Since November 2021, the Omicron variant and its subvariants of the novel coronavirus SARS-CoV-2 have been spreading widely across the globe. Compared to previous variants, the Omicron variant has lower virulence and milder symptoms following infection. However, some patients infected with Omicron can still progress to critical illness, especially elderly individuals or those with weakened immune systems. The latest predictive model suggests that age, neutrophils, lymphocytes, IL-2, IL-10, and procalcitonin are the major variables in predicting progression to severe illness, particularly white blood cell count and procalcitonin inflammatory index, which are commonly used in clinics to judge sepsis [23]. As Omicron continues to spread globally, it has been estimated that the Omicron subvariant XBB.1.9 will surpass XBB.1.16 and become the dominant strain [24, 25], resulting in an increase in COVID-19 cases in countries such as India, China, and the United States.

In COVID-19 hospitalized patients, there is an increased number of bloodstream infections (BSI) in the intensive care unit (ICU) because they require invasive devices such as central venous catheters, extracorporeal membrane oxygenation (ECMO), or renal replacement therapy [26, 27]. However, there are still few studies focusing on the detection and clinical relevance of bacteremia in these COVID-19 patients. Establishing a rapid detection method for bacterial infections in critically ill patients with SARS-CoV-2 infection has broad clinical application prospects and social value. Although current debates surround whether secondary bacterial infections affect the prognosis of SARS-CoV-2 infection, some studies suggest that bacterial infections could worsen pulmonary inflammation and increase mortality rates. Some studies found that COVID-19 patients with concomitant bacterial infections required mechanical ventilation and longer ICU stays [26, 27], while another retrospective study found that the mortality rate among COVID-19 patients with bacterial infections was significantly higher than those without bacterial infections (43.1% vs 12.3%) [28]. Therefore, early and accurate selection of antibacterial drugs is essential in controlling the spread of the pathogens.

Currently, the positivity of pathogen detection in COVID-19 critically ill patients with BSI using blood culture ranges from 10 to 28% [6]. However, these studies are often based on small sample sizes. In a study conducted in Mexico [27], common pathogens found in primary bacteremia were Chryseobacterium indologenes, E. coli, and Streptococcus, while P. aeruginosa and Enterococcus Marina were found in secondary bacteremia. Cuntrò et al.[29] identified E. coli, K. pneumoniae, P. aeruginosa, and A. baumannii as Gram-negative strains, and E. faecalis, E. faecium, S. aureus, and S. pneumoniae as the Gram-positive cocci responsible for COVID-19 BSI. Moreover, another study revealed that the isolated bacteria from COVID-19 BSI patients are different from non-COVID-19 BSI patients [30]. Specifically, COVID-19 patients had higher incidence rates of Enterococcal (20.5% vs. 9%) and Acinetobacter spp. (18.8% vs. 13.6%). To address this issue, we designed a multiplex droplet digital PCR (ddPCR) panel based on global and local pathogen epidemiology targeting 18 of the most common pathogens and seven AMR genes. For the first time, our study evaluated and compared the detection of pathogens responsible for COVID-19 BSI in critically ill patients using both blood culture and multiplex ddPCR methodology.

The study found that the majority of patients (n = 93) were between 46 and 60 years old. Blood culture results disclosed that gram-negative pathogens were predominantly identified (60.7%), followed by gram-positive pathogens (30.4%) and yeast (8.9%). Among these isolates, K. pneumoniae, A. baumannii, and Enterococcus sp. were the most commonly identified pathogens, each accounting for 19.6%. The vast majority of bacteria identified through blood culture positive (BC +) (91.7%) were covered by our multiplex ddPCR targets, with only three strains falling outside of the coverage area. To broaden the scope of detection, we also performed identification at the species level for some bacteria such as Enterococcus, Staphylococcus, Streptococcus, and Candida, resulting in a wide range of pathogen detection covering bacteria and fungi. The ddPCR test had significantly higher positivity rates than the blood culture method. Of the 200 blood samples, 113 were positive (61.4%) through ddPCR, including 76 cases that were blood culture negative, and a total of 214 pathogens were detected, with 67.3% being gram-negative bacteria, 24.8% being gram-positive bacteria, and 7.9% being Candidas. A combination of blood culture and ddPCR identified 230 strains of bacteria in the targets, with A. baumannii (n = 60), K. pneumoniae (n = 35), and gram-positive Enterococcus (n = 26) and Streptococcus (n = 20) comprising the most frequently detected pathogens. In addition, Candida (n = 19) was also frequently detected. These findings are significantly different from those of Jing Wu [17], who researched non-COVID-19 bloodstream infection in ICU inpatient population using ddPCR and blood culture, identifying K. pneumoniae, P. aeruginosa, and E. faecium as the most common pathogens.

However, 11 blood culture-positive samples within the target range tested negative by ddPCR in our study. Out of the 45 BC-positive cases detected, 16 pathogens were BC positive while ddPCR negative, including 10 g-positive strains (2 E. faecalis, 5 E. faecium, 1 Oral Streptococci, 1 S. aureus, and 1 S. epidermidis), 4 g-positive strains (2 A. baumannii, 1 K. pneumoniae, and 1 E. cloacae), and 2 Candida glabrata. It is suspected that the bacteria count in the blood may be too low to be detected by PCR but can be detected by blood culture, which uses a larger volume of samples. However, the sensitivity of blood culture for gram-negative bacteria and fungi was found to be 50% and 75%, respectively, which was significantly lower than the sensitivity of 90.3% for gram-negative bacteria. This indicates that the reaction conditions for our ddPCR method are still suboptimal, especially for gram-positive bacteria. Further optimization of the ddPCR reaction system and conditions is required.

Polymicrobial bacteremia (PMB) is a frequently encountered condition where multiple microorganisms concurrently infect the bloodstream. It has been reported to comprise around 10–11% of the positive blood culture cases in recent studies[31]. Immunocompromised status, the presence of foreign objects, and recent surgical procedures increase the risk of PMB[32]. Notably, patients with PMB remain at high risk for death compared to cases of bacteremia caused by a single microorganism [32,33,34]. Unexpectedly, in our study, the rate of mixed infections among COVID-19 patients has been found over a half (60 out of 113 cases). In some instances, up to 6 different pathogenic species were detected in a single sample using ddPCR. Fungi positivity was also significantly increased, with Candida presented together with other bacterial pathogens in the mixed infections. This highlights the inadequacy of culture-based methods such as blood culture in detecting mixed bacterial and fungal infections. PMB, especially with invasive fungal diseases (IFDs), is a known area of reduced diagnostic fidelity for various pathogen detection methods, and it is still a challenge for high-quality detection requirement [35]. The ddPCR system utilized in this study exhibited a high detection rate of mixed pathogens, which may be attributed to the study population consisting of critically ill COVID-19 patients, suffering from severe lung damage, requiring mechanical ventilation, and possibly receiving extracorporeal membrane oxygenation therapy. When comparing the detection results of mixed pathogens by ddPCR with isolates from other body sites of the same patient within 7 days, we found that only 5 out of 60 mixed pathogen episodes were inconsistent (Table S1). Nonetheless, we need to be cautious to conclude that these mixed pathogens represent truly clinically determined PMB. In fact, some critically ill COVID-19 patients experienced rapid disease progression, making it difficult to allow sufficient time for clinicians to observe more data to make an informed decision. Further exploration is required to elucidate whether this ddPCR method truly provides a high detection rate of mixed infections.

This multi-ddPCR method can detect different bacteria at once, greatly increased the detection efficiency and highlighted the advantage of molecular detection that is independent of bacterial growth [36]. However, some bacterial species included in our system, such as Salmonella, Citrobacter, and Morganella morganii, were not detected. So when diagnosing BSI in COVID-19 patients, it may be necessary to optimize the design scheme and recommend adjustments such as identifying E. faecalis and E. faecium to the species level, and also the Candida albicans and Candida glabrata strains, because they have different AST pattern with each other.

For the no-COVID-19 BSI ICU inpatients, ddPCR displayed a sensitivity ranging from 58.8 to 86.7% and an aggregate specificity ranging from 73.5 to 92.2% [17]. Compared to blood culture, ddPCR showed similar sensitivity but lower specificity. However, ddPCR had a satisfactory extra detection rate, indicating that it was able to detect additional cases that blood culture could not. Importantly, 38.0% (76/200) of all tests had discordant results, with ddPCR positive while blood culture negative. Further review of clinical circumstances revealed that most of these cases were either probable (22.5%, 45/200) or possible (6.5%, 13/200) BSIs. When clinically diagnosed BSIs criteria were used as a comparator, the overall sensitivity and specificity of ddPCR were 78.1% and 90.5%, respectively. These values increased to 84.9% and 92.5% when clinically diagnosed BSI was used as true positive for the no-COVID-19 BSI ICU inpatients [17].

In addition, we designed an AMR genes detection channel in the ddPCR system. However, predicting bacterial resistance using resistance genes has always been controversial due to the many reasons for bacterial resistance and the lack of a one-to-one mapping between bacterial resistance and resistance genes. Among the 11 cultured strains of K. pneumoniae, 1 sensitive strain and 10 CRE strains were found, including 8 strains producing SCARB and 2 strains producing MBL. Seven out of eight strains producing SCARB tested positive for the blakpc gene, while the antimicrobial sensitivity strain and 2 MBL-producing strains did not have the blakpc gene detected. Therefore, the detection of the blakpc gene showed better consistency with clinical test results. Only a few other AMR genes were detected, indicating limited application of other AMR gene detection in the COVID-19 critically ill patient population. Notably, two K. pneumoniae strains that were blood culture positive were shown to be MBL-producing strains by carbapenemase typing, and the blaNDM gene was detected by ddPCR but not by the blaIMP/NDM channel in the ddPCR method. Additionally, the majority of A. baumannii detected using the cultivation method were multidrug-resistant strains, which was not anticipated in the design of the ddPCR-based AMR gene detection system. This finding highlights the necessity for a targeted design of a more suitable AMR gene detection model when applying ddPCR in the population of COVID-19 critically ill patients. Traditional bacterial drug sensitivity tests take longer to obtain results as they have to be performed after bacterium isolation and identification. The application of ddPCR for AMR genes in predicting bacterial drug sensitivity through resistance genes has always been a topic of discussion and needs further verification.

In recent years, there have been several successfully commercialized molecular assays that detect pathogens and AMR genes either directly from whole blood samples or positive blood cultures. Differently from BCID (Blood Culture Identification), assays performed with original blood samples offer the advantage of being independent of the time-consuming culture. Some narrow-based platforms primarily utilize multiplex PCR to determine target pathogens, while more extensive platforms combine broad-range PCR with amplicon sequencing. Although the multiplex ddPCR assay falls into the narrow-based category, it covers the majority of bacteria and yeast as target pathogens. Target pathogens in our study were built on the epidemiological analysis and 15-year blood culture data of our lab. The 18 pathogens included in the multiplex ddPCR panels covered over 80% of the identified positive isolates in our lab. Currently, this multiplex ddPCR platform is for research use only and costs approximately RMB 420–500 ($60–70) for one test, and the price is much lower than $135–175 per test for similar assays that are CE-IVD marked or FDA cleared [35, 37]. The experimental procedure for the multiplex ddPCR assay is relatively simple, with all steps performed in a pouch after reagent hydration. It can be semi-automatically operated with manual intervention or full-automatically handled, while the latter costs higher. The turnaround time from testing start to result is 2.0–2.5 h, which saves 1.0–3.0 h in contrast to 3.5–5 h for detection of target pathogens and AMR genes using multiplex real-time PCR-based methods, such as SeeGene MagicPlex® Sepsis Test and Roche Lightcycler® SeptiFas [38, 39]. More importantly, incorporated with droplet technology, the multiplex ddPCR is more sensitive than the real-time fluorescence quantitative PCR. This implies that the multiplex ddPCR assay possesses certain advantages; however, it has to be performed on specific and expensive instruments, and has higher environmental requirements for detection as well.

In summary, this study evaluated the detection efficacy of a multiple droplet digital PCR system for identifying pathogens in COVID-19 critical patients by comparing to conventional culture and clinical diagnosis. The multi-ddPCR method significantly improved the detection of mix pathogens and fungi, exhibited higher sensitivity, specificity, and faster turnaround times, enabling early diagnosis and timely targeted treatment for patients, especially those with sepsis. It suggests that the application of ddPCR in clinical settings has the potential to improve patient outcomes and reduce the burden of sepsis on the healthcare system.

Limitations of this study include the limited coverage of the ddPCR method compared to metagenomic next-generation sequencing (mNGS), as some bacteria were not detectable. The study was conducted at a single center, limiting its generalizability. The predicted drug sensitivity of multi-drug-resistant A. baumannii was suboptimal in this system. As this was an observational study without intervention treatments, the clinical benefits of ddPCR could not be accurately evaluated. The correlation between quantitative detection and disease progression remains unclear and requires further investigation.