The Next Fifty Years

TL;DR. Four technologies have already crossed from promise into practice and will define the next few decades: mRNA platforms that turn a genetic sequence into a vaccine in weeks, gene editing that has already cured sickle cell disease, GLP-1 based drugs that turned out to treat far more than diabetes, and machine learning applied to images, proteins, and records. Set against them are three forces pushing the other way: antimicrobial resistance, climate change, and population ageing. The single largest determinant of how the next fifty years actually go is none of these. It is whether the treatments that already exist reach the people who need them, because the gap between what medicine can do and what most of the world receives is currently wider than the gap between what medicine can do and what it will be able to do.

Key takeaways

• mRNA went from sequence to authorised vaccine in 11 months during the COVID-19 pandemic, compressing a decade-long process. The same platform is now in trials for cancer, influenza, and RSV.
• The first CRISPR therapy was approved in 2023. It works, and it costs 2 to 3 million US dollars, in a disease concentrated in the world's poorest countries.
• AI has genuinely arrived in two places: image interpretation and protein structure prediction. Its record in clinical prediction from records is far more mixed than the coverage suggests.
• Antimicrobial resistance could reverse a century of gains, with forecasts of tens of millions of cumulative deaths by 2050.
• The number of people over 60 will roughly double by 2050, mostly in countries with the least developed health systems, which makes dementia and multimorbidity the defining clinical problem of the century.
• Do not expect ageing to be cured. Expect incremental compression of the years spent ill, which would be a large achievement.

What has actually changed already

Before forecasting, it is worth marking how much of this book would have been unrecognisable in 2000: hepatitis C was incurable, metastatic melanoma was a death sentence within a year, cystic fibrosis was a childhood disease, HIV treatment involved handfuls of pills with disfiguring side effects, sickle cell had no cure, spinal muscular atrophy killed infants, and the idea that a diabetes drug would become the most effective obesity treatment ever developed would have been dismissed. All six changed in about twenty years.

That is the base rate to keep in mind when reading what follows.

The mRNA platform

In short: Designing a vaccine became a text-editing problem, which is why it took 11 months rather than a decade.

The idea is simple and took thirty years of unglamorous work to make practical: instead of manufacturing a protein and injecting it, deliver the instructions and let the patient's own cells make it. Katalin Karikó and Drew Weissman's key contribution, published in 2005 and awarded the Nobel Prize in 2023, was discovering that modifying one of RNA's building blocks (pseudouridine) stopped the immune system from destroying the message before it could be read.

Why it matters beyond COVID-19:

• Speed. Designing a new vaccine becomes a text-editing problem. The manufacturing process is identical whatever the sequence, which is what allowed 11 months from published genome to authorised vaccine.
• Individualised cancer vaccines. Sequence a patient's tumour, identify the mutated proteins unique to it, and encode them in an mRNA vaccine given alongside a checkpoint inhibitor. Phase 2 results in melanoma reported substantial reductions in recurrence, and phase 3 trials are running.
• Vaccines against hard targets: universal influenza, RSV, and, in development, HIV, malaria, and TB.
• Protein replacement, delivering instructions for a missing enzyme rather than infusing it.

The limitations are real: mRNA is fragile and requires cold storage (improving), the immune response to repeated dosing complicates chronic use, and delivering it to tissues other than muscle and liver remains difficult.

Gene editing and gene therapy

In short: Eleven years from the founding paper to an approved cure, with the remaining obstacle being delivery and price rather than science.

CRISPR-Cas9, adapted from a bacterial immune system, provides a way to cut DNA at a chosen sequence, after which the cell's repair machinery can be steered to disable a gene or insert a correction. It went from the founding 2012 paper to an approved therapy in eleven years, an unusually short interval.

Where it stands:

• Approved: exa-cel for sickle cell disease and beta thalassemia (Chapter 48), plus viral-vector gene therapies for spinal muscular atrophy, haemophilia, some inherited retinal disease, and several metabolic conditions.
• In vivo editing is the crucial next step. Current therapies require removing a patient's stem cells, editing them in a laboratory, destroying their bone marrow with chemotherapy, and reinfusing, which is a months-long process available only in specialist centres. Editing inside the body would remove all of that. Early results in transthyretin amyloidosis and in lowering cholesterol through PCSK9 editing have shown durable effects from a single infusion.
• Base and prime editing change individual DNA letters without cutting both strands, which is safer and covers most disease-causing mutations. In 2025 an infant with a fatal urea cycle disorder was treated with a bespoke base-editing therapy designed and manufactured in about six months, which points toward a future of individualised treatments for ultra-rare disease.

The obstacle is not technical. It is cost, delivery infrastructure, and the fact that the diseases most amenable to gene therapy are concentrated in countries that cannot pay millions per patient. Whether that gets solved is a political and economic question, not a scientific one.

The GLP-1 story, and what it signals

In short: A large share of chronic disease shares metabolic roots, so one upstream intervention produces benefits across unrelated specialties.

Semaglutide and tirzepatide have produced weight loss previously achievable only by surgery, and then, in outcome trials, reduced cardiovascular events, slowed kidney disease progression, improved heart failure with preserved ejection fraction, and reduced sleep apnoea severity. Trials are running in Alzheimer's disease, alcohol and nicotine use, and metabolic liver disease.

The reason to feature this is less the drug than what it reveals: a large fraction of chronic disease shares metabolic and inflammatory roots, so an intervention upstream of many of them can produce benefits across apparently unrelated specialties. Expect more of this pattern, and expect some of the current trials to disappoint, since the same breadth that makes the hypothesis exciting makes it easy to over-extend.

The open questions are long-term safety across decades of use, the loss of lean mass alongside fat, what happens on discontinuation (weight returns, because the drug lowers the defended weight rather than resetting it permanently), and cost at population scale.

Machine learning in medicine

In short: It genuinely works on images and protein structures, its record on predicting from records is far weaker, and models trained on one population fail on others.

The realistic assessment separates three very different things.

Where it works now. Image interpretation: diabetic retinopathy screening from retinal photographs is deployed and approved, and works particularly well where ophthalmologists are scarce. AI assistance improves adenoma detection in colonoscopy and cancer detection in mammography in prospective trials. Automated detection of large vessel occlusion on stroke CT speeds thrombectomy pathways. AlphaFold solved protein structure prediction to a useful accuracy and released structures for essentially every known protein, which has accelerated drug discovery and won a share of the 2024 Nobel Prize in Chemistry.

Where it is promising and unproven. Clinical prediction from electronic records has an uneven record: a widely deployed sepsis prediction model was found in external validation to perform far worse than advertised. Ambient documentation tools that draft notes from a consultation are spreading rapidly and appear to reduce administrative burden. Large language models pass medical examinations comfortably, which measures something other than clinical competence.

Where the risks are. Models trained on one population perform worse on others, which can entrench existing inequity. A widely used US algorithm allocating care management resources was found to systematically under-refer Black patients because it used healthcare spending as a proxy for need, and less had historically been spent on them. Automation bias leads clinicians to defer to confident-sounding output. And a model that improves detection without improving outcomes may simply industrialise overdiagnosis.

The honest expectation for the next decade is substantial gains in throughput, documentation, and image reading, with clinical decision-making changing more slowly than the announcements imply.

Ageing biology

In short: The realistic goal is compressing the years spent ill rather than extending maximum lifespan, and commercial longevity products are far ahead of their evidence.

The field has moved from fringe to serious, and it is still oversold.

What is established: ageing involves identifiable, measurable processes, catalogued as the hallmarks of ageing (genomic instability, telomere attrition, epigenetic alteration, loss of proteostasis, mitochondrial dysfunction, cellular senescence, stem cell exhaustion, altered intercellular communication, and others). Interventions that extend lifespan in mice are numerous: caloric restriction, rapamycin, and several genetic manipulations.

What is not established: that any of this extends healthy human lifespan. Senolytics, which selectively kill senescent cells that accumulate with age and secrete inflammatory signals, have produced striking results in mice and preliminary results in humans. Epigenetic clocks measure biological age from DNA methylation patterns and predict mortality better than chronological age, and whether they are a cause, a consequence, or a correlate is unresolved. Partial reprogramming, using the Yamanaka factors to rejuvenate cells without turning them into stem cells, has produced remarkable results in animals and carries a real cancer risk.

The plausible near-term goal is not extending maximum lifespan but compressing morbidity: shortening the period of illness and dependency at the end of life. If drugs targeting ageing biology delayed several age-related diseases simultaneously by a few years, the aggregate benefit would exceed that of curing any single one.

Treat any commercial longevity product with the scepticism its evidence deserves, which is considerable.

Replacing organs

In short: Pig kidneys are already being transplanted into people, and lab-grown insulin-producing cells are in trials.

Xenotransplantation has moved from theory to human recipients. Genetically modified pig kidneys and hearts, with the genes causing hyperacute rejection removed and human regulatory genes added, have been transplanted into a small number of patients from 2022 onward, with survival so far measured in weeks to months. If rejection and the risk of transmitting porcine viruses can be managed, the organ shortage becomes solvable in principle, and the ethical debates about animal use and about informed consent in desperate patients become urgent in practice.

Organoids and tissue engineering: lab-grown miniature organs already serve as disease models and drug-testing platforms (CF organoids can test which modulator works for an individual patient's mutation), and transplantable engineered tissue remains distant for complex organs.

Machine perfusion of donor organs, keeping them functioning and assessable outside the body, is already increasing the number of usable organs.

Stem-cell derived cell therapies: insulin-producing islets for type 1 diabetes and dopamine neurons for Parkinson's disease are both in clinical trials with encouraging early results, and both face the same immune rejection problem, which gene-edited hypoimmune cells may solve.

The forces pushing the other way

In short: Antimicrobial resistance, climate, pandemics, ageing populations, and mental health, any of which could outweigh the advances above.

Antimicrobial resistance. The most serious. Modern surgery, chemotherapy, transplantation, and neonatal intensive care all assume working antibiotics. Forecasts of tens of millions of cumulative deaths by 2050 are uncertain and the direction is not. The solutions are known (stewardship, diagnostics, vaccination, infection control, agricultural restriction, and new economic models for antibiotic development) and are being implemented slowly.

Climate change. Heat mortality is already measurable and rising, particularly among outdoor workers and older people. The ranges of Aedes and Anopheles mosquitoes are shifting to higher latitudes and altitudes, carrying dengue and malaria into populations with no immunity and no control programmes. Pollen seasons are lengthening, worsening asthma and allergy. Crop yields, water security, and displacement all have downstream effects, and the largest health impacts will fall on populations that contributed least to the cause.

Pandemics. COVID-19 was not the worst plausible pandemic. A pathogen with influenza's transmissibility and a higher fatality rate remains possible, and the drivers of spillover (agricultural expansion, dense livestock production, wildlife trade, air travel) have not diminished. Preparedness has improved technically, through platform vaccines and surveillance networks, and has weakened politically, through fractured international cooperation and eroded trust in public health.

Ageing and multimorbidity. The number of people over 60 will roughly double by 2050, with most of the growth in low- and middle-income countries. The clinical consequence is that the typical patient will have four or five conditions and take ten medications, and medicine is organised around single diseases treated by single-disease specialists following single-disease guidelines. This is the most predictable and least addressed problem in the list.

Mental health. Over a billion people already live with a mental health condition, treatment coverage is low everywhere and negligible in many countries, and no new mechanism of drug action has reached psychiatric practice in decades except for ketamine and, in 2024, a muscarinic antipsychotic.

What will probably not happen

Worth stating, because prediction failures in this field are dominated by over-optimism about timelines.

• Ageing will not be cured, and human maximum lifespan is unlikely to increase substantially in fifty years.
• Cancer will not be cured as a category, because it is hundreds of diseases evolving inside individual patients. Expect more cancers becoming chronic and more prevented outright.
• There will not be a single cure for dementia, and there may be a combination of prevention, early detection by blood test, and partial disease modification.
• AI will not replace clinicians, and it will change what a large part of their work consists of.
• Personalised genomic medicine will not deliver what was promised in 2003, when the human genome was completed. It has delivered enormously in cancer, rare disease, and pharmacogenomics, and very little in the common polygenic diseases that account for most illness.

The question that matters most

In short: The gap between what medicine already knows and what most of humanity receives is larger than the gap between now and fifty years from now.

Every technology in this chapter is expensive, and most of them arrive first in the countries with the least disease. Meanwhile:

• Insulin, discovered in 1922 and sold for a dollar, is unaffordable for people who die without it.
• Inhaled corticosteroids, which transformed asthma mortality, are unavailable in much of the world.
• Hepatitis C is curable in eight weeks and most people with it are undiagnosed.
• A cheap blood test in pregnancy and one penicillin injection prevent congenital syphilis, which is rising.
• Oral rehydration salts cost cents and diarrhoea still kills hundreds of thousands of children.
• Cervical cancer is vaccine-preventable and screening-detectable and kills over 300,000 women a year.

The gap between what medicine already knows and what most of humanity receives is larger than the gap between what medicine knows now and what it will know in fifty years. Closing the first gap requires no discoveries at all: it requires supply chains, health workers, financing, and political attention.

That is not a reason to be pessimistic about the science. It is a reason to be precise about where the remaining problem is.

A closing note

In short: Three things worth taking away: disease is mechanism, treatment is trade, and the largest gains were never dramatic.

This book has covered the diseases that most people get and the ones that shaped history. If it has done its job, three things should now be clearer than they were.

Most disease is a mechanism, not a mystery. Something regulates something, the regulation fails, and the consequences follow logically from what that thing was for. Once you can say what is broken, the treatments stop being an arbitrary list and become interventions at specific points in a chain.

Every treatment is a trade. Drugs work because they interfere with a process, and the side effects come from the same interference happening where you did not want it. That is a reason to understand the trade rather than to fear the drug.

The largest gains were never dramatic. Clean water, sewers, vaccination, salt iodisation, road safety law, and tobacco taxation saved more lives than every operating theatre in history. The diseases in this book are mostly the residue left after those measures did their work, and the places where those measures have not been implemented are where the residue is still enormous.

If you take one practical thing away, make it this: understanding what a treatment is for is the difference between following instructions and managing a condition. Most of the diseases in this book are managed, day after day, by the person who has them.

Sources and notes

mRNA vaccine platform: Karikó and Weissman, Immunity, 2005; Nobel Prize in Physiology or Medicine, 2023. Individualised neoantigen mRNA cancer vaccines: KEYNOTE-942 phase 2b melanoma results, 2023. CRISPR: Jinek et al., Science, 2012; Nobel Prize in Chemistry, 2020 to Doudna and Charpentier; exa-cel approvals, late 2023. In vivo editing: NTLA-2001 in transthyretin amyloidosis, NEJM, 2021; VERVE PCSK9 base editing early results. Bespoke base editing in an infant: reported in NEJM, 2025. GLP-1 outcome trials: SELECT (NEJM, 2023), FLOW (NEJM, 2024), STEP-HFpEF, SURMOUNT-OSA. AlphaFold: Jumper et al., Nature, 2021; Nobel Prize in Chemistry, 2024. Sepsis model external validation: Wong et al., JAMA Internal Medicine, 2021. Algorithmic bias in care management: Obermeyer et al., Science, 2019. Hallmarks of ageing: López-Otín et al., Cell, 2013 and 2023. Senolytics and epigenetic clocks: reviewed in Nature Aging and Cell Metabolism. Xenotransplantation: case reports from 2022 onward in NEJM and Nature Medicine. Antimicrobial resistance forecasts: GRAM Project, The Lancet, 2024. Population ageing: UN World Population Prospects. Access gaps: WHO Essential Medicines and access reports.

Open questions. Essentially everything in this chapter. Forecasts about medicine have a poor record, and the specific failure mode is consistently over-optimism about timelines and under-estimation of how long it takes for a proven treatment to reach the people who need it.

That is the end of the argument. Two reference chapters follow, and they exist because of the last point above: a drug you cannot name the mechanism of is a drug you are taking on trust. First the GLP-1 class in full, since it is the most prescribed, most discussed, and most poorly explained group of drugs in current medicine, and then a decoder for drug names, lab values, and medical words in general. After those, the glossary collects the terms and the sources page explains where the numbers came from. 👉