The editorial frames the 147-minute survival as predictable physics rather than medical exceptionalism: cerebral metabolic rate of oxygen drops 6-7% per 1°C of cooling, so a 17°C delta reduces brain oxygen demand by roughly an order of magnitude. The neurons aren't surviving on stored ATP — the cellular clock itself got slower, making two wall-clock hours affordable. This reframes the case as a biological analog to CPU deep sleep states and suspend-to-disk.
The case report documents asystolic hypothermic arrest at a core temperature in the 20s°C reversed by ECMO-driven rewarming and prolonged CPR, with neurologically intact discharge. By publishing the 147-minute submersion outcome, the authors implicitly argue that the resuscitation ceiling for severe hypothermia is far higher than legacy 4-6 minute anoxic thresholds suggest, and that aggressive ECMO programs should not call time of death based on submersion duration alone.
By submitting the case report to HN with framing that highlights the survival outlier, js2 surfaces the clinical maxim that has been quietly displacing prior records (the Norwegian skier resuscitated at ~80 minutes and 13.7°C). The 139-point thread treats the case as confirmation that ECMO programs in Norway, Switzerland, Japan, and the Pacific Northwest have been steadily expanding the boundary of what counts as recoverable.
A case report published in *JACC Case Reports* documents a patient pulled from ice water after 147 minutes of submersion, presenting with asystolic hypothermic cardiac arrest and a core temperature deep into the 20s°C. Standard resuscitation logic — that the brain begins dying after four to six minutes without perfusion at body temperature — would have made this a recovery of a corpse. Instead, after extracorporeal membrane oxygenation (ECMO)-driven rewarming and prolonged CPR, the patient was discharged neurologically intact.
The Hacker News thread (139 points) zeroed in on the line clinicians have been quoting for decades: *"Nobody is dead until warm and dead."* It is not a slogan. It is an operational rule about when to stop resuscitating, and the threshold has been pushed steadily outward by ECMO programs in Norway, Switzerland, Japan, and the U.S. Pacific Northwest. The previous well-known outlier — a Norwegian skier resuscitated after roughly 80 minutes of cardiac arrest at 13.7°C — has been quietly displaced.
The numbers are the story: a brain that would suffer irreversible anoxic injury in four to six minutes at 37°C tolerated more than two hours of no circulation at roughly 20°C. That is not a miracle. That is a temperature coefficient.
Every first-year medical student memorizes the rule of thumb: cerebral metabolic rate of oxygen (CMRO₂) drops about 6-7% for every 1°C of cooling. Run the math. Going from 37°C to 20°C is a 17°C delta. Compound the reduction and the brain's oxygen demand falls by roughly an order of magnitude. The neurons aren't surviving on stored ATP for two hours; the *clock got slower*, so two wall-clock hours became something the cellular machinery could afford.
This is the part developers should care about, because the same physics shows up in three places we already pay attention to:
1. CPU deep sleep states. A modern x86 core in C6 cuts dynamic power by stopping the clock and dropping voltage, while preserving architectural state to SRAM or DRAM. The wakeup latency is microseconds — much worse than active execution — but the package power drops 50-100x. ARM's WFI and big.LITTLE migration are the same trade: slow down, preserve state, pay a latency tax later to come back. Cold water does to a neuron what C6 does to a core: it preserves the state vector while collapsing the energy budget.
2. Storage tiering. S3 Standard costs ~$0.023/GB-month and gets you millisecond reads. S3 Glacier Deep Archive costs ~$0.00099/GB-month and gets you 12 hours to first byte. The bytes aren't different. The *response time was traded for survival cost.* That's the hypothermic patient. A normothermic arrest is hot storage you can't afford to keep running. A 20°C arrest is Deep Archive — slow as hell to wake up, but the data is still there if you're patient with the rewarming.
3. Checkpoint/restore systems. CRIU on Linux, Firecracker snapshots, AWS Lambda SnapStart — all built on the same insight: a process's *state* is small and durable; its *execution* is expensive and fragile. If you can freeze fast enough that the inflight work doesn't corrupt the state, you can resume arbitrarily far in the future. ECMO is, mechanically, an external scheduler running the cardiopulmonary loop while the original process is paged out.
The medical literature has a name for the failure mode when this goes wrong: rewarming injury. Bring the patient back too fast and you get reperfusion damage, electrolyte chaos, and arrhythmia. The analog in our world is the cold-start storm: spin up a thousand paused Lambdas simultaneously and you melt the downstream database. State preservation is the easy part; controlled resumption is where most cold systems actually fail.
The community reaction on HN was a useful split. The medical commenters emphasized that this only works when cooling precedes the arrest — drowning in 4°C water cools the brain on the way down, before circulation stops. Cool *after* arrest and you get a corpse with a low temperature. The engineering commenters immediately recognized the order-of-operations problem: you have to checkpoint *before* the crash, not during it. Post-hoc state capture is impossible in both biology and distributed systems.
The practitioner takeaway isn't "build a freezer." It's that the cost of preserving state through a failure scales with how aggressively you slow the system down before the failure hits. Three places this should change your defaults:
Idle-tier design. If your service has a long tail of dormant tenants — abandoned accounts, rarely-accessed projects, seasonal workloads — the dominant cost is keeping them warm, not waking them up. Most teams over-provision the hot path. The hypothermia case is a reminder that the latency penalty of a cold tier is almost always cheaper than the energy penalty of keeping everything ready. If your warm-to-cold ratio is anywhere north of 10:1, you are probably running a normothermic ICU on a population that could survive in a snowbank.
Crash-only design. The medical literature has converged on a brutal rule: don't try to resuscitate slowly. Bypass the failing system entirely (ECMO) and rewarm under controlled conditions. The software equivalent is the crash-only architecture argument Candea and Fox made twenty years ago — don't write graceful shutdown paths, write fast restart paths. The patient who survives 147 minutes does so because the team didn't try to nurse the failing heart; they moved circulation to an external machine and dealt with the original organ on a separate timeline. Your incident response should look more like ECMO than like CPR: stop trying to keep the failing component alive and route around it while you debug.
Snapshot economics. Firecracker snapshot/restore got Lambda cold starts down from seconds to ~200ms by preserving the guest's memory state to disk and restoring it on demand. That's a 10-50x improvement, and it comes from exactly the realization the ECMO programs operationalized: the expensive part is the boot sequence, not the running state. If your platform doesn't have a snapshot path for its expensive-to-warm components — JIT caches, ML model weights, prepared statement pools — you are paying full price every time something fails.
The medical frontier is targeted temperature management *before* an anticipated arrest — cardiac surgery, neuro ICU, even some stroke protocols — explicitly to buy the same metabolic slack the drowning victim got for free. The engineering frontier is the same: pre-emptive throttling and checkpointing of services that are about to be evicted, migrated, or upgraded. We've been doing the inverse for years (auto-scaling *up* in response to load) while ignoring the much larger savings from controlled cool-downs of things that aren't currently earning their keep. The 147-minute number is the upper bound nobody believed existed. The lower bound on how aggressively your stack could be cooling itself when nobody is looking is almost certainly lower than you think.
That is incredible. 2.5 hours underwater, 1.5 hours of CPR. They were instructed not to start rewarming him until he could be given more comprehensive treatment at a hospital. They list 'death' as a differential diagnosis...He didn't come out unscathed though. They describe his progre
"Survival" here being, of course, not a black-and-white thing: > Outcome and Follow-Up > On day 59, the boy was discharged to inpatient neurorehabilitation. At 6-month follow-up, he was giving short commands, standing without support, riding a tricycle, eating soft foods, and relearn
Reminds of Chris Lemons, who survived for 30+ minutes without oxygen at the bottom of the North Sea. Cold water (and experience, like staying calm) probably played a large part. He went back to diving a few weeks after!They made a movie about it: https://en.wikipedia.org/wiki/Las
I'd be curious to read about 1, 2, 5, 10, 20 year follow-up.Party pooper warning.I'm afraid I don't have rose tinted glasses, due to personal experience with a family member with TBI (accident at age 16, 3 weeks in a coma). The aftereffects are profoundly destabilizing to his environm
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
Something similar happened recently in Norway. A tourist was found severely cold in the mountains, in a storm where extracting him took hours. After a while hypothermia got to him, his heart stopped, and only 8 hours later they got it starting again, was technically dead for 20 hours. In this case,