Multimodal discordance triage

1. Three patient vignettes#

Vignette A. Three discordances in one shift#

Daniel, 11 years, 33 kg, severe TBI day 4 with ICP probe, bilateral NIRS, TCD probe, cEEG, NPi every 2 h. Across one nursing shift: (i) ICP 22 (rising), NIRS rSO2 unchanged at 65% (discordance 1: ICP says raised pressure, NIRS says oxygen delivery preserved); (ii) TCD MFV stable at 80, PI 1.3 (discordance 2: TCD does not corroborate the ICP rise); (iii) NPi falling from 3.8 to 2.4 on the right side (discordance 3: pupillometry shifts in one direction while other modalities are stable). Total time to interpret all three: 20 minutes. Question: how does the bedside team structure the response when three discordances arrive at once? 1 2

Vignette B. The mismatched ECMO patient#

Esmé, 4 years, 16 kg, VA-ECMO post-cardiac arrest day 3. Non-pulsatile circulation. NIRS bilateral 65% (intact). TCD shows minimal flow signal on the right (poor window or genuine flow change). cEEG continuous on both sides. ICP not measured (ECMO). NPi 3.5 bilaterally. Question: when one of the modalities cannot give a signal (TCD on the right), how do we triage the remaining modalities, and what does this teach about modality robustness? 3 4

Vignette C. The two-monitor agreement, one-monitor outlier#

Imani, 8 years, 26 kg, severe TBI day 2. ICP 14, PRx +0.05 (intact), bilateral NIRS rSO2 70/72 (symmetric), cEEG continuous, NPi 4.0 bilaterally; PbtO2 18 (borderline low). Question: five modalities agree on "stable, intact autoregulation, adequate perfusion"; one modality (PbtO2) is the outlier. Is PbtO2 a true signal of regional tissue oxygen distress, or is it a probe artefact? 5 6

2. The clinical question#

For each of these children: when multimodal monitors disagree, what is the structured triage that converts noisy multimodal data into a bedside decision, and how does the structure differ across resource settings and patient subtypes?

3. Background#

Multimodal neuromonitoring in the modern neuro-ICU produces an information stream that can include ICP, MAP, CPP, PRx, NIRS rSO2, COx, TCD MFV and PI, Mx, PbtO2, Px, cEEG, NPi, microdialysis L/P and glucose, and clinical exam every 1 to 2 hours. With this many channels, disagreement is the norm, not the exception. The clinical task is to interpret disagreement, not to demand perfect concordance.

Three principles from the consensus literature.

The 2014 Neurocritical Care Society and ESICM MMM consensus established that no single modality is the gold standard for all aspects of brain monitoring; the value of multimodality lies in cross-validation. 1

The 2025 Figaji paediatric MMM consensus formalised a resource-stratified approach: bundles defined by available resources, with explicit guidance on how to interpret combinations and discordances. 2

The 2024 Helbok paediatric MMM update emphasised the role of clinical exam as the anchor for interpretation; even with the most sophisticated monitoring stack, clinical exam is the modality with the longest validation history and the lowest cost. 7

Sources of discordance. Disagreement between modalities arises from several mechanisms:

1. Signal quality issues. Probe drift, transducer zero error, electrode contact problems, motion artefact, sensor displacement. These are the most common cause of apparent discordance and the first to check.
2. Regional vs global measurements. ICP is global (whole intracranial space); NIRS is regional (frontal cortex under the probe); TCD is vessel-specific (MCA, ACA, PCA); PbtO2 is local (within millimetres of the probe). Regional injury or focal disease produces real regional discordance.
3. Macrovascular vs microvascular physiology. PRx (macro) and COx (micro) can disagree in sepsis-driven shunting and other microvascular pathologies.
4. Time-scale differences. PRx integrates over 5 to 30 minutes; clinical exam is instantaneous; NIRS shows changes within seconds. Different modalities respond on different time-scales.
5. Calibration and reference values. PRx thresholds (0.0 to +0.3 borderline; > +0.3 impaired) and COx thresholds (similar) are population-derived and have individual variation.
6. True physiological signals. Discordance is sometimes the real signal: sepsis microvascular shunting, regional injury, autoregulation transition, or evolving brain injury.

Why does triage matter? The bedside team often has minutes, not hours, to interpret the multimodal stack and act. A structured triage prevents action paralysis (waiting for full concordance) and over-reaction (treating every transient discordance). The goal is reliable, justifiable, time-bounded bedside decisions.

4. The multimodal picture#

5. Decision tree#

flowchart TD
  Disc[Modalities disagree] --> Clinical[Step 1: clinical exam]
  Clinical --> Sigq[Step 2: signal quality on each modality]
  Sigq --> Region{Regional vs global?}
  Region -->|Regional| Local[Interpret per probe location]
  Region -->|Global| Validated[Step 3: trust most-validated for context]
  Local --> Action[Decision: act on the highest-confidence signal]
  Validated --> Action
  Action --> Default{Still uncertain?}
  Default -->|Yes| Threshold[Step 4: fall back to default thresholds]
  Default -->|No| Document[Step 5: document and hand over]
  Threshold --> Document
  Document --> Reassess[Reassess in 30 to 60 min]

6. Step-by-step bedside actions#

For Daniel (11 y, severe TBI, three discordances). Times are from the first observed discordance.

1. 0 to 5 min: pause; do not act on a single transient value. Confirm the discordance is sustained over 2 to 3 consecutive 5-minute windows. Single-window outliers are usually artefact.
2. 5 to 15 min: clinical exam. Quantified GCS (eye, motor, verbal where possible); pupillary exam with the pupillometer; focal signs (asymmetric movement, posturing); NPi trend. Document.
3. 15 to 20 min: signal quality check across all monitors.
   • ICP: transducer zeroed against atmosphere; waveform morphology (P1 sharp, P2 smaller, P3 smaller still in healthy compliance); no damping.
   • NIRS: probe contact (no lifting; no hair interference; no ambient light leak); recent reapplication if needed.
   • TCD: insonation depth and angle; spectral envelope quality; vessel identity confirmed (carotid tap response).
   • PbtO2 if present: probe position on imaging; recent calibration; not recently inserted (probe trauma settles over 24 to 48 h).
   • cEEG: technician notes; electrode contact; montage; artefact-free epochs.
4. 20 to 25 min: regional vs global interpretation. Is the discordance regional (one probe shows something the others do not, but the others are in a different anatomic location)? Or global (all probes in similar regions but reading differently)? Regional discordance is often informative (focal injury, evolving haematoma, vasospasm) and is interpreted locally.
5. 25 to 30 min: modality validation hierarchy for the population. In paediatric severe TBI: clinical exam, ICP, and PRx are the most validated. NIRS rSO2 trending is validated. TCD vasospasm detection is validated. NPi is validated for objective brainstem function. cEEG is validated for seizures.
6. 30 to 40 min: act on the highest-confidence signal. For Daniel: the ICP rise is sustained, ICP is the most-validated TBI modality, clinical exam shows falling NPi on the right (regional sign), and TCD does not contradict the ICP rise (PI 1.3 is consistent with elevated distal resistance). Plan: empirical osmotherapy (3% saline 5 mL/kg = 165 mL), urgent CT head, consider sedation deepening, watch the right NPi trend.
7. 40 to 60 min: post-action reassessment. Recheck all modalities after osmotherapy. Has ICP fallen? Has NPi recovered? Has clinical exam improved? Document.
8. 60 to 90 min: imaging if not already. CT head for the discordance investigation; especially when regional signs (NPi asymmetry) suggest a focal lesion.
9. Ongoing: document the discordance interpretation in the chart. The next shift should not have to rediscover the discordance from scratch. Pattern recognition over hours is the bedside skill.
10. Family communication. Brief the family if a significant intervention occurred; align with the broader prognostic conversation.

7. Management endpoints#

Success looks like: the discordance is interpreted and acted on; the action produces an observable change (ICP falls, NPi recovers, NIRS asymmetry resolves); subsequent discordances are interpreted in the context of the established pattern.

Failure looks like: the discordance is dismissed without investigation; action is taken on a single transient value; signal quality is not checked; subsequent shifts rediscover the same discordance.

When to escalate:

• Sustained discordance with clinical deterioration, urgent imaging and consultant review.
• Suspected probe malfunction in a critical modality, replace or recalibrate; bridge with non-invasive substitute.
• Multiple modalities deteriorating together (concordant worsening), this is no longer discordance but real global decline; escalate aggressively.

When to de-escalate:

• Discordance resolves (signal quality fix; physiology recovers).
• Patient stable; modalities trending back toward baseline.
• Family communication current.

8. Variant subsections#

8.1 The "everything looks fine but the patient does not" scenario#

Multimodal monitors show acceptable values; clinical exam is deteriorating. This is the most concerning discordance: the patient knows something the monitors do not. The clinical exam is the anchor. Possible explanations: subtle seizure activity (get cEEG on if not already); evolving regional ischaemia missed by NIRS probe placement; medication effect; metabolic disturbance not yet captured by labs; psychological state in the conscious patient. Trust the clinical exam and investigate broadly.

8.2 The "monitors look bad but the patient is fine" scenario#

Multimodal monitors show worrying values; clinical exam is stable or improving. Possible explanations: signal quality issue across multiple modalities (movement artefact, repositioning); medication effect (e.g., bolus of vasopressor altering all upstream variables briefly); recovery phase (post-osmotherapy ICP can drop quickly; PRx can take longer to normalise). Check signal quality; recheck in 30 to 60 minutes; do not over-react.

8.3 The persistent unilateral NIRS asymmetry#

One NIRS probe consistently reads 10 to 15% lower than the contralateral side over hours. Differential diagnosis: focal injury under the lower probe; scalp pathology (haematoma, oedema); probe positioning over different cortex; chronic asymmetry from prior injury. Action: ultrasound under the probe to exclude scalp haematoma; verify probe positioning; CT if a focal lesion is suspected.

8.4 PRx-NIRS-TCD all suggest different autoregulation states#

PRx +0.2 (borderline), COx +0.5 (impaired), Mx 0.0 (intact). The three indices sample three different physiological windows. The interpretation: macrovascular (Mx) is intact at the proximal vessel level; ICP-MAP correlation (PRx) is borderline; tissue-level (COx) is impaired. The pattern fits microvascular dysfunction (sepsis, evolving inflammation). The action: investigate the inflammatory driver; trust the macrovascular indices for CPP targeting; do not over-react to the tissue-level signal without investigating the cause.

8.5 PbtO2 outlier when other modalities agree#

Five modalities agree on "stable"; PbtO2 reads low. Possible explanations: probe in peri-contusional tissue (true regional signal; should target higher CPP for this region per BOOST-II / BOOST-III); recently inserted probe (the 24 to 48 hour settling period); probe malfunction. Imaging to verify probe position; calibration check; if real, raise CPP or address the upstream haemodynamic cause. 5 8 6

8.6 cEEG showing seizures while clinical exam appears stable#

Non-convulsive seizures in a sedated patient. Discordance is between cEEG (showing electrographic ictus) and clinical exam (stable, sedated, paralysed). The cEEG is correct; the clinical exam is uninformative under paralysis and sedation. Treat the NCSE; do not delay because the patient "looks fine". 9 10

9. Multimodal integration matrix#

10. Worked alternative scenarios#

10.1 What if all monitors agree on rapid deterioration?#

Concordant worsening across modalities is no longer discordance; it is global deterioration. Escalate aggressively: imaging, surgical consult, ICP-directed therapy, family communication. The discordance triage was for when monitors disagree; concordance changes the action.

10.2 What if signal quality is degraded across multiple monitors simultaneously?#

Possible causes: bed movement, repositioning, suctioning, family contact. Check the nursing notes for recent activity. Recheck values 5 to 10 minutes after the perturbation. If signals settle, the discordance was movement-related; if they persist, investigate physiologically.

10.3 What if there is a new monitor of uncertain validation?#

Emerging modalities (skin-on-temple non-invasive ICP devices, novel NIRS subtraction algorithms, automated qEEG seizure detection) have less validation base. Treat their signals as supportive, not primary, until validated. The 2025 paediatric MMM consensus explicitly addresses this issue.

11. Outcome data#

• Leroux 2014 MMM consensus: the foundational document on multimodal monitoring; no single modality is gold standard for all aspects; the value of multimodality is cross-validation. 1
• Figaji 2025 paediatric MMM consensus: resource-stratified bundles; explicit guidance on combination interpretation and discordance. 2
• Helbok 2024 paediatric MMM update: clinical exam as the anchor; emphasises the role of structured trending. 7
• BOOST-II (Okonkwo 2017): the regional PbtO2 outlier scenario; BOOST-II showed that PbtO2-directed CPP management is feasible and may improve outcomes; informs how to interpret PbtO2 outliers. 5
• BOOST-3 (Bernard 2025): continues the PbtO2-directed line of evidence. 8
• NIRS limitations (Andresen 2014; Davies 2017): addresses signal quality issues; informs the signal-quality step of triage. 11 12
• Brady 2010 paediatric COx; Lee 2009 non-invasive autoregulation: the validation base for non-invasive autoregulation indices. 13 14
• Rivera-Lara 2017 autoregulation review: comprehensive review of indices and their respective strengths. 15
• Cerebral microdialysis consensus (Hutchinson 2015): addresses the role of MD in the multimodal stack and the importance of regional interpretation. 16

12. Pitfalls#

• Treating discordance as error. Discordance is information; interpret it.
• Demanding concordance before acting. The bedside often requires action on the highest-confidence signal before all modalities agree.
• Ignoring signal quality. Probe contact, transducer zero, electrode drift are the most common cause of apparent discordance.
• Acting on a single 5-minute window. Confirm sustained discordance over 2 to 3 windows before major action.
• Forgetting clinical exam. The exam is the anchor; quantified GCS and NPi every 1 to 2 hours is the floor.
• Mistaking regional for global. ICP is global; NIRS and PbtO2 are regional. Anatomical context matters.
• Over-reliance on the newest monitor. Validation matters; emerging modalities are supportive, not primary.
• Failing to document the discordance interpretation. The next shift should inherit the analysis, not redo it.

13. Pediatric considerations#

14. Combine with#

• ICP modality page: invasive ICP, waveform morphology, calibration.
• NIRS modality page: rSO2 interpretation, scalp contamination.
• TCD / TCCD modality page: MFV, PI, vessel-specific monitoring.
• PbtO2 modality page: tissue oxygen, BOOST-II and BOOST-III.
• Pupillometry / NPi page: quantified pupillometry.
• Integration: PRx vs COx discordance: the specific PRx-COx case.
• Integration: Resource-limited bedside MNM: triage when fewer modalities are available.
• Foundations: autoregulation: the Lassen curve and the indices.

15. Evidence summary#

16. Recent literature (2022 to 2025)#

• Figaji 2025 paediatric MMM consensus formalises the resource-stratified discordance triage. 2
• Helbok 2024 paediatric MMM update addresses the integration of emerging modalities into the triage framework. 7
• BOOST-3 (Bernard 2025) extends BOOST-II PbtO2-directed care to a larger trial; informs the regional PbtO2 outlier scenario. 8
• Pediatric PbtO2 review (Figaji 2024) consolidates current practice; addresses signal interpretation and placement effects. 6
• Smartphone-based pupillometry and other emerging low-cost modalities are entering the triage framework with provisional status.
• Automated qEEG seizure detection is improving; positive predictive value varies and is supportive rather than primary in current practice.

17. Self-check#