Biostatistics Basics: Part I Leroy R. Thacker, PhD Associate Professor Schools of Nursing and Medicine.

Biostatistics Basics: Part ILeroy R. Thacker, PhDAssociate Professor

Schools of Nursing and Medicine

Outline

• Characteristics of a good hypothesis• Hypotheses• Underlying statistical principals

Preliminary Thoughts

• First steps of a research study include:– Identifying a measurable/testable hypothesis– Choosing the appropriate study design– Choosing the appropriate set of variables– Conducting a power/sample size study

• Notice that all of these steps would go better/smoother with the assistance of a biostatistician. So consult with your friendly biostatistician early in the study planning process

Funny Thought/Setting the Tone

"A judicious man looks on statistics not to get knowledge, but to save himself from having ignorance foisted on him." Thomas Carlyle

Characteristics of a Good Hypothesis

• Topic 1


• “A good hypothesis must be based on a good research question. It should also be simple, specific and stated in advance.”

• Plan your analysis based upon your hypothesis AND the types of data you will be working with

• Keep the PICOT criteria in mind when making your hypothesis– This is especially important should you move

into funded research


• Population– What specific patient population are you interested in?

• Intervention– What is your investigational intervention?

• Comparison Group– What is the main alternative to compare with the intervention?

• Outcome of interest– What do you intend to accomplish, measure, improve or affect?

• Time– What is the appropriate follow-up time to assess the outcome?

(This is optional)


• Hypothesis – “Daily eye of newt improves outcomes for high-risk cardiac patients”

• Not bad, in fact I see A LOT of hypotheses just like this– Do you hear a “But” in this statement? If not,

you should!• Why do you think I might have issue with this

hypothesis based on the PICOT criteria in the last slide?


• Is the population defined?• Is the intervention defined?• Is the comparison group defined?• Is the outcome defined?• Is the time defined?

– This is not always done, so don’t get bent out of shape if it isn’t there


• Other considerations:– What does “improve mean”?

• Clinically significant vs. statistically significant– How much change will occur?– Be specific


• Improved Final Hypothesis: “For high risk cardiac patients, daily 50mg eye of newt will produce a reduction in resting systolic BP, measured 7 days post-initiation of treatment of 20mmHg greater than seen in the placebo group”

• THIS is a good hypothesis, one that a biostatistician (or you with the correct software or formulas) could use to come up with a sample size/power calculation


• It is a simple hypothesis– A simple hypothesis contains one predictor

and one outcome variable• It is a specific hypothesis

– A specific hypothesis leaves no ambiguity about the subjects, variables or how the tests of statistical significance will be applied

• It is (I assume) being made BEFORE the start of the study


• A simple hypothesis is easier to work with than a complex hypothesis– In contrast to a simple hypothesis which has

one predictor and one outcome variable, a complex hypothesis will have more than one predictor variable and one dependent variable or one predictor variable and more than dependent variables

• A complex hypothesis is usually broken down into multiple simple hypotheses

Hypotheses

• Topic 2

Hypotheses

• Research questions typically come in two forms– Is there a difference between…?– Is there a relationship between…?

• Questions about differences involve comparing some characteristics between groups

• Questions about relationships involve assessing the relationship/association of one characteristic with another

Hypotheses• Statistical hypotheses are phrased in a very

specific way• The process begins by creating the null

hypothesis from the research hypothesis• The null hypothesis is a hypothesis of “no

difference” or “no relationship”– The null hypothesis will become the formal

basis for testing statistical significance when your study ends and will help estimate the probability that your observed results may be due to chance

Hypotheses

• You will also need to create the alternative hypothesis

• This is a hypothesis that states that there is a difference or there is an association– The alternative hypothesis matches the

research hypothesis if you are doing inference• The possible outcomes are “Reject the null

hypothesis” or “Fail to reject the null hypothesis”– Semantics – You NEVER accept the null

hypothesis

Hypotheses

• Alternative hypotheses are either One-sided or Two-sided

• A one-sided alternative implies some kind of directionality while a two-sided alternative simply implies there is a difference, one way or the other

• The most conservative thing to do (and what most biostatisticians will do) is to use a two sided alternative

Underlying Statistical Principals

• Topic 3

Type 1 Error, Type II Error and PowerNull Hypothesis

True State of NatureAlternative Hypothesis

True State of Nature

Fail to reject theNull Hypothesis

Correct Decision(1-α)

True Negative

Type II Error(Denoted by β)

False Negative

Reject theNull Hypothesis

Type I Error(Denoted by α)

(Size of the test)False Positive

Correct Decision(Power = 1 – β)

True Positive

Effect Size

• The likelihood that a study will be able to detect a difference or an association in a sample depends upon the magnitude of the difference/association in the population

• But, we don’t know the magnitude of the difference/association…we are trying to estimate it with the study!

• When planning the study, the investigator must determine “size” of the difference/association they expect to see or wish to determine

Effect Size

• This expected/desired difference/association is called the effect size

• The effect size may be determined by published data, pilot data the investigator has access to or by investigator discretion– Rank those three methods in order of

preference• Many studies produce multiple effect sizes

because of multiple measures

α, β and Power• In the planning stages of a study, the

investigator establishes the maximum chance that they will tolerate of making a Type I error (false positive; α) and a Type II error (false negative; β)– Power is defined as 1-β and can be thought of as “doing what

you ant to do”, that is, rejecting the null hypothesis when it is false

• Most studies set a significance level of 0.05 and a power of 0.80; there is nothing magical about these numbers– But, if you use other values than these you will be criticized!

Activities• Scenario 1: An investigator is interested in looking at the association

between suicide rates in patients with chronic pain and the effectiveness of their pain management. Draft a “good” research hypothesis using the PICOT criteria, state the null and alternative hypotheses based on the research hypothesis, propose a study design and sampling scheme to address the investigator’s question.

• Scenario 2: An investigator is interested in seeing if using high dose chemotherapy and autologous stem cell transplants as salvage therapy for primary mediastinal non-seminoma germ cell tumors will improve outcomes. Draft a “good” research hypothesis using the PICOT criteria and state both the null and alternative hypotheses based on the research hypothesis. As the investigator presented the question, do you think this is a simple or a complex hypothesis?

Summary• When developing a research hypothesis use the

PICOT criteria• Make your hypothesis simple (if possible),

specific, and make it BEFORE you collect your data

• Do your homework; read the literature and talk to your mentors to determine effect sizes for your studies

• Oh yeah, talk to a biostatistician early!

Questions?

Biostatistics Basics: Part I Leroy R. Thacker, PhD Associate Professor Schools of Nursing and Medicine.

Documents