Placental growth factor testing to assess women with suspected pre-eclampsia: a multicentre, pragmatic, stepped-wedge cluster-randomised controlled trial (PARROT)
Sunday, 23 August at 04:31 SAST
Who’s joined · 84
Read the paper
Members read the paper inside Cogneez, page by page.
Make your appraisal
AI walks this paper’s appraisal one decision at a time — allocation, blinding, power, the lot — the way an evidence-based clinician reads. Pro members appraise every week and watch themselves get sharper at telling strong evidence from weak.
Create a free account to read alongThe debrief — what the numbers really say
Why this paper matters in South Africa
A note on our numbers vs the paper's. Our figures are simple arithmetic on the raw counts (22/573 vs 24/447 — check them yourself, that is the point). The paper's headline adjusted odds ratio of 0.32 (95% CI 0.11–0.96) is smaller, because PARROT is a stepped-wedge trial: sites switched from usual care to revealed testing over time, so the analysis must adjust for site and calendar time. When a crude number and an adjusted number disagree this much, the adjustment is doing real work — asking why is exactly the appraisal skill this club builds.
The numbers, computed from the paper
Calculated from the paper’s own data — the AI never invents a number.
The paper’s own two-by-two
| Event | No event | Total | |
|---|---|---|---|
| Treatment | 22 | 551 | 573 |
| Control | 24 | 423 | 447 |
Every statistic below is calculated from these four cells — check any of them by hand.
Risk ratio
0.715
One arm's event rate divided by the other's. A value of 0.8 means the treated group had 80 events for every 100 in the control group. Exactly 1 means the treatment changed nothing.
Absolute risk reduction
1.53 in 100
The difference that actually reaches your patient: out of 100 people treated, how many fewer had the event. Small numbers here can hide behind impressive-sounding ratios, so always look at both.
Number needed to treat
66
How many patients you would treat for ONE extra person to benefit. Treating 10 to help 1 is a strong drug; treating 300 to help 1 is a policy decision.
p value (Fisher's exact)
0.287
Fisher exact test on 22/573 versus 24/447 gives p = 0.287375. That is the chance of seeing a split at least this lopsided if the treatment did nothing at all. It says nothing about how big the effect is.
If the treatment truly did nothing, this is the chance of seeing a split at least this lopsided anyway. It is NOT the chance the result is wrong — a small p with a tiny effect can still be clinically boring.
Fragility index
6
6 more event(s) the other way would have made this significant.
How many patients would have to swap outcomes before the conclusion flipped. A fragility of 1 or 2 means the finding balances on one or two people — hold it lightly.
Odds ratio
0.704
Like a risk ratio but built from odds, which behave strangely: when events are common, the odds ratio exaggerates. If it looks more dramatic than the risk ratio, that is why.
E-value
2.15
How strong a hidden bias would have to be to explain this result away with no real effect at all. A big E-value means only a whopping unmeasured confounder could fake it; a small one means be careful.
Play with the trial
The paper's own numbersRisk ratio
0.715
ARR
1.53/100
NNT
66
p value
0.31
Learn the statistics behind this
Discussion
No comments yet — the first question usually breaks the ice.
Create a free account to join the discussion.