Baby KJ, and the million-dollar problem

Part 3 — Baby KJ, and the million-dollar problem In February 2025, a seven-month-old named KJ Muldoon became the first person to receive a gene-editing therapy built for his mutation alone. He was born with severe CPS1 deficiency, a urea-cycle disorder affecting about 1 in 300,000 births, in which ammonia builds up to levels that damage the brain and liver. A team at the Children's Hospital of Philadelphia and Penn Medicine identified his specific variant shortly after birth, then designed, tested, manufactured and got regulatory clearance for a base-editing therapy delivered to his liver in lipid nanoparticles — in roughly six months. He received three infusions. His ammonia levels dropped and his protein tolerance improved. The remarkable part isn't that one child was saved, extraordinary as that is. It's that the therapy was largely the same drug each time, with one component swapped out — the guide that tells the editor where in the genome to go. That is the definition of a platform. It means the second patient shouldn't cost what the first one did. It doesn't yet Here is what approved one-time genetic therapies cost today: Therapy Condition List price Lenmeldy Metachromatic leukodystrophy $4.25M Hemgenix Hemophilia B $3.5M Elevidys Duchenne muscular dystrophy $3.2M Lyfgenia Sickle cell disease $3.1M Zolgensma Spinal muscular atrophy $2.125M These prices are not simply the cost of the science. Independent value assessments by ICER have repeatedly landed below list — $2.6–2.9M against Hemgenix's $3.5M, and $0.9–1.5M against Zolgensma's $2.125M. The gap comes from a structural problem, not a scientific one. A developer has to recover a decade of investment from a few hundred patients, using bespoke manufacturing, with no reusable asset left over afterwards. What a platform changes A manufactured therapy has a marginal cost per patient that doesn't fall. A model does. That's the bet behind what we're building at SaJa Bio: a diagnostic and predictive layer that starts from a single drop of capillary blood, can be collected close to where patients actually live, and delivers risk estimates, molecular subtype and evidence-linked options — organized through a knowledge graph and digital twin — to the clinicians making the call. Once the fixed investment sits in the graph, the twin and the trained predictors, each additional person costs an assay and an inference. It's the only architecture I know of where the thousandth patient is meaningfully cheaper than the first. Why it matters My father's brother died of an infection in his abdomen because surgery cost more than the family had. Seventy years later that surgery is trivial, and a child like KJ can have a gene editor designed for him in six months — at a price that would have been just as far out of my grandparents' reach as the operation was. That's the work. Not to make medicine more extraordinary, but to make the extraordinary ordinary, and then to make it cheap. Nobody should have to sit beside someone they love and do arithmetic.