# What Medicine Could Be > [!abstract] Thesis > Medicine became much better, and most persons do not know how much. It can become better more quickly. The limit is not the science. The limits are funds, institutions and incentives. Source: Saloni Dattani, "What Medicine Could Be", *Works in Progress*, Issue 25. Dattani is an editor at *Works in Progress*. Dattani is also an advisor to Coefficient Giving on the reform of clinical trials. ## Core idea Most persons do not know how much medicine became better in their life. Life expectancy became two times longer in 100 years. This number is abstract to most persons. A description of life in 1950 shows the change more clearly. Medical innovation is a continuous stream. It is not a set of rare accidents. New treatments, vaccines and tools come each year. Many good treatments are in the pipeline at this time. Some of them will not be available to patients for years, and some for decades. The cause is not the science. The causes are funds, institutions and incentives. Persons can change these three items. ## How far medicine moved ### Heart attack - In 1950, almost no person near a patient with a heart attack knew how to help. CPR was not in use. A doctor gave morphine and told the patient to stay in bed. - Doctors thought that it was not possible to do more after the heart stopped. - In the 1970s, cardiologists found that each minute without oxygen kills more heart muscle. Dead muscle stays dead. - Doctors then opened the blocked vessels with a thrombolytic drug or a balloon catheter. This made the damage smaller. - After that, deaths in the first month after the most dangerous type of heart attack decreased by two thirds. - At this time, a person in a rich country has one quarter of the 1950s risk of death from cardiovascular disease, at the same age. ### Other examples - In 1925, a person with type 1 diabetes had some months of life. At this time, a small device on the arm monitors the blood sugar and releases insulin. - Before the 1970s, 15% of children with leukemia were alive five years after the diagnosis. At this time, 85% are alive. - In 1925, penicillin was not available. A simple wound infection caused sepsis and death in some cases. - Smallpox killed millions of persons each year. ### Survival across birth cohorts - In France, two thirds of the 1900 cohort were alive at age 60. For the 1960 cohort, 90% were alive at age 60. - In 1950, the shortest-lived 1% of babies in France were alive for less than three months. For the 1990 cohort, the same 1% were alive for 25 years or more. - The change was largest at the bottom of the range. Thus the difference in life expectancy between persons became smaller. - Disease kills a smaller number of children. A person has more friends and family that are alive during a long life. > [!example] Economic value > Economists give an estimate of $95 trillion for the value of the longer life expectancy in the US from 1970 to 2000. Half of this value came from the lower number of deaths from heart disease. ## What the charts show | Chart | What it shows | Primary point | | --- | --- | --- | | 1. US cardiovascular deaths | Deaths for each 100,000 persons, adjusted for age, from 1933 to 2020. | The rate was near 750 in 1950 and near 200 in 2020. The rate starts to decrease near 1970. The curve is flat after 2010. | | 2. French birth cohorts | The age at the end of life for each share of a cohort, by year of birth. | All lines increase. The lines for the shortest-lived persons increase the most. | | 3. Childhood leukemia | The share of children that are alive in trials, by year of diagnosis. | In acute lymphoblastic leukemia, the 10-year survival increased from near 10% to near 85%. The periods are 1968 to 1970 and 2010 to 2015. In acute myeloid leukemia, it increased from near 25% to near 70%. | | 4. New-use approvals | The cumulative share of approvals for new uses, by years from patent expiry. | More than 90% of the approvals come before patent expiry. Almost none come after. | | 5. Eroom's law | FDA-approved drugs for each billion dollars of R&D. | The rate decreased from tens of drugs in the 1950s to near one drug in the 2010s. It becomes half each nine years. After 2010, the rate increased again for rare disease drugs. | ## How the methods changed For most of history, a new treatment was rare. The second vaccine came almost 90 years after Jenner found the first. Persons used plants and natural remedies. They did not know which ingredient had an effect. Contamination was frequent. It was not possible to make the remedies at scale with the same quality each time. In the 1800s, chemistry became a science. Then it was possible to isolate compounds, to make new compounds, and to make a test of thousands of compounds at the same time. Soil gave antibiotics. Fungi gave statins. Coal tar and dyes gave sulfa drugs. Random discoveries became a regular method. In the late 1900s, scientists found the structure of the proteins that cause disease. They made molecules that fit into these proteins and stop them. This method has the name "rational drug design". Scientists also changed bacteria and yeast. These then make insulin, clotting factors and antibodies in large quantities. AI extends this method. AlphaFold calculates the shape of a protein in minutes. This task needed years before. Other tools make new proteins that are not in nature. Some tools find microbial genes for antibiotics that the microbes do not use in standard laboratory conditions. New CRISPR enzymes are much smaller than the enzymes in use at this time. They can cut and replace long parts of DNA. They can change many genes at the same time. The cost of the tools decreased very quickly: - The Human Genome Project finished in 2003. One genome cost $50 million and six months of work. - At this time, one genome costs some hundred dollars and four hours or less of work. - In 200 years, the resolution of the best microscopes increased more than 10,000 times. - Until the 1930s, no person saw a virus. At this time, we can see viruses down to their atoms. Recent results show the rate of change: - A new HIV drug prevents infection with almost 100% efficacy. A patient gets one dose each six months. - New drugs decrease cholesterol by 60% in patients that use statins. - In some cancers, new treatments decrease the rate of progression by more than half. - In the last five years, we got the first vaccines against malaria, chikungunya and RSV, and against Covid-19. ## Where the system has problems 1. **No person supplies funds for neglected diseases.** - The first malaria vaccine came from research in the 1990s. It was not possible for the researchers to find funds for each step. - Aid funds and philanthropy supplied the funds for it. Children got the vaccine only after decades. - All tropical diseases together get $4 billion of R&D each year. This is six cents for each $1,000 of income in rich countries. - In 2023, the R&D funds for trachoma decreased to zero. Trachoma makes 400,000 persons blind. 2. **The market for a rare disease is small.** - A rare disease occurs in less than 1 in 2,000 persons. There are thousands of rare diseases. Together, they occur in a maximum of 6% of persons. - 95% of them have no approved treatment. 3. **Drug repurposing has a time limit.** - GLP-1 drugs went from diabetes to weight loss. SGLT2 inhibitors went from diabetes to heart failure and kidney disease. - After the patent ends, generic drugs come on the market. Then no firm supplies funds for large trials of new uses. - Thus no person knows if the drug has an effect on other diseases. 4. **Firms do not give sufficient funds to basic research.** - If a firm makes its results public, other firms use them for free. If it does not, other persons cannot make a test of the results. - Thus academic labs and governments do the first research. Firms do the subsequent research for large markets. - AZT came from researchers at the National Cancer Institute. Burroughs Wellcome did the trials. - Academic researchers found the GLP-1 hormone in the 1980s. Novo Nordisk then made drugs from it. 5. **Drug development becomes less productive.** - The cost to get a drug to the market became two times larger each nine years for decades, after inflation. - Scannell gave this trend the name Eroom's law. It is the opposite of Moore's law. - One idea is that the easy discoveries are done. The author does not agree, because the tools of discovery are much better. 6. **Trials are slow and costly.** - The total time to get a drug approved increased from six years in the 1970s to eight or nine years in the 2010s. - Each trial starts from zero, with different contracts and administration. Each site must do a check of the protocol, make contracts, get ethics approval and give instruction to the staff. - About half of all trials do not get the planned number of participants. Only one trial in five is on time. The median trial is late by more than one year. - One late-stage trial of an infectious disease drug costs $54 million in the US, on average. 7. **Rules from the 1990s cause waste.** - Global guidelines from the 1990s are the source of the trial rules today. Firms use these guidelines to make sure that the regulator does not reject the trial. - Regulators removed some requirements, but firms are slow to change. - Source data verification is an example. A consultant does a manual check of each data point against paper records. This is a third of the cost of a trial. - The FDA recommends against it, because it finds only a small number of errors. - Less than 10% of the drugs in trials get approval. Thus the industry prevents financial risk and does more administrative work than is necessary. 8. **A new drug must be better than the current standard.** - The author gives this the name "better than The Beatles" problem. A new drug must be safe, and it must have an effect. It must also be better than the standard treatment. - This bar becomes higher with each generation of treatments. Many drugs have an expired patent. The generic versions of these drugs cost less. - As treatments get better, it is not easy to show a benefit. Thus trials must be larger or longer. ## Proposed solutions | Solution | Method | Result | | --- | --- | --- | | Advance market commitment | Donors promise a set price for a vaccine. They give the funds only if the vaccine is safe and has the intended effect. | In 2009, some countries and philanthropists supplied the funds for new pneumococcal vaccines with this method. The vaccines prevented the death of more than 700,000 children. | | Platform trial | Some treatments for one disease share one control group. This decreases the cost of each test. | The RECOVERY trial in the UK included more than 12 Covid drugs. It showed that dexamethasone, a generic drug, decreased deaths by one third in the sickest patients on ventilators. | | Shared trial network | Researchers get patients from many countries for large trials. | Networks for childhood leukemia started in the 1960s. They helped researchers to find quickly which treatment had an effect. | | Platform approval for gene therapy | A regulator gives approval one time for the platform that moves the therapy into the cells. Then a team changes the genetic sequence without a new approval. | Not in use today. It is an idea of the author. | | Risk-based monitoring | A team does a check of the data with the highest risk. It does not do a manual check of each data point. | The FDA recommends against the manual check of each data point. | | Large simple trials | These trials use the electronic health records that are in use. Randomization continues after approval. | A team finds the best dose, treatment time and drug combination. | | Ring vaccination trial | A team waits for a patient with Ebola. Then it gives the vaccine to the contacts of the patient. | WHO controlled the trials. The vaccine got a license in 2019, and there is a global stockpile. | > [!important] Rigor stays > The author does not want less testing. Clinical trials are the best method to find if a drug has an effect. Without trials, random changes and bias can make a drug look good when it has no benefit or causes injury. The author wants a procedure with less work and the same rigor. ## Cures that are missing - **Alzheimer's disease.** More than 30 million persons have it. We have almost no treatment for it. The brain has many parts that work together. The blood-brain barrier also stops most drugs. - **Cancer.** It is hundreds of different diseases. Different mutations cause them. The cancers change quickly and become resistant to each drug. Chemotherapy and radiation cause damage to healthy tissue. - **Pancreatic cancer.** The five-year survival rate is 13%. Mutations in the KRAS protein cause it. Scientists thought that no drug can stop this protein. In a phase 3 trial, the drug daraxonrasib made survival about two times longer than standard chemotherapy. Its side effects are bad. - **Solved or almost solved.** Hepatitis C is curable. Vaccines prevent cervical cancer and many liver cancers. CAR-T cell therapy puts some blood cancers into remission for a long time. New treatments give a near-normal life to patients with cystic fibrosis. > [!quote] Last point > The author thinks that disease is a problem and not a fact of life. Persons can find a solution. Better medicine is not automatic. It depends on how innovation works in practice. Cuts to science, global health and aid make this more urgent. ## Connections in the vault **Biology as software, and the cost curve** - [[Digitalisation of Biology]] and [[TechBio 101]] show the same trend: the cost of sequencing decreased. [[TechBio MoC]] is the map for this theme. - [[Moore's Law]], [[Wrights Law]] and [[Predicting Tech Progress]] give models of cost curves. Eroom's law is the opposite curve. The tools cost less, but drug development costs more. **Speed of change** - [[Directional Arrows of Progress]] and [[The directional arrows of inevitable progress]] show that change for the better is sure. Dattani gives the other view. Change is not sure. Institutions and incentives set its speed. - [[2025 Year-End Reflection - What Landed, What Didn't]] shows that biotech moves at FDA speed and not at software speed. Dattani gives the cause. The cause is the trial system and the incentives. It is not the science. - [[Productivity Paradox]] shows the same pattern. A new tool does not give a higher output until persons change the procedure. AI tools for discovery come, but trials stay slow. **Verification and rules** - [[AI Verification]] shows that verification is the bottleneck, not generation. A clinical trial is the verification step of medicine. Source data verification is verification cost with small value. - [[Where Domain Evals Matter Most]] shows that rules from a regulator make an optional check necessary. Drug development is an example. - [[IP Strategy for Deep Tech Startups]] gives the exclusivity periods. The end of a patent ends the incentive to find new uses for a drug. **Drugs and the science** - [[Drug Types]], [[Small Molecule Drugs]], [[Peptide Drugs]], [[Biologics]] and [[GLP-1 Drugs]] give the background on the drug classes that Dattani names. - [[Drug Discovery]], [[Computer Aided Drug Design (CADD)]] and [[Quantum x Pharma]] give the background on the tools that make discovery faster. - [[Protein Language Models]] and [[Biological Sequence Modelling]] give the background on the AI tools for proteins. - [[Tenaya Deep Dive]] is about a company that works on heart disease. Chart 1 shows that deaths decreased most in this area. - [[Biodefense Imperative]] shows that speed is the defense. The Ebola ring vaccination trial is an example of a fast trial. **Funds and bottlenecks** - [[Funding the Commons]] and [[Decentralized Science]] give other methods to fund open basic research. - [[Bottleneck Business]] and [[Healthcare]] give two more views of where value is blocked. ## Open questions - Where does AI remove a bottleneck in trials? Where does it only make discovery faster, which is not the primary bottleneck? - Which of the eight problems can a new company remove? Which problems only a government or a donor can remove? - Does Eroom's law show a limit of the science or a limit of the procedure? The rebound for rare disease drugs shows that better targets can be better than a slow procedure.