Vitamin A
| 證據等級: L5 | 預測適應症: 10 個 |
目錄
Vitamin A: From Nutritional Deficiency to Perinatal & Neonatal Disease Prevention
One-Sentence Summary
Vitamin A (DrugBank DB00162) is an essential fat-soluble vitamin whose established clinical role is treating and preventing vitamin A deficiency. TxGNN’s single highest-scoring prediction (“congenital prothrombin deficiency”) is almost certainly a knowledge-graph artifact confusing Vitamin A with Vitamin K and is not clinically credible. Among the ten model-ranked candidates, the strongest evidence-backed signal is Perinatal Disease — reflecting Vitamin A’s well-documented role in preventing bronchopulmonary dysplasia (BPD) and reducing neonatal/maternal mortality — supported by five Cochrane systematic reviews and a Phase 3 RCT enrolling 100,000 participants.
Quick Overview
| Item | Content |
|---|---|
| Original Indication | Vitamin A deficiency (no structured indication text on file; drug is a nutritional/essential vitamin, not a single-indication pharmaceutical) |
| Predicted New Indication (evidence-led) | Perinatal disease (Vitamin A supplementation in very low birth weight / preterm infants) |
| TxGNN Prediction Score (this candidate) | 99.63% (rank 9411/~17,000 in disease vocabulary) |
| Evidence Level | L1 (≥2 completed Phase 3 RCTs / Cochrane systematic reviews) |
| US Market Status | Not marketed |
| Number of NDAs | 0 |
| Recommended Decision | Proceed with Guardrails |
⚠️ Note: The disease with the highest raw TxGNN score in this evidence pack (congenital prothrombin deficiency, 99.97%) was not selected as the headline indication — see “Full Candidate Summary” and “Why is This Prediction Reasonable” below for the rationale.
Full Candidate Summary (All 10 Model-Ranked Indications)
| Rank | Disease | TxGNN Score | Evidence Level | Recommendation | Reviewer Note |
|---|---|---|---|---|---|
| 1 | Congenital prothrombin deficiency | 99.97% | L5 | Hold | Likely false positive — Vitamin A/K confusion; no mechanistic link (Factor II synthesis depends on Vitamin K, not A) |
| 2 | Biotin metabolic disease | 99.85% | L5 | Hold | Likely false positive — unrelated coenzyme pathway |
| 3 | Non-syndromic esophageal malformation | 99.72% | L5 | Hold | Plausible embryology (RA signaling in foregut development) but zero human evidence |
| 4 | Injury (wound healing) | 99.69% | L3 | Research Question | Real RAR-mediated mechanism; evidence is observational/mixed (fracture risk both ways) |
| 5 | Cell proliferation disorder | 99.64% | L4 | Research Question | Mechanism valid via RAR/RXR, but strongest clinical evidence belongs to ATRA (a distinct drug entity), not Vitamin A itself |
| 6 | Perinatal disease | 99.63% | L1 | Proceed with Guardrails | Strongest evidence tier — established use, Cochrane-level support |
| 6 | Florid cemento-osseous dysplasia | 99.63% | L5 | Hold | No trials, no literature — no clinical basis |
| 8 | Segmental odontomaxillary dysplasia | 99.63% | L5 | Hold | No trials, no literature — no clinical basis |
| 9 | Disease by subcellular system affected | 99.63% | L5 | Hold | Ontology node, not a real disease entity — should be excluded from candidate list |
| 10 | Radiation/chemically induced disorder | 99.63% | L3 | Research Question | Valid for the photoaging subset (topical retinoids) but the label is too broad; most trials unrelated to Vitamin A |
(Perinatal disease is technically rank 7 by raw score ordering but is elevated here because it is the only candidate reaching L1 evidence and an actionable S3 decision stage.)
Why is This Prediction Reasonable?
On model reliability: Six of the ten candidates (ranks 1, 2, 3, 6, 8, 9) have no supporting clinical trials or literature at all, or evidence pack notes explicitly flag them as probable knowledge-graph embedding artifacts (e.g., Vitamin A conflated with Vitamin K for the top-scoring candidate). One node (“disease by subcellular system affected”) is an ontology category, not a real disease. This report does not treat raw TxGNN score as a proxy for clinical credibility, and features the candidate with the strongest actual evidence instead.
Mechanism: Vitamin A (retinol) and its active metabolite retinoic acid signal through nuclear retinoic acid receptors (RAR/RXR), which are essential for normal lung growth, alveolar epithelial differentiation, and immune maturation. Preterm and very-low-birth-weight (VLBW) infants are born with low hepatic vitamin A stores and low serum retinol-binding protein, placing them at elevated risk of chronic lung disease (bronchopulmonary dysplasia) and infection-related mortality.
Relationship to original use: This is not a novel repurposing hypothesis so much as a well-established, evidence-dense clinical practice (vitamin A supplementation in VLBW infants) that the TxGNN model independently rediscovered from the knowledge graph — a positive validation signal for the model, even though the underlying use is not “new.” A genuinely novel finding within this evidence set (wound healing / cell proliferation, ranks 4–5) has real mechanistic plausibility but only observational or preclinical support today.
Clinical Trial Evidence
(Perinatal disease — Vitamin A–specific trials only; many other trials in the evidence pool concern unrelated vitamins (D, C, B) or micronutrient combinations and are excluded here as not directly relevant.)
| Trial Number | Phase | Status | Enrollment | Key Findings |
|---|---|---|---|---|
| NCT00211341 | Phase 3 | Completed | 100,000 | Randomised, double-blind, placebo-controlled trial of weekly vitamin A supplementation in reproductive-age women (Ghana); hypothesized 33% reduction in maternal mortality — largest RCT in this evidence pack directly testing Vitamin A |
| NCT03971604 | N/A | Unknown | 300 | Correlation between maternal Vitamin A/E levels and preeclampsia risk |
| NCT00860470 | Phase 3 | Completed | 44,567 | Antenatal multiple micronutrient supplementation (including vitamin A) to improve infant survival and health, Bangladesh |
| NCT02300155 | Phase 4 | Completed | 1,370 | Tolerability of a multivitamin (containing vitamin A) vs. a standard prenatal vitamin in pregnant women with morning sickness |
Literature Evidence
| PMID | Year | Type | Journal | Key Findings |
|---|---|---|---|---|
| 27552058 | 2016 | Cochrane Systematic Review | Cochrane Database Syst Rev | Vitamin A supplementation reduces mortality and short/long-term morbidity (including BPD) in very low birth weight infants — most recent update of a repeatedly-confirmed Cochrane series |
| 22742601 | 2012 | Systematic Review/Meta-analysis | Paediatr Perinat Epidemiol | Vitamin A/carotenoid supplementation during pregnancy improves maternal, fetal, neonatal and infant health outcomes |
| 23445845 | 2013 | Review | J Pediatr | Vitamin A and D deficiencies in preterm/LBW infants; supplementation needed to prevent detrimental consequences |
| 25222155 | 2014 | Review/Editorial | JAMA Pediatrics | Vitamin A shortage directly linked to risk of bronchopulmonary dysplasia in preterm infants |
| 21975731 | 2011 | Cochrane Systematic Review | Cochrane Database Syst Rev | Earlier update of the same Cochrane review confirming consistent effect across multiple update cycles (2000, 2002, 2007, 2011, 2016) |
| 2672772 | 1989 | Review | Am J Clin Nutr | Premature infants have lower retinol reserves than term infants; parenteral supplementation strategy discussed |
| 29895848 | 2018 | Case report | Eur J Clin Nutr | Vitamin A/micronutrient deficiency post-bariatric surgery causing maternal and fetal complications across multiple pregnancies |
| 17952233 | 2007 | Burden-of-disease study | S Afr Med J | Estimated disease burden attributable to vitamin A deficiency in children and pregnant women, South Africa |
US Market Information
Vitamin A is currently recorded as not marketed (0 licenses on file) in the source regulatory dataset for this evidence pack. No license records are available to summarize.
Safety Considerations
Structured safety fields (key warnings, contraindications, DDI) in this evidence pack are all marked as data gaps.
Relevant point from the literature evidence base: Vitamin A has a narrow therapeutic window in pregnancy and infancy — both deficiency and excess (hypervitaminosis A) are associated with adverse outcomes. Case-level literature in this pack (PMID 29895848) documents maternal/fetal complications from vitamin A excess following bariatric-surgery-related deficiency correction, and separate systematic reviews (not part of the perinatal evidence set) associate high-dose vitamin A intake with increased fracture risk. Dosing in neonatal/perinatal use should follow established NICU protocols rather than general supplement dosing.
Beyond this, please refer to the package insert for complete safety information.
Conclusion and Next Steps
Decision: Proceed with Guardrails (for Perinatal Disease / VLBW infant supplementation specifically)
Rationale: This is the only candidate in the full set reaching L1 evidence (multiple Cochrane systematic reviews plus a 100,000-participant Phase 3 RCT), and it reflects an already-established area of neonatal clinical practice rather than a speculative new mechanism. The remaining nine model-ranked candidates should be held: six have no clinical trial or literature support at all (several explicitly flagged as likely knowledge-graph artifacts), and the two with mechanistic plausibility (wound healing, cell proliferation) currently rest only on observational or preclinical data.
To proceed, the following is needed:
- Structured MOA and formal package-insert safety data (currently data gaps) to complete a S1 safety pre-screen
- Confirmation of dosing protocols and route of administration specific to preterm/VLBW neonatal populations
- Independent verification that “congenital prothrombin deficiency” and “biotin metabolic disease” predictions are graph-embedding artifacts (recommend flagging to the TxGNN pipeline team for model QA)
- Exclusion of ontology-category nodes (e.g., “disease by subcellular system affected”) from future candidate lists at the data-processing stage
Disclaimer
This content is for research purposes only and does not constitute medical advice. Clinical validation is required before any clinical application.