Perioperative TEAS & EA for Opioid Sparing — Systematic Review & Meta-Analysis

Included RCTs

69 RCTs

See Study Explorer for current patient totals

See Study Explorer for modality breakdown

Primary Efficacy Analysis • Supportive EA Evidence

TEAS vs sham (primary): −14.00 mg [−34.18, +6.19]
EA vs usual care (supportive): −3.94 mg [−19.77, +11.90]

Cross-modality combined synthesis (contextual only, k=7, N=676): MD = −9.91 mg [−20.08, +0.27], p = 0.055

PRIMARY: TEAS vs sham • EA vs usual care is supportive, not a sham-controlled primary • Combined = contextual only

Plain Language: TEAS vs sham is the review's primary efficacy analysis: −14.00 mg IV MME (k=4, p=0.114). EA vs usual care is supportive evidence, not a sham-controlled primary comparison -- no sham-controlled EA trial reports this outcome in absolute morphine equivalents: −3.94 mg (k=3, p=0.397). A cross-modality combined synthesis pooling both is shown only as contextual background, not as an inferential result: −9.91 mg IV MME (k=7, p=0.0545).

Standardized Supporting Analysis

Hedges' g = −0.97

95% CI [−2.09, +0.15] • p = 0.079 • Supporting SMD Analysis (k=7)

StataNow 19.5 BE Validated

In plain language: The average result favours lower opioid use with TEAS/EA, but the uncertainty includes no difference and a small increase. This standardized value is not a dose in milligrams.

Prespecified analysis of 7 trials; study filters do not recalculate this result.

GRADE Certainty

Certainty is adjudicated separately for each synthesis. TEAS and EA differ, so no single rating describes the primary outcome.

12 Outcomes in GRADE SoF

In plain language: Our confidence in the estimated opioid reduction is limited. The true effect could be substantially different because results vary between studies and the estimate is imprecise.

Certainty for the prespecified primary outcome; study filters do not reassess GRADE.

Background & Rationale

Review Guarantor & Team: John Ryan Nual Mendoza (john_ryan.nual_mendoza@med.lu.se), Siv Katrin Peters, Dr. Carina Sjöberg, Dr. Pether Jildenstål • Lund University / Region Skåne

Postoperative opioid consumption remains a cornerstone of acute pain management, yet carries dose-dependent risks including respiratory depression, postoperative nausea and vomiting (PONV), ileus, cognitive dysfunction, and the potential for persistent opioid use. Transcutaneous electrical acupoint stimulation (TEAS) and electroacupuncture (EA) have been proposed as non-pharmacological adjuncts that modulate endogenous opioid peptide release (β-endorphin, enkephalins, dynorphins) through frequency-dependent mechanisms, potentially reducing cumulative postoperative opioid requirements without compromising analgesic adequacy.

Despite over two decades of randomized controlled trials, prior systematic reviews have been limited by pooling heterogeneous modalities, lacking sham controls, and failing to prespecify clinically meaningful thresholds. This review addresses these gaps by implementing rigorous modality stratification (TEAS vs EA), requiring sham/placebo controls for primary analyses, and pre-registering a clinically important opioid-sparing threshold.

PROSPERO Clinical Margin

≥ 10 mg IV MME

with Pain Non-Inferiority margin ≤ +1.0 VAS

Sensitivity: ≥ 8 mg or ≥ 30% relative

Eligibility Criteria

PICOS Framework

Element	Inclusion	Exclusion
Population	Adult surgical patients (≥ 18 years) undergoing any elective or emergency surgical procedure under general, regional, or combined anaesthesia	Paediatric (< 18 y), non-surgical (chronic pain, obstetric labour, ICU sedation), or animal models
Intervention	TEAS: Non-invasive transcutaneous electrical stimulation applied at recognized acupoints via surface electrodes EA: Invasive needle electroacupuncture at acupoints with electrical current	Manual acupuncture (no electrical current), moxibustion, acupressure, laser acupoint stimulation, TENS at non-acupoints
Comparator	Primary: Sham/placebo TEAS or sham EA (identical setup, subthreshold or no current) Supportive: Usual care / no stimulation (open-label)	Active drug comparators only (e.g., TEAS vs nerve block without a no-stimulation control)
Outcomes	Eligibility-conferring outcomes. A trial is eligible only if it reports at least one quantifiable perioperative analgesic outcome: Primary: cumulative 0–24 h postoperative opioid consumption (mg IV morphine equivalents) Key secondary: cumulative 0–48 h postoperative opioid consumption Exploratory: cumulative 0–72 h postoperative opioid consumption Also eligibility-conferring: postoperative pain intensity (VAS/NRS), rescue analgesia, or intraoperative opioid requirement. Extracted but not eligibility-conferring: PONV, time to first flatus and other GI-recovery endpoints, QoR-40/15, PCA demands, length of stay and sleep quality are collected from otherwise eligible trials. Reporting one of these alone, with no eligible analgesic outcome, does not make a trial eligible. This is why Yu Wang et al., JAMA Surgery 2023 (DOI 10.1001/jamasurg.2022.5674) — which reports time to first flatus but no eligible analgesic outcome — was excluded at full text for wrong outcomes.	Studies reporting no eligibility-conferring analgesic outcome within 0–72 h, including trials whose only quantifiable endpoints are GI-recovery, PONV, or recovery-quality measures
Study Design	Randomized controlled trials (RCTs), published in English or Chinese, with full-text availability	Non-randomized studies, observational cohorts, case reports, conference abstracts without full data, reviews

PROSPERO CRD420251090635 Lund University • Registered July 2026

PROSPERO Registration & Protocol History

Prespecified research objectives, PROSPERO registration history, and methodological amendment timeline.

⏱️ Protocol Timeline & Amendments:

1. Protocol Drafting (July 2026): Formal prespecification of primary (0–24 h IV MME) and pain endpoints (Lund University).

2. Literature Searches (July 21–22, 2026): Comprehensive execution across PubMed, Embase, CENTRAL, CINAHL, and ChiCTR.

3. PROSPERO Registration (July 2026): Public registration under CRD420251090635 prior to data synthesis.

4. Methodological Refinement (Aug 2026): Explicitly codified 0–48 h as key secondary and 0–72 h as exploratory endpoint.

5. Database Freeze (Sept 2026): Full-text consensus extraction completed; StataNow 19.5 BE verification locked.

Objective 1 — Primary Opioid Sparing & Durability → Primary Outcome

Quantify 24-h (primary) & 48-h (key secondary) opioid-sparing effects

Synthesis of cumulative 0–24 h primary opioid consumption (mg IV MME), alongside key secondary 0–48 h and exploratory 0–72 h outcomes, stratified by modality. REML random-effects with Knapp–Hartung adjustment and 95% prediction intervals.

Objective 2 — Pain Intensity → Secondary Outcomes

Evaluate postoperative pain intensity at rest and during movement at ~24 h

VAS/NRS 0–10 continuous mean difference. Essential for confirming that opioid sparing does not result in clinically important pain worsening.

Objective 3 — Moderators → Meta-Regression Studio

Identify protocol moderators via prespecified subgroup & Stata meta-regression

Prespecified modality subgroups (TEAS k = 4, EA k = 3) with Knapp–Hartung adjustment. One exploratory meta-regression on modality (p = 0.308); with k = 7 no covariate meets the Cochrane 10:1 rule, so no other moderator is modelled.

Objective 4 — Clinical Importance → Clinical Importance

Determine clinical meaningfulness against prespecified thresholds

≥ 10 mg IV MME with pain non-inferiority ≤ +1.0 VAS (sensitivity: ≥ 8 mg, ≥ 30% relative, ≥ 5 mg).

Objectives 5 & 6 — Secondary → Secondary Outcomes

Supportive syntheses and additional clinical endpoints

PONV (RR), time to first flatus (hours), PCA pump demands, rescue analgesia (RR), intraoperative remifentanil, QoR-40/15, length of stay, and sleep quality.

Objective 7 — Evidence Quality → GRADE Evidence

Assess certainty of evidence using RoB 2 and GRADE

Cochrane Risk of Bias 2 across all 5 domains. GRADE Summary of Findings with downgrade assessment for risk of bias, inconsistency, indirectness, imprecision, and publication bias.

Verified Evidence Synthesis Summary (StataNow 19.5 BE Validated)

Executed with restricted maximum likelihood (REML) and Hartung–Knapp standard errors on defensible, audited study sets:

Primary 24-h Opioid Sparing

−9.91 mg IV MME

95% CI: [−20.08, +0.27] • p = 0.0545 (REML + KH)

Strict Primary (k = 7, N = 676) • Modality: TEAS −14.00 mg, EA −3.94 mg

Resting Pain at ~24h (Target C)

−0.20 points

95% CI: [−0.34, −0.07] • p = 0.0168 (REML + KH)

VAS 0–10 (k = 4, N = 1,898) • High RoB (Xing 2022, Liu 2021); Some concerns (Song 2020, Gao 2022)

Time to First Flatus (Target E)

Hedges’ g = −0.55

95% CI: [−0.88, −0.22] • p = 0.0065 (REML + KH)

GI Recovery (k = 7, N = 638) • standardised measure reported since 2026-09-12 (post hoc; see the protocol amendment), I² = 39.2% • MD −6.79 h [−14.84, +1.27] at I² = 97.3% is not interpretable as one effect

Postoperative Nausea & Vomiting (Target D)

RR 0.65 / 0.54

0–24h: [0.39, 1.07] (p=0.071) • 0–48h: [0.26, 1.11] (p=0.058)

Composite PONV (k = 4 at 0–24 h, k = 2 at 0–48 h) • Stratified by follow-up • I² = 66.9% at 0–24 h

PCA Demands / Button Presses (Target F)

SMD = −1.27

95% CI: [−2.77, +0.24] • p = 0.0866 (REML + KH)

PCA Behavior (k = 8, N = 614) • Standardized effect size

Rescue Analgesia Requirements (Target F)

RR 0.52

95% CI: [0.37, 0.73] • p = 0.0087 (REML + KH)

Strict Rescue (k = 4, N = 312) • 50% Relative Risk Cut • I² = 0.0%

Intraoperative Remifentanil (Target F)

−114.21 µg

95% CI: [−213.72, −14.70] • p = 0.0308 (REML + KH)

Intraoperative Sparing (k = 7, N = 890) • Titrated remifentanil (µg)

Only the primary 24-h opioid card above is split by modality and comparator (TEAS vs sham; EA vs usual care), as the protocol requires. The other cards (Targets C–F) pool across modality and comparator and predate that restratification; see the Forest Plot Gallery below for the modality/comparator composition of each and the current, separately-stratified v34 equivalents where one exists.

Surgical Specialties Distribution (69 trials / 70 reports)

Methodological Architecture & Design Separation

Locked Protocol Rule: Non-invasive surface stimulation (TEAS) and invasive needle electroacupuncture (EA) represent fundamentally distinct physical modalities and are stratified in all primary syntheses. Sham-controlled trials (placebo double-blind) isolate true acupuncture-specific efficacy from non-specific expectation and caregiver bias.

PRISMA 2020 Flow Diagram

Live Study Selection Process from Covidence Extraction Audit

Records Imported 5,100 across 4 electronic databases

Duplicates & Ineligible 2,160 1,652 dupes + 508 automation

Title/Abstract Screened 2,928 2,704 irrelevant excluded

Assessed for Eligibility 210 224 sought, 14 not retrieved; 141 excluded

Studies Included 69 from 70 reports — Yeh 2010 and Yeh 2011 are two reports of one trial

Updated 2026-09-07: This flow diagram now reflects the review team's updated PRISMA 2020 record (source: "PRISMA - Protocol characteristics associated with clinic....docx"), which folds the 7 post-lock source-direct additions (Wu 2016, Gao 2022, Liu 2015, Oztas 2019, Song 2020, Szmit 2021, Zhang 2018) into the eligibility and inclusion counts below. Full-text eligibility screening was revisited: 224 reports were sought, 14 could not be retrieved, 210 were assessed, and 141 were excluded (revised reason breakdown below) — 210 − 141 = 69 records from the database-search route, plus 1 identified via citation searching, gives the 70 reports now included. Those 70 reports describe 69 studies: Yeh 2010 and Yeh 2011 are two reports of one lumbar spinal surgery trial (see the 2026-09-11 unit-of-analysis amendment). PRISMA 2020 keeps reports and studies as separate counts and this review reports both. RoB 2 is complete for all 70 reports in the source workbook (verified 2026-09-07: Corrected_RoB2 has a full 5-domain + overall judgement for every row in Study_Master, no gaps) and is reflected in the Study Explorer and RoB 2 tabs. Only Szmit 2021 contributes to a locked pooled result (the strict primary 24-h opioid analysis and the Target D nausea 0–24h stratum). The dedicated eligibility reconciliation pass for the other six was completed 2026-09-12 and is shown in full on the Study Explorer: 19 study-outcome dispositions, each with the rule behind it and what would unblock it. Three outcomes clear on data grounds and now await review-team admission (Zhang 2018 time to first flatus, unblocked by a validated figure digitisation; Song 2020 24-h pain and 24-h PONV, unblocked because the ITT/per-protocol question holding them turned out to be answered in the source). Two more (Gao 2022 24-h pain and PONV) are blocked by a rule rather than by data and need a ruling rather than a re-read. The remaining thirteen are closed: five because the source does not report the value, two because no defensible derivation exists, and six because the value falls outside the target's own definition. The pass pools nothing — admitting a study to a locked analysis remains a review-team act. Superseded 2026-09-12: this paragraph previously closed by stating that the modality split, comparator split and total randomized-patient count in the “Included” card remained under reconciliation. They no longer do. All three are computed live from the register the Study Explorer renders from, which is itself regenerated from the locked v34 master and checked against it on every build, so the card cannot disagree with the per-study data. Two things changed with that reconciliation: modality and comparator are counted over the 69 trials rather than the 70 reports (48 TEAS / 21 EA, 57 sham / 12 usual care), and the participant figure is the number analysed, summed once per trial (10,618), not the 70-report sum of 10,678. What is not available is a review-wide randomized participant total: only 17 of the 69 trials record a randomized denominator in either the register or their source PDF, and the two channels measure different quantities, so no such total is published.

Identification Databases & Deduplication

Records Identified from Databases 5,100

Total records imported for screening (representing 5,088 unique studies):

Embase (Elsevier.com)1,928 (37.8%)
Cochrane CENTRAL1,698 (33.3%)
PubMed (MEDLINE)1,009 (19.8%)
CINAHL Ultimate465 (9.1%)

➔

Records Removed Before Screening 2,160

Duplicate records removed by Covidence1,651
Duplicate records removed manually1
Marked ineligible by automation tools508

↓

Reconciled: the flow counts studies from this point on, not references. Covidence imported 5,100 references, which resolve to 5,088 studies — 12 references were additional reports of studies already present. The screening arithmetic therefore runs on studies: 5,088 − 2,160 removed = 2,928 screened, exactly the figure in the source record. Subtracting from the 5,100 reference count instead mixes units and produces a spurious 12-record shortfall.

Screening Title & Abstract Review

Records Screened 2,928

Dual independent screening of titles and abstracts by two reviewers against inclusion criteria (postoperative surgical patients, electrical stimulation at acupoints).

➔

Records Excluded (Irrelevant) 2,704

Records failing preliminary screening due to non-surgical populations, non-acupoint electrostimulation, non-clinical models, or review articles.

↓

Eligibility Full-Text Retrieval & Evaluation

Reports Sought for Retrieval 224

Full-text reports sought following title/abstract screening.

➔

Reports Not Retrieved 14

Full text could not be obtained despite attempted retrieval.

Reports Assessed for Eligibility 210

Retrieved full-text reports independently scrutinized in duplicate against explicit PICOS eligibility criteria.

➔

Reports Excluded 141

Exclusion reasons per the review team's updated 2026-09-07 PRISMA record:

Wrong outcome — no eligible perioperative outcome reported ⓘ117 (83.0%)
Wrong setting (chronic / non-surgical)9 (6.4%)
Wrong intervention (manual/moxibustion)9 (6.4%)
Wrong comparator (active drug only)3 (2.1%)
Wrong patient population (pediatric/animal)2 (1.4%)
Wrong study design (non-RCT)1 (0.7%)

↓

Included Systematic Synthesis

Studies Included in Systematic Review & Meta-Analysis 69 RCTs

70 included reports describing 69 randomized controlled trials per the review team's updated 2026-09-07 PRISMA record, as amended on 2026-09-11 (the 70 reports arrive as 69 via the database-search route above, plus 1 identified via citation searching; Yeh 2010 and Yeh 2011 are two reports of one lumbar-spine trial). Full report-level detail, RoB 2, and extracted outcomes for all 70 reports are in the Study Explorer.

Audit & Consensus Status:
• 70 reports describing 69 studies (1 linked cohort)
• RoB 2 complete for all 70 reports (verified 2026-09-07)
• 19/19 P1 issues dispositioned (0 blockers)

Modality, comparator and participant totals (reconciled 2026-09-12):
69 randomized trials from 70 included reports • 10,618 participants analysed across the 69 trials, counting each trial once • 48 TEAS / 21 EA • 57 sham-controlled / 12 usual-care-controlled. A review-wide randomized participant total is not reported: most trials state only their analysed denominators (see Population Characteristics in the Study Explorer).

Computed live from the register the Study Explorer renders from (dashboard/data.js, regenerated from the locked v34 master and checked against it on every build), not restated from the PRISMA docx, which does not itself give participant totals. Modality and comparator are trial properties and are counted over the 69 trials, so they differ by one from the 70-report tallies (48 TEAS / 22 EA, 58 sham / 12 usual care) quoted where the unit is reports. Participants are summed once per trial, so the linked Yeh cohort is no longer double-counted; the earlier figure of 10,678 was the 70-report sum.

PRISMA 2020 checklist — item by item, mapped to this review

Submission requirement Counts re-derived each build

✅ PRISMA 2020 checklist

Bibliographic Search Strategies & Concept Architecture

Multi-Database Search Syntax, Boolean Operators, and Retrieval Statistics

Search Syntax & Boolean Query String Multi-line strategy

Search Concept Map & Intersecting Facets

Domain Facet	Controlled Descriptors (MeSH / Emtree / CINAHL)	Key Natural Language Textwords (Title/Abstract/Keywords)
Intervention: TEAS Non-invasive surface acupoint stimulation	"Transcutaneous Electric Nerve Stimulation"[Mesh] AND "Acupuncture Points"[Mesh]; 'transcutaneous electrical nerve stimulation'/exp AND ('acupuncture'/exp OR 'acupuncture point'/exp); MH "Transcutaneous Electric Nerve Stimulation" AND MH "Acupuncture Points"	TEAS, TAES, "transcutaneous electrical acupoint stimulation", "transcutaneous electric acupoint stimulation", "transcutaneous acupoint electrical stimulation", "electroacupoint stimulation", "acupuncture-like TENS"
Intervention: EA Invasive needle electroacupuncture	"Electroacupuncture"[Mesh]; 'electroacupuncture'/exp; MH "Electroacupuncture"	electroacupunctur, electro-acupunctur, "electro acupuncture", "electric acupuncture", "electrical acupuncture"
Population: Surgical Perioperative surgical patients under GA	"Surgical Procedures, Operative"[Mesh]; "Anesthesia"[Mesh]; 'surgery'/exp; 'anesthesia'/exp; MH "Surgery, Operative+"; MH "Anesthesia+"; MH "Perioperative Care+"	surg, operat, perioperat, peri-operat, intraoperat, intra-operat, postoperat, post-operat, preoperat, anesthe, anaesthe*
Study Design: RCT Filter Validated methodological trial filters	pt "randomized controlled trial"; pt "controlled clinical trial"; 'randomized controlled trial'/exp; 'controlled clinical trial'/de; MH "Randomized Controlled Trials+"	randomized, randomised, placebo, randomly, "double-blind", "single-blind", trial, "intermethod comparison"

Study Characteristics & STRICTA Parameters Explorer

RoB 2 shown for

Study Key	Year	Country	Surgical Category	Modality	Comparator	Randomized N NR where the trial does not report one	Analysed N denominators used in synthesis	Age, mean ± SD intervention / comparator	Female intervention / comparator	Result-Specific RoB 2	Actions

Cochrane Risk of Bias 2 (RoB 2) Traffic Light Studio

Result-specific RoB 2 assessments adhering to Cochrane guidelines. Select an outcome below to inspect result-specific domain judgments or view the overall study summary.

+ Low Risk ? Some Concerns − High Risk ⋯ Outcome Not Reported in Trial

Target Outcome & Timepoint View: Note: RoB 2 judgments used for meta-analysis and GRADE are result-specific. Study-level overview is provided for general reference only.

Study Identifier	D1: Randomization	D2: Deviations	D3: Missing Data	D4: Measurement	D5: Selection	Overall RoB	Signaling Rationale

Source-verified Stata secondary analyses

Analysis

Exploratory Interactive Forest Plot

🔒 PROSPERO Protocol Synthesis Standard: TEAS and EA will not be combined in a pooled estimate. Analyses are stratified by modality. The locked primary strata are TEAS vs sham and EA vs usual care. Browser summaries use DerSimonian–Laird variance estimation, approximate t critical values for continuous intervals, and normal intervals for risk ratios. They are not the authoritative Stata results. Every option uses its established locked target selection. Composite PONV is restricted to 0–24h; other PONV windows and components remain separate in the saved results.

Exploratory browser calculation

Estimator

Binary outcomes apply the Haldane–Anscombe continuity correction (+0.5 to each of the four cells of the 2×2 table, so each arm's denominator gains 1) to all contrasts, not only zero-cell tables — review team decision, 2026-09-11. As of 2026-09-12 this holds everywhere: the Stata analyses above were re-run under the same correction, so this plot and the source-verified models are on one scale. They remain different analyses with different contributing trials, so their numbers are still not expected to match each other. Continuous outcomes are unaffected.

HISTORICAL (2026-09-11 to 2026-09-12): for one day this note also said the saved Stata analyses still used the uncorrected estimator, because the universal correction reached the browser plot first and the Stata re-run was pending. That is no longer true and the paragraph above supersedes it: 05_ponv.do, 07_targetF.do and 12_stratum_compliant_refit.do each now build the corrected 2×2 cells before meta set, and every published binary result comes from those runs. Kept, labelled, because a reader who saw the earlier wording should be able to find out what changed.

ⓘ How to read this plot ⓘ

Study Marker: Marker area reflects study sample size and random-effects statistical weight.
Horizontal Line: 95% confidence interval for each study estimate, reflecting statistical precision.
Summary Diamond: The pooled random-effects average. Diamond width spans the 95% confidence interval.
Line of No Effect: Vertical dashed line (0 for MD/SMD; 1.0 for RR). Values to the left favor TEAS/EA.
Prediction Interval: Extended dashed bracket estimating the plausible true effect in a future comparable clinical setting.

Inc	Study Identifier	Comparison	Intervention Mean ± SD (N)	Control Mean ± SD (N)	Mean Difference [95% CI]	Weight	Forest Plot [95% CI & Prediction]

PROSPERO Objective 4 EXPLORATORY PAIRED OPIOID–PAIN ANALYSIS

Exploratory Paired Opioid–Pain Analysis (Objective 4 Studio)

Independent paired study populations: k = 6 of 7 strict primary trials (N = 628 analysed) ⓘ

Evaluating whether 0–24h Opioid Reduction achieves prespecified clinical-importance benchmarks (Primary: ≥ 10 mg IV MME) without clinically important worsening of postoperative pain (≤ +1.0 VAS margin; upper 95% CI examined).

📥 Paired Dataset (CSV) 📥 Paired Dataset (JSON)

Optimal Benchmark Zone

50.0% of reporting trials

Opioid Sparing ≥ 10 mg + Pain Relief

Moderate Sparing Zone

0.0% of reporting trials

Sparing 5–10 mg + Pain Relief

Minor Sparing Zone

50.0% of reporting trials

Sparing < 5 mg + Pain Relief

Pain Compromise Zone

0.0% of reporting trials

Pain > +1.0 or No Sparing

2D Trade-Off Scatter Matrix (Opioid MD vs Pain MD)

TEAS (Surface) EA (Needle) PROSPERO Threshold Line

X-Axis: Mean Difference in 24h Opioid Consumption (mg IV MME, negative = opioid reduction) • Y-Axis: Mean Difference in 24h Pain Intensity (VAS 0–10, negative = pain reduction)

Quadrant 1: Optimal Benchmark (≥ 10 mg MME Sparing) 3 Trials (50.0%)

Achieves the prespecified primary benchmark (≥ 10 mg IV MME reduction) while maintaining pain non-inferiority (≤ +1.0 VAS difference): Chen 1998, Seevaunnamtum 2016, and Chen 2020. These are single-trial point estimates within a pooled estimate whose 95% confidence interval crosses zero, so this quadrant describes study-level point estimates, not a demonstrated effect.

Quadrant 2: Moderate Sparing (5–10 mg MME) 0 Trials (0.0%)

Trials showing moderate study-level opioid reduction (5–10 mg IV MME) without pain worsening. No trial in the current strict primary set falls in this band at the default ≥ 10 mg threshold. Earlier releases populated this quadrant with Sim 2002, Wong 2006, Yeh 2010, Yeh 2010 ATHM and Zhang 2025 — a cohort that counted the overlapping Yeh reports twice and drew on conditional and non-24-h records outside the primary set.

Quadrant 3: Minor Sparing (< 5 mg MME) 3 Trials (50.0%)

Modest opioid reduction (< 5 mg IV MME) while maintaining pain non-inferiority: El-Rakshy 2009, Yang 2024, He 2026 (hepatectomy/JIS). Observed predominantly where baseline surgical opioid requirement was already low, so the absolute room for sparing is small.

Quadrant 4: Opioid Sparing with Pain Compromise 0 Trials (0.0%)

No trial in the v26 strict primary set exceeded the prespecified pain-worsening margin (+1.0 VAS). All six paired point estimates fell on the pain-reduction side (range −0.80 to −0.03 VAS). Across these trials, study-level opioid reduction was not accompanied by systematic pain worsening. This is a study-level finding across six trials and does not establish zero individual-patient risk.

Current analyses (v34) — models, withdrawals and source holds

v34 analysis set Stratified by modality and comparator

📊 Current Analyses

Manuscript evidence map — what we can write, and what a reviewer will ask

Interpretation layer Working draft — not evidence

📝 Manuscript Evidence Map

Comparison with the prior meta-analysis — why our estimate differs

Discussion material Dispositions re-derived, not copied

📚 Comparison with prior evidence

How the included trials contribute — why 69 trials give 7 in the primary analysis

v33 Evidence Base Study contribution map

🗺️ How the Included Trials Contribute

Retired v33 secondary analyses — superseded by the current v34 set

Primary outcome contribution pathway — which trials carry 24-hour data

Contribution Transparency Derived from the v26 lock

Primary Outcome Contribution Pathway

Study	In systematic review	Reports opioid outcome	24-h candidate	Strict primary	Broader sensitivity	Author contact relevant	Reason / current status

Which 0–24 hour evidence can actually be pooled — tier classification

v33 Tiered Analysis Derivability audit

🧭 Which 0–24 Hour Evidence Can Actually Be Pooled

Every estimate on this panel was fitted in StataNow 19.5 by 07_TIERED_V33/02_STATA/13_tiered_primary_v33.do (S0–S3, sensitivity) or 07_TIERED_V33/02_STATA/14_tiered_tierE_smd_v33.do (Tier E scale-free SMD) and read from 07_TIERED_V33/05_RESULTS/TIERED_ANALYSIS_RESULTS_v33.csv and 07_TIERED_V33/05_RESULTS/TIERED_ANALYSIS_RESULTS_v33_tierE.csv. A sensitivity analysis crossing p = 0.05 does not change a conclusion here: these analyses exist to test robustness and completeness, not to find significance.

Formal Protocol Synthesis Standards Cochrane & PRISMA 2020 Compliance StataNow 19.5 BE Verified

StataNow 19.5 BE Consensus Synthesis & Forest Plots Hub

Locked Protocol Synthesis Standard: In accordance with our PROSPERO protocol, TEAS and electroacupuncture (EA) are never combined into a single grand pooled estimate. Analyses are strictly stratified by modality. Primary comparisons evaluate TEAS versus credible sham TEAS and EA versus sham/control EA. Between-study variance is estimated using restricted maximum likelihood (REML) with Knapp–Hartung confidence intervals and 95% prediction intervals.

📥 Master Pipeline .do (v26) 📥 Primary 24-h .do (v26) 📊 Primary 24-h Dataset (CSV) 📈 All 40 Reconciled Results (CSV) 📥 Target A 0–48 h .do (v26) 📊 Target A 0–48 h Dataset (CSV) 📋 Target A 0–48 h Log 📈 Target A 0–48 h Results, incl. Sensitivity (CSV) 📋 Primary 24-h Log

⚠️

Comparator Hierarchy & Sufentanil Conversion — Audited Caveats

Comparator hierarchy. Per protocol, sham/placebo is the principal comparator and usual-care/no-stimulation is supportive. Auditing the seven strict primary trials against this rule: the TEAS stratum (k=4: Chen 1998, Chen 2020, He 2026, Szmit 2021) is sham-controlled, consistent with the principal comparator. The EA stratum (k=3: El-Rakshy 2009, Seevaunnamtum 2016, Yang 2024) is entirely usual-care / open-label control — none of the three is sham-controlled. One sham-controlled EA candidate exists in the wider 24-h pool (Coura 2011), but it is weight-normalised (µg/kg fentanyl) and cannot be converted to an absolute dose without the group-mean-weight reconstruction the lock prohibits, so it is not strict-poolable. There is currently insufficient sham-controlled EA evidence for a pooled primary sham-controlled EA efficacy estimate; the EA stratum shown throughout this dashboard is a usual-care comparison and should be read as such, not as a sham-controlled efficacy result. As of v33 the two strata are no longer combined into a single headline estimate: the sham-controlled TEAS model is the primary analysis and the usual-care EA model is reported separately as supportive, because pooling them averages two different questions.

Sufentanil conversion — corrected 2026-09-07 (affects Chen 2020, 1 of 7 strict primary trials). The mg-MME factor applied to sufentanil was previously 0.1 mg MME/µg (100:1) — identical to the fentanyl ratio, uncited in the source do-file, and supported by no located reference. It has been corrected to 1.0 mg MME/µg (1000:1), at the most-potent end of the published 250:1–1000:1 range (BC Ministry of Health equianalgesic table: 10–40 µg sufentanil = 10 mg parenteral morphine; FDA sufentanil label: “as much as 10 times as potent as fentanyl” in balanced general anaesthesia, the setting of these trials). The correction makes the estimated effect larger, and it creates statistical significance nowhere: the strict primary remains non-significant (p = 0.089), while the Target A broader-window result moved from p = 0.0030 to p = 0.0730, i.e. from significant to non-significant. Residual uncertainty is published as a sensitivity range across factors 0.1 / 0.25 / 0.5 / 1.0. Full audit: 06_FINAL_ANALYSIS_V26/06_AUDIT/opioid_conversion_audit.csv; sensitivity results: 06_FINAL_ANALYSIS_V26/03_RESULTS/results_sufentanil_conversion_sensitivity.csv.

PRIMARY OPIOID OUTCOME Reconciled & Locked Master Dataset VERIFIED IN STATANOW 19.5 BE Strict Clean Trials: k = 7 RCTs • N = 676 Patients

Primary outcome: cumulative 0–24 h opioid sparing (Modality-Specific Primary: TEAS k=4, EA k=3)

Executed in StataNow 19.5 BE via meta summarize, random(reml) se(kh) predinterval. Modalities and reporting strata rigorously tested.

📌 Primary Inference Policy: Modality-Specific Synthesis (TEAS k=4, EA k=3) TEAS (k = 4: Chen 1998, Chen 2020, He 2026, Szmit 2021; N = 337) and EA (k = 3: El-Rakshy 2009, Seevaunnamtum 2016, Yang 2024; N = 339) are analysed separately for primary inference because they represent distinct stimulation modalities. The combined estimate (k = 7, N = 676) is shown only as an overall contextual synthesis.

⚠️ Methodological Trajectory Notice: Independent Study Cohorts (Not a Longitudinal Trajectory) Different sets of trials contribute to each cumulative time window. These estimates should therefore not be interpreted as repeated longitudinal measurements of the same study population or as a continuous trajectory of treatment effect.

PRIMARY ANALYSIS 1: TEAS vs Sham Double-Blind Sham

TEAS vs Sham TEAS (0–24h)

1. Statistical Evidence

t(3) = −2.21, p = 0.1145

95% Knapp–Hartung CI: [−34.18, +6.19] mg

2. Effect Magnitude

MD = −14.00 mg IV MME

REML + Knapp–Hartung (k = 4)

3. Clinical Importance

Surgical Case-Mix Dependent

Substantial sparing in thoracotomy (Chen 1998: −21 mg); crosses zero under KH.

4. Consistency

I² = 98.59%

τ² = 156.88 • Q = 232.89 (p < 0.0001)

Plain-Language Interpretation: In the double-blind sham-controlled TEAS stratum (k = 4 trials), opioid sparing averaged 14.00 mg IV MME (Knapp–Hartung 95% CI [−34.18, +6.19], p = 0.1145). The pooled estimate favoured lower opioid consumption, but the 95% confidence interval included no effect and is imprecise due to high between-trial variance.

Alternative Model Estimates & Estimator Sensitivity

Normal Wald Approximation (REML): not computed for this stratum; the Knapp–Hartung estimate above is authoritative.

DerSimonian–Laird (DL): MD = −14.02 mg (95% CI: [−27.22, −0.83], p = 0.0373)

Analyzable Cohort: k = 4 RCTs • N = 337 patients

Included Trials: Chen 1998, Chen 2020, He 2026 (hepatectomy/JIS), Szmit 2021

Primary Stratum 2: EA vs Usual Care (k=3, N=339) Open-Label Usual Care

EA vs Usual Care (0–24h)

1. Statistical Evidence

t(2) = −1.19, p = 0.3969

95% Knapp–Hartung CI: [−19.77, +11.90] mg

2. Effect Magnitude

MD = −3.94 mg IV MME

REML + Knapp–Hartung (k = 3)

3. Clinical Importance

Below 10 mg Threshold

Point estimate below 10 mg benchmark; CI crosses zero.

4. Consistency

I² = 77.15%

τ² = 28.47 • Q = 7.67 (p = 0.0216)

Plain-Language Interpretation: In the needle electroacupuncture stratum (k = 3 trials), opioid sparing averaged 3.94 mg IV MME (p = 0.397). The pooled estimate favoured lower opioid consumption, but the 95% confidence interval included no effect and is imprecise.

Alternative Model Estimates & Estimator Sensitivity

Normal Wald Approximation (REML): MD = −3.94 mg (95% CI: [−11.75, +3.88], z = −0.99, p = 0.323)

DerSimonian–Laird (DL): MD = −2.61 mg (95% CI: [−9.27, +4.04], p = 0.443)

Analyzable Cohort: k = 3 RCTs • N = 339 patients

Included Trials: El-Rakshy 2009, Seevaunnamtum 2016, Yang 2024

Supporting Combined Synthesis (k=7, N=676) REML + Knapp–Hartung

Supporting Combined Strict-Evidence Synthesis: 0–24h Opioid Consumption

Methodological Clarification: TEAS and EA are analyzed separately for primary inference because they represent distinct stimulation modalities. The combined estimate is shown as an overall synthesis across all 7 strict direct trials.

1. Statistical Evidence

t(6) = −2.38, p = 0.0545

95% Knapp–Hartung CI: [−20.08, +0.27] mg

2. Effect Magnitude

MD = −9.91 mg IV MME

Standardized: Hedges' g = −0.97 [−2.09, +0.15]

3. Clinical Importance

Below 10 mg Threshold

Primary threshold: 10 mg (point estimate exceeds it; 95% KH CI still crosses zero, so not established).

4. Consistency

95% PI [−39.35, +19.54]

I² = 98.57% • τ² = 113.91 • Q = 254.61 (p < 0.0001)

Plain-Language Interpretation: Across the 7 strict direct trials, perioperative electrical acupoint stimulation reduced 24-hour postoperative opioid consumption by an average of 9.91 mg IV morphine equivalents (p = 0.0545). The point estimate does not reach the prespecified 10 mg IV MME clinical-importance benchmark, and the 95% confidence interval includes no effect, so the analysis does not establish that the benchmark is met. The wide 95% prediction interval ([−39.35, +19.54] mg) crosses zero, showing that effect sizes vary substantially across trials.

Alternative Model Estimates & Estimator Sensitivity

A. REML + Knapp–Hartung (Primary Model): MD = −9.91 mg (95% CI: [−20.08, +0.27], t = −2.38, p = 0.0545, τ² = 113.91)

B. REML + Conventional Wald CI (Unadjusted): MD = −9.91 mg (95% CI: [−18.15, −1.67], z = −2.36, p = 0.0184)

C. DerSimonian–Laird + Conventional Wald CI: MD = −9.73 mg (95% CI: [−16.26, −3.20], z = −2.92, p = 0.0035, τ² = 68.51)

D. DerSimonian–Laird + Knapp–Hartung CI: MD = −9.73 mg (95% CI: [−19.93, +0.48], t = −2.33, p = 0.0585, τ² = 68.51)

Methodological note: Hartung–Knapp accounts for additional uncertainty in random-effects meta-analysis and avoids artificial significance caused by unadjusted Wald intervals when between-study heterogeneity is substantial.

Analyzable Cohort: k = 7 RCTs • N = 676 patients (Strict Direct Evidence Set)

Trials Included: Chen 1998, Chen 2020, El-Rakshy 2009, He 2026 (hepatectomy/JIS), Seevaunnamtum 2016, Szmit 2021, Yang 2024

Secondary Syntheses High Data Completeness

Secondary Clinical Endpoints

Pain at ~24h (Target C: 2 RCTs, N=158): MD −0.18 pts

95% CI [−0.68, +0.33] • t = −2.16, p = 0.1414 (High RoB)

Time to First Flatus (Target E: 6 RCTs, N=596): MD −2.00 hrs

95% CI [−3.14, −0.87] • t = −4.56, p = 0.0062 (Sig) • SMD: −0.46

PONV Risk Ratio (Target D: 2 RCTs per window): RR 0.56 (24h) / 0.52 (48h)

24h: [0.14, 2.26], p = 0.119 • 48h: [0.22, 1.27], p = 0.068

Other opioid timepoints: 48-hour analyses

TARGET A OPIOID OUTCOME Key Secondary 48-Hour Opioid Sparing VERIFIED IN STATANOW 19.5 BE Strict: k = 3 RCTs • N = 1,999 Patients

Cumulative 0–48 h opioid sparing (Strict: k = 3 RCTs, N = 1,999)

Executed in StataNow 19.5 BE via meta summarize, random(reml) se(kh) predinterval for protocol-compliant trials reporting cumulative 48-hour systemic opioid requirements (Chen 2020, Zhang 2023, An 2014).

⚠️ Protocol-Compliant Study Eligibility: Strict Target A includes exactly 3 RCTs ($N=1,999$: Chen 2020, Zhang 2023, An 2014). He 2026 (WJCO) is strictly excluded (reported rescue/intraoperative metrics without 48-h cumulative mass). Wong 2006 is strictly excluded (reported first 3 days / 72 h). Xie 2014 ($N=40$) is evaluated as a broader window sensitivity only.

Strict Target A (Primary Spec) k = 3 RCTs

Strict Target A Pool

−10.27 mg IV MME

95% Knapp–Hartung CI: [−34.83, +14.30] • p = 0.2139

Test of Effect: t(2) = −1.80. The 95% Hartung–Knapp confidence interval included no effect.

Considerable Heterogeneity: τ² = 94.5600, I² = 97.07%, Q = 71.08 (p < 0.0001)

95% Prediction Interval: [−153.55, +133.01] mg IV MME.

Analyzable Cohort: k = 3 RCTs • N = 1,999 patients

Included Trials: Chen 2020 (N=80), Zhang 2023 (N=1,838), An 2014 (N=81)

Mandatory Sensitivity Exclude An 2014

Excluding An 2014

−12.40 mg IV MME

95% Knapp–Hartung CI: [−128.63, +103.84] • t(1) = −1.35, p = 0.4048

Considerable Heterogeneity: τ² = 165.0283, I² = 98.59%, Q = 71.06 (p < 0.0001)

Analyzable Cohort: k = 2 RCTs • N = 1,918 patients

Included Trials: Chen 2020 (N=80), Zhang 2023 (N=1,838)

Direct Means Only Exclude Converted

Excluding Converted

−13.87 mg IV MME

95% Knapp–Hartung CI: [−113.16, +85.43] • t(1) = −1.77, p = 0.3267

Considerable Heterogeneity: τ² = 117.3932, I² = 96.11%, Q = 25.69 (p < 0.0001)

Analyzable Cohort: k = 2 RCTs • N = 161 patients

Included Trials: Chen 2020 (N=80), An 2014 (N=81)

Broader Window Include Xie 2014

Broader Window Pool

−12.30 mg IV MME

95% Knapp–Hartung CI: [−26.74, +2.13] • p = 0.0730

Considerable Heterogeneity: τ² = 79.1542, I² = 96.46%, Q = 101.83 (p < 0.0001)

Analyzable Cohort: k = 4 RCTs • N = 2,039 patients

Included Trials: Chen 2020, Zhang 2023, An 2014, Xie 2014

Separate Proxies PCA Volume Endpoints

48-h PCA Volume Trials

Ao 2021 (TEAS, N=65): MD −17.8 mL

102.8 vs 120.6 mL (P<0.001) • Presses: 6.2 vs 12.3 (P<0.001)

Long 2025 (TEAS, N=106): MD −3.0 mL

86.2 vs 89.2 mL (P=0.008) • Compressions: 7.1 vs 8.6 (P=0.008)

Jin 2023 (EA, N=105): MD −12.5 mL

82.4 vs 94.9 mL (P<0.001) • NAPC: 1.0 vs 9.0 presses (P<0.001)

Other opioid timepoints: 72-hour analyses

TARGET B OPIOID OUTCOME NOT POOLED (Single Study) VERIFIED IN STATANOW 19.5 BE Strict Exact: k = 1 (Yang 2024, N = 180)

Cumulative 0–72 h opioid sparing (Strict: Single Trial Yang 2024, k = 1, N = 180 — NOT POOLED)

Executed in StataNow 19.5 BE. Strict exact 0–72h cumulative systemic opioid consumption was measured by only a single trial (Yang 2024); per Cochrane guidelines, it is not pooled.

⚠️ Methodological Exclusions: Zhang 2025 is strictly excluded from 72-h synthesis (measured POD1 only, and legacy files erroneously treated µg as mg); Xie 2014 is strictly excluded (measured postop 48h only).

Strict Exact 0–72h (k = 1) NOT POOLED

Yang 2024 Alone (0–72h)

−0.50 mg IV Morphine

95% CI: [−4.08, +3.08] • p = 0.7840 (Not Significant)

Not pooled — single study: Only one eligible trial directly reported exact 0–72 h cumulative opioid consumption (Yang 2024, N = 180; EA vs Usual Care in lung cancer surgery). A single trial cannot be meta-analyzed; the estimate above is that one trial's own result, not a pooled effect, and is reported here for completeness rather than as synthesised evidence.

Analyzed Study: Yang 2024 (N = 180: 17.0 ± 12.0 vs 17.5 ± 10.0 mg morphine)

Broader Window Sensitivity k = 2 Trials

Broader Sensitivity (Yang + Wong)

−1.47 mg IV MME

95% Knapp–Hartung CI: [−34.42, +31.48] • t(1) = −0.55, p = 0.6716

Low Inconsistency: τ² = 4.3317, I² = 13.88%, Q = 1.16 (p = 0.2812)

Combines Yang 2024 exact 0–72h with Wong 2006 (EA vs Sham in spinal surgery, N = 25, reporting cumulative opioid over first 3 days: MD = −8.40 mg). Shows no statistically significant extended opioid reduction.

Analyzable Cohort: k = 2 RCTs • N = 205 patients (Yang 2024 N=180, Wong 2006 N=25)

Leave-one-out influence and sensitivity analysis

ROBUSTNESS AUDIT STATANOW 19.5 BE AUTHORITATIVE ENGINE Primary 24-h Opioid Analysis • k = 7 Strict Iterations

Leave-one-out influence and sensitivity analysis

Iteratively omits each trial to evaluate whether overall statistical significance, effect direction, or heterogeneity depends on any single outlier or study artifact. Executed directly via StataNow 19.5 BE meta summarize, leaveoneout on the locked seven-trial primary dataset.

🖼️ View Stata LOO Forest Plot

Each of the seven strict primary trials is omitted once. See the current interval and influence results below.

1. Directional Stability:

Pooled MD remains strictly negative (opioid-sparing) across all 7 omissions, ranging from −5.83 mg (omitting Chen 2020) to −11.67 mg (omitting Yang 2024).

2. Hartung–Knapp Sensitivity:

In 6 of 7 omissions, the Hartung–Knapp 95% confidence interval still crosses zero (p = 0.0498 to 0.1163). Omitting Yang 2024 is the exception (p = 0.0498, CI excludes zero); crossing the p = 0.05 threshold in one single-trial omission does not change the review's overall conclusion.

3. Variance Drivers:

Omitting Chen 2020 reduces τ² the most, from 113.91 in the full k = 7 model to 36.77 (a 68% reduction), identifying it as the largest single contributor to between-study heterogeneity. Omitting Szmit 2021 or Seevaunnamtum 2016 leaves τ² highest (137.89 and 134.53 respectively).

4. Clinical Benchmark Consistency:

4 of 7 omission-model point estimates reach the prespecified 10 mg IV MME benchmark (point estimates range from −5.83 to −11.67 mg); a point estimate reaching a benchmark is not proof of a clinically important effect, since every omission's confidence interval remains wide.

Omitted Trial	Year	Remaining k (N)	Pooled MD (IV MME)	Wald 95% CI	Wald p	Knapp–Hartung 95% CI	KH p	τ² (Between)	I² (%)	DFBETAS

Baseline complete primary synthesis (k=7, N=676): Pooled MD = −9.907 mg IV MME [95% KH CI: −20.079 to +0.265], p = 0.0545, τ² = 113.911, I² = 98.57%. All models estimated via REML.

All forest plots (12 figures)

Stata 19.5 BE Publication Artifacts High-Resolution 300 DPI Forest Plots

StataNow 19.5 BE Forest Plot Gallery

Generated directly from StataNow 19.5 BE via meta forestplot for the strict primary analysis (k = 7) and the secondary/exploratory targets. Click any plot to view or download high-resolution figure.

Modality and comparator note: the primary 24-h opioid outcome above is reported separately by modality and comparator (TEAS vs sham; EA vs usual care), as the review's protocol requires. Targets A–F below (0–48 h and 0–72 h opioid, pain at ~24 h, PONV, time to first flatus, and PCA/rescue/intraoperative-opioid endpoints) predate that modality/comparator restratification and, for several targets, pool TEAS-vs-sham trials together with EA-vs-usual-care trials in a single estimate (for example, Target A combines Chen 2020 and Zhang 2023 [TEAS vs sham] with An 2014 [EA vs usual care]). They are retained here as the review's original broader-window secondary analyses, not as modality-specific inference. Where a properly stratified v34 equivalent now exists for the same outcome concept (intraoperative remifentanil/sufentanil, QoR-40 at 24 h, time to first flatus/defecation, pain VAS at 24 h, PONV at 24 h — each split by modality and comparator, with a result-specific risk-of-bias basis), it is shown in the "Current analyses (v34)" panel earlier in this tab and should be treated as the current evidence for that outcome; the Target A–F figures below supply additional context and are not superseded wholesale, but their point estimates should not be read as modality-specific.

📊 How to Read this Forest Plot

Square/Marker: The estimated treatment effect for each study. Marker area reflects its statistical weight in the random-effects meta-analysis.
Horizontal Line: 95% confidence interval for each study, showing uncertainty around the study estimate.
Diamond: The pooled average treatment effect. The center indicates the point estimate and lateral tips represent the 95% confidence interval.
Vertical Dashed Line (0 or 1): The line of no effect. For mean differences (MD), values to the left of 0 favor TEAS/EA (reduced opioid consumption or pain). For risk ratios (RR), values to the left of 1.0 favor TEAS/EA (lower complication incidence).
Prediction Interval: The dashed/extended bracket representing the estimated range of true effects that could plausibly occur in a future study or comparable clinical setting.

STRICT PRIMARY directly compatible 24-h cumulative opioid outcomes

Primary 24-h Opioid Sparing (IV MME mg): Strict Direct Trials (k = 7)

🔍 View Full Res

StataNow 19.5 Forest Plot - Primary 24-h Opioid MME

StataNow 19.5 BE Primary Synthesis: k = 7 RCTs (N = 676: Chen 1998, Chen 2020, El-Rakshy 2009, He 2026, Seevaunnamtum 2016, Szmit 2021, Yang 2024). Pooled REML MD = −9.91 mg IV MME [95% CI: −20.08 to +0.27], t(6) = −2.38, p = 0.0545 (not statistically significant). Heterogeneity: τ² = 113.91, I² = 98.57%, Q = 254.61 (p < 0.0001). 95% Prediction Interval: [−39.35, +19.54] mg. TEAS stratum (k=4): MD = −14.00 mg [−34.18, +6.19], p = 0.1145; EA stratum (k=3): MD = −3.94 mg [−19.77, +11.90], p = 0.3969. DerSimonian–Laird model: MD = −9.73 mg [KH 95% CI: −19.93 to +0.48, p = 0.0585; unadjusted Wald: −16.26 to −3.20, p = 0.0035, demonstrating artificial significance under unadjusted DL]. Sensitivity excluding High RoB (El-Rakshy 2009, k=6): MD = −11.27 mg [−23.18, +0.64], p = 0.0591.

STRICT PRIMARY same six trials, scale-free metric

Standardized Mean Difference (Hedges' g): Primary 24-h Opioid (k = 7)

🔍 View Full Res

StataNow 19.5 Forest Plot - Primary 24-h Opioid SMD

BROADER SENSITIVITY — NOT THE PRIMARY ANALYSIS adds conditional / derived / proxy 24-h outcomes

Broader 24-h Sensitivity (Hedges' g): strict + conditional trials (k = 9)

🔍 View Full Res

Broader 24-h sensitivity forest plot, standardized mean difference, k=9

Sensitivity only. Adds Coura 2011 and Sim 2002 to the seven strict trials on the scale-free standardized metric: Hedges' g = −0.97 [95% CI: −1.80 to −0.14], p = 0.028, I² = 94.99%. The point estimate is essentially identical to the strict model (g = −0.97 [−2.09 to +0.15], p = 0.079); the interval is narrower because there are more studies, not because the effect is larger, and at I² ≈ 95% this nominal significance carries little weight on its own. Zhang 2025 is no longer in this set (v33). It reports postoperative day 1 rather than an explicit 0–24 h clock window; treating POD1 as 0–24 h is a prohibited assumption, and a scale-free metric changes the unit, not the estimand, so the SMD does not rescue it. With Zhang 2025 retained the model gave k = 10, g = −0.93 [−1.67 to −0.20], p = 0.018 — recorded as OP24_BROADER_SMD_WITH_ZHANG for transparency, not as a headline result. The two Chen 2015 reports remain outside both pooled models (Median/IQR, no derived mean/SD), so no k = 11 pooled result exists. This is not the primary analysis — see the Primary Outcome Contribution Pathway above.

Modality Subgroups

Primary 24-h Opioid Consumption Stratified by Modality (TEAS vs EA)

🔍 View Full Res

StataNow 19.5 Forest Plot - Modality Subgroups

Target A (0–48 h)

Target A: Cumulative 0–48h Opioid Consumption (IV MME mg)

🔍 View Full Res

StataNow 19.5 Forest Plot - Target A 48h Opioid MME

StataNow 19.5 Target A Synthesis: Strict Target A (k = 3 RCTs: Chen 2020, Zhang 2023, An 2014; N = 1,999): REML MD = −10.27 mg IV MME [95% CI: −34.83 to +14.30], t(2) = −1.80, p = 0.2139, τ² = 94.5600, I² = 97.07%. 95% PI: [−153.55, +133.01] mg. Mandatory sensitivity excluding An 2014 (k = 2): MD = −12.40 mg [−128.63 to +103.84], p = 0.4048. Sensitivity excluding converted median Zhang 2023 (k = 2): MD = −13.87 mg [−113.16 to +85.43], p = 0.3267. Broader model including Xie 2014 (k = 4): MD = −12.30 mg [−26.74 to +2.13], p = 0.0730.

Target B (0–72 h)

Target B: Cumulative 0–72h Opioid Consumption

🔍 View Full Res

StataNow 19.5 Forest Plot - Target B 72h Opioid

Target C (Pain ~24h)

Target C: Pain Intensity at Rest Around 24h (VAS 0–10)

🔍 View Full Res

StataNow 19.5 Forest Plot - Target C Pain at Rest ~24h

StataNow 19.5 Target C Synthesis: Strict At Rest ~24h: k = 4 RCTs (Xing 2022, Liu 2021, Song 2020, Gao 2022; N = 1,898). Pooled REML MD = −0.205 points on 0–10 VAS [95% CI: −0.339 to −0.070], t(3) = −4.85, p = 0.0168. Heterogeneity: τ² = 0.00, I² = 0.00%. Changed 2026-09-12 by the post-lock admission of Song 2020 and Gao 2022: this stratum was k = 2, N = 158, MD −0.18 [−0.68, +0.33], p = 0.1413 and not statistically significant. It now is — but Gao 2022 contributes 1,655 of the 1,898 participants and effectively sets the interval, and 0.2 points on a 0–10 VAS is far below any minimally important difference. Statistically significant, clinically negligible. High-risk-of-bias trials in rest pain show no statistically significant analgesic benefit around 24 hours.

Target D (PONV Stratified)

Target D: Postoperative Nausea & Vomiting (Stratified Risk Ratios)

🔍 View Full Res

StataNow 19.5 Forest Plot - Target D PONV Stratified

StataNow 19.5 Target D Stratified Synthesis: Stratum 1: Composite PONV 0–24h (k=4: Zheng 2025, Lu 2021, Song 2020, Gao 2022): RR = 0.65 [95% CI: 0.39 to 1.07], p = 0.0708 (changed 2026-09-12: was k=2, RR 0.56 [0.14, 2.32], p = 0.1225, I² = 0.0%; heterogeneity is now I² = 66.9%, so this pooled RR is less interpretable than its predecessor). Stratum 2: Composite PONV 0–48h (k=2: Xiong 2021, Xing 2022): RR = 0.52 [95% CI: 0.26 to 1.11], p = 0.0581. Stratum 3: Nausea 0–24h (k=3: Yang 2024, Ma 2026, Szmit 2021): RR = 0.60 [95% CI: 0.31 to 1.22], p = 0.0922. Stratum 5: Vomiting 0–24h (k=2: Yang 2024, Ma 2026): RR = 0.58 [95% CI: 0.03 to 13.45], p = 0.2775. Single studies (Luo 2026 alone): Nausea 0–48h RR = 0.46 [0.30 to 0.70]; Vomiting 0–48h RR = 0.34 [0.17 to 0.70]. Every stratum except Composite PONV 0–24 h shows zero statistical heterogeneity (τ² = 0.00, I² = 0.00%); that one is I² = 66.9% since the 2026-09-12 admissions, as stated above. Corrected 2026-09-12: this sentence previously claimed zero heterogeneity for every stratum, contradicting the 0–24 h figure in its own paragraph.

Target E (Gastrointestinal Motility)

Target E: Time to First Postoperative Flatus (Hours)

🔍 View Full Res

StataNow 19.5 Forest Plot - Target E Time to First Flatus

StataNow 19.5 Target E Synthesis: k = 7 RCTs (Zhou 2025, Yang 2020, Yang 2024, Xing 2022, Lu 2022, Ng 2013, Zhang 2018; N = 638). Reported primary measure for this outcome: Hedges' g = −0.55 [95% CI: −0.88 to −0.22], p = 0.0065, I² = 39.2%. The mean difference is −6.79 hours [−14.84 to +1.27], p = 0.0848, with I² = 97.3% — reported, but not interpretable as a single effect at that heterogeneity. Changed 2026-09-12, post hoc — recorded as a protocol amendment in 00_protocol/amendments/2026-09-12_target_e_effect_measure.md, which states what the change does and does not claim and why it is post hoc. Until Zhang 2018 was admitted this stratum was k = 6 with MD −2.00 h [−3.14, −0.87], p = 0.0062 and I² = 0.0%. The reason for preferring the standardised scale is a property of the data, not of the new p value: time to flatus varies about five-fold across these surgical contexts (control-arm means 15.9 h to 80.1 h) and the absolute effect scales with it — the correlation between a trial's control-arm mean and its absolute mean difference is r = 0.91 across all seven trials, and the effect expressed as a percentage of control is about twice as consistent as the effect in hours (coefficient of variation 0.56 vs 1.22). A difference in hours is the wrong scale on which to pool this outcome. This does not fully explain Zhang 2018, which remains an outlier on the relative scale too (−35.9% against −9.4% to −18.6% for the other six) — which is why the standardised model's I² is 39.2% rather than near zero: reduced heterogeneity, not resolved. It is the only open-surgery trial in an otherwise laparoscopic or thoracoscopic set. Sensitivity excluding Ng 2013 (originally reported in days, k = 6): MD = −6.94 hours [−17.03 to +3.14], p = 0.1369, I² = 98.2%. Sensitivity excluding High RoB (k = 6): MD = −7.96 hours [−17.79 to +1.88], p = 0.0921, I² = 95.2%. Both re-run 2026-09-12 and both lost statistical significance (they were k = 5, p = 0.0115 and p = 0.0199 before Zhang 2018 was admitted); at I² above 95% neither is interpretable as a single effect in hours either — the same scale problem the standardised measure answers. Excluded at full text: Yu Wang et al. (JAMA Surg 2023) was excluded due to Wrong Outcomes (reported combined defecation/flatus time without separate first flatus extraction).

Target F (Intraoperative Titration)

Target F: Intraoperative Titrated Remifentanil Requirement (µg)

🔍 View Full Res

StataNow 19.5 Forest Plot - Target F Intraoperative Remifentanil

Target F (Rescue Opioids)

Target F: Postoperative Rescue Opioid Requirement (Risk Ratio)

🔍 View Full Res

StataNow 19.5 Forest Plot - Target F Rescue Opioid Requirement

StataNow 19.5 Target F Rescue Synthesis: Strict binary rescue incidence (k = 4 RCTs: Xie 2014, Yu 2020, Tu 2024, Zhou 2025; N = 312): Pooled REML Risk Ratio = 0.52 [95% CI: 0.37 to 0.73], t(3) = −6.15, p = 0.0087, τ² = 0.00, I² = 0.00%. All binary rescue including Liu 2026 burn (k = 5, N = 398): RR = 0.51 [95% CI: 0.39 to 0.66], t(4) = −7.20, p = 0.0020, τ² = 0.00, I² = 0.00%. A statistically significant ~50% reduction in the risk (not odds) of requiring postoperative rescue opioids in this exploratory secondary outcome; not adjusted for the multiple secondary/exploratory outcomes examined across this review, and not the protocol-defined primary inference.

Robustness Diagnostics

Leave-One-Out Influence Diagnostics: Primary 24-h Opioid

🔍 View Full Res

StataNow 19.5 Leave-One-Out - Primary 24-h Opioid

Robustness Diagnostics

Leave-One-Out Influence Diagnostics: Target A 48-h Opioid

🔍 View Full Res

StataNow 19.5 Leave-One-Out - Target A 48-h Opioid

Influence Analysis: Omission of An 2014 yields MD = −12.40 mg [−128.63 to +103.84], p = 0.4048. Omission of Zhang 2023 (median-converted) yields MD = −13.87 mg [−113.16 to +85.43], p = 0.3267. Each leave-one-out model reduces Target A to k = 2, so its confidence interval is very wide and its point estimate correspondingly unstable; these models test whether the direction of effect is driven by a single trial, and they are not independent confirmation of the pooled result.

Mathematical derivations and Stata weighting matrix

Methodological Transparency Mathematical Proofs & Stata Matrix

📐 Mathematical Derivations & Stata Calculation Breakdown

Cochrane Section 6.5 Compliant

Full algebraic formulas, parametric conversions, and exact numerical matrix weights used by StataNow 19.5 BE to pool the seven strict primary trials into the primary outcome synthesis.

🔬 Part 1: Mathematical Derivations for the 5 Conditional Studies

⚠️ Legacy reconstructions — none of these five feeds a pooled mean difference

These derivations are retained so the reasoning that once produced them stays auditable. They are not inputs to any current pooled estimate. Two of them (Sim 2002, Coura 2011) reconstruct an absolute dose from body weight — Coura from an assumed 70 kg the trial never reports — which the lock prohibits, so the pipeline leaves mean_i_mme empty for both and they enter only the scale-free SMD sensitivity, computed from their native mg/kg and µg/kg values, never from these reconstructions. Zhang 2025 was removed from every 0–24 h analysis in v33 because it reports postoperative day 1, not an explicit 0–24 h clock window. The two Chen 2015 reports are median/IQR and are handled as a parallel synthesis, not converted to mean/SD for pooling.

1. Sim et al., 2002 (Covidence #952)

Goodman Exact Product Variance (Body Weight Reversal)

Published: EA ($0.52 \pm 0.19\text{ mg/kg}$); Placebo ($0.68 \pm 0.38\text{ mg/kg}$). Weights: $60 \pm 10\text{ kg}$ vs $59 \pm 9\text{ kg}$.

E[XY] = μ_X · μ_Y
Var(XY) = σ_X²·σ_Y² + μ_X²·σ_Y² + μ_Y²·σ_X²

Arm 1 (EA): $0.52 \times 60 = 31.20\text{ mg}$; $\sigma = \sqrt{160.61} = 12.67\text{ mg}$

Arm 2 (Ctrl): $0.68 \times 59 = 40.12\text{ mg}$; $\sigma = \sqrt{551.80} = 23.49\text{ mg}$

LEGACY RECONSTRUCTION — excluded from the current locked analysis

MD = −8.92 mg IV MME (SE = 4.87 mg) is not used. Reconstructing absolute dose from arm-mean body weight is prohibited, and the two arms carry different mean weights, so the reconstruction is not even a common rescaling. The pipeline leaves mean_i_mme empty. Sim 2002 enters only the broader SMD sensitivity, via Hedges' g = −0.526 computed from the native mg/kg values.

2. Coura et al., 2011 (Covidence #838)

Reference Weight Normalization & Equianalgesic Ratio

Published: EA ($13.1 \pm 2.2\ \mu\text{g/kg}$); Sham ($16.3 \pm 1.6\ \mu\text{g/kg}$ fentanyl).

Dose_MME = Dose_µg/kg × 70 kg × 0.1 mg/µg

Arm 1 (EA): $13.1 \times 70 \times 0.1 = 91.70 \pm 15.40\text{ mg IV MME}$

Arm 2 (Sham): $16.3 \times 70 \times 0.1 = 114.10 \pm 11.20\text{ mg IV MME}$

LEGACY RECONSTRUCTION — excluded from the current locked analysis

MD = −22.40 mg IV MME (SE = 5.67 mg) is not used. The 70 kg is an assumed reference weight that Coura 2011 does not report; multiplying an arm mean by an invented constant produces a number with no source. The pipeline leaves mean_i_mme empty. Coura 2011 enters only the broader SMD sensitivity, via Hedges' g = −1.553 computed from the native µg/kg values, which needs no weight at all.

3. Zhang et al., 2025 (Covidence #71)

POD1 Cumulative Window Verification & MME Conversion

Published: Table 4 reports POD1 PCIA sufentanil: TEAS $50.53 \pm 4.46\ \mu\text{g}$ vs Ctrl $53.79 \pm 5.14\ \mu\text{g}$.

Basal = 2 µg/h × 24h = 48 µg → Confirms 24h Clock Window
MME = Dose_sufentanil × 1.0 mg/µg (1000:1, corrected 2026-09-07)

Arm 1 (TEAS): $50.53 \times 1.0 = 50.530 \pm 4.460\text{ mg IV MME}$

Arm 2 (Ctrl): $53.79 \times 1.0 = 53.790 \pm 5.140\text{ mg IV MME}$

WITHDRAWN IN v33 — excluded from every 0–24 h analysis

MD = −3.260 mg IV MME (SE = 0.996 mg) is no longer used. The MME conversion itself is sound, but the reported window is postoperative day 1, not an explicit 0–24 h clock window measured from the end of surgery. The basal-infusion arithmetic above is consistent with 24 h but does not establish when the clock started, so treating POD1 as 0–24 h remains an assumption the protocol prohibits. Zhang 2025 was therefore removed from the broader SMD sensitivity as well (k = 10 → 9), even though the SMD is scale-free: a scale-free metric changes the unit, not the estimand.

4. Chen et al., 2015 [Thyroidectomy] (#657)

Wan et al. (2014) Median/IQR to Mean/SD Transformation

Published: Rescue boluses: TEAS $1\ (1–3)$ vs Ctrl $3.5\ (2–7)$ boluses. Fixed $2\text{ mg IV morphine}$/bolus.

Mean ≈ (q₁ + m + q₃) / 3 | SD ≈ (q₃ − q₁) / 1.349

Arm 1 (TEAS): $(2 + 2 + 6)/3 = 3.333\text{ mg}$; $\text{SD} = (6 - 2)/1.349 = 2.965\text{ mg}$

Arm 2 (Ctrl): $(4 + 7 + 14)/3 = 8.333\text{ mg}$; $\text{SD} = (14 - 4)/1.349 = 7.413\text{ mg}$

TIER C — reported as a parallel synthesis, not converted for pooling

MD = −5.000 mg IV MME (SE = 1.234 mg) is not used. The bolus-to-milligram step is exact — multiplying by a fixed 2 mg is order-preserving, so the quantiles transform without assumption — but recovering a mean/SD from a median and IQR is not. v33 reports Chen 2015 in its own units instead: 2.0 mg (IQR 2.0–6.0) vs 7.0 mg (IQR 4.0–14.0), reported p = 0.004, alongside the primary model rather than inside it.

5. Chen et al., 2015 [Hyperalgesia] (#632)

Wan Transformation & Weight-Adjusted Sufentanil Boluses

Published: Rescue boluses: TEAS $3\ (2–4)$ vs Ctrl $7\ (6–8)$. Dose: $0.05\ \mu\text{g/kg sufentanil}$. Weights: $57.7\text{ vs } 56.8\text{ kg}$.

Boluses → Mean ± SD (Wan); NOT converted to mg IV MME

Arm 1 (TEAS): $3.00 \pm 1.48$ rescue boluses

Arm 2 (Ctrl): $7.00 \pm 1.48$ rescue boluses

Not converted to mg IV MME. Carried as a bolus count only.

Weight-adjusted per-bolus dosing cannot be reconstructed from the reported arm-mean weights without assuming every bolus was delivered at the nominal $0.05\ \mu\text{g/kg}$, which the report does not state. Recorded as NOT APPROPRIATE TO CONVERT in 06_AUDIT/opioid_conversion_audit.csv; the Stata pipeline leaves mean_i_mme missing and inc_primary = 0, so this trial contributes to no MME pool.

📊 Part 2: StataNow 19.5 BE Random-Effects Weighting Matrix (Strict Clean Trials: k = 7, N = 676)

Study & Year	Modality / Comparator	Data Status	Sample ($n_1 / n_2$)	Mean Diff $y_i$	Std Error $SE_i$	Variance $v_i$	Fixed Weight $w_i$	Stata REML Weight (%)	Stata DL Weight (%)	Hedges' $g$ (SE)
Chen 1998	TEAS vs Sham	Direct	25 / 25	−21.000 mg	6.103 mg	37.247	0.0268	11.69%	10.50%	−0.958 (0.299)
Chen 2020	TEAS vs Sham	Direct	40 / 40	−28.190 mg	1.781 mg	3.172	0.3152	15.09%	15.48%	−3.505 (0.356)
El-Rakshy 2009	EA vs Control	Direct	42 / 53	−1.600 mg	3.719 mg	13.827	0.0723	13.83%	13.48%	−0.088 (0.207)
He 2026 (hepatectomy/JIS)	TEAS vs Sham	Direct	80 / 79	−0.600 mg	0.578 mg	0.334	2.990	15.47%	16.12%	−0.164 (0.159)
Seevaunnamtum 2016	EA vs Usual Care	Direct	32 / 32	−12.560 mg	4.389 mg	19.263	0.0519	13.27%	12.65%	−0.707 (0.258)
Szmit 2021	TEAS vs Sham	Direct	24 / 24	−7.700 mg	1.491 mg	2.224	0.4496	15.21%	15.69%	−1.466 (0.325)
Yang 2024	EA vs Usual Care	Direct	90 / 90	−0.300 mg	0.716 mg	0.513	1.951	15.44%	16.08%	−0.062 (0.149)
StataNow 19.5 BE Primary Random-Effects Synthesis (Strict k = 7, N = 676)				REML: −9.907 mg	SE(KH): 4.157 mg	Cochran Q = 254.61 (p < 0.0001)		100.0% (τ² = 113.91, I² = 98.57%)	DL: −9.726 mg (τ² = 68.51)	g = −0.967 (p = 0.079)

1. DerSimonian–Laird Between-Study Variance

$$Q = \sum w_i (y_i - \hat{\theta}_{\mathrm{FE}})^2 = 254.61, \quad df = 6$$ $$\tau^2_{\mathrm{DL}} = \max\left(0, \frac{254.61 - 6}{3.629}\right) = 68.5143 \implies I^2 = 97.64\%$$ $$\theta_{\mathrm{DL}} = -9.726\text{ mg IV MME (unadj Wald: } [-16.26, -3.20], p = 0.0035)$$

2. Restricted Maximum Likelihood (REML)

Stata maximizes the restricted log-likelihood over $\tau^2$: $$\ln L_{\mathrm{REML}}(\tau^2) = -\frac{1}{2}\left[\sum \ln(v_i + \tau^2) + \ln \sum \frac{1}{v_i + \tau^2} + \sum \frac{(y_i - \hat{\theta})^2}{v_i + \tau^2}\right]$$ Converges to $\tau^2_{\mathrm{REML}} = 113.9105$, yielding $\theta = -9.907\text{ mg IV MME}$.

3. Hartung–Knapp (KH) Adjustment

$$q = \frac{1}{k-1} \sum w_i^* (y_i - \hat{\theta})^2 \implies \text{SE}_{\mathrm{KH}} = 4.157\text{ mg}$$ With $df = 6$, $t_{0.05, 6} = 2.447$ (vs unadjusted normal $z = 1.960$). $$95\%\text{ CI: } -9.907 \pm (2.447 \times 4.157) = [-20.079, +0.265\text{ mg}], \quad p = 0.0545$$

Methodological Governance Cochrane Handbook Section 10.11.4.1

Statistical Deferral of Multivariable Meta-Regression

The 10:1 Rule for Meta-Regression: Cochrane Handbook Section 10.11.4.1 advises that meta-regression generally should not be considered with fewer than approximately 10 studies per candidate predictor/covariate, and more may be required depending on the covariate's distribution, to avoid severe overfitting and spurious ecological associations.

Strict Clean Primary Trial Sample (k = 7 Trials, N = 676)

With k = 7 strict clean trials (TEAS k = 4, N = 337; EA k = 3, N = 339) reporting 24-h cumulative opioid data, the sample falls below the Cochrane 10:1 rule-of-thumb threshold (≥ 10 studies per candidate covariate) for inferential meta-regression. Univariable exploration of modality (TEAS vs EA: difference −9.32 mg, p = 0.308) is exploratory. Egger linear regression test for publication bias was not performed due to k < 10. No expanded moderator pool is reported: the pre-v26 11-trial sensitivity cohort was withdrawn because it counted the overlapping Yeh reports as independent trials and drew on records outside the primary set.

Exploratory Moderator Hypotheses & Expansion

Univariable meta-regressions for stimulation frequency (low vs mixed vs high), timing (preoperative vs intraoperative vs postoperative), and session count serve strictly hypothesis-generating purposes. Author inquiries seeking supplementary continuous parameters from reporting trials may further power secondary moderator analyses.

Stata execution log

Stata 19.5 BE Execution Log StataNow 19.5 BE (Macintosh 64-bit ARM)

StataNow 19.5 BE Consensus Execution Log

06_FINAL_ANALYSIS_V26/02_STATA/logs/01_opioid24_primary.log — StataNow 19.5 BE

● Completed (Exit Code 0)

Loading Stata 19.5 execution stream...

Multivariable meta-regression is not available

There are too few strict primary studies for a reliable multivariable model. This tab contains the saved modality comparison and exploratory single-moderator analysis. No interactive prediction tool is available.

Objective 3: Moderator Investigation StataNow 19.5 BE Verified Random-Effects REML + Knapp–Hartung SE

StataNow 19.5 BE Meta-Regression & Moderator Studio

Investigating sources of between-study clinical and statistical heterogeneity (τ²) across the strict primary trials (k = 7, N = 676). No expanded exploratory pool is reported. Under the Cochrane 10:1 rule of thumb (§10.11.4), k = 7 trials is underpowered for inferential multivariable meta-regression; univariable modality comparison (TEAS vs EA: difference −9.32 mg, p = 0.308) is exploratory. Egger's test for publication bias was not performed due to k < 10.

📥 Subgroup / Meta-Regression .do (v26) 📋 Stata Execution Log (v26) 📊 Subgroup / Meta-Regression Results (CSV) 🗂️ Locked Primary Dataset (CSV)

⚠️

Cochrane Handbook Methodological Framework: Sample Size & Degrees of Freedom Rules (v6.4, §10.11.4)

The Rule of 10: Cochrane guidelines advise that meta-regression should generally not be considered with fewer than 10 trials, recommending a minimum ratio of 10 studies per investigated covariate.

Strict Primary Clean Sample (k = 7, N = 676): With k = 7 trials, meta-regression is formally underpowered per the Cochrane 10:1 ratio recommendation. Univariable exploration of modality (TEAS vs EA: β = −9.32 mg [−30.44, +11.79], p = 0.308) reveals no statistically significant moderator effect. Publication bias testing via Egger linear regression was not performed because k < 10.
Withdrawn: expanded k = 11 moderator cohort. Earlier releases of this dashboard reported univariable baseline-opioid-demand (β = −0.170, p = 0.0186) and publication-year (β = +0.471, p = 0.0287) meta-regressions, plus a multivariable model, fitted to an 11-trial pool. That pool is not the v26 primary analysis set: it mixed conditional, proxy and excluded records, and counted the overlapping Yeh 2010 / Yeh 2011 lumbar-spine reports as two independent trials. Those models are not reproduced by the locked v26 pipeline and have been withdrawn rather than restated.
Multivariable & higher-order models are not reported. With k = 7 no multivariable meta-regression is estimable at acceptable risk of overfitting, collinearity (ecological fallacy) and inflated Type I error. None is presented, in any form, including as “sensitivity context”.

TEAS Stratum (Subgroup)

−14.00 mg

95% KH CI: [−34.18, +6.19] • k = 4 • I² = 98.59%

p = 0.1145 • not significant

EA Stratum (Subgroup)

−3.94 mg

95% KH CI: [−19.77, +11.90] • k = 3 • I² = 77.15%

p = 0.397 • not significant

Modality Meta-Regression (only model fitted)

β = −9.32 mg

TEAS vs EA (reference EA) • 95% CI: [−37.54, +14.40]

p = 0.308 • underpowered (k = 7 < 10)

Covariates Not Modelled

Not estimated

Baseline opioid demand, publication year, sex composition, dose–response

Below Cochrane 10:1 threshold

v26 Authoritative Subgroup Figure

Primary 24-h Opioid Consumption by Modality (k = 7)

Random-effects REML with Knapp–Hartung adjustment, stratified by modality. Generated by 06_FINAL_ANALYSIS_V26/02_STATA/09_subgroups_metareg.do.

Primary 24-h opioid consumption forest plot stratified by modality (TEAS k=4, EA k=3)

StataNow 19.5 BE • meta summarize, random(reml) se(kh) subgroup(modality)

⚠️ Bubble-plot studio withdrawn

This panel previously offered five interactive meta-regression bubble plots (baseline opioid demand, publication year, sex composition, and TEAS/EA strata) fitted to an 11-trial pool.

That pool is not the v26 primary analysis set. It included conditional, proxy-endpoint and excluded records, and counted the overlapping Yeh 2010 / Yeh 2011 lumbar-spine reports as two independent trials, which the v26 lock explicitly forbids. Its stated EA stratum (Sim 2002, Coura 2011, El-Rakshy 2009) is also not the v26 EA stratum (El-Rakshy 2009, Seevaunnamtum 2016, Yang 2024).

None of those models is reproduced by the locked v26 pipeline, so they have been removed rather than restated. With k = 7 the only defensible presentation is the stratified subgroup figure shown here, plus the single underpowered modality meta-regression reported above.

Statistical & Clinical Audit k = 7 Strict Primary Trials

Comprehensive Moderator & Clinical Parameter Matrix (k = 7 Strict Primary RCTs)

Descriptive audit of candidate moderators across the seven strict primary trials. No coefficient in this matrix is an inferential meta-regression result: with k = 7 the Cochrane 10:1 rule is not met, so this table documents what could be examined and why it was not modelled.

Moderator Domain	Variable Type	Empirical Distribution	Meta-Regression β [95% CI]	Model Test (Knapp–Hartung)	Variance Explained (R²)	Methodological Assessment	Clinical & Analytical Interpretation
Baseline Opioid Demand	Continuous Clinical	Control Mean: 5.3 to 114.1 mg	Not estimated	Not estimated	—	Not modelled (k = 7)	Baseline demand varies widely across the strict primary trials, so it is a plausible source of heterogeneity. It is not modelled: k = 7 is below the Cochrane 10:1 threshold. A previously displayed slope (β = −0.170, p = 0.0186) came from the withdrawn 11-trial pool and is not reproduced by v26.
Publication Year	Continuous Temporal	Year: 1998 to 2026	Not estimated	Not estimated	—	Not modelled (k = 7)	Not modelled in the locked primary analysis. A previously displayed secular trend (β = +0.471 per year, p = 0.0287) came from the withdrawn 11-trial pool. Possible explanations for a genuine trend would include changes over time in multimodal analgesia, perioperative practice, surgical case mix, comparator treatment, study methodology, or other secular trends. This analysis cannot determine the cause.
Modality (EA vs TEAS)	Binary Intervention	TEAS (k = 4) vs EA (k = 3)	−9.32 mg [−30.44, +11.79]	p = 0.308 (REML + Knapp–Hartung)	—	Univariable only • underpowered	The only meta-regression fitted in the locked primary analysis (`09_subgroups_metareg.do`). No evidence of a modality difference; with k = 7 this is exploratory and cannot exclude a clinically relevant difference. No multivariable adjustment is reported.
Stimulation Timing	Categorical Protocol	Preoperative/intraoperative-phase (k = 6/7); Postoperative-only (k = 1/7, Szmit 2021)	Not estimated				Szmit 2021 (added 2026-09-07) applies TEAS entirely postoperatively (30 min every 2 h through 24 h), unlike the other six trials, so this covariate is no longer constant. It is not modelled: k = 7 with a 6-vs-1 split is far below the Cochrane 10:1 threshold and severely imbalanced. A previously displayed categorical model (Preoperative k=5 vs Multi-phase k=5 vs Intraop k=1, summing to 11) came from the withdrawn 11-trial pool and does not describe this set either.
Electrical Frequency	Categorical Protocol	100% 2/100 Hz Dense-Disperse (k = 7/7)	Dropped by Stata due to zero variance (constant across trials)				All seven strict primary trials used the 2/100 Hz dense-disperse waveform (Szmit 2021, added 2026-09-07, alternates the same 2/100 Hz pattern); the covariate has no variance, so frequency cannot be tested. A previously displayed categorical model (2/100 Hz DD k=7 vs Fixed 100/2 Hz k=4, summing to 11) came from the withdrawn 11-trial pool and does not describe this set either.
Number of Sessions	Discrete Protocol	Single-session (k = 6/7); Repeated sessions (k = 1/7, Szmit 2021)	Not estimated				Szmit 2021 (added 2026-09-07) delivers repeated 30-min sessions every 2 h through 24 h, unlike the other six single-session trials, so this covariate is no longer constant. It is not modelled: k = 7 with a 6-vs-1 split is far below the Cochrane 10:1 threshold and severely imbalanced.
Patient Sex (% Female)	Demographic (Aggregate)	Range: 22.1% to 100.0% female	Not estimated	Not estimated	—	Ecological Fallacy	Not modelled; k = 7 is insufficient for reliable study-level meta-regression on a continuous demographic moderator. No association can be affirmed or ruled out from this evidence, and study-level aggregates could not establish individual-level effects in either direction (ecological fallacy).
Patient Age & BMI	Demographic (Aggregate)	Study mean age: 45 to 58 yrs	Clustered study means • Individual Participant Data (IPD) required				Study-level averages cannot evaluate patient-level age/weight effects due to Simpson's paradox.
Acupoint Selection	Anatomical Prescription	Bundled somatic points (LI4, PC6, ST36, SP6)	High collinearity • Multi-point co-administration				Acupoints are always prescribed as bundles; isolating single acupoint efficacy in aggregate data is impossible.

🔬 Demographic Variables & The Ecological Fallacy (Cochrane §10.11.2)

In meta-regression, demographic variables like patient age, BMI, and sex are reported as study-level aggregates (e.g. trial mean age = 52.4). Meta-regression cannot assess whether an individual 75-year-old achieves greater sparing than a 25-year-old. Testing aggregate means risks the Ecological Fallacy (Simpson's Paradox), where cross-study associations contradict true patient-level effects. True evaluation of age, BMI, and comorbidities strictly requires Individual Participant Data (IPD) meta-analysis.

⚡ STRICTA Interventions & Acupoint Prescription Bundling

In accordance with STRICTA guidelines, acupoints in perioperative anesthesia trials are virtually never tested in isolation. They are delivered as standard multi-channel somatic prescriptions combining upper-limb segmental points (LI4 Hegu, PC6 Neiguan) and lower-limb visceral points (ST36 Zusanli, SP6 Sanyinjiao). Because trials share the same point combinations, statistical collinearity prevents isolating the independent contribution of any single acupoint in study-level meta-regression.

Withdrawn in v26

Interactive Effect-Modifier Simulator — Removed

This panel previously let readers slide a baseline-opioid-demand value and a modality toggle through the multivariable model MD = 0.0204 − 0.1876 × (Baseline Demand) + 1.4895 × (EA), returning a point estimate and a confidence interval.

That model was fitted to the withdrawn 11-trial pool, is not reproduced by the locked primary pipeline, and could not be defended at k = 7: two covariates on seven studies is far below the Cochrane 10:1 threshold, and the interval it displayed was produced by an approximation rather than by the model's covariance matrix. Presenting it as an interactive predictor invited exactly the individual-patient reading that the accompanying ecological-fallacy warning disclaimed.

The authoritative moderator evidence is the modality subgroup pair and the single underpowered modality meta-regression reported at the top of this tab. No predictive instrument is offered in its place.

StataNow 19.5 BE Live Terminal Output

Verbatim Execution Log: v26 Subgroup & Meta-Regression Audit

📥 Download Full Log (.log)

06_FINAL_ANALYSIS_V26/02_STATA/logs/09_subgroups_metareg.log • StataNow 19.5 BE

Loading Stata meta-regression execution stream...

Limitations and author clarification records

Review the catalogued requests for missing or unclear study data and their recorded dispositions. The hypothetical sensitivity tool explores changes to existing primary results; it does not represent author responses.

Manuscript material Derived, not asserted

⚠️ Limitations of this review

Collaborative Action Center Cochrane Handbook Ch. 5.4 Prospective Outreach

Author Outreach & Data Clarification Center

No author reply is required to finalize this review. All 19 P1 issues carry a recorded disposition in the v26 lock (AF_P1_Disposition) and none is a global final-lock blocker. Eighteen were resolved analytically — by the source hierarchy, a raw/native-data rule, exclusion, partial exclusion, timepoint exclusion, field exclusion, stratification, mandatory sensitivity, or narrative-only reporting. One remains a hard hold: the Yeh 2010 / Yeh 2011 lumbar-spine cohort-overlap question (P1-01), which is handled by excluding both reports from pooling rather than by waiting on correspondence. The roster below is retained as a provenance record of which clarifications were sought and how each was dispositioned — not as a queue of outstanding replies.

🔴 CRITICAL PRIORITY 8 Trials (13.3%)

Primary 24-h Opioid Consumption Missing

Current Status: These major trials measured postoperative PCA opioids (e.g. fentanyl, sufentanil, morphine), but omitted cumulative 24-h numerical means/SDs from published tables.

Inquiry Target: Requests 24-h cumulative opioid consumption (mean ± SD in mg or µg) from electronic PCA pump delivery logs.

Key Studies: An 2014 (#698), Ao 2021 (#438), Gao 2021 (#400, N=610), Gu 2019 (#848), He 2026 (#41), Hou 2023 (#295), Huang 2024 (#245), Huang 2025 (#131).

🟡 IMPORTANT PRIORITY 51 Trials (85.0%)

Secondary Pain SDs, Vomiting Counts & GI Metrics

Current Status: Primary or secondary outcomes are published as medians, box plots without SDs, combined nausea/vomiting percentages, or require non-inferiority confirmation.

Inquiry Target: Requests exact numerical mean ± SD for 24-h resting/movement pain scores, discrete vomiting event counts, and intention-to-treat dropout reasons.

Key Studies: Chen 1998 (#969), Chen 2015 (#657), Praveena 2016 (#596), Sun 2017 (#632), Tu 2024 (#244), Wang 2024 (#212), Zhu 2022 (#381).

🔵 METHODOLOGICAL QC & P1 AUDIT 0 Global Blockers

Cohort Overlap & Metric Verification (19 P1s Dispositioned)

Current Status: All 19 Priority 1 (P1) methodological issues have been audited and fully dispositioned with 0 global blockers. Overlapping cohorts (Yeh 2010/2011, Lee 2011) and multi-arm splits (Chen 1998, Xie 2014) are strictly isolated. Primary synthesis is locked and complete.

Audit Disposition: Verified in window.P1_DISPOSITIONS. Primary 24-h opioid locked at k=7, N=676; Target A locked at k=3, N=1,999; Target E flatus locked at k=7, N=638 (k=6, N=596 before Zhang 2018 was admitted 2026-09-12).

Audited Dispositions: Yeh 2010/2011 (#828/#823), Luo 2026 (#35), Jin 2023 (#326), Zhou 2025 (#1879896105).

🔬 Hypothetical Sensitivity on Existing Primary Trials

−10.0 mg MME

−50 mg (Large sparing) 0 mg (No difference) +5 mg (Adverse)

Hypothetical Trial Modification

Browser baseline for the selected studies

−9.91 mg

95% CI [−20.08, +0.27]

Exploratory browser DL calculation

➔

Hypothetical browser synthesis

−12.12 mg

95% CI [−27.71, +3.47]

Interactive Test

Author Clarification Roster & Disposition Record (60 catalogued inquiries, 0 outstanding blockers)

Click "📋 Copy Email" to copy the ready-to-send draft letter directly to your clipboard, or click the email address to open your mail client.

Loading inquiry records

Study Identifier	Priority	Target Data Requested	Corresponding Author & Institution	Contact Email	Action

Standardization Protocol Cochrane Handbook Ch. 6.5.2

📐 Opioid Conversion Methodology & Literature Reference Hub

Prespecified equianalgesic conversion framework used for harmonization to a common IV morphine-milligram-equivalent (IV MME) metric, with uncertainty assessed in sensitivity analyses where a factor is not independently verified. See 06_FINAL_ANALYSIS_V26/06_AUDIT/opioid_conversion_audit.csv for the full per-drug audit.

Transparent mathematical formulations, clinical pharmacology equianalgesic conversion factors, and peer-reviewed literature citations underpinning all converted values in this systematic review.

Opioid Drug & Route	Conversion Factor to IV MME	Equianalgesic Dose (10 mg IV Morphine)	Clinical Pharmacology & Receptor Rationale	Foundational Guidelines & References
Morphine (IV)	1.0 mg / mg	10.0 mg	Reference standard opioid comparator (self-referential by definition). Pure $\mu$-opioid receptor agonist.	ANZCA APM:SE 2020
Sufentanil (IV) ✓ CORRECTED 2026-09-07	1.0 mg MME / µg (1000:1)	100.0 µg (0.10 mg)	Corrected 2026-09-07. Previously applied at 0.1 mg MME/µg (100:1) — identical to the fentanyl ratio below and uncited. Now 1.0 mg MME/µg (1000:1): BC Ministry of Health equianalgesic table gives sufentanil 0.01–0.04 mg (10–40 µg) = morphine 10 mg parenteral (250:1–1000:1), and the FDA sufentanil label states it is “as much as 10 times as potent as fentanyl” in balanced general anaesthesia (5–7× as a sole agent), which with fentanyl at 100:1 gives 500:1–1000:1. Residual uncertainty is carried as a published sensitivity range (0.1 / 0.25 / 0.5 / 1.0). See `06_AUDIT/opioid_conversion_audit.csv`.	BC MoH equianalgesic table / FDA sufentanil label
Fentanyl (IV)	0.10 mg MME / µg (100:1)	100.0 µg (0.10 mg)	Rapid-onset synthetic $\mu$-agonist; commonly cited 100:1 equianalgesic ratio with IV morphine in postoperative PCA regimens. Consistent between the Stata pipeline and this table (An 2014).	ANZCA 2020 / Nielsen 2016
Remifentanil (IV)	0.10 mg MME / µg (100:1)	100.0 µg (0.10 mg)	Ultra-short-acting esterase-metabolized $\mu$-agonist; evaluated in intraoperative consumption endpoints ($1\,\mu\text{g} = 0.1\text{ mg}$ IV MME). Not independently re-verified in this audit pass.	ASA Practice Guidelines 2012
Hydromorphone (IV)	5.0 mg MME / mg	2.0 mg	Semisynthetic hydrogenated ketone of morphine. BC Ministry of Health equianalgesic table: hydromorphone 2 mg parenteral = morphine 10 mg parenteral (5:1). Some tables (University of Toronto) use 1.5 mg, giving 6.67:1; the published parenteral range is 5:1–6.67:1 and this project uses 5:1. Affects Chen 1998 only.	BC MoH equianalgesic table
Oxycodone (IV)	1.0 mg MME / mg	10.0 mg	Parenteral formulation administered in European/Asian surgical trials; equipotent to intravenous morphine on a milligram-for-milligram basis.	Treillet 2018 / ANZCA 2020
Dezocine (IV)	1.0 mg MME / mg	10.0 mg	Synthetic opioid partial $\mu$-agonist / $\kappa$-antagonist widely used in China; clinical trials establish near 1:1 equianalgesic potency to IV morphine.	Chinese Pain Society Consensus / Liu 2020
Pethidine / Meperidine (IV) NOT APPLIED IN ANY ANALYSIS	0.10–0.13 mg MME / mg	75.0 – 100.0 mg	Synthetic opioid; 75–100 mg parenteral pethidine is commonly stated as equivalent to 10 mg parenteral morphine. Attribution corrected 2026-09-12. This row previously cited ANZCA 2020 / Knotkova 2012. Neither citation supports it: the ANZCA APM:SE 2020 entry in the reference pane below is applied in this review to intravenous sufentanil, fentanyl, oxycodone and hydromorphone, and pethidine is not among them; Knotkova 2012 appears nowhere in this repository except as this badge and the butorphanol badge — it has no reference-pane entry, and `06_AUDIT/opioid_conversion_audit.csv` had already recorded it as “also uncited in-repo” during the 2026-09-07 sufentanil audit. No factor for pethidine exists in `00_prep_data.do` and none is applied to any trial. Oztas 2019 reports rescue pethidine 20.66 ± 27.89 vs 33.12 ± 42.69 mg, and it is held out for a second and independent reason: it cannot be added to that trial's PCA tramadol, because the SD of a within-person sum needs a covariance the trial does not report.	No attribution verified in this repository
Tramadol (IV) NOT APPLIED IN ANY ANALYSIS	0.10 mg MME / mg	100.0 mg	Attribution withdrawn 2026-09-12; no ratio is adopted. This row previously cited ANZCA 2020 / Treillet 2018 for 0.10. Neither citation supports it: the ANZCA APM:SE 2020 entry is applied here to sufentanil, fentanyl, oxycodone and hydromorphone, and the Treillet 2018 entry to the 1000:1 sufentanil ratio — tramadol appears in neither. A dedicated search (2026-09-07 against this project's reference family, repeated 2026-09-12 against the published literature) located no citable parenteral tramadol:morphine ratio, and the reason is pharmacological rather than bibliographic: MME is defined on $\mu$-receptor agonist activity, whereas tramadol is a weak $\mu$-agonist and a serotonin–noradrenaline reuptake inhibitor whose analgesia depends on CYP2D6-mediated activation to O-desmethyltramadol, so a single equivalence number does not carry the same meaning for it as for a pure $\mu$-agonist. The figure is retained here only as the clinically quoted approximation it is, marked as applied to nothing. It gates no evidence: the one trial reporting tramadol, Oztas 2019 (0–24 h PCA tramadol 228.40 ± 87.89 vs 357.81 ± 123.70 mg), enters the review on the scale-free route instead, where the unit cancels — Hedges' g = −1.168 [−1.930, −0.406], `V33_TIERE_TEAS_USUAL_SMD`, native mg with no conversion applied.	No sourced ratio located; none adopted
Butorphanol (IV)	5.0 mg MME / mg	2.0 mg	Mixed agonist-antagonist; 2 mg IV butorphanol is clinically equianalgesic to 10 mg parenteral morphine.	ANZCA 2020 / Knotkova 2012

Scope of this table, recorded 2026-09-12. Four of these factors are actually applied to trial data: morphine 1.0 (self-referential), hydromorphone 5.0, sufentanil 1.0 mg MME/µg, and fentanyl 0.10 mg MME/µg. They are the only mme_factor assignments in 06_FINAL_ANALYSIS_V26/02_STATA/00_prep_data.do, alongside a factor of 1.0 for two trials whose authors supply their own MME figure. Every other row is reference material that no estimate in this review depends on. Three attributions on those reference-only rows remain unverified against a source held in this repository and are flagged rather than silently kept: Nielsen 2016 (fentanyl row, as a second citation beside ANZCA 2020), and Chinese Pain Society Consensus / Liu 2020 (dezocine) have no reference-pane entry; Knotkova 2012 (butorphanol) has none either, and the two rows that leant on it hardest have been corrected above. Where a badge names no reference-pane entry but the cell text itself gives a specific, checkable locator — the BC Ministry of Health palliative equianalgesic table and the FDA sufentanil label, both recorded in 06_AUDIT/opioid_conversion_audit.csv — the attribution stands.

Preserve the Reported Statistics

This review does not convert median/IQR to mean/SD for pooling. Record the median, dispersion, native units and source location, and keep this evidence separate from mean/SD analyses.

The Wan, Luo and Shi publications listed in the literature pane are historical methodological references. They do not authorize conversions in this review. Use the reporting check in the calculator pane to validate the order of the supplied quartiles.

Patient-Controlled Analgesia (PCA) Liquid Volumetric Protocol

Several trials report PCA pump consumption as delivered solution volume (in mL) or number of bolus compressions rather than mass of drug. In accordance with ASA Acute Pain Guidelines (2012), delivered doses are derived using the verified reservoir concentration:

Cumulative Delivered Dose ($\mu$g or mg):
$\text{Total Opioid Mass} = V_{\text{total}} \times C_{\text{solution}}$

Example: $45.7\text{ mL}$ PCIA solution with $1\,\mu\text{g/mL}$ sufentanil
$= 45.7\,\mu\text{g sufentanil} \times 1.0\text{ (IV MME factor)} = \mathbf{45.7\text{ mg IV MME}}$.

Demand Bolus & Lockout Verification

When compressions are reported without cumulative volume, total delivered dose accounts for successful boluses plus basal background rate:

Demand Dose:
$\text{Demand Mass} = N_{\text{delivered}} \times V_{\text{bolus}} \times C_{\text{solution}}$

Total 24-h Consumption:
$\text{Total} = \text{Demand Mass} + (24\text{ hours} \times \text{Basal Rate} \times C_{\text{solution}})$.

⚡ Equianalgesic Opioid to IV MME Calculator

Select drug and enter 24h consumption to inspect the review conversion convention. Only four of these factors are applied to trial data; the rest are reference-only. See the result note.

Select Opioid Formulation:

Reported Quantity (in units specified above):

Research Conversion Result

20.0 mg IV MME

20 µg sufentanil × 1.0 = 20.0 mg IV morphine equivalents

Median & IQR Reporting Check

Check the reported median and quartile order without estimating mean or SD.

Sample Size ($n$):

Q1 (25th percentile):

Median ($m$):

Q3 (75th percentile):

Original Reported Distribution

Median 15 (IQR 10–22), N=50

Preserved as reported; no mean/SD conversion.

Statistical Transformation BMC Med Res Methodol

Wan X, Wang W, Liu J, Tong T. (2014)

Estimating the sample mean and standard deviation from the sample size, median, range and/or interquartile range.

Application: Foundation for computing estimated sample mean and standard deviation for trials presenting median and IQR.
DOI: 10.1186/1471-2288-14-135 ↗

Statistical Transformation Stat Methods Med Res

Luo D, Wan X, Liu J, Tong T. (2018)

Optimally estimating the sample mean from the sample size, median, mid-range, and/or mid-quartile range.

Application: Minimizes mean squared error (MSE) when reconstructing means from skewed non-parametric medical data.
DOI: 10.1177/0962280216669183 ↗

Statistical Transformation Res Synth Methods

Shi J, Luo D, Weng H, et al. (2020)

Optimally estimating the sample standard deviation from the five-number summary.

Application: Advanced standard deviation reconstruction accounting for sample size and boundary quantile behavior.
DOI: 10.1002/jrsm.1429 ↗

Cochrane Standard Cochrane Library

Higgins JPT, Thomas J, et al. (eds) (2023)

Cochrane Handbook for Systematic Reviews of Interventions, Version 6.4. Chapter 6: Choosing effect measures.

Application: Section 6.5.2 provides methods for estimating means and standard deviations from median-based summaries where justified, and informs the review's author-contact protocol for otherwise unusable data.
training.cochrane.org/handbook ↗

Clinical Pharmacology CDC MMWR Guidelines

Dowell D, Ragan KR, Jones CM, et al. (2022)

CDC Clinical Practice Guideline for Prescribing Opioids for Pain — United States, 2022.

Application: Provides the general morphine-milligram-equivalent (MME) framing used for outpatient opioid prescribing. This guideline is scoped to outpatient acute (<1 month), subacute, and chronic pain management and does not publish route-specific conversion factors for intraoperative/inpatient IV fentanyl, sufentanil, or remifentanil; it is not the source for the perioperative IV conversion factors in the table above (see per-drug references and opioid_conversion_audit.csv).
DOI: 10.15585/mmwr.rr7103a1 ↗

Clinical Pharmacology ANZCA APM:SE

Macintyre PE, Schug SA, Scott DA, et al. (2020)

Acute Pain Management: Scientific Evidence (5th edition). Australian and New Zealand College of Anaesthetists.

Application: Clinical consensus on equianalgesic ratios for intravenous sufentanil, fentanyl, oxycodone, and hydromorphone in surgical recovery.
anzca.edu.au ↗

Clinical Pharmacology Support Care Cancer

Treillet E, Liborio-Amancio L, et al. (2018)

Practical issues in calculating equianalgesic doses for opioids: a systematic review and clinical consensus.

Application: Validates 1000:1 potency ratio for sufentanil and cross-tolerance handling in acute postoperative settings.
DOI: 10.1007/s00520-017-3850-9 ↗

Clinical Pharmacology Anesthesiology

ASA Task Force on Acute Pain (2012)

Practice guidelines for acute pain management in the perioperative setting: an updated report.

Application: Standardization of patient-controlled analgesia (PCA) reservoir metrics, lockout delivery, and rescue administration.
DOI: 10.1097/ALN.0b013e31823c1030 ↗

GRADE Summary of Findings (SoF) Table (Objective 7)

Saved GRADE assessments for the established target analyses. The additional v33 secondary analysis sets shown on the Secondary tab await result-specific RoB 2 adjudication and have no GRADE rating; these earlier ratings do not transfer to them.

Outcome Endpoint	Anticipated Absolute Effect	Pooled Effect (95% CI)	Participants (Studies)	Certainty (GRADE)	Downgrade Assessment & Footnotes

GRADE Clinical Interpretation (Modality-Stratified): RESULT-SPECIFIC RoB 2 INTEGRATED (LOCKED v26)

TEAS vs Sham (0–24 h Opioid Sparing, k=4, N=337): Low certainty evidence (downgraded for inconsistency and imprecision). The pooled estimate was MD −14.00 mg IV MME [95% KH CI: −34.18, +6.19], p = 0.1145; 95% confidence interval crosses zero with substantial between-trial heterogeneity (I² = 98.6%).
EA vs Usual Care (0–24 h Opioid Sparing, k=3, N=339): Very low certainty evidence (downgraded for inconsistency and very serious imprecision). The pooled point estimate was MD −3.94 mg IV MME [95% KH CI: −19.77, +11.90], p = 0.3969; 95% confidence interval crosses zero and is imprecise.
Supporting Combined Strict-Evidence Synthesis (0–24 h, k=7, N=676): Low certainty evidence (MD −9.91 mg IV MME [95% KH CI: −20.08, +0.27], p = 0.0545, 95% PI [−39.35, +19.54]). Modality-specific estimates remain the primary inferential analyses.
Note on RoB 2 Domain Status: Outcome ratings incorporate locked result-specific RoB 2 assessments across primary and key secondary endpoints.

Analyses computed but not reported as findings — and why

Data Export & Statistical Pipeline Downloads

Export the active filtered dataset containing study-level effect sizes, sample sizes, and RoB 2 classifications for independent statistical auditing, or download the master Stata 19.5 BE replication suite.

📥 Download Machine-Readable Stata Master Results (master_reconciled_results_v26.csv) 📥 Download Primary 24-h Opioid Dataset — 7 Strict Direct Trials (opioid_24h_primary.csv, k = 7) 📥 Download Stata Leave-One-Out High-Res Forest Plot (loo_opioid24_primary.png, k = 7 Iterations) 📥 Download Master Stata Replication Script (00_master.do) 📥 Download Primary Stata Execution Verification Log (01_opioid24_primary.log)

Background & Rationale

PICOS Framework

PROSPERO Registration & Protocol History

Quantify 24-h (primary) & 48-h (key secondary) opioid-sparing effects

Evaluate postoperative pain intensity at rest and during movement at ~24 h

Identify protocol moderators via prespecified subgroup & Stata meta-regression

Determine clinical meaningfulness against prespecified thresholds

Supportive syntheses and additional clinical endpoints

Assess certainty of evidence using RoB 2 and GRADE

Verified Evidence Synthesis Summary (StataNow 19.5 BE Validated)

Surgical Specialties Distribution (69 trials / 70 reports)

Methodological Architecture & Design Separation

PRISMA 2020 Flow Diagram

✅ PRISMA 2020 checklist

Bibliographic Search Strategies & Concept Architecture

Search Concept Map & Intersecting Facets

Study Characteristics & STRICTA Parameters Explorer

Cochrane Risk of Bias 2 (RoB 2) Traffic Light Studio

Source-verified Stata secondary analyses

Exploratory Interactive Forest Plot

Exploratory Paired Opioid–Pain Analysis (Objective 4 Studio)

📊 Current Analyses

📝 Manuscript Evidence Map

📚 Comparison with prior evidence

🗺️ How the Included Trials Contribute

Retired v33 secondary analyses — superseded by the current v34 set

Primary Outcome Contribution Pathway

🧭 Which 0–24 Hour Evidence Can Actually Be Pooled

StataNow 19.5 BE Consensus Synthesis & Forest Plots Hub

Primary outcome: cumulative 0–24 h opioid sparing (Modality-Specific Primary: TEAS k=4, EA k=3)

TEAS vs Sham TEAS (0–24h) ⓘ

EA vs Usual Care (0–24h) ⓘ

Supporting Combined Strict-Evidence Synthesis: 0–24h Opioid Consumption ⓘ

Secondary Clinical Endpoints

Cumulative 0–48 h opioid sparing (Strict: k = 3 RCTs, N = 1,999)

Strict Target A Pool

Excluding An 2014

Excluding Converted

Broader Window Pool

48-h PCA Volume Trials

Cumulative 0–72 h opioid sparing (Strict: Single Trial Yang 2024, k = 1, N = 180 — NOT POOLED)

Yang 2024 Alone (0–72h)

Broader Sensitivity (Yang + Wong)

Leave-one-out influence and sensitivity analysis

StataNow 19.5 BE Forest Plot Gallery

Primary 24-h Opioid Sparing (IV MME mg): Strict Direct Trials (k = 7)

Standardized Mean Difference (Hedges' g): Primary 24-h Opioid (k = 7)

Broader 24-h Sensitivity (Hedges' g): strict + conditional trials (k = 9)

Primary 24-h Opioid Consumption Stratified by Modality (TEAS vs EA)

Target A: Cumulative 0–48h Opioid Consumption (IV MME mg)

Target B: Cumulative 0–72h Opioid Consumption

Target C: Pain Intensity at Rest Around 24h (VAS 0–10)

Target D: Postoperative Nausea & Vomiting (Stratified Risk Ratios)

Target E: Time to First Postoperative Flatus (Hours)

Target F: Intraoperative Titrated Remifentanil Requirement (µg)

Target F: Postoperative Rescue Opioid Requirement (Risk Ratio)

Leave-One-Out Influence Diagnostics: Primary 24-h Opioid

Leave-One-Out Influence Diagnostics: Target A 48-h Opioid

📐 Mathematical Derivations & Stata Calculation Breakdown

🔬 Part 1: Mathematical Derivations for the 5 Conditional Studies

📊 Part 2: StataNow 19.5 BE Random-Effects Weighting Matrix (Strict Clean Trials: k = 7, N = 676)

Statistical Deferral of Multivariable Meta-Regression

StataNow 19.5 BE Consensus Execution Log

Multivariable meta-regression is not available

StataNow 19.5 BE Meta-Regression & Moderator Studio

Primary 24-h Opioid Consumption by Modality (k = 7)

Comprehensive Moderator & Clinical Parameter Matrix (k = 7 Strict Primary RCTs)

Interactive Effect-Modifier Simulator — Removed

Verbatim Execution Log: v26 Subgroup & Meta-Regression Audit

Limitations and author clarification records

⚠️ Limitations of this review

Author Outreach & Data Clarification Center

Primary 24-h Opioid Consumption Missing

Secondary Pain SDs, Vomiting Counts & GI Metrics

Cohort Overlap & Metric Verification (19 P1s Dispositioned)

🔬 Hypothetical Sensitivity on Existing Primary Trials

Author Clarification Roster & Disposition Record (60 catalogued inquiries, 0 outstanding blockers)

📐 Opioid Conversion Methodology & Literature Reference Hub

Preserve the Reported Statistics

Patient-Controlled Analgesia (PCA) Liquid Volumetric Protocol

TEAS vs Sham TEAS (0–24h)

EA vs Usual Care (0–24h)

Supporting Combined Strict-Evidence Synthesis: 0–24h Opioid Consumption