Objectives and Endpoints

1
Objectives and Endpoints

• ? ? ?
• ? ? ? ?
• ???????????

2
Introduction

• The objectives of a trial are the research
  questions phrased in concise quantitative terms.
  An example is to determine which therapy being
  investigated has superior efficacy or fewer side
  effects. Achieving objectives does not depend on
  the outcome of the trial, only on obtaining a
  valid result.
• Endpoints are the quantitative measurements
  implied or required by the objectives. An
  endpoint is determined in each study subject,
  whereas the objectives are met by the aggregate
  of endpoints. The best endpoints to use depend
  on specific clinical objectives stated in
  quantitative terms.

3
Example A clinical objective is to determine if
a new surgical procedure reduces peri-operative
morbidity compared with standard method.

• The measurement of operative morbidity can be
  partly subjective, which may be an issue if more
  than one surgeon participates in the study. At
  least three aspects of morbidity might need to be
  defined
• 1. A window of time, during which adverse
  events could be attributed plausibly to the
  operative procedure.
• 2. A list of diagnoses or complications to be
  included.
• 3. Specification of procedures or tests
  required to establish each diagnosis
  definitively.
• Using these three criteria, a morbid event can
  be defined without much subjective interpretation
  and each patient can be classified as having or
  not having the endpoint.

4

• A clinical objective may have more than one way
  of being quantified or may be described by more
  than one endpoint. For example, improve
  survival might mean prolonged median survival,
  higher 5-year survival, or a lower death rate in
  the first year. These three definitions may
  require different methods of assessment and need
  not yield the same size or analysis plan.
• Knowing which endpoint and method of
  quantification to use in particular clinical
  circumstances is an essential task for the
  statistician.

5

• Trials typically have a single primary objective
  or endpoint with additional secondary ones. It
  may not be feasible, or there may not be
  sufficient resources, to answer more than one
  primary question reliably. For example, we can
  only actively design control over the type II
  error for one objective. Secondary objectives
  have a statistical power that is determined
  passively by the sample size for the primary
  objective. Trials with many objectives require a
  multiplicity of statistical analyses, some of
  which may be based on subsets of the study
  cohort. This increase the possibility of error.

6

• Prefer hard endpoints
• Hard endpoint are clinical landmarks that
  are well defined in the study protocol,
  definitive with respect to the disease process,
  and require no subjectivity. Examples include
  death, disease relapse or progression, and many
  laboratory measurements. Soft endpoints are
  those that do not relate strongly to the disease
  process or require subjective assessments by the
  investigator or patient. Trying not to abuse
  terminology, we might say that hard endpoints are
  objectively measured, whereas soft endpoints are
  subjective.

7

• An example of why subjectivity is undesirable in
  clinical trial endpoints is the so-called
  Hawthorne effect, named after experiments at the
  Hawthorne plant of the General Electric Company
  in the 1920s and 1930s. These studies tested the
  effects of working conditions on productivity and
  demonstrated that even adverse changes could
  improve productivity. The research subjects were
  affected by the knowledge that they were being
  tested, illustrating that study participants can
  respond in unexpected ways to support the
  research hypothesis if they are aware of it.
  While such effects may be more likely when
  subjective endpoints are used, they can also
  influence hard outcomes that depend on changes in
  behavior.

8

• Some useful and reliable endpoints fall between
  the extremes of hard and soft. An example is
  pathologic classification, which is usually based
  on expert, experienced, and objective, judgment.
  Such endpoints are likely to be useful in
  clinical trials and prognostic factor studies,
  because they are valid and reliable. The
  underlying issue with endpoints is not their
  degree of hardness or subjectivity, but rather
  how error prone they are. The best endpoints not
  prone to error and are repeatable. Even so, a
  good endpoint, such as vital status, can be made
  unreliable if investigators use poor methods of
  ascertainment.

9
Objectives

• Estimation is the most common objective
• Most clinical objectives translate into a
  need to estimate an important quantity. In Phase
  I, the primary purpose is usually to estimate the
  MTD. In phase II, the objective of SE studies is
  usually to estimate response and toxicity
  probabilities using a fixed dose of drug or a
  specific treatment. In phase III, trials
  typically estimate treatment differences. Phase
  IV trials estimate rates of complications or side
  effects.
• Selection can also be an objective
• In some circumstances, trials are intended
  primarily to select a treatment that satisfies a
  set of important criteria, as opposed to simply
  estimating the magnitude of an effect. For
  example, the treatment with highest response rate
  could be selected from among several alternatives
  in a multi-armed randomized SE trial. (by using
  the logistic regression analysis).

10

• Objectives require various scales of measurement
• Clinical trial objectives require
  measurements that fall into one of four numeric
  categories
• 1. The 1st type of objective is
  classification, or to determining into which of
• several categories an outcome fits. In
  these cases, the numerical character of
• the endpoint is nominal, i.e., it is
  used for convenience only.
• 2. The 2nd type of objective requires
  ordering of outcomes that are measured
• on a degree, severity, or ordinal scale,
  e.g., severity of side effects or toxicity.
• 3. The 3rd type of objective is to estimate
  differences and lead to interval scales
• of measurement. On interval scales,
  ordering and differences between values
• is meaningful.
• 4. The 4th type of objective is to estimate
  ratios. These lead to ratio scales on
• which sums, differences, ratios, and
  products are meaningful.

11
Endpoints

• Endpoints can be quantitative or qualitative
• The most important beneficial
  characteristics of the endpoint used in a study
  are it must correspond to the scientific
  objective of the trial and the method of
  endpoint assessment must be accurate and free of
  bias. These are important, not only for
  subjective endpoints like functional status or
  symptom severity, but also for more objective
  measures such as survival and recurrence times.
• Measures are useful and efficient endpoints
• Measurements that can theoretically vary
  continuously over some range are common and
  useful types of assessments in clinical trials.
  E.g. laboratory values, blood or tissue levels,
  functional disability scores, or physical
  dimensions. In a study population, these
  measurements have a distribution, often
  characterized by a mean or other location
  parameter, and variance or other dispersion
  parameter. Consequently, these outcomes will be
  most useful when the primary effect of a
  treatment is to raise or lower the average
  measure in a population. Typical statistical
  tests that can detect differences such as these
  include t-test, ANOVA, regression,etc.

12

• Some outcomes are summarized as counts
• Count data also arise frequently in
  clinical trials. Count data are most common when
  the unit of observation is a geographic area or
  an interval of time.
• Ordered categories are commonly used for severity
  or toxicity (mild, moderate, or severe)
• Unordered categories are sometimes used
• Dichotomies are simple summaries
• Some assessments have only two possible
  values, for example, present or absent. In the
  study population, these outcomes can often be
  summarized as a proportion of successes or
  failures. Comparing proportions leads to tests
  such as the chi-square or exact conditional
  tests. Another useful population summary for
  proportions is the odds, log-odds, or odds ratio
  for the outcome.
• Event times may be censored
• Measurements of the time interval from
  treatment, diagnosis, or other baseline landmarks
  to important clinical events such as death (event
  times) are common and useful outcomes in chronic
  disease clinical trials. The potential problem is
  the possibility of censoring.

13

• Event time data require two numerical values
• Composite outcomes instead of censoring
• When two or more failure processes affect a
  population (competing risks), investigators
  cannot expect to obtain unbiased estimates of
  risk for one cause by censoring all other events.
  If failure processes are not independent of one
  another, as is usually the case, events
  attributable to one cause contain some
  information about events of other types. For
  example, a myocardial infarction listed as the
  primary cause of death may be secondary to an
  advanced state of the underlying disease.
• Waiting for good events complicates censoring
• In most event time studies, the interval of
  interest is measured from a hopeful clinical
  landmark, such as treatment, to a bad event, such
  as disease progression or death. Our perspective
  on censoring is that events such as disease
  progression or death could be seen with
  additional follow-up. In a few studies, however,
  investigators measure the waiting time to a good
  event. Short event times are better than long
  ones and censoring could be a more difficult
  problem than it is in survival studies.

14
Surrogate Endpoints

• A surrogate endpoint is one that is measured in
  place of the biologically definitive or
  clinically most meaningful endpoint. A good
  surrogate endpoint needs to be convincingly
  associated with a definitive clinical outcome so
  that it can be used as a reliable replacement.
  Investigator choose a surrogate when the
  definitive endpoint is inaccessible due to cost,
  time, or difficulty of measurement.
• Prentice(19xx) offered a rigorous definition of a
  surrogate endpoint as a response variable for
  which a test of null hypothesis of no
  relationship to the treatment groups under
  comparison is also a valid test of the
  corresponding null hypothesis based on the true
  endpoint.
• Surrogate endpoints are sometimes called
  surrogate markers, intermediate, or replacement
  endpoints.

15
Surrogate endpoints are disease-specific

• Some important characteristics of surrogate
  endpoints can be listed as follows
• 1. A good surrogate can be measured
  relatively simply and without invasive
• procedures.
• 2. A surrogate that is strongly associated
  with a definitive outcome will likely by
• part of, or close to, the causal pathway
  for the true endpoint. For example,
• cholesterol level, it fits into the
  model of disease progression
• High cholesterol ? Atherosclerosis ?
  Myocardial Infarction ? Death
• This is in contrast to a surrogate like
  prostatic specific antigen (PSA), which is
• a reliable marker of tumor burden but is
  not in the chain of causation (indirect
• surrogate).
• 3. We would expect a good surrogate endpoint
  to yield the same inference as the
• definitive endpoint.
• 4. We would like the surrogate to have a
  short latency with respect to natural
• history of the disease.
• 5. A good surrogate should be responsive to
  the effects of treatment.

16
Examples of Surrogate Endpoints used in Clinical
Trials

• Disease
• HIV Infection
• Cancer
• Colon Cancer
• Prostate Cancer
• Cardiovascular
• Disease
• Glaucoma

• Definitive Surrogate
• Endpoint Endpoint
• AIDS (or Death) CD4 Count
• Mortality Tumor Size
  Reduction
• Disease Progression CEA Level
• Disease Progression PSA Level
• Hemorrhagic Stroke Blood Pressure
• Myocardial Infarction Cholesterol Level
• Vision Loss Intraocular pressure

17

• Surrogate endpoints can make trials more
  efficient
• Most clinical trials require an extended
  period of accrual and observation for each
  patient after treatment. This is especially true
  of comparative studies with event time as a
  primary outcome. Good surrogate endpoints can
  shorten such clinical trials, which explains why
  they are of particular interest in prevention
  trials. For example, suppose we wish to test the
  benefit of a new anti-hypertensive agent against
  standard therapy in a randomized trial. Survival
  is a definitive endpoint and blood pressure is a
  surrogate.
• Surrogate endpoints have significant limitations
• Surrogate endpoints have serious
  limitations, e.g., sources of difficulty include
  the validity of the surrogate, coping with
  missing data, having the eligibility criteria
  depend on the surrogate measurement, and the fact
  that trials using these endpoints may be too
  small to reliably inform us about uncommon but
  important events. Moreover, the problem with
  trials using surrogates is that treatment effects
  on the definitive endpoints may not be predicted
  accurately by treatment effects on the surrogate.
  This problem can occur for two reasons. First is
  the imperfect association between the surrogate
  and the true outcome. Second is the possibility
  that treatment affects the true outcome through a
  mechanism that does not involve the surrogate.

18
Some Special Endpoints

• Repeated measurements are not common in clinical
  trials
• It is occasionally necessary in clinical
  trials to summarize endpoints repeatedly over
  some interval of time. A major difficulty after
  implementing such a scheme is using all of the
  information collected for analysis. It can be
  difficult to implement statistical techniques
  that simultaneously use all of the longitudinal
  information collected, are robust to the
  inevitable missing data, and flexible enough to
  permit valid inferences concerning a variety of
  questions.
• Quality of life
• Quality of life assessments are a special
  category of endpoints that are broadly used in
  clinical trials and other medical studies. These
  types of endpoints attempt to capture
  psychosocial features of the patients condition,
  symptoms of the disease that may be distressing,
  and/or functional (physical) status and are
  important in chronic diseases like cancer.
  Quality of life assessments are often made by
  summarizing items from a questionnaire
  (instrument) using a numerical score. Individual
  responses or assessments on the quality of life
  instrument might be summed, for example, for an
  overall score.