Biostatistics

ACCP: Volume 2 Module 6 (Biostatistics & Study Designs). MedEd101: Biostatistics study sheet +/- slide set not sure yet

What is a Null Hypothesis/when it is accepted? (Ho)

There is NO STATISTICALLY SIGNIFICANT DIFFERENCE between the control and the treatment groups

What is an Alternative Hypothesis (aka when it is accepted)? (Ha)

There IS A STATISTICALLY SIGNIFICANT DIFFERENCE between the control and the treatment groups

Aka you reject the null / Ho to accept the alternative / Ha

What does Alpha mean conceptually and when is Alpha created during the trial? (and what is standard alpha)

Alpha is the acceptable margin of error, the threshold for rejecting the Ho (need less than 0.05% error to say there is a difference between the control and treatment group)

Alpha is established BEFORE trial begins, when initially design trial.

alpha traditionally is 0.05 (if not otherwise stated)

What value are you comparing to alpha to determine if you can reject the null hypothesis and accept the alternative (Ha)?

Compare P value (calculated using results of trial) to alpha (hypothetical margin of error determined during trial design) to see if treatment has statistically significant difference versus control

What value(s) are you comparing the confidence interval’s range (CI) to in order to determine if you can reject the null hypothesis (Ho) and accept the alternative (Ha)?

ZERO for Difference type data:

• primary end point’s confidence interval is calculated using the difference btwn the drug and placebo’s data

• if range INCLUDES zero then NOT STAT SIG (-0.38 to 0.88)

ONE for Ratio type data:

• (ex, Relative Risk, Odds Ratio, Hazard Ratio)

• primary end point confidence interval data is calculated by dividing the drug / placebo’s data

• if range INCLUDES one, then NOT STAT SIG (0.229 to 1.776), because 1/1 = 1 and not sig

what equation uses alpha? / how do you calculate alpha?

CI = 1-alpha

What is the Confidence Interval CONCEPTUALLY and CONCEPTUALLY what variable is it tied to? (hint the equation)

CI is our hypothetical error margin but calculated using the data/results from the study. aka The hypothetical range (from the results of our data) that we are 95% confident the true value lies within that range

Ex: It is a range of values calculated using study results. it states you are 95% confident (from 1- alpha (1-0.05 = 0.95)) that the TRUE effect of a medication is within the given range

It is tied to alpha, in fact its the inverse of alpha (pre determined error margin needed to reject Ho)

Another way to conceptually utilize alpha but with the actual data from the study and not hypothetical benchmark like alpha is

Fixx??????? REdudnant??????/

what is the relationship between alpha and confidence interval and how do you calculate it? What type of error is Confidence Interval (CI) related to and how do you calculate CI?

CI is inverse of alpha

CI = 1-alpha

It is related to Type 1 error (you think there is a difference between control and treatment but in reality there is NO DIFFERENCE, aka falsely reject Null Hypothesis)

Example:

The difference in Drug vs placebo in FEV1 is 38 (46-8).
The difference in Drug vs placebo in FEV1/FVC(%) is 0.313 (0.314-0.001).

Which one is statistically significant? (using CI for “difference btwn two variables” data like Means)

Drug (N=745)	Placebo (N=745)	Difference drug - placebo (95% CI)
FEV1 = 46	FEV1 = 8	38 (18 to 58)
FEV1/FVC = 0.314	FEV1/FVC = 0.001	0.313 (-0.26 to 0.89)

How do you determine statistic significance using Confidence Interval without the P value with “difference data” aka MEAN type values

For DIFFERENCE type data, it’s statistically significant when range DOES NOT include 0

• FEV1 CI = 18 to 58 so STAT SIG (range does not cross zero)

• FEV1/FVC* CI = -0.26 to 0.89 so NOT STAT SIG (range CROSSES ZERO)

  • *even though FEV1/FVC is calculated using division, the data we are making a confidence interval is the DIFFERENCE between those two numbers, not dividing those two numbers. so the 0 thing applies here not the 1.

Example:

The Ratio (relative risk) of Drug vs placebo Severe COPD exacerbation is 0.92 (0.11 / 0.12)
The Ratio (relative risk) of Drug vs placebo in Moderate COPD exacerbation is 0.85 (0.94 / 1.11).

Which one is statistically significant? (using CI for “ratio btwn two variables” data like relative risk, odds ratio and hazard ratio data)

Drug (N=745)	Placebo (N=745)	Relative Risk, a RATIO of drug / placebo (95% CI)
Severe Exacerbate = 0.11	Severe Exacerbate = 0.12	0.92 (0.61 to 1.29)
Moderate Exacerbate = 0.94	Moderate Exacerbate = 1.11	0.85 (0.72 to 0.99)

How do you determine statistic significance using Confidence Interval without the P vale with “ratio data” aka relative risk, odds ratio and hazard ratio

For RATIO type data, it’s statistically significant when range DOES NOT include 1

• Severe Exacerbation CI = 0.61 to 1.29 so NOT STAT SIG (range CROSSES ONE)

• Moderate Exacerbation CI = 0.72 to 0.99 so STAT SIG (range does not cross one)

do you want a wide or narrow CI? (DELETE????)

You want a Narrow CI (because smaller margin of error means less variability to data and easier to find the “true” answer within that margin of error)

Wide range = larger dispersion of data, data is less precise and less useful

• Narrow range: 6% to 10% BP lowering

• Wide range: 4% to 68% BP lowering (well is it 4% lower BP or 68%?? thats very wide range)

What is a Type 1 Error? (definition of concept)

The likelihood that Ho is rejected in error (accepting Ha), when the TRUTH is Ho

The likelihood that we THOUGHT there was a difference between the control and treatment groups, but in REALITY there was NO DIFFERENCE between the control/treatment groups, and should reject Ha

The percentage or chance of detecting a difference when one does NOT actually exist, it was random chance

“we thought there was a difference btwn control and treatment groups, but in reality that perceived difference was just random chance! there was no difference after all”

What type of error (type 1 or 2) is P value associated with and explain the corresponding concept of the P value (not just the number)

bonus: definition of that error

Type 1 error (falsely reject Ho, reality = no difference btwn control and tx)

P value = % chance of a type 1 error occurring

The P value is percentage or chance of detecting a difference when one does NOT actually exist, it was random chance,
“the chance that what you found in your study is actually different from reality” MedEd101

Is it better to have a big P value or a small P value?

If big P value what does that conceptually mean (related to type of error) and vice versa with a small number

Better to have a Small Number/percentage

• big number = very high chance study results are different than reality (high type 1 error chance)

• while very low number = very low chance study results are different than reality (low type 1 error chance)

Example: A drug manufacturer has a new beta-blocker (madeuprolol) they are testing. They are comparing its blood pressure lowering ability against placebo. The null hypothesis (Ho) would be that there is no difference between the two groups. The alternative hypothesis (Ha) would be that there is a difference in blood pressure. In their study, the researchers found that madeuprolol significantly lowers blood pressure more than placebo with a p-value of 0.01. (assuming that study was designed for 0.05 as alpha or whatever)

Do we reject or accept the null hypothesis with this P value?

What does the P value mean (conceptually) in this scenario/in regards to type of error? (using the example in the explanation and not using the example in the answer)

P value in study is less than 0.05 (predetermined) so we would REJECT the null hypothesis (there IS a difference between the control and treatment groups)

There is a 1% chance that a TYPE 1 error occurred (we think there is a difference between control and treatment groups but in reality we were tricked! there is no difference)

aka: There is a 1% chance that madeuprolol does NOT lower blood pressure more than placebo.

if type 1 error occurs, how do we fix it?

What is Power CONCEPTUALLY and when is Power created during the trial? (and what is standard power)

(or maybe do beta conceptually here,, might make more sense)

What is a Type 2 Error (definition of concept)

The likelihood that Ho is kept (Ha is NOT accepted) in error, when the TRUTH is Ha

The likelihood that we THOUGHT there was no difference between the control and treatment groups, but in REALITY there WAS A DIFFERENCE between control and treatment, so we should accept Ha instead and reject Ho

What type of error (type 1 or 2) is Beta associated with and explain the corresponding concept of Beta

bonus: definition of that error

Type 2 error (falsely accepting Ho)

The percentage or chance of NOT detecting a difference when one actually exists

Is it better to have a high power or low power?

What is the relationship between beta and power and how do you calculate it?

Power is inverse of beta

Power = 1 - beta

is it better to have a big beta or a small beta?

if there is a type 2 error, how do we fix it?

Example: Researchers are going to study the effect of various NSAIDs on the decline of cognition in the elderly. They will use the BIMS (cognition scale) score and place patients in three groups. Group 1 will be patients on 800 mg daily of ibuprofen, Group 2 will be patients on Naproxen 500 mg daily, and Group 3 will receive no NSAID therapy.

what are the independent variable(s)?

HINT: independent variable is something the scientists can control/are changing in test subjects

The various NSAIDs given to patients: Ibuprofen and Naproxen

(could also be the patient archytypes we choose to enroll in the trial like age, gender etc)

Example: Researchers are going to study the effect of various NSAIDs on the decline of cognition in the elderly. They will use the BIMS (cognition scale) score and place patients in three groups. Group 1 will be patients on 800 mg daily of ibuprofen, Group 2 will be patients on Naproxen 500 mg daily, and Group 3 will receive no NSAID therapy.

what is the DEPENDENT variable?

what changes in response to the independent variable(s): the BIMS/cognition scale

(more often than not it’s the variable we are measuring to try to determine if our drug works)

NEXT STEPS:

• Fix the beta/power definition cards (originally set them up like i phrased the alpha ones but might not work, compare the naplex and meded101 definitions to make them)

• how to fix type 1 and type 2 errors (if there is even a way to fix type 1 error like how for type 2 its just increase patient population)

• types of data like nominal and ordinal and stuff (should be A LOT faster than the earlier stuff in here)

• HR, OR, RRR, etc