A diagnostic stewardship bundle approach to facilitate same-day optimal antimicrobial therapy in gram-negative sepsis: a quasi-experimental study

Abstract

It is challenging to reduce time to optimal antimicrobial therapy in patients with sepsis for microbiology laboratories in low to middle-income countries. We evaluated the diagnostic accuracy and impact of a diagnostic stewardship bundle titled “Sepsis-24” to reduce turnaround time (TAT) of provisional blood culture reports (pBCR) ≤ 24 h in patients with gram-negative bacteraemia. During the preintervention period (January-May 23), the key preanalytical and analytical parameters of automated blood culture diagnostics were optimized in a multiphasic manner. Early microbial identification and susceptibility testing were performed by direct inoculation of VITEK-2 identification cards from flagged blood culture bottles (+ BCs) and EUCAST RAST method, read at 8-hour. During the intervention period (June-December 2023), Sepsis-24 was implemented in adult ICUs to provide pBCR for four RAST reportable gram-negatives (RRGNs). The agreements of direct microbial identification and RAST for tested drug-bug combinations were 94% [95%CI: 90–98] and 93% [95%CI: 91–94], respectively during both periods. There was a statistically significant reduction in BC loading, unloading and performance of direct VITEK/RAST from + BC during intervention period [median (minutes): 32 versus 25, 12 versus 2, and 181 versus 70, p ≤ 0.001, respectively]. Of 49 pBCRs released, 48 (98%) were concordant in species-level microbial identification with a median TAT of 1473 min [IQR: 1635 − 1321], from sample receiving. Sepsis-24 facilitated early review of antimicrobial regimen in 71% (34/48) patients leading to therapy change in 64.7% (22/34) patients. Sepsis-24 was found to be diagnostically accurate and facilitated early review of antimicrobial therapy in our resource-limited setting.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-46677-6.

Introduction

Gram-negative sepsis (GNS) is a leading cause of mortality worldwide, with the highest burden in low- and middle-income countries (LMICs)¹. Each hour delay in administering appropriate antimicrobial therapy increases the odds of mortality by 9% in patients with sepsis². Unfortunately, the microbiological diagnostic tools necessary for reducing the time to optimal antimicrobial therapy (TOAT) are often unavailable or underutilised in LMICs because of inadequate infrastructure, an insufficient workforce, and high costs. Consequently, it promotes empiricism in broad-spectrum antimicrobial usage, contributing directly or indirectly to antimicrobial resistance (AMR).

Rapid phenotypic AST approaches that reduce the clinical reporting turnaround time (TAT) to ≤ 24 h after blood culture bottle flagging (+ BC) have resulted in a reduction in actionable TOATs^3,4, shorter hospital stays^3,5, reduced mortality risk⁶ and cost savings⁵ in observational studies conducted in high-income settings. However, they employed newer AST systems, such as ACCELERATE PHENO (Accelerate Diagnostics, USA), ALFRED 60/AST (Alifax, Italy), paired with rapid microbial identification by mass spectrometry, technologies scarcely available in LMICs. A recent network meta-analysis revealed a survival benefit when rapid diagnostic tests on + BCs or direct blood samples were paired with antimicrobial stewardship⁷. Reducing TOAT is challenging for clinical microbiologists in resource-limited settings, where the testing process utilises culture-dependent methods such as conventional biochemicals for microbial identification, Kirby-Bauer disk-diffusion AST, or previous generation automated systems such as VITEK-2 (bioMérieux, France), BD PHOENIX (Becton Dickinson, USA) or MICROSCAN (Beckman Coulter, USA).

Diagnostic stewardship (DS) emphasises appropriate testing and accurate and timely delivery of results to clinicians, thereby supporting efficient patient management⁸. Implementing the bundle-care approach is a key strategy for efficient infection control and antimicrobial stewardship programs; however, its usefulness in facilitating DS is rarely studied^9–11. A bundle is a set of 3–5 evidence-based interventions that result in improved outcomes when implemented concertedly. We formulated a diagnostic stewardship bundle (DSB) titled “Sepsis-24” to provide provisional blood culture reports (pBCRs) for patients with gram-negative bacteraemia. We conducted this study to evaluate the diagnostic accuracy and impact of this DSB to facilitate early review of antimicrobial therapy in critical care settings.

Materials and methods

Study design and setting

This study was conducted in an academic, tertiary care hospital in India between January and December 2023 at the bacteriology laboratory and intensive care units (ICUs), namely, the 30-bed medical ICU (MICU), the 6-bed respiratory ICU (RICU) and the 4-bed medicine high dependency unit (MHDU). It was a prospective, quasi-experimental study done in two periods, preintervention (P1, January–May 23), and intervention period (P2, June–December 23). During P1, the testing components of the tool i.e., direct microbial identification using VITEK was validated on BC in a multiphasic manner (Fig. 1). During P2, the Sepsis-24 DSB tool was formulated and implemented in the study locations and pBCRs were issued based on the tool. The Institutional Human Ethics Committee (IHEC) of AIIMS Bhopal approved the study vide letter no: IHEC-LOP/2022/IL072. All methods were carried out in accordance with relevant guidelines and regulations. Informed consent was obtained from subjects or their legal representatives, as applicable, where antimicrobial therapy was changed based on pBCR.

Fig. 1.

Flowchart depicting the multiphasic progression of the study. BC blood culture bottle, RAST rapid antimicrobial susceptibility testing, RRGN RAST reportable gram-negative, +nBC positively flagged blood culture bottle showing gram-negatives, DSB diagnostic stewardship bundle, ICUs intensive care units, pBCR provisional blood culture report.

Diagnostic stewardship intervention

We developed the ‘Sepsis-24’ DSB clinical decision support tool with the goal of providing pBCRs ≤ 24 hours of sample collection by optimising the key preanalytical and analytical parameters of automated blood culture diagnostics (aBCD), as described in Table 1. The key step of early microbial identification via direct VITEK (dID) was validated during P1 in 3 phases (Fig. 1). During phase 1 and Phase 2 of P1, the BCs were spiked with standard bacterial strains and processed for direct identification by single and multiple laboratory operators, respectively. During the Phase 3 of P1, the validation was performed on BCs received from any clinical area for routine diagnosis; however, no pBCR was released using the Sepsis-24 tool during this phase.

Table 1.

Components of Sepsis-24 diagnostic stewardship bundle.

S. No.	Title	Intervention rationale	Parameter
	Sepsis-24 diagnostic stewardship bundle	To provide provisional blood culture reports within 24 h of sample collection	Time to provisional report (TAT1)
	Clinical arm
Bundle 1	Using chlorhexidine-based skin antisepsis for sample collection	To reduce the contaminating organisms in BCs	% contamination rate
Bundle 2	Collection of at least 1 set of paired blood culture samples	To increase the diagnostic yield of BCs	% Paired Blood culture events
Bundle 3	Collection of blood sample before start or administering next dose of antibiotic	To increase the diagnostic yield of BCs	None
Bundle 4	Transporting BC to laboratory ≤ 30 min of sample collection	To reduce the intervening time duration in processing of BCs	Transportation Time (TT)
	Laboratory Arm
Bundle 5	Placement of BC in CBCMS ≤ 30 min of sample receiving	To reduce the intervening time duration in processing of BCs	Loading Time (LT)
Bundle 6	BC unloading within 30 min of positive signal by CBCMS		Unloading Time (UT)
Bundle 7	Reporting Gram stain from + BC within 30 min of BC unloading		Time to Gram stain (TTG)
Bundle 8	Performing direct identification by VITEK and RAST from + BC within 60 min of gram-stain reporting		Time to direct processing (TTD)
Bundle 9	To complete provisional testing within ≤ 24 h of sample receiving		Time to result completion (TTRC)

BC blood culture, CBCMS continuous blood culture monitoring system, RAST rapid antimicrobial susceptibility testing.

After the diagnostic accuracy of the direct microbial identification and AST methods was analysed during P1, the Sepsis-24 DSB was formulated in consultation with multidisciplinary stakeholders, including critical care and laboratory physicians. The DSB was implemented for BC samples from patients admitted to the MICU, RICU and MHDU. The printed DSB was attached to the clinical requisition form of BC samples to serve as a reminder for the clinical and laboratory personnel to adhere to the components of the testing bundle and to mention the time stamps of the various steps involved in aBCD from sample collection to clinical reporting. During P2, training was given to resident doctors and laboratory personnel in June month. Subsequently, it was implemented in the diagnostic pathway of patients with sepsis admitted to the ICUs from July-December in Phase 4.

Microbiological methods

Blood culture bottles (BCs) were processed in the bacteriology laboratory in BACT/ALERT-3D (bioMérieux, France) system followed by VITEK-2 COMPACT for microbial identification and AST. The laboratory functions around the clock, with resident doctors available for reporting during emergencies (5 pm–9 am). The standard-of-care (SoC) method for processing all flagged BCs (+ BCs) involves subculturing them on conventional plated media, such as chocolate agar, sheep blood agar, and MacConkey agar (HIMEDIA, India), and then incubating them aerobically for 16–24 h, followed by automated identification and AST. The results of the SoC method were considered the “reference” for the subsequent evaluation of the diagnostic accuracy of the tool. Before this study, the laboratory routinely provided pBCRs for Escherichia coli, Klebsiella pneumoniae, Pseudomonas aeruginosa and Acinetobacter baumannii complex, the four EUCAST rapid antimicrobial susceptibility testing (RAST) reportable GNs (RRGNs), which were identified via conventional biochemicals inoculated with bacterial pellets isolated from flagged blood culture bottles showing gram-negative bacteria (+ nBC), and direct antimicrobial susceptibility testing (AST) results were read after 16–20 h of incubation by the EUCAST RAST method.

In Sepsis-24, early microbial identification was performed by inoculating the GNID (bioMérieux, France) card in VITEK from the bacterial pellet isolated from + nBC, as described by Vishwakarma et al.¹². For spiking experiments during phases 1 and 2, E. coli ATCC 25922, K. pneumoniae BAA 700603, P. aeruginosa ATCC 27853 and a clinical laboratory strain of Acinetobacter baumannii complex were used for spiking along with 10 ml of sheep blood. +nBCs were processed for direct identification using VITEK-2 COMPACT after flagging. Briefly, 5 ml of the blood-broth mixture was transferred to a serum separator tube, which was centrifuged at 160 × g for 10 min to separate the human cells. The supernatant was transferred to a sterile plain (red top) blood collection vacutainer and centrifuged at 2000 × g for 10 min to pellet the bacterial cells. The supernatant was carefully discarded, and from the pellet, the turbidity was adjusted to 0.47–0.63 McFarland in a VITEK polystyrene tube (bioMérieux, France) containing VITEK saline (bioMérieux, France), which was subsequently used to inoculate the GNID card. Purity was checked for each processed inoculum tube after the GNID card was loaded by subculturing on cystine lactose electrolyte-deficient (CLED) media (HIMEDIA, India).

To perform direct AST from + nBC, the EUCAST RAST method was followed¹³. Briefly, 125 ± 25 µL of blood-broth inoculum was aspirated in a 1 mL syringe, and lawn culture was performed on Mueller‒Hinton agar (MHA, HIMEDIA, India) plates. Antibiotic discs (HIMEDIA, India) were applied to inoculated plates and subsequently incubated at 35 ± 2 °C, viz., piperacillin-tazobactam (30–6 µg), ceftazidime (10 µg), meropenem (10 µg), imipenem (10 µg), ciprofloxacin (5 µg), levofloxacin (5 µg), amikacin (30 µg), gentamicin (10 µg), and trimethoprim-sulphamethoxazole (1.25–23.75 µg). Zone diameters (ZDs) after 8 h of incubation were read and interpreted as per species-specific breakpoint tables, as per the RAST standard¹⁴. For dID/RAST, weekly quality control testing was performed with E. coli ATCC 25922 as per the RAST standard¹⁵.

Clinical reporting

During phase 4, only the + nBC received from the study ICUs were processed according to the Sepsis-24 tool and pBCRs were issued if an RRGN was identified in dID and ZDs were readable at 8 h on RAST plates. The pBCR was identified as ‘provisional’, and any antimicrobial change after it was at the clinicians’ discretion. To assess the change in antibiotic prescription during combination empirical antimicrobial therapy, a beta-lactam antimicrobial was considered ‘pivotal’, whereas a drug of another antimicrobial class was considered a ‘companion’¹⁶. Antimicrobial de-escalation was defined as either a) replacing the broad-spectrum antimicrobial with a narrow-spectrum agent or having a lower ecological impact or b) stopping components of an antimicrobial combination therapy¹⁶. The hierarchy of the spectrum and resistance-promoting potential of β-lactams was based on the consensual ranking by Weiss et al.¹⁷. Clinical isolates reported in pBCR were considered multidrug resistant (MDR), extensively drug resistant (XDR), or pan-drug resistant (PDR) according to the consensus definition¹⁸ after availability of susceptibility testing results by SoC method.

Data collection

Compliance with the clinical arm of the tool was monitored, and to assess its impact, the area wise BC contamination rate (CR) and paired BC events (PBE) were calculated and compared between phases 3 and 4. Common commensals in blood cultures were identified according to the 2023 NHSN laboratory-confirmed bloodstream infection checklist¹⁹. The intervention durations for the laboratory arm were calculated as described in Table 2. The time stamps available from the automated systems for calculating loading time (LT), unloading time (UT) and time to performance of direct VITEK (TTD) were available, and these durations were compared for all received BCs between phases 3 and 4, along with the proportion of compliance to the specified targets. The data were managed and analysed in MICROSOFT EXCEL 365.

Table 2.

Time taken for the intervening steps of automated blood culture diagnostics during Phases 3 & 4.

Intervening steps	Calculation	Duration Median, minutes (IQR)			Compliance % (n/d)
		Phase 3	Phase 4	p-value1	Phase 3	Phase 4	p-value2
TT ≤ 30 min	Time of BC receiving- Time of sample collection	--	230 (410–120)		--	2.7 (47/1147)
LT ≤ 30 min	Time of BC loading in BACT/ALERT- Time of BC receiving	32 (85 − 11)	25 (49 − 11)	0.001	49.5 (163/329)	58.8 (1036/1763)	0.001
UT ≤ 30 min	Time of BC unloading- Time of BCB flagging by BACT/ALERT	12 (43 − 2)	2 (10 − 1)	< 0.00001	67.9 (76/112)	87.6 (440/502)	< 0.00001
TTG ≤ 30 min	Time of Gram stain reporting- Time of BC unloading	--	32 (54–20)		--	47.2 (237/502)
TTD ≤ 60 min	Time of performance of dID/RAST- Time of flagging	181 (243 − 105)	70 (139 − 41)	< 0.00001	8.3 (2/24)	43.4 (46/106)	< 0.00001
TTRC ≤ 24 h	Time of Result completion by dID/RAST- Time of BC receiving	--	1358 (1547–1211)		--	68.75 (33/48)
TAT 1 ≤ 24 h	Time of providing provisional blood culture report to the clinicians - Time of sample collection	--	1510 (1650 − 1321)		--	40 (10/25)
TAT 2 ≤ 24 h	Time of providing provisional blood culture report to the clinicians - Time of sample receiving in lab	--	1473 (1635 − 1321)		--	48 (23/48)

IQR interquartile range, n numerator, d denominator, TTT time to transport of blood culture bottle, min minutes, BC blood culture, LT loading time, TTU time to bottle unload, TTG time to gram-stain, TTD time to dID/RAST performance, dID direct microbial identification by VITEK, RAST rapid antimicrobial susceptibility testing TTRC, time to dID/RAST reading completion, hr hour, TAT turnaround time.

¹Mann-Whitney U test.

²Chi-square test.

Statistical analysis

Polymicrobial +nBCs were excluded from the analysis. To estimate the diagnostic accuracy of the dID method, the exact identification (till species) with SoC was considered ‘concordant’ and other results as ‘discordant’. The agreement rate of dID was calculated as the proportion of results which were concordant per total no. of isolates tested. The disagreement rate was calculated as the proportion of discrepant results between the two methods per total number of isolates tested. The subset analysis was done for RRGNs and non-RRGNs, as a group and species-wise.

For AST, the results categorised under ‘area of technical uncertainty (ATU)’ in RAST or SoC were excluded from analysis, as the EUCAST RAST standard prohibits their reporting¹³. The results were in categorical agreement (CA) when the interpretive category of a particular drug-bug result was agreed upon by both methods. Disagreements were further categorised as major error (ME), very major error (VME), or minor error (mE), and their rates were calculated²⁰. ME% = number of major errors by RAST*100/number of sensitive results reported by SoC, VME% = number of very major errors by RAST*100/number of resistant results reported by SoC. The positive predictive value (PPV) and negative predictive value (NPV) of the RAST for each tested drug-bug combination at the 8-hour reading time were also calculated as follows: PPV% = all concordant susceptible results*100/all susceptible results reported by the RAST; NPV% = all concordant resistant results*100/all resistant results reported by the RAST. Mann‒Whitney U tests and chi‒square (χ²) tests were used for continuous and categorical variables, respectively, considering a two-sided P value of ≤ 0.05 as statistically significant.

Results

The organism wise results for dID accuracy during P1 are shown in Table 3, whereas the detailed results of dID and RAST during Phase 3 are shown in Supplementary Tables S1 and S2, respectively. All spiked blood cultures processed by dID were concordant for all four RRGN strains during Phases 1 & 2 (Table 3). As shown in Fig. 2, during Phase 3, 166 + nBCs were processed with direct testing protocols, 20 (12%) of which were polymicrobial and excluded from analysis. Among the remaining 146 + nBCs, RRGN and non-RRGN were identified in 77 (52.7%) and 69 (47.3%) +nBCs, respectively, by the SoC method. The overall agreement rates in identification by dID method for RRGNs and non-RRGNs were 93.5% [95% CI: 85.5–97.9] and 65.2% [95% CI: 52.8–72.3], respectively.

Table 3.

Accuracy of direct identification by VITEK-2 COMPACT for RAST reportable gram-negatives in all phases.

Phases of validation of direct microbial identification	Escherichia coli1 % (n/d) [95% CI]	Klebsiella pneumoniae1 % (n/d) [95% CI]	Acinetobacter baumannii complex1 % (n/d) [95% CI]	Pseudomonas aeruginosa1 % (n/d) [95% CI]	All % (n/d) [95% CI]
Phase 11 (January 23)	100 [5/5]	100 [5/5]	100 [5/5]	100 [5/5]	100 (20/20) [83.2–100]
Phase 21 (January 23)	100 [5/5]	100 [5/5]	100 [5/5]	100 [5/5]	100 (20/20) [83.2–100]
Phase 3 (January-April 23)	92.1 (35/38) [78.6–98.3]	94.4 (17/18) [72.7–99.9]	91.7 (11/12) [61.5–99.8]	100 (9/9) [66.7–100]	93.5 (72/77) [85.5–97.9]
Phase 4 (June-December 23)	94.4 (17/18) [72.7–99.9]	80 (4/5) [28.4–99.5]	96.4 (27/28) [81.6–99.9]	--	94.1 (48/51) [83.8–98.8]
Combined results for Phase 3 & Phase 4	92.9 (52/56) [82.7–98]	91.3 (21/23) [72-98.9]	95 (38/40) [83.1–99.4]	100 (9/9) [66.7–100]	93.7 (120/128) [88.1–97.2]
Median time to identification, in hours (IQR)	4.8 (5.8–3.9)	4 (6.9–3.8)	5.7 (7.2–4.8)	5.8 (8.9–4.9)	4.9 (6.8–4.6)

n numerator, number of +nBCs having concordant identification with standard method; d, denominator, total number of +nBCs processed for direct identification in VITEK-2 COMPACT; CI, confidence interval; IQR, interquartile range.

Fig. 2.

Flowchart of the study depicting the results of direct testing method during Phase 3 and 4 in comparison to the standard of care (SoC) method. dID, direct microbial identification by VITEK; +nBC, flagged blood culture bottles with gram-negatives, RAST reportable gram-negative either of Escherichia coli, Klebsiella pneumoniae, Pseudomonas aeruginosa or Acinetobacter baumannii complex; non-RRGN, any micro-organism other than RRGN.

The diagnostic accuracy of 8-hour breakpoints by RAST was assessed in 69 concordant RRGNs, including 33 (47.8%) E. coli, 17 (24.6%) K. pneumoniae, 11 (15.9%) A. baumannii complex and 8 (11.6%) P. aeruginosa, leading to 583 drug-bug reading ZDs. Overall, of the 557 bug-drug combinations, after 26 (4.5%) ZDs in the ATU interpretive category were excluded, the CA, ME and VME rates were 93% [95% CI: 90.5–95], 5.1% [95% CI: 2.2–9.8], and 6.1% [95% CI: 4–9], respectively, as shown in Supplementary Table S2.

After establishing the diagnostic accuracy of the tool, we implemented it in the study ICUs during Phase 4. The details of the BCs received from each study ICU and their compliance with various components of the clinical arm of the tool are provided in Supplementary Table S3. There was a nonsignificant reduction in CR during P2 (phase 3 vs. 4: 14.3 [95% CI: 10.7–18.5] vs. 12 [95% CI: 10.5–13.6], p = 0.24, χ² test). In contrast, the PBE increased significantly during P2 (7.2 [95% CI: 4.5–10.6] vs. 81.6 [95% CI: 79.3–84.1], p < 0.00001, χ² test). The time taken for the intervening steps of aBCD, including LT, UT, and TTD, during phases 3 and 4 is shown in Table 2, which shows a statistically significant reduction and corresponding increase in compliance to achieve the target duration during phase 4.

During phase 4, the flow of performance of the laboratory arm of the DSB is shown in Fig. 2. The detailed results of dID/RAST during Phase 4 are shown in Supplementary Tables S4 and S5, respectively. During Phase 4, 106 + nBCs were processed with direct testing protocols, of which 21 (19.8%) were polymicrobial and excluded from analysis. Among the remaining 85 + nBCs, RRGN and non-RRGN were identified in 51 (60%) and 34 (40%) +nBCs, respectively, by the SoC method. The agreement rate of identification of an RRGN by dID was 94.1% [95% CI: 83.8–98.8], whereas the disagreement rate was, as shown in Supplementary Table S4. For AST, the diagnostic accuracy was assessed in 48 concordant RRGNs, including 17 (47.8%) E. coli, 4 (24.6%) K. pneumoniae and 27 (15.9%) A. baumannii complex, leading to 378 drug-bug reading ZDs. Overall, of the 361 bug-drug combinations, after 17 (4.5%) ZDs in the ATU interpretive category were excluded, the CA%, ME% and VME% values were 92.8% [95% CI: 88.3–94.3], 5.8% [95% CI: 1.2–15.9] and 6.2% [95% CI: 3.8–9.6], respectively.

The combined results of phases 3 and 4 for dID and RAST are shown in Tables 3 and 4, respectively. The results were comparable during both phases in terms of dID and RAST accuracy. Overall, of the 272 dIDs/RASTs performed, an RRGN was identified in 128 (47%) + BCs by the SoC method, of which 120 (93.7%) were concordant in dID. Among the 8 discordant results, 3 were not identified, and 5 + nBC were misidentified via the dID method. Among the 56 E. coli strains, 2 were misidentified as Salmonella serotype diarizonae and A. hemolyticus. One each + nBC containing K. pneumoniae and A. baumannii complex was misidentified as Ralstonia pickettii and Cupriavidus pauculus, respectively. The possibility of incorrect reporting due to identification as an RRGN occurred in 3 instances where in one + nBC each with Salmonella species, Sphingomonas paucimobilis and A. lwoffi, the identification in dID was E. coli, K. pneumoniae ssp. ozaenae and A. baumannii complex, respectively. On subset analysis, we found that the agreement rate for the identification of an RRGN by dID was significantly greater than that for a non-RRGN (RRGN vs. non-RRGN; 94.1% [95% CI: 88.1–97.3] vs. 60.2% [95% CI: 50.1–69.7], p < 0.00001, χ² test).

Table 4.

Diagnostic accuracy (CA, ME and VME) of EUCAST RAST method (version 6.1) in comparison to antimicrobial susceptibility testing by VITEK-2 COMPACT (interpreted as per EUCAST) at 8-hour during both Phase 3 and Phase 4.

Variables1	TZP	CAZ	MEM	IMP	AMIK	GEN	CIP	LEV	TRI-SUL	ALL (n/d) [95% CI]
Escherichia coli (50)
CA (%) (n/d)	90.2 (37/41)	91.1 (41/45)	84.8 (39/46)	97.8 (45/46)	94 (47/50)	91.8 (45/49)	97.8 (44/452)	93.8 (45/48)	94 (47/50)	92.8 (390/420) [90-95.1]
ME (%) (n/d)	6.7 (1/15)	0 (-/1)	0 (-/29)	0 (-/29)	2.9 (1/34)	3.8 (1/26)	- (-/-)	0 (-/1)	11.1 (1/9)	2.8 (4/144) [0.8-7]
VME (%) (n/d)	11.5 (3/26)	9.1 (4/44)	0 (-/10)	5.9 (1/17)	12.5 (2/16)	13 (3/23)	2.2 (1/45)	0 (-/44)	4.9 (2/41)	6 (16/266) [3.5–9.6]
mE (%) (n/d)	0 (-/41)	0 (-/45)	15.2 (7/46)	0 (-/46)	0 (-/50)	0 (-/49)	0 (-/45)	6.3 (3/48)	0 (-/50)	2.4 (10/420) [1.1–4.3]
ATU (%)	18 (9/50)	10 (5/50)	8 (4/50)	8 (4/50)	0 (-/50)	2 (1/50)	2 (1/50)	4 (2/50)	0 (-/50)	5.8 (26/446) [3.8–8.4]
PPV (%)	82.3 (14/17)	20 (1/5)	96.7(29/30)	96.7 (29/30)	94.3 (33/35)	89.3 (25/28)	0 (-/1)	25 (1/4)	80 (8/10)	87.5 (140/160) [81.4–92.2]
NPV (%)	95.8 (23/24)	100 (40/40)	62.5 (10/16)	100 (16/16)	93.3 (14/15)	95.2 (20/21)	100 (44/44)	100 (44/44)	97.5 (39/40)	96.1 (250/260) [93-98.1]
Klebsiella pneumoniae (21)
CA (%) (n/d)	85 (17/20)	100 (20/20)	100 (21/21)	100 (20/20)	100 (19/19)	90.5 (19/21)	90 (18/20)	94.1 (16/17)	95.2 (20/21)	93.8 (168/179) [89.3–96.9]
ME (%) (n/d)	0 (-/2)	- (-/-)	0 (-/5)	0 (-/5)	0 (-/4)	0 (-/5)	- (-/-)	- (-/-)	25 (1/4)	4 (1/25) [0.1–20.3]
VME (%) (n/d)	16.7 (3/18)	0 (-/20)	0 (-/16)	0 (-/15)	0 (-/15)	12.5 (2/16)	10 (2/20)	0 (-/17)	0 (-/17)	3.9 (6/153) [1.4–8.3]
mE (%) (n/d)	0 (-/20)	0 (-/20)	0 (-/21)	0 (-/20)	0 (-/19)	0 (-/21)	0 (-/20)	5.9 (1/17)	0 (-/21)	0.5 (1/179) [0-3.1]
ATU (%)	4.8 (1/21)	4.8 (1/21)	0 (-/21)	4.8 (1/21)	9.6 (2/21)	0 (-/21)	4.8 (1/21)	19 (4/21)	0 (-/21)	5.3 (10/189) [2.6–9.5]
PPV (%)	40 (2/5)	- (-/-)	100 (5/5)	100 (5/5)	100 (4/4)	71.4 (5/7)	0 (-/2)	0 (-/1)	100 (3/3)	75 (24/32) [56.6–88.5]
NPV (%)	100 (15/15)	100 (20/20)	100 (16/16)	100 (15/15)	100 (15/15)	100 (14/14)	100 (18/18)	100 (16/16)	94.4 (17/18)	99.3 (146/147) [96.3–100]
Acinetobacter baumannii complex (38)
CA (%) (n/d)	NA	NA	100 (38/38)	100 (38/38)	64.9 (24/37)	89.2 (33/37)	97.4 (37/38)	97.4 (37/38)	86.5 (32/37)	90.9 (239/263) [86.7–94.1]
ME (%) (n/d)	NA	NA	0 (-/1)	0 (-/1)	0 (-/1)	0 (-/2)	50 (1/2)	50 (1/2)	33.3 (4/12)	28.6 (6/21) [11.3–52.2]
VME (%) (n/d)	NA	NA	0 (-/37)	0 (-/37)	36.1 (13/36)	11.4 (4/35)	0 (-/36)	0 (-/36)	4 (1/25)	7.4 (18/242) [4.5–11.5]
mE (%) (n/d)	NA	NA	0 (-/38)	0 (-/38)	0 (-/37)	0 (-/37)	0 (-/38)	0 (-/38)	0 (-/37)	0 (-/263) [0-1.4]
ATU (%)	NA	NA	0 (-/38)	0 (-/38)	2.6 (1/38)	2.6 (1/38)	0 (-/38)	0 (-/38)	2.6 (1/38)	1.1 (3/266) [0.2–3.2]
PPV (%)	NA	NA	100 (1/1)	100 (1/1)	7.1 (1/14)	33.3 (2/6)	100 (1/1)	100 (1/1)	88.9 (8/9)	45.4 (15/33) [28.1–63.6]
NPV (%)	NA	NA	100 (37/37)	100 (37/37)	100 (23/23)	100 (31/31)	97.3 (36/37)	97.3 (36/37)	85.7 (24/28)	97.4 (224/230) [94.4–99]
Pseudomonas aeruginosa (8)
CA (%) (n/d)	100 (8/8)	87.5 (7/8)	100 (8/8)	100 (7/7)	87.5 (7/8)	NA	100 (8/8)	100 (8/8)	NA	96.4 (53/55) [87.5–99.6]
ME (%) (n/d)	0 (-/2)	0 (-/2)	0 (-/3)	0 (-/3)	0 (-/3)	NA	0 (-/3)	0 (-/3)	NA	0 (-/19) [0-17.6]
VME (%) (n/d)	0 (-/6)	16.7 (1/6)	0 (-/5)	0 (-/4)	20 (1/5)	NA	0 (-/5)	0 (-/5)	NA	5.6 (2/36) [0.7–18.7]
mE (%) (n/d)	0 (-/8)	0 (-/8)	0 (-/8)	0 (-/7)	0 (-/8)	NA	0 (-/8)	0 (-/8)	NA	0 (-/56) [0-6.4]
ATU (%)	0 (-/8)	0 (-/8)	0 (-/8)	12.5 (1/8)	0 (-/8)	NA	0 (-/8)	0 (-/8)	NA	1.8 (1/56) [0-9.5]
PPV (%)	100 (2/2)	66.7 (2/3)	100 (3/3)	100 (3/3)	75 (3/4)	NA	100 (2/2)	100 (3/3)	NA	90 (18/20) [68.3–98.8]
NPV (%)	100 (6/6)	100 (5/5)	100 (5/5)	100 (4/4)	100 (4/4)	NA	100 (6/6)	100 (5/5)	NA	100 (35/35) [90–100]
All Isolates (117)
CA (%) (n/d)	90 (62/69)	93.1 (68/73)	93.8 (106/113)	99.1 (110/111)	85.1 (97/114)	90.6 (97/107)	96.4 (107/1112)	95.5 (106/111)	91.7 (99/108)	92.9 (852/917) [91-94.5]
ME (%) (n/d)	5.3 (1/19)	0 (-/3)	0 (-/38)	0 (-/38)	2.4 (1/42)	3 (1/33)	20 (1/5)	16.7 (1/6)	24 (6/25)	5.3 (11/209) [2.7–9.2]
VME (%) (n/d)	12 (6/50)	7.1 (5/70)	0 (-/68)	1.4 (1/73)	22.2 (16/72)	12.2 (9/74)	2.8 (3/106)	0 (-/101)	3.6 (3/83)	6.2 (43/697) [4.5–8.2]
mE (%) (n/d)	0 (-/69)	0 (-/73)	6.2 (7/113)	0 (-/111)	0 (-/114)	0 (-/107)	0 (-/110)	3.6 (4/111)	0 (-/108)	1.2 (11/917) [0.6–2.1]
ATU (%)	12.7 (10/79)	7.6 (6/79)	3.4 (4/117)	5.1 (6/117)	2.6 (3/117)	1.8 (2/109)	1.7 (2/117)	5.1 (6/117)	0.9 (1/109)	4.2 (40/9572) [3-5.6]
PPV (%)	75 (18/24)	37.5 (3/8)	97.4 (38/39)	97.4 (38/39)	71.9 (41/57)	78 (32/41)	57.1 (4/7)	55.5 (5/9)	86.4 (19/22)	80.5 (198/246) [75-85.2]
NPV (%)	97.8 (44/45)	100 (65/65)	91.9 (68/74)	100 (72/72)	98.2 (56/57)	98.5 (65/66)	99 (103/104)	99 (101/102)	93 (80/86)	97.5 (654/671) [96-98.5]

n numerator, d denominator, CA categorical agreement, ME major error, VME very major error, TZP piperacillin-tazobactam, CAZ ceftazidime, IMP imipenem, MEM meropenem, AMIK amikacin, GEN gentamicin, CIP ciprofloxacin, LEV levofloxacin, TRI-SUL trimethoprim-sulphamethoxazole.

Note: If a particular drug-bug combination having ≥ 30 denominator is not having requisite diagnostic accuracy as per CLSI M52, i.e., CA: ≥90%, ME: ≤3% and VME: ≤1.5%) or NPV/PPV is < 90%, it is highlighted in bold letters.

¹Excludes results falling under category of “Area of technical uncertainty, ATU” in RAST or VITEK.

²Four isolates which had “ATU” results in ciprofloxacin by standard of care method (4 in E. coli) were excluded from analysis.

The individual drug-bug rates for CA, ME, VME, mE, PPV and NPV, cumulatively during phases 3 & 4, are provided in Table 4. The RAST readings at 8 h revealed a CA rate of 93% (852/917, 95% CI: 91–94.5) for all tested drug‒bug combinations, whereas the VME and ME rates were 5.3% (11/209) and 6.2% (43/609), respectively. Among the individual drug-bug combinations tested with a denominator of ≥ 30, CA was ≤ 90% for meropenem-Escherichia, amikacin/gentamicin/trimethoprim-sulphamethoxazole-Acinetobacter. For K. pneumoniae (21 isolates) and P. aeruginosa (9 isolates), the CAs were poor for piperacillin-tazobactam and ceftazidime/amikacin, respectively. Unacceptably high VME rates were observed for the ceftazidime/trimethoprim-sulphamethoxazole-Escherichia (9.1%, 4/44; 4.9%, 2/41, respectively) and amikacin/gentamicin-Acinetobacter combinations (36%, 13/36; 11.4%, 4/35, respectively). No individual drug-bug combination had ≥ 30 susceptible isolates, except for amikacin-Escherichia hence, the ME rates for most of the tested combinations were unreliable.

There were 49 pBCRs issued during phase 4, including one reporting error wherein Acinetobacter lwoffi was identified as an A. baumannii complex in dID. Among the 48 concordant pBCRs, 35 (73%), 7 (14.6%) and 6 (12.5%) were issued to the MICU, MHDU and RICU, respectively. The median TAT for providing pBCR to clinicians from the time of receiving the sample in the laboratory was 1473 min (24 h, 23 min). The median durations to complete the testing by the Sepsis-24 DSB tool and SoC methods were 1358 [IQR: 1547 − 1211] and 3746 [IQR: 4059 − 3392] minutes, corresponding to 22.6 and 62.4 h, respectively. The targeted end-to-end TAT (from sample collection to report receipt) of ≤ 24 h was calculated in 25 patients for whom the sample collection time was known. The median TAT was 1510 [IQR: 1650 − 1321] minutes, and the target was met in 40% (10/25) of the patients.

Among the 48 isolates reported in the pBCR, 25 (52%), 19 (43%) and 3 (5%) had PDR, XDR and MDR, respectively. The action taken by the treating clinician after receiving pBCR is described in Fig. 3. Overall, patient outcomes occurred before the pBCR could reach the clinician in 29.2% (14/48) of patients, whereas an opportunity for an early review of the antimicrobial regimen by pBCR was provided in 34 (71%) patients. The antimicrobial regimen was modified in 64.7% (22/34) of patients, whereas it was continued in 12 patients. The majority of VMEs were reported with aminoglycosides (58.1%, 25/43), particularly amikacin-Acinetobacter, leading to the incorrect addition of amikacin in the revised antimicrobial regimen based on pBCR in 2 patients.

Fig. 3.

Algorithm of action taken by the clinical teams after the receipt of the provisional blood culture report (pBCR). DAMA, discharge against medical advice; ab, antibiotic.

Discussion.

This study evaluated the diagnostic accuracy and feasibility of the “Sepsis-24” DSB tool to facilitate early review of antimicrobial therapy in patients with gram-negative bacteraemia. As per bundle, diagnostic testing was completed ≤ 24 h in two-thirds, and clinical reports were provided in half of the patients after sample collection, demonstrating the efficacy of this approach to enable same-day review of empirical antimicrobial therapy. This TAT is comparable to what is possible with new age culture-dependent rapid AST systems²¹, which are a distant possibility for microbiology laboratories in LMICs. Hence, this strategy of expediting microbial identification with direct AST from + BC can serve as a model for resource-limited settings to reduce the TAT associated with aBCDs.

We also demonstrated that the diagnostic approach was accurate in both early microbial identification and AST determination for RRGNs, achieving ~ 94% and 93% agreement during phases ¾ combined (Tables 3 and 4). Most laboratories implementing direct phenotypic AST have used mass spectrometry-based identification after early incubation^4,22 or directly from + BC²³. Few authors in resource-limited settings have used conventional biochemical tests for identification to facilitate EUCAST RAST but could reduce the reporting time to only 24 h post-flagging^24,25. The utility of directly identifying gram-negative bacteria from + BCs using previous-generation automated systems has been demonstrated^26,27. It took < 5 h to identify the RRGNs by the dID approach, which is comparable to other faster approaches, including mass spectrometry-based identification after early incubation²² and the ACCELERATE PHENO²⁸. We found that the dID approach was significantly more accurate in identifying RRGNs than other gram-negative approaches were and perfectly complemented the EUCAST RAST method for early clinical reporting. It solves the conundrum of identifying microorganisms before the RAST results for a resource-limited setting. This will increase the likelihood of adapting direct AST techniques to the workflow of clinical laboratories.

We found that the 8-hr CA rates of the RAST were acceptable; however, the VME% was higher than the specified value (6.5%) than the target of ≤ 1.5%²⁰. Most VMEs were observed for aminoglycosides [58.1%, 95% CI: 42.1–73], especially the amikacin-Acinetobacter combination, where ~ one-third of the isolates (13/36) demonstrated an error leading to incorrect administration of amikacin in the revised antimicrobial regimen in two patients. Peculiarly, we observed that there was a clear inhibition zone against amikacin at 8 h, which disappeared after extending the incubation period to 16–20 h in 71% (10/14) of the A. baumannii complex isolates, indicating that ZD breakpoints at 16–20 h may be more accurate with this drug–bug combination. Pariona et al.²³ also reported that the VME% was high for the tobramycin-Acinetobacter combination, but other studies have reported limited number of Acinetobacter blood stream infections^22,24. Various guidelines on management of carbapenem resistant A. baumannii (CRAB) recommends aminoglycosides only in combination with another active agent such as durlobactam-sulbactam or high dose ampicillin-sulbactam and if susceptibility is demonstrated^29,30. Hence, the reporting of a sensitive result for aminoglycosides in RAST should be met with caution and the laboratory should advise following up with the results at 16–20 h. From a clinician’s perspective, in our study, reporting a resistant isolate in pBCR was highly accurate for all drugs, as NPV rates were > 90% for all tested drug–bug combinations (Table 4). However, the PPV rates were not reliable because of the low proportion of susceptible isolates in our study.

In our setting, there was a very high burden of AMR, as 91.7% (44/48) of pBCRs reported an XDR/PDR strain. The mortality rate associated with these pathogens was very high, as 29% (14/48) of patients died before the pBCR could reach the clinician. For the remaining 34 patients, the antimicrobial regimen was changed in ~ two-thirds (22/34) and escalated in 50% (17/34) of patients, as shown in Fig. 3. This finding underscores the usefulness of LMICs with a high AMR burden, as reports of AMR can help escalate the use of antimicrobial agents. This contrasts with the literature from developed countries, where RAST has proven its utility as a de-escalation strategy^22,31.

Our study underscores the importance of implementing quality improvement measures for aBCD in a clinical microbiology laboratory as well as the benefits of a continuously operating laboratory to enable round-the-clock reporting. Recently, Paluch et al. reported that conversion to full automation and a 24/7 workflow for blood culture diagnostics improved laboratory efficiency, significantly reducing the TAT and time to first antibiotic modification³². There are multiple processing steps in aBCD, and delays in any of them lead to delays in downstream processing. Our bundle approach also aimed to reduce the intervening durations for these steps and reduce the cumulative median duration for sample loading, unloading and dID/RAST processing from 225 to 97 min during Phases 3 and 4, respectively. Even though the clinical arm of the tool could not reduce the transportation of BCs to the laboratory within 30 min, it improved other parameters, such as the proportion of paired blood culture sets and a nonsignificant reduction in contamination rates.

This study has several strengths. First, the use of direct microbial identification using VITEK to enable phenotypic direct AST for clinical reporting is novel and is comparable to the use of mass spectrometry in terms of the TAT. The widespread presence of these previous generation automated systems in LMICs will increase the likelihood of the adoption of direct phenotypic AST methods in resource-limited settings. Second, it demonstrates the feasibility of implementation in a real-world LMIC setting in multiple critical care units without compromising diagnostic accuracy. Third, the tool’s multiphasic validation strategy involving ≥ 100 isolates provide scientific rigour to prove the accuracy of the reported results.

The limitations of this study include that it was a single-centre study, and generalizability is limited. Secondly, due to unavailability of mass-spectrometry in our setting, we relied on microbial identification by VITEK as both “test” and “comparator” method which has the potential of introducing incorporation bias in the results. Thirdly, pBCRs were issued in only approximately 50% of instances where dID/RAST was performed for various reasons, such as polymicrobial cultures, procedural failures and RAST breakpoints validated for RRGNs only. By excluding such results from the denominator, the test performance could be an overestimation. Fourth, the tool facilitated a review of the antimicrobial regimen in one-third of patients and an actual change in only one-fifth of patients, indicating the financial burden the strategy would impose on diagnostic services. Fifthly, we did not calculate the exact time of change of optimal antimicrobial therapy after the clinical team received the provisional report; however, all changes in antimicrobial therapy, shown in Fig. 3, were between the receipt of the provisional report by Sepsis-24 and the final report by the SoC method. Sixthly, we could not assess the impact of the tool on patient outcomes because of limited issued reports and lack of comparator group for patient centred clinical outcomes.

Nevertheless, we suggest specific measures for even more effective implementation in resource-limited settings. To reduce overprocessing, we suggest performing direct testing only if the BC flags within 24 h, as most RRGNs flag within this duration. Second, skin disinfection should be aggressively pursued to reduce the proportion of polymicrobial cultures. Third, the clinician should be made aware of specific drug–bug combination testing errors. Future research should evaluate the direct identification approach using other previous generation automated systems, such as BD PHOENIX and MICROSCAN WALKAWAY. Additionally, the evaluation should be performed on bacteraemia due to gram-positive microorganisms and in multicentric, experimental study designs to assess the impact on patient outcomes.

To conclude, GNS is a medical emergency in which clinical microbiologists must overcome the challenge of providing clinical reports promptly to facilitate antimicrobial stewardship, especially in LMICs. This study underscores the importance of implementing DS interventions in resource-limited clinical microbiology laboratories. Clinical reporting by direct identification and antimicrobial susceptibility testing significantly reduces TOAT. Direct microbial identification using previous-generation automated systems perfectly complements the direct phenotypic AST based on Kirby–Bauer disk diffusion in resource-limited settings.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

AG: Conceptualisation, Methodology, Data curation and analysis, Writing-original draft, project administration. SS: Conceptualisation, project administration, patient management, Writing-reviewing & editing. SK: Conceptualisation, Project administration, Patient management, Writing-reviewing & editing. SP: Conceptualisation, Methodology, Writing-original draft. AGo: Conceptualisation, Project administration, Patient management, Writing-reviewing & editing. AK: Project administration, Patient management, Writing-reviewing & editing. PKK: Data Curation and analysis, Writing-reviewing & editing. All authors approve the final draft of submitted manuscript.

Funding

The study was funded by the intramural research grant given to Dr Ayush Gupta by the parent institute i.e. All India Institute of Medical Sciences (AIIMS) Bhopal. The part of the work was presented in the ESCMID Global 2024 as poster presentation and published in the ESCMID Global 2024 Abstract book available at 10.1016/j.cmicom.2024.100013. Partial results of the study are published in the article available at https://doi.10.3791/66588.

Data availability

The data will be submitted to the online repository. Full access to the data will be available on request.

Declarations

Competing interests

The authors declare no competing interests.