Gastroesophageal reflux disease — the clinical name for chronic acid reflux — affects roughly one in five adults in Western countries. For most patients, treatment means proton pump inhibitors (PPIs) like omeprazole: effective at suppressing acid, but increasingly questioned for long-term use. Studies have linked prolonged PPI exposure to kidney disease, bone fractures, and vitamin B12 and iron deficiency, and the drugs treat the symptom — acid — without clearly reducing the likelihood of relapse.

A new study, published this month in Frontiers in Pharmacology, asks whether Chinese herbal medicine can do better. It pools 20 randomized controlled trials involving 2,246 patients and compares herbal formulas directly against PPIs. But what makes this paper genuinely novel is not the comparison itself — it is the method. For the first time, the authors apply a "component-level" analysis to traditional formulas, trying to disentangle which individual herbs are doing the therapeutic work inside multi-herb prescriptions.

That is an unusual and ambitious thing to attempt. Whether it succeeds is worth examining carefully.

What the Researchers Did

The team screened nine literature databases — six English-language and three Chinese-language — for randomized controlled trials through April 2026, and pre-registered the protocol with PROSPERO (CRD42024612778). All 20 included trials enrolled adults with GERD, used PPIs as the control (not placebo), and each trial arm had at least 15 patients. Sixteen of the 20 studies were conducted in China, three in Japan, and one in Taiwan.

The standard machinery was all present: dual independent reviewers, Cochrane risk-of-bias assessment, random-effects meta-analysis, Egger's test for publication bias. On top of that, the authors added two advanced layers:

  • A network meta-analysis ranking 17 different treatments (16 herbal formulas plus PPI) against each other, even where formulas were never directly compared in a single trial.
  • A component network meta-analysis (CNMA) — the novel part — which models each formula's effect as the sum of its constituent herbs, and estimates a contribution coefficient (β) for every botanical drug appearing in more than three formulas.

For readers familiar with evidence reviews on this site, this echoes questions raised by our earlier look at a 33-trial review of herbs for dry eye: herbal research often evaluates formulas as indivisible packages. This study tries to crack the package open.

What the Numbers Show

The headline results are striking:

  • Clinical efficacy: herbal medicine outperformed PPIs with an odds ratio of 4.13 (95% CI 2.73–6.25).
  • Symptom relief: a standardized mean difference of 0.71 (95% CI 0.49–0.94) favoring herbs, measured across validated scales like the GERD-Q and the Frequency Scale for Symptoms of GERD.
  • Recurrence: six trials with follow-up found lower symptom recurrence after stopping herbal treatment than after stopping PPIs — 13.8% versus 28.6% in one trial, for example.
  • Safety: twelve trials reported adverse events. Minor complaints (nausea, dizziness, diarrhea) were few and similar to controls; laboratory measures — liver enzymes, kidney function, blood counts — showed no differences. No treatment-related toxicity or organ damage was reported.

The network ranking put Zhang's Banxia Houpu Decoction (ZBXHP) at the top, followed by Shugan Weian and Chaishao Tongjian. PPI ranked toward the bottom of the distribution — an eye-catching result, but as the authors themselves warn, one to handle with care (more on that below).

The component analysis identified 54 botanical drugs across the trials. The most frequently prescribed were Pinelliae rhizoma (bàn xià, pinellia tuber, in 62.5% of formulas), Coptidis rhizoma (huáng lián, coptis rhizome, 50%), Zingiberis rhizoma (jiāng, dried ginger, 50%), Bupleuri radix (chái hú, bupleurum root, 43.75%), Glycyrrhizae radix (gān cǎo, licorice root, 43.75%), and Evodiae fructus (wú zhū yú, evodia fruit, 43.75%). Pinellia and coptis formed the "central nucleus" of the co-prescription network — the hub around which the other herbs orbit.

Some component coefficients align neatly with known pharmacology: Citri reticulatae pericarpium (chén pí, dried tangerine peel) showed the largest positive contribution, consistent with its hesperidin-driven prokinetic effects on gastric emptying; berberine from coptis has been shown to relax the lower esophageal sphincter and protect esophageal mucosa. In the traditional logic, GERD often corresponds to "liver-stomach disharmony" — stagnant liver qi interfering with the stomach's downward function — and the top-ranked formulas are precisely those designed to disperse that stagnation and regulate the stomach.

What Weakens the Numbers

This is where the paper's most interesting lessons sit.

Most trials were open-label. Thirteen of 20 trials did not blind participants or clinicians. In GERD, where the primary outcomes are symptom scores reported by the patient, unblinded designs reliably inflate effect sizes. This is the single biggest methodological soft spot, and the CINeMA confidence assessment (the network-meta-analysis analogue of GRADE) penalized it accordingly: 38% of treatment comparisons were rated very-low confidence, only 33% moderate.

The network is sparse. There were no closed loops — no pair of treatments compared both directly and indirectly — which means statistical inconsistency couldn't even be formally tested. The transitivity assumption had to be assessed qualitatively instead. This doesn't invalidate the rankings, but it makes them more fragile than a well-connected network's would be.

The top ranking rests on one trial. ZBXHP's spectacular effect size (SMD 2.60) came largely from a single study with an unusually large estimate. Sensitivity analysis — removing that outlier — dropped overall heterogeneity from I² = 81% to 65%, and the authors explicitly caution against reading the ranking as a shopping list. Their clinical advice is blunt: patients should not select formulas based on the treatment rankings, but use them only under a qualified practitioner's prescription.

Publication bias is visible. Egger's test was significant (p = 0.0019). The authors handle this honestly, noting the asymmetry could reflect either true bias or heterogeneity — but either way, the missing negative trials question remains open.

The component analysis has a built-in paradox. Several herbs with well-established gastroprotective pharmacology — licorice root, ginger — received negative or near-zero contribution coefficients. The authors explain this clearly: the additive model can't disentangle herbs that are almost always prescribed together, and licorice is typically used as a "harmonizing" ingredient rather than the principal therapeutic agent. But this is precisely the limitation that matters for interpretation. The β coefficients describe associations within this particular prescription network, not the pharmacological worth of the individual herb. A reader who mistakes one for the other will draw exactly the wrong conclusion — and that is the paper's most valuable warning for anyone tempted to "optimize" formulas by cherry-picking top-scoring ingredients.

Availability is limited outside East Asia. The authors note that only a small number of CHM products hold formal registration in Europe, and comparative cost-effectiveness data were absent. The findings may not travel well across healthcare systems.

How to Read This Study — A Short Field Guide

  1. Reward the novelty, discount the headlines. The component-level analysis is a genuine methodological advance — a quantitative framework for studying formulas that are philosophically and pharmacologically designed to work as wholes. But novelty of method doesn't add certainty to evidence. Read the results as a proof of concept for the approach, not as a prescription guide.

  2. Watch for the paradox test. Whenever a statistical decomposition assigns a known-active ingredient a negative or null effect, something structural is happening — collinearity, confounding by co-prescription, or a role (like licorice's harmonizing function) that the model can't see. The authors pass this test by explaining it; the reader should notice it anyway.

  3. Separate "better than PPIs in trials" from "safer for you." The safety data are reassuring but thin: 12 of 20 trials reported adverse events, none with serious toxicity, but the sample is small and the follow-up short. The authors also flag real interaction risks — herbs can modulate CYP2C19 and CYP3A4, the same enzymes that metabolize PPIs — and advise liver monitoring for combined therapy.

  4. Ranking is not recommending. Treatment rankings from sparse networks are the most misread output in network meta-analysis. Here the authors say so themselves, in plain language: do not use the P-scores to pick a formula. If more papers did that, fewer misleading headlines would exist.

The Bottom Line

This is a well-conducted review with honest authors: 20 trials, 2,246 patients, a registered protocol, favorable symptom and recurrence results for herbal formulas versus PPIs — wrapped in open-label designs, visible publication bias, and a component analysis that is exciting precisely because it reveals how hard it is to attribute effects to individual herbs inside traditional prescriptions.

The lasting contribution here may be methodological rather than clinical: a first demonstration that you can model formulas as sums of herbs without losing the whole. The clinical verdict — whether these formulas reliably beat PPIs in well-blinded trials — remains a question for the next generation of studies: larger, double-blinded, standardized, and run across more than one healthcare system.

For those of us watching from the outside, the study is a reminder of something the Chinese medical tradition has always claimed: the prescription is not a list of ingredients. It is a composition — and compositions are harder to take apart than they look.

If you are curious how the classical Chinese medical tradition thinks about digestion and balance more broadly, our book Huangdi Neijing: A Modern Introduction offers a beginner-friendly guide to its foundational text.

This article is for educational purposes only and is not medical advice. Do not change your medication based on this article — if you have GERD or any medical condition, please consult a qualified healthcare professional.

References

  • Frontiers in Pharmacology, 2026. Treatment effects and safety of Chinese herbal medicine in gastroesophageal reflux disease: a systematic review of randomized controlled trials and a network meta-analysis with component analysis. DOI: 10.3389/fphar.2026.1920738. https://www.frontiersin.org/journals/pharmacology/articles/10.3389/fphar.2026.1920738/full
  • PROSPERO registration CRD42024612778. https://www.crd.york.ac.uk/PROSPERO/view/CRD42024612778