BioSTATSsh1

Question 1

Definitions:

Mean

Answer 1

Mean : most commonly used. Used for continuous data.
Continuous data is data that is measurable and can be broken down into intervals. Mean is used for measure central tendency. Mean is reliable unless there is extreme outlier.

Median: middle value in the set of data. Median is generally used for ordinal data. Or ranked data. Type of data that has order, classes, and rank. Eg) CHF classifications. Class 1, 2,3, 4 : order and ranked.
- if its a set of numbers grab middle. If even numbers take middle two and get average

Mode: most frequently. Set of number if 1 number occurred more times than others youre getting that number for modes.

Range: interval between the smallest number and largest number and that whole distribution is called the variance
Variance: measure of the data spread or distribution.

Standard deviation ( SD ): square root of variance

Question 2

Bell shaped distribution / normal distribution

Answer 2

When you collect your continuous data and plot them in a graph, majority of the times you’ll end up with a bell shaped graph that looks bell shaped. Mean is right in the middle of the bell. Standard deviation is the deviation from the mean. -1 SD to + 1 SD deviation is 68% of the data. -2SD to +2 SD covers 95% of population.
-half of that : -2SD to mean is half of 95% so that’s 47.5%

Now look at 47.5 % : 34 % of it is from mean to + 1 standard deviation and leftover is 13.5%

Now look at leftover after 95% taken. To the left 2.5 % and to the right is 2.5 %

Question 3

Answer 3

Question 4

Relative risk, relative & absolute risk reduction, NNT ( number need to treat or harm ) & odds ratio

Answer 4

This is the most important part of biostatistics section.

We need to be able to describe the strength of correlation. Strength of correlation between two things. Is it a little or significantly.

Eg) control group had 15% event and exp had 10 % . Difference is 5%. Absolute risk reduction.

Absolute risk reduction: AAR= CER- EER = 15%-10% = 5%

Question 5

Relative risk

Answer 5

What is the baseline risk?

We use relative risk:

Ratio: event rate /Control event rate = 10%/15% = 0.67 or 67%

RRR= 1-RR = 1-0.67 = 0.33 ( or 33% )

The event rate in the aspirin was .67 times the event rate in the controlled group.

How much did it reduce risk: RRR relative risk reduction = 33%

Question 6

Relative risk reduction example

Answer 6

Shopping example
Item is usually 100 dollars. Today it’s on sale for 80 dollars. Today’s price is 0.8 times the original price. Or you could say 20% off. 1- 0.8 = .2 or 20%
Reduced it by 20%

Question 7

NNT or NNH
Number need to treat/harm

Answer 7

NNT = 1 / AAR

1/5% = 20

5% instead of 5/100 ( 5% ) , inverse is 100/5 or 20

20 people needed to be treated for 1 % to benefit

Why do we have number needed to treat. It’s a cost issue. If you have a medication, and NNT is 500. It means you have to treat 500 people for 1% to benefit.
- if it’s an expensive medication then maybe the small benefit you get out of it it wont be worth it.

Generally NNT less than 50 considered valuable

Question 8

Odds ratio:

Answer 8

∆ in likelihood ( odds ) of either increase or reduction of the likelihood of the event occurring in the experimental group compared to my control group. OR strength of correlation between two tested groups.

EER = experimental event rate / ENR =experimental non event rate
Controlled event rate/ controlled non event rate

ODDs ratio ( OR ) = [ EER/ ENR ] / [ CER / CNR] = EER x CNR / ENR x CER = a x d/ b x c = 10% x 85% / 90% x 15% = 0.63
- change in the likelihood occurring

0.63 means likelihood of the MI occurring in the aspirin group is 0.63 times the odds of it occurring in the control group

So how much have i reduced this likelihood, if odds of it occurring is 100% -63% = 37% percent

Eg odds ratio is 1.0 then it means numerator on top and denominator on bottom is the same.
Numerator on top was all the experimental data. Denominator was all the controlled group data
Odds ratio of 1.0 means no change in the likelihood of event occurring as a result of taking medication.

If odds ratio> 1.0 then it means the likelihood of event occurring is now more.
If odds ratio < 1.0 then it means you reduced the likelihood of the event.

Question 9

Odds ratio example

Answer 9

If OR 1.3 then it means the likelihood of the event occurring in the experimental group is 1.3 times the likelihood of it occurring in the controlled group.
-how much have I increased this likelihood? You’ve increased it by 30%. Make OR into percentages 1.3 = 130% / 100% , 130%-100% = 30% you’ve increased it by 30%
- how about if you say odds ratio is 2.0. Likelihood of event occurring in experimental group is 2 times more that the rate or occurrence in control group. So 2.0 =200%/100% . Increase likelihood by 100%
-OR 4.0 increased likelihood of event 400%/100% = 300 % increase in likelihood

Question 10

AAR

Answer 10

Absolute risk reduction
The difference between the two

10-6 = 4%

Question 11

NNT

Answer 11

Number needed to treat
Inverse the absolute risk reduction

Instead of 4% just say 100/4 = 25

Question 12

Relative risk

Answer 12

A ratio
EER/ CCR

Say EER =6% / CER = 4% = . 6 or 60%

Question 13

Relative risk reduction

Answer 13

1- RR
1- 0.6 = 0.4 or 40%

Drug reduced the risk by 40% in duration of the study

Question 14

NNT for 1 year

Answer 14

NNT for 1 year = NNT of the study x Duration of study in years
NNT = 25 patients x 5 years = 125

Question 15

2 types of data
Discrete and continuous

Answer 15

Discrete: whole numbers, cannot be divided. Number of people 32 or 33 no 32.5

Continuous data: type of data. Measurable and can broken down into intervals. You can use decimal points to break them down into intervals
-eg) blood test. Usually if they give you levels, it’s continuous data. HgBA1C of 7 or 8 or 7.1, 7.2, or 7.3 etc.

Question 16

Subtypes of discrete data
Nominal
Ordinal

Answer 16

Nominal ( categorical ) data: type of discrete data. Dividing your data into separate groups. Nominal comes from name. So you can give names to each category.
-eg) group of people who had stroke vs didn’t : yes stroke or no stroke

Ordinal ( ranked ) data: collecting data on people pain scale. Say pain scale from 1 to 10. It has an order. Ranked format.

Both of these are discrete data. Means it cannot be broken down into decimals

Question 17

Continuous data subtypes
Interval data
Ratio data

Answer 17

Interval data: you can break your data into measurable equal intervals. Example drug level 12.5 or 12.6

Ratio data: equal measurable intervals and absolute zero

Difference between interval data vs ratio data. Ratio data has an absolute zero.
Weight in pounds that’s continuous. But it has an absolute zero. Zero weight means no weight.
What is temperature. Zero temperature doesn’t mean no temperature. So temperature is not a ratio data .
Zero drug concentration means no concentration. So thats ratio data. You cant have negative concentration so thats ratio data.

Question 18

Nominal data

Answer 18

A subtypes of discrete
Nominal is discrete categories with NO order or rank

The categorical

Question 19

Ordinal data

Answer 19

A subtype of discrete data.
Ordinal data is discrete categories with order and rank

Could be pain scale, HF class, colon cancer stages

Question 20

Interval data

Answer 20

A subtype of continuous data.
Interval is equal measurable intervals

Question 21

Ratio data

Answer 21

A subtype of continuous data

It’s equal, measurable intervals and absolute zero

No negative numbers

Question 22

Statistical tests

Answer 22

The type of test you choose depends on 3 things
1) what type of data? 3 major ( continuous, nominal, or Ordinal )
2) how many groups do have? How many groups are you comparing. 2 groups or more than that?
3) are the groups independent or are they paired ( matched )

Independent group: sample a group of patients and they take exp drug and other group takes placebo. They’re separated and you check data ( event rate )
Paired data: group is only 1 group but they act as their own control. So when you check you data, at one point the group will act as placebo, and then later, the same group of people will take experimental medication and you test the group before and after. They act as their own control

Question 23

Continuous data tests

Answer 23

T test: used with continuous data when the means of two groups are being compared. Comparing the means of 2 different groups. Eg) 2 groups cholesterol levels.
-if the 2 groups are independent and separate - we call that student’s t-test
- if the 2 groups are paired ( test group act as its own control ) then we called it - Paired t-test

Eg) students t test- new diabetes med, determining amount of HgBA1C reduction in 2 separate independent groups.
Eg2) what if with same med, now youre testing hgbA1c before and after treatment. Once you give the placebo, then 3 months later, you give the experimental drug and then check A1C levels again. Same group acting as their own control. -paired

If you have 3 or more groups means being compared then the name of that test is ANOVA ( analysis of variance ).
- continuous data but you want 3 more groups. ANOVA used whether they are independent or paired.

Question 24

Test for nominal categorical data
- CHI SQUARE TEST

Answer 24

2 separate categories. Yes MI or No MI. If you only have 2 groups, sometimes they call it categorical dichotomous groups.

When you want to use nominal categorical data the main test is CHI- SQUARE test
So aspirin vs placebo , event is GI bleed so EER vs CER and ENR vs CNR

What if testing GI bleed of another medication like plavix. You can plavix vs placebo and Asa vs plavix
- all groups are independent

Question 25

McNemar’s Test

Answer 25

This is nominal categorical data
If the subjects are used as their own control then chi square test is called mcnemars test

Similar to chi square test except instead of having independent groups you have two paired groups

If you have 3 or more paired groups then the name changes to Cochran’s Q test

Question 26

3 tests for nominal categorical data

Answer 26

1) Chi Squared: nominal categorical data and 2 or more independent groups
2) McNemar’s test: nominal categorical data and 2 paired groups
3) Cochrans Q test: nominal categorical data with 3 or more paired groups then you have Cochran’s Q tests

Question 27

Ordinal ( ranked ) data tests
4 tests

Answer 27

If your ordinal data ( pain scale or HF classes ) - data of this kind - if 2 independent groups we use Wilcoxon Rank Sum test

2) if 2 paired groups with ordinal ranked data, then the test is Wilcoxon signed Rank test

3) Kruskal - Wallis Test: ordinal data of ≥ 3 or more independent groups are being compared

4) Friedman test : ordinal data of 3≥ paired groups are being compared being compared

Question 28

Answer 28

Question 29

Research study designs

Answer 29

1) Randomized controlled trials ( RCT ) - a high quality statistical test
Randomized: want the 2 groups to be randomized. Eg) testing effect of heart failure medication on patients. It wouldn’t be fair if you put all HF patients that are 50 in one group and all HF patients that are 80 y/o in other group and then check the effect of a medication vs placebo. It would be better if groups were randomized
-control vs experimental group

2) Cohort studies: clinical treatment trials. cohort means group of people that share a trait. Eg) group of people with intermittent asthma. Take cohort , randomize samples on both sides, expose them to medication that were testing, and then start them at this point in time, and follow them throughout months or years to see the effect of medication. Going forward, A prospective study.

3) Case control study:
Vioxx ( same class as celebrex a cox2 inhibitor )
Same cardiologist started suspecting maybe this drug causes more heart attacks than general population.
So they designed a study: ( it wouldn’t be ethical a prospective study on this. Can’t say take vioxx now and we’ll follow you throughout the years and see if you get a heart attack ) so they went backwards. Looked at patients who had a heart attack and people who did not have a heart attack, and compared their exposure status with regards to vioxx.
Cases of people who had MI and control group ( people who had not had the outcome ).

In case control study when you use term control it is different than when you say control in cohort study. In cohort study when you say control it means people who were taking placebo. In case control study when you say control it means people who haven’t had the outcome ( people who have not had the MI ) so cases with MI and cases with no MI and lets see if vioxx was a factor in this.

Question 30

Case control study are quicker and cheaper bc data is already there, don’t have to follow people throughout the years.

Answer 30

T

Question 31

Categories of study designs

Cross sectional study:

Cross over study :

Intent to treat analysis vs Per Protocol analysis

Answer 31

Cross sectional study: You check the event rate in one cross section of time. When were are checking prevalence of a disease. We’re not going forward or backwards, just one cross section of time.

Cross over study: a paired group. They act as their own control group. First they take placebo then they crossover to experimental group and take drug. Then they take drugs and we check event rate.
- when groups are paired or matched.

Intent to treat analysis vs Per Protocol analysis :
•Intent to treat : takes data and takes into account all the data from all patients who started this study regardless of whether they finished it or whether they were compliant it doesn’t matter. They had the intention to do this. - intent to treat analysis.
• Per protocol analysis: collects data and takes into account data only from people who finished the study according to the protocol. They were compliant.

Benefit to each:
Intent to treat: people not being compliant is closer to real life scenario.
Per protocol study: they actually give you a better more accurate estimate of the effectiveness of the medication.

Question 32

Meta analysis / systematic review

Answer 32

When we get results from many different but relevant studies and we try to put all the results together to come up with a practical conclusion. In order to do this, we have to choose high quality studies ( want randomized control ).

Next, you want to avoid what is called a publication bias when you are choosing your study. What is publication bias? ( if at the end of study, it doesn’t find a correlation or difference, it not’s exciting so chances are it wont get published and study will be difficult to find. -publication bias). When you are deciding which studies to put in your meta analysis, you want to make sure you find even the ones that were not published and get an average of all of those. We get the individual studies and we find the odds ratio of each individual study and plot on a graph. The odds ratio of each individual study. Two lines going to the side of them is the confidence interval. And the bottom line is the composite ratio ( like an average odds ratio of all the studies you found. Put them together, and this was the conclusion you found at the end. )

Question 33

Cohort study

Answer 33

Generally a prospective study. Follows a group ( cohort ) of populations of treated and non-treated individuals through the years and then compares their data. - clinical trials of new meds are typically these.

Question 34

Case - control studies

Answer 34

Retrospective, meaning they go back and look at populations with a certain outcome and then compare incidence of this outcome in exposed and non-exposed individuals.

Question 35

Cross-sectional study

Answer 35

Risk factors and health status are compared for a group are studied in one specific point in time ( not going forward and not going backward).

Question 36

Meta analysis study

Answer 36

Results of many different but related studies are compared to find a conclusion.

Question 37

Intent to treat vs per protocol

Answer 37

Everyone who intended to finish the study even the ones who didnt finish vs the subjects who followed the protocol exactly

Question 38

Categorical dichotomous data vs continuous data vs ordinal data vs interval data vs ratio data

Answer 38

Categorical dichotomous data : dividing study subjects into two groups based on occurrence of outcomes : eg ) those who had stroke vs those who did not.

Continuous data: Most collected data that are measured and can come up with a decimal point. More of a continuum: eg) weight in people in a study.

Ordinal data: ordered data. Ranks

Interval data: similar to ordinal except it has exact units and the interval between them is exactly the same. Eg) difference between 42 and 43 degrees same as difference between 68 and 69 degrees

Ratio data: a type of interval data with absolute zero. Eg ) weight. Can’t be negative pounds. Temperature on the other hand can be negative.

Question 39

Power

Answer 39

Power is the ability of a study to find an association.
Power = 1- Beta
Beta is the chance of a type 2 error ( missing an association when one exists)

Lower the beta the higher the power.