ARA0103 A

1
ARA0103 Aðferðafræði Rannsókna
Australian Medical Sheepskin

• Fyrirlestrar 7 og 8
• Experimental design I/Tilraunasnið I

quantitative/megindlegur
2
Hvað er vandamálið?
Dæmi...

• Langtímasjúklingar með legusár.
• bed sores/pressure ulcers

3
What are the variables?Hvað eru breyturnar?

• Independent variables/óháðar breytur
• Support mattress in use.
• Frequency of patient turning.
• Time in hospital.
• Dependent variables/háðar breytur
• Number of pressure ulcers.
• Patient comfort.
• Using a sheepskin mattress cover might make
  patient too hot.

Athugasemd in student projects, start by
considering only one independent variable and one
dependent variable.
4
What is the intervention/inngrip?

• Medical Sheepskin
• http//www.healingfibres.com/
• http//www.medicalsheepskins.com/index.html
• http//www.classicsheepskin.com/medical.htm
• 2x frequency/tíðni of patient turning...
• Menntun...
• If you can move yourself, change position every
  15 minutes.
• Eat properly/Borða rétt
• ...
• Use pressure relieving supports...

olnbogi
ökkli
5
Að bera saman gögn...

• Applying an intervention may be unnecessary if
  you can compare data from different
  wards/deildir, hospitals/sjúkrahús or nursing
  homes/hjúkrunarheimili.

gögn hér
gögn hér
gögn hér
More turning of patients?
Less turning of patients?
6
State the null hypothesis H0Tilgreina
núlltilgátuna H0- áhrif er ekki til -
dæmi

• A sheepskin mattress cover has no effect on the
  number of bed sores.
• Doubling frequency of patient turning has no
  effect on the number of bed sores.
• Educating patients on how to prevent bed sores
  has no effect on the number of bed sores.

Yfirleitt er leitast við að hafna núlltilgátuna.
7
State the alternative hypothesis HaTilgreina hin
tilgátuna H0- áhrif er til -
dæmi

• A sheepskin mattress cover reduces the number of
  bed sores.
• Doubling frequency of patient turning reduces the
  number of bed sores.
• Educating patients on how to prevent bed sores
  reduces the number of bed sores.

Ef þú ert að nota t-próf, til dæmis, vonin er að
segja núlltilgátan er ekki rétt, frekar hin
tilgátan hlýtur að vera rétt.
hin tilgátan aðaltilgátan
8
Select measurement toolsVelja mælitæknar

• Hjúkrunarkona getur lagt saman legusár beint.
• Rannsóknarmaður getur lagt saman legusár að nota
  myndir sjúklings.
• Hjúkrunarkona getur mælt hita sjúklings.
• A questionnaire/spurningalisti can be used to
  assess patient bed comfort.
• Rannsóknarmaður má skrifa sjálfur
• eða nota/breyta spurningalista hjá
  http//www.medalreg.com/,

9
Reliability and ValidityÁreiðanleiki og Réttmæti
mælitæknar

• Measurement tools are reliable if you get the
  same answer by repeating the measurement.
• A nurse might fail to count a Stage I bed sore on
  one ankle which was counted by another nurse.
• Procedures/aðferðir have to be followed to
  measure blood pressure.
• Measurement tools are valid if they measure what
  you want them to measure.
• On a questionnaire, patients may say they are too
  hot, but their temperatures measured using
  thermometers suggest they were not too hot.
• Maybe they completed the questionnaires on a very
  hot sunny day?

10
Is the intervention potent?Er inngrip sterkur?

• A sheepskin mattress cover that is too thin might
  not work.
• A video showing someone die in the final stages
  of smoking-related lung cancer is a more potent
  intervention than an information
  leaflet/bæklingur.
• Toe-grasp training for 30 seconds every month is
  unlikely to improve balance.

11
Pilot study/Forathugun,Forrannsókn

• If possible, conduct a pilot experiment with a
  small number of people beforehand (4-6).
• Pilot subjects should be similar to the subjects
  in the actual experiment.
• Pilot studies check on
• Wording of questionnaires/orðalag spurningalista.
• How often are patients turned?
• Aldrei Stundum Oft Mjög Oft
• Hvað þýða stundum, oft, og mjög oft?
• Unclear instructions/óvissar leiðbeiningar .
• How often should the sheepskin mattress cover be
  washed?

12
Use a control groupNota samanburðarhóp
confounding variable/truflandi breyta

• The control group do not receive the
  intervention.
• From an ethical point of view, the control group
  should receive normal treatment/venuleg meðferð.
• Using a control group is the only way to deal
  with confounding variables.
• age, sex, weight, surgery, mobility, ...
• Patient characteristics in the control group
  should be similar to the intervention group.
• Gengur ekki að bera saman 20-25 ár með 70-75 ár...

13
Randomized Controlled Trial hrein tilraun
eligibility criterion/úrtaksskilgreining
Identify eligible subjects
overview/yfirlit
Obtain informed consent/Fá upplýst samþykki
Þátttakendur
Non-participants
Willing
Unwilling
How many unwilling?
Randomization/Tilviljunarval
between subjects/ milli hópa
Intervention group (I)
Control group (C)
Participants
Drop outs
Þátttakendur
Brottföll
Data collection
Data collection
Safna gögn
Safna gögn
Number of bed sores
Number of bed sores
How many drop outs?
How many drop outs?
14
Randomization/Tilviljunarval

• Ideally, a third party allocates subjects to the
  control (C) or intervention (I).
• This is called allocation concealment.
• Eliminates experimenter bias/skekkja.
• An experimenter could allocate younger patients
  to the sheepskin group (I).
• An experimenter could allocate only
  post-operative patients to the control group (C).
• Subjects can be given a number and random numbers
  can be selected from a table of random numbers or
  using a software tool.

http//www.graphpad.com/quickcalcs/randomize1.cfm
15
(No Transcript)
16
Randomized Controlled Trial (RCT)
C
I
At some point we begin to believe there is a real
difference between the two distributions
mismunurinn er til ( t-próf, p-gildi minna en
0,01). Sheepskin mattress covers reduce the
number of bed sores on average.
17
Within subjects design/innanhópasnið
sample size/úrtaksstærð

• Single subject/einn þátttakandi

úrtaksstærð n1
AB design/snið. Leggja saman legusár. Fyrr (A)
og á eftir (B). Ekki hægt að nota tölfræðipróf,
svo er ekki hægt að álykta um þýði. Líka ABAB,
ABABAB snið.
A
B

• An individual patient may be so unique and in
  need of such help, that only they receive the
  intervention.

Dr House...
18
Within subjects design/ innanhópasniðimbalanced/ó
jafnvægt
n10
Subject mars (C) apríl (I)
S1 control sheepskin
S2 control sheepskin
S3 control sheepskin
S4 control sheepskin
S5 control sheepskin
S6 control sheepskin
S7 control sheepskin
S8 control sheepskin
S9 control sheepskin
S10 control sheepskin
paired or related or dependent data -gt nota háð
t-próf
safna gögn lok mars og lok apríl
19
What if ?/Hvað ef?

• During the control period, there are nursing
  staff shortages and patients are turned only 50
  of the time.
• This would effect the measurements of bed sores
  at the end of the control period. (skekkja)
• This within subjects design is said to be
  imbalanced and is fatally flawed.
• gengur ekki að nota slíkt snið

20
Within subjects design/ innanhópasniðbalanced/í
jafnvægi
cross-over
n10
Subject Mars (C) Apríl (I)
S1 control sheepskin
S2 sheepskin control
S3 control sheepskin
S4 sheepskin control
S5 control sheepskin
S6 sheepskin control
S7 control sheepskin
S8 sheepskin control
S9 control sheepskin
S10 sheepskin control
safna gögn lok mars og lok apríl
21
Within subjects design/ innanhópasnið

• By using the same subjects, there is better
  control of variability between subjects.
• Sometimes it is impossible for subjects to
  receive both the control and the intervention.
• Post-operative patients can be measured only
  once. They are no longer post-operative after
  receiving either the control or the intervention.

post-operative/eftiraðgerðar-
22
Within subjects design/ innanhópasnið

• Sometimes the cross-over design leads to ethical
  problems.
• If the intervention works really well for an
  individual patient, should they be taken of the
  intervention and put on the control ?
• Patients are no longer blinded.
• They can realise that the use of the sheepskin
  matress cover is the intervention.
• Sometimes the cross-over itself is imbalanced.
• Going from a sheepskin to a normal mattress might
  not be the same as going from a normal mattress
  to a sheepskin.
• Sometimes a patients health status can change
  over time.
• Becoming post-operative as you are about to
  cross-over between control (C) and interventiion
  (I).
• Data for such patients may have to be left out.
  (brottfall)

23
Between subjects design/Millihópasnið.
Dæmi Does the diastolic blood pressure between
men and women differ? Men 76, 76, 74, 70,
80, 68, 90, 70, 76, 80, 68, 72, 96, 80, 90,
72 Women 76, 70, 82, 90, 68, 60, 62, 60, 62, 72,
68, 80, 74 Calculate the results of an
independent t-test. (2-tailed, unequal
variances) What significant level is reached?
__________ (marktektarstig) What is the size of
the effect? __________ (stærð
áhrifa)
Null hypothesis/Nulltilgátan There is no
difference in diastolic blood pressure between
men and women.
24
It is better to have equal sample sizes.
t-Test Two-Sample Assuming Unequal Variances t-Test Two-Sample Assuming Unequal Variances t-Test Two-Sample Assuming Unequal Variances

  Variable 1 Variable 2
Mean 77,375 71,07692308
Variance 69,71666667 85,07692308
Observations 16 13
Hypothesized Mean Difference 0 null hypothesis
df 25
t Stat 1,907486345 the value of t
P(Tltt) one-tail 0,034004546
t Critical one-tail 1,708140745
P(Tltt) two-tail 0,068009092 p- gildi
t Critical two-tail 2,059538536  
The value of p needs to be less than 0,05 for
statistical significance. The smaller p is, the
greater the statistical significance.
25
More subjects?
power of a statistical test/styrkur tölfræðiprófs

• p gildi er ekki minna en 0,05 en er lítið (0,07).
• The difference between means is 6.
• If there were more subjects, it might be possible
  to detect a real difference.
• A negative result in a statistical test might be
  because you do not have enough subjects. The
  statistical test might not have enough
  statistical power to detect the difference
  between means.
• What if we have another set of measurements of
  diastolic blood pressure that are similar to the
  first?
• Is it possible to obtain statistical
  significance?
• Er hægt að ná tölfræðileg marktekt

26
How many subjects?
power of a statistical test/styrkur tölfræðiprófs

• Ef þú ert að reikna út meðaltal og staðalfrávik,
  úrtaksstærðin hlytur að vera gt10.
• ngt20 is better
• ngt30 is even better.
• The greater the number of subjects the greater
  the statistical power.

27
Standard error of the mean/Staðalvilla

• The standard deviation of the sampling
  distribution of sample means is called the
  standard error of the mean.
• As n increases, the standard error of the mean
  decreases.

28
RVLS Sampling Distributions
s5,00
S2,27
n5
29
RVLS Sampling Distributions
s5,00
S1,54
n10
30
RVLS Sampling Distributions
s5,00
n25
S1,02
31
Central Limit Theorem/Markgildissetningin
skewed/skekkt
parametric test/ stikapróf
non-parametric test/ stikalaust próf
n25

• Even if the population distribution is skewed,
  the sampling distribution of means is normal if
  the sample size n is large enough.
• If the population distribution is skewed and the
  sample size is small, experimenters must use
  non-parametric statistics. These tests have less
  statistical power. The equivalent of the
  parametric independent t-test is called the
  Mann-Whitney U test.

32
Diabetes/Sykursýki
Within subjects design/ Innanhópasnið
Dæmi Does diabetes education improve scores on
tests about diabetes knowledge?
S1
Before 75 62 67 70 55 59 60 64 72 59
After 77 65 68 72 62 61 60 67 75 68
paired or related or dependent t-test/háð t-próf
Calculate the results of a dependent t-test.
(2-tailed) What significant level is reached?
__________ (marktektarstig) What is the size of
the effect? __________ (stærð áhrifa)
Null hypothesis/Nulltilgátan There is no
difference in knowledge scores after receiving
diabetes education.
33
t-Test Paired Two Sample for Means t-Test Paired Two Sample for Means

  Variable 1 Variable 2
Mean 64,3 67,5
Variance 42,23333333 33,61111111
Observations 10 10
Pearson Correlation 0,906845904
Hypothesized Mean Difference 0 null hypothesis
df 9
t Stat -3,692307692
P(Tltt) one-tail 0,002489772
t Critical one-tail 1,833112923
P(Tltt) two-tail 0,004979545 p gildi
t Critical two-tail 2,262157158  
p gildi/marktektarstig er minn en 0,05. Við erum
með tölfræðileg marktekt.
34
A two-tailed or one-tailed t-test?Einhliða próf
eða tvíhliða próf?

• Diabetes education will not decrease peoples
  knowledge about diabetes.
• We can use the one-tailed t-test result.
• The value of p is 0,00248.
• statistically significant/tölfræðilega marktæk

t-3,69
35
Clinically significant ?Klínísk marktæk ?

• The paired t-test result (one-sided) showed that
  diabetes education improves knowledge about
  diabetes.
• En stærð áhrif er aðeins 3,2.
• Er það klínísk marktæk?
• Do patients control their blood sugar levels
  better as a result of only slightly improved
  knowledge ?

36
Statistical significance does not imply clinical
significance

• Reducing the mean number of bed sores by 1,0
  means one less bed sore for every patient.
• Að meðaltali og er klínísk marktæk
• Reducing the mean number of bed sores by 0,1
  means one less bed sore every 10 patients.
• Reducing the mean number of bed sores by 0,01
  means one less bed sore every 100 patients.
• Reducing the mean number of bed sores by 0,001
  means one less bed sore every 1000 patients.
• stærð áhrifa er ekki klínísk marktæk

tölfræðilega marktæk er ekki sama sem klínísk
marktæk