How to Optimize ICU Decision Making with Clinical Intelligence

The volume of physiologic data in the ICU has outgrown the way clinicians process it. Monitors, ventilators, laboratory systems and the electronic health record (EHR) each hold part of the picture, and the work of assembling that picture still falls to the care team at the bedside, under time pressure and amid constant alarms.

This paper sets out the four challenges critical care teams face: delayed recognition of deterioration, inefficient escalation and de-escalation of support, extended length of stay, and unplanned readmission. It then describes what a clinical intelligence layer needs to do to close that gap, and the four requirements against which any platform in the category can be assessed. The evidence cited is drawn from peer-reviewed literature, US agency data and published outcome studies.

The ICU decision-making problem

Critical care teams manage large volumes of physiologic data from monitors, ventilators, laboratories and the EHR, under tight time constraints. A single ICU bed can generate thousands of data points an hour, and the patient’s state is held across several systems that were designed for different purposes.

Most of the data that reaches the patient record is captured at intervals. Typically, flowsheet vital signs are documented every hour, so a trend that develops over minutes is visible only in retrospect. In a study of 857 pediatric cardiac ICU patients, significant deviations in heart rate, blood pressure, respiratory rate and oxygen saturation were absent from the EHR between 48% and 58% of the time when compared with automatically acquired data.¹

The bedside monitor holds the continuous signal and communicates through alarms. Alarm fatigue is well documented and is linked to desensitization and missed deterioration.² The care team is therefore working from an EHR record that arrives late and a live signal that is difficult to trust.

Four places decisions can go wrong

Each of the four challenges below follows the same pattern. The information needed to act exists somewhere in the unit, and the decision is delayed or misjudged because that information was not synthesized at the required time.

Delayed recognition of deterioration

Serious deterioration usually develops over hours, with abnormal vital signs detectable before the event. Patients admitted to intensive care from the ward have frequently had documented physiologic abnormalities for hours before transfer, and registry analysis of in-hospital cardiac arrest shows abnormal vital signs to be common in the one to four hours preceding the event.³

Early warning scores were developed to formalize this window. A systematic review of 95 studies found that many scores in clinical use had methodological weaknesses in their development and validation and might perform less well than expected.⁴ Recognition therefore depends heavily on the vigilance of individual clinicians.

Inefficient escalation and de-escalation

While escalation decisions often receive the most clinical attention, decisions to end ventilation or vasoactive support also carry consequence. Damuth et al.’s meta-analysis found that patients treated with prolonged mechanical ventilation have poor long-term survival, and Gaies et al. showed that, in infants after cardiopulmonary bypass, higher vasoactive support in the first 48 hours is associated with greater morbidity and mortality.⁵ Additional hours of support are a cost to the patient as well as to the unit.

De-escalation is also harder to standardize. While the evidence that a patient is deteriorating tends to present itself, the evidence that a patient is improving must be assembled from stable or gradually improving trends across several parameters. One problem is that readiness assessments for extubation or weaning are often performed at fixed times and rely on the data available in the record at that moment.

Extended ICU length of stay

ICU beds account for around 12% of staffed beds in US acute care hospitals and a disproportionate share of hospital cost.⁶ Under DRG reimbursement, each case is paid at a fixed rate that assumes a benchmarked length of stay. Each day beyond that is an additional cost borne by the hospital.⁷

Unplanned readmissions

Readmission carries a cost at two levels. A patient readmitted to the ICU is exposed to further risk and stays in hospital longer, and a patient readmitted to hospital after discharge exposes the hospital to penalty under the CMS Hospital Readmissions Reduction Program.⁸ Physiologic abnormalities at the point of ICU discharge are a recognized risk factor for readmission,⁹ and where care teams have reviewed continuous physiologic data during rounds and built a data-informed care plan, unplanned ICU readmissions have been shown to fall.¹⁰

Why the existing technology does not close the gap

Bedside monitors present the current moment. They are designed to display live waveforms and to alarm when a parameter crosses a threshold, and they perform that role well. They hold little context about the patient’s course and do not combine parameters into a view of trajectory.

The EHR holds the longitudinal record and is designed for documentation, billing and compliance. It captures discrete data points at intervals and displays them in separate tabs and charts so the patient’s complete situation is difficult to see in one view.

Both systems are essential, and neither is designed to synthesize continuous physiologic data into a decision-ready view.

What a clinical intelligence layer needs to do

Clinical intelligence software organizes high-frequency physiologic data and EHR data into time-aligned views and provides adjunctive risk information to support timely clinician review. It sits above the EHR and alongside the monitors, and its outputs supplement clinician judgment, alarms and monitoring devices.

Four requirements define whether a platform in this category can address the challenges above.

• Continuous, contextual analysis of physiologic data. It should take high-resolution data directly from monitors, ventilators and laboratory systems, combine it with EHR data, and analyze it continuously. Any risk signal it produces should show the inputs behind it, so the clinician can question the output and rule out a sensor anomaly.
• Hospital-configured clinical pathways. It should run the institution’s own protocols for extubation readiness, vasoactive weaning and transfer against the patient’s live data, and make adherence to those protocols visible to the team.
• Bedside visualization that supports the care team. It should present current and historical data in one time-aligned view that can be used on rounds, at handoff and between disciplines, so that everyone works from the same picture of the patient’s course.
• Regulatory clearance and published evidence. Any risk index used to inform care should carry FDA clearance with documented indications for use, and the platform’s association with clinical outcomes should be published in the peer-reviewed literature for the patient population in question.

The Etiometry Platform

The Etiometry Platform is a vendor-neutral clinical intelligence platform that aggregates continuous physiologic data from patient monitors, ventilators, laboratory systems and the EHR into one longitudinal patient view. Backed by 11 FDA clearances, it supports clinician-directed escalation and de-escalation decisions, and its use has been associated with published outcomes in cardiac ICUs, including a reduction of up to 36% in ICU length of stay,¹¹ a 41% reduction in unplanned ICU readmissions,¹² a 30% reduction in time on ventilator with an automated spontaneous breathing trial tool,¹³ and a 29% reduction in vasoactive infusion duration.¹⁴†

References:

1. Lowry AW, Futterman CA, Gazit AZ. Acute vital signs changes are underrepresented by a conventional electronic health record when compared with automatically acquired data in a single-center tertiary pediatric cardiac intensive care unit. J Am Med Inform Assoc. 2022;29(7):1183–1190. https://pmc.ncbi.nlm.nih.gov/articles/PMC9196691/
2. Agency for Healthcare Research and Quality. Alarm Fatigue. In: Making Healthcare Safer III. Rockville, MD: AHRQ; 2020. https://www.ncbi.nlm.nih.gov/books/NBK555522/
3. Hillman KM, Bristow PJ, Chey T, et al. Duration of life-threatening antecedents prior to intensive care admission. Intensive Care Med. 2002;28(11):1629–1634. https://doi.org/10.1007/s00134-002-1496-y Andersen LW, Kim WY, Chase M, et al. The prevalence and significance of abnormal vital signs prior to in-hospital cardiac arrest. Resuscitation. 2016;98:112–117. https://pmc.ncbi.nlm.nih.gov/articles/PMC4715919/
4. Gerry S, Bonnici T, Birks J, et al. Early warning scores for detecting deterioration in adult hospital patients: systematic review and critical appraisal of methodology. BMJ. 2020;369. https://pmc.ncbi.nlm.nih.gov/articles/PMC7238890/
5. Damuth E, Mitchell JA, Bartock JL, Roberts BW, Trzeciak S. Long-term survival of critically ill patients treated with prolonged mechanical ventilation: a systematic review and meta-analysis. Lancet Respir Med. 2015;3(7):544–553. https://doi.org/10.1016/S2213-2600(15)00150-2 Gaies MG, Gurney JG, Yen AH, et al. Vasoactive-inotropic score as a predictor of morbidity and mortality in infants after cardiopulmonary bypass. Pediatr Crit Care Med. 2010;11(2):234–238. https://doi.org/10.1097/PCC.0b013e3181b806fc
6. Society of Critical Care Medicine. Critical Care Statistics. Reports 112,369 ICU beds out of 916,752 staffed beds (12.2%) in US acute care hospitals, based on the AHA 2023 annual survey. https://www.sccm.org/communications/critical-care-statistics
7. Centers for Medicare and Medicaid Services. Acute Inpatient PPS. https://www.cms.gov/medicare/payment/prospective-payment-systems/acute-inpatient-ppsAmerican Hospital Association. The Cost of Caring: Challenges Facing America’s Hospitals as They Care for Patients in 2026. March 2026. https://www.aha.org/costsofcaring
8. Centers for Medicare and Medicaid Services. Hospital Readmissions Reduction Program (HRRP). https://www.cms.gov/medicare/payment/prospective-payment-systems/acute-inpatient-pps/hospital-readmissions-reduction-program-hrrp
9. Kramer AA, Higgins TL, Zimmerman JE. Intensive care unit readmissions in U.S. hospitals: patient characteristics, risk factors, and outcomes. Crit Care Med. 2012;40(1):3–10. https://doi.org/10.1097/CCM.0b013e31822d751e
10. Gaies M, Olive MK, Owens GE, et al. Methods to Enhance Causal Inference for Assessing Impact of Clinical Informatics Platform Implementation. Circ Cardiovasc Qual Outcomes. 2023;16(2). https://pubmed.ncbi.nlm.nih.gov/36727516/
11. Salvin JW, Baronov D, Laussen PC, et al. The Impact of a Real-Time Physiologic Data Analytic Index on Length of Stay in Neonates Following Surgery for Congenital Heart Disease. Circulation. 2017;136(Suppl 1). https://www.ahajournals.org/doi/10.1161/circ.136.suppl_1.20603
12. [ii] Gaies M, Olive MK, Owens GE, et al. Methods to Enhance Causal Inference for Assessing Impact of Clinical Informatics Platform Implementation. Circ Cardiovasc Qual Outcomes. 2023;16(2). https://pubmed.ncbi.nlm.nih.gov/36727516/
13. Clark MG, Borasino S, Zaccagni HJ, et al. Automated spontaneous breathing trial performance tool is associated with improved outcomes following pediatric cardiac surgery: a single-center retrospective study from Alabama, USA. J Pediatr Crit Care. 2025;12(1):1–7. https://doi.org/10.4103/jpcc.jpcc_60_24
14. Gazit AZ, Futterman C, Baronov D, et al. Risk Analytics Clinical Decision Support Decreases Duration of Vasoactive Infusions Following Pediatric Cardiac Surgery: A Multicenter Before and After Clinical Trial. Crit Care Med. 2025;53(7)–e1364. https://pubmed.ncbi.nlm.nih.gov/40298483/