Rapid and Accurate Diagnosis and Prognosis of Acute Infections and Sepsis from Whole Blood Using Host Response mRNA amplification and Result Interpretation by Machine-Learning Classifiers

doi:10.21203/rs.3.rs-5194992/v1

Download PDF

Article

Rapid and Accurate Diagnosis and Prognosis of Acute Infections and Sepsis from Whole Blood Using Host Response mRNA amplification and Result Interpretation by Machine-Learning Classifiers

https://doi.org/10.21203/rs.3.rs-5194992/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Many patients in the emergency department present with signs and symptoms that arouse concern for sepsis; however, other explanations are also possible. There are currently no rapid tests used in clinical practice that reliably distinguish the presence of a bacterial or viral infection vs. a non-infectious etiology and can predict a patient’s likelihood to decompensate. The diagnostic and prognostic uncertainty in “gray zone” patients complicates the decision to begin therapy as clinicians need to balance the risk of withholding therapy vs. the risk of the therapy itself (e.g., overtreatment with antibiotics and hospitalization, which is costly, potentially harmful, and contributes to antibiotic resistance).

The TriVerity^™ Test uses isothermal amplification and machine-learning algorithms to quantify and interpret mRNA expression levels to determine both likelihood of bacterial infection, viral infection, or no infection, and whether the patient will likely require one or more critical interventions within 7 days. The three scores each fall into one of five interpretation bands ranging from Very high to Very low. Testing takes approximately 30 minutes using the proprietary Myrna^™ Instrument with an operator hands-on-time of under one minute.

We enrolled 1,222 patients from 22 emergency departments (ED) to validate the performance of the TriVerity Test. Patients were treated as per local standard of care and were followed for 28 days. Bacterial and viral TriVerity results were validated against clinically adjudicated infection status; the illness severity TriVerity result was validated against the need for at least one critical interventions within 7 days.

The bacterial TriVerity result had high AUROC for the diagnosis of bacterial infection (0.83; 80% CI 0.81–0.85) and divided bacterial infection likelihood scores into five interpretation bands with increasing likelihood ratios of infection ranging from Very low (LR- 0.08, 80% CI 0.06–0.11) to Very high (LR + 8.04, 80% CI 5.72–11.78). The AUROC for the bacterial TriVerity result was significantly higher compared to AUROCs for C-reactive protein, procalcitonin or white blood cell count. Similarly, the viral TriVerity score showed high AUROC for the diagnosis of viral infection (0.91; 80% CI 0.90–0.93) and likelihood ratios from Very low (LR- 0.09, 80% CI 0.05–0.14) to Very high (LR + 40.93; 80% CI 29.11–79.23). The TriVerity Illness Severity score showed a high AUROC for the prediction of illness severity (0.77; 80% CI 0.77–0.81) with scores divided into five interpretation bands with increasing likelihood ratios ranging from Very low (LR- 0.22; 80% CI 0.14–0.33) to Very high (LR + 11.33; 80% CI 7.31–17.00). TriVerity illness severity results allowed marked re-classification of the risk for “ICU-level care” as compared to clinical assessment (qSOFA scores) alone.

In conclusion, TriVerity provides rapid, highly accurate and actionable results for the diagnosis and prognosis of patients with suspected acute infection and/or sepsis, supporting a major unmet medical need. TriVerity may improve personalized management of patients with suspected acute infections and suspected sepsis for improved overall healthcare outcomes.

Health sciences/Diseases/Infectious diseases/Bacterial infection

Health sciences/Biomarkers/Diagnostic markers

Sepsis, an often fast-progressing and potentially fatal condition, is defined as a life-threatening organ dysfunction caused by a dysregulated host response to an infection ¹. During sepsis, the body’s immune system, already fighting an infection, becomes dysregulated and damages its own tissues and organs or becomes functionally paralyzed, which permits a primary or a secondary infection to progress dangerously. Sepsis requires rapid administration of antibiotics and fluids ² since the risk of death from sepsis continues to be high and increases by 7–8% with every hour of delay in beginning optimal treatment ^3,4. Sepsis is often subtle and difficult to detect, and the diagnosis is not infrequently missed, with potentially deadly consequences. As a result of difficult diagnosis and an imperative for early treatment of all sepsis as bacterial, many patients with suspected sepsis who have alternative diagnoses are overtreated with antibiotics and hospitalization, which is costly, potentially harmful, and contributes to antibiotic resistance⁵.

The two main components of sepsis treatment are: 1) anti-infective therapy (e.g. antimicrobials or surgical source control) and supportive therapy (e.g. intravenous fluids, vasopressor medications, oxygen for hypoxia etc.). These two therapeutic approaches underscore the need to assess a patient with suspected infection along two ‘axes’: infection presence and severity ^6,7. Furthermore, along the infection axis, there is a need to further distinguish whether the cause of the patient’s illness is likely bacterial, viral, co-infection (both viral and bacterial infection), or non-infectious in order to guide anti-infective treatment strategies.

There are currently no tests used in clinical practice that rapidly and reliably determine the presence and severity of infection ^6,8. For example, while protein-based biomarkers such as C-reactive protein (CRP) and procalcitonin (PCT) or diagnostic tests based on direct pathogen-detection can assist in the the diagnosis of infection, they often lack accuracy and/or speed to allow a physician at initial presentation to accurately identify the presence of an infections and distinguish bacterial from viral infections, important for antibiotic decision-making at the time of presentation ^9,10. Secondly, irrespective of a patient’s infection status, predicting their likelihood of clinical decompensation remains an important challenge. Early admission to the intensive care unit (ICU) is imperative for patient survival advantages ¹¹ but biomarkers and clinical scores for estimating risk (e.g. lactate, cellular morphology tests, clinical scores such as sequential organ failure assessment (SOFA) and quick SOFA (qSOFA)) have limited accuracy for predicting the likelihood of decompensation in broader populations ^12,13.

Inflammatix’s TriVerity™ Test addresses the unmet need by measuring the host’s blood messenger RNA levels for determining both presence of, and quantifying the severity of infection. This is achieved in a timely and actionable manner by real-time measurement of the expression of 29 host mRNAs in a blood sample, which were previously reported as differentially expressed in response to infection and predictive of the severity of infection ^14–20. TriVerity uses a novel isothermal amplification technique and machine-learning algorithm to quantify and interpret RNA expression levels to determine the likelihood of bacterial infection, viral infection or no infection, and whether the patient is likely to require critical interventions within 7 days (vasopressors, mechanical ventilation, and/or renal replacement therapy [RRT]). Testing takes approximately 30 minutes using the proprietary cartridge-based Myrna™ Instrument with an operator hands-on-time of under one minute. This rapid turnaround around time enables immediate actionability for ruling in or ruling out infections and severe illness through the expression of scores that fall into one of five interpretation bands ranging from Very high to Very low.

In this large multicenter clinical study (SEPSIS-SHIELD, CT04094818), building on previous observational studies ^21–24 we investigated whether TriVerity can both diagnose bacterial and viral infections vs. non-infectious mimics and whether TriVerity can predict the need for mechanical ventilation, vasopressor use and/or renal replacement therapy within 7 days (“ICU-level care”) in patients with suspected acute infection or suspected sepsis presenting to 22 EDs in the USA and Europe. We compared the accuracy of TriVerity to established biomarkers for diagnosing bacterial and viral infections. We also investigated whether TriVerity’s severity score together with clinical scores further improves the accuracy of predicting the severity of illness.

Characteristics of study participants at the time of ED presentations

Mean age, sex, race and ethnicities were representative of a US ED population (Table 1). Among patients evaluable for the diagnostic endpoint the mean age was 50.6 (SD 17.4) and 47.3% of the patients were female. The majority of patients were White (63.2%), followed by African American (30.6%); 13.3% of patients were Hispanic/Latino. Similar percentages were observed among those evaluated for the prognostic endpoint (Table 1). Metabolic/endocrinological, respiratory and cardiovascular diseases were the most prevalent medical conditions. There were no marked differences between patients evaluable for the diagnostic endpoint vs. those for the prognostic endpoint. Overall, 132 (18.1%) patients evaluable for the diagnostic endpoint were immunosuppressed compared to 206 (18.4%) patients evaluable for the prognostic endpoint. Malignancies were the most frequently found type of immunosuppression (approximately 10% of patients), followed by solid organ transplantation, steroid treatment and HIV/AIDS.

Table 1

**Overview of demographics and medical history of patients evaluable for the diagnostic and prognostic endpoint**
	Consensus Adjudicated (Diagnostic Outcome) (N = 729)	Evaluable for 7-Day ICU-Level Care (Prognostic Outcome) (N = 1,120)
Age
Mean (SD)	50.6 (17.4)	51.3 (17.6)
Median (Range)	52.0 (18.0, 90.0)	52.0 (18.0, 90.0)
Sex at Birth
Female	345 (47.3%)	544 (48.6%)
Male	384 (52.7%)	576 (51.4%)
Race
American Indian or Alaska Native	0 (0.0%)	1 (0.1%)
Asian₁	6 (0.8%)	8 (0.7%)
Black or African American	223 (30.6%)	321 (28.7%)
Native Hawaiian or Other Pacific Islander	2 (0.3%)	2 (0.2%)
White¹	461 (63.2%)	734 (65.5%)
Other	37 (5.1%)	56 (5.0%)
Ethnicity
Hispanic or Latino	97 (13.3%)	128 (11.4%)
Not Hispanic or Latino	623 (85.5%)	972 (86.8%)
Unknown	9 (1.2%)	20 (1.8%)
Medical History
Blood Disorders	51 (7.0%)	80 (7.1%)
Cardiovascular Diseases	303 (41.6%)	485 (43.3%)
Kidney Diseases	148 (14.0%)	225 (20.1%)
Gastrointestinal Tract Diseases	167 (22.9%)	265 (23.7%)
Metabolic/Endocrinological Diseases	315 (43.2%)	490 (42.8%)
Musculoskeletal Diseases	106 (14.5%)	167 (14.9%)
Neuropsychiatric Diseases	162 (22.2%)	264 (23.6%)
Respiratory Tract Diseases	229 (31.4%)	350 (31.3%)
Skin or Soft Tissue Diseases	36 (4.9%)	58 (5.2%)
Urogenital Diseases	57 (7.8%)	90 8.0%)
Other	111 (15.2%)	187 (16.7%)
Immunosuppression
Bone Marrow, Stem Cell and/or Other Cell Transplant	2 (0.3%)	3 (0.3%)
Malignancies	73 (10.0%)	121 (10.8%)
HIV/AIDS	16 (2.2%)	22 (2.0%)
Solid Organ Transplant	27 (3.7%)	34 (3.0%)
Steroid Treatment	17 (2.3%)	30 (2.7%)
Other Immunosuppression	17 (2.3%)	29 (2.6%)
¹ Two patients gave their race as biracial (Asian, White)

The rate of abnormal vital signs at presentation in the ED did not differ significantly between patients evaluable for the diagnostic and prognostic endpoint (data not shown). The mean leukocyte count in patients eligible for the diagnostic endpoint was 11.8 x 10⁹/L, the mean neutrophil percentage was 76.9 and the mean lymphocyte percentage was 12.5; similar numbers were observed in patients eligible for the prognostic endpoint (Supplementary Table 3A). Mean concentrations of biomarkers are shown in Supplementary Table 3B.

The majority (57.2%) of patients were admitted to a regular ward, 13.4% were admitted to an ICU whereas 23.5% were discharged home from ED (Supplementary Table 4). The mean length of hospital stay was 6.3 days, and the mean length of ICU stay was 5.7 days. A total of 122 (10.9%) patients required “ICU-level care”, the primary severity endpoint, which included mechanical ventilation (n = 63, 51.6%), vasopressor use (n = 99, 81.2%) and/or renal replacement therapy (23, 18.9%) (Supplementary Table 5). Among 147 patients who were transferred to the ICU, 98 (66.7%) received the “ICU-level care” composite endpoint.

Infection status and anatomical location of infection

A total of 61.5% of participants had a bacterial infection compared to 22.6% viral infections and 1.6% bacterial-viral coinfections, while 14.3% of patients were adjudicated not to have an infection (Table 2). Among bacterial infections, urinary tract (30.9%) and skin or soft tissue infections (28.9%) were most frequently observed, followed by respiratory, bloodstream, and gastrointestinal tract infections. Among viral infections, the distribution was markedly different: infections were most often adjudicated to as respiratory tract infections (95.5%) followed by gastrointestinal tract infections (2.3%). Three parasitic infections were diagnosed (Giardia lamblia enterocolitis, n = 1; Trichomonas vaginalis vulvovaginitis, n = 2)). The percentage of bacterial infections was relatively high (61.5%), driven by seasonal epidemiology and the broad inclusion criteria allowing for enrollment of patients with all types of suspected infections (not only respiratory infections).

Table 2

Clinically adjudicated infection status and anatomical localization of infection
	Type of Infection/ Anatomical Location of Infection^1,2	% (n/N)
Type of Infection	Bacterial Infection	61.5% (448/729)
	Viral Infection	22.6% (165/729)
	Viral-Bacterial Coinfection	1.6% (12/729)
	Non-infected	14.3% (104/729)
Bacterial Adjudication	Bloodstream	20.0% (92/460)
	Central Nervous System	0.4% (2/460)
	Gastrointestinal Tract	15.0% (69/460)
	Joint	1.7% (8/460)
	Respiratory Tract	12.6% (58/460)
	Skin or Soft Tissue	28.9% (133/460)
	Urinary Tract	30.9% (142/460)
	Unknown/ Other	18.9% (87/460)
Viral Adjudication¹	Bloodstream	0.6% (1/177)
	Central Nervous System	1.1% (2/177)
	Gastrointestinal Tract	2.3% (4/177)
	Respiratory Tract	95.5% (169/177)
	Skin or Soft Tissue	1.1% (2/177)
	Unknown/ Other	3.4% (6/177)
¹ Patients may have multiple sources. ² Anatomical localization of infection was determined by clinical adjudicators after review of the clinical adjudication report.

Accuracy of TriVerity for the diagnosis of bacterial infection

TriVerity Bacterial Score had an AUROC of 0.83 (80% CI 0.81–0.85) for detecting bacterial infections. TriVerity divided the bacterial scores into five interpretation bands based on likelihood of infection which showed increasing risk of bacterial infection by interpretation band with Very low likelihood of 0.08 (95% CI 0.06–0.11), Low 0.54 (0.47–0.66), Moderate 1.14 (0.94–1.43), High 2.5 (2.04–3.31) and Very high 8.0 (5.72–11.78) (Table 3A). TriVerity demonstrated very high specificity for the two rule-in interpretation bands, Very high (specificity = 95.5%) and High (specificity = 90.7%). Simultaneously, TriVerity had Very high sensitivity for the two rule-out interpretation bands: Very low (sensitivity = 97.2%) and Low (sensitivity = 81.5%) (Table 3A). Lastly, 81.3% of the patients fell into one of the clinically actionable Very low, Low, High, and Very high interpretation bands. At a prevalence of 63.1% for bacterial infections, the probability of having a bacterial infection for the outer Very high interpretation band was 93.2% whereas the Very low interpretation band ruled out a bacterial infection with probability 87.9%.

Table 3

A | Accuracy of TriVerity bacterial results for the diagnosis of bacterial infections
TriVerity Bacterial Band (Score)	Clinically Adjudicated Bacterial Infection¹		Sensitivity (%)	Specificity (%)	Likelihood Ratio (80% CI)	Relative Frequency of Result (% in Band)	Probability of Bacterial Infection (%)
TriVerity Bacterial Band (Score)	Yes (N)	No (N)	Sensitivity (%)	Specificity (%)	Likelihood Ratio (80% CI)	Relative Frequency of Result (% in Band)	Probability of Bacterial Infection (%)
Very high (40–50)	165	12	35.9	95.5	8.04 (5.72–11.78)	24.3	93.2
High (30–39)	107	25	23.3	90.7	2.50 (2.04–3.31)	18.1	81.1
Moderate (21–29)	90	46	19.6	82.9	1.14 (0.94–1.43)	18.7	66.2
Low (11–20)	85	92	81.5	34.2	0.54 (0.47–0.66)	24.3	48
Very low (0–10)	13	94	97.2	34.9	0.08 (0.06–0.11)	14.7	12.1
¹ using consensus adjudication as the reference standard

Accuracy of TriVerity for the diagnosis of viral infection

The TriVerity Viral score had very high accuracy for diagnosing viral infections with an AUROC of 0.91 (80% CI 0.90–0.93) when compared to clinical adjudication. TriVerity divided the viral scores into five interpretation bands based on likelihood of infection which showed increasing risk of viral infection by interpretation band with Very low likelihood of 0.09 (80% CI 0.05–0.14), Low 0.32 (80% CI 0.23–0.43), Moderate 0.87 (80% CI 0.64–1.10), High 2.36 (80% CI 1.59–3.30) and Very high 40.93 (80% CI 29.11–79.23). The TriVerity Viral score had very high specificity for the two rule-in interpretation bands, Very high (specificity = 98.6%) and High (specificity = 94%) (Table 3B). Simultaneously, TriVerity had very high sensitivity for the two rule-out interpretation bands: Very low (sensitivity = 95.5%) and Low (sensitivity = 90.4%) (Table 3B). Lastly, 86.1% of TriVerity Viral results fell into the clinically actionable Very high, High, Low and Very low bands. At a prevalence of 24.3% for viral infections, the probability of having a viral infection in the outer (Very High) interpretation band was 92.9%. In contrast, the Very Low interpretation band correctly ruled out viral infections in 97.1% of cases. Of interest, the accuracy of the TriVerity viral results did not differ significantly in patients diagnosed with SARS-CoV-2 vs. those without SARS-CoV-2 (data not shown), demonstrating its applicability and generalizability to emerging pathogens.

Table 3

B | Accuracy of TriVerity viral results for the diagnosis of viral infections
TriVerity Viral Band (Score)	Clinically Adjudicated Viral Infection¹		Sensitivity (%)	Specificity (%)	Likelihood Ratio (80% CI)	Relative Frequency of Result (% in Band)	Probability of Viral Infection (%)
TriVerity Viral Band (Score)	Yes (N)	No (N)	Sensitivity (%)	Specificity (%)	Likelihood Ratio (80% CI)	Relative Frequency of Result (% in Band)	Probability of Viral Infection (%)
Very high (40–50)	105	8	59.3	98.6	40.93 (29.11–79.23)	15.5	92.9
High (30–39)	25	33	14.1	94.0	2.36 (1.59–3.30)	8	43.1
Moderate 21-29)	22	79	12.4	85.7	0.87 (0.64–1.10)	13.9	21.8
Low (11–20)	17	166	90.4	30.1	0.32 (0.23–0.43)	25.1	9.3
Very low (0–10)	8	266	95.5	48.2	0.09 (0.05–0.14)	37.6	2.9
¹ using consensus adjudication as the reference standard

In summary, TriVerity Bacterial and Viral results readouts were consistent, indicating a clear relationship between TriVerity scores and the increasing likelihoods of bacterial and/or viral infections across TriVerity interpretation bands. Importantly, the TriVerity Bacterial and Viral scores increased monotonically and the 80% CIs for adjacent bands did not overlap in any of the analyses (Fig. 2).

Accuracy of TriVerity for the diagnosis of bacterial infections compared to commonly used biomarkers, across different races and stratified by immune status

The AUROC of the TriVerity Bacterial Score (0.83, 80% CI 0.80–0.87) was significantly higher than those of commonly used biomarkers for the diagnosis of infections, including PCT (AUROC = 0.69, 80% CI 0.65–0.72), CRP (AUROC = 0.72, 80% CI 0.68–0.77), and white blood cell counts (0.75, 80% CI 0.72–0.79) (p < 0.0001 for all comparisons, Supplementary Table 6A).

TriVerity had significantly higher generalizability for the accuracy of diagnosis across different races than other methods. Specifically, while the overall AUROC for PCT was 0.69 for diagnosis of bacterial infection, it was lower in Black/African Americans (AUROC = 0.63) and other races (AUROC = 0.57) compared to Whites (AUROC = 0.74), highlighting the variability across races. In contrast, the variability between AUROCs for TriVerity across different races was substantially lower. Its overall AUROC of 0.83 was virtually identical in Whites (0.82) and Black/African Americans (0.83), and higher in other races (0.92) (Supplementary Table 6B). Furthermore, when comparing the AUROCs for TriVerity and PCT for each race, TriVerity consistently had significantly higher AUROC than PCT (Supplementary Table 6B).

Because TriVerity measures the host’s immune response to infection, it is of interest to determine whether the scores maintain predictive ability for diagnosing infections in immunocompromised patients, who also have an increased risk of infection. Therefore, we investigated the accuracy of the bacterial and viral TriVerity results stratified by the patient’s immune status. For diagnosis of bacterial infection, there was no significant difference in the AUROC for the TriVerity Bacterial score between immunocompromised (0.80, 80% CI 0.75–0.85) and immunocompetent (0.83, 80% CI 0.81–0.85) patients (Supplementary Table 7A); similarly, the AUROCs for the TriVerity Viral score were virtually the same between immunocompetent (0.89, 80% CI 0.86–0.94) and immunocompromised (0.91, 80% CI 0.75–0.85) patients (Supplementary Table 7B).

Prognostic accuracy of the TriVerity Severity score

We assessed the accuracy of the TriVerity Severity score for predicting the need for “ICU-level care” (defined as the need for mechanical ventilation, vasopressor use, and/or renal replacement therapy) within 7 days, the primary severity endpoint of our study. TriVerity had an AUROC of 0.77 (80% CI 0.77–0.81) for predicting severity of illness. TriVerity divided the illness severity scores into five interpretation bands based on likelihood of need for ICU-level care which showed increasing risk of severe illness by interpretation band with Very low likelihood of 0.22 (80% CI 0.14–0.33), Low 0.43 (80% CI 0.30–0.56), Moderate 1.63 (80% CI 1.36–1.92), High 2.41 (80% CI 2.06–2.84) and Very high 11.33 (80% CI 7.30-7.00). The specificity of the Very high and High illness severity (rule-in) interpretation bands for predicting severe illness were 98.7% and 86.1%, respectively, and the LR + were 11.33 and 2.41, respectively (Table 4). Conversely, the sensitivity of the Very low and Low illness severity (rule-out) bands were 91.8% and 87.7%, respectively, and LR- of 0.22 and 0.43, respectively. The majority (79.6%) of all results were observed in the clinically actionable Very high, High, Low and Very low interpretation bands. At a prevalence of 10.9%, the probability of having severe illness was 58.1% for the Very High band whereas the Very Low interpretation band correctly ruled out severe illness in 97.3% of cases.

Table 4

Accuracy of TriVerity Illness Severity scores for the prediction of the need of mechanical ventilation, vasopressors, and/or renal replacement therapy within 7 days
TriVerity Illness Severity Band (Score)	Need for “ICU-Level Care” Within 7 Days¹		Sensitivity (%)	Specificity (%)	Likelihood Ratio (80% CI)	Relative Frequency of Result (% in Band)	Probability of Severe Illness (%)
TriVerity Illness Severity Band (Score)	Yes	No	Sensitivity (%)	Specificity (%)	Likelihood Ratio (80% CI)	Relative Frequency of Result (% in Band)	Probability of Severe Illness (%)
Very high (40–50)	18	13	14.8	98.7	11.33 (7.31–17.00)	2.8	58.1
High (30–39)	41	139	33.6	86.1	2.41 (2.06–2.84)	16.1	22.8
Moderate 21-29)	38	191	31.1	80.9	1.63 (1.36–1.92)	20.4	16.6
Low (11–20)	15	288	87.7	28.9	0.43 (0.30–0.56)	27.1	5
Very low (0–10)	10	367	91.8	36.8	0.22 (0.14–0.33)	33.7	2.7
^{1 ”}ICU-level care”: need for mechanical ventilation, vasopressors, and/or renal replacement therapy within 7 days

The TriVerity Illness Severity scores increased monotonically and the 80% CIs for adjacent bands did not overlap (Supplementary Fig. 1). Lactate, a biomarker for the severity of hypoxia and lactic acidosis, which is mandated as part of the SEP-1 bundle within 1 hour of suspicion of sepsis, had an AUROC of 0.77 (80% CI 0.73–0.79) for predicting the need for ICU-level care. In the same patients, TriVerity had an AUROC of 0.78 (80% CI 0.75–0.80). When stratifying lactate into concentrations ranges of < 2, 2–4, and > 4 ng/ml, the specificity of the rule-in concentration of > 4 ng/ml was 95.8% whereas the rule-out sensitivity for the lowest (< 2 ng/ml) concentration was 66.7% (Supplementary Table 8A). In these patients, the TriVerity illness severity Very high and High bands correctly predicted “ICU-level care” in 35 of 49 (71.4%) patients whereas lactate concentrations of > 4 ng/ml only identified 23 of 49 (46.9%) patients. The TriVerity Very Low and Low illness severity bands correctly ruled out “ICU-level care” in 532 of 636 (83.6%). In contrast, lactate concentrations of < 4 did rule out ICU-level care in 95.1% of patients (Supplementary Table 8B).

The severity of the patient’s clinical presentation at the time of presentation in the ED is often assessed using clinical scores such as the quick serial organ failure assessment (qSOFA) score. We observed that AUROCs for predicting the need for “ICU-level care” within 7 days were similar for the TriVerity Illness Severity score and qSOFA (0.78, 80% CI 0.76–0.81 vs. 0.78, 80% CI 0.76–0.81). Therefore, we investigated whether integrating the clinical presentation, as measured by qSOFA, with host response, as measured by the TriVerity Illness Severity score, further improves the accuracy of predicting “ICU-level care”. We determined pre- and post-test probabilities of sequential qSOFA plus TriVerity Illness Severity results (Table 5 and Fig. 3A, B). Overall, 10.9% of the prognostic study population required “ICU-level care” within 7 days. Among patients with low-risk qSOFA scores (0–1), 7.7% required ICU-level care (Fig. 2A), whereas 46.3% of those with high-risk qSOFA scores (2–3) required the same (Fig. 2B). However, using TriVerity Illness Severity scores sequentially with qSOFA scores markedly increased predictive accuracy in both the low and high clinical risk patients. Within patients with low- risk qSOFA and Very high, High, and Moderate TriVerity Illness Severity score, the percentage of those needing of “ICU-level care” increased substantially to 52%, 14%, and 13%, respectively. Further, in patients with Low and Very low TriVerity Illness Severity scores with low-risk qSOFA score, the percentage of patients requiring ICU-level care decreased to 4% and 2%, respectively.

Table 5

Pre- and post-test probabilities for the need for “ICU-level care” within 7 days stratified by qSOFA scores of 0–1 (low risk) or 2–3 (high risk) and TriVerity Illness Severity interpretation band
Patients with Need of “ICU-Level Care”¹ Within 7 Days
Overall	Stratified by qSOFA Score on Day 0	Stratified by Triverity Illness Severity Band	No	Yes	Positive Rate (%)
10.9% (122/1,120)	qSOFA Score 0–1 (Low Risk) 7.71% (74/960)	Very high	11	12	52.17
		High	127	21	14.19
		Moderate	159	24	13.11
		Low	249	10	3.86
		Very low	340	7	2.02
	qSOFA Score 2–3 (High Risk) 46.34% (38/82)	Very high	2	5	71.43
		High	6	16	72.73
		Moderate	16	12	42.86
		Low	14	3	17.65
		Very low	6	2	25.00
^{1 “}ICU-level care”: Need for mechanical ventilation, vasopressor use, and/or renal replacement therapy

At the same time, in patients with high-risk qSOFA scores of 2–3 (46.3% requiring “ICU-level care”), the proportion of patients requiring “ICU-level care” increased substantially to 71% and 73% in those with Very high and High TriVerity Illness Severity scores, respectively (Fig. 3B). In contrast, in patients with high-risk qSOFA scores and Very Low or Low TriVerity Illness Severity scores, the proportion of patients requiring “ICU-level care” was markedly reduced to 25% and 18%, respectively.

Collectively, these results demonstrate that TriVerity Illness Severity scores can accurately predict the need for “ICU-level care” within 7 days. Furthermore, integrating TriVerity Illness Severity scores with qSOFA further improves the accuracy of predicting the need for “ICU-level care” by up to 7-fold compared to qSOFA alone thereby reducing both false positive and false negative results of clinical presentation.

As sepsis is one of the most important severe conditions, we investigated whether the combined use of TriVerity Bacterial and Severity scores can assist predicting sepsis (Table 6). 64.0% of patients with Very high TriVerity Bacterial scores and Very high TriVerity Illness Severity scores had sepsis using the Sepsis-3 definition, whereas only 1.1% of patients with Very low TriVerity Bacterial plus Illness Severity scores were diagnosed with sepsis.

Table 6

Number and percent of patients with vs. without sepsis stratified by TriVerity Bacterial and Illness Severity scores
		TriVerity Severity Score
		Very high (N = 15)	High (N = 21)	Moderate N = 26)	Low (N = 14)	Very low (N = 9)
TriVerity Bacterial Score	Very high	16/25 (64.0%)	26/92 (28.3%)	14/49 (28.6%)	0/4 (0.0%)	0/0 (N/A)
	High	1/2 (50.0%)	2/16 (12.5%)	10/53 (18.9%)	7/51 (13.7%)	0/2 (0.0%)
	Moderate	0/0 (N/A)	0/5 (0.0%)	5/21 (23.8%)	4/80 (5.0%)	2/19 (10.5%)
	Low	0/0 (N/A)	0/1 (0.0%)	0/12 (0.0%)	3/48 (6.3%)	6/103 (5.8%)
	Very low	0/0 (N/A)	0/0 (N/A)	0/1 (0.0%)	0/6 (0.0%)	1/89 (1.1%)
¹ defined as bacterially infected (consensus adjudication) and having delta SOFA > = 2 and/or
7-day ICU level care

In clinical practice, when making combined use of the diagnostic and prognostic TriVerity Test results we found very strong supporting actionability: the vast majority of TriVerity results (99.6% for consensus adjudicated cases) fell into one of the four clinically actionable interpretation bands (i.e., Very low, Low, High and Very high) for at least one of the diagnostic and prognostic result axis (i.e., infection or severity). This is a unique feature of TriVerity that holds clinical importance.

The diagnosis and prognosis of patients with acute infections and/or suspected sepsis is an ongoing clinical challenge. The clinical presentation in sepsis is often non-specific and current diagnostic tools have limited sensitivity and/or require a long time to obtain results and do not assess prognosis or etiology ^6,8,25. This manuscript presents results of a large registrational clinical trial establishing performance criteria for the TriVerity Test. The TriVerity Test combines innovative transcriptomic measurement of 29 host response mRNAs and interpretation by machine-learning algorithms. It rapidly provides simultaneous results for presence (infected or not), type (bacterial or viral), and illness severity in patients with suspected acute infection and suspected sepsis. Clinical actionability is made possible by high test accuracy, intuitive result readouts as scores that fall into defined interpretation bands and the test’s rapid turnaround time of approximately 30 minutes from an easy-to-use cartridge-based instrument.

The present SEPSIS-SHIELD study demonstrated very high overall accuracy of the TriVerity Test with AUCs of 0.83 and 0.91 for identifying bacterial and viral infections, respectively, compared to the reference standard of clinically adjudicated infection status. We then calculated specificity and sensitivity of each of the TriVerity interpretation bands that innovatively permit to rule-in and rule-out bacterial and viral infections. Importantly, we simultaneously observed very high specificities for the diagnosis of bacterial and viral infection (95.5% and 98.6%, respectively) and very high sensitivities for ruling-out bacterial and viral infections (97.2% and 95.5%, respectively). When incorporating prevalences, these findings translated into high probabilities for ruling-in or ruling-out of bacterial and viral infections. The clinical value of TriVerity results is further supported by the finding that most results fell into the highly actionable but not the less actionable Moderate (middle) band.

The TriVerity Tests offers the potential to assist the ED clinician to make more informed decisions about several clinically relevant and common decisions including the administration of antibiotics, helping to select those patients who would benefit from anti-infective therapy while mitigating anti-biotic overuse, thereby assisting in reducing common side effects of antimicrobials and combating the further emergence of antimicrobial resistance ^26,27. Notably, TriVerity met the expert consensus target product profile for a diagnostic assay to differentiate between bacterial and non-bacterial infections, which can reduce antimicrobial overuse ²⁸. In addition, to the best of our knowledge, the TriVerity Test is the first direct “biomarker” for detection of viral illness; recently introduced protein-based host response tests only distinguish bacterial vs. non-bacterial etiologies not allowing to fully distinguish viral infections from noninfectious inflammation or co-infections ^29,30. Specific and accurate identification of patients with viral infections will be adjunctive to other approaches for ED clinicians in helping them to decide on specific viral diagnostics, antiviral treatment and isolation measures to prevent the spread of infection. Further, the simultaneous presentation of TriVerity Bacterial and Viral scores may reinforce the impact of a single bacterial result in support of antimicrobial stewardship. We also found that TriVerity viral results are stable across different viruses, including SARS-CoV-2. Of importance, TriVerity can also detect bacterial-viral coinfections. While the prevalence of coinfection is low in the ED, ruling out bacterial co-infections is an important feature to allow withholding antibiotics, especially when radiographic or other clinical features introduce diagnostic uncertainty, and during pandemic threats ^31,32. Finally, TriVerity results indicating low probabilities of bacterial and viral infection allow clinicians to focus their efforts on non-infectious diagnoses thereby avoiding ordering of expensive pathogen identification tests and motivating clinicians to find the non-infectious cause of illness.

While the overall accuracy of the bacterial and viral TriVerity results was lower using the forced vs consensus adjudication (an expected finding as cases with uncertain infection status are included in the analysis), the accuracy remained high. It is important to remember that in the context of this study accuracy was determined post-hoc using the best possible reference standard, clinical adjudication ³³. In daily practice in the ED it is envisioned that TriVerity results will be used in addition to provider assessment based on medical history and clinical presentation which is known to be far from accurate.

Bacterial infections were most frequent (61%) compared to viral infections (22.0%) and co-infections (1.6%) while 14.3% of participants did not have an infection. A high proportion of patients were immunosuppressed, but we did not observe significant differences in the accuracy of TriVerity in immunocompetent vs. immunocompromised patients. These findings indicate that the TriVerity Test can be safely used across patients with diverse infections and immune statuses, an important feature for patients presenting to the ED. Importantly, the performance in immunocompromised patients has not been determined in clinical studies for other host response tests as these were validated in patient cohorts excluding immunosuppression.

Overall, TriVerity AUROCs for detecting bacterial infections were significantly higher than those for commonly used biomarkers including PCT, CRP and white blood cell counts. These findings are remarkable as the results of these biomarkers were provided to clinical adjudicators when adjudicating the infection status and thus may have introduced bias favoring the accuracy results for these biomarkers. Notably, the TriVerity Test as the first direct biomarker for the diagnosis of acute viral infections also detected influenza and SARS-CoV-2 infections during the present study (enrollment into SEPSIS-SHIELD occurred before, during, and after the COVID pandemic). The TriVerity Test’s ability to provide both highly accurate bacterial and viral diagnoses could contribute to more appropriate management of patients with suspected infections. The rapid and early identification or ruling-out of bacterial and viral infections could assist in the timely administration of antibiotic or antiviral drugs (or withholding of antimicrobials) supporting antimicrobial stewardship, (cohort) isolation to combat the spread of infections and the appropriate ordering of follow-up tests.

The early identification and treatment of suspected sepsis are critical³⁴ but there is disagreement among major societies regarding exact guidance and timing for optimizing clinical care. For example, the Centers for Medicare & Medicaid Services (CMS)’s SEP-1 bundle, and the Surviving Sepsis Campaign guidelines strongly recommend the rapid administration of antibiotics to patients with suspected sepsis but physician-based societies (e.g., the Infectious Diseases Society of America) have argued that they do not have adequate tools to judge who has sepsis. These challenges contribute to overtreatment, and substantial costs and morbidities associated with patients given unnecessary antibiotics^5,35 including the rise of antimicrobial resistance. At the same time (and driving some of the practice patterns), occult sepsis continues to be overlooked in a significant proportion of ED patients. While qSOFA greater than or equal to 2 conveys a high percentage mortality, further classification of patients with qSOFA 0–1 is needed as the percentage mortality is low, but because the group is larger, still can account for 30% of sepsis deaths³⁶. In the present study, the TriVerity Illness Severity score demonstrated an overall AUC of 0.85 for the prediction of “ICU-level care” within 7 days with very high specificity (98.7% for the Very high band) to rule-in severe illness while also showing high sensitivity (91.8% for the Very low band) for ruling-out severe illness. The TriVerity Test also demonstrated high accuracy in ruling in and ruling out ICU transfer. In a post-hoc analysis, TriVerity was able to fine-tune the risk prediction based on clinical scores, i.e., qSOFA with conjectured use of TriVerity dramatically improved accuracy in those patients clinically at low risk but at significantly increased risk of severe illness (by 7-fold). TriVerity also correctly identified those patients with significantly lower risk that had clinically been identified as high risk. The TriVerity Illness Severity score also provided additional information to aid in identification of patients with sepsis.

Taken together, TriVerity could assist in the fine-tuning of patient management in the ED, with results offering potential for improved clinical decision making, including disposition decisions. Remarkably, the TriVerity severity results not only predicted the short-term outcome of “ICU-level care” but also the longer-term outcome of 28-day mortality (the secondary prognostic endpoint of the study, data not shown), thereby further strengthening the validity of TriVerity illness severity results.

Results of the current study, along with findings of other studies^21–24 point towards broad generalizability of our findings. First, demographics of enrolled participants were representative of common US ED patients with respect to age, sex, race and ethnicity, underlying conditions, immune status and clinical outcomes. While PCT performs ‘moderately’ well in identifying bacterial infections, results have historically been less accurate in Black/African Americans and Hispanic subgroups³⁷. Our finding of significantly superior accuracy of the bacterial TriVerity result across White, Black/African American and other races in the current study offers the opportunity for improved accuracy in the diagnosis of bacterial infections in minority populations where PCT is known to underperform.

Our study has limitations. First, without a true gold standard, we used clinical adjudication as the reference standard to determine the patient infection status³³. As expected, the accuracy of the TriVerity Test for diagnosing bacterial and viral infections was highest in the consensus adjudication population (certain infection status). While consensus adjudication appears to be the most accurate comparator to assess infection status, it removes a substantial fraction of patients. Adding patients with uncertain infection status (forced adjudication, Supplementary Table 2) allowed accuracy to be calculated across the entire cohort of SEPSIS-SHIELD. Second, the prevalence of bacterial (58%) compared to viral (20%) infections and non-infectious diseases (19%) may appear high. However, the observed high bacterial prevalence was likely driven by a) the broad inclusion criteria resulting in approximately 50% of all patients having abdominal, skin and soft tissue or urinary tract diseases (rarely caused by viral pathogens), and b) the seasonality of bacterial vs. viral infection (i.e. respiratory tract infections). Depending on local epidemiology and setting, TriVerity will demonstrate varying positive and negative predictive values in different settings. However, SEPSIS-SHIELD enrolled many patients and over cold and other seasons, and patient characteristics were characteristic of a US ED population. Therefore, we believe that the validity of the results reported here is not affected by these factors. Furthermore, the development of the machine learning-based TriVerity classifiers was based on a large number of datasets from highly diverse cohorts of patients with suspected infections and/or suspected sepsis²⁰. Lastly, the percentage of patients with severe illness (“ICU-level care”) within 7 days was only 10.9%, lower than expected; thus, while results of the TriVerity Illness Severity score indicate high accuracy for the prediction of “ICU-level care”, future validation studies are warranted. Also, we have previously shown consistently high accuracy among patients admitted to the ICU^17,38,39. Interventional and randomized studies to demonstrate the clinical benefit of TriVerity for diagnosing bacterial and viral infections and the prognosis of illness severity in the ED are in the planning or preparation phase.

Importantly, our statistical analysis approach in SEPSIS-SHIELD ensured inclusion of all patients as equivocal adjudication was not allowed (adjudication categories were Yes, Probable, Uncertain or No) and included all patients with results that fell into the (middle) Moderate TriVerity band. This strategy is important because several studies have been published in which statistical analysis excluded both patients with uncertain adjudication (a clinically critical “gray zone” population) and those in the middle (e.g., equivocal) interpretation bands thereby unfairly over-stating the test overall accuracy^40–43.

In conclusion, the TriVerity Test provides rapid, highly accurate and actionable results for the diagnosis and prognosis of patients with suspected acute infection and sepsis, serving a major unmet medical need. The test is highly innovative as it can detect bacterial infection, viral infection, bacterial-viral coinfections and rule out bacterial and viral infections, thereby showing superior accuracy compared to routine biomarkers. TriVerity can also predict the severity of illness, and combined with clinical findings, TriVerity adds granularity and important potential clinical utility for predicting severe infection. Together, these properties together suggest that TriVerity can improve personalized management of patients with suspected acute infections and suspected sepsis for improved overall healthcare outcomes.

Methods

Study design

The “TriVerity in the Diagnosis and Prognosis of Emergency Department Patients with Suspected Infections and Suspected Sepsis” (SEPSIS-SHIELD, clinicaltrials.gov NCT04094818) was conducted between 02 March 2020 and 28 May 2024. SEPSIS-SHIELD was a prospective, non-interventional, minimal-risk study to analyze gene expression and other laboratory data from biological samples collected from patients presenting to the ED with suspected acute infections with at least one abnormal vital sign, or patients with suspected sepsis of any cause with at least two abnormal vital signs and a blood culture order (see below for precise enrollment criteria). Study sites included emergency departments (EDs) of community and academic hospitals and were located at 21 geographically diverse locations throughout the United States and one site in Europe (Supplementary Table 1). The enrollment period was divided into a “FROZEN” phase (02 March 2020 to 16 Feb 2023) and a “FRESH” phase (08 Dec 2023 to 28 May 2024). During the “FROZEN” phase of the study, blood samples were frozen and sent to one of two reference laboratories to perform the TriVerity Test after thawing the frozen samples. During the “FRESH” phase of the study, the TriVerity Test was performed on freshly obtained blood samples (no freezing) at the enrolling site.

Patients, sample, and data collection

Eligibility included ED patients who met all of the following inclusion criteria: 1) age ≥ 18 years, 2) suspected acute infection, e.g., respiratory, urinary, abdominal, skin and soft tissue infection, meningitis/encephalitis, or any other infection and 3) at least one vital sign change (heart rate > 90 beats/minute, temperature > 38°C or < 36°C, respiratory rate > 20 breaths/minute or PaO₂ of < 60 mmHg or SpO₂ < 90%, systolic heart pressure < 100 mmHg, altered mental status per clinical exam) OR suspected sepsis of any cause defined by a blood culture order by the treating physician and at least 2 vital sign changes, 3) able to provide informed consent or consent by a legally authorized representative. Patients were ineligible if they met any of the following exclusion criteria: 1) Patient-reported treatment with systemic antibiotics, systemic antiviral agents, or systemic antifungal agents within the past 7 days prior to the ED visit (use of antiviral treatment for HIV, hepatitis B and hepatitis C, topical antibiotics, topical antivirals, or topical antifungal agents, anti-herpes prophylaxis, peri-operative [prophylactic] antibiotics and a single dose of antimicrobials during the present ED visit [< 10 hours before blood draw] did not result in exclusion), 2) patients who were receiving palliative or hospice care, or those who were receiving limited interventional care, 3) prisoners or those mentally disabled or unable to give consent (consent by legally authorized representatives were accepted), 4) patients who were receiving experimental therapy or who were already enrolled in an interventional clinical trial in which a patient received some type of intervention, which could include but was not limited to investigational drugs, medical devices, or vaccines and 5) patients previously enrolled in this clinical trial. Of note, patients with any form of immunosuppression were eligible for SEPSIS-SHIELD.

All patients had routine samples collected as per the standard of care and all patients were managed as per the standard of care independent of the study conduct. The study intervention consisted of blood collection via venipuncture (2.5 ml whole blood into PAXgene^Ⓡ Blood RNA Tube (PreAnalytix), 5 ml of whole blood for central laboratory testing of serum C-reactive protein and PCT) and a nasopharyngeal swab collection from those patients with suspected upper respiratory tract infections. Patients were followed for up to 28 days, and those discharged earlier were contacted via phone-call to collect follow-up information (re-admission etc.). Data collection occurred from the day of presentation in the ED (covering 7 days before enrollment) to the follow-up phone call after day 28 using a case report form. Data collected included detailed information on demographics, medical history, clinical, laboratory, and imaging findings as well as information related to management including treatment with antibiotics, ICU-level treatment etc. All data were stored pseudonymized in a secure database.

Endpoints

The primary endpoints for the bacterial etiology and viral etiology were determined by independent adjudicators following a transparent and standardized clinical adjudication process that was developed for the SEPSIS SHIELD trial by a multidisciplinary team of physicians and laboratorians described elsewhere (Whitfield et al., DMID 2024). In brief, two independent physicians reviewed comprehensive clinical, laboratory and other patient information to adjudicate the presence or absence of bacterial and viral infections into Yes, Probable, Unlikely and No categories (no equivocal adjudication allowed); these adjudicators were randomly chosen from a pool of 12 ED physicians. Discordant cases were resolved by one of two expert physicians (experienced and involved in generating the adjudication protocol) who were blinded to the initial reviewer results. Cases adjudicated as Yes and No by the adjudicators (subgroup with certain infection status) formed the consensus adjudication cohort presented in the main body of this report. Cases adjudicated as Probable or Unlikely were forced into categories of Yes/Probable and No/Unlikely to form the forced adjudication cohort (all patients but including uncertain and certain adjudications). The severity outcome of patients required “ICU-level care” was met if a patient received any of the following during hospitalization: mechanical ventilation, vasopressor use, and/or renal replacement therapy.

TriVerity Test

The TriVerity Test (formerly HostDx or InSep, pending FDA clearance) is a gene expression profiling assay that quantifies the relative expression of 29 host response genes from 2.5 ml whole blood collected in a PAXgene Blood RNA tube. The TriVerity System (Supplementary Fig. 2A) is comprised of the TriVerity Cartridge and the Myrna Instrument with embedded software and proprietary pre-processing and machine-learning classification algorithms²⁰ that process the data and deliver the results. The system is designed for single-sample testing. The TriVerity Test generates three scores corresponding to 1) the likelihood of a bacterial infection, 2) the likelihood of a viral infection, and 3) the severity of the patient’s illness. They fall into actionable interpretation bands ranging from Very high to High, Moderate, Low, and Very low (Supplementary Fig. 2B). Analytical and other details of the TriVerity System are described elsewhere (Rasania et al., in preparation).

Statistical Methods

The analysis population consisted of all participants who provided consent, met all inclusion criteria, met no exclusion criteria and had their PAXgene blood sample collected and successfully processed (resulting in three TriVerity scores). The overall analysis population was further divided into diagnostic and prognostic analysis populations with additional requirements. The diagnostic endpoint analysis population included participants who had a completed adjudication for bacterial and viral infection to define the ground truth defined as either a certain (Yes or No) or uncertain (Probable or Unlikely) infection status. The diagnostic population was further classified as Consensus (only contains participants with a certain [Yes or No] adjudication) and Forced (all patients adjudicated as [Yes or Probable] vs. [No or Unlikely]). The primary prognostic endpoint analysis population included participants who had information available on the presence of severe illness (acute need for mechanical ventilation, vasopressor, and/or RRT) within 7 days. The secondary prognostic endpoint included those participants that met the primary endpoint above and/or 28-day mortality.

Each of the bands for the TriVerity Test were assessed in terms of likelihood ratio, sensitivity, specificity, and predictive values (post-test probabilities). The calculation is dependent on the band (whether intended as a rule-in band with a case being a true positive, or a rule-out band with a case being a false negative). For continuous variables (e.g., age, white blood cells), results were summarized with the numbers of observations, means, standard deviations, medians and ranges and confidence intervals. The Fisher's exact test was used to determine the significance of differences in the need of “ICU-level care” across patients with different infection types. P-values for differences between the AUROCs of the TriVerity Test and biomarker concentrations or biomarker counts, clinical scores (qSOFA), and differences across races were calculated using DeLong’s test, whereas p values comparing immunocompromised and immunocompetent patients as well as COVID-19 vs non-COVID-19 patients were calculated using bootstrapping. 80% confidence intervals are given where appropriate. Statistical analyses were performed in SAS version 9.4, or R software version 4.2.

Ethics statement

This study was approved by local or central Institutional Review Boards or Independent Ethics Committees and written permission was obtained from each patient. The study was conducted with the highest respect for the individual patients and in accordance with the protocol, the ethical principles that have their origin in the Declaration of Helsinki, the informed consent regulations stated in Title 21 Code of Federal Regulations (CFR), Part 50, International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use (ICH), GCP (E6) § 4.8, as well as all applicable local and FDA regulations.

Data availability

Ethics statement

This study was approved by local or central Institutional Review Boards or Independent Ethics Committees and written permission was obtained from each patient. The study was conducted with the highest respect for the individual patients and in accordance with the protocol, the ethical principles that have their origin in the Declaration of Helsinki, the informed consent regulations stated in Title 21 Code of Federal Regulations (CFR), Part 50, International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use (ICH), GCP (E6) §4.8, as well as all applicable local and FDA regulations.

Data availability

Acknowledgements

We like to thank all the participating patients, local staff at enrolling sites, central and reference laboratories. We are very grateful for the expert support of the clinical trial team at Inflammatix (Shailee Rasania, Ashley Prasse Miller, … , Lara Gettys, Chance Koy, Ashwin Harani,…, Cici Lu, Michelle West).
(anybody from PD, Ragheb?).

Authors and Affiliations

Oliver Liesenfeld¹

Clinical Affairs

Inflammatix Inc.

540 Oakmead Pkwy

Sunnyvale, CA 94085

Sanjay Arora²

University of Southern California

Department of Emergency Medicine

1200 North State Street

Room 1011

Los Angeles, CA 90033

Thomas (Tom) Aufderheide³

Medical College of Wisconsin

Department of Emergency Medicine

8701 Watertown Plank Road, 3rd Floor

Milwaukee, WI 53226

Casey M. Clements⁴

Mayo Clinic

200 First Street SW

Rochester, MN 55905

Elizabeth DeVos⁵

University of Florida, Jacksonville

655 West 8th Street

Jacksonville, FL 32209

Miriam Fischer⁶

Medstar

Medstar Washington Hospital Center

100 Irving St, NW. East Bldg 4111

Washington, DC 20010

Evangelos J. Giamarellos-Bourboulis⁷

University of Athens

ATTIKON University Hospital

4th Department of Internal Medicine;

1 Rimini Street

Athens, Greece 12462

Stacey House⁸

Washington University

660 S Euclid Avenue

St. Louis, MO 63110

Roger L. Humphries⁹

University of Kentucky

800 Rose Street Room M53

Lexington, KY 40536

Jasreen Kaur Gill¹⁰

Henry Ford Hospital

2799 W Grand Blvd.,

Department of Emergency Medicine, CFP 268

Detroit, MI 48202

Edward Liu¹¹

Hackensack Meridian Health

Hope Tower at Jersey Shore University Medical Center

19 Davis Avenue,

Neptune, NJ 07753

Sharon E Mace¹²

Cleveland Clinic

9500 Euclid Ave. E-19

Cleveland, OH 44195

Larissa May¹³

University of California, Davis

4150 V Street, PSSB Suite 2100

Sacramento, CA 95817

Edward Michelson¹⁴

Texas Tech University

5001 El Paso Dr. MSC 41003

El Paso, TX 79905

Tiffany Osborn¹⁵

Washington University

660 S Euclid Avenue

St. Louis, MO 63110

Edward Panacek¹⁶

University of South Alabama

2451 University Hospital Drive

10th Floor, Suite L

Mobile, AL 36617

Richard Rothman¹⁷

Johns Hopkins Hospital

Johns Hopkins Adult Emergency Department, Room G1076

1800 Orleans Street

Baltimore, MD 21287

Wesley H. Self¹⁸

Vanderbilt University Medical Center

2525 West End Avenue

Suite 600

Nashville, TN 37203

Howard Smithline¹⁹

Baystate Medical Center

759 Chestnut Street

Springfield, MA 01199

Jay Steingrub¹⁹

Baystate Medical Center

759 Chestnut Street

Springfield, MA 01199

Paul Van Heukelom²⁰

University of Iowa

200 Hawkins Drive, 1008 RCP,

Iowa City, IA 52242

Alexandra Weissman²¹

University of Pittsburgh School of Medicine, Department of Emergency Medicine

Iroquois Building, Suite 400A, 3600 Forbes Ave

Pittsburgh, PA 15213

Mathew Wilson⁶

Medstar Washington Hospital Center

100 Irving St, NW. East Bldg 4111

Washington, DC 20010

Donna M. Wolk²²

Geisinger Medical Center

100 North Academy Ave

Danville, PA 17822

David Wright²³

Emory University

Department of Emergency Medicine

Emory University School of Medicine 49 Jesse Hill, Jr. Drive, SE

Suite 126

SE Atlanta, GA 30303

Ljubomir Buturovic¹

Inflammatix Inc.

540 Oakmead Pkwy
Sunnyvale, CA 94085

Yehudit Hasin-Brumshtein¹

Inflammatix Inc.

540 Oakmead Pkwy
Sunnyvale, CA 94085

Nandita Damaraju¹

Inflammatix Inc.

540 Oakmead Pkwy
Sunnyvale, CA 94085

Cici Lu¹

Inflammatix Inc.

540 Oakmead Pkwy

Sunnyvale, CA 94085

Natalie Whitfield¹
Inflammatix Inc.

540 Oakmead Pkwy
Sunnyvale, CA 94085

Purvesh Khatri²⁴

Institute for Immunity, Transplantation and Infection

Center for Biomedical Informatics Research, Department of Medicine

School of Medicine, Stanford University

Stanford, CA 94304

Timothy E. Sweeney¹

Inflammatix Inc.

540 Oakmead Pkwy
Sunnyvale, CA 94085

Nathan I. Shapiro²⁵

Beth Israel Deaconess Medical Center

1 Deaconess Road

Boston, MA 02215

Corresponding author

Correspondence to Oliver Liesenfeld, [email protected]

Singer, M., et al. The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3). Jama 315, 801-810 (2016).
Evans, L., et al. Surviving Sepsis Campaign: International Guidelines for Management of Sepsis and Septic Shock 2021. Critical care medicine (2021).
Gaieski, D.F., et al. Impact of time to antibiotics on survival in patients with severe sepsis or septic shock in whom early goal-directed therapy was initiated in the emergency department. Critical care medicine 38, 1045-1053 (2010).
Ferrer, R., et al. Empiric antibiotic treatment reduces mortality in severe sepsis and septic shock from the first hour: results from a guideline-based performance improvement program. Critical care medicine 42, 1749-1755 (2014).
Rhee, C., et al. Infectious Diseases Society of America Position Paper: Recommended Revisions to the National Severe Sepsis and Septic Shock Early Management Bundle (SEP-1) Sepsis Quality Measure. Clinical infectious diseases : an official publication of the Infectious Diseases Society of America (2020).
Prescott, H.C. & Iwashyna, T.J. Improving Sepsis Treatment by Embracing Diagnostic Uncertainty. Annals of the American Thoracic Society 16, 426-429 (2019).
Ducharme, J., et al. A multi-mRNA host-response molecular blood test for the diagnosis and prognosis of acute infections and sepsis: proceedings from a clinical advisory panel. Journal of personalized medicine (2020).
Gunsolus, I.L., Sweeney, T.E., Liesenfeld, O. & Ledeboer, N.A. Diagnosing and Managing Sepsis by Probing the Host Response to Infection: Advances, Opportunities, and Challenges. Journal of clinical microbiology 57(2019).
Jain, S., Self, W.H., Wunderink, R.G. & Team, C.E.S. Community-Acquired Pneumonia Requiring Hospitalization. The New England journal of medicine 373, 2382 (2015).
Ohnuma, T., et al. Association of Appropriate Empirical Antimicrobial Therapy With In-Hospital Mortality in Patients With Bloodstream Infections in the US. JAMA Netw Open 6, e2249353 (2023).
Simchen, E., et al. Survival of critically ill patients hospitalized in and out of intensive care. Critical care medicine 35, 449-457 (2007).
Boussina, A., et al. Impact of a deep learning sepsis prediction model on quality of care and survival. NPJ Digit Med 7, 14 (2024).
Sundrani, S., et al. Predicting patient decompensation from continuous physiologic monitoring in the emergency department. NPJ Digit Med 6, 60 (2023).
Sweeney, T.E., Wong, H.R. & Khatri, P. Robust classification of bacterial and viral infections via integrated host gene expression diagnostics. Science translational medicine 8, 346ra391 (2016).
Sweeney, T.E., Shidham, A., Wong, H.R. & Khatri, P. A comprehensive time-course-based multicohort analysis of sepsis and sterile inflammation reveals a robust diagnostic gene set. Science translational medicine 7, 287ra271 (2015).
Sweeney, T.E., et al. A community approach to mortality prediction in sepsis via gene expression analysis. Nature communications 9, 694 (2018).
Mayhew, M.B., et al. A generalizable 29-mRNA neural-network classifier for acute bacterial and viral infections. Nature communications 11, 1177 (2020).
Ducharme, J., et al. A Multi-mRNA Host-Response Molecular Blood Test for the Diagnosis and Prognosis of Acute Infections and Sepsis: Proceedings from a Clinical Advisory Panel. Journal of personalized medicine 10(2020).
He, Y.D., et al. The Optimization and Biological Significance of a 29-Host-Immune-mRNA Panel for the Diagnosis of Acute Infections and Sepsis. Journal of personalized medicine 11(2021).
Buturovic, L., Mayhew, M., Luethy, R., Choi, K., Midic, U., Damaraju, N., Hasin-Brumshtein, Y., Pratap, A., Adams, R.M., Fonseca, J., Srinath, A., Fleming, P., Pereira, C., Liesenfeld, O., Khatri, P., Sweeney, T.E. Development of Machine Learning Classifiers for Blood-based Diagnosis and Prognosis of Suspected Acute Infections and Sepsis. ArXiv(2024).
Galtung, N., et al. Prospective validation of a transcriptomic severity classifier among patients with suspected acute infection and sepsis in the emergency department. Eur J Emerg Med (2022).
Bauer, W., et al. A Novel 29-Messenger RNA Host-Response Assay From Whole Blood Accurately Identifies Bacterial and Viral Infections in Patients Presenting to the Emergency Department With Suspected Infections: A Prospective Observational Study. Critical care medicine 49, 1664-1673 (2021).
Safarika, A., et al. A 29-mRNA host response test from blood accurately distinguishes bacterial and viral infections among emergency department patients. Intensive Care Med Exp 9, 31 (2021).
Kostaki, A., et al. A 29-Mrna Host Response Whole-Blood Signature Improves Prediction of 28-Day Mortality and 7-Day Intensive Care Unit Care in Adults Presenting to the Emergency Department with Suspected Acute Infection and/or Sepsis. Shock 58, 224-230 (2022).
Mi, M.Y., Klompas, M. & Evans, L. Early Administration of Antibiotics for Suspected Sepsis. The New England journal of medicine 380, 593-596 (2019).
Pulia, M., Redwood, R. & May, L. Antimicrobial Stewardship in the Emergency Department. Emergency medicine clinics of North America 36, 853-872 (2018).
Tamma, P.D., Avdic, E., Li, D.X., Dzintars, K. & Cosgrove, S.E. Association of Adverse Events With Antibiotic Use in Hospitalized Patients. JAMA Intern Med 177, 1308-1315 (2017).
Dittrich, S., et al. Target Product Profile for a Diagnostic Assay to Differentiate between Bacterial and Non-Bacterial Infections and Reduce Antimicrobial Overuse in Resource-Limited Settings: An Expert Consensus. PloS one 11, e0161721 (2016).
Shapiro, N.I., et al. Diagnostic Accuracy of a Bacterial and Viral Biomarker Point-of-Care Test in the Outpatient Setting. JAMA Netw Open 5, e2234588 (2022).
Novak, D., et al. MeMed BV testing in emergency department patients presenting with febrile illness concerning for respiratory tract infection. Am J Emerg Med 65, 195-199 (2023).
Ram-Mohan, N., et al. Using a 29-mRNA Host Response Classifier To Detect Bacterial Coinfections and Predict Outcomes in COVID-19 Patients Presenting to the Emergency Department. Microbiol Spectr 10, e0230522 (2022).
Bauer, W., et al. Detection of Viral Infection and Bacterial Coinfection and Superinfection in Coronavirus Disease 2019 Patients Presenting to the Emergency Department Using the 29-mRNA Host Response Classifier IMX-BVN-3: A Multicenter Study. Open Forum Infect Dis 9, ofac437 (2022).
Whitfield, N.N., et al. A standardized protocol using clinical adjudication to define true infection status in patients presenting to the emergency department with suspected infections and/or sepsis. Diagnostic microbiology and infectious disease 110, 116382 (2024).
Newman-Toker, D.E. & Sharfstein, J.M. The Role for Policy in AI-Assisted Medical Diagnosis. JAMA Health Forum 5, e241339 (2024).
Pakyz, A.L., et al. Impact of the Centers for Medicare and Medicaid Services Sepsis Core Measure on Antibiotic Use. Clinical infectious diseases : an official publication of the Infectious Diseases Society of America 72, 556-565 (2021).
Freund, Y., et al. Prognostic Accuracy of Sepsis-3 Criteria for In-Hospital Mortality Among Patients With Suspected Infection Presenting to the Emergency Department. Jama 317, 301-308 (2017).
Linnander, E.L., et al. Mitigating structural racism to reduce inequities in sepsis outcomes: a mixed methods, longitudinal intervention study. BMC health services research 22, 975 (2022).
Brakenridge, S.C., et al. A Transcriptomic Severity Metric That Predicts Clinical Outcomes in Critically Ill Surgical Sepsis Patients. Crit Care Explor 3, e0554 (2021).
Brakenridge, S.C., et al. Evaluation of a Multivalent Transcriptomic Metric for Diagnosing Surgical Sepsis and Estimating Mortality Among Critically Ill Patients. JAMA Netw Open 5, e2221520 (2022).
van Houten, C.B., et al. A host-protein based assay to differentiate between bacterial and viral infections in preschool children (OPPORTUNITY): a double-blind, multicentre, validation study. The Lancet. Infectious diseases 17, 431-440 (2017).
Halabi, S., et al. Host test based on tumor necrosis factor-related apoptosis-inducing ligand, interferon gamma-induced protein-10 and C-reactive protein for differentiating bacterial and viral respiratory tract infections in adults: diagnostic accuracy study. Clin Microbiol Infect 29, 1159-1165 (2023).
Bachur, R.G., et al. A rapid host-protein test for differentiating bacterial from viral infection: Apollo diagnostic accuracy study. J Am Coll Emerg Physicians Open 5, e13167 (2024).
Carroll, K.C. Assessment of MeMed BV assays for differentiating between bacterial and viral respiratory infections. Expert review of molecular diagnostics (2024).

Yes there is potential Competing Interest. O. Liesenfeld, L. Buturovic, Y. Hasin-Brumshtein, N. Damaraju, C. Lu, N. Whitfield, T. Sweeney are employees and stock option holders of Inflammatix Inc.. P. Kathri is stock option holder of Inflammatix. T. Aufderheide received research funding from Inflammatix, Zoll, Abbott, Memed, and Cytovale, and consulting honoraria from Medtronic. E. Giamarellos-Bourboulis reports honoraria and consultation fees from Abbott Products Operations, bioMérieux, Brahms GmbH, GSK and Sobi (granted to the National and Kapodistrian University of Athens); independent educational grants from AbbVie, InCyte, Novartis and UCB (granted to the National and Kapodistrian University of Athens) and from Abbott Products Operations, bioMérieux Inc, Johnson & Johnson, MSD, and Swedish Orphan Biovitrum AB (granted to the Hellenic Institute for the Study of Sepsis); and funding from the Horizon 2020 European Grants ImmunoSep and RISCinCOVID and the Horizon Health grants EPIC-CROWN-2, POINT and Homi-Lung (granted to the Hellenic Institute for the Study of Sepsis). L. May has received honoraria from Inflammatix and Thermofisher and has been an advisor to Cytovale, Roche, and Biomerieux. R. Rothman served as a PI for the Inflammatix Reproducibility trial, received a grant from NIH POC Center grant (NIBIB) and served on advisory boards from Cytovale, Memed, and Cepheid. W. Self serves as a consultant for Bayer and Cytovale. N. Shapiro received Blue Jay Diagnostics, Inflammatix, and Lumos, and speaker honoraria from Diagnostic Robotics and Prenosis. All other authors do not declare COIs.

NMEDA136305rs.pdf
Reporting Summary
NMEDA136305epc.pdf
Editorial Policy Checklist
SupplementaryTablesandFigs.docx

Download PDF

Version 1

posted

You are reading this latest preprint version

Rapid and Accurate Diagnosis and Prognosis of Acute Infections and Sepsis from Whole Blood Using Host Response mRNA amplification and Result Interpretation by Machine-Learning Classifiers

Status:

Version 1

Abstract

Figures

Main

Results

Characteristics of study participants at the time of ED presentations

Infection status and anatomical location of infection

Accuracy of TriVerity for the diagnosis of bacterial infection

Accuracy of TriVerity for the diagnosis of viral infection

Prognostic accuracy of the TriVerity Severity score

Discussion

Methods

Study design

Patients, sample, and data collection

Endpoints

TriVerity Test

Statistical Methods

Ethics statement

Data availability

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1