Optimizing Clinical Indicators in Hematologic Malignancies and Sepsis Using ddPCR: A Retrospective Study

Abstract

Introduction

Early antibacterial treatment is critical for patients with hematologic malignancies (HMs) and sepsis. Droplet digital polymerase chain reaction (ddPCR) can rapidly detect pathogens and antimicrobial resistance (AMR) genes, but its clinical value in HMs is unknown. This study aimed to systematically evaluate the role of ddPCR in diagnosis, clinical outcomes, and antimicrobial stewardship.

Methods

From January 2023 to March 2025, 400 patients with hematologic malignancies (HMs) and sepsis were enrolled in the study. Of these, 150 received both ddPCR and blood culture (BC), while 250 underwent BC alone. Using propensity score matching (PSM), as well as subgroup and sensitivity analyses, we evaluated ten indicators, including 28-day mortality, treatment efficacy, and antibiotic use density (AUD).

Results

ddPCR showed a 49.33% positive rate (vs. BC’s 17.50%, P < 0.01) with a 4.06-h diagnostic turnaround (vs. 72.47 h for BC, P < 0.01), achieving 70.37% sensitivity and 55.28% specificity. The ddPCR group had lower 28-day mortality (HR = 0.55, P = 0.01), higher clinical response rates, and greater inflammatory marker decline. Antimicrobial optimization via ddPCR improved efficacy to 85.11%, with reduced AUD (OR = − 28.93, P < 0.01), the quantity and proportion of combined antimicrobial usage. However, a non-significant difference was observed in the proportion of antibacterial treatment costs (P = 0.14). PSM and sensitivity analysis results were consistent, indicating data robustness.

Conclusions

ddPCR outperforms BC in diagnostic efficiency for patients with HMs and sepsis, accelerating pathogen and AMR genes identification, optimizing antibacterial therapy and management, improving clinical effectiveness, and reducing 28-day all-cause mortality. The findings support the application of ddPCR in immunosuppressed populations.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at 10.1007/s40121-025-01207-1.

Key Summary Points

Why carry out this study?

In patients with hematologic malignancies (HMs) and sepsis, immunosuppression, low blood culture (BC) positivity, and prolonged BC turnaround time collectively contribute to high mortality. Early pathogen identification and precise antimicrobial therapy are critical for improving clinical outcomes.

Can digital droplet PCR (ddPCR) for rapid pathogen and antimicrobial resistance (AMR) gene detection enhance clinical outcomes and optimize antimicrobial stewardship in patients with HMs and sepsis?

What was learned from this study?

ddPCR’s rapid pathogen and AMR genes detection enabled early precise antimicrobial therapy in patients with HMs and sepsis.

ddPCR enhanced clinical efficiency and reduced the 28-day all-cause mortality.

ddPCR effectively optimized scientific antimicrobial stewardship.

Introduction

Sepsis, a life-threatening disorder of acute organ dysfunction due to the body’s dysregulated response to infection, poses a significant global health risk. Globally, approximately 48.90 million sepsis cases are reported annually, leading to 11 million deaths [1]. The patients with hematologic malignancies (HMs) face heightened sepsis risk (15-fold higher than non-HM populations) [2, 3]. Bloodstream infections (BSIs) are a major sepsis driver in HMs, with mortality rates up to 42% [4, 5], exacerbated by antimicrobial resistance (AMR). The prevalence of sepsis in neutropenic patients is up to 45% [6], leading to mortality rates as high as 69% [7, 8]. Yu (2024) [9] conducted a study on 201 patients with HMs and carbapenem-resistant Gram-negative BSIs, reporting a 28-day all-cause mortality rate of 58.70% and an overall mortality rate of 65.20%.

In patients with HMs, immunosuppression, neutropenia, and comorbidities contribute to atypical manifestations of the systemic acute inflammatory response in sepsis. This poses significant diagnostic and therapeutic challenges and is associated with high mortality from severe infections [10]. Early, accurate pathogen identification enabling precise treatment is critical for optimizing anti-infection strategies, improving outcomes, and reducing mortality in sepsis management [11–13]. Traditional blood culture (BC), the gold standard for BSI diagnosis, has limitations: low positivity (10–30%) [14] and sensitivity (≤ 50%) [15], delayed results (3–5 days) [16, 17], and reliance on empirical broad-spectrum antibiotics, risking treatment delays in immunocompromised patients [18].

Droplet digital PCR (ddPCR), a third-generation PCR technology with high sensitivity (10 CFU/ml) [19], a rapid 4-h turnaround time [20], and the ability to identify pathogens, AMR genes and quantify infection load, enables early precision antimicrobial therapy. Based on pathogen distribution patterns in patients with HMs, our ddPCR assay was designed with specific primers/probes targeting conserved nucleic acid regions of 13 pathogens and seven bacterial resistance genes, enabling rapid quantitative detection of their nucleic acids via digital PCR. It is expected to overcome limitations of BC and existing molecular techniques to facilitate infection diagnosis and treatment.

Metagenomic next-generation sequencing (mNGS), a strong competitor to ddPCR, also shows great potential in pathogen detection. However, both ddPCR and mNGS have their own advantages in the etiological diagnosis of BSIs. By combining microfluidic technology with PCR, ddPCR can provide absolute quantitative results as low as 10 CFU/ml within approximately 4 h, making it suitable for rapid detection of known targets and efficacy monitoring. Nevertheless, it is limited by preset targets and prone to missed diagnosis of non-targeted pathogens. In contrast, mNGS uses high-throughput unbiased sequencing to cover bacteria, fungi, viruses, and drug resistance/virulence genes in one go. Its sensitivity is further increased by 18–25% in culture-negative samples, with reports available within 24–48 h, enabling the simultaneous detection of emerging pathogens. However, it has high requirements for equipment and bioinformatics, high costs, reduced sensitivity in the presence of high human background, and complex result interpretation [21].

Therefore, we believe that based on their respective limitations and advantages, ddPCR is more suitable for the rapid detection of common isolated pathogens and AMR genes in critically ill patients with suspected BSI, while mNGS is more applicable to BSI diagnosis where pathogens cannot be identified by classical microbiological or molecular diagnostic methods. Considering accessibility, economy, and the particularity of patients with hematologic malignancies complicated by infections, this study deems it highly necessary to carry out ddPCR and clarify its clinical value in patients with hematologic malignancies.

ddPCR platforms exhibit variable sensitivity and specificity for pathogens across populations [22]. ddPCR shows higher accuracy and sensitivity than BC for BSI diagnosis in emergency or intensive care unit (ICU) settings [23–25]. Currently, only one retrospective study on febrile hematological patients has reported that ddPCR exhibits a diagnostic sensitivity of 87.50% and a turnaround time (TAT) of 7.56 h [26], with diagnostic performance superior to BC. Limited research has been conducted on the application of ddPCR for patients with HMs and sepsis. Key gaps include: (1) association with 28-day mortality; (2) potential to optimize antimicrobial therapy, reduce inflammation, and improve efficacy; (3) impact on antimicrobial use density (AUD) and combined antimicrobial usage.

To address these critical knowledge gaps, this study systematically evaluated the clinical utility of ddPCR in patients with HMs and sepsis. By systematically assessing outcomes, including 28-day all-cause mortality, clinical outcomes, and antimicrobial stewardship indicators (e.g., AUD), we aimed to establish the diagnostic and prognostic value of ddPCR in immunosuppressed populations and advance the integration of precision molecular diagnostics into clinical practice.

Methods

Design and Setting

We conducted a retrospective cohort study of inpatients with HMs and sepsis who underwent ddPCR and/or BC testing at Tongde Hospital of Zhejiang Province (a 2300+ bed tertiary hospital) from January 2023 to March 2025. The study was approved by the Ethics Committee of the Tongde Hospital of Zhejiang Province (No. 2025-087 (JY)). Inclusion and exclusion criteria are shown in Table 1.

Table 1.

Inclusion criteria and exclusion criteria

Inclusion criteria:
1. Adults ≥ 18 years old
2. Inpatients with HMs, including leukemia, lymphoma, multiple myeloma (MM), myelodysplastic syndromes (MDS), paroxysmal nocturnal hemoglobinuria (PNH), etc
3. Meeting the diagnostic criteria of sepsis 3.0: It is defined as a life-threatening organ dysfunction caused by a dysregulated host response to infection. Clinically, for patients with a suspicion of infection, an increase of at least 2 points in the SOFA score is used as the judgment criterion for organ dysfunction in sepsis. If the baseline SOFA score is unknown, it is presumed to be zero
4. Inpatients who have a ddPCR and/or BC on the same day as enrollment (day 0)
5. Having complete clinical data
Exclusion criteria:
1. Younger than 18 years old
2. Non-HMs, including idiopathic thrombocytopenic purpura (ITP), anemia, chemotherapy-induced thrombocytopenia (CIT), etc
3. No blood culture during hospitalization
4. Inpatients who were discharged from the hospital or refused active treatment prior to the availability of specimen test results

HMs hematologic malignancies, BC blood culture, ddPCR droplet digital polymerase chain reaction, SOFA Sequential Organ Failure Assessment

The Principle of ddPCR

Detection was performed using the 6-fluorescence-channel ddPCR system from Linghang Gene Technology (Hangzhou) Co., Ltd. The complementary kit for this system was designed with specific primers and fluorescent probes targeting the conserved nucleic acid regions of 13 pathogenic microbial genera and seven bacterial resistance genes, enabling rapid and quantitative detection of pathogen and resistance gene nucleic acids in samples via digital PCR technology. The range of detected pathogens is detailed in Table S1.

The experimental procedure is briefly described as follows: nucleic acids were extracted using the complementary nucleic acid extraction reagents; after droplets were generated by a droplet generator, the chip was placed in a PCR amplifier for amplification; finally, analysis was performed by scanning with a biochip reader. According to the kit instructions, the detection thresholds for Candida, Streptococcus, and coagulase-negative Staphylococcus are 1.00 copies/μl, and those for the others are 0.50 copies/μl. Samples are considered positive if concentrations exceed these thresholds.

Clinical Adjudication for ddPCR Results

ddPCR detected eight types of Gram-negative bacteria, four types of Gram-positive bacteria, one type of fungus, and seven AMR genes (Table S1). Clinicians independently evaluated whether ddPCR-identified pathogens caused sepsis using lab results, imaging, and clinical data, then analyzed antimicrobial regimen adjustments (discontinuation, dose modification, or maintenance). Therapeutic efficacy was assessed by clinical symptom improvement within 7 days post-adjustment and declines in CRP/PCT levels to determine ddPCR’s impact on treatment decisions and prognosis.

Data Sources

Data were retrieved from the electronic medical record system, and survival data were gathered via standardized telephone follow-up interviews. Demographic and clinical data, including age, gender, BMI, primary diseases status, aCCI (age-adjusted Charlson Comorbidity Index), Sequential Organ Failure Assessment (SOFA) score, hemoglobin (HB) level, albumin level, D-dimer level, and inflammatory markers, were collected. This trial enrolled patients with sepsis who underwent BC and/or ddPCR. Follow-up ended when an outcome event occurred, patients were lost to follow-up, or the observational period concluded.

Outcomes

Primary outcomes included 28-day all-cause mortality, AUD, and the quantity of combined antimicrobial usage. Secondary outcomes comprised clinical treatment effectiveness, the rate of patients with CRP/PCT down to 50% within 48 h or 72 h, the proportion of combined antimicrobial usage, and the ratio of antimicrobial costs to total medication costs.

Statistical Analysis

Continuous variables are presented as mean ± standard deviation or median (IQR) based on the data distribution. Categorical variables are described using counts and frequencies. Appropriate tests were used for baseline comparisons. Variables with over 20% missing data were excluded from the analysis. For those with less than 20% missing data, missing values were handled through multiple imputation using the "MICE" package (multiple imputation by chained equations), and estimates were combined following Rubin’s rules to ensure unbiased results. The Chi-square test evaluated diagnostic test performance.

Propensity score matching (PSM) was performed to address non-random treatment assignment. Patients were matched 1:1 using nearest neighbor matching (caliper 0.20). Balance was assessed via standardized mean differences (SMD < 0.10). Outcomes were analyzed using time-dependent Cox regression (HRs and 95% CIs), Kaplan–Meier survival estimates, linear regression (β coefficients), and logistic regression (ORs). Subgroup analyses (age ≥ 65, BMI ≥ 24, SOFA ≥ 5, etc.) explored treatment effect heterogeneity for primary outcomes. Sensitivity analyses were performed using multiple models to validate data robustness, including multivariate regression models, stabilized inverse probability of treatment weighting (sIPTW), and propensity score adjustment. Additionally, sensitivity analyses evaluated data robustness by deleting missing values. Statistical significance was set at P < 0.05. All analyses were performed in R 4.4.2.

Results

Clinical Characteristics

From Jan 2023 to Mar 2025, 597 inpatients with HMs underwent BC and/or ddPCR. Following the exclusion of 197 patients (non-HMs, SOFA < 2, discharge and discontinued treatment before results), a total of 400 patients were enrolled: 150 in the ddPCR + BC group and 250 in the BC-alone group (Fig. 1). A total of 400 patients were included in the study, with a mean age of 61.8 ± 16.42 years and 56% being male (Table 2). The mean BMI was 22.32 ± 3.42 kg/m², and 48.25% of the patients had primary diseases in progressive status. Median SOFA score was 4.00 (IQR: 3.00, 5.00), aCCI score 6.00 (IQR: 4.00, 7.00), and 59.75% had multisite infections. 47.25% patients had agranulocytosis. Demographic characteristics (age, gender, BMI, disease status, SI, neutropenia, CRP) did not differ between groups. Compared to the BC group, the ddPCR + BC group had significantly higher SOFA and aCCI scores (P < 0.01), lower HB/albumin levels, and higher PCT/D-dimer levels.

Fig. 1.

Study cohort. HMs hematologic malignancies, BC blood culture, ddPCR droplet digital PCR

Table 2.

Baseline characteristics of participants receiving BC or ddPCR combined with BC before and after propensity score matching (PSM)

Baseline characteristics	Before PSM					After PSM
	Total	BC group	ddPCR + BC group	P value	SMD	Total	BC group	ddPCR + BC group	P value	SMD
	(n = 400)	(n = 250)	(n = 150)			(n = 260)	(n = 130)	(n = 130)
Age (years)	61.80 ± 16.42	61.88 ± 15.65	61.67 ± 17.67	0.90	− 0.01	61.04 ± 17.10	60.36 ± 17.27	61.72 ± 16.97	0.52	0.08
Sex, n (%)				0.21					0.71
Male	224 (56.00)	134 (53.60)	90 (60.00)		0.13	143 (55.00)	70 (53.85)	73 (56.15)		0.05
Female	176 (44.00)	116 (29.00)	60 (40.00)		− 0.13	117 (45.00)	60 (46.15)	57 (43.85)		− 0.05
BMI (kg/m2)	22.32 ± 3.42	22.27 ± 3.43	22.40 ± 3.41	0.71	0.04	22.20 ± 3.60	22.09 ± 3.74	22.31 ± 3.47	0.62	0.06
PD, n (%)	193 (48.25)	119 (47.60)	74 (49.33)	0.74	0.04	137 (52.69)	69 (53.08)	68 (52.31)	0.90	− 0.02
SOFA	4.00 (3.00, 5.00)	4.00 (2.00, 5.00)	4.00 (3.00, 6.00)	< 0.01	0.34	4.00 (3.00, 5.00)	4.00 (3.00, 5.00)	4.00 (3.00, 5.00)	0.94	0.04
aCCI	6.00 (4.00, 7.00)	5.00 (4.00, 7.00)	6.00 (4.00, 8.00)	0.03	0.21	5.00 (4.00, 7.00)	5.00 (4.00, 7.00)	6.00 (4.00, 7.00)	0.59	0.06
Multi-site infection, n (%)	239 (59.75)	137 (54.80)	102 (68.00)	0.01	0.28	168 (64.62)	83 (63.85)	85 (65.38)	0.80	0.03
Shock Index (SI)	0.98 ± 0.30	0.97 ± 0.30	0.98 ± 0.29	0.86	0.02	0.97 ± 0.28	0.97 ± 0.28	0.98 ± 0.29	0.60	0.06
Agranulocytosis, n (%)	189 (47.25)	125 (50.00)	64 (42.67)	0.16	0.15	126 (48.46)	67 (51.54)	59 (45.38)	0.32	0.12
Hemoglobin (g/l)	78.31 ± 23.34	81.47 ± 24.59	73.04 ± 20.07	< 0.01	− 0.42	73.43 ± 20.94	73.00 ± 21.50	73.86 ± 20.44	0.74	0.04
Albumin (g/l)	32.19 ± 5.29	33.11 ± 5.44	30.65 ± 4.66	< 0.01	− 0.53	31.35 ± 4.72	31.44 ± 4.84	31.26 ± 4.61	0.77	− 0.04
PCT (ng/ml)	0.51 (0.19, 2.99)	0.46 (0.15, 2.33)	0.86 (0.24, 4.40)	0.01	0.15	0.59 (0.22, 2.96)	0.59 (0.21, 2.51)	0.64 (0.22, 3.16)	0.73	0.04
CRP (mg/l)	65.00 (27.00, 129.88)	60.22 (26.10, 128.55)	75.10 (28.9, 137.20)	0.34	0.11	66.80 (28.35, 128.30)	63.20 (33.45, 127.80)	70.40 (27.30, 128.70)	0.65	0.07
D-dimer (mg/l)	2.07 (0.94, 4.72)	1.83 (0.77, 4.45)	2.44 (1.23, 5.54)	0.02	0.13	1.98 (0.99, 4.64)	1.96 (0.91, 4.91)	1.98 (1.06, 4.55)	0.64	− 0.03

PSM propensity score matching, BC blood culture, ddPCR droplet digital polymerase chain reaction, BMI body mass index, PD progressive disease, SOFA Sequential Organ Failure Assessment, aCCI age-adjusted Charlson Comorbidity Index, PCT procalcitonin, CRP C-reactive protein, SMD standardized mean difference

In this study, 59.75% (239/400) of the pathogens causing sepsis in patients originated from multiple sites, with 78.50% (314/400) derived from the respiratory tract and 49.75% (199/400) from the bloodstream. Compared to the BC group, the ddPCR + BC group had a significantly higher multi-site infection rate (68.00% vs. 54.80%, P < 0.01). Specific differences in infection sources between the two groups are shown in Table S2. In our study, the proportion of missing data for all variables was below 10.00%, specifically: 3.00% (12/400) for body mass index (BMI), 1.75% (7/400) for procalcitonin (PCT), and 9.75% (39/400) for D-dimer.

Performance and Concordance Rate of ddPCR and BC

The positive rate of BC was 17.50% (70/400), with a median TAT of 72.47 h (IQR: 66.22–92.76 h; Table S3-S4 and Fig. 2). For ddPCR + BC group, the positive rate was 49.33% (74/150) with a median TAT of 4.06 h (4.00–6.68 h), all detected faster than BC group. Concordance between ddPCR and BC was 58.00% (87/150), including 19 ddPCR⁺/BC⁺ and 68 ddPCR⁻/BC⁻ cases (Table S4). Among BC-positive patients, 73.08% (19/26) benefited from ddPCR’s early diagnosis, and ddPCR provided early etiological evidence in 72.00% (54/75) of suspected infections. For BC-verified blood specimens, ddPCR showed 70.37% sensitivity, 55.28% specificity, 25.68% PPV, and 89.47% NPV. ddPCR’s diagnostic performance significantly improved when combined with clinical BSI diagnosis (Table 3).

Fig. 2.

Comparison of turnaround time (TAT) between two groups. A Inter-group comparison of positive rates and TAT across the entire study population. B Comparison of TAT between digital droplet PCR (ddPCR) and blood culture (BC) in patients with positive test results

Table 3.

The sensitivity, specificity, PPV, and NPV of ddPCR in BSIs

Samples (n = 150)	ddPCR+	ddPCR−	Sensitivity	Specificity	PPV	NPV
Positive by BC	19	8	70.37%	55.28%	25.68%	89.47%
Negative by BC	55	68
Positive by clinical diagnosis	73	27	73.00%	98.00%	98.65%	64.47%
Negative by clinical diagnosis	1	49

ddPCR droplet digital PCR, PPV positive predictive values, NPV negative predictive values, BC blood culture, BSI bloodstream infections

In the 74 ddPCR-positive specimens, a total of 127 pathogenic strains and 24 AMR genes were identified (Fig. 3). Gram-negative bacteria accounted for 66.14% of the detected pathogens, while Gram-positive bacterial strains and fungal strains accounted for 24.41% (31 strains) and 9.45% (12 strains), respectively. Among the AMR genes, 75.00% were bla_KPC, 12.50% were bla_NDM/bla_MIP, and 12.50% were mecA (Table S4).

Fig. 3.

Pathogen distribution detected by ddPCR and BC. A Pathogens identified using ddPCR and BC, with bar lengths indicating case numbers. B In the inner ring, light gray represents fungi, dark gray indicates Gram-negative bacteria (Gram−), and the rest represents Gram-positive bacteria (Gram+). The ring sizes are proportional to case counts. dd+ positive ddPCR test results, dd− negative ddPCR test results, BC+ positive BC results, BC− negative BC results, Salmonella ser. Newport, Salmonella enterica subsp. enterica serovar Newport. KPC Klebsiella pneumoniae carbapenemase, NDM New Delhi metallo-β-lactamase, IMP imipenem-hydrolyzing β-lactamase

In the 26 BC-positive specimens, a total of 29 pathogenic strains were identified (Fig. 3), including six carbapenem-resistant Enterobacteriaceae (CRE) strains. Gram-negative bacteria accounted for 66.14% of the detected pathogens, while Gram-positive bacterial strains and fungal strains comprised 20.68% (six strains) and 10.34% (three strains), respectively. Furthermore, 44.59% (33/74) of the ddPCR-positive specimens harbored 2–7 pathogens, whereas merely two of the BC-positive specimens contained two species (Table S4).

Propensity Score Matching (PSM)

Figure S1 illustrates the distribution of estimated propensity scores for undergoing ddPCR testing among inpatients. After PSM, 130 matched pairs of patients were obtained (Table 1), with significantly attenuated group differences in variables in the PSM samples (Figure S2). The baseline characteristics were balanced after PSM.

Clinical Efficacy Indicators

During the follow-up period, the 28-day all-cause mortality rates of the ddPCR + BC group and BC group after PSM were 21.54% and 36.15% (Table 4). The implementation of ddPCR led to a significant reduction in the 28-day all-cause mortality, with an HR of 0.55 and a 95% CI of 0.34–0.88 (P = 0.01; Fig. 4). After adjusting for confounders in model 3 (Table 4), ddPCR was associated with lower 28-day all-cause mortality (adjusted HR = 0.49, P < 0.01), consistent across three PS analyses (Table 5).

Table 4.

The outcomes before and after PSM analysis

Comprehensive indicators	Before PSM					After PSM
	Total (n = 400)	BC group (n = 250)	ddPCR + BC group (n = 150)	HR, OR or β (95% CI)	P value	Total (n = 260)	BC group (n = 130)	ddPCR + BC group (n = 130)	HR, OR or β (95% CI)	P value
28-day all-cause mortality	106 (26.5%)	75 (30.0%)	31 (20.57%)	0.66 (0.43,1.0)	0.05	75 (28.85%)	47 (36.15%)	28 (21.54%)	0.55 (0.34, 0.88)	0.01
AUD	190.46 ± 90.01	201.34 ± 99.31	172.33 ± 68.43	− 29.01 (− 47.03, − 11.00)	< 0.01	190.37 ± 87.06	204.83 ± 100.92	175.91 ± 67.91	− 28.93 (− 49.84, − 8.02)	< 0.01
Rate of clinical effectiveness	295 (73.75%)	175 (70.0%)	120 (80.0%)	1.71 (1.10, 2.80)	0.03	187 (71.92%)	84 (64.62%)	103 (79.23%)	2.12 (1.19, 3.77)	0.01
Quantity of combined antimicrobial usage	3.00 (2.00, 3.00)	3.00 (2.00, 3.00)	2.00 (2.00, 3.00)	− 0.46 (− 0.64, − 0.27)	< 0.01	3.00 (2.00, 3.00)	3.00 (2.00, 3.00)	2.00 (2.00, 3.00)	− 0.49 (− 0.71, − 0.28)	< 0.01
Proportion of combined antimicrobial usage	353 (88.25%)	231 (92.4%)	122 (81.33%)	0.36 (0.19,0.67)	< 0.01	228 (87.69%)	123 (94.62%)	105 (80.77%)	0.14 (0.04,0.48)	< 0.01
Rate of patients with CRP down to 50% within 48 h	60 (15.00%)	26 (10.40%)	34 (22.67%)	2.53 (1.45, 4.41)	< 0.01	41 (15.77%)	10 (7.69%)	31 (23.85%)	3.63 (1.66, 7.93)	< 0.01
Rate of patients with PCT down to 50% within 48 h	70 (17.50%)	32 (12.80%)	38 (25.33%)	2.31 (1.37, 3.90)	< 0.01	47 (18.08%)	15 (11.54%)	32 (24.62%)	2.31 (1.21, 4.42)	0.01
Rate of patients with CRP down to 50% within 72 h	107 (26.75%)	55 (22.00%)	52 (34.67%)	1.88 (1.20, 2.95)	0.01	68 (26.15%)	21 (16.15%)	47 (36.15%)	3.36 (1.72, 6.59)	< 0.01
Rate of patients with PCT down to 50% within 72 h	119 (29.76%)	57 (22.80%)	62 (41.33%)	2.39 (1.54, 3.70)	< 0.01	77 (29.62%)	24 (18.46%)	53 (40.77%)	3.64 (1.87,7.09)	< 0.01
Ration of antimicrobial agent cost to total medication cost	48.06 ± 22.75%	48.09 ± 23.49%	48.02 ± 21.53%	− 0.07 (− 4.68, 4.54)	0.98	48.36 ± 22.71	50.44 ± 23.48%	46.28 ± 21.82%	− 4.16 (− 9.67,1.35)	0.14

PSM propensity score matching, BC blood culture, ddPCR droplet digital polymerase chain reaction, AUD antibiotic use density, HR hazard ratio, OR odds ratio, β regression coefficient, CI confidence interval

Fig. 4.

Mortality comparison between two groups for patients with HMs and sepsis. Kaplan–Meier curves depicting 28-day all-cause mortality in the main cohort (A) and the propensity score matching (PSM) cohort (B)

Table 5.

The outcomes in the regression analysis and propensity-score analysis

Analysis	28-day all-cause mortality		AUD		Quantity of combined antimicrobial usage		Proportion of combined antimicrobial usage		Rate of patients with CRP down to 50% within 48 h		Rate of patients with PCT down to 50% within 48 h		Rate of patients with CRP down to 50% within 72 h		Rate of patients with PCT down to 50% within 72 h		Ration of antimicrobial agent cost to total medication cost		Quantity of combined antimicrobial usage
Regression analysis; values shown are no. of events/no. of patients at risk (%) or median (IQR), days
BC (n = 250)	75 (30.00%)		201.34 ± 99.31		175 (70.00%)		3.00 (2.00, 3.00)		231 (92.40%)		26 (10.40%)		32 (12.80%)		55 (22.00%)		57 (22.80%)		48.09 ± 23.49%
ddPCR + BC group (n = 150)	31 (20.57%)		172.33 ± 68.43		120 (80.00%)		2.00 (2.00, 3.00)		122 (81.33%)		34 (22.67%)		38 (25.33%)		52 (34.67%)		62 (41.33%)		48.02 ± 21.53%
Crude analysis, HR or β (95% CI), P value	0.66 0.43; 1.00	0.05	− 29.01 − 47.03, 11.00	< 0.01	1.71 1.06, 2.78	0.03	− 0.46 − 0.64, − 0.27	< 0.01	0.36 0.19, 0.67	< 0.01	2.53 1.45, 4.41	< 0.01	2.31 1.37, 3.90	< 0.01	1.88 1.20, 2.95	0.01	2.39 1.54, 3.70	< 0.01	− 0.07 − 4.68, 4.54	0.98
Multivariable analysis: model 1, HR or β (95% CI), P value	0.56 0.37, 0.86)	0.01	− 28.90 − 46.81, − 10.99	< 0.01	2.37 1.33, 4.22	< 0.01	− 0.45 − 0.63, − 0.27	< 0.01	0.34 0.18, 0.64	< 0.01	2.50 1.41, 4.42	< 0.01	2.33 1.36, 3.98	< 0.01	1.88 1.18, 3.00	< 0.01	2.42 1.54, 3.79	< 0.01	− 0.37 − 4.95, 4.22	0.88
Multivariable analysis: model 2, HR or β (95% CI), P value	0.49 0.32, 0.76	< 0.01	− 24.52 − 42.51, − 6.52	0.01	3.25 1.68, 6.29	< 0.01	− 0.47 − 0.65, − 0.29	< 0.01	0.31 0.15, 0.64	< 0.01	2.94 1.59, 5.44	< 0.01	1.99 1.13, 3.53	0.02	2.12 1.29, 3.49	< 0.01	2.14 1.34, 3.42	< 0.01	− 1.65 − 6.34, 3.04	0.49
Multivariable analysis: model 3, HR or β (95% CI), P value	0.49 0.31, 0.77	< 0.01	− 25.02 − 43.33, − 6.71	0.01	3.41 1.73, 6.74	< 0.01	− 0.52 − 0.70, 0.34	< 0.01	0.24 0.11, 0.52	< 0.01	2.70 1.43, 5.10	< 0.01	1.84 1.01, 3.37	0.05	2.21 1.32, 3.71	< 0.01	2.35 1.41, 3.91	< 0.01	− 2.03 − 6.72, 2.65	0.40
Propensity-score analysis
PSM: Adjust: ANC_0.5 HR or β (95% CI), P value (n = 260,1:1)	0.560 0.35, 0.90	0.02	− 25.78 − 45.82, − 5.74	0.01	2.12 1.18, 3.81	0.01	− 0.46 − 0.67, − 0.25	< 0.01	0.17 0.05, 0.59	0.01	3.57 1.61, 7.93	< 0.01	2.36 1.20, 4.66	0.01	3.30 1.67, 6.53	< 0.01	3.65 1.83, 7.27	< 0.01	− 4.19 − 9.72, 1.34	0.14
PS-adjusted, HR or β (95% CI), P value	0.47 0.29, 0.78	< 0.01	− 25.6 − 44.9, − 6.34	0.01	2.38 (1.41, 4.1)	< 0.01	− 0.52 − 0.72, − 0.32	< 0.01	0.34 0.17, 0.65	< 0.01	2.55 1.41, 4.66	< 0.01	1.85 1.06, 3.25	0.03	2.05 1.27, 3.24	< 0.01	2.08 1.31, 3.33	< 0.01	− 2.12 − 7.03, 2.78	0.40
sIPTW, HR or β (95% CI), P value	0.58 0.36, 0.92	0.01	− 25.45 − 42.06, − 8.83	< 0.01	2.79 1.47, 5.28	< 0.01	− 0.45 − 0.65, − 0.25	< 0.01	0.29 0.13, 0.65	< 0.01	2.72 1.44, 5.16	< 0.01	1.87 1.02, 3.40	0.04	2.15 1.27, 3.62	< 0.01	2.35 1.42, 3.90	< 0.01	− 0.02 − 0.07, 0.03	0.40

BC blood culture, ddPCR droplet digital polymerase chain reaction, AUD antibiotic use density, PSM propensity score matching, PS-adjusted adjusted for propensity score, sIPTW stable-inverse probability of treatment weighting, HR hazard ratio, OR odds ratio, β regression coefficient, CI confidence interval

Model 1 was adjusted for males, age, BMI, state_2

Model 2 was adjusted as for model 1 plus infection_multiple, SOFA, aCCI, SI, ANC_0.5

Model 3 was adjusted as for model 2 plus HB, PCT, CRP, albumin

Post-PSM, ddPCR use was associated with faster CRP/PCT decline, including higher odds of ≥ 50.00% CRP reduction within 48 h (OR 3.63; P < 0.01), PCT reduction within 48 h (OR 2.31; P = 0.01), and within 72 h (two sets: OR 3.63, P < 0.01 and OR 3.64, P < 0.01) (Table 4). Clinical effectiveness rates were 79.23% (ddPCR + BC) vs. 64.62% (BC; OR 2.12, P = 0.01), consistent with multivariate and PS analyses (Table 5).

Antimicrobial Stewardship

In this study, clinicians adjusted antibacterial agents in 47 cases (63.51%) using positive ddPCR results; 13 (17.57%) with negative ddPCR results continued original therapy, and 14 (18.92%) adjusted agents based on BC results and symptoms (Fig. 5A). For patients with negative results in both ddPCR and BC, clinicians administered empirical anti-infective therapy based on the patients’ clinical symptoms, inflammatory markers (such as CRP and PCT), and responses to empirical anti-infective regimens. Among ddPCR-guided adjustments, 85.11% (40/47) showed effective anti-infection treatment, versus 68.96% (220/319) in unadjusted/empirical groups (Fig. 5B, C). Post-PSM, ddPCR significantly reduced AUD (OR = − 28.93, P < 0.01) (Fig. 5D, Table 4), consistent across regression and PS-analyses (Table 5).

Fig. 5.

Effectiveness of antibiotic therapeutic strategies. A Post-diagnostic antibiotic regimen modifications. B Efficacy of antibiotic regimen adjustments informed by ddPCR-based pathogen identification. C Efficacy of empiric antibiotic therapy administered prior to definitive diagnosis. D The antibiotic use density (AUD) of two groups. E The quantity of combined antimicrobial usage in two groups

The results of this study showed that the proportion of combined antimicrobial usage in the ddPCR group was 80.77%, which was significantly lower than that in the BC group (94.62%) (P < 0.01) (Table 4). Although the proportion of combined antimicrobial usage in the two groups was similar, ddPCR could significantly reduce the quantity of combined antimicrobial usage (P < 0.01) (Fig. 5E, Table 4). However, the proportion of antibacterial expenditure exhibited a downward trend in the ddPCR + BC group, though the difference fell short of statistical significance (OR = − 4.16; P = 0.14).

Subgroup and Sensitivity Analysis

Supplementary Figures S3-S5 show stratification and interaction analyses. ddPCR had consistent mortality effects across subgroups (all interaction P ≥ 0.05). A significant interaction (P < 0.05) was observed between the AUD and the quantity of combined antimicrobial usage, suggesting the possible existence of a subgroup effect. In ddPCR + BC group, patients with HMs in regression had lower AUD (OR = − 54.93, interaction P = 0.02) and those with SOFA ≥ 5 had even greater AUD reduction (OR = − 63.5, interaction P = 0.02). Regarding the quantity of combined antimicrobial usage, ddPCR was associated with reduced use in patients with progressive disease (OR = − 0.63, interaction P = 0.03) and those with SOFA ≥ 5 (OR = − 0.96, interaction P < 0.01). Subgroup analysis showed that ddPCR significantly reduced AUD and the quantity of combined antimicrobial usage in critically ill HM subgroups. ddPCR optimized antimicrobial stewardship by targeting pathogen and resistance data.

The outcomes of the six models (Table 5) and analyses of complete data after excluding missing values (Tables S5–S6) were consistent with the primary analysis (PSM), confirming the reliability and consistency of our findings. Additionally, sensitivity analyses incorporating hematologic malignancy subtypes and duration of agranulocytosis (Tables S7–S8) further validated the robustness and reliability of these results.

Discussion

Immunosuppression increases sepsis risk in patients with HMs. Early and accurate pathogen identification is crucial for improving prognosis [11]. While ddPCR shows promise for early pathogen diagnosis [27], further data are needed to support broader clinical use. In our study of 400 patients with HMs and sepsis, ddPCR significantly outperformed traditional BC in pathogen and AMR genes diagnosis.

However, this study found ddPCR’s diagnostic performance in patients with HMs remains improvable. dPCR’s diagnostic performance in infectious diseases, ICU [18] and emergency patients (88.89% sensitivity, 55.61% specificity) [20] surpassed that in patients with HMs. We speculate that the population-specific characteristics of patients with HMs may be one of the reasons for the difference in performance. Additionally, it has been reported in the literature [25] that the sensitivity of ddPCR in patients without empirical antibiotic therapy can reach 100%, while the sensitivity decreases to 55.56% after receiving such therapy. In this study, 75% (300/400) of patients used broad-spectrum antibiotics within 3 days before ddPCR testing, which may be another important reason for the difference in performance.

Further comparative studies revealed that in a retrospective study involving 71 febrile hematological patients [26], ddPCR demonstrated better detection performance than our study, potentially due to differences in pathogen coverage. Unlike that study, our ddPCR did not include five viruses (detection rate: 62.90% in the prior study) and five Gram-negative bacteria (rarely detected, only once). We speculate that the exclusion of virus detection is one of the reasons for the difference in performance. In addition, the detection range of ddPCR in this study did not include Aspergillus and Mucor, which are increasingly prevalent in patients with hematological malignancies, and this further limited its diagnostic efficacy. In summary, key reasons for the potential improvement in ddPCR’s diagnostic performance in this study include the unique characteristics of patients with HMs, prior exposure to broad-spectrum antibiotics, and limited detectable pathogen range. Thus, in subsequent prospective studies, our team will focus on this special population, optimize ddPCR timing, and expand detectable pathogens. This is expected to further clarify and enhance ddPCR’s diagnostic performance in patients with HMs.

Mixed infections are relatively common in patients with HMs. Notably, ddPCR showed a higher detection rate of mixed pathogens (44.59% vs. 26.39% for BC), highlighting its potential in identifying mixed infections in immunosuppressed patients. Additionally, ddPCR detected seven AMR genes, overcoming BC’s inability to provide resistance genes. This enables ddPCR to facilitate more precise use of antimicrobial agents, timely response to drug-resistant bacteria, and rapid control of disease progression.

The timeliness and accuracy of early antibacterial treatment are closely linked to sepsis prognosis. A study [28] showed BC-based sensitive antimicrobial therapy takes 52 h, while the median time to death of carbapenem-resistant Enterobacteriaceae BSI was 4 days. Delayed antibiotic access correlates with higher mortality. A review found 6-h timely antimicrobial therapy reduces death risk by 57.00% (OR = 0.57, 95% CI 0.39–0.82). Each hour delay increases in-hospital mortality by 9.00%, and the absolute mortality rate of patients with septic shock increases by 1.80%/h [10, 29]. Our ddPCR assay, with a TAT 18-fold faster than BC, enabled early pathogen detection in 73.08% of BC-positive cases and 72.00% of clinically suspected infections, facilitating early targeted therapy consistent with prior findings by Li et al. [26]. Thus, the rapidity and accuracy of ddPCR imply its potential to optimize clinical decision-making [22, 30] and possibly decrease the mortality associated with BSIs.

While ddPCR enables rapid pathogen detection, its impact on clinical outcomes and antimicrobial stewardship remains unclear. Prior studies on other rapid diagnostic methods have yielded conflicting results: bla_KPC gene identification reduced mortality in CRE infections [31], whereas MALDI-TOF-based detection showed no effect on 30-day mortality [32]. Our study is the first to demonstrate that ddPCR accelerates the decline of inflammatory biomarkers (CRP/PCT) by > 50.00% within 48–72 h in patients with HMs and sepsis, enhances treatment efficacy, and reduces 28-day all-cause mortality.

To further explore the clinical potential of ddPCR, we assessed its value for antimicrobial stewardship. Our study in patients with HMs aligns with prior research in emergency, infectious disease, and ICU settings, confirming ddPCR’s role in real-time antibacterial treatment optimization [18, 23, 25]. Our study first confirms that ddPCR optimizes antimicrobial stewardship via real-time treatment adjustments, reducing AUD and combination therapy using in patients with HMs and sepsis.

Thus, we conclude that ddPCR rapidly detects pathogens and AMR genes, enabling clinicians to timely optimize antimicrobial regimens, avoid treatment delays, and reduce broad-spectrum antibiotic abuse, thereby improving clinical efficacy, reducing mortality, and optimizing antimicrobial stewardship. For patients with HMs and sepsis, it is recommended to simultaneously submit BC and nucleic acid-based rapid pathogen identification tests, along with rapid detection of antimicrobial resistance genes. This study suggests that ddPCR represents a preferable option for these purposes.

Although ddPCR reduced the proportion of antimicrobial agents in total drug expenditure, no significant difference was observed between the two groups. This phenomenon is speculated to be attributed to the higher detection rate of AMR genes in the ddPCR group: while the combined use of antimicrobials was reduced, the utilization of higher-tier and more expensive antimicrobials offset potential cost savings, ultimately leading to no intergroup difference in the proportion of antimicrobial expenditure. Furthermore, numerous factors continue to influence the proportion of antimicrobial agents in total drug expenditure. Consequently, this conclusion should be interpreted with caution and warrants further validation.

In this study, the cost of ddPCR ($203.19) is 5.5 times that of traditional BC ($36.81). Furthermore, as ddPCR remains a self-funded item in China at present, patient acceptance is relatively low, which hinders its application in patients with mild infections to a certain extent. Meanwhile, despite its advantages of high sensitivity and absolute quantification in pathogen detection, ddPCR has inherent limitations: it may produce false positives or false negatives due to factors such as pre-set targets, nucleic acid extraction efficiency, contamination risks, target gene mutations, and matrix interference. In addition, ddPCR still cannot distinguish between viable and non-viable bacteria. When multiple pathogens are detected or the quantified copy number is low, result interpretation remains highly dependent on the clinical experience of physicians.

mNGS exhibits potential in pathogen detection for patients with suspected BSIs due to its advantage of unbiased broad coverage, enabling simultaneous detection of multiple pathogens, including bacteria, fungi, and viruses. ddPCR, characterized by high sensitivity and rapid absolute quantification, also performs prominently. Both methods show promise in pathogen detection for suspected BSIs; however, no studies have clarified differences in their diagnostic performance among sepsis patients with HMs. A study involving 60 ICU critically ill patients with suspected BSIs [24] demonstrated that ddPCR had a higher positive rate (83.30%) than mNGS (68.30%, excluding viruses) and blood culture (16.70%). ddPCR featured faster detection speed and higher detection rate of AMR genes, albeit with a narrower detection range. Consistent with our findings, this study confirms the advantages of ddPCR in terms of high positive rate, high AMR gene detection rate, and rapid detection.

Our study did not compare these two methods, primarily considering factors such as the clinical accessibility of ddPCR, the unique features of infections in patients with HMs, patients’ economic affordability, and the limited sample size of mNGS-tested cases in our institution. As mNGS is increasingly applied in clinical practice, our team will further investigate the diagnostic differences between the two methods in patients with HMs to more accurately define their clinical application value.

This study offers key strengths: (1) It was the first study to systematically evaluate ddPCR in patients with HMs and sepsis. (2) This is the first time that ddPCR has been linked to mortality, the decline rate of inflammatory indicators, and antimicrobial stewardship (e.g., AUD and combined antimicrobial usage). By integrating 10 comprehensive indicators spanning mortality, clinical outcomes, and antimicrobial stewardship, this study represents the most comprehensive exploratory research in this field to date. (3) Multiple analytical approaches were employed to mitigate confounding and strengthen the robustness and reliability of findings, including propensity score matching (PSM), regression analyses, stabilized inverse probability of treatment weighting (sIPTW), and propensity score adjustment.

This study has several limitations: (1) Single-center retrospective design may cause selection bias, needing multicenter prospective validation despite PSM and sensitivity analyses. (2) The detection panel has limited pathogens, missing rare ones (e.g., Cryptococcus), requiring cautious interpretation. (3) We did not compare ddPCR with emerging technologies (e.g., mNGS). This aspect needs to be addressed in future research. (4) The sample sizes of some subgroups were relatively small, limiting the statistical power of the study.

Conclusions

This real-world study demonstrates that ddPCR provides dual benefits for patients with HMs and sepsis, including guiding early antimicrobial therapy, improving clinical outcomes, promoting rational drug use, reducing AUD and mortality in immunocompromised patients. These results highlight ddPCR’s utility in optimizing antimicrobial stewardship, clinical efficacy, and patient prognoses in this cohort. For patients with HMs presenting with sepsis, combining BC with ddPCR is a favorable approach for rapid identification of pathogens and AMR genes. The findings offer robust evidence-based support for precision infection management in immunosuppressed populations.

Supplementary Information

Below is the link to the electronic supplementary material.

Author Contributions

Bei Zheng, Huifang Jiang, Meiling Zhang and Yuexing Tu contributed to the conception and design of the work and take responsibility for the integrity of the data and the accuracy of the data analysis. Bei Zheng was involved in the acquisition, analysis, and interpretation of the data, and wrote and edited the manuscript, and the drawing of the graphical abstract. Chuanwei Xin and Bo Yang provided support with the data analysis. Yizhuo Liu, Enhui Lv, Hong Jiang, Wenjuan Yang, and Yuxia Jiang were involved in the acquisition of data. All authors revised the draft and approved the final version of the manuscript.

Funding

This work was supported by the Zhejiang Provincial Program for the Development of Traditional Chinese Medicine Science and Technology [2024ZL036], the Zhejiang Pharmaceutical Society Project [2022ZYJ20], Zhejiang Medical Association Projects [2B01802], and the Shanghai Medical Innovation and Development Foundation [SMIDF-142-5]. The graphical abstract is original, and the funding for its publication is the same as that of the original manuscript. The authors funded the journal’s Rapid Service Fee.

Data Availability

The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Confict of Interest

All of the authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Bei Zheng, Chuanwei Xin, Yizhuo Liu, Enhui Lv, Hong Jiang, Wenjuan Yang, Yuxia Jiang, Bo Yang, Huifang Jiang, Meiling Zhang, Yuexing Tu have nothing to disclose.

Ethical Approval

The study was approved by the Ethics Committee of the Tongde Hospital of Zhejiang Province (No. 2025-087 (JY)), and a waiver of patient consent was granted.