The Blood Test That Could Detect Thousands of Diseases | Helmy Eltoukhy, Guardant Health

Guardant Health bets a blood test can catch cancer early

Guardant Health is betting that a blood test can catch cancer early enough to matter, and eventually detect many diseases at once. The sequencing was hard but solvable. The harder problem is that in diagnostics the person choosing the test does not pay, so validation, reimbursement and physician education decide who wins, and Helmy Eltoukhy argues business model innovation may matter more than the science.

Takeaways

• Guardant raised usable cell-free DNA to 70%. Only about 5% of cell-free DNA was entering the sequencing reaction, and sequencer error ran near 1%. Helmy Eltoukhy's team pushed utilization to roughly 70%, a 10x to 14x sensitivity gain that made blood-based cancer testing practical.
• Early detection drove every step of the roadmap. The first product was a therapy-selection assay, but the founding pitch promised a blood test in the annual physical. Helmy compares the path to Tesla's Roadster to Model 3, compounding 10x improvements rather than waiting for one leap.
• Diagnostics rewards business model innovation over technology. A test needs validation, regulatory approval, reimbursement, physician education and selling, and the person choosing it does not pay. Helmy says business model innovation may matter more than the science, and rivals who priced tests 10x lower imploded.
• Pharma work started as a consulting project. Guardant assumed pharma was too small a market and ran a project to test demand. It now serves 200 pharma companies, and the business doubles as a leading indicator of where oncology is heading.
• AI disease claims are mostly overhyped. Helmy puts about 90% of the excitement around AI curing disease or discovering drugs down to undersampled biological data. Roughly 60% of a cell is housekeeping genes, so models learn little that is disease-specific.
• Epigenomics opens diseases beyond cancer. Guardant has tested 1.3m cancer patients and holds 700,000 epigenomic profiles. Conditions such as diabetes and cardiovascular disease leave the genome unchanged but alter the epigenome, and about 500 labeled diseases sit in the database.

The 2015 sales pullback

• Helmy pulled the sales team back in 2015 when test volumes grew faster than reimbursement. He says it cost Guardant at least 18 months of market share and was probably the wrong call.
• Mark Jacobstein argues the discipline was still right, because the alternative was burning capital the company could not easily replace.

What comes next

• A test costing around $1,000 that detects many conditions instead of one, heading toward $1 per diagnosis. Sequencing costs have fallen from $3bn to $1, and Guardant's tests will collect 10x to 1,000x more data each.