📄 intensive care and critical care medicine

Development of an Intelligent Predictive Indicator System for Weaning and Extubation Timing in Mechanically Ventilated Patients Based on the Delphi Method

This study utilized the Delphi method to develop a comprehensive, expert-validated intelligent predictive indicator system comprising six primary and 58 secondary indicators to guide the timing of weaning and extubation for mechanically ventilated patients, thereby establishing a robust framework for clinical decision support and intelligent assessment in critical care.

Original authors: Li, P., Wang, Y., Zhang, Y., Meng, X., Zhang, H.

Published 2026-08-04

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Li, P., Wang, Y., Zhang, Y., Meng, X., Zhang, H.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Imagine the Intensive Care Unit (ICU) as a high-stakes cockpit where a patient's life support system is the plane. When a patient can't breathe on their own, they need a machine to do the work for them. But there's a tricky moment in every flight: deciding exactly when to let the pilot take the controls back. If you let them take over too early, the plane might crash (the patient struggles to breathe again). If you wait too long, the pilot's muscles might get weak from sitting idle, or the plane might get damaged from the extra weight. Right now, doctors and nurses often have to guess this timing based on experience, like a pilot flying by the seat of their pants without a perfect dashboard. This is risky because guessing wrong can lead to more time on the machine, more infections, or worse outcomes. The big question is: how do we build a super-smart, digital dashboard that tells us the exact right moment to switch from machine breathing to human breathing?

This paper is about building the blueprint for that digital dashboard. A team of researchers from Peking University People's Hospital decided to stop guessing and start designing a "smart" system. They didn't just make up a list of rules; they used a method called the Delphi method, which is like hosting a very serious, multi-round game of "expert consensus." Imagine gathering 21 of the world's best ICU doctors, nurses, and respiratory therapists in a room (virtually, in this case) and asking them, "What are the most important signs that tell us a patient is ready to breathe on their own?"

The researchers started by digging through thousands of old medical studies to find every possible clue they could. They ended up with a messy pile of 204 potential clues, ranging from blood test results to how a patient's lungs feel. Then, they sent this list to their 21 experts. The experts rated every single clue on how important it was, from "not important" to "super important." They also argued about the wording and added new ideas they thought were missing.

After the first round of voting, the team realized they needed to tidy things up. They added 13 new clues (like checking for pain or how much air leaks around the breathing tube) and split some categories apart. They sent a revised list back to the same experts for a second round. This time, the experts were even more sure of their answers. The result? A clean, organized, and highly agreed-upon list of 6 main categories and 58 specific sub-rules that cover the entire journey of a patient on a ventilator.

Here is what they found:

The System is Complete: The new list covers everything from the patient's medical history before they even got to the ICU, to the moment they are taken off the machine, and even what happens right after. It's like a checklist that follows the patient from the moment they arrive until they are safe.
It's Built for Nurses and Machines: The experts made sure the list includes things nurses can actually measure at the bedside, like pain scores and how alert the patient is. They also organized the data so that a computer program (an "intelligent prediction system") could read it easily later.
The Experts Agreed: The doctors and nurses were almost 100% in agreement on the most critical signs, such as the patient's breathing speed, the strength of their lungs, and whether they have a "difficult airway."
The "Maybe" List: There were a few items, like the patient's gender or specific blood sugar levels, where the experts didn't fully agree. The paper suggests these might still be useful, but they need more testing to prove they are truly necessary for the final system.

The paper concludes that this new "indicator system" is ready to be used as the foundation for building a smart computer program. It's not a finished computer program yet, but it is the perfect instruction manual for one. By having a standardized, expert-approved list of what to look for, hospitals can eventually build tools that help nurses and doctors make safer, faster, and more accurate decisions about when to take a patient off a ventilator. The authors are confident that this system is scientifically solid and ready for the next step: testing it in real hospitals to see if it actually helps patients breathe easier.

Technical Summary: Development of an Intelligent Predictive Indicator System for Weaning and Extubation Timing

Problem Statement
Mechanical ventilation is a critical life-support intervention in Intensive Care Units (ICUs), yet the timing of weaning and extubation remains a complex clinical challenge. Current clinical practice relies heavily on physician experience, lacking a standardized, unified evaluation indicator system. Existing tools are often confined to single dimensions, failing to support comprehensive, dynamic, and precise assessments. Premature extubation risks reintubation and mortality, while delayed extubation increases complications and resource consumption. Furthermore, critical care nursing lacks standardized support tools for auxiliary assessment, limiting the effectiveness of nursing interventions. With the rise of artificial intelligence in critical care, there is a pressing need for a scientific, systematic indicator system to serve as the foundation for intelligent prediction and decision-support systems.

Methodology
This study employed a rigorous Delphi method combined with a systematic literature review to construct an indicator system.

Initial Item Pool Generation: Researchers conducted a comprehensive search across six databases (CNKI, Wanfang, VIP, PubMed, Embase, Cochrane Library) up to December 2025, identifying 1,869 articles. After screening and expert interviews with five senior clinicians, an initial pool of 204 items (5 primary, 46 secondary, 153 tertiary indicators) was established.
Expert Panel: A panel of 21 experts from tertiary hospitals across China was selected, comprising critical care physicians, respiratory therapists, and critical care nurses (with a focus on nursing representation). All experts had ≥10 years of experience and intermediate or senior titles.
Delphi Process: Two rounds of consultation were conducted in February and April 2026 via email and WeChat.
- Round 1: Experts rated 203 items on a 5-point Likert scale for importance and provided qualitative feedback.
- Round 2: A revised questionnaire with 211 items was distributed based on Round 1 statistical analysis and feedback.
Statistical Analysis: Data were analyzed using SPSS 26.0. Key metrics included:
- Expert Authority ( $C_r$ ): Calculated as the average of the familiarity coefficient ( $C_s$ ) and judgment basis coefficient ( $C_a$ ).
- Consensus: Measured using Kendall's W coefficient and the coefficient of variation (CV).
- Item Screening: Items were retained, revised, or deleted based on mean importance scores (≥4.00), CV (≤0.25), and qualitative expert recommendations.

Key Results

Expert Engagement and Authority: Both rounds achieved a 100% effective response rate. The expert authority coefficient increased from 0.89 in Round 1 to 0.93 in Round 2, well above the 0.70 threshold for acceptability. The familiarity coefficient notably improved from 0.80 to 0.96, indicating deepening expert understanding of the system.
Consensus: Kendall's W coefficients were statistically significant ( $P < 0.001$ ) in both rounds (0.244 in Round 1; 0.079 in Round 2). The slight decrease in W in Round 2 was attributed to the addition of new indicators and structural reclassification, a common phenomenon in multi-round Delphi studies, rather than evaluation bias.
System Structure: The final system comprises 6 primary indicators and 58 secondary indicators (with associated tertiary indicators). The primary indicators cover the full clinical trajectory:
1. Medical History
2. ICU Admission Assessment
3. Pre-Weaning/Extubation Assessment
4. Weaning Assessment
5. Extubation Assessment
6. Post-Weaning/Extubation Outcomes
Indicator Refinement: The process resulted in 13 additions, 23 modifications, and the separation of the "Weaning/Extubation" dimension into distinct "Weaning" and "Extubation" assessments. Highly endorsed nursing-specific indicators included the RASS score, pain score, cuff leak test, and bedside driving pressure.
Discordant Items: Indicators such as sex, blood glucose, and blood urea nitrogen retained some disagreement (CV > 0.25) even after two rounds. The authors attribute this to a lack of direct causal evidence and differing perspectives between medical and nursing specialties regarding their immediate predictive value. These are designated as "pending verification indicators."

Significance and Claims
The authors claim that this study successfully developed a comprehensive, nursing-operable, and intelligent-system-compatible indicator framework. Key contributions include:

Standardization: The system provides a unified, standardized evaluation tool that bridges the gap between physician decision-making and nursing assessment, addressing the current lack of standardized support tools in critical care nursing.
Intelligent Readiness: By splitting assessment dimensions and standardizing terminology with clear time points, the system is structurally designed to meet the data format requirements of machine learning algorithms and intelligent prediction models.
Clinical Operability: The inclusion of specific, bedside-nursing indicators (e.g., cuff leak test, RASS score) ensures the system is practical for clinical application, not just theoretical.
Foundation for AI: The system serves as a core theoretical basis for developing risk warning and intelligent prediction systems for weaning and extubation timing.

The authors maintain a modest stance, acknowledging that the system requires future weight assignment (via Analytic Hierarchy Process) and large-scale multicenter prospective validation to confirm its predictive performance and practical impact on patient outcomes. They emphasize that the current work establishes the necessary indicator framework but does not yet constitute a fully validated clinical prediction model.

Technical Summary: Development of an Intelligent Predictive Indicator System for Weaning and Extubation Timing

More like this