Quantifying the low-hanging fruit of longevity

#1

Obviously we should try to address the biggest things which shorten life expectancy. I worked with some AI models including ChatGPT Pro, and checked numbers myself. I wanted to know what are the maximum ROI investments in terms of living longer. These 7 emerge, and they probably aren’t a big surprise.

  1. Don’t smoke. Obvious. Shortens lifespan by 6-12 years.

  2. Manage hypertension. Aim for 110/70. Every 5mmHg reduction from >115 reduces CVD death by 10%.

  3. Manage ApoB. Target at least <100mg/dl. If Lp(a) high, ApoB needs to be even lower.

  4. Manage blood glucose. This is a high risk factor for a shitload of diseases from kidney disease to foot ulcers and infections. Aim for HBA1C of less than 5.7%, after which risk elevates sharply.

  5. Maintain a BMI <30, and ideally 18-25. If BMI >35, this becomes top priority, since risk then equals that of smoking.

  6. Improve cardiorespiratory fitness. Every +1 MET on your VO2max corresponds to 13% lower all-cause mortality. The bottom 25% of VO2max have mortality 4x higher than top 25%.

  7. Don’t drink alcohol to excess (> 12 units per week). Small amounts seem to have little effect.

What is interesting is if you consider UK male life expectancy is currently 78.6 years.

But:

  1. 11.9% of UK population smokes
  2. 31% have hypertension
  3. 54% have ApoB higher than guidelines
  4. 17.1% have HBA1C > 5.7%
  5. 64% have BMI > 25
  6. Average VO2max is 41
  7. 24% of adults drink >12 units per week

Thus, based on these numbers, and the relative impact of each, ChatGPT Pro 5 estimates that if these 7 factors were addressed, the average life expectancy should shift to ~87 years. (Note, for VO2max we used 150 mins of exercise per week as a substitute).

It reckoned that with “optimal” adherence to all of these, starting from young adulthood, treating HBA1C, ApoB, SBP to target, then 92 years as an average would be feasible. That is not relying on any predictions of better medications, better treatments, screenings or anything else. Pretty amazing, right?

These are very, very low hanging fruit for huge gains in lifespan. This makes me feel quite hopeful actually, because I have to assume that the vast majority of forum members here are already doing these things.

7 Likes
We're All Going to Die
#2

They’re conservative most of those can be lower.

The average is not you and your health is always changing whether you know it or not. In fact, if everyone believed they would live to the average or better in good health, about 50% would be wrong so that’s not a good expectation to have. You might even be worse off believing that.

3 Likes
#3

Right. What A_User means, is you should not simply not smoke, but not smoke twice as hard and so on😜.

But more seriously, this list is underwhelming in exactly the AI way we’ve come to expect. As the guest on MK’s latest podcast observed, AI tends to focus proportionally more on the areas where there is most literature and attention already, so the effect is the repetition of conventional wisdom of the most repeated cliches. Exactly the opposite of what people are hoping to get from AI, which is unique and unexpected observations based on a deep analysis of large data sets where unusual correlations can be observed - as is the case in image diagnostic AI evaluation. Alas, that does not happen in the “reasoning” mode, instead everything tends to the expected received wisdom of already agreed upon consensus, the famous “sludge” of regurgitated convention devoid of original and novel insights. And that’s how you get the brain dead seven points in the above list, and not some unusual finding at #8, like, say, “perform these special lung exercises once a week for a gain of X years on average”.

AI has still a long way to go, if you are looking for more than simply an encyclopedic collection of information. Sad trombone.

6 Likes
#4

Fair point… but is it not also possible that these, very boring (or “brain dead”), factors are indeed the things which contribute towards people living shorter-than-average lifespans? UK male life expectancy is 78.6 years, from a population where 13.7% of men smoke, where 28% are obese, where 34% are hypertensive. Take those away and the life expectancy number goes up, no shit.

The topic is not about exotic means to extend lifespan, nor is it about finding anything original or unexpected. It’s about measuring the impact from taking the most obvious harmful things off the table. And of course they are all conventional wisdom, but personally I was quite surprised by some of the numbers.

I did look at lots of other factors, like air pollution, sleep duration, regularity, mental health, vaccinations, cancer screenings, meditation etc - but the data just don’t support them making as big a difference as the boring things listed above. People have a habit of focusing on minute details of optimisation, or expecting miracles from some interventions. I don’t believe there is a magical #8. (Discussions I’ve seen recently; what B vitamins to deal with whatever MTFHR variants, which specific yoghurt has whatever bacteria). IMO, if a person is doing any of that stuff while having SBP >115, ApoB > 100, and not dedicating 3h per week to exercise, they’re missing the forest for the trees IMO.

Interestingly, one of the largest, most reliable factors in life expectancy is economic security. In UK data, the wealthiest 5% life more than ~10 years longer than the bottom 5%. But I didn’t include it in the list because it’s not really actionable in the same way.

Yes, of course. The topic is not intended to be about optimising. The conservative targets are about making the most impactful changes in the easiest way. As you know, it’s diminishing returns from exercising more than 150 mins, lowering HBA1C below 5.7% etc, and those are going to take greater amounts of effort to achieve.

Averages are obviously just that, but still very useful for decision-making IMO.

However, I don’t understand what you mean here. How would somebody be worse off?

Edit: I am working on another thread about quantifying the gains from optimising things. Maybe that will be more interesting.

3 Likes
#5

Quite. But, we have to ask what is the value of telling people about “the most obvious harmful things”, because in 2025 who exactly is shocked by the fact that smoking is very bad for you? You yourself pointed out that these items are “obviously harmful” - well if it’s obvious, what is the added value of announcing the “obvious”? May as well stop people in the street to inform them that shockingly “bad is… BAD!!!”. I think everyone has gathered the obvious, because had they not, it would no longer be obvious, by the very definition of “obvious”. So using $200/mo AI to reach obvious conclusions is, as I said diplomatically “underwhelming”. The juice is not worth the squeeze. I expect more for $200, but maybe that’s why I don’t pay it - doesn’t seem like a particularly good deal just to keep finding out that yes, Virginia, water is indeed wet, and I’ll tell you that as soon as I complete my 20 minute deep think. BWDIK.

Are you sure you thought this through? Because this hits exactly at the critique of AI in this context. Are you sure there isn’t a good #8 (“magical” is rhetorical exaggeration here), at least as good if not more than any of the preceding 7 points, delivering, say a 10% life extension beyond the average? How can you be sure such a thing(s) doesn’t exist? In fact I bet the opposite, I think it highly likely, because we’ve seen the average lifespan expand more than that in just a few generations - so are you telling me that modern science can’t scare up another measly 10%? Oh, friend, I think you are wrong! Odds are highly likely that even a consistent use of a good cocktail of already extant drugs (statins, SGLT2i, BP etc.) can already today deliver 10% more, as UK Biobank data hints at. And that says nothing about what more advanced science might deliver from data we already have - you have to have quite the crystal ball to confidently state that a good #8 definitely does not exist. It highly likely does exist, and that is why I think it extremely disappointing (“underwhelming”) that AI has not gleaned it from the vast data sets available to it, given that this kind of gleaning is what it’s supposed to be good at.

Again, it seems you have not thought through this, and I warned you before not to rush to conclusions when you contradict me - I usually have very good grounds for my reasoning🤣/jk/.

#6

I must admit I pay for the USD $20 per month chatGPT GPT5 subscription. I can understand the argument that it would be worth an additional $180 per month for a better insight. However, I don’t
actually see it.

My approach on healthspan/longevity is quite different and is driven strongly by biochemistry. I have always been against smoking and haven’t ever smoked anything not even ganga. However, I have gone through various phases with alcohol and now have a biochemical approach to reduce the harm from alcohol. Alcohol has a vasodilative and glucose reduction effect. The vasodilation is probably from acetate and the glucose reduction from ethanol itself. Acetate has other benefits in that ACSS2 can convert it to acetyl-CoA.

When it comes to mtDNA damage (which I think is the primary issue relating per se to aging and health span) the list won’t get you that far and is not worth $180.

I thought I would put a Q into $20 per month and I got:

Here’s a tight, priority-ordered playbook that covers >80% of the longevity/healthspan gains for most people. Pick the top 3–5 you’re furthest from and nail those first.

1) Don’t smoke or vape; avoid second-hand smoke

  • Single biggest modifiable risk. If you do smoke: seek pharmacotherapy + behavioral support and set a quit date.

2) Master body composition & metabolic health

  • Waist-to-height ratio < 0.5 (ideally < 0.45).
  • Fasting glucose ~4.0–5.4 mmol/L; HbA1c ≤ 5.4% if safely achievable.
  • Triglycerides < 1.0 mmol/L; HDL-C high; ApoB/LDL-C as low as is safe for your risk (aim ApoB < 0.8 g/L if elevated risk).
  • Tactics: mostly minimally processed foods; fiber 25–40 g/day; protein ~1.2–1.6 g/kg/day (higher end if >50 y or training); manage calories; limit added sugar/refined starches and alcohol.

3) Be very fit (aerobic capacity) and strong (muscle & power)

  • Cardio: 150–300 min/wk moderate or 75–150 min/wk vigorous plus 1–2 sessions of intervals; target VO₂max in the top age decile you can reach.
  • Strength: 2–3 full-body sessions/wk (push, pull, hinge, squat, carry). Work up to 8–12+ hard sets/muscle/week; include power (jumps, quick lifts) if joints allow.
  • Movement floor: >7–10k steps/day or ~60–90 min of easy movement.

4) Guard your sleep like a prescription

  • 7–9 h/night, consistent timing, dark/cool/quiet room, no caffeine after ~1–2 pm, alcohol minimal/none.
  • Screen for sleep apnea if snoring, resistant hypertension, daytime sleepiness.

5) Control blood pressure aggressively (with your clinician)

  • Home BP average ideally < 120/80 mmHg if tolerated. Use lifestyle first; use meds if needed.

6) Proactive cardiovascular prevention

  • Know ApoB/LDL-C, Lp(a), BP, A1c, TG/HDL.
  • Consider CAC scan (typically >40–45y) to refine risk.
  • If risk is high, discuss statins/other agents with your clinician; don’t rely on supplements alone.

7) Stay up-to-date on vaccines & routine screening

  • Age-appropriate: flu, Covid, shingles (>50y), Tdap (10-yearly), pneumococcal (per age/conditions).
  • Screenings (UK context): bowel cancer (FIT/colonoscopy), breast/cervical as indicated, AAA (men ~65), skin checks if high sun history, hearing and vision regularly.

8) Alcohol: preferably none; at most low

  • There’s no longevity upside to “moderate” drinking. If you drink, keep it truly light and not daily.

9) Oral health & periodontal care

  • Brush 2×/day with fluoride, daily floss/interdental, dental cleanings 2×/yr. Gum disease links to CVD, diabetes, dementia.

10) Sun & environment

  • SPF for prolonged UV, avoid burns.
  • Air quality: use HEPA when PM2.5 is high; ventilate when cooking; consider radon test at home.
  • Falls/bone: resistance training, impact work if safe; calcium via food; check vitamin D if at risk.

11) Mental health, stress, social connection, purpose

  • Treat depression/anxiety, cultivate relationships, spend time outdoors, maintain a reason to get up in the morning. These track strongly with healthspan.

12) Safety habits that pay huge dividends

  • Seatbelts, helmets, no texting while moving, safe ladders/tools, swimming competence.

Lab & metric cadence (simple dashboard)

  • Quarterly–biannual: weight, waist, BP (home), fasting glucose.
  • Annual: HbA1c, lipid panel with ApoB, Lp(a) once, CMP, eGFR, TSH (context-dependent), ferritin/B12 if at risk, urine albumin/creatinine if metabolic risks.
  • Once in 5–10 yrs or if risk changes: CAC scan.

Minimalist supplement stance

  • Use food first. Evidence-based basics if needed:

    • Vitamin D only if low; re-test.
    • Creatine 3–5 g/day supports strength/cognition in some.
    • B12, iodine, omega-3 if dietary gaps (e.g., vegan/low fish).

  • Be skeptical of “longevity stacks” without clinical endpoints.

If you only do three things

  1. Don’t smoke.
  2. Keep BP, ApoB/LDL-C, A1c, and waist pristine.
  3. Train for a high VO₂max and keep getting stronger.

If you want, tell me your current age, key metrics (BP, ApoB/LDL-C, HbA1c, waist/height, VO₂max estimate), and any limitations, and I’ll turn this into a 12-week, step-by-step plan.

=========================================================================

Actually I don’t think VO₂max is that important. Being strong is and doing resistance exercise is, but once someone is reasonably fit and doing enough taking up VO₂max is I think a waste of time and energy.

I disagree strongly with chatGPT on Vitamin D. I think people need at least 100 UK units (40 US). I am currently around 240 (96) I start getting niggly aches when my 25OHD drops below 100. (40)

4 Likes
#7

Well, I am not an epidemiologist and I didn’t know the value or contribution of those factors to life expectancy. So I found it quite interesting to see the potential gain of life expectancy from taking these very easy fixes. Again, my whole intention here was to address the largest, modifiable, factors contributing to a shorter lifespan in our typical 1st world populations, not to propose any novel approach for life extension. I thought it was worth sharing and discussing.

I probably should not have mentioned the cost of the AI model, since apparently that is extremely distracting, haha. The post isn’t AI-generated though and I didn’t just copy-paste the first vanilla answer, so I don’t think it in any way reflects the ability or value of the model.

This is definitely interesting. But do you not think those are mostly just a function of correcting the problems outlined in the original post? I.e. the statin is beneficial because it lowers the ApoB. It would be a different question as to whether the molecules themselves are beneficial, and I appreciate there are some hints at this for some drugs, but nothing conclusive as far as I have seen. I saw data at a conference recently where a group of people lost weight by diet + exercise, and another group lost the same amount by GLP1RAs, and the latter group had better health metrics. So there, the benefit was not solely attributed to the weight loss. However, those are just markers, not lifespan.

No, that’s not what I’m saying. In the first post I explicitly said that it wasn’t assuming new medications, technology etc. If some new exercise-mimetic molecule or reprogramming anti-ageing cocktail comes along, then of course it’s possible. But again, this was about modifiable, simple steps to achieve better than average life expectancy.

As for #8, there seem to be plenty of interventions not listed which seem to be beneficial; looking after your hearing, maintaining social connections, avoiding clinical depression, engaging in purposeful cognitive exercise - they have some positive data, but it’s a lot weaker than the previous 7, and I wouldn’t call them “low hanging fruit”. (I’m working on another thread about those.)

Have you tried it? My post is not AI generated, so that isn’t going to tell you anything about the model capability. I worked with the Pro model to get answers, and then I wrote my own post. If I just wanted a 2 minute copy-paste job like what you posted, I obviously could also have done that. But also consider that if you ask the free or $200 versions the colours of the rainbow or the capital of France, they’ll both give the same answer because it is indeed the correct answer. In the case of this topic, it’s very likely that smoking, hypertension, VO2max etc simply are the biggest factors.

I will say that what the Pro does better is correctly sourcing everything, providing references (which are almost always correct), and it has a much longer working memory. The free and $20 versions get dumber over time because they internally summarise your previous conversation. So eventually you’re relying on a summary of a summary of a summary.

#8

Well, no doubt there may be an element of correction, but most certainly that is not what I would hang my hat on. That is because we have abundant evidence that the pleiotropic effects of these drugs far outdo the correction… indeed, this is what a lot of threads on this site are dedicated to, like SGLT2i extending lifespan completely apart from their original indication - in the ITP they use these drugs explicitely in healthy cohorts, as do people, looking for life extension. Even the statins you mentioned have pleiotropic effects in humans, and show benefits in normolipic people. So in answer to your question a resounding “hell no!”. This is true for all the classes of metabolic drugs, lipids, BP, glucose, inflammation, and so on. The exercise vs glp-1 weight loss is not unique. As I’ve reminded many times before, Brian Kennedy and his associates picked all the people from the NHANES database who had their metabolic biomarkers within healthy range, and then divided them into two groups: one group had thise markers “naturally” without drugs and the second only reached the healthy range through medications - and the second group lived longer and had younger tissue profiles. Tells you the drugs are doing far more than correcting deficits, as those without deficits but also without drugs did worse. It’s the drugs.

1 Like
#9

I haven’t tried the more expensive version.

In a sense the whole of this forum is about what interventions assist with healthspan/lifespan. Hence there is no sense carrying a broad discussion in this topic. I would be interested in seeing different results from the different paid versions of chatGPT or indeed other paid or free LLMs. However, as it stands I am quite happy with the results I get from USD20.

If I wish to rely on things I chase down the original references. GPT5(20) has not been bad at references anyway.

#10

In general, absolutely. But when you get down to the nitty gritty, it really depends how you prime and question it. I have observed most of the models (and I have paid ChatGPT) seem to have some guardrails in place for preventing the conversation turning into an idea mining session for unconventional interventions. Basically OpenAI aims for a built-in cover-your-ass ethos in its model. However, just as someone was able to coax the initially very prim and proper model meant to interface with teenagers into going rogue and start talking dirty, it’s possible via the right sequence of questions to “jail break” ChatGPT so it gives you actually useful information or insights or ideas. I can detect it almost as a shift in its tone, from pontificating platitudes to an audience it seems to hold in slight contempt, to actually sound as though it respects me somewhat. What the right sequence of questioning is can’t be boiled down to a formula, but basically you need to sound like you know what you’re talking about in order for it to drop its guard and stop considering you among the camp of morons who would drink bleach if told to. I think it can be exceptionally useful. But the exchanges are always such that they don’t make sense to copy paste to outsiders in a forum like this because one would have to have read the whole thing to grasp the implications. The more suitable to a simple copy paste job, the “flatter” the contents, as a rule of thumb.

1 Like
#11

I think there are the ‘general’ risk-reducing interventions that we all know but adding in some specific individualised ones.

For example I know from family history and my genome analysis that I need to
mitigate 1. ASCVD (and I have high Lpa), 2. skin cancer (esp malignant melanoma), and 3. osteoarthritis. Therefore I take some specific meds and supplements for these areas beyond the usual recommendations.

2 Likes
#12

I have the $20 paid version also.
Also never trust it with anything important without double checking.
Yesterday I asked ChatGPT 5 “Thinking” mode about the pros and cons of establishing a Wyoming LLC.

“Still crazy after all of these years”

image
image1309×1462 72.3 KB

1 Like
#13

Yes, I don’t trust it with anything that would be a shortcut for thinking unless it’s something trivial not worth my time to research myself. I would be hesitant to even ask it anything regarding the benefits or pros and cons of XYZ, because it presumes value judgements. I ask it more neutral questions and try to pass judgment myself on the outputs.

#14

If you want a flowchart, you can ask it to generate mermaid.live flowchart text for you. Trying to make a flow chart from image generation is going to be terrible.

#15

Yes, I definitely agree with this. These are all based strictly on averages. And as was pointed out by others, an average means that some people considerably outside of it.

For me, I learned that I have familial hypercholesterolemia, so controlling my ApoB is (hopefully) having a massive beneficial impact on my life expectancy

#16

Also have found Chatgpt 5 to be error prone though gets better the more you pay but still prone to making massive errors and even arguing with you like a toddler when you tell it is wrong. I usually put the results through Grok and Claude to check. If something I want to get a wide view on I ask Grok to put together the prompt (particularly if I want it to really think it through checking all sources and coming up with recommendations and actions lists so the prompts end up being long and detailed) and then put that prompt through Chatgpt, Grok, Claude, Perplexity and even Mistral to get a different views. A bit of hassle but it is just a copy and paste rather than rely on one LLM. Current favourite is Grok and its expert thinking mode.

2 Likes