kmf.plot() last point extra line to zero

[avkitex]

Last point censored could result in extra line from last value to zero (see red line on picture at ~5000 days)

[avkitex]

I'd expect red line to end at point (4990, 0.2), not at (4990, 0)

[CamDavidsonPilon]

@avkitex This is very interesting to me. I haven't been able to reproduce the error though. Here's my test data:

        T = np.arange(0, 50, 0.5)
        E = np.random.binomial(1, 0.7, 100)
        E[-1] = 0
        kmf = KaplanMeierFitter()
        kmf.fit(T, E)

The survival curve at the last point is > 0. Are you able to send me your data for group 2, or provide a reproducible example?

[CamDavidsonPilon]

Looking at the plot too, the last point is censored, but the curve does not drop to 0.

[avkitex]

Try this:

groups = pd.Series([1, 1, 1, 2, 2, 2, 3, 3, 3, 3])
survival = pd.DataFrame({'events': [0, 1, 0, 0, 1, 1, 0, 1, 0, 0], 'duration': [10, 20, 55, 10, 30, 40, 10, 40, 60, 70]})
palette={1:'b', 2:'r', 3:'g'}

from lifelines import KaplanMeierFitter
kmf = KaplanMeierFitter()

_, ax = plt.subplots(figsize=(8, 4))

for group in groups.groupby(groups):
    kmf.fit(survival.duration[group[1].index], survival.events[group[1].index], label=str(group[0]))
    kmf.plot(ax=ax, ci_show=False, show_censors=True, color=palette[group[0]])
ax.set_ylim((0, 1))
plt.show() 

Green and blue line end censored. While red line ends with an event. This results in extra stretch down.

[CamDavidsonPilon]

Last point censored could result in extra line from last value to zero (see red line on picture at ~5000 days)

Green and blue line end censored. While red line ends with an event. This results in extra stretch down.

Is your original post mistyped? Do you mean to say last point uncensored?

[CamDavidsonPilon]

R seems to agree with lifelines: it should hit 0

t <- c(10, 30, 40)
e <- c(0,1,1)
s = survfit(Surv(t, e) ~ 1)
summary(s)

Call: survfit(formula = Surv(t, e) ~ 1, type = "kaplan-meier")

 time n.risk n.event survival std.err lower 95% CI upper 95% CI
   30      2       1      0.5   0.354        0.125            1
   40      1       1      0.0     NaN           NA           NA

[avkitex]

Is your original post mistyped? Do you mean to say last point uncensored?

Yes, I mistyped.

I just wanted to point, that event at the middle of the curve cases small height decrease while the single last one (the last point if uncensored) cases many times more height decrease.
As for myself I'd end the curve with small vertical line (same length as others on event). But it's my own opinion.
I've searched pubmed and found that almost all articles use r survfit and so similar to yours line end.

I guess the issue could be closed. Thanks for conversation!

[CamDavidsonPilon]

thanks for the follow up ❤️