)37,931 (13.8) Emergency Department402,180 (23.1)129,334 (47.1) Hospital ward184,975 (10.6)75,179 (27.4) Other hospital94,281 (5.4)31,105 (11.3) Other/unknown source5182 (0.3)834 (0.3)Hospital classification [number of patients (%)] Public rural/regional208,373 (12.0)57,290 (20.9) Public metropolitan250,775 (14.4)70,572 (25.7) Public tertiary715,861 (41.1)114,300 (41.7) Private567,186 (32.6)32,221 (11.7)Hours in hospital prior to ICU admission [number of hours (median, IQR)]10.1 (4.9–28.3)8.3 (3.5–25.9)Therapies during ICU admission [number of patients (%)]* Invasive ventilation640,505 (36.8)72,992 (26.6) Renal replacement therapy26,857 (3.6)10,721 (8.2) Vasoactive medication266,006 (35.4)73,080 (54.9) Extracorporeal membrane oxygenation2224 (0.3)590 (0.5)Illness severity and organ failure scores SOFA-2 score [median (IQR)]3 (1–5)4 (2–7) APACHE III/IV score [mean (SD)]50.4 (23.5)63.1 (26.1) ANZROD [mean (SD)]6.9 (15.4)13.9 (19.1) ANZROD [median (IQR)]1.2 (0.4–4.7)5.8 (2.0–16.9)Admission category [number (%)] Emergency surgical admission279,275 (16.0)24,493 (8.9) Planned, elective surgery794,269 (45.6)13,220 (4.8) Medical admission668,651 (38.4)236,670 (86.3)Comorbidities [number (%)] Diabetes204,074 (11.7)44,177 (16.1) Immunosuppressed (by disease or therapy)86,083 (4.9)36,595 (13.3) Metastatic cancer66,196 (3.8)12,487 (4.6) Chronic cardiovascular149,505 (8.6)24,286 (8.9) Chronic respiratory118,664 (6.8)26,561 (9.7) Chronic dialysis dependence53,306 (3.1)14,944 (5.4) Chronic cirrhosis28,087 (1.6)6674 (2.4) Leukaemia, lymphoma or other haematological conditions27,825 (1.6)14,417 (5.3) History of lymphoma11,039 (0.6)6221 (2.3) History of leukaemia10,967 (0.6)8364 (3.0)Primary cause of ICU admission was haematological condition9221 (0.5)0 (0.0)APACHE diagnostic group [number (%)] Cardiovascular256,099 (14.7)3155 (1.1) Cardiac surgery235,005 (13.5)0 (0.0) Gastrointestinal307,685 (17.7)27,848 (10.1) Neurological243,473 (14.0)7320 (2.7) Other386,192 (22.2)14,089 (5.1) Respiratory218,930 (12.6)75,379 (27.5) Sepsis (as primary diagnosis)0 (0.0)146,592 (53.4) Trauma94,811 (5.4)0 (0.0%)Met criteria for Sepsis-3 definition at admission0 (0.0)247,618 (90.2)Physiological variables in first 24 h in Intensive Care Unit [mean (SD)] Highest temperature in degrees Celsius37.1 (0.7)37.5 (1.0) Lowest temperature in degrees Celsius35.9 (0.8)36.1 (0.8) Highest heart rate in beats per minute96.4 (20.8)106.6 (23.7) Lowest heart rate in beats per minute67.8 (14.9)73.2 (16.7) Highest mean arterial blood pressure in mmHg97.9 (16.2)95.2 (16.6) Lowest mean arterial pressure in mmHg68.3 (12.0)64.9 (11.4) Highest respiratory rate in breaths per minute22.9 (6.2)27.3 (7.8) Lowest respiratory rate in breaths per minute12.2 (3.4)14.4 (4.2)Outcomes In-hospital mortality [number (%)]115,613 (6.6)37,350 (13.6) In-ICU mortality [number (%)]72,737 (4.2)23,332 (8.5) Duration of stay in ICU in days [median (IQR)]1.7 (0.9–3.0)2.7 (1.4–5.1) Duration of stay in hospital in days [median (IQR)]7.9 (4.3–14.2)9.7 (5.3–18.7)ICU Intensive Care Unit, IQR interquartile range, SD standard deviation, SOFA Sequential Organ Failure Assessment, APACHE Acute Physiology and Chronic Health Evaluation, ANZROD Australia and New Zealand Risk of Death*Data available regarding therapies during ICU admission (other than invasive ventilation) for patients admitted from 2018 onlyNote all p values for Table 1 are < 0.001 Table 1Demographics of patients in non-infective and infection groupsPatients’ demographicsNon-infective group (n = 1,742,195)Infection group (n = 274,383)Peak leukocyte count in first 24 h in ICU [median n × 109/L (IQR)]12.0 (9.1–15.8)13.1 (8.7–19.0)Age (years) [mean (SD)]62 (17.7)63.4 (17.1)Men [number of patients (%)]995,805 (57.2)153,186 (55.8)Pregnancy status [number of patients (%)] Pregnant3684 (0.2)1398 (0.5) Post-partum17,296 (1.0)1235 (0.5) Not pregnant1,721,215 (98.8)271,750 (99.0)Source of admission to ICU [number of patients (%)] Operating theatre1,055,577 (60.6)37,931 (13.8) Emergency Department402,180 (23.1)129,334 (47.1) Hospital ward184,975 (10.6)75,179 (27.4) Other hospital94,281 (5.4)31,105 (11.3) Other/unknown source5182 (0.3)834 (0.3)Hospital classification [number of patients (%)] Public rural/regional208,373 (12.0)57,290 (20.9) Public metropolitan250,775 (14.4)70,572 (25.7) Public tertiary715,861 (41.1)114,300 (41.7) Private567,186 (32.6)32,221 (11.7)Hours in hospital prior to ICU admission [number of hours (median, IQR)]10.1 (4.9–28.3)8.3 (3.5–25.9)Therapies during ICU admission [number of patients (%)]* Invasive ventilation640,505 (36.8)72,992 (26.6) Renal replacement therapy26,857 (3.6)10,721 (8.2) Vasoactive medication266,006 (35.4)73,080 (54.9) Extracorporeal membrane oxygenation2224 (0.3)590 (0.5)Illness severity and organ failure scores SOFA-2 score [median (IQR)]3 (1–5)4 (2–7) APACHE III/IV score [mean (SD)]50.4 (23.5)63.1 (26.1) ANZROD [mean (SD)]6.9 (15.4)13.9 (19.1) ANZROD [median (IQR)]1.2 (0.4–4.7)5.8 (2.0–16.9)Admission category [number (%)] Emergency surgical admission279,275 (16.0)24,493 (8.9) Planned, elective surgery794,269 (45.6)13,220 (4.8) Medical admission668,651 (38.4)236,670 (86.3)Comorbidities [number (%)] Diabetes204,074 (11.7)44,177 (16.1) Immunosuppressed (by disease or therapy)86,083 (4.9)36,595 (13.3) Metastatic cancer66,196 (3.8)12,487 (4.6) Chronic cardiovascular149,505 (8.6)24,286 (8.9) Chronic respiratory118,664 (6.8)26,561 (9.7) Chronic dialysis dependence53,306 (3.1)14,944 (5.4) Chronic cirrhosis28,087 (1.6)6674 (2.4) Leukaemia, lymphoma or other haematological conditions27,825 (1.6)14,417 (5.3) History of lymphoma11,039 (0.6)6221 (2.3) History of leukaemia10,967 (0.6)8364 (3.0)Primary cause of ICU admission was haematological condition9221 (0.5)0 (0.0)APACHE diagnostic group [number (%)] Cardiovascular256,099 (14.7)3155 (1.1) Cardiac surgery235,005 (13.5)0 (0.0) Gastrointestinal307,685 (17.7)27,848 (10.1) Neurological243,473 (14.0)7320 (2.7) Other386,192 (22.2)14,089 (5.1) Respiratory218,930 (12.6)75,379 (27.5) Sepsis (as primary diagnosis)0 (0.0)146,592 (53.4) Trauma94,811 (5.4)0 (0.0%)Met criteria for Sepsis-3 definition at admission0 (0.0)247,618 (90.2)Physiological variables in first 24 h in Intensive Care Unit [mean (SD)] Highest temperature in degrees Celsius37.1 (0.7)37.5 (1.0) Lowest temperature in degrees Celsius35.9 (0.8)36.1 (0.8) Highest heart rate in beats per minute96.4 (20.8)106.6 (23.7) Lowest heart rate in beats per minute67.8 (14.9)73.2 (16.7) Highest mean arterial blood pressure in mmHg97.9 (16.2)95.2 (16.6) Lowest mean arterial pressure in mmHg68.3 (12.0)64.9 (11.4) Highest respiratory rate in breaths per minute22.9 (6.2)27.3 (7.8) Lowest respiratory rate in breaths per minute12.2 (3.4)14.4 (4.2)Outcomes In-hospital mortality [number (%)]115,613 (6.6)37,350 (13.6) In-ICU mortality [number (%)]72,737 (4.2)23,332 (8.5) Duration of stay in ICU in days [median (IQR)]1.7 (0.9–3.0)2.7 (1.4–5.1) Duration of stay in hospital in days [median (IQR)]7.9 (4.3–14.2)9.7 (5.3–18.7)ICU Intensive Care Unit, IQR interquartile range, SD standard deviation, SOFA Sequential Organ Failure Assessment, APACHE Acute Physiology and Chronic Health Evaluation, ANZROD Australia and New Zealand Risk of Death*Data available regarding therapies during ICU admission (other than invasive ventilation) for patients admitted from 2018 onlyNote all p values for Table 1 are < 0.001 Demographics of patients in non-infective and infection groups ICU Intensive Care Unit, IQR interquartile range, SD standard deviation, SOFA Sequential Organ Failure Assessment, APACHE Acute Physiology and Chronic Health Evaluation, ANZROD Australia and New Zealand Risk of Death *Data available regarding therapies during ICU admission (other than invasive ventilation) for patients admitted from 2018 only Note all p values for Table 1 are < 0.001 The mean peak leukocyte count was lower in the non-infective group than in the infection group. Physiological variables such as temperature, heart rate, mean arterial blood pressure and respiratory rate were less severely deranged in the non-infective group compared with the infection group. Severity of illness as assessed by ANZROD [13, 14], SOFA-2 [15], and APACHE III/IV [7] scores was lower in the non-infective group compared with the infection group. All comorbidities were less common in those admitted with non-infective illnesses compared with those admitted with infection. Those in the non-infective group had higher rates of invasive ventilation but lower rates of renal replacement therapy and vasoactive medication compared with the infection group. Mortality in ICU was lower in the non-infective group than in the infection group. Length of stay in ICU, and in hospital, were shorter in the non-infective group than in the infection group. Primary outcome There were 115,613 (6.6%) deaths in the non-infective group and 37,350 (13.6%) deaths in the infection group (Table 1). Comparison of survivors to non-survivors in each group is proved in Supplemental Tables 3 and 4. In-hospital mortality was highest in both groups when patients were severely leukopenic. The lowest mortality was observed in both groups with peak leukocyte counts of approximately 8 × 109/L (Table 2; Fig. 2). Table 2Observed in-hospital mortality and adjusted mortality risk across different early peak leukocyte countsPeak leukocyte count (109/L)Non-infective groupInfection groupNumberObserved mortality% (99% CI)Adjusted odds ratioa(99% CI)Risk-adjusted mortalityb% (99% CI)p valueNumberObserved mortality% (99% CI)Adjusted odds ratioa (99% CI)Risk-adjusted mortalityb% (99% CI)p value< 4684528.4 (27.0–29.8)3.53 (3.21–3.88)11.9 (11.1–12.7)< 0.001974231.1 (29.9–32.3)2.02 (1.85–2.20)19.8 (18.7–20.9)< 0.0014 to < 890,3427.6 (7.4–7.9)1.26 (1.21–1.32)6.3 (5.9–6.6)< 0.00124,45916.2 (15.6–16.8)1.32 (1.24–1.41)15.3 (14.6–16.0)< 0.0018 to < 12459,3604.4 (4.3–4.5)Reference group5.4 (5.1–5.6)55,20111.0 (10.6–11.3)Reference group12.8 (12.2–13.3)12 to < 16570,1974.9 (4.8–4.9)1.16 (1.13–1.20)5.9 (5.7–6.2)< 0.00162,53411.3 (11.0–11.6)1.01 (0.95–1.06)12.8 (12.3–13.4)0.7416 to < 20340,1526.8 (6.7–6.9)1.42 (1.37–1.46)6.8 (6.4–7.1)< 0.00147,13512.4 (12.0–12.8)1.01 (0.96–1.08)12.9 (12.3–13.5)0.5120 to < 24153,02110.1 (9.9–10.3)1.67 (1.62–1.73)7.5 (7.2–7.9)< 0.00130,46413.7 (13.2–14.2)1.04 (0.97–1.10)13.1 (12.4–13.7)0.1624 to < 2865,84514.0 (13.6–14.3)1.89 (1.81–1.97)8.1 (7.7–8.5)< 0.00118,67715.0 (14.3–15.6)1.04 (0.96–1.12)13.1 (12.4–13.8)0.2128 to < 3229,76017.1 (16.6–17.7)1.97 (1.86–2.09)8.3 (7.9–8.8)< 0.00111,29915.6 (14.7–16.5)0.97 (0.89–1.06)12.5 (11.7–13.3)0.3532 to < 3614,21620.2 (19.3–21.0)2.11 (1.96–2.28)8.7 (8.2–9.2)< 0.001682916.2 (15.1–17.4)0.93 (0.84–1.04)12.2 (11.3–13.1)0.1036 to < 40698822.7 (21.4–24.0)2.19 (1.99–2.42)8.9 (8.3–9.6)< 0.001416518.5 (16.9–20.0)1.03 (0.90–1.17)13.0 (11.8–14.1)0.62≥ 40589025.9 (24.3–27.4)2.40 (2.16–2.66)9.4 (8.6–10.1)< 0.001423120.3 (18.7–22.0)1.03 (0.91–1.17)13.0 (11.8–14.1)0.50aAdjusted for confounders including age, sex, acute illness severity (APACHE III/IV acute physiology score with age and leukocyte components removed), chronic comorbidities, diagnosis, source of admission to ICU and hospital, treatment limitation at ICU admission and hospital type (mixed effects logistic regression with site as a random intercept). Complete multivariable models are provided in Supplemental Tables 5 and 6bRisk-adjusted mortality estimates represent the marginal model-derived probabilities for the ‘typical patient in the typical ICU’ and show the expected change in mortality with varying leukocyte counts independently of change in baseline illness severity Table 2Observed in-hospital mortality and adjusted mortality risk across different early peak leukocyte countsPeak leukocyte count (109/L)Non-infective groupInfection groupNumberObserved mortality% (99% CI)Adjusted odds ratioa(99% CI)Risk-adjusted mortalityb% (99% CI)p valueNumberObserved mortality% (99% CI)Adjusted odds ratioa (99% CI)Risk-adjusted mortalityb% (99% CI)p value< 4684528.4 (27.0–29.8)3.53 (3.21–3.88)11.9 (11.1–12.7)< 0.001974231.1 (29.9–32.3)2.02 (1.85–2.20)19.8 (18.7–20.9)< 0.0014 to < 890,3427.6 (7.4–7.9)1.26 (1.21–1.32)6.3 (5.9–6.6)< 0.00124,45916.2 (15.6–16.8)1.32 (1.24–1.41)15.3 (14.6–16.0)< 0.0018 to < 12459,3604.4 (4.3–4.5)Reference group5.4 (5.1–5.6)55,20111.0 (10.6–11.3)Reference group12.8 (12.2–13.3)12 to < 16570,1974.9 (4.8–4.9)1.16 (1.13–1.20)5.9 (5.7–6.2)< 0.00162,53411.3 (11.0–11.6)1.01 (0.95–1.06)12.8 (12.3–13.4)0.7416 to < 20340,1526.8 (6.7–6.9)1.42 (1.37–1.46)6.8 (6.4–7.1)< 0.00147,13512.4 (12.0–12.8)1.01 (0.96–1.08)12.9 (12.3–13.5)0.5120 to < 24153,02110.1 (9.9–10.3)1.67 (1.62–1.73)7.5 (7.2–7.9)< 0.00130,46413.7 (13.2–14.2)1.04 (0.97–1.10)13.1 (12.4–13.7)0.1624 to < 2865,84514.0 (13.6–14.3)1.89 (1.81–1.97)8.1 (7.7–8.5)< 0.00118,67715.0 (14.3–15.6)1.04 (0.96–1.12)13.1 (12.4–13.8)0.2128 to < 3229,76017.1 (16.6–17.7)1.97 (1.86–2.09)8.3 (7.9–8.8)< 0.00111,29915.6 (14.7–16.5)0.97 (0.89–1.06)12.5 (11.7–13.3)0.3532 to < 3614,21620.2 (19.3–21.0)2.11 (1.96–2.28)8.7 (8.2–9.2)< 0.001682916.2 (15.1–17.4)0.93 (0.84–1.04)12.2 (11.3–13.1)0.1036 to < 40698822.7 (21.4–24.0)2.19 (1.99–2.42)8.9 (8.3–9.6)< 0.001416518.5 (16.9–20.0)1.03 (0.90–1.17)13.0 (11.8–14.1)0.62≥ 40589025.9 (24.3–27.4)2.40 (2.16–2.66)9.4 (8.6–10.1)< 0.001423120.3 (18.7–22.0)1.03 (0.91–1.17)13.0 (11.8–14.1)0.50aAdjusted for confounders including age, sex, acute illness severity (APACHE III/IV acute physiology score with age and leukocyte components removed), chronic comorbidities, diagnosis, source of admission to ICU and hospital, treatment limitation at ICU admission and hospital type (mixed effects logistic regression with site as a random intercept). Complete multivariable models are provided in Supplemental Tables 5 and 6bRisk-adjusted mortality estimates represent the marginal model-derived probabilities for the ‘typical patient in the typical ICU’ and show the expected change in mortality with varying leukocyte counts independently of change in baseline illness severity Observed in-hospital mortality and adjusted mortality risk across different early peak leukocyte counts aAdjusted for confounders including age, sex, acute illness severity (APACHE III/IV acute physiology score with age and leukocyte components removed), chronic comorbidities, diagnosis, source of admission to ICU and hospital, treatment limitation at ICU admission and hospital type (mixed effects logistic regression with site as a random intercept). Complete multivariable models are provided in Supplemental Tables 5 and 6 bRisk-adjusted mortality estimates represent the marginal model-derived probabilities for the ‘typical patient in the typical ICU’ and show the expected change in mortality with varying leukocyte counts independently of change in baseline illness severity Fig. 2Observed in-hospital mortality versus peak total leukocyte count in the first 24 h in ICU, in the non-infective (blue) and infection (green) groups Observed in-hospital mortality versus peak total leukocyte count in the first 24 h in ICU, in the non-infective (blue) and infection (green) groups When peak leukocyte count exceeded 24 × 109/L observed mortality was higher in those with non-infective illnesses than in those with infection. At all peak leukocyte counts less than 24 × 109/L, in-hospital mortality was higher in the infection group. In the non-infective group, there was an almost linear increase in in-hospital mortality as the peak leukocyte count increased to 40 × 109/L. In the infection group, an increase in peak leukocyte count resulted in a less pronounced increase in observed mortality. A significant interaction was observed between peak leukocyte count and infection in determining mortality risk (p < 0.001). A mixed-effects model demonstrated a near-linear increase in odds of mortality with progressive leukocytosis from 8 to 40 × 109/L among patients admitted with non-infective illnesses. This was equivalent to a rise in risk-adjusted mortality in the non-infective group from 5.4% (99% CI 5.1–5.6) for leukocyte counts of 8–12 × 109/L, to 9.4% (99% CI 8.6–10.1) for leukocyte counts at 40 × 109/L. For those admitted with infection, there was no increase in mortality risk with a rising leukocyte count. The infection group had a risk-adjusted mortality of 12.8% (99% CI 12.2–13.3) for leukocyte counts of 8–12 × 109/L and 13.0% (99% CI 11.8–14.1) for leukocyte counts at 40 × 109/L (Table 2; Fig. 3). Fig. 3Adjusted odds ratio for in-hospital mortality versus peak leukocyte count in the first 24 h in ICU for patients in the non-infective (blue) and infection (green) groups. Adjusted for confounders, including age, sex, acute illness severity (APACHE III/IV acute physiology score with age and leukocyte components removed), chronic comorbidities, diagnosis, source of admission to ICU and hospital, treatment limitation at ICU admission and hospital type (mixed effects logistic regression) Adjusted odds ratio for in-hospital mortality versus peak leukocyte count in the first 24 h in ICU for patients in the non-infective (blue) and infection (green) groups. Adjusted for confounders, including age, sex, acute illness severity (APACHE III/IV acute physiology score with age and leukocyte components removed), chronic comorbidities, diagnosis, source of admission to ICU and hospital, treatment limitation at ICU admission and hospital type (mixed effects logistic regression) There were 6845 (0.4%) patients in the non-infective group and 9742 (3.5%) patients in the infection group who were leukopenic. In both groups, leukopenia correlated with an increase in observed and adjusted mortality risk. This relationship between decreasing leukocyte count and mortality risk was seen even within the normal range as peak leukocyte count fell below 8 × 109/L (Table 2; Fig. 3). Similar results were obtained when modelled using restricted cubic splines (Supplemental Fig. 1) and when risk adjustment models used the SOFA-2 score or ANZROD (Supplemental Figs. 2 and 3), and amongst 247,618 patients who met Sepsis-3 criteria (90.2% of the infection group) (Supplemental Table 7). Results were consistent after exclusion of 23,736 pregnant women (1.2% of the non-infective group and 0.96% of the infection group) (Supplemental Fig. 4), after exclusion of 44,766 patients with haematological conditions including leukaemia and lymphoma (1.7% of the non-infective group and 5.5% of the infection group) (Supplemental Fig. 5) and after exclusion of 102,319 patients listed as immunosuppressed by disease or therapy (4.3% of the non-infective group and 9.8% of the infection group) (Supplemental Fig. 6). Similar results were obtained when only immunosuppressed patients were examined (Supplemental Fig. 7), when the peak leukocyte count was replaced with lowest leukocyte count in the first 24 h (Supplemental Fig. 8), or was replaced with the average of these two values (Supplemental Fig. 9). Amongst the 26,765 patients with an infective diagnosis who did not meet Sepsis-3 criteria, there was a trend to increased mortality with progressive leukocytosis (Supplemental Table 7). Discussion We evaluated over two million patients admitted to ICUs in Australia and New Zealand over a fourteen-year period. Having accounted for illness severity and other confounders, we observed that the prognostic significance of early peak leukocyte count was different depending on whether infection was the primary reason for ICU admission. The adjusted odds of mortality increased with progressive leukocytosis in the non-infective group. This relationship was not present in those admitted with infection. We observed an increased risk of death in both groups when peak leukocyte count was less than 8 × 109/L. In normal subjects with a functioning bone marrow, the stress response can result in a substantial increase in the number of leukocytes in the peripheral circulation within minutes to hours [16]. Previous studies have linked leukocytosis to mortality in multiple conditions including cardiovascular, cerebrovascular and respiratory diseases, COVID-19, head trauma, and malignancy [17–19]. We also found a strong correlation between leukocytosis and risk of mortality in patients with non-infective illnesses, where the degree of leukocytosis may relate to the extent of injury or inflammation. Uniquely, our study demonstrates a different association in the setting of infection: where escalating leukocytosis was not associated with mortality. This is consistent with differing host responses in the presence of infection, although causal or mechanistic inferences cannot definitively be made from these observational data [20]. A leukocyte count below the normal reference range has also been associated with increased risk of death in critically ill patients [21]. Leukopenia is defined as a leukocyte count of less than 4 × 109/L [3]. Our study found a progressive mortality risk as peak leukocyte count fell below a level of 8 × 109/L (i.e. within the normal range) in all groups examined. Reasons include the failure of critically ill patients to mount an adequate immune response to combat the insult (immune exhaustion), bone marrow suppression or depletion of circulating leukocyte numbers due to extravasation. Our study indicates that using the leukocyte count to prognosticate for critically ill patients is more nuanced than previously understood. If a patient is admitted to ICU with a non-infective illness, escalating leukocytosis is related almost linearly with adjusted odds of death, and therefore a useful additional indicator for that patient. However, if the underlying reason for ICU admission is infection, progressive leukocytosis lacks prognostic significance. In addition, ‘low normal’ leukocyte counts in all patients may also indicate an increased mortality risk. This suggests that the leukocyte count should be interpreted in the context of the underlying diagnosis. Our study has numerous strengths. It included over 2 million critically unwell patients. The large study population adds to generalisability of our results. Data were collected for audit and benchmarking purposes independent of our study and were therefore unlikely to be affected by bias. Our multivariable analysis included validated illness severity scoring systems and recognised, accessible biomarkers of illness severity, reducing the risk of bias related to confounding factors. Additionally, there was a statistically robust and distinct association in each group between peak leukocyte count and outcome which was consistent in all sensitivity analyses. Our study included pregnant and postpartum patients. Because the normal reference range for leukocytes would not apply in this cohort, we repeated our analysis having excluded these patients and found no change in the relationships between peak leukocyte count and mortality in either group [22] (Supplemental Fig. 4). We undertook a further analysis excluding patients with known haematological diseases and confirmed that inclusion of this cohort in our original analysis had not skewed our results (Supplemental Fig. 5). Similar findings were present in both non-immunosuppressed and immunosuppressed patients (Supplemental Figs. 6 and 7). The diagnosis in each case was made by a clinician. The codes used to allocate to the infection group have previously been validated and this diagnosis represents the primary cause of the patient’s admission to the ICU [23]. Our study has limitations. Despite the large size of the study cohort, 4.6% of patients were excluded because of missing data. We used the ANZICS modification of APACHE IV admission diagnostic codes to assign the non-infective or infection groups, based on the primary reason for admission to ICU. Because there is only one diagnostic code available for each patient, this classification does not account for patients who had overlap of non-infective and infective illnesses at admission, or those who were assigned to the non-infective group who developed infection later in their admission. In addition, the diagnostic system does not differentiate bacterial and viral infections for most patients. Our study primarily used the peak recorded leukocyte count in the first 24 h of ICU admission. Sensitivity analyses using lowest and mean leukocyte counts, showed similar findings. However, lack of knowledge about the relative timing of the two available values or any additional values during or after the first 24 h, limit interpretation about how response to therapy or recovery trajectory might influence the relationship with mortality. Erroneous measurement of leukocyte count may occur, for example due to platelet clumping, incomplete erythrocyte lysis or cryoglobulinaemia [24]. Inaccurate reporting of leukocyte count may have occurred in either group. We cannot comment on mortality relationships above leukocyte counts of 40 × 109/L. Leukocyte numbers may not correlate with phenotype or function [25]. Finally, we could not estimate the effect of antimicrobial therapies, drugs such as steroids or granulocyte-macrophage colony-stimulating factor or the impact of more detailed information about haematological or other diseases, which may affect the leukocyte count [26–29]. Our study identifies an important point of difference for the leukocyte count in prognostication based on the presence of an underlying infective diagnosis. Further studies should elucidate the reasons for this, whether this relationship persists after the first day in ICU, how it is modified by treatment, microbiological aetiology, leukocyte differential, subpopulation ratios and markers of leukocyte function. If the leukocyte count is to be incorporated in future illness severity scores, stratification of patients on the basis of infective aetiology should be considered. Conclusions The relationship between leukocytosis and mortality risk in critically ill patients depends on whether the primary reason for admission to ICU is infection. In those with a non-infective illness there was a direct correlation between degree of leukocytosis and mortality risk. This was not seen in those admitted with infection, even at extremes of leukocytosis. There was a strong correlation between leukopenia and odds of mortality in both groups. This was noted even at leukocyte counts at the lower end of the normal range (< 8 × 109/L). Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1