SKILL.md
Lifelines - Survival Analysis
In medicine, we often care about "Time to Event" (death, recovery, relapse). Lifelines handles the complexity of "censored" data (patients who left the study).
When to Use
- Analyzing clinical trial data (time to death, disease progression).
- Comparing survival between treatment groups.
- Identifying risk factors using Cox Proportional Hazards regression.
- Building survival models for prognosis.
- Epidemiology studies (time to infection, recovery).
Core Principles
Censoring
Patients who haven't experienced the event by the end of the study are "censored". Lifelines properly accounts for this.
Hazard Ratios
In Cox regression, a hazard ratio > 1 means increased risk; < 1 means decreased risk.
Survival Curves
Kaplan-Meier estimates the probability of survival over time without assuming a distribution.
Quick Reference
Standard Imports
from lifelines import KaplanMeierFitter, CoxPHFitter
from lifelines.statistics import logrank_test
import pandas as pd
Basic Patterns
# 1. Kaplan-Meier (Visualizing survival)
kmf = KaplanMeierFitter()
kmf.fit(durations=df['days'], event_observed=df['died'])
kmf.plot_survival_function()
kmf.median_survival_time_ # Time when 50% have died
# 2. Cox Proportional Hazards (Risk factors)
cph = CoxPHFitter()
cph.fit(df, duration_col='days', event_col='died')
cph.print_summary() # See hazard ratios for age, drug type, etc.
cph.plot_partial_effects_on_outcome(covariates=['age'], values=[30, 50, 70])
Critical Rules
✅ DO
- Use event_observed correctly - 1 = event occurred, 0 = censored.
- Check proportional hazards assumption - Use
cph.check_assumptions()to validate Cox model. - Compare groups with logrank test - Statistical test for survival curve differences.