Statistics Terms & Definitions: EBCR3 EX2 Comprehensive Study

You're reading a study that is evaluating the effect of different vitamin supplements on the development of colon cancer. The independent variable is _________ and the dependent variable is ___________.

A. Vitamin supplements; colon cancer

B. Colon cancer; vitamin supplements

C. A continuous variable; a nominal variable

D. A nominal variable; a continuous variable

A. Vitamin supplements; colon cancer

In the study from KC #1, data on the use of vitamin supplements (VS) is divided into "never taken", "taken 1-10 days/month", "taken 11-20 days/month", or "taken >20 days/month". The development of colon cancer (CC) over the study period is classified as "yes" or "no". VS is best described as a ____variable and CC is best described as a_____ variable.

A. Continuous; nominal

B. Ordinal, nominal

C. Nominal, continuous

D. Nominal, ordinal

B. Ordinal, nominal

Two types of numeric variables?

Continuous (interval; no meaningful zero OR ratio; meaningful zero)

Discrete (integer data)

Categorical data types? (two)

Nominal

Ordinal

Which measure of central tendency is most sensitive to outliers in the data?

A. Mean

B. Median

C. Mode

A. Mean

When are medians preferred?

With numeric or continuous data that is not normally distributed; also for some ordinal data.

When are means or averages preferred?

Numeric or continuous data that is normally distributed

When data is arranged in numeric order, the middle 50% of data values is called the?

A. Range

B. Standard deviation

C. The interquartile

C. The interquartile range

Describe the interquartile range (IQR)

Middle 50% of the data values when arranged in numerical order

Describe a standard deviation

Measures the spread of data around the mean

The larger the SD, the more spread out or variable the data is

A frequency distribution that contains data values which are much higher than the mean ("outliers") is

A. Left skewed

B. Right skewed

C. Normal

B. Right skewed

What does left skewed data look like on a graph?

(has outliers much lower than the mean)

Normal distribution is otherwise known as what?

Gaussian distribution

T/F: In normally distributed data, the mean/median/mode are the same value.

True

What percentage of data values are within one standard deviation of the mean in gaussian distribution? Two standard deviations?

68% in one

95% in two

T/F: As the SD (variation in the data) increases, the curve becomes flatter and vice versa.

True

An investigator would like to compare Drug A (new) to Drug B (standard of care) in the treatment of rheumatoid arthritis. The null hypothesis for this study is:

A. Drug A is superior to Drug B

B. Drug A is inferior to Drug B

C. Drug A is no different than Drug B

C. Drug A is no different than Drug B

Population parameters (4 examples)

1. Mean

2. Variance

3. Standard deviation

4. Odds ratio

Describe point estimates in inferential statistics

Sometimes called statistics, estimate the corresponding population parameter

T/F: Inferential statistics starts with a research question.

True

An investigator, Dr. Z, would like to compare Drug A (new) to Drug B (standard of care) in the treatment of lung cancer. Alpha is 0.05 and beta is 0.20. Dr. Z would like to detect a 20% difference in in the response rate between A and B.

What is the likelihood of concluding there is a difference when there really isn't one?

A. 5%

B. 95%

C. 20%

D. 80%

A. 5%

Describe the significance of an alpha level

Expression of the likelihood of concluding there is a difference when there really isn't one

Type 1 error

False positive

Describe the significance of a beta level

Expression of the likelihood of concluding there is

Define "risk"

Risk = number of subjects who develop an event (incidence) divided by total number in that group

Define "relative risk"

Ratio of risk in the exposed treatment group divided by the risk in the unexposed or control group

Relative risk measurements are inappropriate in which of the following study types

A. Case-control studies

B. Cohort studies

A. Case-control studies

How to calculate risk in exposed group?

Exposed with event / Exposed with event + Exposed without event

How to calculate risk in unexposed group?

Unexposed with event / Unexposed with event + Unexposed without event

How to calculate relative risk between exposed and unexposed groups?

Risk in exposed / Risk in unexposed

How to interpret a relative risk ratio of > 1?

Risk of outcome is increased in the treatment or exposed group

How to interpret a relative risk ratio of < 1?

Risk of outcome is decreased in the treatment or exposed group

How to calculate absolute risk reduction?

Risk with standard treatment (%) - Risk with new treatment (%)

Describe number needed to treat/harm (NNT/NNH)

Number of patients needed to be treated in order for one patient to benefit or number of patients needed to be treated in order for one patient to experience an adverse event

How to calculate NNT?

1/ARR

round up for NNT, round down for NNH

Define "Odds ratio"

Odds of an event occurring in "exposed" individuals divided by the odds of an event occurring in "unexposed" individuals

In what study type are odds ratios most commonly used?

Case-control studies

Your PLRP group will be doing a chart review study to evaluate the incidence of hepatotoxicity of a new antibiotic recently approved for children hospitalized for meningitis. You will compare this to the standard antibiotic used.

Your PLRP mentor asks what statistical test you will use to analyze your results.

Which of the following tests is the correct choice?

A. Chi-square

B. Unpaired t-test

C. ANOVA

D. Logistic regression

A. Chi-square

During a journal club presentation, your preceptor asks about the appropriate statistical test that should be used when evaluating the difference in pain scores (measured by the scale shown below) before and after a group of patients are given a new combination opioid and NSAID analgesic.

Which test would be correct?

A. Fisher's exact test

B. Paired t-test

C. Wilcoxan signed-rank test

D. Mann Whitney U test

D. Wilcoxan signed-rank test

Two sets of paired ordinal data

Describe "paired" data

Samples are dependent

(before and after evaluation in the same group of individuals)

Describe "unpaired" data

Samples are independent

(two different groups of individuals)

When do we use parametric tests?

When the dependent variable is numeric AND normally distributed

(i.e. t-test, ANOVA)

When do we use nonparametric tests?

When the dependent variable is numeric, but NOT normally distributed

(i.e. Wilcoxan rank sum, Kruskal-Wallis)

Which two tests are used for unpaired nominal data? (<2 groups)

Chi squared test

Fisher's exact test

Which test is used in paired nominal data? (<2 groups)

McNemar's test

Which two tests are used for unpaired ordinal data? (<2 groups)

Wilcoxon two sample test

Wilcoxan rank sum test (aka Mann Whitney U test)

Which test is used in paired ordinal data? (<2 groups)

Wilcoxon paired sign rank test

Which test do we use for unpaired continuous data? (<2 groups)

Unpaired t-test

Which test do we use for paired continuous data? (<2 groups)

Paired t-test

When do we use a Fisher's Exact test as opposed to a chi-squared test?

Small sample sizes (<5)

When do we use a Marginal Homogeneity test?

Paired nominal data (>2 groups)

When do we use a Kruskal-Wallis test?

Unpaired ordinal data (>2 groups)

When do we use a Friedman test?

Paired ordinal data (>2 groups)

When do we use an ANOVA test?

Unpaired continuous data (>2 groups)

When do we use a repeated-measure ANOVA?

Paired continuous data (>2 groups)

You're preparing for a journal club presentation fo a study assessing the mean change in cholesterol in 3 different groups of patients. Results show: A=12mg/dL, B=16mg./dL, C=24 mg/dL. An ANOVA test was used to evaluate the results. The p value is less than alpha so you can conclude a statistically significant result.

The appropriate conclusion based on the ANOVA test is:

A. C is better than B

B. B is better than A

C. C is better than A

D. One of the groups is different from at least one of the other groups

D. One of the groups is different from at least one of the other groups

T/F: When testing more than 2 groups, determination of between which two groups a difference lies is not possible.

True

T/F: As you conduct more and more tests, your chance of a false positive increases (alpha increases).

True

What variety of test do we use to determine which groups are different from each other (after testing more than 2 groups)?

Post hoc test

Examples of post hoc tests?

Bonferroni test

Tukey's test

Predictor = ____________

A. Independent variable

B. Dependent variable

A. Independent variable

If our predictor is continuous and our outcome is continuous, what test do we use?

Correlation/linear regression

If our predictor is continuous and our outcome is categorical, what test do we use?

Logistic regression

If our predictor is categorical and our outcome is continuous, what test do we use?

ANOVA/General linear model

If our predictor is categorical and our outcome is categorical, what test do we use?

Kappa or kendall/spearman correlation

Describe "correlation" (positive and negative)

Increase in independent variable -> increase in dependent variable, positive correlation

Increase in independent variable -> decrease in dependent variable, negative correlation

What is a phase III study?

A prospective study comparing the effect of an investigational intervention to a control in human beings

T/F: Phase III studies can be comprised of animal participants.

False

T/F: Phase III studies usually employ randomization and blinding.

True

What is the general purpose of phase III studies?

The purpose is generally to demonstrate the clinical use of the intervention and establish a safety profile.

Describe "clinical equipoise"

A state of genuine uncertainty as to the advantages/disadvantages of one therapy versus another

When is a phase III trial needed?

When clinical equipoise exists

When Phase II data exists to suggest a new therapy might be better than the standard, but more evidence is needed

Challenges associated with Phase III trial design?

Medical care has improved dramatically; thus harder to show major therapeutic differences

Requires a larger sample size to achieve respectable power

Newer therapies may offer other advantages besides better efficacy; similar efficacy but lower cost, better compliance, fewer side effects etc.

Key characteristics of superiority designs?

Generally ask "Is new therapy different (hopefully better) than standard?"

"How much is "better/different?" (effect size)

Statistical analyses are generally two-tailed (interested in either outcome) (null hypothesis being that there is no difference)

May be one-tailed - easier to find a difference; if the results are in the unexpected direction, cannot state the difference is significant

When can selection bias occur?

Can occur if randomization is predictable or known (also called allocation bias)

Can also occur if subjects selected for the study are different than population of interest

When can accidental bias occur?

Can occur if the randomization process does not achieve balance of risk factors

More of an issue in smaller studies

When can investigator or assessment bias occur?

Can occur when investigator is assessing outcome in unblinded studies

When can follow up bias occur?

Lost to follow up occurs preferentially in one group over the other

T/F: Phase III interventions can be one or a combination of diagnostic, preventative or therapeutic drugs, biologics, devices, or procedures.

True

How are interventions applied in Phase III trials?

Standard, prespecified method to all participants throughout the course of the study

Randomization minimizes what?

Allocation bias and confounding

T/F: An independent party should be responsible for developing and implementing the randomization process.

True

T/F: Randomization is usually generated by a computer program that is "invisible" to investigators.

True

Three types of fixed randomization?

Simple

Blocked

Stratified

Describe "simple" randomization

Basic, like flipping a coin or using a random number generator

Easy

Can result in imbalance of the arm allocation, especially if the sample size is small

Advantage + disadvantage of block randomization?

Advantage: Allocation is equal at anytime during the trial

Disadvantage: May be able to predict final assignment in each block if block size is known (allocation bias)

T/F: Stratified randomization requires that prognostic factors be known at the time of randomization.

True

T/F: The number of strata in stratified randomization is restricted by sample size.

True

What is an alternative to stratification?

Prespecify a statistical adjustment for potential confounding factors (i.e. regression analysis)

When are "dummy" designs employed?

Used when control and intervention medications are given by different routes or on different schedules

As a method of maintaining blinding

What is the most commonly used Phase III randomized trial design?

Parallel trials

What is the intention behind parallel trial designs?

Subject stays on the same treatment arm throughout the study

Describe "crossover" randomized trial designs

Participants are their own controls since each patient is in the control arm and the experimental treatment arm

Requires a smaller sample size than do parallel trials

Washout period may be appropriate

In which of the following Phase III randomized trial designs are losses to follow up most detrimental

A. Crossover

B. Parallel

A. Crossover

T/F: Crossover trials are a good design for "waxing and waning" diseases or diseases that require urgent treatment.

False; not really good to withhold treatment in these participants.

Key features of composite outcomes?

Outcomes should be related

Generally reduces the sample size needed

MACE events are a good example

Interpreting composite outcomes becomes problematic if results are inconsistent

Measurement of outcome variables should be what? Especially when?

Should be standardized, validated, and free of bias; ESPECIALLY if trial is not double blinded

Describe a surrogate outcome/endpoint?

For example decrease in LDL-C -> reduction in ASCVD

An outcome that achieves something if the outcome is not specific, like MACE

Which of the following outcome types are typically described as "exploratory?"

A. Primary

B. Secondary

C. Tertiary

C. Tertiary

T/F: As more statistical tests (comparisons) are done, a "significant" finding may be found just by chance, thus increasing the likelihood of making a type I error.

True

What can be done to adjust for multiple comparisons?

Use a smaller alpha (Bonferroni's correction: if k comparisons are being made, then for each comparison, alpha should be divided by k)

Hierarchal procedures (Holm's and Hochberg's procedures) - these are rank comparisons done by level of importance