What really predicts biotech success

• Specialist biotech fund on the cap table

  2.0xAny exit

  95% CI 1.8 to 2.3 · p <0.001 · Single variable

• Two or more specialist funds

  2.9xTakeout of $100M+

  95% CI 2.1 to 3.9 · p <0.001 · Single variable

• Pharma investor on the cap table

  2.2xTakeout of $100M+

  95% CI 1.5 to 3.0 · p <0.001 · Single variable

• Trial stopped in the prior 3 years

  0.37xAny acquisition

  95% CI 0.16 to 0.61 · p 0.004 · Single variable

• 12+ rivals entered Phase 2 on its target in the last 3 years

  0.54xAny acquisition

  95% CI 0.24 to 0.96 · p 0.04 · Single variable

• A rival drug on the same target is already in the clinic for the same disease

  2.4xPhase 2 to Phase 3

  95% CI 1.9 to 2.9 · p <0.001 · Full model

Risk ratios against companies or programmes without the signal, adjusted for year and data depth. Single variable: the signal on its own. Full model: the signal held against our controls, such as sponsor size, the disease's base rate and human genetics. Intervals from clustered bootstraps or clustered errors. Associations, not causes. These are a few of the variables our models in development combine, dozens at a time.

Leak: the variable encodes the outcome or was recorded after it. Circular: it is built from the outcome. Model output: another model's score, not a measurement.