Biostats test

Mean preferred for what kind of data

continuous data that is normally distributed

Median preferred for what kind of data

Ordinal or continuous data that is NOT normally distributed

Mode preferred for what kind of data

Nominal data

Descriptive statistic

present, organize, and summarize data in a basic sense - “One dimensional”

Inferential statisitcs

test for differences or relationships within a study sample - “result” or “outcome”

What would show data set normally distributed

Mean, median and mode are the same

Discrete data include

Nominal and ordinal

Nominal data

2 or more mutually exclusive groups (yes/no, dead/alive, event/no event)

Ordinal

intrinsic order but no equal interval in between (pain scales, NYHA HF class, CHA2DS2-VASc)

Continuous data

Interval and ratio

Interval

measurable data with equal interval WITHOUT absolute zero (BMI, IQ, CrCl)

Ratio

measurable data with equal interval WITH absolute zero (height, weight, HR, BP, length of stay)

Null hypothesis

there is no difference between groups

What is alpha value?

We accept % chance of the results observed are due to chance (false positive)

Confidence interval for ratios (HR, Odds ratios)

If it does NOT include 1, then it is statistically significant

Confidence internval for continuous values

If it does not include 0, then it is statistically significant

Ordinal

Nominal

C

D

Descriptive epidemiologic

Describe distribution of disease in the population

Experimental epidemiology

Planned experiment to confirm causal relationship

Analytical epidemiology

Relationship between exposure and disease

How is RCT different from observational study

RCT: measure efficacy under ideal and controlled circumstances - drug causes outcomes

Observational: real world circumstances, effectiveness - drug is related to outcomes

Why do we need observational studies

Better generalization: with more population study and better size, more indications, longer follow-up, cheap

Regulatory, marketing, legal and clinical purpose

Supplement data from premarketing study: new patient population, DDI, drug-disease, relative to other drug used for the same indication, undetected benefits and adverse effects, drug utilization patterns, overdose, economic implications

What kind of experimental setting is observational studies can use

Cohort and case-study

Prevalence

prevalent case = both new and pre-existing

Incidence risk

New cases (no exposure, no outcome)/at risk

Incidence rate

Odd ratio

How to measure ratio effects

Risk ratio, rate ratio, odd ratio, hazard ratio

Risk ratio

If RD (risk difference) =0 : no difference (same in both groups)

• If RD >0 : (+) risk in the exposed is greater than risk in the unexposed

• If RD <0 : (1) risk in the exposed is lower than risk in the unexposed

Rate ratio

CUMMULATIVE event rate

Hazard ratio

instantaneous event rate, i.e., the probability of event at a particular given point in time

HR=1 (no association)

HR >1 (increased risk)

HR <1 (decreased risk)

Number needed to treat (NNT)

Number of patients needed to treat to gain additional beneficial effects of a drug intervention

NNT= 1/Risk difference or 1/Rate Difference

What is a cohort?

Well-defined group of subjects

• Followed an at-risk population over time from exposure

Closed cohort = fixed cohort

No subjects are added after cohort is formed

Everyone (A-G) starts at the same time (t0) and is followed until t1 unless an event occurs.

Open Cohort - dynamic

Subjects can enter and exit; study population is constantly changing.

Individuals in the cohort start at a different time

Individuals in the cohort leave at a different time

When is a cohort study warranted?

When the (alleged) exposure is known

• When exposure is rare and incidence of disease among the exposed is high

• When adequate funding for prospective cohorts is available

• When the time between exposure and disease is relatively short (induction and latent)

Drug exposure definition

Exposure is always assessed before outcome

• Unit of analysis

o At the patient level if drug exposure occurs at a single point in time (e.g., use during hospital stay)

o At the patent-time level if drug exposure status changes over time

Exposure during induction/latent period should NOT be included - additional exposure doesn’t affect the outcome

Users with prevalent exposure

patients who have been taking drugs before they enter the study

Users with incident (new) exposure:

patients who initiate drugs at study entry

Problem with prevalent user design

• Underestimate outcome events

o New users who have early events are automatically removed before cohort entry under prevalent user design

o Prevalent users are “survivors” of the early period of drug therapy,thereby having a lower risk, compared to new users

cons of cohort studies

More expensive, inefficient for rare outcome, losses to f/u

Case-control study

Cases and controls are defined based on whether patients have the outcome of interest

Retrospective

only one outcome is studied

When is a case-control study warranted?

• Disease being studied is rare (prevalence < 10%)

• Exposure data are difficult or expensive to obtain

• Little is known about the disease

• Time between exposure and disease (i.e., induction and latent period) is long

How are cases defined and selected

• Cases should represent the target (source) population with the disease of interest

• Cases should be well-defined before measuring exposure

• Cases should be incident cases rather than prevalent cases

Select controls for case

Controls are subjects without the disease of interest

• Controls should come from the same population from which cases arise

• Controls should be selected similar to cases on key factors (e.g., gender and age)

• Selection of controls should be independent of the key exposure

Types of controls

• Convenient control: Hospital/clinic controls, Neighborhood or family/friends controls

• Population control: Random sample

Hospital-based controls criteria

Controls are individuals without the disease of interest

Controls are unrelated to exposure being studied

Controls have similar referral patterns as the cases

Population-based controls

What is matching in case-control studies?

Make cases and controls similar to each other in variables

• Variables matched are key variables associated with exposure and outcome (e.g., age and gender)

Matching increases statistical precision

o Matching leads to a narrower confidence interval (CI) of the association estimate, ensures similar numbers of cases and controls in confounder strata

Potential limitations of matching

Sources for exposure measures

Can incidence or absolute risk/rate be calculated in case-control?

No; only odds ratio can be calculated

Cons of case control studie

Case-control studies increase the chances for introducing bias

• Case-control studies cannot calculate incidence and absolute risk or rate

• Case-control studies cannot do survival analysis (not follow patients over time)

• Case-control studies cannot evaluate multiple outcomes simultaneously

• Selection of suitable controls is difficult in case-control studies

Surrogate

a biomarker that is used as a substitute for a clinical endpoint that predicts trial outcome

Clinical outcome

a characteristic or variable that reflects how a patient feels, functions, or how long a patient survives

Composite outcome

a group of clinical outcomes (usually morbidity and mortality-related) that are combined into a single analysis to increase power and ability to find a difference

Statistically significant vs. clinically significant

Statistically significant: the difference that exists between study groups is unlikely

to be explained by chance

• Clinically significant: the difference that exists between study groups is substantial

enough to be clinically useful

Null Hypothesis of Non-inferiority

New treatment is not noninferior to active control by specified noninferiority margin

Confounding Factors

Guidance for patients on any lifestyle aspects they should keep constant

Post Hoc Analyses

All analysis methods should be defined a priori

• Post-hoc analyses should be specified as such and interpreted as correlation, not causation