Closed jharenza closed 2 years ago
I'm not sure about including this panel, because the upper CI's are all NA
and sometimes both are NA
. I was reading more about how the survival
package does these calculations, it seems that it is just not possible to calculate an upper limit when the events are <50% for the given group.
But, that's not the case here. Instead, we have the opposite case - nearly all samples have an event, and it's for most of the samples. The only grouping with a bounded CI is DMG, H3 K28, TP53 loss
, and I expect this is because of it's relatively larger sample size (but 26 is still small!). Also, as I noted here, two of the subtype groups have only N=1 which means there's no variation to model there at all.
The model doesn't reach median survival of 50% because pretty much nobody survives, i.e. lack of variation in the data. I've copied the model output here and added some comments to highlight non-events:
n events median 0.95LCL 0.95UCL
molecular_subtype=DMG, H3 K28 13 13 536 355 NA
molecular_subtype=DMG, H3 K28, TP53 activated 7 7 294 18 NA
molecular_subtype=DMG, H3 K28, TP53 loss 26 26 275 244 394
molecular_subtype=HGG, H3 G35 1 1 2681 NA NA
molecular_subtype=HGG, H3 wildtype 12 7 582 375 NA # 5 NON-EVENTS
molecular_subtype=HGG, H3 wildtype, TP53 activated 3 2 1179 1139 NA # 1 NON-EVENT
molecular_subtype=HGG, H3 wildtype, TP53 loss 8 8 392 349 NA
molecular_subtype=HGG, IDH, TP53 activated 1 1 1311 NA NA
Because of the very small sample sizes here and the relative lack of variation which preclude estimating upper bounds (or both lower/upper), I do not think this figure contributes much to the paper beyond what could be added in a results sentence.
My idea was to include the plot, not any of the stats - this would be for visualization only, to go along with the forest plot discussion.
That's fine with me as long as we have a clear discussion, and caption information, contextualizing! One of the plotting issues is that we can't show CIs that are infinite, so we'll have to really contextualize those stats.
Yes of course, definitely will be clear! We don't have to plot the CIs, typically i haven't seen them plotted unless there are two groups (Even then, not usually) because it's too much for a figure.
What needs to be done?
From #1332, create panel 4H for KM curve of HGG subtypes
Who will complete this?
@sjspielman