二甲双胍与长寿 & 信念效应的力量
摘要
Peter Attia博士与Andrew Huberman共同开展一期期刊俱乐部，分析两篇论文：一篇是2023年重新评估二甲双胍在2型糖尿病患者中生存获益的回顾性队列研究，另一篇是以尼古丁为模型药物探讨剂量依赖性安慰剂效应的研究。讨论提供了一个批判性解读流行病学研究与实验性研究的详细框架，最终对二甲双胍在非糖尿病人群中能否带来有意义的长寿获益这一长期观点提出了挑战。
核心要点
- 2014年具有里程碑意义的Banister研究显示，接受二甲双胍治疗的糖尿病患者比非糖尿病患者寿命更长，但该研究很可能存在缺陷，原因在于信息性截尾（informative censoring）——病情进展或死亡的患者被从二甲双胍组中剔除，人为夸大了其表观获益。
- 2023年Keys等人的研究纳入约50万名丹麦患者，发现服用二甲双胍的糖尿病患者与匹配的非糖尿病对照相比，全因死亡率高出32–48%，颠覆了Banister的研究结论。
- 二甲双胍通过弱抑制线粒体电子传递链复合体I发挥作用，降低肝脏葡萄糖输出——但其潜在长寿获益的确切机制仍存在争议。
- 二甲双胍未能在**干预测试项目（Interventions Testing Program，ITP）**中延长寿命，而该项目是临床前长寿研究的金标准平台；雷帕霉素则取得了成功。
- 静息血乳酸升高（超过约1.0 mmol/L）可能提示二甲双胍正在损害线粒体功能和2区有氧运动能力——这是Attia本人停用二甲双胍的个人原因。
- 胰岛素抵抗起始于肌肉，由细胞内脂肪堆积驱动，最早可检测到的迹象是空腹或餐后胰岛素升高，而非血糖升高。
- 运动可能是预防胰岛素抵抗最重要的单一因素；即便是体型偏瘦、年轻但缺乏运动的个体也会出现葡萄糖处置能力受损。
- 仅仅一周的睡眠剥夺（每晚4小时）即可使葡萄糖处置能力下降约50%，相当于诱发严重的胰岛素抵抗。
- 安慰剂效应遵循剂量-反应曲线——对剂量大小的信念会按比例影响生理反应，这一点已通过尼古丁实验得到证实。
- 小檗碱（Berberine）提取自树皮，作用机制与二甲双胍类似，被称为”穷人的二甲双胍”——能抑制mTOR并降低血糖——且无需处方即可获得。
详细笔记
什么是二甲双胍？
- 商品名：Glucophage；作为2型糖尿病一线用药已有40–50余年历史的通用药物
- 作用机制：弱抑制线粒体电子传递链复合体I
- 净效应：降低肝脏葡萄糖输出（即抑制糖尿病患者肝脏过度向循环系统释放葡萄糖的倾向）
- 也能弱抑制mTOR、减轻炎症，并可能抑制衰老细胞活性——这些”脱靶”效应引发了将其作为衰老保护剂（geroprotective agent）的研究兴趣
- 副作用：明显的恶心感，尤其在剂量未缓慢递增时；可能升高静息乳酸水平
背景：2014年Banister研究
- 纳入来自英国生物样本库的约95,000名受试者
- 将仅使用二甲双胍的2型糖尿病患者与匹配的非糖尿病对照进行比较
- 研究发现：在2.8年随访期内，糖尿病患者使用二甲双胍后全因死亡率低15%（风险比0.85）——即其寿命似乎比无糖尿病人群更长
- 这一反直觉的结果在学界引发广泛兴奋，推动了将二甲双胍作为衰老保护剂的研究
关键缺陷——信息性截尾：
- 二甲双胍组中停药、失访或进展至需要额外用药的患者被从分析中剔除
- 这一做法系统性地排除了病情较重的二甲双胍患者，仅保留了应答良好者
- 类比：研究吸烟与肺癌的关系，却将每个因此死亡的吸烟者从数据集中删除
2023年Keys等人的研究
引用： Reassessing the evidence of a survival advantage in type 2 diabetics treated with metformin compared with controls without diabetes: A retrospective cohort study
相较Banister研究的设计改进：
- 纳入丹麦健康登记系统约50万名受试者（样本量更大）
- 增加了不同性别同卵双胞胎差异分析——一名双胞胎患有糖尿病，另一名未患——提供了遗传和环境层面的匹配
- 对有无信息性截尾的情况分别进行了敏感性分析
主要研究结果（粗死亡率，每1,000人年）：
| 组别 | 死亡数/1,000人年 |
|---|---|
| 非糖尿病匹配对照 | 约16.9 |
| 服用二甲双胍的糖尿病患者（单例组） | 约24.9 |
| 非糖尿病同卵双胞胎 | 约12.9 |
| 服用二甲双胍的糖尿病患者（双胞胎组） | 约24.7 |
风险比（单例组，未调整）： 1.48 → 二甲双胍/糖尿病组死亡风险高48%
调整药物、心血管疾病、精神科用药、教育程度后：
- 单例组风险比：约1.32–1.33（死亡率仍升高32–33%）
- 双胞胎组未调整风险比：2.15；调整后：约1.70–1.80
截尾敏感性分析：
- 不截尾（所有患者均计入）：HR = 1.48（单例组）
- 截尾处理（复制Banister方法）：HR = 1.39
- 截尾处理仅小幅改善结果——未能重现Banister研究中的保护性结论
Keys研究结论： Banister研究中的表观生存优势在更严格的方法学下消失。服用二甲双胍的糖尿病患者死亡率仍显著高于匹配的非糖尿病患者。
重要注意事项： 这并不能证明二甲双胍有害或无效。由于缺乏未治疗糖尿病患者的对照组，我们无法判断二甲双胍是否正在延缓病情进展。该研究无法建立因果关系。
TAME试验
- 以二甲双胍靶向衰老（Targeting Aging With Metformin）——一项前瞻性随机临床试验
- 由Nir Barzilai主导
- 旨在前瞻性地确定二甲双胍能否延长非糖尿病成年人的健康寿命
- 目前已获资助并正在进行中；将提供最终的人类数据
二甲双胍在ITP中的表现（动物数据）
- 干预测试项目是美国国立卫生研究院资助的临床前长寿研究项目，同时在三个独立实验室中使用非近交系小鼠品系开展
- 雷帕霉素：多次取得成功，甚至在晚年开始给药（20月龄小鼠）也有效；寿命延长15%以上
- 二甲双胍：未能在ITP中延长寿命
- 17-Alpha雌二醇：在雄性小鼠中的寿命延长效果与雷帕霉素相当；在雌性小鼠中无效
- 坎格列净（Canagliflozin）（SGLT2抑制剂）：在ITP中取得成功
胰岛素抵抗：机制与原因
- 胰岛素与细胞受体结合→触发GLUT转运体插入肌肉细胞膜→葡萄糖被动进入细胞（无需泵，顺浓度梯度运动）
- 随着**肌细胞内脂肪（intramyocellular fat）**积累（耶鲁大学Gerald Shulman的研究），该信号级联被破坏→需要越来越多的胰岛素才能达到相同的葡萄糖摄取效果 = 胰岛素抵抗
- 最早期生物标志物：空腹或餐后胰岛素升高，而非血糖升高
- 正常血糖总循环量约为5克；2型糖尿病本质上就是多出一茶匙的葡萄糖
胰岛素抵抗的主要原因：
- 缺乏运动——最重要的单一因素；即便体型偏瘦的不活动个体也会发生胰岛素抵抗
- 睡眠剥夺——每晚4小时持续一周可使葡萄糖处置能力下降约50%
- 高皮质醇血症（慢性压力/皮质醇升高）
- 能量失衡——过量热量摄入导致脂肪从皮下储存溢出至肌肉、肝脏和胰腺
Attia
English Original 英文原文
Metformin for Longevity & The Power of Belief Effects
Summary
Dr. Peter Attia and Andrew Huberman conduct a joint journal club analyzing two papers: a 2023 retrospective cohort study reassessing metformin’s survival benefits in type 2 diabetics, and a study on dose-dependent placebo effects using nicotine as a model drug. The discussion provides a detailed framework for critically interpreting epidemiological versus experimental research, ultimately challenging the long-held belief that metformin offers meaningful longevity benefits in non-diabetic populations.
Key Takeaways
- The landmark 2014 Banister study showing metformin-treated diabetics outliving non-diabetics was likely flawed due to informative censoring — patients who progressed or died were removed from the metformin group, artificially inflating its apparent benefit.
- The 2023 Keys et al. study using ~500,000 Danish patients found diabetics on metformin had 32–48% higher all-cause mortality compared to matched non-diabetic controls, reversing the Banister finding.
- Metformin works by weakly inhibiting Complex I of the mitochondrial electron transport chain, reducing hepatic glucose output — but its exact mechanism of potential longevity benefit remains debated.
- Metformin failed to extend lifespan in the Interventions Testing Program (ITP), the gold standard preclinical longevity research platform, while rapamycin succeeded.
- Elevated resting blood lactate (above ~1.0 mmol/L) may indicate metformin is impairing mitochondrial function and Zone 2 aerobic capacity — this was Attia’s personal reason for stopping metformin.
- Insulin resistance begins in the muscles, driven by intracellular fat accumulation, and the earliest detectable sign is elevated fasting or postprandial insulin, not elevated glucose.
- Exercise is likely the single most important factor in preventing insulin resistance; even lean, young, but inactive individuals show impaired glucose disposal.
- Sleep deprivation of just one week (4 hours/night) can reduce glucose disposal by approximately 50%, equivalent to inducing profound insulin resistance.
- The placebo effect follows a dose-response curve — belief about dose magnitude scales the physiological response, demonstrated with nicotine.
- Berberine, derived from tree bark, acts similarly to metformin as a poor man’s version — inhibiting mTOR and reducing blood glucose — without requiring a prescription.
Detailed Notes
What Is Metformin?
- Brand name: Glucophage; generic drug used for 40–50+ years as a first-line agent for type 2 diabetes
- Mechanism: Weakly inhibits Complex I of the mitochondrial electron transport chain
- Net effect: Reduces hepatic glucose output (the liver’s tendency to over-release glucose into circulation in diabetics)
- Also weakly inhibits mTOR, reduces Inflammation 炎症, and may suppress senescent cell activity — these “off-target” effects prompted interest in it as a geroprotective agent
- Side effects: Significant nausea, especially when dose is not titrated up slowly; may elevate resting lactate levels
Background: The 2014 Banister Study
- Used ~95,000 subjects from a UK biobank
- Compared type 2 diabetics on metformin-only to matched non-diabetic controls
- Finding: Diabetics on metformin had 15% lower all-cause mortality over 2.8 years (hazard ratio 0.85) — meaning they appeared to outlive people without diabetes
- This counterintuitive result generated widespread excitement about metformin as a geroprotective agent
Critical flaw — Informative censoring:
- Patients in the metformin group who stopped the drug, were lost to follow-up, or progressed to needing additional medications were removed from the analysis
- This systematically excluded the sicker metformin patients, leaving only those who responded well
- Analogy: Studying smoking and lung cancer but removing every smoker who dies from the dataset
The 2023 Keys et al. Study
Citation: Reassessing the evidence of a survival advantage in type 2 diabetics treated with metformin compared with controls without diabetes: A retrospective cohort study
Design improvements over Banister:
- ~500,000 subjects from the Danish Health Registry (larger sample)
- Added a discordant same-sex twin analysis — one twin had diabetes, one did not — providing genetic and environmental matching
- Performed sensitivity analysis with and without informative censoring
Key findings (Crude mortality per 1,000 person-years):
| Group | Deaths/1,000 person-years |
|---|---|
| Non-diabetic matched controls | ~16.9 |
| Diabetics on metformin (Singletons) | ~24.9 |
| Non-diabetic co-twins | ~12.9 |
| Diabetics on metformin (Twins) | ~24.7 |
Hazard ratios (Singletons, unadjusted): 1.48 → 48% higher risk of death in metformin/diabetic group
After adjusting for medications, cardiovascular conditions, psychiatric meds, education:
- Singleton hazard ratio: ~1.32–1.33 (still 32–33% elevated mortality)
- Twin hazard ratio unadjusted: 2.15; adjusted: ~1.70–1.80
Censoring sensitivity analysis:
- Without censoring (all patients counted): HR = 1.48 (Singletons)
- With censoring (replicating Banister method): HR = 1.39
- Censoring improved the result only modestly — it did not reproduce Banister’s protective finding
Conclusion from Keys: The apparent survival advantage in the Banister study disappears under more rigorous methodology. Diabetics on metformin still die at significantly higher rates than matched non-diabetics.
Important caveat: This does not prove metformin is harmful or unhelpful. Without a group of untreated diabetics as comparison, we cannot know whether metformin is slowing the progression. The study cannot establish causation.
The TAME Trial
- Targeting Aging With Metformin — a prospective randomized clinical trial
- Led by Nir Barzilai
- Aims to determine prospectively whether metformin extends healthy lifespan in non-diabetic adults
- Currently funded and underway; will be the definitive human data
Metformin in the ITP (Animal Data)
- The Interventions Testing Program is an NIH-funded preclinical longevity program run concurrently across three independent labs using non-inbred mouse strains
- Rapamycin: Repeatedly successful, even when started late in life (20-month-old mice); 15%+ lifespan extension
- Metformin: Failed to extend lifespan in the ITP
- 17-Alpha estradiol: Extends lifespan in male mice comparably to rapamycin; does not work in female mice
- Canagliflozin (SGLT2 inhibitor): Successful in ITP
Insulin Resistance: Mechanism and Causes
- Insulin binds to cell receptors → triggers insertion of GLUT transporters into the muscle cell membrane → glucose flows passively into the cell (no pump needed, moves with concentration gradient)
- As intramyocellular fat accumulates (work by Gerald Shulman at Yale), this signaling cascade is disrupted → more and more insulin is required to achieve the same glucose uptake = insulin resistance
- Earliest biomarker: Elevated fasting or postprandial insulin, not elevated glucose
- Normal blood glucose is ~5 grams total in circulation; type 2 diabetes is essentially one extra teaspoon of glucose
Primary causes of insulin resistance:
- Physical inactivity — the single most important factor; even lean inactive individuals develop it
- Sleep deprivation — 4 hours/night for one week reduces glucose disposal by ~50%
- Hypercortisolemia (chronic stress/elevated Cortisol 皮质醇)
- Energy imbalance — excess caloric intake causing fat spillover from subcutaneous stores into muscle, liver, and pancreas