Platelets and mortality in bloodstream infection: a multicenter cohort study

Abstract

Objective:

Our objective was to determine whether thrombocytopenia is independently associated with mortality in patients with bloodstream infections (BSI), and to compare the impact of platelets on mortality to that of white blood cells (WBCs) and neutrophils.

Methods:

Retrospective cohort study using two U.S. cohorts of patients with BSI: (1) Patients at a multi-hospital network in the metropolitan Houston, TX area between 7/1/2016–6/17/2023, and (2) patients in the publicly available MIMIC-IV database (2008–2022). We included patients who had platelets checked in the 48 hours before positive blood culture collection. We created multivariable logistic regression models to determine whether 30-day in-hospital mortality was impacted by degree of thrombocytopenia (severe [platelets <50 k/μL], moderate [50–99 k/μL], mild [100–149 k/μL], and none [≥150 k/μL]).

Results:

We included 21,105 patients in the Houston cohort and 2,710 in the MIMIC-IV cohort. 30-day mortality in the Houston cohort was 12.0% (2524/21105) and was significantly associated with platelet count. After controlling for confounders, the adjusted odds ratio (aOR) for 30-day mortality with severe thrombocytopenia was 4.66 (95% CI 3.91–5.55); aOR for moderate thrombocytopenia 2.61 (95% CI 2.25–3.02); and aOR for mild thrombocytopenia 1.55 (95% CI 1.37–1.76), all compared to normal platelet counts (≥150 k/μL). The adjusted odds of death with severe thrombocytopenia were greater than that with neutropenia, leukopenia, or leukocytosis. Results were similar in multiple sensitivity analyses and in the MIMIC-IV cohort.

Conclusions:

Thrombocytopenia was independently associated with mortality among patients with BSI. Platelet counts can provide clinicians a readily available way to risk stratify patients with BSI, and future research should examine mechanisms by which platelets are protective in BSI.

Introduction

Bloodstream infections (BSI) are common and associated with 15–30% mortality.[1, 2] White blood cells (WBCs), particularly neutrophils, are considered the principal immune cells for combatting pathogens in the bloodstream.[3] However, there are several immune mechanisms by which platelets defend against intravascular bacteria and fungi.[4, 5] Platelets are known to be associated with adverse outcomes during infection as part of the Sequential Organ Failure Assessment (SOFA) score,[6] however only limited clinical evidence has examined the importance of platelets independently in bloodstream infection.[5]

While hospitalized patients often have their platelets checked daily as part of a “complete blood count,”[7] platelet counts are used primarily to assess bleeding risk. Strengthening the known link between this commonly assayed laboratory parameter and mortality in patients with BSI could prompt clinicians to assess not only WBCs but also platelets to determine infection severity. The identification of a direct link between platelets and mortality could also spur additional research into the specific mechanisms by which platelets improve immune responses to BSI. We therefore conducted a U.S. multi-center retrospective cohort study to evaluate the association between platelets and mortality in BSI, as compared to WBCs and neutrophils.

Methods

This study was approved by the Houston Methodist Institutional Review Board (#00037862). We analyzed two cohorts of patients with BSI: (1) Patients ≥18 years old in the Houston Methodist Hospital (HMH) system (Houston, TX, USA) between 7/1/2016–6/17/2023; (2) patients from MIMIC-IV, a publicly available, deidentified dataset of patients in the emergency department or an intensive care unit at the Beth Israel Deaconess Medical Center (Boston, MA, USA) from 2008–2022.[8] We excluded patients with a blood culture positive for a commensal organism unless a second blood culture was positive with the same organism within three calendar days of the index blood culture.[9, 10] We did not analyze polymicrobial BSIs and limited our analysis to the first BSI per patient. Additional information on variables collected are listed in the Supplementary Appendix.

The primary exposure was platelet count closest to time of collection of positive BSI culture within 48 hours prior to blood culture collection, stratified a priori into severe (<50 k/μL), moderate (50–99 k/μL), mild (100–149 k/μL), and no thrombocytopenia (≥150 k/μL).[11] The primary outcome was 30-day in-hospital mortality. We compared characteristics between groups of interest using χ², t-test, Fisher’s exact test, Mann-Whitney test, or ANOVA as appropriate. To determine the relationship between platelet count and 30-day mortality, we constructed multivariable logistic regression models selected based on clinical relevance and controlling for likely confounders of the relationship between thrombocytopenia and mortality. Our primary model included platelet strata, age, gender, race, ECI, vasopressor administration, receipt of renal replacement therapy, oxygen modality, and white blood cell (WBC) count (not stratified). To assess the relative importance of platelets compared to WBCs and neutrophils specifically, we constructed similar models using WBC strata (<5.0 k/μL, 5.0–9.9 k/μL, 10.0–14.9 k/μL, 15.0–19.9 k/μL, and ≥20.0 k/μL) or neutrophil strata (<1000/μL, ≥1000/μL) respectively (strata assigned a priori), each adjusted for confounders as in the primary model, in addition to platelet count (not stratified). We additionally performed several sensitivity analyses (Supplementary Appendix methods and tables).

Data pre-processing and imputation was performed with Python 3.10.9 (Python Software Foundation, Wilmington, DE, USA), and analysis was performed with SAS University Edition (SAS, NC, USA).

Results

We included 21,105 BSI patients in the HMH cohort (Table S1A). Several demographic variables including younger age, male gender, and higher ECI were associated with severe thrombocytopenia, as were clinical variables including receipt of renal placement therapy or vasopressors and elevated lactic acid (Table S2A). Thirty-day mortality in the HMH cohort was 12.0% (2524/21105) overall and was significantly associated with platelet count: 29.2% (292/1001) for severe thrombocytopenia (platelets <50 k/μL, N=1001), 22.3% (374/1675) for moderate thrombocytopenia (platelets 50–99 k/μL, N=1675), 14.3% (444/3102) for mild thrombocytopenia (platelets 100–149 k/μL, N=3102), and 9.2% (1414/15327) for no thrombocytopenia (platelets ≥150 k/μL, N=15327) (p<0.0001).

In the primary model, the adjusted odds ratio (aOR) for 30-day mortality with severe thrombocytopenia was 4.66 (95% confidence interval [CI] 3.91–5.55); aOR for moderate thrombocytopenia 2.61 (95% CI 2.25–3.02); and aOR for mild thrombocytopenia 1.55 (95% CI 1.37–1.76), all compared to normal platelet counts (≥150 k/μL). Full model results are shown in the Table. In models including WBC and neutrophil strata while controlling for platelet count (and other covariates as in the main platelet model), the association between each of WBCs and neutrophils and mortality was less than that of severe thrombocytopenia (Figure). Results were similar in multiple sensitivity analyses (Tables S3–S5). We performed the same analyses on 2,710 patients from the MIMIC-IV cohort; the impact of platelet strata (and WBC/neutrophil strata) on 30-day mortality was similar to that in the HMH cohort (Figure). MIMIC-IV cohort characteristics and sensitivity analyses are shown in the Supplementary Appendix.

Table.

Adjusted odds ratios for the impact of platelets and other clinically relevant covariates on 30-day mortality in patients with bloodstream infection.

	Houston Methodist Cohort	MIMIC-IV Cohort
Variable	aOR (95% CI)
Platelet strata (k/μL)
< 50	4.66 (3.91–5.55)	4.28 (2.97–6.17)
50–99	2.61 (2.25–3.02)	2.27 (1.66–3.12)
100–149	1.55 (1.37–1.76)	1.39 (1.02–1.89)
≥ 150	Referent	Referent
Age (per year)	1.02 (1.02–1.02)	1.03 (1.02–1.03)
Female gender	0.99 (0.90–1.08)	1.03 (0.82–1.28)
Race
Black	1.09 (0.97–1.23)	0.96 (0.65–1.40)
Asian	1.13 (0.92–1.39)	1.14 (0.62–2.08)
Other/unk.	1.44 (1.15–1.80)	1.88 (1.43–2.47)
White	Referent	Referent
ECI (per point increase)	1.05 (1.04–1.06)	1.08 (1.04–1.12)
Vasopressors	2.90 (2.33–3.64)	2.89 (1.97–4.24)
Any RRT	1.06 (0.92–1.23)	1.31 (0.93–1.85)
Increasing oxygen modality	1.56 (1.52–1.59)	1.43 (1.34–1.53)
WBC (per k/μL increase)	1.01 (1.01–1.02)	1.03 (1.01–1.04)

Abbreviations: CI, confidence interval; ECI, Elixhauser comorbidity index; OR, odds ratio; RRT, renal replacement therapy; unk., unknown; WBC, white blood cell.

Figure.

Adjusted odds of 30-day mortality across different strata of platelet, white blood cell (WBC), and neutrophil counts among patients with bloodstream infection. Adjusted odds ratios (aOR) were determined in separate multivariable logistic regression models for each cell type. Odds ratios were adjusted for age, gender, race, Elixhauser comorbidity index, vasopressor use, renal replacement therapy, oxygen modality, and white blood cell count (for platelets) or platelet count (for white blood cells and neutrophils). We analyzed patients in the Houston Methodist Hospital (HMH) system (Houston, TX) and in the publicly available MIMIC-IV cohort[8] (Boston, MA) separately.

Discussion

In two large cohorts of patients with BSI, thrombocytopenia was independently associated with mortality, and odds of mortality increased with severity of thrombocytopenia. While residual confounding is possible, the impact of platelets on mortality was greater than that of WBCs and neutrophils after controlling for key confounders. Given the clinically accepted platelet strata used in this analysis,[11] these findings indicate that platelets can provide clinicians a quick, widely available method to assess risk of mortality in patients with BSI.

Platelets are known to impact outcomes in critical illness and factor into the SOFA score, which is used to determine whether a patient has sepsis and the associated risk of sepsis-related mortality.[6, 12] Despite this, the SOFA score is not widely used at the bedside, and our results suggest that platelets alone are strongly associated with mortality in BSI, regardless of additional SOFA score inputs. Although our multivariable models suggest that the protective effects of platelets were independent of severity of illness and underlying comorbidities, these findings will require further validation with more granular datasets. These results imply that clinicians should weigh platelet counts more heavily than WBC or neutrophil counts when assessing risk of adverse outcomes in these patients. Future research should prospectively confirm these results and investigate the specific mechanisms by which platelets contribute to immunity in BSI. Adjunctive approaches to BSI aside from pharmaceutical antimicrobials and supportive care could help reduce high BSI-associated mortality, and platelet administration is one potential avenue that should be explored.

Our study is subject to several limitations. We used retrospective, electronic-medical records derived databases which are not routinely subject to data audits and validation and therefore some data may be inaccurate or incomplete. Comorbidities, important potential confounders of the relationship between platelets and mortality, rely on ICD-10 codes which may not be reliable or reflect the status of the comorbidity at the time of infection. We were not able to extract several important infection-related variables (including source of infection, source control, treatment etc.) which may have impacted our observed associations. Lastly, our outcome was 30-day in-hospital mortality and therefore we may have missed patients died after discharge within 30 days.

In summary, platelet count was significant associated with mortality in two large cohorts of patients with BSI. Our results validate the known association between platelets and adverse outcomes and may provide clinicians a quick way to risk-stratify patients with BSI according to clinically-accepted platelet thresholds.