CHAPTER 4
Clinical Research
KEY POINTS
1 Clinical research includes all research involving human participants and represents a
range of research disciplines and approaches, including patient-oriented research,
clinical trials, epidemiology, and outcomes research.
2 Clinical trials are a subset of clinical research where human participants are
prospectively assigned to an intervention.
3 Randomized clinical trials are a subset of clinical trials that use a controlled
experimental design to assess the effectiveness of an intervention on an outcome.
4 Outcomes research and health services research include studies that seek to identify
the most effective and efficient intervention, treatments, and services for patient
care.
5 Study designs include experimental studies (clinical trials), observational studies
(cohort studies, case-control studies, and cross-sectional studies), and descriptive
studies (case reports and case series).
6 Scientific validity of a research study is evaluated by understanding the study
question, how the study was designed, and whether chance, bias, or confounding
could have accounted for the findings.
7 All study designs have inherent strengths and weaknesses and the quality of the
methods of the individual study and its scientific validity must be evaluated to
determine the study’s overall quality of evidence.
STUDY DESIGNS
Medical practice is evolving to include complex options for patient treatment and
preventive care, in part because clinical research methods and techniques to guide
patient care have advanced. To evaluate whether new treatments and diagnostic
approaches should be integrated into clinical practice or decide whether study
results reported in the literature are valid, clinicians should understand the
fundamental strengths and limitations of clinical research methods and the level
of evidence different types of studies provide.
[1] Clinical research, which encompasses all research involving human
participants, includes patient-oriented research involving understanding
mechanisms of human disease, studies of therapies or interventions for
disease, clinical trials, and epidemiology (1). It does not include secondary
121studies using publicly available datasets or studies using [2] existing biologic
specimens. Clinical trials, a subset of clinical research, are studies that evaluate
the effects of an intervention on participants by prospectively assigning
participants to intervention groups (1).
Epidemiologic methods and behavioral research are used in clinical
research to examine the distribution of disease and the factors that affect
health and how people make [4] health-related decisions. Outcomes research
and health services research include studies that seek to identify the most
effective and efficient intervention, treatments, and services for patient care.
The purpose of a research study is to test a hypothesis and to measure an
association between exposure (or treatment) and an outcome (e.g., disease
occurrence, prevention, symptom score, quality of life). The type of study
design influences the way the study results should be interpreted. Selection of
study design should be based on the specific research question being addressed.
[5] Analytic studies are often subdivided into experimental studies (clinical
trials) and observational studies (cohort studies, case-control studies, and crosssectional studies).
Descriptive studies (case reports and case series) can provide useful
information for informing future analytic studies.
The common types of clinical research study methods, strengths and
weaknesses of the specific study method, and interpretation of the results are
presented. Well-designed and executed clinical trials have been presumed to
represent the highest level of evidence for evaluating health interventions;
however, work in evidence-based medicine has urged that study design is
only one of several factors that drives the [7]quality of evidence within a
study. It should be noted that all study designs have inherent strengths and
weaknesses and the quality of the methods of the individual study and its
scientific validity must be evaluated to determine the study’s overall quality of
evidence (2,3). [6] Scientific validity of a research study is evaluated by
understanding the study question, how the study was designed, and whether
chance, bias, or confounding could have accounted for the findings.
ANALYTIC STUDIES
Clinical Trials
[2] Clinical trials include any study where participants are prospectively assigned
to receive an intervention or treatment (which may include being assigned to a
control group) and the outcomes to be measured are clearly defined at the time the
trial is designed. [4] Features of randomized clinical trials include
randomization (in which participants are randomly assigned to exposures),
122unbiased assessment of outcome, and analysis of all participants based on the
assigned exposure (an “intention to treat” analysis).
There are many different types of clinical trials, including studies designed to
evaluate treatments, prevention techniques, community interventions, quality-oflife improvements, and diagnostic or screening approaches (4). Investigators
conducting randomized clinical trials are expected to register the trial to comply
with mandatory registration and results reporting requirements (5,6).
Clinical Trial Phases
New investigational drugs or treatments are usually evaluated by clinical trials in
phases which begin with small trials designed to determine treatment safety
(phase 1) and progress to large-scale studies designed to determine efficacy and
side effects (phase 3) (Table 4-1) (7).
Randomized Controlled Double-Blinded Clinical Trial
The randomized controlled double-blinded clinical trial design minimizes bias
when evaluating the effect of an intervention on an outcome because
randomizing treatment assignment minimizes the influence of confounders
and blinding both the participant and the investigator minimizes the
possibility that ascertainment of the outcome is influenced by treatment
group assignment. When studies are not randomized or blinded, bias may
result from preferential assignment of treatment based on patient
characteristics or an unintentional imbalance in baseline characteristics
between treatment groups, leading to confounding (8).
Although not all studies can be designed with blinding, the efforts used in the
trial to minimize bias from nonblinding should be explained. Investigators are
expected to provide evidence that the factors that might influence outcome, such
as age, stage of disease, medical history, and symptoms, are similar in patients
assigned to the study protocol compared with patients assigned to placebo or
traditional treatment. Published reports from the clinical trial are expected to
include a table showing a comparison of the treatment groups with respect to
potential confounders and to demonstrate that the groups did not differ in any
important ways before the study began.
CONSORT Checklist
Clearly defining the outcome or criteria for successful treatment helps ensure
unbiased assessment of the outcome. A well-designed clinical trial has a sufficient
number of subjects enrolled to ensure that a “negative” study (one that does not
show an effect of the treatment) has enough statistical power to evaluate the
predetermined (a priori), expected treatment effect. The Consolidated Standards
123of Reporting Trials (CONSORT) Statement is an evidence-based, minimum
set of recommendations for reporting on randomized controlled trials
developed by the CONSORT Group to alleviate the problems arising from
inadequate reporting of randomized controlled trials. The 25-item CONSORT
checklist (Table 4-2) and flow diagram (Fig. 4-1) offer a standard way for
authors to prepare reports of trial findings, facilitating their complete and
transparent reporting, and aiding their critical appraisal and interpretation (9).
Table 4-1 Phases of Clinical Trials
Stage of
Testing
Trial Aim Numbers of
Participants
Phase
1
Evaluate treatment safety, determine safe dosage
range. Data are collected on the treatment (dose,
when, and how it is taken) and how participants
respond (in terms of effects and side effects).
Involve 20–
100 healthy
volunteers
or people
with the
disease or
condition
Phase
2
Evaluate treatment efficacy, further evaluate safety
and tolerability.
Involve up
to several
hundred
people with
the disease
or condition
Phase
3
Definitively determine the efficacy of treatment for
the intended population, compare with other available
treatments, assess adverse events and side effects.
Involve 30–
3,000
participants,
often within
randomized
clinical
trials
Phase
4
Evaluate for uncommon serious side effects and
adverse effects, optimal use including identifying
subgroups that may benefit more or less from the
treatment under study. These trials are particularly
important for identifying rare adverse events when
medications and devices are utilized in larger
populations. These studies are conducted after the
intervention has been approved by the FDA.
Large trials
or
observations
studies,
registries,
with
thousands of
participants
124FIGURE 4-1 CONSORT flow diagram.
Clinical Trial Design Considerations
[3] Well-designed and correctly performed clinical trials can clearly
determine the efficacy of an intervention and identify causal relationships,
125because they provide information about the relative and absolute risks and
minimize concerns about bias and confounding (see section Presenting and
Understanding the Results of Analytic Studies). Potential weaknesses of clinical
trials include cost restraints, length of time required to complete the study,
feasibility of recruitment and implementation, and issues with applicability to
populations outside of the strictly controlled study environment. Some clinical
research questions are not amenable to clinical trials because of ethical issues
with assigning patients to treatment groups, nature of the exposure being studied
(such as socioeconomic status, weight, environmental exposures, and other
patient characteristics), and urgent and emerging medical issues (e.g., the impact
of Zika virus on pregnancy outcomes) (2).
When evaluating the results from a clinical trial, consider how restrictive
inclusion and exclusion criteria may narrow the participant population to an
extent that there may be concerns about external validity or generalizing the
results. Other concerns include blinding, loss to follow-up, and clearly
defining the outcome of interest. When the results of a randomized controlled
trial do not show a significant effect of the treatment or intervention, the methods
should be evaluated to understand what assumptions (expected power and effect
size) were made to determine the necessary sample size for the study.
Intention-to-Treat Analysis
Randomized controlled trials should be evaluated with an intention-to-treat
analysis, which means that all of the people randomized at the initiation of
the trial should be accounted for in the analysis with the group to which they
were assigned. Unless part of the overall study design, even if a participant
stopped participating in the assigned treatment or “crossed over” to another
treatment during the study, they should be analyzed with the group to which they
were initially assigned. All of these considerations help to minimize bias in the
design, implementation, and interpretation of a clinical trial (8).
Table 4-2 CONSORT Checklist
126127Observational Studies
Observational studies can evaluate exposures and outcomes that are not amenable
to an experimental design (e.g., exposure is known or suspected to have harmful
effects). Not all research questions can be addressed with clinical trials because of
ethical issues (unethical to willingly subject participants to known harmful
exposures), the nature of the exposures being studied (inherent participant
characteristics), and feasibility issues (rare outcomes such as unusual adverse
events and diseases).
Observational studies, including cohort, case-control, and cross-sectional
studies, are analytic studies that take advantage of “natural experiments” in
which exposure is not assigned by the investigator; rather, the individuals
are assessed by the investigator for a potential exposure of interest (present
or absent) and outcomes (present or absent). The timing of the evaluation of
the exposure and outcome defines the study type.
Cohort Studies
Cohort studies often are referred to as longitudinal studies. Cohort studies
involve identifying a group of exposed and unexposed individuals and
following both groups over time to compare the rate of disease (or outcome)
in the groups. Cohort studies may be prospective, meaning that the exposure is
identified before outcome, or retrospective, in which the exposure and outcome
have already occurred when the study is initiated. Even in a retrospective cohort
study, the study is defined by the fact that the cohorts were identified based
on the exposure (not the outcome), and individuals should be free of disease
128(outcome) at the beginning time point for the cohort study (Fig. 4-2).
In a study that includes a survival analysis, the two cohort groups (exposed
and unexposed) begin with a population that is 100% well (or alive) at the
beginning of the study. The groups are followed over time to calculate the
percentage of the cohort that is still well (or alive) at different time points
during the study and at the end of the study. Although a survival analysis
typically describes mortality after disease (i.e., cancer patients who died within 5
years), it can be adapted to other events and outcomes (e.g., the percentage of
women who become pregnant while using long-acting contraceptives).
Cohort Study Design
Strengths of cohort studies include the ability to obtain attributable and
relative risks (RRs) because the occurrence of the outcome is being compared
in two groups (see section Presenting and Understanding the Results of Analytic
Studies). In these studies, only associations can be established, not causality.
Because randomization is not part of the study design, the investigator must
consider that a factor associated with the exposure may lead to the outcome rather
than the exposure itself. Misclassifying the exposure or the outcome and
confounding variables are potential sources of bias in cohort studies.
Given that truly prospective cohort studies can be expensive and take a long
time for completion, there should be compelling evidence for the public health
importance of the exposure(s) and association(s) being addressed. Issues related
to sample size and participant retention in the study protocol are as important in
cohort studies as they are in randomized controlled trials.
129FIGURE 4-2 Schematic of prospective and retrospective cohort study designs.
Case-Control Studies
A case-control study starts with the identification of individuals with a
disease or outcome of interest and a suitable control population without the
disease or outcome of interest. The controls should represent a sample of the
population from which the cases arose and who were at risk for the disease
or outcome but did not develop it. The relationship between a particular
attribute or exposure to the disease is retrospectively studied by comparing how
the cases and controls differed in that exposure (Fig. 4-3).
Odds Ratio
The measure of association for a case-control study is the odds ratio (OR),
which is the ratio of exposed cases to unexposed cases, divided by the ratio of
exposed to unexposed controls (see section Presenting and Understanding the
Results of Analytic Studies). If an entire population could be characterized by
its exposure and disease status, the exposure OR would be identical to the
RR obtainable from a cohort study of the same population. Although the RR
cannot be calculated directly from a case-control study, it can be used as an
130estimate of the RR when the sample of cases and controls are representative of all
people with or without the disease and when the disease being studied is
uncommon. Attributable risk is not directly obtainable in a case-control study.
Case-Control Study Considerations
The advantages of case-control studies are that they are lower in cost and
easier to conduct than other analytic studies. Case-control studies are most
feasible for examining the association between a relatively common exposure
and a relatively rare disease. Disadvantages include greater potential for
selection bias, recall bias, and misclassification bias.
FIGURE 4-3 Schematic of case-control study design.
Case-control studies may be especially prone to selection bias and recall
bias. Investigators need to understand sampling issues around which cases and
controls were selected for their study and how these may have affected exposure
rates. Subtle issues, such as interviewer technique, may affect the likelihood that
cases may recall or report exposures more readily than controls.
Cross-Sectional Studies
Cross-sectional studies assess both the exposure and the outcome at the same
point in time. Individuals are surveyed to provide a “snapshot” of health events
in the population at a particular time. Cross-sectional studies are often called
131prevalence studies because the disease exists at the time of the study, and the
longitudinal follow-up and disease duration are not known. Prevalence (PR) is
the existing number of cases at a specific point in time.
Cross-sectional studies are often done to evaluate a diagnostic test. The
value of the test (predictor) is compared with the outcome (disease). The
results of these evaluations are often presented as sensitivity and specificity. The
sensitivity and specificity represent the characteristics of a given diagnostic test
and do not vary by population characteristics. In contrast, the negative predictive
value (NPV) and positive predictive value (PPV) of a test do vary with the
baseline characteristics of a population such as PR of a disease (Fig. 4-4).
Cross-Sectional Study Considerations
Although cross-sectional studies are primarily descriptive, they may contribute
information about suggested risk factors for a disease by showing how that
disease varies by age, sex, race, or geography. In ecologic studies, disease rates in
various populations are correlated with other characteristics measured at the
population level (such as diet, ultraviolet radiation exposure, work or home
environment).
Caution must be used in interpreting findings from a cross-sectional study
because there is no temporal relationship between the exposure and the
outcome; therefore, causality cannot be established. Cross-sectional data can
be valuable in informing analytic study designs or used as supporting data for
documenting the consistency of an association.
DESCRIPTIVE STUDIES
Descriptive studies, case reports and case series, do not include comparison
groups.
Case Reports and Case Series
In a case report or case series, the characteristics of individuals who have a
particular exposure or outcome are described. Examples include series of patients
with a particular disease, patients who had a particular surgical procedure, or an
adverse outcome (such as ureteral injury at the time of hysterectomy). A case
report usually describes an unusual clinical scenario or procedure in a single
patient, whereas a case series usually includes a larger group of patients with
similar exposures or outcomes. Although members of a case series share a
particular characteristic, it cannot be assumed that there is a cause-andeffect relationship.
Hypotheses about exposures and disease may be developed from
132descriptive studies that should be explored in analytic studies. Because a case
series has no comparison group, statistical tests of association between the
exposure and outcome cannot be performed. A case series usually does not
yield any measure of association other than estimates of the frequency of a
particular characteristic among members included in the case series.
PRESENTING AND UNDERSTANDING THE RESULTS OF
ANALYTIC STUDIES
To present the results of clinical trials or observational studies, a variety of rates
and measures may be derived, as summarized below. To judge the scientific
validity of the results of clinical studies, an investigator needs to consider
whether the finding could have occurred simply by chance, by performing
appropriate statistical testing, or if there are other possible explanations for
the reported association, including bias or confounding. Besides statistical
significance and freedom from bias or confounding, there are several additional
criteria that can be applied to judge whether the treatment truly did affect disease
outcome or whether an exposure truly caused the outcome, as outlined below.
Rates and Measures
The terminology associated with rates and measures include the following
(Fig. 4-5):
Incidence (IR)—frequency of newly identified disease or event
(outcome).
Prevalence (PR)—frequency of an existing disease or outcome during a
specified period or point in time.
Odds Ratios (OR)—ratio of the probability of an exposure in one group
(cases) compared with probability of the exposure in another group
(controls).
Relative Risk (RR)—ratio of risk in the exposed group compared with
the risk in the unexposed group. If the RR = 1 (or not significantly
different from 1) then the risk in the exposed group is equal to the risk in
the unexposed group. RR >1 may suggest a positive association with the
exposed group having greater risk than the unexposed group, whereas RR
<1 implies a negative association with the exposed group having less risk
than the unexposed group.
Absolute Risk Reduction (ARR)—the difference in risk between the
unexposed (control) group and the exposed (treatment) group.
Relative Risk Reduction (RRR)—the percentage of reduction in the risk
133comparing the unexposed (control) group to the exposed (treatment)
group.
Number Needed to Treat (NNT)—represents the number of people who
would need treatment (or the intervention) to prevent one additional
outcome (to calculate the NNT, take the inverse of the ARR, i.e., 1 ï
ARR).
Sensitivity—among the people who have the outcome, this is the
proportion who have a positive test.
Specificity—among the people who do not have the outcome, this is the
proportion who have a negative test.
Negative Predictive Value (NPV)—among the people who have a
negative test, this is the proportion who do not have the outcome.
Positive Predictive Value (PPV)—among the people who have a positive
test, this is the proportion who have the outcome.
134135FIGURE 4-4 Comparison of sensitivity, specificity, and predictive values when the
prevalence of the disease varies.
FIGURE 4-5 Calculating rates and measures.
Statistical Testing
Statistical testing is used in clinical research for hypothesis testing in which
the investigator is evaluating the study results against the null hypothesis
(that there is no difference between the groups). Results from statistical testing
136allow the investigator to evaluate how likely it is that the study result is caused by
chance rather than an intervention or exposure (p value). In the case where a study
failed to find a significant difference, it is equally important to describe the
likelihood that the study conclusion was wrong and that a difference truly exists.
Finally, it is important to provide as precise a measure of the treatment effect or
association as possible and convey to the reader the plausible range in which the
“true” effect resides (or confidence interval [CI]).
p Value and Statistical Significance
The p value is a reflection of the probability of a type I error (alpha). This
reflects the probability that a difference between study groups could have
arisen by chance alone. In other words, it is the probability that there is a
difference between therapies, interventions, or observed groups when a true
difference does not exist.
Historically in the medical literature, a p value of less than or equal to 0.05
was used to determine statistical significance. This reflects a probability of 1
in 20 that the null hypothesis was incorrectly rejected based on the results
from the study sample. This p value may be adjusted downward if multiple
associations are being tested and the chances of false discovery are high. In
genome-wide association studies, in which hundreds of thousands of genetic
variants are tested between groups, p values are frequently set at 10−7
(0.0000001).
Beta Error and Power
Type II (or beta) error reflects the probability of failing to reject the null
hypothesis when in reality it is incorrect (i.e., there truly was a treatment
effect or a difference between the observed groups). In clinical trials it is
important for the investigator to address the beta error, even in the design stage of
the study. Study planners should determine the power (or 1 - the beta error) that
they would like their study to have to detect an association and, given
assumptions made about the differences expected between treatments, design the
study size accordingly. Be aware that small clinical trials cited as evidence for
“no effect of therapy,” may not have an adequate sample size to address the study
question; in essence, the study is not powered to detect the difference.
Confidence Intervals
CI provide the investigator an estimated range in which the true statistical
measure (e.g., mean, proportion, and RR) is expected to occur. A 95% CI
implies that if the study were to be repeated within the same sample
population numerous times, the CI estimates would contain the true
137population parameter 95% of the time. In other words, the probability of
observing the true value outside of this range is less than 0.05. When
evaluating measures of association, such as OR or RR with 95% CI, values that
include 1 (no difference) are not considered statistically significant.
Meta-analysis
Meta-analysis is a quantitative study design that involves the systematic
collection of data from previous studies, evaluation of the data, and
combination of the data (where appropriate). Meta-analyses can improve
precision of the effect measure and narrow the CI by aggregating treatment
effects data from several clinical trials to provide a summary measure.
Commonly performed by the Cochrane Collaboration within their systematic
evidence reviews, well-designed meta-analyses can provide high-quality evidence
that can be used for clinical decision making (10). There are important
considerations in interpreting the meta-analysis, including whether studies were
similar enough in their design, study population, and outcome measurement to be
aggregated and whether or not there were sufficient good-quality studies available
for analyses (2). Guides for systematic reviews and meta-analysis that involve
randomized controlled trials (i.e., the Preferred Reporting Items for Systematic
Reviews and Meta-Analyses [PRISMA] statement) and observational studies (i.e.,
the Meta-analysis Of Observational Studies in Epidemiology [MOOSE]
guidelines) are excellent resources for the investigator and the reviewer (11,12).
Bias
Bias is a systematic error in the design, conduct, or analysis of a study that
can result in invalid conclusions. It is important for an investigator to anticipate
the types of bias that might occur in a study and correct them during the design of
the study, because it may be difficult or impossible to correct for them in the
analysis.
Information bias occurs when participants are classified incorrectly
with respect to exposure or disease. This may occur if records are
incomplete or if the criteria for exposure or outcome were poorly defined,
leading to misclassification.
Recall bias is a specific type of information bias that may occur if cases
are more likely than controls to remember or to reveal past exposures.
In addition to establishing well-defined study criteria and accessing
complete records, information bias may be reduced by blinding
interviewers to a participant’s study group.
Selection bias may occur when choosing cases or controls in a case-
138control study and when choosing exposed or unexposed subjects in a
cohort study. A systematic error in selecting participants may
influence the outcome by distorting the measure of association between
the exposure and the outcome. Including an adequately large study
sample and obtaining information about nonparticipants may reduce bias or
provide information to evaluate potential selection bias.
Confounding
A confounder is a factor that is associated with the outcome (e.g., disease) and the
exposure. The confounder may account for the apparent effect of the exposure on
the disease or mask a true association. Confounders have unequal distributions
between the study groups.
Age, race, and socioeconomic status are potential confounders in many
studies. Results may be adjusted for these variables by using statistical
techniques such as stratification or multivariable analysis. Adjusting for
confounding variables aids in understanding the association between the
outcome and exposure if the confounding variables were constant.
Multivariable analysis is a statistical technique commonly used in
epidemiologic studies that simultaneously controls a number of
confounding variables. The results from an adjusted analysis include the
adjusted OR or RR that reflects an association between the exposure and
the outcome and accounts for the specific known confounders that were
included in the analysis.
Causality and Generalizability
The criteria needed to establish a causal relationship between two factors,
especially exposure and disease, are defined (13). Although there are nine
separate criteria for judging whether an association is likely to be causal, several
of these criteria are most relevant for clinical studies.
Biologic gradient or dose response refers to a relationship between
exposure and outcome such that a change in the duration, amount, or
intensity of the exposure is associated with a corresponding increase or
decrease in disease risk.
Plausibility refers to knowledge of the pathologic process of the disease or
biologic effects of the exposure that would reasonably support an
association. Plausibility overlaps with another concept, coherence, which
also refers to compatibility with the known biology of the disease.
Experiment refers to the evidence that the disease or outcome can be
139prevented or improved by an experiment that eliminates, reduces, or
otherwise counters the exposure.
Consistency refers to whether the association was repeatedly observed by
different investigators, in different locations and circumstances.
Temporality refers to the concept that cause must precede effect. For
example, is it possible in a case-control study that symptoms of preclinical
disease could lead to the exposure? Investigators must demonstrate that the
exposure was present before the disease developed.
Strength refers to the strength of association. The further the deviation of
the RR or OR from 1, the stronger the association and the easier it is to
accept that the study results are real. For example, studies have shown that
the possession of a BRCA1 mutation may increase the lifetime risk for
breast cancer by 14- to greater than 30-fold (varies by age) (14).
SUMMARY
Reviewing the medical literature is part of the ongoing education for those
who provide clinical care. Incorporating research findings into clinical care
is enhanced by understanding different study designs, their strengths and
weaknesses, and the measures of association they are able to provide.
Evaluating whether there is enough evidence available to support changing a
specific medication, procedure, or protocol used to care for patients is the
cornerstone to improving clinical practice. In a field that is rapidly
progressing, understanding clinical research
Nhận xét
Đăng nhận xét