Research Methods: Design, Data Analysis, And Presentation

1
Research Methods Design, Data Analysis, And
Presentation

• Chapter 1.
• H. Singh, Ph.D.
• NTRS 511

2
Why research?

• Backbone of nutrition and dietetics or any
  subject
• To observe association
• To test hypothesis
• To compare programs
• Strong research means strong profession
• Practice related research is important

3
Classification Descriptive and Analytical

• Descriptive
• include qualitative, case reports, case series
  and survey research.
• Useful in generating hypothesis regarding the
  determinant of condition or disease
• Can establish - associations among factors

4
Analytical

• Include clinical and observational research
• Cohort or follow up studies
• Case-control studies
• Designed to test hypotheses to determine cause
  and effect relationship by intervention by
  researcher

5
The research question

• The research process is a problem-solving,
  decision making procedure involving a series of
  interrelated decision
• Researcher should focus on one decision at a time
• Research question should expand current knowledge
  and practice of profession
• Select issue important to your profession

6
What can be answered?

• How can improve health
• To increase effectiveness of services and
  products
• Untested concepts in published literature
• Researcher should
• Break the problem in to its components parts and
  select a component feasible for study
• Start with a simple problem and generate data
  new questions will develop

7
Prepare for the project

• Review literature
• Both on computer and library
• Critical reviews indicate shortcomings and new
  areas of possible research
• Contact and talk to experts
• Assess available resources
• patient population,
• lab and library facility databases, computer,
  personnel,
• collaborations and funds needed

8
State clearly the question

• Concise, simple and straightforward it is
  easy to design the experiment
• Who is studied?
• Patient, food items. . .
• What is studied?
• Iron intake, labor cost. . .
• How it will be assessed?
• Alteration in food selection, labor cost per
  patient
• Know your variables
• Dependent variables is effected
• Serum cholesterol
• Independent variable is changed
• Fiber intake

9
Research design

• Cross-sectional study data collected at single
  point of time
• Longitudinal collected more that one point of
  time
• Prospective study effect of fiber on CVD with
  time
• Retrospective study goes back in time to find
  relationship

10
Prepare research protocol

• Include
• The research question
• Literature review
• Importance and potential value of research
• The research design
• Who, what and how assessed
• Methods
• Data analysis
• Appropriate statistical analysis

11
Research protocol( contd.)

• For funding use NIH guidelines
• Title page
• Abstract of research plan
• Table of content
• Research Plan ( expanded in next slide)
• biographical sketch
• Other personnel
• Existing and pending resources
• Available resources
• environment

12
Research Plan

• First
• Specific aims and objectives
• Second
• significance of research and review of
  literature
• Third
• Published and unpublished preliminary studies by
  author
• Fourth
• Research design and methods
• Fifth
• Human subject approval procedure (IRB)
• Sixth
• Justification for animal used
• Seventh
• Consultants approvals
• Eighth
• Literature cited
• http//era.nih.gov/ElectronicReceipt/
• http//grants.nih.gov/grants/guide/index.html
• http//grants.nih.gov/grants/funding/424/index.htm
  

13
Conduct a pilot study

• Strongly advised
• Provides experience
• Refinements
• Redrafting of proposal
• Unknowns come up
• Collect and process data

14
Assure ethics in research

• Informed consent
• Voluntary participation
• Anonymity
• Confidentiality
• Risk of harm

15
Qualitative research

• Explore the phenomenon of interest by
• Observations
• Interviews
• Structured or less structured
• Group together focus group
• Questionnaires
• Delphi technique
• answer separately --- circulated ----questioned
  again 2 -3 times till consensus is reached
• Good for planning, solving a problem and
  forecasting

16
Case Series

• Is a report of observations on more than one
  subject
• The purpose of this design to describe
  quantitatively the experience of a series of
  cases with a disease
• Can lead to
• Hypothesis for future studies
• Evidence for association between disease and
  condition
• Identifying the variables

17
Case Series contd.

• Steps. . . .
• Institutional approval and confidentiality
• Features and Subject Selection
• Patients with gastric cancer referred in last six
  months
• The results from this study cannot be generalized
• Data collection
• Existing record
• Concurrent data

18
Case Series contd.

• Statistical analysis
• Means, medians, standard deviation, ranges and
  frequencies are appropriate
• Inference can not be made
• Sample is not representative of all of large
  population

19
Survey

• Design to describe and quantify characteristics
  of a defined population
• Lacks a specific hypothesis
• May suspect a relationship
• Provide statistical profile of the population
• Uses -
• Establishing association among variable
• Provide clue for further study
• baseline data
• Planning health services
• E.g. cold vs hot tray

20
Features

• Sample is selected based on probability design
• subjects are questioned and screened
• Target population should be defined
• More the number of samples more representative
  the results are.

21
Survey contd.

• Data collection
• Using interviews or surveys
• Most difficult part is designing questionnaire
• Length, format also effect
• Do a pilot study
• Correct and improve according to the objective
• Interviewer must be well trained

22
Survey contd.

• Validity and Accuracy
• Validity the instrument is measuring what it is
  suppose to
• Quantitative measure of validity is accuracy
• Reliability and Precision
• Reliability is consistency or repeatability
• Quantitative measure of reliability is precision

A
P
23
Sensitivity and specificity

• Part of Diagnostic test
• Sensitivity refers to how good a test is at
  correctly identifying people who have the
  disease.
• e.g. Less than 6n mol folate - who are actually
  deficient is sensitivity
• Specificity, on the other hand, is concerned with
  how good the test is at correctly identifying
  people who are well
• Proportion of non afflicted individual who test
  negative is specificity
• if we increase sensitivity we loose specificity
• http//bmj.bmjjournals.com/cgi/content/full/327/74
  17/716

24
Statistical Analysis and Interpretation

• Means
• Medians
• Standard deviation and frequency
• Scatter diagram to show correlations
• If data is skewed use Histograms
• limitation
• Low response rate affects the confidence in the
  study
• Measurements are not taken at the same point of
  time
• (risk and prognostic factors should be
  distinguishable)

25
Ex post Facto Analysis (after the fact)

• Should be used with caution since there was no
  control on the design and variables
• Must be exploratory and no causal relationship
  should be developed
• e.g. Bone density and food consumption in women

26
Experimental Design

• Gold standard of analytic research all factors
  are held constant
• To establish causal relationship
• Important consideration during analytic study
  are
• Ethical guidelines to protect rights, privacy and
  welfare of individuals NIH has detailed
  guidelines on this
• Institutional review board (IRB) is another
  agency to educate investigators

27
Randomized clinical trial

• Most common design
• Used to prove the feasibility and safety of a
  treatment
• To compare the efficacy of two or more treatments
• Features
• Subjects are informed
• Assigned to groups
• With randomness
• Observe the outcome

28
Selecting the subjects

• If the study group is not representative results
  cannot be generalized
• Select subject relatively homogeneous
• Compliance should be enhanced by incentives
• Pre-study of collecting urine every 24 hours
• Setting for the study should be natural

29
Choice of Intervention or treatment

• The more the treatment of intervention group
  differs from the comparison group the better the
  results
• Equivalent treatment or placebo should be given
• Blind study- subject are unaware of treatment
  assigned
• Blinding should be maintained
• If subject and investigator are unaware of the
  treatment it is double blind

30
Assignment to treatment groups

• Random method is essential
• Age and gender is important
• If the sample size is less than 200, chances of
  imbalance is there
• Crossover design
• Require fewer subjects. Same subject is used to
  check the effect of treatment is not long lasting
  and the condition is chronic subjects are
  referred to two kind of treatments
• Shortcoming of this method is that subject may
  carry over effects of treatment to next cycle.

31
Size of Sample

• Choose the main endpoint and its method of
  measurement
• Choose the statistical test
• Specify the magnitude of difference smaller
  difference require bigger sample (sensitivity)
• From published results look for variability or
  best guess
• Specify chance of being wrong
• N(zs2)/E where z critical value of confidence
  interval standard deviation and E is plus or
  minus error allowed

32
End point and data collection

• End point of a study is the variable by which
  the treatments are compared
• Real end point may be difficult to record
• Surrogate end point is observed
• Hard (objective) and soft (subjective) end points
• cholesterol vs degree of head ache
• Variables of subjects should be recorded

33
Statistical Analysis and Interpretation

• Non-compliance and loss of subjects is main
  problem
• Report side effects

34
Factorial DesignIf there are more than one
variables

• Check job performance after 3 months

Factor A
Factor B
Bp bonus points B - monetary bonus
35
Partially controlled or quasi experimental design

• Cohort study
• Follow-up studies are observational
• Cohort is a group of patients followed over time
  to observe the effects of treatments
• Group is exposed to a potential risk factor and
  followed for detection of disease
• Time of exposure to disease should be short
• A large group is required

36
Cohort contd.

• Subjects should be free of disease at start
• Samples size
• low frequency of disease need a large number
• End point should be unambiguous
• Statistical analysis descriptive type
• Shortcoming relationship is not causal - only
  exposure was observed other factors may effect

37
Case control study

• Observational design to investigate hypothesis of
  causal relationship
• Retrospective
• No intervention - So not experimental
• Individual cases or comparison is studied
• Exposure status is determined after disease status

38
Statistical Analysis and Presentation of results

• Descriptive
• Inferential
• Probability tells whether these results are due
  to chance rather than due to experimental
  variable
• Frequency histograms
• Mean and standard deviation when symmetrical
• Median and percentile range when unsymmetrical

39
Stat.

• Inferential
• Comparing expected and observed results
• Expected is null hypothesis
• There is no association between factors
• Mean and variance are compared
• Parametric test assume normal distribution and
  similar variability within the group
• Non-parametric test for interval and category
  type data but are less significant statistically

40
Significance test

• Two sided - more common
• on either side of mean
• Standard deviation
• is descriptive
• Describe the spread
• Standard error is
• Is inferential
• Describe the variation relative to sample size

41
Confidence interval

• Degrees of freedom
• P value
• Statistics give us
• Estimate of the quantity mean difference
• Probability that this is due to chance
• Parameters are represented with values to lie
  with in a range of values confidence interval
• Calculated from estimated and its standard error

42
Repeated and Replicated measurements

• Repeat at several point of time
• T-test to measure the difference among the means
• Replicated at the same time from the single
  subject
• Analysis of variance can be used to evaluate the
  variability it need same number of measurements

43
Stat

• Multiple significance test
• If multiple tests are done on many subgroups or
  many variables
• T-test
• LSD values (least square difference)
• Outliers
• Not consistent with normal data
• Missing values
• Must not ignored
• Discuss how it could effect your results

44
Regression and Correlations

• Relationship between the variables is often
  expressed by mathematical model
• Correlation coefficient indicate strength of
  relationship
• Probability and Variance also need to be
  mentioned

45
Other

• Multivariate analysis
• Meta- analysis
• Now
• Conclusion are made keeping in mind limitations
  and assumptions made.
• The new questions arisen are explored and further
  studies are recommended
• Negative studies also important

46
Some points

• This class is about steps involved and tolls
  required to do research and not to finalize a
  proposal - because all areas of research are
  different and need expert to guide
• You can take help from you supervisor to select a
  topic
• It is not your thesis but a proposal your
  research will not be done in this class

47
Some points

• Read Read Read -- is the key to find a topic,
  refine the topic and write the proposal, So Read
  read read read -----
• Discuss, Discuss and Discuss with peers and your
  superviser
• Write your plan - Planning ahead is super
  important