Сергей Пузанов – Multiple Sclerosis. Everything you need to know (страница 9)
As a result, tracking the global population of sclerotics over time has turned out to be an incredibly non-trivial task. Searching Yandex and Google yields practically zero concrete data prior to the year 2000. I used to think that cases only started being registered once doctors gained the ability to confirm diagnoses via MRI, but that turned out to be a myth.
I was convinced that over the 25 years since MRI was universally adopted for MS diagnosis, the patient count must have doubled or tripled, but there was no clean way to verify it. So, I messaged my editor-in-chief and asked for the publishing house's help. Maria turned out to be far more resourceful than me and instantly figured out how to solve the bottleneck: look to ChatGPT, the artificial intelligence. By the way, last year I read halfway through
Ultimately, in the absence of official stats, using ChatGPT felt like the only accessible shortcut to rapidly aggregate fragmented historical data. Crucial clarification: the neural network doesn't act as a "fortune teller"; it merely summarizes official, scattered statistical registries. Since ChatGPT speaks every human language fluently, I asked it to gather published data from various countries—health ministry reports, epidemiological reviews, national registries, and massive meta-analyses—and use these sources to estimate the global patient population back in 1980. For the record, the neural network flagged this as a highly problematic task, since a standardized tracking system for such diseases didn't exist back then. Naturally, I didn't stop at 1980.
Global MS Patient Population Estimates (Aggregated by AI):
1980: 0.8 to 1.0 million
1990: 1.1 to 1.3 million
2000: 1.5 to 2.0 million
2010: 2.1 to 2.3 million
2020: 2.8 million
2025: 3.2 to 3.5 million
I think the trajectory toward a rapid population explosion is obvious to anyone, even those far removed from mathematics. Armed with these statistics, I couldn't resist asking: "Based on historical incidence data, project the global number of multiple sclerosis patients in the year 2100." ChatGPT complimented me on the unconventional question and then ran the projections.
The baseline scenario (50–60% probability) assumed that multiple sclerosis treatment variables would remain stagnant, with only diagnostic tools improving. In this track, the patient count would scale to 13–14 million by 2100. The pessimistic track, which the AI hit with a 20–30% probability, projected a surge up to 30 million cases. There was also an optimistic scenario: the arrival of radical new medical technologies, including genetic engineering, alongside a massive leap in differential diagnostics. In that track, the planet would harbor 6.2 million MS patients in 2100, but the probability of this outcome sat at a measly 10–20%.
As someone whose favorite school subject was math, I am deeply convinced that numbers do not lie. On the contrary, using mathematically sound models usually yields the most pinpoint projections—provided, of course, that no error sneaked into the baseline data. Furthermore, if a mathematically sound model starts projecting absolute, unadulterated nonsense, the error is almost certainly baked right into the initial inputs.
Based on official numbers, ChatGPT’s baseline projection (50% probability) stated that by the year 2200, the planet would see about 100 million cases of MS, while the pessimistic track (20% probability) claimed the patient population would swell to 270 million. Yes, I also immediately factored in proportional global population growth. However, those same mathematical models stated that the planet's human population in 2200 would cap out between 8 and 10 billion people.
So, the artificial intelligence suggested that if current trends hold, out of 10 billion people living on earth in 2200, the baseline track would yield 100 million MS patients, and the pessimistic track would yield 270 million. The pessimistic model calculated that by 2200, one out of every thirty-seven people on earth would suffer from multiple sclerosis; by 2500, one out of every six; and by roughly the year 3400, sclerosis would be diagnosed in literally every single human being without exception.
When I followed up with, "Doesn't it bother you that your projection requires the relative proportion of MS patients to skyrocket 20- to 30-fold by 2200?" the neural network replied, "A very pertinent observation—and yes, that absolutely raises flags." Shortly after, it heavily adjusted its projections, and after a few more clarifications and roasts on my part, it essentially told me it was "done with this nonsense" and that the global MS population would never exceed 15 million, no matter what happens, period. It seems I managed to train the neural network.
Obviously, such extreme conclusions don't map reality; they merely expose the bankruptcy of the initial data. A model built on numbers stripped of common sense rarely yields a solid forecast, but mapping a real forecast wasn't my objective anyway. The point lies elsewhere—this exercise unevocably proves that the MS diagnostic framework is corrupted by systematic errors. My next question was: "Given your previous answers, what percentage of people diagnosed with G35 do you think received that label by mistake?" The neural network praised the angle and stated quite clearly that the data (backed by a dozen research papers it attached to its reply) indicates that in recent years, MS has been misdiagnosed in 25 to 30% of all cases.
On "Active Demyelination Lesions" on MRI Results
Today, it’s hard to believe, but 40 years ago, clinical practice possessed zero technology capable of visualizing what was happening inside a living brain. A person experiencing strange neurological symptoms would visit a doctor, undergo a long, tedious physical exam, and head home completely stripped of a definitive diagnosis. There were no McDonald criteria, no contrast enhancement, no "demyelination lesions"—or rather, they existed, but you could only find them during an autopsy. Before the arrival of MRI, a patient could feel systematically unwell for years without any way to map the structural warfare unfolding inside their skull.
One of the first to map strange neurological symptoms to physical changes in the brain was Jean-Martin Charcot. In 1868, he observed a patient presenting with tremors, nystagmus, and slurred speech. After her death, he conducted an autopsy and discovered characteristic "plaques" scattered through the brain's white matter. It was Charcot who coined the term
The first successful attempt to systemize this guesswork came in 1965 with the introduction of the "Schumacher criteria." According to these rules, a diagnosis could only be made if a patient experienced at least two relapses, each lasting more than 24 hours and separated by an interval of at least one month. Doctors also had to find objective neurological signs of damage in at least two separate areas of the central nervous system. The patient’s age had to sit strictly between 10 and 50, and any alternative explanations for the symptoms had to be completely ruled out. These criteria heavily prioritized the value of a physical exam and clinical observation over laboratory data, demanding a massive level of qualification, experience, and a willingness to monitor a patient over long stretches of time.
Without an MRI, a diagnosis was assembled literally "by eye." Everything hinged on the doctor’s ability to track and interpret subtle clinical cues: nystagmus, reflex asymmetries, pathological babinski signs, coordination, gait, and sensory maps. Any single detail could turn out to be the master key, but only when framed by the total clinical picture and how it shifted over time. Physicians had to know their patients practically intimately, not just as charts—otherwise, tracking the slippery dynamics of the disease was impossible.
An audit rarely wrapped up in a single visit; observation stretched across months or even years. A doctor had to recall and contrast today’s symptoms against how the patient presented six months or a year ago, ideally factoring in variables like weather, fatigue, stress, and other ambient triggers. Finding multiple sclerosis in a perfectly healthy person was incredibly easy; missing the disease entirely was just as easy; and objectively proving or disproving either was flat-out impossible. Naturally, a system wrapped in so much subjectivity and individual bias had zero chance of long-term stability: it simply left too much room for human error.