A Conceptual Approach to

Survival Analysis
Laura Lee Johnson, Ph.D.
Associate Director
Division of Biostatistics III
Center for Drug Evaluation and Research
US Food and Drug Administration
IPPCR Course Fall 2015

Chapter 23, 3rd Edition

 Disclaimer
• This presentation reflects the views of
the author and should not be construed
to represent FDA’s views or policies.
 Objectives
• “Conceptual Approach”
• Recognize some vocabulary used in
survival analysis and a few commonly
used statistical methods for time to
event data in medical research
• Interpret Kaplan Meier graph
• Interpret a covariate from a Cox model
• Understand a few of the assumptions
for basic types of survival models
 Preventing Mother-Infant HIV
Transmission (D.O. Dixon, NIAID)
• Zidovudine able to slow progression of
HIV in adults with advanced disease
• AIDS Clinical Trials Group Protocol 076
designed to assess both safety and
efficacy of Zidovudine in preventing
transmission of HIV from infected (not
advanced) women to their babies
 Preventing Mother-Infant HIV
Transmission
• Powered (80%) to detect a 33%
reduction of transmission rate (through
78 weeks of baby’s life) relative to
projected rate of 30%
• Target N was 748; began April 1991
• Projected accrual to take at least 5
years and 15% dropouts
 Preventing Mother-Infant HIV
Transmission
• DSMB met twice a year to monitor safety
• Efficacy reviews planned after each 1/3 of
projected infant infections
• 1st efficacy review took place in February
1994, based on mothers enrolled up to
December 1993 and their babies
 At First Interim Analysis
(1/3 of projected infant infections)

P = 0.00006
DOD/BRB/NIAID
 (Note: Survival Analysis)
• What is that graph in the last slide?
– Kaplan Meier curves
• What are they telling me?
 Define the Outcome Variable
• What is the event?
– What is the y-axis?
– Mother-Infant HIV transmission; testing
positive
• Where is the time origin?
– Infant birth?
• What is the time scale?
– Weeks
• How is time at which the event ‘occurs’
defined?
 Kaplan Meier (KM)
• One way to estimate survival
– Or in this case, transmission of HIV
• Nice, simple, can compute by hand
• Can add stratification factors
• No sensible interpretation for
competing risks
• Cannot evaluate covariates like Cox
model
• Kaplan Meier is a workhorse
At First Interim Analysis

P = 0.00006
 Preventing Mother-Infant HIV
Transmission
• DSMB recommended stopping (after careful
review of data quality and completeness,
toxicity, transmission rates)
• Trial leaders and NIH agreed
• Zidovudine provided to those in control
group
• United States Public Health Service (USPHS)
national guidelines modified
SURVIVAL OR TIME-TO-EVENT
ANALYSIS
 Why Survival (Time to Event)
Analysis
• New cancer treatment
– Want to know if it extends a person’s life 5
months longer than current treatments
• Survival is about events and when they
happen
– Death, infection, MI, hospitalization
– Recurrence of cancer after treatment
– Marriage, soccer goal
– Light bulb fails, computer crashes
– Balloon filling with air bursts
14
 Why Survival Analysis?
Hypertension
• Treat it
– Lower/Normal blood pressure (BP)
– Want to extend healthy life, prevent heart
attacks…events in time
• Follow people, see how many die and have
events like MI and when
• What if Intervention lowers blood pressure
but after 20+ years people die two years
before a similar person not on medicine?
– Goal accomplished?
– COX 2 inhibitors
15
 People with lower X live longer!
• Many times from an observational
study
• Instigate a change in blood pressure,
weight, something
– Do you get a similar change in
outcome?
– Perhaps

16
People live too long to follow!
• We can only hope (and that they live
well and feel great, are productive, and
all that)

• Many surrogates
• Even more surrogates for events (in
particular death)
• Sometimes follow people a long time

17
 What is Survival
• Survival analysis deals with making
inference about EVENT RATES
• Rate at t = Rate among those at risk at t
• Look at Median survival (50%) not Mean
survival
– Mean: need everyone to have an
event

18
 Outline
How to Measure Time and Events
Survival and Hazard Functions
• Truncation and Censoring
• Competing Risks
• Models and Hypothesis Testing
• Example
• Conclusions

19
 What is a Model?
• Basic
Y = β0 + β1 X1 +…+ βp Xp
– Y = outcome or response variable
– β = coefficient
– X = covariate, variable
• Survival
λ(t) = λ0(t) exp{ β1 X1 +…+ βp Xp}
– λ0(t) = baseline hazard
– β1,…, βp = regression coefficients
– X1,…, Xp = prognostic factors
20
 Vocabulary
• Survival vs. time-to-event
• Outcome variable = event time
• Examples of events:
– HIV positive test, AIDS defining event,
Mother-Infant HIV transmission
– Systolic blood pressure or Cholesterol
below a cut
• BMJ 17 October 2009 article on follow-up of
MIST trial: incidence of pregnancy
– Time in years to first live birth after index
miscarriage
21
 Time Notation
• t: for time axis
– t = 0 is the time origin
• T: random outcome variable
– time at which event occurs

22
 Vocabulary
• t = time
– Baseline = 0 months
– 6, 12, 18, 24 months, etc.
• S(t) = Survival at time t
• P[ T ≥ t ] = Probability Time of event is
greater than time t

23
 Define the Outcome Variable
• What is the event? (Death?)
• Where is the time origin? (Diagnosis?)
• What is the time scale? (Weeks? Years?)

• Then what?
– Need to know the time the event occurs
– May or may not use covariates
• Could do a logistic regression model
– Yes/No outcome
– Not focus of lecture
24
 Choice of Time Scale
Scale Origin Comment
Study time Dx or Rx Clinical Trials
Study time First Exposure (Occupational)
Epidemiology
Age Birth (subject) Epidemiology

25
 Treatment for a Cancer
• Event = death
• Time origin = date of surgery
• Time scale = time (months)
• T = time from surgical treatment to
death
• Graph = P[ T ≥ t ] vs t

26
 Proportion Alive
0.0 0.2 0.4 0.6 0.8 1.0

3 4 0
1
2 5 6
Months since surgery
7
8
9
27
 Example Numbers
• S(9) = P[ T ≥ 9 ] = 0.25
• 25% is the probability the time from
surgical treatment to death is greater
than 9 months
• “9 month post-resection survival is
25%”
• 0 ≤ S(t) ≤ 1

28
 Survival Function
• S(t) = P[ T ≥ t ] = 1 – P[ T < t ]
• Plot: Y axis = % alive, X axis = time
• Proportion of population still without
the event by time t
 Survival Function in English
• Event = death, scale = months since Rx
• “S(t) = 0.3 at t = 60”
• “The 5 year survival probability is 30%”
• “70% of patients die within the first 5
years”

• Everyone dies → S(∞) = 0

 Hazard Function
• Incidence rate, instantaneous risk, force
of mortality
• λ(t) or h(t)
• Event rate at t among those at risk for an
event
• Key function
• Estimated in a straightforward way
– Censored
– Truncated
 Hazard Function in Words
• Event = death, scale = months since Rx
• “λ(t) = 1% at t = 12 months”
• “At 1 year, patients are dying at a rate
of 1% per month”
• “At 1 year the chance of dying in the
following month is 1%”
 Hazard Function: Instantaneous
• 120,000 die in 1 year
• 10,000 die in 1 month
• 2,500 die in a week
• 357 die in a day
• Instantaneous: move one increment in
time
 Herpes Example
• Recurrence of Herpes Lesions After
Treatment for a Primary Episode
• Event = recurrence
– needs well defined criteria
• Time origin = end of primary episode
• Time scale = months from end of primary
episode
• T = time from end of primary episode to first
recurrence

34
Toxin Effect on Lung Cancer Risk
• Occupational exposure at nickel refinery
• Event = death from lung cancer
• Origin = first exposure
– Employment at refinery
• Scale = years since first exposure
• T = time: first employed to death from LC

35
 Population Mortality
• Event = death
• Time origin = date of birth
• Time scale = age (years)
• T = age at death

36
 Volume of Air a Balloon Can
Tolerate
• Event = balloon bursts
• t = ml of air infused
• Origin = 0 ml of air in the balloon
• T = ml of air in balloon when it bursts

37
 Unique Features of Survival
Analysis
• Event involved
• Progression on a dimension (usually
time) until the event happens
• Length of progression may vary among
subjects
• Event might not happen for some
subjects

38
 Sample Size Considerations
• Event may not ever happen for some
subjects
– Sample sizes based on number of
events
– Work backwards to figure out # of
subjects
• Covariates must be considered (age,
total exposure, etc)

39
 Outline
How to Measure Time and Events
Survival and Hazard Functions
Truncation and Censoring
• Competing Risks
• Models and Hypothesis Testing
• Example
• Conclusions

40
 Truncation and Censoring
• Truncation is about entering the study
– Right: Only sample those with Event of
interest (cancer registry) (underestimate)
– Left: short survival may be overlooked
(>65 years of age) (overestimate)
• Censoring is about leaving the study
– Right: Incomplete follow-up (common)
– Left: Observed time > survival time (know
the subject exists)
• Independence is key
41
 Left Truncation
• Mention more in epi vs medical studies
– Medical: zero-out at time of diagnosis or
treatment
• Key Assumption
– Those who enter the study at time t are a
random sample of those in the population
still at risk at t
– Allows one to estimate the hazard function
λ(t) in a valid way
• Who is the audience and will they generalize
42
 Censoring
• Incomplete observations
• Right
– Incomplete follow-up
– Common and Easy to deal with
• Left
– Event has occurred before T0, but
exact time is unknown
– Not easy to deal with

43
 Left Censoring
• Age smoking starts
– Data from interviews of 12 year olds
– 12 year old reports regular smoking
– Does not remember when he started
smoking regularly
• Study of incidence of CMV infection in
children
– Two subjects already infected at
enrollment

44
 One Form of Right Censoring:
Withdrawals
• Must be unrelated to the subsequent
risk of event for ‘independent
censoring’ to hold
• Accidental death is usually ok
• Moves out of area (moribund unlikely to
move)

45
 Right Censoring
10
8 6
Patient
4 2
0

0 5 10 15 20
Age (months)
46
 Types of Censoring
• Type I censoring
– T* same for all subjects
– Everyone followed for 1 year
• Type II censoring
– Stop observation when a set number of
events have occurred
– Replace all light bulbs when 4 have failed
• Random censorship
– Our focus, more general than Type I

47
 Notation
• T = event time
• T* = observation time
– T if event occurs
– Follow-up time otherwise
• ∂ = failure indicator
– 1 if T* = T
– 0 if T* < T
– “censor” or “censor indicator”

48
 Key Assumption:
Independent Censoring
• Those still at risk at time t in the study
are a random sample of the population
at risk at time t, for all t

• This assumption means that the hazard

function (λ(t)) can be estimated in a
fair/unbiased/valid way

49
 Independent Censoring:
If you have Covariates
• Censoring must be independent within group
– Censoring must be ‘independent’ given X
– Censoring can depend on X
• Among those with the same values of X,
censored subjects must be at similar risk of
subsequent events as subjects with
continued follow-up
• Censoring can be different across groups

50
 Censoring Examples
• At five year follow-up patients have not died
• Follow school group from age 5 until 25
– Ask students when they start smoking
– Answering “never” is an example of right
censoring
• Early in trial older subjects are not enrolled
– Amount of time could be on study?
– Condition on age: ok
– Do not condition on age: the estimates will
be biased because censoring is not
independent
51
 Take Away: Study Types
• Clinical studies
– Time origin = enrollment, treatment begins
– Time axis = time on study
– Right censoring common
• Epidemiological studies
– Time axis = age
– Right censoring common
– Left truncation common

52
 Bottom Line
• Standard methods to deal with right
censoring and left truncation
• Key assumption is that those at risk at t
are a random sample from the
population of interest at risk at t

53
 Survival Analysis
• Models mostly for the hazard function
• Accommodates incomplete observation
of T
• Censoring
– Observation of T is ‘right censored’ if
we observed only that T > last follow-
up time for a subject

54
 Typical Intervention Trial
• Accrual into the study over 2 years
• Data analysis at year 3
• Reasons for exiting a study
– Died
– Alive at study end
– Withdrawal for non-study related
reasons (LTFU)
– Other reasons, including dying from
other causes (which leads us to…)
55
 Outline
How to Measure Time and Events
Survival and Hazard Functions
Truncation and Censoring
Competing Risks
• Models and Hypothesis Testing
• Example
• Conclusions

56
 Competing Risks
• Multiple causes of death/failure
• Special considerations of competing
risk events described in the literature
• Example:
– event = cancer diagnosis/recurrence
– death from myocardial infarction (MI)
= competing risk
• No basis for believing the
independence assumption
57
 Competing Risks
• Interpretation of λ(t) = “risk of cancer at
t when the risk of death from MI does
not exist” isn’t practically meaningful
• Rather, interpret λ(t) = “risk of cancer
among those at risk of cancer at t”
– This will exclude MI deaths (if you are
dead from an MI you are not at risk of
cancer) and that is ok

58
 Polar Bear Plunge Club Death
Rates (fiction)
• Annual death rates
– 3% taking dip 1Jan in Lake Michigan
– 2% Males all other causes
– 1% Female all other causes
• Over a decade
– 25% of women died from taking a dip
in Lake Michigan 1 Jan
– 24% of men died from taking a dip in
Lake Michigan 1 Jan
59
 Polar Bear Club Death Rates
(fiction)
• Why does it harm women?
• Over a decade
– 33.5% of women died from all other
causes
– 40% of men died from all other
causes
• There are more women to harm
• People die of something
– Which means they cannot die from
something else
60
 Bottom Line
• We make inference about λobs(t) = event
rate among subjects under observation
at t
• We can interpret it as λ(t) = event rate
among subjects with T ≥ t, if censoring
is independent

61
 Outline
How to Measure Time and Events
Survival and Hazard Functions
Truncation and Censoring
Competing Risks
Models and Hypothesis Testing
• Example
• Conclusions

62
 Kaplan Meier
• One way to estimate survival
– Or other events (transmission of HIV)
• Nice, simple, can compute by hand
• Can add stratification factors
• No sensible interpretation for
competing risks
• Cannot evaluate covariates like Cox
model
• Kaplan Meier is a workhorse
63
 Kaplan Meier
• Multiply together a series of conditional
probabilities
Time ti # at risk # events Ŝ
0 20 0 1.00
5 20 2 [1-(2/20)]*1.00=0.90
6 18 0 [1-(0/18)]*0.90=0.90
10 15 1 [1-(1/15)]*0.90=0.84
13 14 2 (1-(2/14)]*0.84=0.72
64
 Proportion Surviving (95% Confidence)
0.6 0.7 0.8 0.9 1.0

10 0
5
Survival Time
15
Kaplan Meier Curve

20
65
 Kaplan Meier Estimator
• One estimate of S(t)
– Use the same idea when looking at 1- S(t)
• Need independent censoring
– If high risk subjects enter the study late
then early on the K-M curve will come
down faster than it should
• Censored observations provide information
about risk of death while on study
• What are those little hash marks on the
curves? Mark times people were censored
66
 Kaplan Meier
• Just the outcome is in many models
• One or more stratification variables
may be added
– Intervention
– Gender
– Age categories
• Quick and Dirty

67
How to Test? At a Given Time
• H0: S1(t) = S2(t)
• Form test statistic

Sˆ1 (t )  Sˆ2 (t )
Z
 2
Sˆ1 ( t )
 2
Sˆ2 ( t )

• “Arbitrary time” – choosing t post hoc

• Not using all of the data

68
 Simple Inference
• For single event data inference about
rates → inference for S(t)
– No time dependent covariates, no
recurrent events, no competing risk
events
• Logrank statistics compare event rates
and allow the same generality as right
censoring, left truncation

69
 Log Rank
• H0: S1(.) = S2(.)
• Test overall survival
• 2 independent samples from the
same population
• Observed # events vs. Expected #
• Software; statistician should check
• Some variations and some
assumptions

70
 Log Rank
• Confounding
• Are prognostic factors balanced
between treatment groups?
• Can see a difference using logrank, but
just bias

71
 Stratified Log Rank
• Compare survival within each stratum
• Essentially perform test within each
stratum
• Can prognostic factor be categorized?
• Enough people per stratum?
• Loss of power
• Significance test, no estimates of
difference

72
 At First Interim Analysis
(1/3 of projected infant infections)

P = 0.00006
DOD/BRB/NIAID 73
Cox Models
 Proportional Hazards: Cox
• Cox Proportional Hazards model
λ(t) = λ0(t) exp{ β1 X1 +…+ βp Xp}
• λ0(t) = baseline hazard
• β1,…, βp = regression coefficients
• X1,…, Xp = prognostic factors
• β = 0 → hazard ratio = 1
– Two groups have the same survival
experience
• eβ = relative rate (relative risk) (RR)

75
Cox Proportional Hazards Model
• Add covariates to the model
• No need to stratify
• Change in a prognostic factor →
proportional change in the hazard (on
the log scale)
• Statistical software
• Can test the effect of the prognostic
factor as in linear regression - H0: β=0

76
 Cox Model for Event Rates
• Provides a framework for making
inference about covariate effects
• Semi-parametric
– λ0(t) completely unspecified
• Multiplicative - eβx
– Effect of covariate is to multiply the
rate by a factor

77
 Cox cont.
• Requires either that
– RR is constant over time
(proportional hazards), or
– That we model RR over time
• Allows time-dependent covariates and
stratification factors

78
 Age Example
• Early in trial older subjects are not
enrolled
• If age is not in the Kaplan Meier then
the KM estimate is biased because
censoring is not independent
• Put age in the Cox model – conditioned
on age; ok

79
 Age Example (cont.)
• If I follow everyone for 1 year, am I ok?
• Not necessarily
– The study is not proportional by age
to the population risk set
– Could try to over sample older people
later in the study to make the final
study more correctly proportional
• Easier to condition on age?

80
Testing Proportional Hazards
• λ(t) = λ0(t) exp{ β1 age + β2 drug}
• exp{ β1age+β2drug+β3age*ln(t)+β4 drug *ln(t)}
• Look at p-values associated with β3 and β4
(Wald tests)
• Do a partial likelihood ratio test comparing
the two models
• Look at Schoenfeld residual plots

81
 Testing Proportional Hazards
Variable Coef SE P-value 95%CI
Drug 0.58 0.25 0.020 (0.09, 1.1)
Age 0.18 0.03 <0.001 (0.12, 0.25)

Drug 0.57 0.25 0.023 (0.08, 1.1)

Age 0.19 0.03 <0.001 (0.12, 0.26)
Drug*ln(t) 0.002 0.16 0.988 (-0.32, 0.31)
Age*ln(t) 0.007 0.02 0.716 (-0.03, 0.05)
Partial likelihood ratio test, 2 degrees of freedom
82
(for adding two interaction terms): p-value=0.94
Testing Proportional Hazards
Variable Coef SE P-value
Drug 4.24 0.61 <0.001
Age 0.17 0.03 <0.001

Drug 8.98 1.88 <0.001

Age 0.19 0.03 <0.001
Drug*ln(t) 2.71 0.84 0.001
Age*ln(t) 0.01 0.02 0.60
Partial likelihood ratio test, 2 degrees of freedom
(for adding two interaction terms): p-value=0.003 83
Time-Dependent Survival Curves
• Failure to account for change in
exposure/treatment over time
– Usually assume there is no change
– Think about HAART example
• Stanford Heart Transplant Study (1971)
– End-stage heart disease
– Not responding
– Seeking transplant

84
 Take Home
• Choose the right method and test
• Kaplan Meier – simple
– Logrank tests – useful, potentially
misleading
• Cox Proportional Hazards – workhorse
– Not everything is proportional –
check
– Consider more general Cox models
• Time matters
• Changes in protocol matter
 Outline
How to Measure Time and Events
Survival and Hazard Functions
Truncation and Censoring
Competing Risks
Models and Hypothesis Testing
Example
• Conclusions

86
 Example
• Series of prospective cohort and
randomized clinical studies evaluating
survival of patients with liver cirrhosis
– Compare a ‘new’ treatment, D-
penicillamine with placebo
– Conflicting reports
• One study appears to report a two year
survival probability of 0.88, calculated
with Kaplan Meier
Lancet 1981, Gut 1985, NEJM 1985, Cochrane review 2009 87
 Trial Information
• Data collected at randomization
– Lots of information (stage)
– Presence/absence of ascites
– Prothrombin time in seconds -10
• Cox model you might see
• λ(t) = λ0(t) exp{ -0.135 XTRT+1.737 XA+0.346 XP}

88
 What is this model?
• λ(t) = λ0(t) exp{ -0.135 XTRT+1.737 XA+0.346 XP}
• XTRT: 1 = D-penicillamine, 0 = placebo
• XA: 1 = ascites, 0 = no ascites
• XP: Prothrombin time – 10
– Continuous, in seconds
• λ0(t) is the event rate at time t in the placebo
arm for subjects without ascites with a
prothrombin time of 10 seconds

89
λ(t) = λ0(t) exp{ -0.135 XTRT+1.737 XA+0.346 XP}

• Relative rate of death two years post

randomization for a subject on this trial
who received the new treatment, had
ascites at randomization and a
prothrombin time of 10 seconds
compared to a similar subject who
received placebo?
• RR = exp { -0.135 } = 0.87

90
 Worked Out
 (t )  0 (t ) exp{-0.135 X TRT +1.737 X A +0.346 X P }

 person1 (t ) 0 (t ) exp{-0.135 *1+1.737*1+0.346 *0}

 
 person 2 (t ) 0 (t ) exp{-0.135 *0+1.737 *1+0.346 *0}

e 0.135 * e1.737 * e0
0 1.737 0

e *e *e

exp{0.135}  0.87 is the relative rate of death for

subjects who received treatment compared to those
who received placebo
91
 RR at Three Years?
• Relative rate does not vary with time
according to the proportional hazards
model.
• At the years the previously described
RR is also exp { -0.135 }
• Can work out RR for lots of other
subject comparisons

92
 But…
• Physicians were initially reluctant to
enter patients with ascites on the trial
because of potential toxicity concerns
• After about a year and a half
recruitment became more
representative of the clinic population

93
 Effects on the Validity of the
Kaplan Meier (KM) Estimator
• Censoring is not independent
• At large t, the risk sets will not include
patients with ascites because they were not
recruited early enough and therefore are
censored early.
• The hazard function will be biased too small
for larger t and so Ŝ will be larger than the
population survival function at large t.

94
 In Short, What If
• From first participant entered until the
end of study: 4 years
• Enroll for 3 years
– Can be on study at least 1 year and
up to 4 years
• Followed enrollment to end of study
• Do not start fully enrolling ascites until
year 1.5

95
 Ascites Participants
• On study at least 1 yr and up to 2.5 yr
• Do not have full population/risk set
information at time t > 2.5 years
• At time points t > 2.5 the study does not
include a representative population
– Ascites → worse prognosis
– KM estimate at t > 2.5 too high
– Hazard is too small at larger t

96
Cox Model: Doomed Regression
Coefficient Estimates?
• No bias because conditional on
covariates (including XA)
• Censoring must be independent GIVEN
X
• Censoring is independent and that is all
that is required for consistency of the
partial likelihood estimator (i.e. the
coefficients)

97
 Outline
How to Measure Time and Events
Survival and Hazard Functions
Truncation and Censoring
Competing Risks
Models and Hypothesis Testing
Example
Conclusions

98
 Survival Analysis
• Survival analysis deals with making
inference about EVENT RATES
• Rate at t = Rate among those at risk at t
• Look at Median survival (50%) not Mean
survival
– Mean: need everyone to have an event
• Cox Regression is the most robust method
• Kaplan Meier curves do not have sensible
interpretations for competing risks

99
Survival Analysis Can Handle
• Right censoring
• Left truncation
• Recurrent events
• Competing risks, etc.

• Available representative risk sets at t

allow us to estimate/model event rates
 Kaplan Meier
• One way to estimate survival
• Nice, simple, can compute by hand
• Can add stratification factors
• Cannot evaluate covariates like Cox
model
• No sensible interpretation for
competing risks

101
 Inference: Log Rank
• Logrank statistics compare event rates
and allow the same generality as right
censoring, left truncation
• For single event data inference about
rates → inference for S(t)
– No time dependent covariates, no
recurrent events, no competing risk
events

102
 Cox Model for Event Rates
• Provides a framework for making
inference about covariate effects
• Semi-parametric
– λ0(t) completely unspecified
• Multiplicative - eβx
– Effect of covariate is to multiply the
rate by a factor

103
 Cox cont.
• Requires either that
– RR is constant over time
(proportional hazards), or
– That we model RR over time
• Allows time-dependent covariates and
stratification factors

104
 Truncation and Censoring
• Independence is key
• Truncation is about entering the study
– Right: Event has occurred (e.g. cancer
registry)
– Left: Have the event and fall out of view
before they can enter to be counted
• Censoring is about leaving the study
– Right: Incomplete follow-up (common)
– Left: Observed time > survival time

105
 Analysis Follows Design
Questions → Hypotheses →
Experimental Design → Samples →
Data → Analyses →Conclusions
• Take all of your design information to a
statistician early and often
– Guidance
– Assumptions

106
 Questions?
• Thanks!
• Please fill out the course evaluations
• Post specific examples
• Send questions to the appropriate
board for each lecture

107