Preventive Paxlovid Fails

Pfizer just read out a Phase 2/3 trial of paxlovid for preventing transmission of COVID-19 to other members of a household when one member is infected. Alas: nope.

Preventive paxlovid

This is a trial that Pfizer start last October, so it’s run for about 7 months now and a top-line readout has become available. They’re testing paxlovid for prevention of COVID-19 in people who are exposed because they live with someone who has tested positive. Best of all, they recruited during the Omicron wave, so any results are relevant to the current variant.

If this had worked, it would have been big news: a way to stop the spread of the virus!

Alas, and also possibly alack:

Ok, but tweets are just cheap talk, not evidence.

However, this tweet points us to a summary article at Reuters ^[1], which has some more information (but still not much). A bit of digging uncovered the same-day press release from Pfizer ^[2], which as a primary source is the best authority we’re gonna get today. Yes, it’s just a press release, but they’re still analyzing the rest of the data and the real scientific publication is yet to come. The press release points us at the actual clinical trial ^[3], which was only modestly helpful because they haven’t entered the result data yet.

Here’s Pfizer’s summary (our emphasis added):

In this trial, compared to placebo, Pfizer observed risk reductions of 32% and 37% in adults who received PAXLOVID for five and ten days, respectively, to prevent infection. These results, however, were not statistically significant and, as such, the primary endpoint of reducing the risk of confirmed and symptomatic COVID-19 infection in adults who had been exposed to the virus through a household contact was not met.

Now, observe: a 32% - 37% risk reduction is not nothing. However, the statement that it is not statistically significant means that it might be nothing, i.e., we could have seen efficacy this large by chance. So, by clinical trial standards, we embrace the null hypothesis that there is no effect.

They didn’t say exactly what statistic they used to assess statistical significance. At some level, it just tells them that the 95% confidence interval on the risk ratio with respect to controls includes 0% efficacy.

There’s not quite enough data revealed in the press release to check this for ourselves:

• The number of participants was $N = 2957$.
• They were randomized into 3 arms: placebo for 10 days, paxlovid for 10 days, and paxlovid for 5 days.
• The randomization was 1:1:1, so the number in each arm was $N_i \sim 2957 / 3 = 986$.
• The number in each arm who got sick, $K_i$, was not reported in the press release. (Media relations people generally can’t do math, so that’s understandable, if frustrating.)
• The risk ratios were 32% and 37%. Let $N_1$ be the control arm, and $N_2$ and $N_3$ be the 5 day and 10 day paxlovid courses. Then this means:
  \(\left\{ \begin{align*} \frac{K_2/N_2}{K_1/N_1} &= 0.32 \\ \frac{K_3/N_3}{K_1/N_1} &= 0.37 \end{align*} \right.\)
• Since $N_1 = N_2 = N_3$ (1:1:1 randomization), the $N$’s drop out:
  \(\left\{ \begin{align*} \frac{K_2}{K_1} &= 0.32 \\ \frac{K_3}{K_1} &= 0.37 \end{align*} \right.\)

That leaves us with 2 equations in 3 unknowns ($K_1$, $K_2$, and $K_3$), so we can’t move further in checking statistical significance. Guess we’ll have to wait for the paper. (C’mon, Pfizer folk: just 3 more small integers in your press release, is that too much to ask?)

Reuters can’t resist reporting the judgment of the stock market, for all the usual silly reasons:

Shares of Pfizer, which fell 3% in regular trading, were down another 1% at $48.53 after hours.

That’s totally meaning-free: Friday 2022-Apr-29 was a terrible day in the US stock market, with the US total stock market index closing down 3.55% as shown here. Pfizer stock didn’t fall because anybody even noticed this trial’s non-result; Pfizer stock fell because the entire market fell dramatically, and Pfizer was just along for the ride.

The Weekend Conclusion

It was a great idea, with the potential to stop the spread of COVID-19 in its tracks. It was a noble effort in testing, with a trial both large enough and properly controlled. It just turned out not to work.

So… we should proceed to the next great idea. For my money, that would be a paxlovid/fluvoxamine combination trial. (I am biased, as I used to be an expert on assessing synergy of drug combinations in oncology. Still, I think it’s a good idea. Since we know fluvoxamine works, why isn’t it more widely prescribed?)

---

Notes & References

1: D Beasley, “Pfizer says COVID treatment Paxlovid fails to prevent infection of household members”, Reuters, 2022-Apr-29. ↩

2: Pfizer Media Relations, “Pfizer Shares Top-Line Results from Phase 2/3 EPIC-PEP Study of PAXLOVID™ for Post-Exposure Prophylactic Use”, Pfizer Press Releases, 2022-Apr-29. ↩

3: Pfizer, Inc., “A Study of a Potential Oral Treatment to Prevent COVID-19 in Adults Who Are Exposed to Household Member(s) With a Confirmed Symptomatic COVID-19 Infection”, ClinicalTrials.gov id NCT05047601, retrieved 2022-Apr-30. ↩