Comment by didibus | Hacker News Reader

Comment by didibus | Hacker News Reader

Parent Full thread

Why does it say:

> We found a clinically important absolute risk reduction of 5·0%, and 32% RR reduction, on the primary outcome of hospitalisation

Why is the absolute 5% and the RR 32%? What's the difference?

Also why does it say:

> There were no significant differences between fluvoxamine and placebo for viral clearance at day 7 (p=0·090) and hospitalisations due to COVID (p=0·10), all-cause hospitalisations (p=0·09), time to hospitalisation (p=0·11), number of days in hospital (p=0·06), mortality (p=0·24), time to death (p=0·49), number of days on mechanical ventilation (p=0·90), time to recovery (p=0·79) or the PROMIS Global Physical (p=0·55) or Mental Scale (p=0·32)

And they say in the Interpretation section:

> Treatment with fluvoxamine (100 mg twice daily for 10 days) among high-risk outpatients with early diagnosed COVID-19 reduced the need for hospitalisation defined as retention in a COVID-19 emergency setting or transfer to a tertiary hospital

Why wouldn't they mention anything else but the reduction in hospitalization in their Interpretation section? Is it because all the other effects were not statistically significant to count?

>Why is the absolute 5% and the RR 32%? What's the difference?

Absolute reduction is compared to control in absolute terms (16% hospitalized in control minus 11% with treatment = 5%).

Relative risk reduction is the absolute reduction divided by hospitalizations in the control (5%/16%)

Absolute % tells you how big the effect is, relative risk tells you how big it is in comparison to the reference.

>Why wouldn't they mention anything else but the reduction in hospitalization in their Interpretation section? Is it because all the other effects were not statistically significant to count?

The P values tell you the % chance that the difference could be due to random sampling.

e.g. hospitalization for COVID P = 0·1 means that the there is an 10% chance this difference is due to random sampling. The standard for reportable finding is P < 0.05, more than 95% confidence the observation is not due to random sampling.

>Why wouldn't they mention anything else but the reduction in hospitalization in their Interpretation section? Is it because all the other effects were not statistically significant to count?

The primary test the study was designed to evaluate was:

>risk of hospitalization defined as either retention in a COVID-19 emergency setting or transfer to tertiary hospital due to COVID-19 for patients allocated fluvoxamine versus placebo

This is why most of the discussion focuses on this question. Death was also statistically significant and called out in the findings, most of the other outcomes didn't pass the 95% threshold

Thanks for the explanation, I can much better understand the data now.

What about the BCI? How does the Bayesian confidence interval and the P value differ ?

Bayesian confidence interval is an estimated range of the difference if it exists, while P value is the chance that a difference exists at all.

E.g. you could do a study and find:

Men are taller than women on average P= 0.001 ( extremely likely) BCI = 5-10 inches taller

> Why is the absolute 5% and the RR 32%? What's the difference?

Example:

(20/100) 20% of patients die without treatment

(10/100) 10% of patients die with treatment

absolute = 10 percentage points difference (20->10)

relative = percent decrease. 20 -> 10 50% difference

Ah cool, that's a great explanation.

So normally people with mean age 50 and varied medical conditions that catch Covid have 16% chance of being hospitalized (based on the study placebo group). And with fluvoxamine administered early, the same people would have 11% chance of being hospitalized.

Makes sense now.

Reply on news.ycombinator.com