Say Nay to Northfield

Subtracting those 126 patients leaves 586 patients in a "per protocol" subset in which there were no data errors. When these patients were analyzed, the death rate in the Polyheme arm was 10.8% compared with 9.1% in the control arm. Polyheme patients still died at higher rates than control patients on a numerical basis, but statistically, the blood substitute passed its non-inferiority test, according to Northfield.

Northfield is now re-examining all protocol violations to see if any adjustment would change the results of the larger 712-patient "intent to treat" population. Regardless, Gould says the company's plan is to also include the more positive per-protocol analysis in its FDA submission because he believes it is the cleanest and clearest picture of Polyheme's efficacy.

I have a few major problems with this plan. First, the FDA almost always uses "intent to treat" analyses as the basis for review; per-protocol or subgroup analyses -- whether they are prospectively defined or not -- take a back seat.

Second, the intent-to-treat population in the Polyheme study represents the real-world scenarios in which a blood substitute would be used. EMTs responding to the scene of an accident won't always have time to go down a checklist to determine if a patient should get Polyheme.

Also, within the 126-patient protocol violator group that Northfield seeks to exclude, the difference in death rate of patients getting Polyheme (17/70 or 23%) vs. those patients in the control arm (7/56 or 12.5%) was striking. Northfield says the patients in the protocol violator group were more severely injured, and that could have contributed to the sharply higher death rate.