APPLIED STATISTICS IN CLINICAL AND MEDICAL RESEARCH :

SURVIVAL ANALYSIS
NEILY ZAKIYAH, PhD., Apt
SURVIVAL ANALYSIS
u To analyze data where the outcome variable is the time until the occurrence
of an event of interest (e.g. death, occurrence of a disease).

u TIME-TO-EVENT DATA
• Time from diagnosis of cancer to death due to cancer
• Time from HIV infection to AIDS
• Time from an asthma exacerbation until the next asthma exacerbation

u The time to event/survival time can be measured in days, weeks, years, etc.
SURVIVAL ANALYSIS

u TIME-TO-EVENT DATA

Time time since diagnosis

age
unit: years, days, minutes, etc

Event mortality
re-admission
SURVIVAL ANALYSIS
Objectives
u Estimate time-to-event for a group of individuals, e.g. time until
second heart-attack for a group of myocardial infarction patients.
u To compare time-to-event between two or more groups, e.g.
treated vs. placebo myocardial infarction patients in a randomized
controlled trial.
u To assess the relationship of co-variables to time-to-event, e.g.
does weight, insulin resistance, or cholesterol influence survival
time of myocardial infarction patients?
WHY USE SURVIVAL ANALYSIS?

u Why not compare mean time-to-event between the groups using linear
regression?
ignores censoring

u Why not compare proportion of events in the groups using risk/ adds
ratios or logistic regression?
ignores time
SURVIVAL ANALYSIS:
Terms
u Time-to-event: the time from entry into a study until a subject has a
particular outcome.
u Censoring: the time until the last known information for that
particular person.
For example
1. A person does not experience the event at last follow-up survey
2. A person is lost to follow-up/withdrawn during the study period
Two-variable outcome (dependent):
u Time variable: ti = time at last disease-free observation or time at
event
u Censoring variable: ci =1 if had the event; ci =0 no event by time ti
SURVIVAL ANALYSIS
Illustration
SURVIVAL AND HAZARD FUNCTION

u The survival and hazard functions are key concepts in survival analysis
for describing the distribution of event times.
u Survival function: the probability of surviving (or not experiencing the
event) up to the given time
Notation:
– T ≡ survival time of a randomly selected individual
– t ≡ a specific point in time.
– S(t) = P(T > t) ≡ Survival Function
– λ(t) ≡ instantaneous failure rate at time t : hazard function
u Hazard function: the potential that the event will occur, per time
unit, given that an individual has survived up to the specified time.
SURVIVAL FUNCTION (KAPLAN MEIER)
u The Kaplan – Meier (KM) estimator is the most widely used for estimating survival function
• product-limit estimator
• nonparametric maximum likelihood estimator.
• When there are no censored data, the KM estimator is simple estimator
u In the case of human longevity, Ti is unlikely to follow a normal distribution, because the
probability of death is not highest in the middle ages, but at the beginning and end of life.
u The survival function estimator is defined as
u S(t) = P r(T > t) = 1 − F(t).
u The hazard function estimator:

fY(y) = the probability density function

of survival time Y,
SY = the Survivor function
Practical example 1:

Survival probability in renal replacement therapy (RRT) patients for end-stage

renal disease (ESRD) due to diabetes mellitus and other causes:

In a sample of 50 RRT patients taken from a study on diabetes mellitus, survival

time started running at the moment a patient was included in the study, in this
case at the start of RRT. Patients were followed until death or censoring. The
survival probability was calculated using the Kaplan–Meier method. Subsequently,
the survival of patients with ESRD due to diabetes mellitus was compared with the
survival of those with ESRD due to other causes.

Kitty J. Jager, Paul C. van Dijk, Carmine Zoccali, Friedo W. Dekker,

The analysis of survival data: the Kaplan–Meier method, Kidney International,Volume 74, Issue 5, 2008
Survival function Hazard function:

= 1 x 49/50 = 1 – 0.9800
= 0.9800 x 48/49 = 1 – 0.9860
= 0.9600 x 47/48 = 1 – 0.9460
Overall cumulative survival in RRT patients :
COMPARING SURVIVAL BETWEEN
GROUPS
(Log-Rank test)
u Non-parametric method, based on ranks
u No effect estimate, only test for differences
u No adjustment for confounders or interaction/ effect modification
u In application, it is a method for comparing the Kaplan-Meier curves
estimated for each group of subjects.
Cumulative survival in RRT patients (overall and by cause of ESRD)

The curves provide a means of

assessing visually whether survival
P value: 0.0425 was different for these subgroups.
When the log-rank test was used
to test formally whether the
difference was statistically
significant, it showed a P-value of
0.0425; therefore, in this small
sample of 50 patients, the
difference was statistically
significant.
COX REGRESSION
(PROPORTIONAL HAZARDS REGRESSION)
u Estimate of the magnitude of the effect on survival.
u Adjustment for confounders (as in multiple logistic and linear regression).
u Cox models the effect of predictors and covariates on the hazard rate but
leaves the baseline hazard rate unspecified.
u Estimates relative rather than absolute risk.

Hazard ratio Interpretation

HR > 1 Exposure to X à higher risk of event, X is a risk factor
HR < 1 Exposure to X à lower risk of event, X is a protective
factor
HR = 1 No effect of X on event
COX REGRESSION
The model
u Components:
• A baseline hazard function that is left unspecified but must be
positive (=the hazard when all covariates are 0)
• A linear function of a set of k fixed covariates that is exponentiated.
(=the relative risk)

log hi (t ) = log h0 (t ) + b1 xi1 + ... + b k xik

Can take on any form
b1 xi1 +...+ b k xik
hi (t ) = h0 (t )e
COX REGRESSION
The model
u The point is to compare the hazard rates of individuals who have
different covariates:
u Hence, called Proportional hazards:

h1 (t ) h0 (t )e bx1 b ( x1 - x2 )
HR = = = e
h2 (t ) h0 (t )e bx2

u Ratio of the hazard is constant over time

Example:

Karim Fizazi, Howard I Scher, Arturo Molina, Christopher J

Logothetis, Kim N Chi, Robert J Jones, John N Staffurth,
Scott North, Nicholas J Vogelzang, Fred Saad, Paul
Mainwaring, Stephen Harland, Oscar B Goodman, Cora N
Sternberg, Jin Hui Li, Thian Kheoh, Christopher M Haqq,
Johann S de Bono,
Abiraterone acetate for treatment of metastatic castration-
resistant prostate cancer: final overall survival analysis of
the COU-AA-301 randomised, double-blind, placebo-
controlled phase 3 study,
The Lancet Oncology, Volume 13, Issue 10, 2012
u Research design: Randomized controlled trial
u Interventions: Abiraterone acetate vs placebo
u Statistical methods
• Dependent variable: mortality due to prostate cancer over
the time period (1 year)
• Primary endpoint of the study: overall survival of patients
with metastatic castration-resistant prostate cancer (time
from randomisation to death from any cause).
• Secondary endpoint: sub-group analyses (prostate-specific
antigen –PSA response rate, time to PSA progression,
radiographic progression-free survival)
• Adjustment to covariates in multivariate analysis
(confounders; baseline stratification factors): ECOG score,
pain, previous chemotherapy regimens and progression
category
• Log-rank test was used to compare treatment differences
• Cox proportional hazards model was used to obtain the
estimated HR and its associated 95% CIs for both the primary
overall survival analyses and by stratification factors.
Overall survival in patients receiving Abiraterone vs placebo:

Median overall survival for the

abiraterone group was longer than in
the placebo group (15.8 months [95%
CI 14.8–17.0] vs 11.2 months [10.4–
13.1]; hazard ratio [HR] 0.74, 95% CI
0.64–0.86;
(p<0.0001).
Results of primary
and secondary
end-point of the
study:
Interpretation:

u Abiraterone acetate significantly prolongs overall survival in patients with

metastatic castration-resistant prostate cancer, compared to placebo (15.8
months vs 11.2 months).
u Hazard ratio [HR] 0.74, 95% CI 0.64–0.86; (p<0.0001)à exposure to
Abiraterone is associated with statistically significant lower risk of mortality
in patients with metastatic castration-resistant prostate cancer.
TYPE OF REGRESSION (to sum-up):

Dependent variable Aim Method

Numerical /continuous Size of the effect of X Linear regression
at Y, independent of
other variables
Categorical/ idem Logistic regression
dichotomous

Categorical/ idem Cox proportional

dichotomous + time hazards (PH) regression