
Prospective Study
Followed forward in time

Retrospective Study
Reviewed back in time

Parallel
 Subject either receives the control or the experimental medication throughout the treatment
 Two or more groups receiving treatment at the same time

Crossover
Each subject serves as his/her own control and receives both the study and control treatments

Give two examples of a descriptive study

Descriptive Study
 quantitative research that describes naturally occuring events
 document and communicate experiences

Explanatory Study
Examine etiology, efficacy, or cause using a comparison strategy

What type of trial is a RCT or educational intervention study?
Experimental

Experimental Study
 Evaluate the efficacy of therapeutic or educational interventions
 Investigator controls allocation

Name three types of observational explanatory studies
 Case Control
 Cohort (follow up)
 Cross sectional

When you think case control think " ".
Disease

Observational study
 Investigator does not control allocation
 Observes nature

Casecontrol
 Retrospective
 Patients with the disease (cases) and without the disease (control) are compared to determine the exposure to the risk factor in question

List some strengths of casecontrol
 Quicker and less expensive than cohort or experimental
 Optimal for evaluating rare diseases
 Appropriate for evaluating disease with long latency periods

List some limitations of casecontrol
 Inefficient for evaluation of rare exposures
 Prone to bias (i.e. recall bias)
 Do not allow for direct calculation of incidence rates
 Temporal sequence of exposure and outcome is not always known

What is another name for a cohort study?
Follow up

When you think "cohort" think " "
Exposure

Cohort (Follow up)
 Prospective or Retrospective
 Design may involve the evaluation of risk factors for disease development in a specified population

List some strengths of cohort
 Appropriate for evaluating rare exposures
 Temporal sequence of exposure and outcome is clear
 Selection and recall bias is minimized
 Allow for direct calculation of incidence rates

List some limitations of cohort
 Time and cost tend to be more than with casecontrol
 Subjects are lost to follow up over course of the study
 notefficient for evaluating rare diseases

When you think cross sectional think " "
Slice in time

Cross sectional
 Prevalence Study
 Gather information about the prevalence of healthrelated states and conditions
 Do not identify cause and effect relationships
 Identify existence of health problems

What type of study would an epidemiology study be categorized as?
Cross sectional

Superiority
 Trial with the primary objective of showing that the response to the investigational product is better to a comparator
 Typically done if we have a novel therapy and a placebo; superiority between ACE inhibitors

What value cannot be crossed by the confidence interval in order to determine superiority?
"0"

Equivalence
 Trial with the primary objective of showing that the respose to 2 or more treatments differs by an amount that is clinically unimportant
 Usually demonstrated by showing that the true treatment difference is likely to lie between a lower and an upper equivalence margin of clinically acceptable differences
 AKA: if confidence intervals are between the margins then the differences are clinically unimportant but if a CI goes outside of the margin then there is a clinically important difference

Noninferiority
Trial with the primary objective of showing that the response to the investigational product is not clinically inferior to a comparative agent

Equivalence margin
 prespecified maximum allowable difference between the new and the standard treatment
 (in a noninferiority trial if the confidence interval crosses this line it shows inferiority)

Nof1
 controlled study conducted in a single subject where periods of exposure to at treatment are compared to periods of exposure to placebo
 i.e. one patient with a very rare disease with a new treatment

Stability
study designed to determine the stability of drugs in various preparations

Bioequivalence
study that evaluates whether products are similar in rate and extent of absorption

Survey research
research where responses to questions asked of subjects are analyzed to determine the incidence, distribution, and reltaionships of sociological and psychological variables

Programmatic research
research focused on the impact of economic value of programs and services provided

Quality of Life
evaluation of a patient's living situation based on the patient's environment, family life, financial situation, education, and health

Postmarketing surveillance
study designed to examine drug use and frequency of side effects following FDA approval

Pharmacoeconomic
study of economic impact of drug therapies or services

List the following in order of weak to increasing strength: casecontrol, case report, case series, RCT, Cohort
 Case report
 Case series
 Casecontrol
 Cohort
 RCT

Metaanalysis
 Process of systematically evaluating and combining the results of clinical trials that have been completed
 Systematically aggregates and quantifies results from multiple clinical trials

Data pooling: Type 1
 Combining raw data to increase study power
 Strongest type of metaanalysis: combine all raw data from individual studies and start the analysis over

Data Pooling: Type 2
 Combining conclusions (i.e. odds ratio) of single trials to create an overall odds ratio
 Done most frequently

Data Pooling: Type 3
 Summating individual results (i.e. 7 of 10 trials support A is more effective than placebo)
 Not very strong
 Weakest type

List some limitations of metaanalysis
 Heterogeneity of clinical trials: goal is for there to be no differences but in reality there will be small differences between the trials
 Number of trials does not necessarily move you closer to the truth
 Garbage in equals Garbage out

Funnel Plot
 Measure precision against treatment effect
 Absense of publication biase is suggested by a symmetrical inverted funnel
 Open circles= unpublished studies; Closed circles= published studies

Blobbogram (Forest Plot)
 Each line represents the confidence interval of a clinical trial
 Diamond is the result of the meta analysis : middle is the estimate of effect and width is the confidence interval
 If confidence interval crosses the line it means there was no effect. So if the diamond also crosses the line it shows no effect.

Noninferiority to Superiority
 Designed for noninferiority trials
 Actual P values for superiority are presented to allow independent assessment
 Analysis according to intentiontotreat principle

Superiority to Noninferiority
 Protocol must contain a prospectively defined  delta for noninferiority
 Intention to treat and perprotocol analyses should show similar findings
 Trial was properly designed and carried out with requirements of a noninferiority trial

Population
every individual in the universe with the specific characteristic(s) or disease state under study

Sample
group of individuals chosen as representatives from the population under study

Inclusion criteria
characteristics that must be possessed by the patient in order to enroll in the study

Exclusion characteristics
Characterstics that prevent patient participation

Patient demographics
baseline characteristics of patients must be as similar as possible or controlled for in order to reduce confounding effect (i.e. age, disease states, medications)

List the 5 types of controls
 Placebo
 Historical
 Crossover
 Standardtreatment
 Withinpatient comparison

Historical Control
Comparing current intervention to data collected in the past because it is no longer ethical to give the patient this historical treatment

Standardtreatment control
 "head to head" study
 New ACE inhibitor in development would be compared to current available treatment, not a placebo

Within patient comparison control
 Receive both treatments at one time
 Seen in dermatological studies

Randomization
 Each subject has an equal and independent chance of being in any of the treatment arms
 Ensures that all factors which can differ prognosis will be equally distributed throughout the groups
 Examples: simple, block (cluster), nonrandomization, stratification

Simple Randomization
 Tossing a coin
 Inexpensive
 May wind up with mismatched groups in terms of numbers

Block Randomization
 At any time if you stopped accepting patients your treatment groups will not be off by more than a block size
 Block size of 4: 6 possible combinations of which 1 will be chosen. Once done with first block another combination is randomly chosen

NonRandomization
 Not ideal but it is a possibility
 Used if you want to describe the patient sample

Stratification Randomization
 Considered a type of randomization but actually occurs in combination with randomization
 Strate= nonoverlapping groups
 Stratify patients by age, sex, heart failure stage and then randomly assign patients

What are some limitations of blinding?
 May be difficult to blind a medication with a distinct tase, physiologic effect, or continuous titration
 Expensive and time consuming

Openlabel trial
 Everyone, including the patient and the researcher, knows who is receiving what treatment
 This is okay if you are measuring objective data (a lab value) and not subjective data

Singleblind trial
 At least one party in the trial is blinded
 Investigators aware of the treatment assignment but the patient is unaware

Doubleblind trial
 When bo the patient and the investigator are blinded
 Minimizes bias

Tripleblind trial example
If a radiologist is involved in the study to perfrom the chest xray he would also not know which patient received what treatment in addition to the patient and investigator not knowing

Doubledummy trial
comparing 2 different dosage forms (inhaled vs. SQ insulin)administer placebo inhaled/active SQ to one group and active inhaled/placebo SQ to the other group so neither knows whicho ne is active

Bias
Systematic error that enters a clinical trial and distorts the data obtained

Selection Bias example
Only going to select men to be in the study but it isn't representative of the entire population

Observation Bias
 direct observation or information bias that may enter usually because the study is not blinded
 example: MD knows what the patient is receiving and his documentation about the effects are bias

Recall Bias
Patients will remember everything if the experience isnegative but if there were no problems they may not remember much

Interviewer bias
When an interviewer is trying to collect information in a study from the patient and they do so in a way that is leading

Misclassification Bias
A patient is classified one way when they should have been classified as another way (Class III HF when they should have been Class IV)

Channeling Bias
Allocation bias that happens when drugs with a similar therapeutic indication are given to patients with different prognostic indicators (i.e. sicker patients vs. not as sick patients)

How can you control for bias?
 Minimize confounders
 Proper selection of patients
 Objective methods of selecting data
 Blinding
 Use of a control group
 Reliable data sources

Intentiontotreat Analysis
 Compares outcome based on the intended inital subjects' assignements
 Determines the effect of treatment under usual conditions (i.e. how it will be in the real world)
 Gives a conservative estimate of the treatment difference
 No data should be eliminated

Astreated Analysis
 Analyzes subjects based on what intervention the subjects actually received
 No data should be eliminated
 Cleaner data regarding method effectiveness compared with intentiontotreat analysis

Perprotocol Analysis
 Analyzes those subjects who precisely followed the protocol
 Gives superior information regarding method effectiveness compared with other analyses
 Provides a more generous estimate of the differences between treatments
 Problematic if compliance is related to prognosis

Internal Validity
 Within the confines of the study: the methods and analysis used stand up to scrutiny, the investigators' interpretation is supported, the results appear accurate
 Affected by controls, randomization, blinding, confounders
 Strong trial design= internal validity

External Validity
 Generalizability
 Ability to extrapolate to the population of interest
 Can the results be applied to other groups, patient, or systems?

Reliability
Reproducibility of measurements

Validity
Measuring what you want to test

Is this reliable? Valid?
 Reliable: Yes
 Valid: No
 Because it is clumped it is reliable because it is consistent (reproduce measurements) but it is not valid because it is not in the center so it didn't measure what we wanted to test

Is this reliable? Valid?
 Reliable: Yes
 Valid: Yes
 This is reliable because the dots are clumped meaning we were able to reproduce the measurement it is also valid because it is centered around the middle meaning we were able to measure what we wanted to.

Attrition
 People dropping out of the study
 Expect 1020% drop rate but not reasonable to be higher than 25%

Interim Analysis
Taking a look at the data before the actual end of the study to make sure that one group isn't at an increased risk vs. another group

Early Stop Rules
 Relate to interim analysis
 Investigators define under what circumstances they would stop the study early

Compliance Assessment
 Direct observation therapy (RN observes pt taking medicine)
 Pill Count
 Serum Concentration
 Ask the patient

Dependent Variable
Outcome of interest within the study

Independent Variable
Intervention or what is being manipulated

Confounding Variable
Factor affecting the patients' conditions and associated statistically with the intervention being evaluated

Univariable Analysis
 Applicable to a set of observations that contains ONE dependent variable and NO independent variables
 example: observational (cohort, case control)

Bivariable Analysis
 Applicable to a set of observations that contains ONE dependent variable and ONE independent variable
 example: clinical trial

Multivariable Analysis
 Applicable to a set of observations that contains ONE dependent variable and MORE THAN ONE independent
 variables
 example: parallel study with more than one intervention

Diseaseoriented Evidence (DOE)
 Involves the measurement of an intermediate outcome in a study such as surrogate marker or end point
 Aimed at increasing understanding of a disease and its incidence, prevalence, diagnosis and treatment, and prognosis
 examples: intermediate outcomes, surrogate markers, disease markers [BP, A1c]

Patientoriented evidence (POE)
 Outcomes that patients care most about
 Example: Quality of life, morbidity, mortality/death

Continuous Data
 Data with constant defined units of measure
 Mathematical relationship between the values
 Equal distance between increments
 ex. Interval Scale and Ratio scale

Interval Scale
 Predetermined order with consistent level of magnitude with an arbitrary zero point assigned
 example: degrees Farenheit

Ratio Scale
 Interval scale with an absolute zero
 examples: BP, HR

Discreet Data
Ordinal and Nominal

Ordinal Data
 Limited number of categories that DO have implied rank or order
 Order of numbers is meaninful but not mathematical
 Ordinal scale can be ranked in a specific order, be it low to high or high to low
 Examples: likerscale questionnaires, pain intensity score

Nominal Data
 Names categories that hae no implied rank or order
 No mathematical relationship between classifications
 Subjects are categorized based on some mutual characterstics, which the entire group possess
 Examples: presence or absence of disease, gender, hair color

Descriptive Statistics
 Describe data set
 Summarizes data into a useful form
 examples: Measures of central tendency (mean, median, mode), measures of variability (range, IQR, percentile, and SD)

Inferential Statistics
 Determine the statistical importance of differences between groups and allow conclusion to be drawn from the data
 Can describe the whole population
 examples: SEM

Mean
 Arithmetic average of the data
 May be computed for CONTINUOUS data
 Extremely sensitive to outliers

Median
 The 50th percentile
 Value above which or below which half of the data points lie
 Not sensitive to outliers
 Useful for continuous or ordinal data

Mode
 Most commonly obtained value in the distribution
 Useful to describe nominal, ordinal, and continuous data

Range
 Difference between the largest and the smalles values in the distribution
 Highly sensitive to outliers

Interquartile Range
 Measure of variability directly related to the median
 Describes the range between the 25th and 75th percentile values
 Clearly defines where the middle 50% of measures occurs and indicates the spread of data
 Used to descrive the variability for ordinal data

Percentile
 Point on the distribution where a value is larger than x% of the other values in the group
 Not sensitive to outliers

Standard Deviation
 Describes the variability of data about the sample mean
 Meaningful only when it is calculated for normally distributed continuous data
 68% of data will fall within + 1 SD and about 95% of data will fall within + 2 SD
 Not sensitive to outliers

Standard Error of the Mean
 Measure of the precision with which a sample mean estimates the population mean
 SEM= SD/sqrt (n)
 Used to calculate confidence intervals
 The true but unmeasured mean of the larger population of interest will lie within two standard errors of the sample mean about 95% of the time
 SEM decreases with increase in sample size

Normal Distribution
 Bellshaped curve
 Gaussian distribution
 Symmetrical scatter around the mean

Bimodal Distribution
 Not appropriate to use a mean
 Should use a median or mode when looking at bimodal data to describe the central value

Is this a postive or negative skew?
Positive

Is this a postive or negative skew?
 Negative
 Could measure with median or measure of variability can be the IQR

Steps in hypothesis testing
 Define the research question to be studied and restate it in terms of the H_{0} and H_{1}
 Collect data on the sample subjects
 Calculate the test statistic
 Evaluate the evidence against H_{0}
 State the conclusion

Null Hypothesis
 Statistical hypothesis denoted by H_{0}
 The hypothesis investigators test when they conduct a study
 Goal is to disprove this hypothesis
 Hypothesis of no difference: Group A=Group B

Research Hypothesis
 AKA alternative hypothesis denoted by H_{1}
 Difference exists between groups Group A does not equal Group B
 Usually the purpose or objective of the study

Test statistic
 The test statistic is a number, computed from the data, that is compared with the associated probabilty distribution to determine the P value
 Associated with a statistical test that uses it

Alpha
 Falsepositive rate the investigators are willing to tolerate (type I error)
 Whe achieved, allows H_{0} to be rejected and H_{1} to be accepted
 Decided upon before the study
 Generally < 0.05

How do you interpret alpha
 Probability of making a type I error is < 0.05
 1 out of 20 times a type I error is made
 5% chance of a false positive

Pvalue
 Probability that the observed results in the study could be due to chance alone
 Detemined after the fact and calculated from the sample data

Interpreting P values
 P < alpha: statistically significant (not zero) and not due to chance
 P > alpha: may be due to chance
 Pvalue means that if the H_{0} is true, then the statistical test should discredit H_{0} 5% of the time or less

Beta
 The falsenegative rate the investigators are willing to accept
 The probability of making a type II error
 < 0.2 acceptable but < 0.1 ideal

Type I error
 alphatype error
 Reject the null hypothesis when it is true (false positive)
 Concluding that there is a difference in fact there is not a difference

Type II error
 Beta type error
 Accept null hypothesis when it is false
 Concluding that there is no difference when in fact there is a difference (false negative)
 many be due to chance or small sample size

Fill in the chart with the type of error:

Power
 Defined by beta (Power= 1 beta)
 Ability to detect a difference if a difference existsminimizes type II error
 If power is not achieved, but difference is found between the groups then the power is irrelevant

What factors will increase power?
 Increased sample size
 increased alpha
 Large differences between population

What factors will decrease power?
 Poor choice of statistical test
 Poor study design

Confidence Interval
Includes the range of possible values for the true treatment difference that are statistically likely given the results of a specific trial

Relationship of confidence interval with sample size
 Width of CI will narrow when sample size is increased
 Width of CI will widen when the sample size is decreased

Relationship of confidence interval with power
 The width of the CI will narrow when power is increased
 The width of the CI will widen when power is decreased

Interpreting the CI of a ratio
If the CI includes "ONE" then significance is not achieved

Interpreting the CI of a nonratio variable
If the CI includes "ZERO" then significance is not achieved

Statistical Significance
The observed difference is associated with a P value that is below the established alpha level, indicating that chance isnot a likely explanation for the difference observed

Clinical Significance
Clinical importance of the data generated in a study, regardless of statistical results

Parametric Test
Only valid when used with continuous, normally distributed data

Nonparametric
Applies to nonnormal distributions or to data that do not meet the criteria for parametric tests

What factors should be considered when chosing a statistical test?
 Type of data collected (continuous, ordinal, nominal)
 Distribution of the data
 Number of groups being compared
 Study design
 Investigator or statistician preference

Onesided vs. Twosided Parametric Test
 One sided tests can detect a difference in one direction
 Two sided tests can detect a difference in two directions (better or worse)

Ttest
 Used to compare the means from 2 independent samples
 i.e. parallel design

Paired Ttest
 Used when 2 groups contain the same people in the groups
 i.e. crossover study design

ANOVA
 Analysis of variance
 Used in place of a ttest when comparing means of 3 or more groups
 i.e. Bonferroni, Tukey or Scheffe

ANCOVA
 Analysis of covariance
 Method used for controlling for the effects of multiple confounding variables

ChiSquare
 Compares percentages between 2 or more independent samples
 Most useful for nominal data
 Used to answer research questions about rates, proportions, or frequencies

Fisher's Exact Test
 Used instead of Chisquare when: The matrix or any cell will be less than 5 or the entire sample size is less than 40
 Used with 2 independent sample groups

McNemar's Test
 Used to compare nominal data for paired samples (i.e. crossover)
 May be used in a casecontrol design in which data are obtained from related patients

MantelHaenszel
Used to compare nominal data while controlling for the effect of a confounder

MannWhitney U test
 Nonparametric equivalence to the ttest
 Used most often when data are measured on an ordinal scale
 Used for independent samples
 May also be used for data that are not normally distributed or when the variable is discrete
 MannWhiteney U= Wilcoxon Rank Sum

Wilcoxon Signed Rank Test
 Nonparametric equivalent of the paired ttest
 used for ordinal data
 Used when the data are matched or paired

KruskalWallis
 Nonparametric equivalent to ANOVA
 Used for ordinal data
 Used for 3 or more groups with independent samples

Friedman
 Used for 3 or more groups where subjects participate in more than one group (paired sample)
 Used for ordinal data

Fill in the blank with the appropriate test.

What is correlation?
Describes the strength of the relationship between variables

What is pearson's correlation?
 Describes the strength of the linear association between two continuous, normally distributed variables
 "r"
 r = +1 : strong, perfectly positive linear association
 r = 0 : indicates no association
 r = 1 : indicates a strong, perfectly negative linear association

Characterstics of Pearson's r
 Variables are continuous
 Variables follow a normal curve
 Observations are independent

When is correlation not defined well by Pearson's r?
 Heteroscedasticity (different spread or scatter from bestfit line)
 Nonlinear relationship
 Outliers present

Spearman's correlation (Spearman's rho)
 Nonparametric equivalent to Perason's r
 Useful when variables are not normally distributed, are on an ordinal scale, or when the relationship is nonlinear
 Useful when outliers are present or when there is a smalls ample size
 r_{s} or p (greek rho)

What is regression?
Used to predict the value of one variable based on the knowledge of another

Simple linear regression
 One continuous dependent variable and one continuous independent variable
 (line cannot be extended)

Characteristics and assumptions for simple linear regression
 The dep. variable is continuous
 Relationship is linear
 Variables follow a normal distribution
 Observations are independent
 Data have the same spread or scatter
 No outliers

What is the coefficient of determination?
 Represents how much of a dependent variable is explained by an independent variable
 r^{2}= simple regression analysis
 R^{2}= multiple regression analysis
 Higher value indicates a stronger relationship

Multiple Linear Regression
Used when there is one continuous dependent variable and 2 or more continuous or discrete or independent variables/covariates

Characteristics and assumptions of multiple linear regression
 Dependent variable is on a continuous scale
 Relationship is linear
 Variables follow a normal distribution
 Observations are independent
 Data have the same spread or scatter
 No outliers

Simple Logistic Regression
 Used when there is one discrete dependent variable and one categorical or continuous independent variable
 Used when the relationship between variables is nonlinear
 Often used in survival analyzes

Multiple Logistic Regression
 Used when there is one discrete dependent variable and two or more continuous or discrete independent variables
 When the relationship is nonlinear

Fill in the the chart below:

Relative Risk/Risk Ratio
 Used for prospective studies
 "Ratio of risk of an event occuring in the experimental group compared with the control group"
 Probability that a condition will occur
 Indicates the risk of the event after the experimental treatment as a percentage of the orginal risk
 < 1= therapy decreased risk, 0= no difference, > 1= therapy increased hte risk
 RR= [A/(A+B)]/[C/(C+D)]

Odds Ratio
 Used for retrospective studies or casecontrols
 Ratio of odds of an event occuring in the experimental group compared with the control group
 Does not take into account the total number of participants
 <1 = therapy decreased the risk, 0= no difference, > 1= therapy increased the risk
 OR= AxD/ CxB

Relative Risk Reduction
 Estimates the percentage of baseline risk that is removed as a result of therapy
 Percent reduction in the experimental group event rate compared with the control group rate
 0= there was no effect of the treatment compared with the control
 RRR= 1RR x 100
 RRR= [% placebo % treatment]/% placebo

Absolute Risk Reduction
 The difference in the event rate between a control group and an experimental group
 Provides the percentage of patients spared the adverse event as the result of the treatment
ARR= % placebo % treatment

Number needed to treat
 The number of patients who require treatment to prevent one event
 Assumes the baseline risk is equal for all patients
NNT= 1/ARR

Number Needed to Harm
The number of people that need to be at risk for a define period of time for a harm to come to one person
NNH= 1/ARR

Survival Analysis
 Evaluating time to event
 Analyze the time between a given starting point and a given event
 Adjusts for the fact that patients are followed for different lengths of time and that the event may not have occured in some subjects

Survival Curve Comparisons
 Logrank test: determine whether there is a statistically significant difference between the two curves
 A hazard ratio can summarize a survival analysis and takes all time points into consideration
 Interpet hazard ratio the same as relative risk

