Since Kehlet introduced Enhanced Recovery After Surgery (ERAS) in the late 1990s, multimodal, evidence-based perioperative care has substantially reshaped surgical practice^[1]. Through preoperative optimization, minimally invasive surgery, and early mobilization and feeding, ERAS shortens hospital stay by ~1.9 days and reduces complications by ~29%^[2,3]. Yet recovery is still deemed to begin within 24 h after surgery - by which time hypothalamic-pituitary-adrenal (HPA)-axis activation, disturbed cerebral perfusion pressure (CPP), systemic inflammation, and transient immunosuppression are already underway^[4].

In thoracic surgery, the tubeless technique developed by Professor Jianxing He’s team has advanced the field^[5]: by avoiding invasive tubes and combining precision sedation with regional blocks, carefully selected patients undergo Video-Assisted Thoracoscopic Surgery (VATS) while breathing spontaneously^[6,7], which avoids or attenuates stressors such as mechanical ventilation and deep sedation rather than removing surgical stress itself. Building on these modifications, we propose using “Immediate Recovery Surgery (IRS)”. This ERAS-based framework brings recovery milestones forward, enabling suitable patients to resume oral intake, mobilize, and communicate on the day of surgery. IRS is offered as a domain-specific refinement within ERAS and as a hypothesis for evaluation, not as a validated replacement, since its application is predominantly thoracic (pulmonary nodules and early-stage lung cancer).

In parallel, machine-learning-based perioperative decision-support tools are gradually moving from retrospective model development toward evaluation in clinical settings. Relevant studies include radiomics- and deep-learning-based approaches that integrate clinical, imaging, and molecular features to support pulmonary nodule risk stratification and individualized pathway planning, as illustrated by the DECIPHER-NODL model^[8]; hypotension warning systems designed to reduce the risk of intraoperative hypotension^[9]; and prospective assessments of machine learning (ML)-generated postoperative risk estimates as adjuncts to clinician judgment^[10]. In China, “Nan-Fang-Zhi-Ma” anesthesia mode and the “Shen-Gong-Shen-Shu” operating-room platform have been publicly reported as emerging implementation examples^[11,12]. Such tools may help refine thoracic perioperative care, though their added value remains to be demonstrated^[13].

We frame the tubeless rationale as a nonlinear, multi-level cascade in which a small reduction in iatrogenic input - avoiding intubation, positive-pressure ventilation, deep neuromuscular blockade, and chest-tube nociception - may attenuate downstream stress, inflammatory, and immune responses. We invoke the “butterfly effect” only as an illustrative heuristic, not a literal description: the surgical stress response is multifactorial, only partly modifiable by any single intervention, and the contribution of each avoided stressor is unquantified. The levels below are therefore a mechanistic hypothesis rather than an established causal chain.

(1) HPA-axis/sympathoadrenal attenuation and cerebral perfusion: nociception and tissue injury drive the HPA axis [corticotropin-releasing hormone (CRH)→adrenocorticotropic hormone (ACTH)→cortisol] and catecholamine release, while intubation, positive-pressure ventilation, and deep sedation impose additional stimuli. Avoiding them may attenuate - not abolish - stress-hormone release; meanwhile, spontaneous negative-pressure ventilation avoids the raised intrathoracic pressure that impedes cerebral venous drainage, helping preserve cerebral perfusion. These effects are plausible but unquantified in tubeless cohorts.

(2) Systemic inflammation: tissue trauma releases damage-associated molecular patterns that activate innate signaling [e.g., nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB)] and drive cytokine release; IL-6 stimulates hepatic C-reactive protein (CRP) production, and mechanical/one-lung ventilation adds ventilator-induced biotrauma. Preliminary thoracic surgery studies suggest a trend toward lower early postoperative IL-6, CRP, and TNF-α levels after non-intubated approaches compared with intubated approaches^[5,15,16,19]. But these data derive from only two thoracic (lung-resection) cohorts and vary with procedure, resection extent, and patient factors; they should therefore be read as preliminary and setting-specific, pending broader validation.

(3) Pharmacological “subtraction”: reducing opioids and neuromuscular blockers lowers the cumulative drug load. Opioids and volatile agents are immunomodulatory, and high opioid exposure has been linked to impaired NK function; an opioid- and relaxant-sparing strategy may attenuate these side effects, though the beneficial effect in tubeless surgery remains unmeasured and likely modest.

(4) Cellular immunity and central sensitization: one high-dimensional single-cell study linked tubeless management with relative NK-cell preservation^[5]. Given the role of NK cells in immunosurveillance, this is a plausible advantage for cancer patients. Still, it rests on a single preliminary dataset, and translation to long-term oncological outcomes remains unproven. For pain, combining regional blockade with reduced upstream drive could act on two contributors to central sensitization - a hypothesized, not demonstrated, dual mechanism.

Whereas the tubeless approach minimizes intraoperative iatrogenic perturbation, artificial intelligence (AI) can shift perioperative management from experience-driven to data-informed. Within the IRS, AI is best viewed not as an autonomous decision-maker but as a decision-support layer spanning the perioperative chain - patient selection, intraoperative control of the narrow margins that spontaneous-breathing anesthesia demands, and earlier detection of deterioration during recovery. Because the IRS depends on well-selected patients tolerating surgery without an artificial airway, catheter, or chest drain, the quality of stratification and vigilance matters more than in conventional pathways. As a concept-proposing Perspective, this article does not present original AI data; the applications below illustrate how AI could support the IRS and are still being developed and prospectively validated.

(1) Preoperative - selection and risk stratification. Previous studies have developed risk scores to stratify postoperative complication risk in patients undergoing non-intubated VATS^[20]. More recently, machine-learning models have also been explored for perioperative risk prediction; for example, one machine-learning analysis of uniportal anatomical segmentectomy reported an accuracy of approximately 70% for predicting prolonged air leak (PAL)^[21]. Air leak is a key safety boundary for omitting chest drainage, such PAL prediction models may help stratify patients and inform selection for non-intubated or tubeless pathways. However, these should be read as illustrative rather than definitive, since they derive from a single-center cohort and performance varies with case mix; any such figure is hypothesis-generating until validated in thoracic tubeless populations. Such tools should support, not replace, judgment - e.g., a high predicted risk could move a borderline candidate toward a conventional intubated approach - with thresholds defined and validated prospectively.

(2) Intraoperative - closed-loop control. Sustaining sedation and analgesia while preserving spontaneous breathing under artificial pneumothorax leaves a narrow margin - inadequate depth risks patient movement and airway compromise, while excessive depth may suppress respiratory drive - making precise anesthetic control essential. AI-based closed-loop systems that optimize the sedation-analgesia balance bear directly on stable spontaneous ventilation and may reinforce opioid/relaxant-sparing goals^[22], though the incremental benefit under non-intubated conditions remains unquantified.

(3) Postoperative - complication prediction. Because the IRS compresses the interval to recovery, early recognition of deterioration is essential as patients mobilize and resume oral intake. Institutional systems such as “Shen-Gong-Shen-Shu” have been described as predicting common postoperative complications (e.g., respiratory or cardiac failure) in real time.

(4) Validation and implementation. The named systems are best regarded as developmental or single-institution prototypes lacking external prospective validation; performance in one specialty or cohort does not automatically transfer to the thoracic tubeless setting. Implementation imposes a learning curve distinct from the surgical/anesthetic one - staff must interpret outputs, recognize unreliable predictions, and avoid automation complacency and alert fatigue - and tools validated at high-volume centers may require recalibration elsewhere. Until externally validated, AI is best positioned as an adjunct that supports rather than replaces clinical judgment.

Perioperative care continues to evolve toward earlier, better-defined recovery. Within an ERAS-based framework, the tubeless approach may reduce intraoperative iatrogenic stress, while AI can support more precise, individualized decisions. For appropriately selected patients - at present principally those undergoing thoracic procedures - this combination offers a plausible route to recovery that is not only faster but also anchored to explicit, measurable milestones. Whether it confers consistent benefit over conventional ERAS and generalizes beyond thoracic surgery remains to be tested in prospective, multicenter studies.