Accessible Kaplan-Meier Survival Curves with py-maidr

Make Kaplan-Meier survival curves accessible with py-maidr: survival probability read by keyboard, censored times announced and reachable with the rotor, the confidence band and the median survival of each arm.

A Kaplan-Meier curve is the share of subjects still free of the event at each time, dropping in steps as events occur. Drawn with ax.step, it is read as a step chart: the probability at each time, and nothing else. What the figure is read for is lost – which times are censored (a subject who left the study without the event, where the curve does not step), the confidence band, and the median survival.

maidr.plot_survival() draws the curve from its numbers and reads it as a survival curve: each censored time is announced as censored and can be reached with the rotor, the band is announced after the estimate, and the description (d) gives each arm’s median survival, or says it was not reached.

SURVIVAL is experimental and may change without a deprecation period: see Plot Type Stability.

Prototype

This is one of the experimental plot types. It has not been through a user study, and it may change without a deprecation period. See Plot type stability.

Setup

import matplotlib.pyplot as plt

# Just import maidr package: plt.show() now renders accessible output 
import maidr

One arm

import matplotlib.pyplot as plt

import maidr  

months = [0, 3, 6, 9, 12, 15, 18]
survival = [1.0, 0.88, 0.71, 0.5, 0.41, 0.34, 0.29]
lower = [0.88, 0.76, 0.59, 0.38, 0.29, 0.22, 0.17]
upper = [1.0, 1.0, 0.83, 0.62, 0.53, 0.46, 0.41]

fig, ax = plt.subplots()
maidr.plot_survival(  
    months, survival, lower=lower, upper=upper, censored=[15], ax=ax  
)  
ax.set_xlabel("Months")
ax.set_ylabel("Survival probability")
ax.set_title("Overall survival")

plt.show()

Left and Right move along the times. Each stop announces the survival probability and the band around it, and month 15 is announced as censored. The curve reaches one half at month 9, which the description gives as the median survival.

Comparing arms

Each call on the same axes adds an arm to one chart, so Up and Down switch between arms at the same time. A censoring time that is not one of the times given is added as a point of its own, at the survival in force then: the Treatment arm holds at 0.79 through month 12, where a subject was censored.

import matplotlib.pyplot as plt

import maidr  

fig, ax = plt.subplots()
maidr.plot_survival(  
    [0, 3, 6, 9, 12, 15, 18],
    [1.0, 0.88, 0.71, 0.5, 0.41, 0.34, 0.29],
    censored=[15],
    label="Control",
    ax=ax,
)
maidr.plot_survival(  
    [0, 3, 6, 9, 15],
    [1.0, 0.95, 0.86, 0.79, 0.72],
    censored=[12, 18],
    label="Treatment",
    ax=ax,
)
ax.set_xlabel("Months")
ax.set_ylabel("Survival probability")
ax.set_title("Overall survival by arm")
ax.legend(title="Arm")

plt.show()

The rotor offers the censored times as a unit of its own, so a reader can jump from one to the next and hear how much of the tail is still backed by subjects. Control reaches one half at month 9; Treatment never does, and the description says its median was not reached.

From lifelines

With lifelines installed, a fitted KaplanMeierFitter is read the same way when you plot it: the estimate, its confidence band and the times the fitter recorded as censored. Nothing in maidr imports lifelines; the reading is wired up when your own code imports it.

import matplotlib.pyplot as plt
from lifelines import KaplanMeierFitter
from lifelines.datasets import load_waltons

import maidr

df = load_waltons()
fig, ax = plt.subplots()
for group, rows in df.groupby("group"):
    KaplanMeierFitter().fit(rows["T"], rows["E"], label=group).plot_survival_function(
        ax=ax, show_censors=True
    )

plt.show()