Replies to post #190786 on Biotech Values

[DewDiligence]

OCRX—They confirmed that the increase in trial size restores power to original design 80%.

Moreover, the slide deck confirms that the restored 80% power is based on the observed efficacy differential at the (50-patient) interim look, extrapolated over the entire trial. This is the more conservative and realistic way to report power when a trial initially under-performs the original modeling assumptions.

(Some DSMBs report power after an interim look by plugging in the original modeling assumptions for the remainder of the trial, even if those assumptions vary widely from the observed data at the interim look; naturally, such a method can overstate the true power to a considerable degree.)

With the upsizing of the trial, I think the [Lactulose] control arm is performing better than the original assumption.

Agreed; as you know, this has always been my main concern about the trial design.

Using the company's method, my initial estimate for the IA is relative symptom improvement of ~35% at worst (vs ~50% by trial design) which would make ~1 day improvement still a possibility.

The original trial design (with 200 patients and no interim look) assumed an HR of 0.80, which equated to a 43% (2.5 days / 5.8 days) reduction in the median primary-endpoint value (5.8 days for the control arm vs 3.3 days for the OCR-002 arm). After the upsizing to 230 patents recommended by the DSMB, I estimated that the HR observed by the DSMB was ~0.85 (#msg-112399186), assuming that the DSMB was calculating power based on observed rather than modeled efficacy (as we now know they did).

To ascertain whether we’re on the same wavelength, it would be helpful for you to show your intermediate arithmetic to arrive at your 35% number and your conclusion that a 1-day median benefit “is still possible.”

Given how they sized the trial, I am wondering if they are censoring patients that improve after the 5 day primary treatment regimen…

I highly doubt they are censoring patients after 5 days, as this would destroy much of the information content of the dataset. Patients in the trial are followed for 19 days, and my assumption is that any improvement (according to the definition of the primary endpoint) during the 19-day observation period is tallied for calculating the primary endpoint.