Biostats tests, & random names

If I have to suffer, you do too.

Comparing proportion of a population to 1 group

Z test, Chi square(X²)?

Comparing proportion of 2 independent groups

Z test, Chi square(X²)

Comparing proportion of 3+ independent groups

Chi square(X²)

Comparing proportion for SMALL SAMPLES

Fisher’s Exact Test

Comparing proportions in:
ONE(dependent/matched/paired) group for
Before/After type exposure or…
Multiple exposures (still only 1 group)

Mcnemar's Test

Comparing proportions or means in
2 groups when NONPARAMETRIC DATA

Mann-Whitney U Test (Wilcoxon rank-sum test)

Comparing proportions or means in
3+ groups comparison when data is NONPARAMETRIC DATA

Kruskal-Wallis Test

Comparing means of:
1 group mean to a hypothesized population mean

One sample t-test

Comparing means of:
2 independent groups

Independent/Unpaired t-test
or just “t-test”
ANOVA

Comparing means of:
3 independent groups

ANOVA

Comparing means of:
2 dependent sample means (1 group)
→ Before/after or 1 group - 2 exposures

Dependent/Matched/Paired t-test

Comparing means of:

2 groups(out of 3+) after ANOVA determined a difference…

Post hoc pairwise group analyses
ie. Tukey’s HSD test
Bonferroni’s Correction Method

Comparing means of:

2 groups(out of 3+) after ANOVA by adjusting α via → α/ total # of comparisons = new alpha

Bonferroni's Correction Method

Comparing means of:
2 groups(out of 3+) after ANOVA by adjusting p-values via statistical software and use same alpha

Tukey’s HSD

Comparing means of:
Groups which differ in 2 factors (ie. smoking & physical activity)

two-way ANOVA

Comparing means of:
Groups which differ in 3 factors (ie. smoking & physical activity & DASH diet)

multi-factor ANOVA

Comparing means of:
Dependent/Matched/Paired groups with 3+ time points or interventions

Repeated Measure ANOVA

Comparing means of:
Dependent/matched/paired for NONPARAMETRIC DATA

Wilcoxon signed-rank test

Formula to traditionally calculate correlation coefficient (r)

Pearson’s product-moment correlation
Pearson’s r
Pearson’s correlation coefficient

Formula to calculate correlation coefficient (r) when NONPARAMETRIC

Spearman’s Rank Correlation

Formula to calculate correlation coefficient (r) when ORDINAL

Kendall’s Rank Correlation

Type of regression with:

1 continuous or categorical predictor variable (x)
1 continuous outcome variable (y)

Simple Linear Regression
Linear Regression

Type of regression with:

Multiple continuous &/or categorical predictor variables (x)
1 continuous outcome variable (y)

Multiple Linear Regression
Multiple Regression

Type of regression with:

1 continuous or categorical predictor variable (x)
1 dichotomous outcome variable (y)

Simple Logistic Regression
Logistic Regression

Type of regression with:

Multiple continuous &/or categorical predictor variable (x)
1 dichotomous outcome variable (y)

Multiple Logistic Regression

Type of regression with:

1+ continuous &/or categorical predictor variable(s) (x)
1 nominal outcome variable with 3 or more categories (y)

Multinomial Logistic Regression

Statistical method which uses fixed/variable intervals to tabulate time until an event is experienced with cumulative event-free probabilities (Survival Analysis)

Life Tables

Statistical method which uses exact times until an event is experienced with cumulative proportions recalculated every time an event occurs (Survival Analysis)

Kaplan Meier Curve

Test comparing:

2+ treatment groups for a statistically significant difference in survival rates(Kaplan Meier-Survival Analysis)

log-rank test (Mantel-Cox test)

2 groups(out of 3+) after log-rank test determined a difference in survival rates

Post-hoc log-rank tests

Statistical method which allows the investigators to predict relationship between explanatory variable(s) (x) and time to an event as an outcome (y) (Survival Analysis)

Cox Proportional Hazards Model

Measure of association provided by Cox Regression

Hazard Ratio

Description of unexplained outcomes to treatment with detailed descriptions of patient characteristics, in an individual patient

Case Report

Description of unexplained outcomes to treatment with detailed descriptions of patient characteristics, in a few patients with the same disease, exposure, or outcome.

Case Series

Benefits:

• Identify new Diseases, drug Side Effects, new drug Uses

• Descriptive, generates hypotheses for other study designs

Limitations

• Definitive associations/causal effects cannot be drawn between the exposure and outcome from such few cases

Case Reports & Series

Prevalence of health information either exposure or outcome at a single point in time in a defined population

Cross-Sectional Studies
Prevalence studies

Benefits:

• Mostly descriptive, but can be used to evaluate associations between exposures and outcomes, generating hypotheses for future studies

• Valuable in assessing health needs/outcomes of populations and comparing them to others.

• Relatively inexpensive & easy to conduct

Limitations

• associations may not be reliable because of assessing both exposure and outcome at a single point in time

• Cannot establish temporality of events

• Do not establish causality

Cross-Sectional

Compare disease(cases) to matched individuals without the disease (controls)

Reviews retrospective data to evaluate relationship between outcome and predictor

Case-control

Benefits

• Data easy to obtain (eMR/databases)

• Useful when RCT is unethical

• Better for rare outcomes

• Less expensive than RCT and prospective cohort

Limitations

• Associations (No causality)

• Matching is not perfect → confounders

• Recall bias (cases better at recalling than controls)

Case control

Follow a group of exposed individuals to matched individuals not exposed to see if they develop an outcome of interest

Prospective or Retrospective

Cohort

Benefits

• Useful when intervention is unethical or when exposure is rare

• Time sequence of events (exposure→ outcome) can be determined

• Less expensive than RCTs

Limitations

• Prospective → expensive & time-consuming

• Prospective → Loss to follow-up is a possibility

• Influenced by confounders

Cohort

Design for establishing causality

Compare experimental treatment to existing/placebo to determine superiority, equivalence, or noninferiority

Sampled with specific inclusion, & exclusion criteria

Randomized to study treatment groups

Subjects are usually blinded (potentially investigators & analysts too)

Randomized Controlled Trials

Benefits

• Randomization ensures equal chance of intervention, and have similar characteristics → minimizes selection bias & confounding (compared to observational)

• Best design of cause-effect between intervention & outcome

Limitations

• Time-consuming & Expensive

• Not for unethical interventions or lack thereof with randomization

• Not for rare outcomes

• May not be real-world (external validity issues due to restrictive exclusion criteria)

Randomized Controlled Trials

Summary of clinical literature - specific topic like treatment options for a condition or side effects of a drug

Begins with a question and systematic literature search and presented in an unbiased, balance summary

Systematic review

Benefits

• Inexpensive since studies already exist

• Several studies included provide stronger evidence

Limitations

• Summary without pooled statistical estimate

Systematic Review

Combine results from multiple studies & provide a pooled single statistical estimate of the effect size or magnitude of treatment effect from all reviewed studies

Meta Analyses

Benefits

• Smaller studies can be pooled instead of 1 large expensive study

  • Pooled = greater statistical power

Limitations

• Studies may not be homogenous

  • design, exposure, outcome, sample size, inclusion/exclusion criteria

• Validity can be compromised if lower quality is weighted equally to higher quality studies

Meta Analyses

If a case is better able to remember details of their exposures this bias is…

Recall bias.

Disease rates, or exposures made on group of people (house, hospital, countries)

Benefits

Useful to compare health in different populations

Generating questions for future hypotheses

Ecological