A phase 2 randomized, placebo-controlled trial of inulin for the prevention of gut pathogen colonization and infection among patients admitted to the intensive care unit for sepsis

Abstract

Background

Patients admitted to the intensive care unit (ICU) often have gut colonization with pathogenic bacteria and such colonization is associated with increased risk for death and infection. We conducted a trial to determine whether a prebiotic would improve the gut microbiome to decrease gut pathogen colonization and decrease downstream risk for infection among newly admitted medical ICU patients with sepsis.

Methods

This was a randomized, double-blind, placebo-controlled trial of adults who were admitted to the medical ICU for sepsis and were receiving broad-spectrum antibiotics. Participants were randomized 1:1:1 to placebo, inulin 16 g/day, or inulin 32 g/day which were given for seven days. The trial primary outcome was a surrogate measure for gut colonization resistance, namely the within-individual change from ICU admission to Day 3 in the relative abundance of short chain fatty acid (SCFA)-producing bacteria based on rectal swabs. Additional outcomes sought to evaluate the impact of inulin on the gut microbiome and downstream clinical effects.

Results

Ninety participants were analyzed including 30 in each study group. There was no difference between study groups in the within-individual change in the relative abundance of SCFA-producing bacteria from ICU admission to ICU Day 3 (placebo: 0.0% change, IQR − 8·0% to + 7·4% vs. combined inulin: 0·0% change, IQR − 10·1% to + 4·8%; p = 0·91). At end-of-treatment on ICU Day 7, inulin did not affect SCFA-producer levels, microbiome diversity, or rates of gut colonization with pathogenic bacteria. After 30 days of clinical follow-up, inulin did not affect rates of death or clinical, culture-proven infection. Patients who died or developed culture-proven infections had lower relative abundance of SCFA-producing bacteria at ICU admission compared to those who did not (p = 0·03).

Conclusions

Prebiotic fiber had minimal impact on the gut microbiome in the ICU and did not improve clinical outcomes.

Trial registration

Clinicaltrials.gov: NCT03865706.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13054-024-05232-3.

Keypoints

Question

Does prebiotic fiber alter the gut microbiome in the medical ICU to decrease gut pathogen colonization and decrease downstream risk for infection?

Findings

This randomized controlled trial found that prebiotic inulin at doses up to 32 grams/day did not alter the gut microbiome among adults admitted to the ICU with sepsis. Even at these high doses, inulin was well tolerated. The taxonomic composition of the gut microbiome, assessed using longitudinal rectal swabs and stool samples for up to thirty days of follow-up, was not substantively altered by inulin; also unchanged by inulin were fecal levels of short chain fatty acids, gut colonization with pathogenic bacteria including vancomycin-resistant Enterococcus and multidrug-resistant Gram-negative bacteria, and clinical outcomes including culture-proven infection or death.

Meaning

Pre- and probiotics may face significant challenges in the ICU. Improved understanding of the dynamic changes within the gut microbiome in the ICU may lead to more targeted therapies.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13054-024-05232-3.

Introduction

There is a need for interventions which improve the gut microbiome of patients hospitalized in the medical intensive care unit (ICU) [1]. The normal, native gut microbiome resists the expansion of pathogenic bacteria such as vancomycin-resistant Enterococcus (VRE) and multidrug-resistant Gram-negative bacteria (MDR-GNB) [2]. Depletion of the native gut microbiota—by antibiotics and by critical illness itself—results in loss of normal colonization resistance against these and other pathogens [3]. As a result, 10–15% patients in the ICU are frequently colonized in the gut with VRE and/or with MDR-GNB at the time of ICU admission; another 10–15% of patients test negative for these pathogens at ICU admission and become gut colonized while in the ICU [4–6].

There is significant morbidity and mortality associated with gut pathogen colonization in the ICU [7]. For example, VRE gut colonization at the time of ICU admission is associated with a 50% increased risk for death or all-cause infection, even after adjusting for other patient factors including severity of illness [8]. If microbiome restitution therapies could improve gut colonization resistance against pathogenic bacteria, many ICU-associated infections could be prevented [9].

While it is unknown which features of the native gut microbiota most contribute to gut colonization resistance, increasing evidence points to fiber-metabolizing and short chain fatty acid (SCFA)-producing bacteria as central to the process. Preclinical studies demonstrate the ability of SCFA producers to prevent VRE and other pathogenic bacteria [10, 11] while clinical studies describe profound losses of SCFA producers during the first few days of ICU hospitalization [12–14]. Promisingly, observational studies suggest that SCFA producer levels are increased among ICU patients who receive higher amounts of prebiotic dietary fiber [15].

Drawing on this literature, we designed a clinical trial to test inulin, a prebiotic fiber derived from chicory root, in the medical ICU. We selected inulin because of animal studies showing protection against pathogen colonization [16] and studies in ambulatory patient cohorts showing that a short course of inulin substantially increases gut microbial SCFA production [17, 18]. Inulin has prior evidence of safety at doses up to 34 g/day, is inexpensive, and is easily obtained [19]. The goal of the trial was to directly assess the effects of inulin and, more broadly, to test whether the gut microbiome can be modified in the medical ICU, where patients inevitably receive a diverse array of medical interventions including antibiotics.

Methods

Trial design and oversight

This was a single-center, randomized, double-blind, placebo-controlled, phase 2 trial of inulin for the purpose of preventing gut pathogen colonization and infection in the ICU. Patients were randomized 1:1:1 to placebo, inulin 16 g/day, or inulin 32 g/day beginning on October 14, 2019. Prior to enrollment of the first patient, a data and safety monitoring board (DSMB) was appointed which had full access to the study data. The DSMB reviewed outcomes and adverse effects after enrollment of 10, 30, and 45 patients and at the end of the study and was empowered to halt the trial at any time. The study was approved by the institutional review board of Columbia University Irving Medical Center. It was sponsored by the Department of Defense (DoD) Peer-Reviewed Medical Research Program (PR181960) and approved by the DoD Human Research Protection Office. Trial registration was completed on clinicaltrials.gov on March 4, 2019 prior to enrollment of the first patient (NCT03865706). Once the COVID pandemic began in 2020, all patients were tested for COVID-19 and all those who tested positive were excluded. No other significant changes were made in the trial after its initial registration.

Participants and setting

The trial entry criteria sought to enroll a standard medical ICU population that would generalize to all medical ICUs. Trial enrollment took place within three medical ICUs affiliated with CUIMC. We included adults who were admitted to the ICU with sepsis as the primary diagnosis and who could be enrolled within 24 h of ICU admission. Sepsis was operationalized using the Sepsis-3 (2016) consensus definition [20] as a known or suspected infection plus a Sequential Organ Failure Assessment (SOFA) score ≥ 2 points above baseline [21]. Participants were further required to have received a broad-spectrum antibiotic within the 24 h before enrollment or to have an antibiotic ordered but pending administration such that it would be received prior to the first dose of the study intervention. Eligible broad-spectrum antibiotics included β-lactam/β-lactamase inhibitor combination antibiotics, cephalosporins, fluoroquinolones, lincosamides (clindamycin), metronidazole, and monobactams. Exclusion criteria were any one or more of the following: inability to receive enteral fluids (because the intervention could not be delivered), inulin allergy, serum sodium ≤ 128 mEq/L (because the intervention was delivered in free water), a CD4 count of < 200 or absolute neutrophil count of < 500, surgery involving the intestinal lumen within 30 days, or limited treatment goals such as “do not resuscitate” or “no escalation of care” orders. Written participant or surrogate consent was obtained prior to gathering any samples or data.

Intervention

Patients were randomly assigned 1:1:1 to placebo, inulin 16 g/day, or inulin 32 g/day. The study intervention was given in two doses daily for seven days, via naso- or orogastric tube as a bolus of consecutive 60 mL pushes. When clinically appropriate, all patients received zero-fiber enteral nutrition to minimize any effects on the microbiome from dietary fiber. No patients received parenteral nutrition. The intervention was fully dissolved in 250 mL of sterile water with a zero calorie sweetener to mask flavor and placed into an opaque bottle such that inulin could not be distinguished from placebo. Inulin powder was manufactured from desiccated chicory root by Cargill, Inc. and distributed by Swanson Health and purchased at fair market value by the study investigators; neither Cargill nor Swanson had any role in the study.

Samples

Rectal swabs were used for the main trial analyses because swabs, unlike whole stools, can be precisely timed. Rectal swabs were performed with the participant in the left lateral decubitus position, with fecal soilage used to ensure adequate sampling. Two deep rectal swabs were gathered by a study nurse on all participants at the time of ICU admission and prior to receipt of the first dose of the study intervention. Subsequent rectal swabs were performed on Day 3, 7, 14, and 30 unless patients died or were discharged from the hospital (i.e., patients on the general wards were also swabbed). At each timepoint, one of the two rectal swabs was flash frozen at − 80 °C for 16S rRNA gene sequencing and the second swab was added to soy broth and cultured. Additionally, freely voided stool samples were gathered to measure absolute levels of SCFAs during the intervals between rectal swabs. For example, between Day 0 and Day 3, we collected the first stool sample that was produced in this timeframe, between Day 3 and Day 7 we again collected the first sample from this timeframe, etc. Enrollment was timed to avoid sample collections falling on weekend days.

16S rRNA gene sequencing

Rectal swabs were thawed and DNA was extracted using positive and negative controls on a 96-well plate handling station (Qiagen). From extracted DNA, the V3V4 region of the 16S rRNA gene was amplified using primers with Illumina adapters and the resulting libraries were barcoded and sequenced on an Illumina MiSeq platform at 600 cycles. Sequencing data was processed using DADA2 and R with all analyzed samples meeting a minimum requirement of 7500 reads after quality filtering [22]. Phylogenetic trees were constructed using the MAFFT and FastTree modules in QIIME2. Taxonomic classification was conducted with a native naïve RDP Bayesian classifier against the Silva version 138 database [23]. SCRuB was used downstream to account for potential well-to-well contamination (additional details in Supplemental Methods) [24].

Bacterial culture

Rectal swabs were plated directly onto chromogenic agar selective for VRE, ESBL-producing Gram-negative bacteria, and CRE (carbapenem-resistant Enterobacteriaceae) and incubated overnight at 37 °C. Unique VRE colonies were subcultured onto blood plates and ESBL and CRE colonies were subcultured onto MacConkey agar. Species identification was obtained using full-length 16S rRNA gene sequencing on picked colonies from blood agar plates (VRE) and MacConkey plates (GNB). Disc diffusion susceptibility testing was performed on each isolate to confirm antimicrobial resistance phenotype [25]. Susceptibility results were analyzed alongside sequencing data to ensure both datasets aligned. For our analyses, any isolates testing positive as ESBL or CRE were classified as MDR-GNB.

Direct measurement of fecal SCFAs

Fecal SCFAs were directly measured from whole stools, which were collected in periods corresponding to the rectal swab samples: a stool from ICU admission to Day 3, from ICU Day 3 to Day 7, and so on. Liquid chromatography-mass spectrometry (LC–MS) was performed on flash-frozen and never thawed whole stools as described by Chun et al. and the absolute levels of fecal SCFAs were measured in mMol as the summed total of fecal acetate, butyrate, formate, isobutyrate, isovalerate, propionate, and valerate [26]. Details of LC–MS are in the Supplemental Methods.

Primary outcome

The primary a priori outcome was the within-individual change in SCFA producers, compared between intervention groups at Day 3. The Day 7 (end of treatment) change in SCFA producers was an additional outcome. Within-individual change in SCFA producers was calculated by subtracting the relative abundance of SCFA producers at baseline from the relative abundance of SCFA producers at follow-up. SCFA producer relative abundance was determined from 16S sequencing of rectal swabs, with Vital et al. used to classify taxa as SCFA producers (Supplemental Table 1) [27]. The rationale for this primary outcome was that SCFA producers levels are a surrogate marker for the overall health and colonization resistance of the gut microbiome.

Secondary outcome and additional outcomes

The secondary outcome was gut pathogen colonization status, which was classified as the presence or absence of VRE or MDR-GNB based on bacterial culture from ICU Day 3 rectal swabs. Additional pre-specified outcomes of interest were differences between study groups in taxonomic changes in the gut microbiome (e.g., alpha- and beta-diversity measurements) and differences in absolute levels of SCFAs. Clinical outcomes of interest were death through 30 days of follow-up, culture-proven infections (which were based on CDC site-specific criteria), [8] feasibility, and safety.

Sample size

The study was powered by referencing levels of SCFA producers within a pilot trial which randomized ICU patients to mixed high- versus low-fiber enteral feeds [28]. In this pilot trial, there was a 6·6% versus 3·7% relative abundance of SCFA producers on ICU Day 3 among those who received high- vs. low-fiber feeds respectively. Using this data, we calculated that a sample size of 90 patients (30 per arm) would yield 80% power to detect a minimum difference of 0·73 standard deviations in SCFA producer levels comparing any one inulin arm to the placebo arm (two-sided t-test with alpha 0·05 and equal variance assumed). We anticipated that some patients would be enrolled but would die prior to ICU Day 3 and therefore a replacement strategy was used to achieve a final sample size of 90 analyzable participants; under this strategy, participants were replaced if they were enrolled but failed to donate a Day 3 sample. Post hoc, the assumptions of normality for SCFA producer levels and equal variances across groups were met.

Statistical analysis

The blinded trial data was locked before it was released to the trial biostatistician who formally adjudicated the primary outcome. The unblinded study data was subsequently released to the investigative team. To determine the primary outcome, we first tested for a difference in SCFA producer levels between inulin groups; when no difference was found, the inulin groups were combined and compared against placebo as a rank-sum test. Additional outcomes were determined by comparing the three study arms using Kruskal–Wallis tests for continuous measures or chi-squared tests for categorical measures. R was used for sequencing analyses, which were adjusted using the Benjamini–Hochberg method to control for false discovery (Supplemental Methods). Other analyses were done using STATA version 14. All statistical tests were performed two-sided at alpha 0·05. No adjustment was made for multiple comparisons in the secondary and exploratory outcomes.

Results

Participants

We randomized 94 individuals who were admitted to the medical ICU with sepsis to enroll 90 participants who completed baseline and ICU Day 3 samples and were included in the analysis (Fig. 1). Enrollment began in October 2019 and concluded in October 2023. Demographic and clinical characteristics were similar between study groups (Table 1 & Supplemental Table 2). The median age was 63.5 years old (IQR 46–72) and 44% of participants were women. Acute severity of illness was high, with an overall median SOFA score of 8 points (IQR 7–10) which predicts a 25–30% 30-day mortality in prior studies [29]. All participants received broad spectrum antibiotics within 24 h of admission, most often β-lactam/β-lactamase inhibitor combination antibiotics (Supplemental Table 3). The observed overall 30-day mortality rate was 26%.

Fig. 1.

Enrollment and randomization

Table 1.

Demographic and clinical characteristics at admission to the intensive care unit

Characteristic	Placebo (N = 30)	Inulin 16 g/day (N = 30)	Inulin 32 g/day (N = 30)	p valuea
Female sex (N, %)	13 (43%)	14 (47%)	13 (43%)	0.96
Age in years (median, IQR)	66 (55–71)	62 (44–69)	65 (40–76)	0.55
Race (N, %)				0.88
White	10 (33%)	12 (40%)	13 (43%)
Black	5 (17%)	7 (23%)	7 (23%)
Hispanic	9 (30%)	6 (20%)	8 (27%)
Other	6 (20%)	5 (17%)	2 (7%)
Ethnicity (N, %)				0.83
Hispanic	11 (37%)	11 (37%)	13 (43%)
Non-Hispanic	19 (63%)	19 (63%)	17 (57%)
Home environment (N, %)				0.63
Independent	12 (40%)	16 (53%)	12 (40%)
Home with help	8 (27%)	8 (27%)	11 (37%)
Assisted living facility	10 (33%)	6 (20%)	7 (23%)
Charlson comorbidity score (N, %)	4 (2–5)	2.5 (2–5)	5 (2–6)	0.22
ICU admission characteristics
Origin prior to ICU admission				0.22
Emergency room	17 (57%)	23 (77%)	19 (63%)
Hospital ward	5 (17%)	0 (0%)	4 (13%)
ICU transfer	8 (27%)	7 (23%)	7 (23%)
Sepsis, based on Sepsis-3 criteria	30 (100%)	30 (100%)	30 (100%)	–
Broad-spectrum antibiotics administered within ± 24 h of ICU admissionb	30 (100%)	30 (100%)	30 (100%)	–
End-organ failure at ICU admission
Use of mechanical ventilationc	21 (70%)	24 (80%)	27 (90%)	0.15
Use of vasopressorsd	20 (69%)	23 (77%)	21 (70%)	0.77
Urine output < 500 mL / 24 h	14 (47%)	15 (50%)	9 (30%)	0.24
SOFA score (median, IQR)e	8 (7–11)	8.5 (7–10)	8 (7–9)	0.56

^aComparison between the three groups; chi-squared test for categorical measures and ANOVA for continuous measures

^bReceipt of broad-spectrum antibiotics was an inclusion requirement and was operationalized as receipt of one or more of any of the following antibiotic classes within ± 24 h of ICU admission: β-lactam/β-lactamase inhibitor combination antibiotics, cephalosporins (gen 2 or greater), fluoroquinolones, lincosamides (clindamycin), metronidazole, and monobactams (e.g., meropenem)

^cMechanical ventilation: classified as use of mechanical ventilation including CPAP or BiPAP at ICU admission

^dUse of vasopressors: classified as receipt of dopamine, dobutamine, epinephrine, norepinephrine, or vasopressin (any dose) at ICU admission

^eSee supplemental Table 8 for complete SOFA score characteristics

IQR, interquartile range; ICU, intensive care unit; SOFA, sequential organ failure assessment; CPAP, continuous positive airway pressure

Primary outcome

Overall, there was no change in the relative abundance of SCFA producers from ICU admission to ICU Day 3 (− 0·1%, IQR − 9·4% to + 6·2%). There was no difference in the within-individual change in the relative abundance of SCFA producers on ICU Day 3 based on study group (placebo: 0·0%, IQR − 8·0% to + 7·4% vs. combined inulin: 0·0%, IQR − 10·1% to + 4·8%; rank-sum p = 0·91; Fig. 2). This null result was unchanged when the two doses of inulin were separately compared against placebo or when end-of-treatment results were examined on ICU Day 7 (placebo: − 0·8%, IQR − 8·3% to + 9·1% vs. combined inulin: − 2·4%, IQR − 11·8% to + 11·0%; p = 0·52) (Table 2 & Fig. 3). Because baseline gut pathogen colonization may influence the extent to which the microbiome can be altered, the above analyses were repeated after restricting the cohort to those who were not gut colonized with VRE or MDR-GNB at ICU admission (n = 60). Again, there was no difference between study groups in the within-individual change in the relative abundance of SCFA producers from baseline to either ICU Day 3 or to ICU Day 7.

Fig. 2.

Within-individual change in the relative abundance of short chain fatty acid-producing bacteria by ICU Day 3, comparing placebo vs. the combined inulin groups

Table 2.

Primary outcome and additional outcomes

End Point	Placebo (N = 30)	Inulin 16 g/day (N = 30)	Inulin 32 g/day (N = 30)	p value*
Primary outcome
Within-individual change in SCFA producers from admission to Day 3	− 0.0% (− 7.9 to + 7.4)	− 1.8% (− 8.9 to + 3.7)	+ 1.1% (− 12 to + 11)	0.72
SCFA producers—% (median, IQR)
ICU admission (N = 90)	7.8% (0.8–27)	9.4% (4.6–23)	12% (1.8–29)	–
ICU Day 3 (N = 90)	9.0% (2.3–27)	9.8% (3.1–26)	12% (2.5–29)	0.91
ICU Day 7 (N = 78)	7.4% (3.2–17)	6.3% (0.8–21)	9.7% (1.3–48)	0.54
ICU Day 14 (N = 60)	16% (0.1–34)	7.3% (0.3–24)	7.7% (0.8–27)	0.71
ICU Day 30 (N = 22)	7.8% (3.4–22)	19% (11–39)	20% (4.9–20)	0.51
Secondary outcome
VRE colonization status
ICU admission	8 (27%)	6 (20%)	4 (13%)	–
At last sample collected	11 (37%)	13 (43%)	6 (20%)	0.14
MDR-GNB colonization status
ICU admission	6 (20%)	4 (13%)	3 (10%)	–
At last sample collected	3 (10%)	4 (13%)	3 (10%)	0.89
Additional outcomes
Shannon index
ICU admission (N = 90)	2.7 (1.8–3.2)	3.1 (1.9–3.6)	2.7 (1.4–3.3)	–
ICU Day 3 (N = 90)	2.3 (1.5–2.9)	2.6 (2.0–3.3)	2.7 (1.5–3.1)	0.24
ICU Day 7 (N = 78)	2.2 (1.2–2.5)	2.1 (1.3–3.0)	2.1 (1.6–3.1)	0.71
Fecal SCFA levels (mMol)
ICU Days 0 to 3 (N = 43)	0 (0–0.18)	0.11 (0–0.66)	0 (0–0.46)	0.40
ICU Days 3 to 7 (N = 48)	0 (0–0.06)	0.29 (0–0.63)	0 (0–0.19)	0.21
Clinical outcomes
ICU length of stay (median days, IQR)	9.5 (6–20)	10.5 (6–23)	9 (6–17)	0.85
Culture-proven infection	15 (50%)	9 (30%)	12 (40%)	0.29
Death	7 (23%)	7 (23%)	9 (30%)	0.79
Death or culture-proven infection	19 (63%)	15 (50%)	16 (53%)	0.56

*Comparisons are across all three study arms. Kruskal–Wallis testing for continuous measures and chi-squared testing for categorical measures

Fig. 3.

Within-individual change in the relative abundance of short chain fatty acid-producing bacteria during the trial in all study groups. Data shown is the relative abundance of SCFA producers at each timepoint minus the relative abundance of SCFA producers at baseline. There were no significant differences between study groups

Secondary outcome

The secondary outcome was gut pathogen colonization status on ICU day 3, which was operationalized as rectal swab positivity for VRE or MDR-GNB. Overall, 34% of patients tested positive for VRE or MDR-GNB on ICU Day 3. There was no difference between gut pathogen colonization status on ICU Day 3, at the end of treatment on ICU Day 7, or based on cultures performed from the last swab acquired (Table 2; Fig. 4). These null results were unchanged when these analyses were repeated after restricting the cohort to patients who tested negative for gut pathogen colonization at ICU admission.

Fig. 4.

Colonization with vancomycin-resistant Enterococcus during the trial, comparing study groups. This figure visualizes dynamic changes in VRE colonization during the study. It shows only patients who tested positive for VRE at some timepoint. Filled circles are samples that cultured positive for VRE. Unfilled circles are samples that cultured negative for VRE. Gray circles (i.e., those without a black outline) are shown when no sample was available for VRE culture at this timepoint. For example, within the placebo group, the first patient cultured positive for VRE on ICU Day 0 to ICU Day 14, and had no sample available for VRE culture on ICU Day 30. MDR-GNB colonization was less common compared to VRE colonization and is not shown in the figure (see Table 2)

Additional outcomes

Alpha and beta diversity

We next tested more broadly for any effect of the intervention on the gut microbiome. We focused initially on alpha diversity (Shannon index), which defines the richness and evenness of a given sample. No consistent differences in Shannon index were observed when comparing study groups (Fig. 5). Results were similar for the Chao index and other measures of alpha diversity. Next we examined beta diversity (unifrac distance), which defines the difference in diversity between two samples. We found significant changes in unifrac distance from baseline to all subsequent timepoints in the placebo group and in the inulin 16 g/day group, but not in the inulin 32 g/day group (Supplemental Fig. 1).

Fig. 5.

Fecal microbial diversity during the trial, in all study groups. Data shown is the Shannon index at each timepoint minus the Shannon index at baseline. There were no significant differences between study groups

Differentially abundant taxa

Next, we used ZacoSeq to look for differential taxa based on study group. All post-intervention samples were combined in this analysis, with a variable added to account for timepoint. After adjusting for false discovery, no differential taxa were observed when the inulin 16 g/day group or inulin 32 g/day group was compared to placebo. When the inulin 32 g/day group was compared to placebo without adjusting for false discovery, relative increases were seen among those who received inulin in three taxa corresponding to Subdoligranulum, Akkermansia Mucinophila, and Coprococcus catus (Supplemental Fig. 2).

Whole stool SCFA levels

SCFA levels were directly measured from whole stools using LC–MS and compared between groups. Across groups, we compared (1) samples collected from ICU Day 0 to 3 and (2) samples collected from ICU Day 3 to 7 (Table 2). Comparing the placebo group vs. the combined inulin groups, there were no difference in directly measured levels of SCFAs either at the Day 0–3 timepoint or at the Day 3–7 timepoint (p = 0·40 and 0·21 respectively, Supplemental Fig. 3).

Clinical outcomes

Last, we assessed clinical outcomes. The overall 30-day mortality rate was 26% and the overall 30-day rate of the composite outcome of death or culture-proven infection was 56%. Infection types are shown in Supplemental Table 4. Comparing study groups, there was no difference in ICU length of stay, 30-day mortality, culture-proven infection, or the composite outcome (Table 2). Patients who died or developed infections had lower relative abundance of SCFA producers at ICU admission compared to those who did not die or develop infections (p = 0·03).

Adherence and safety

All participants received one or more study doses and the overall median number of doses received was 13 (IQR 11–14) out of a total of 14 intended doses (Supplemental Table 5). Prespecified categories of adverse events of interest included (1) bloating and changes in stool frequency and consistency, because these are established side effects of dietary fiber; and (2) serum electrolyte changes, because the study intervention was delivered in 250 mL of free water per dose. We found no differences in these prespecified adverse events, comparing between study groups (Supplemental Table 6). There was one episode of vomiting deemed related to the study intervention, in a patient randomized to inulin 32 g/day. No other adverse events were deemed related to the study intervention.

Discussion

In this randomized controlled trial, we investigated the effects of prebiotic inulin on the gut microbiome and on gut colonization resistance against pathogenic bacteria in the ICU. The trial primary outcome was the change in relative abundance of SCFA-producing bacteria. This outcome was designed with the understanding that SCFA producers serve as a surrogate for important downstream clinical events such as infection with gut-derived bacteria, and in fact this was one of the findings from this trial. However, high doses of inulin did not alter the relative abundance of SCFA producers. Rates of positive rectal swab cultures for VRE and MDR-GNB, the most clinically significant gut pathogens in the ICU, were also similar across study groups. Minor differences between groups were observed in beta-diversity and in specific bacteria, but without correlates to suggest a clinically meaningful benefit for inulin. Overall, there were minimal differences in the gut microbiome among those randomized to inulin versus placebo.

This trial differentiated itself from prior trials of pre- or probiotics [30, 31] because it tested a specific prebiotic fiber (inulin) and employed a highly rigorous study design which may set a standard for future microbiome-based interventions in the ICU. Adherence to the study intervention was excellent and samples were collected at pre-specified timepoints, without any samples missed. This trial successfully demonstrated that an oral prebiotic fiber supplement can be delivered in the ICU with high fidelity, without any adverse effects.

In ambulatory adults, inulin has large effects on the gut microbiome whereas this trial showed that inulin had few effects on patients in the ICU [17, 18]. Several factors are likely to explain these differences across patient populations. First, unlike cohorts of ambulatory patients or healthy volunteers, the participants in our study all received intravenous broad-spectrum antibiotics concurrent with prebiotic inulin including many patients who received multiple broad-spectrum antibiotics. Intravenous vancomycin, which was received by half of the cohort, achieves high gut concentrations; surprisingly sparse data exists on the gut penetration of other intravenous antibiotics [32–34]. One possible reason that we did not observe greater effects for inulin is that vancomycin or other intravenous antibiotics may have achieved high gut concentrations. These antibiotics, rather than inulin, may have dictated microbiome composition. If this interpretation is correct, it would represent a major obstacle for all gut microbiome-based interventions in the ICU (i.e., for pre- and probiotics alike). In other high-risk populations such as nursing home residents, inulin and other gut microbiome-based interventions could still be viable therapies. Second, in this trial, the characteristics of the gut microbiome at ICU admission was an important determinant of the gut microbiome at subsequent timepoints. Participants were relatively “flat” during the course of the study in SCFA producer levels, alpha-, and beta-diversity. This speaks to the importance of the baseline microbiome at ICU admission, which may already have been shaped by prior antibiotics. More research is needed to delineate the dynamics of the microbiome in the ICU and to understand whether specific factors such as baseline VRE gut colonization may be used to distinguish patients who are unlikely to benefit from a gut microbiome-based intervention. This trial studied prebiotic inulin, and it did not test a probiotic or symbiotic approach to improve the gut microbiome in the ICU. Our results cannot directly comment on alternative strategies, but they argue that gut microbiome-based interventions will need to be highly potent in order to be effective.

The trial has limitations. In hindsight, the selection of Day 3 as the primary outcome may have been too short a duration of treatment to realistically hope to see an effect from inulin. The trial was single center, and the population studied had a high burden of chronic illness, even for the medical ICU. The patient cohort was likely to have prior hospitalizations and they may have received prior antibiotics of varying durations. These cohort characteristics may have been reflected in the very high rates of VRE gut colonization observed, and they may have impacted the ability of inulin to affect the microbiome (i.e., inulin might be more effective in a more pristine ICU patient population). The trial targeted patients with sepsis, who are the plurality of ICU patients but not the only patients. The intervention was delivered over a timeframe of seven days rather than continuously during hospitalization and it is possible that greater effects would have been observed if inulin was continued past exposure to antibiotics (although this would be unlikely to impact short-term outcomes). In future studies, later sample timepoints—after discontinuation of antibiotics and microbiome recovery—would be of interest. A challenge of such samples is minimizing loss to follow-up, because substantial numbers of patients are likely to die or to be discharged after ICU Day 14. Last, as previously noted, the trial tested only one prebiotic (inulin). These results do not necessarily mean that all pre- or probiotic strategies are doomed in the ICU.

In sum, this trial did not find evidence that inulin improved gut pathogen colonization or infection in the ICU. Alternative strategies should be considered.

Author contributions

The following contributions were made by each author: Heekuk Park: analysis Elissa Lynch: data curation Alice Tillman: data curation Kristen Lewis: data curation Zhezhen Jin: analysis Anne-Catrin Uhlemann, MD: methodology, supervision Julian A. Abrams, MD: conceptualization, funding Daniel E. Freedberg, MD: analysis, data curation, methodology, supervision, conceptualization, funding, investigation, resources, validation, project administration, and manuscript writing, reviewing, editing.

Funding

The trial was funded by the Department of Defense Peer Reviewed Medical Research Program (PR181960). Additional funding was received from the Columbia Division of Digestive and Liver Disease (P30DK132710).

Availability of data and materials

No datasets were generated or analysed during the current study.

Declarations

Competing interests

The authors declare that they have no conflict of interest.