Ashok S Gavaskar
Asst. Editor - Indian Journal of Orthopaedics
Retrospective
study design
How to set it up?
Research workshop - IOACON’ 16
Retrospective study - design
• most common form of analysis
(Data originally collected for other reasons)
• quick
• not expensive
• rare outcomes
• long latent period
• generates hypothesis
• Bias
• Cannot provide valid solutions
`
Outcome
-measurable parameter of clinical interest
“Has already occurred”
Retrospective design: Key points
Retrospective design: Key points
Exposure: ‘Factor of interest’

Interventional (can only be prospective)
you control the factor of interest
Observational (prospective/retrospective)
“you just observe”
Retrospective design: Key points
Cross-sectional
Cohort
Case control
Observational
(retrospective)
Cross sectional design
No direction

One time

(eg: Survey)
1
3
2
Different groups
compared at
ONE time
• Descriptive purposes

(states the problem)

• Poor inference
Case control design
Unexposed
Exposed
Exposed
Unexposed
DISEASE

(cases)
DISEASE

(controls)
Review

records
Review

records
• Rare outcome

(only one outcome)

• Multiple exposures

• Inference - moderate
Cohort design
Unexposed
Exposed
outcomes

(study begins)
records

review
Disease
NoDisease
Disease
NoDisease
• common outcomes

(multiple outcomes)

• Multiple exposures

• Strongest - Observational
Doing a good retrospective study
Research Question
• Description
• Relationship
• Comparison
what is going on? (incidence/prevalence research)
proportion/percentage/ central tendency/ variability
how phenomena are related?
correlation co-efficients
variable of interest (difference among groups)
central tendency
Literature review
• An essential pre-requisite

• Systematic review

(study’s area of focus, demographics, criteria)

• Multiple databases

• Background - key concepts & variables
Study proposal
• abstract

• introduction

• research question

• literature review

• methodology

• significance

• limitations

• budget

• references

Sample
Design
Variables
Instruments
Key elements: Sampling issues
• Sample size
• Sampling strategy
• Key element in any research proposal
Sample size
• Power analysis

(probability of rejecting the null hypothesis)

related to sample size

(10 cases per variable)

Tools:

• textbooks
• journal articles
• downloadable software programs (G*Power 3.0)
Sampling strategy
• Convenience sampling
what is available at disposal (e.g:cases with in a particular time frame)
• rare cases, outcomes
• small sample size
Sampling strategy
Gold standard, (has equal chance)
• suitable for multi-centre trials
• common disorders
• Random sampling
Sampling strategy
every nth case is selected (not truly random)
access to large number of records
• Systematic sampling
Study proposal
• abstract

• introduction

• research question

• literature review

• methodology

• significance

• limitations

• budget

• references

• Future prospective studies
• Variables
Define
Operationalise (literature review)
translating a construct to its
manifestation
Study proposal
• Design
• flow of information
• go through a few charts
• on site clinicians (multi-centre)
• abstract

• introduction

• research question

• literature review

• methodology

• significance

• limitations

• budget

• references
Methodology
• Instruments (paper/digital)
Paper
cost effective
pre-printed form
(avoids coder’s interpretation of data)
not so good…..

• Handwriting

• storage

• maintenance
Methodology
• Instruments
• Digital
• large RCRs
• centralisation of data storage
• entry and transcription errors
• can be generated from software packages
Data abstraction
Inclusion/ Exclusion criteria
• lack of sufficient variables recorded
• presence of excessive/confounding co-
morbidities
• confounding factors that can degrade the
validity of data
Constant review to assess excluded data
Data abstraction
• Coding/procedure manual

to ensure accuracy, reliability & consistency of data

• clear definitions

• protocols

• steps for data extraction
Data abstraction
• Data abstractors:

• selection & training

• blinding

reviewer bias

• Intra and inter -rater reliability
Data abstraction
• Intra & Inter rater reliability
• (statistical estimate to report
consistency in coding)
Inter:
Cohen kappa
(extent of agreement
-1 to +1, for RCR: 0.6)
Intra:
calculating ICC (intra class correlation)
Data management
• Data management
Software package
(Microsoft access, Medquest)
• data input

• statistics

• reporting
Pilot study
• Very useful
helps to assess
study design
feasibility
evaluate methodology
• 10% of the target population
Summary
• Well defined research questions
• Sampling: size & strategy
• Operationalise variables
• Data abstraction process: most important
• Inclusion and exclusion criteria
• Observer reliability
• Pilot test
For a good RCR…

Retrospective Study Design

  • 1.
    Ashok S Gavaskar Asst.Editor - Indian Journal of Orthopaedics Retrospective study design How to set it up? Research workshop - IOACON’ 16
  • 2.
    Retrospective study -design • most common form of analysis (Data originally collected for other reasons) • quick • not expensive • rare outcomes • long latent period • generates hypothesis • Bias • Cannot provide valid solutions
  • 3.
    ` Outcome -measurable parameter ofclinical interest “Has already occurred” Retrospective design: Key points
  • 4.
    Retrospective design: Keypoints Exposure: ‘Factor of interest’ Interventional (can only be prospective) you control the factor of interest Observational (prospective/retrospective) “you just observe”
  • 5.
    Retrospective design: Keypoints Cross-sectional Cohort Case control Observational (retrospective)
  • 6.
    Cross sectional design Nodirection One time (eg: Survey) 1 3 2 Different groups compared at ONE time • Descriptive purposes (states the problem) • Poor inference
  • 7.
  • 8.
    Cohort design Unexposed Exposed outcomes (study begins) records review Disease NoDisease Disease NoDisease •common outcomes (multiple outcomes) • Multiple exposures • Strongest - Observational
  • 9.
    Doing a goodretrospective study Research Question • Description • Relationship • Comparison what is going on? (incidence/prevalence research) proportion/percentage/ central tendency/ variability how phenomena are related? correlation co-efficients variable of interest (difference among groups) central tendency
  • 10.
    Literature review • Anessential pre-requisite • Systematic review (study’s area of focus, demographics, criteria) • Multiple databases • Background - key concepts & variables
  • 11.
    Study proposal • abstract •introduction • research question • literature review • methodology • significance • limitations • budget • references Sample Design Variables Instruments
  • 12.
    Key elements: Samplingissues • Sample size • Sampling strategy • Key element in any research proposal
  • 13.
    Sample size • Poweranalysis (probability of rejecting the null hypothesis) related to sample size (10 cases per variable) Tools: • textbooks • journal articles • downloadable software programs (G*Power 3.0)
  • 14.
    Sampling strategy • Conveniencesampling what is available at disposal (e.g:cases with in a particular time frame) • rare cases, outcomes • small sample size
  • 15.
    Sampling strategy Gold standard,(has equal chance) • suitable for multi-centre trials • common disorders • Random sampling
  • 16.
    Sampling strategy every nthcase is selected (not truly random) access to large number of records • Systematic sampling
  • 17.
    Study proposal • abstract •introduction • research question • literature review • methodology • significance • limitations • budget • references • Future prospective studies • Variables Define Operationalise (literature review) translating a construct to its manifestation
  • 18.
    Study proposal • Design •flow of information • go through a few charts • on site clinicians (multi-centre) • abstract • introduction • research question • literature review • methodology • significance • limitations • budget • references
  • 19.
    Methodology • Instruments (paper/digital) Paper costeffective pre-printed form (avoids coder’s interpretation of data) not so good….. • Handwriting • storage • maintenance
  • 20.
    Methodology • Instruments • Digital •large RCRs • centralisation of data storage • entry and transcription errors • can be generated from software packages
  • 21.
    Data abstraction Inclusion/ Exclusioncriteria • lack of sufficient variables recorded • presence of excessive/confounding co- morbidities • confounding factors that can degrade the validity of data Constant review to assess excluded data
  • 22.
    Data abstraction • Coding/proceduremanual to ensure accuracy, reliability & consistency of data • clear definitions • protocols • steps for data extraction
  • 23.
    Data abstraction • Dataabstractors: • selection & training • blinding reviewer bias • Intra and inter -rater reliability
  • 24.
    Data abstraction • Intra& Inter rater reliability • (statistical estimate to report consistency in coding) Inter: Cohen kappa (extent of agreement -1 to +1, for RCR: 0.6) Intra: calculating ICC (intra class correlation)
  • 25.
    Data management • Datamanagement Software package (Microsoft access, Medquest) • data input • statistics • reporting
  • 26.
    Pilot study • Veryuseful helps to assess study design feasibility evaluate methodology • 10% of the target population
  • 27.
    Summary • Well definedresearch questions • Sampling: size & strategy • Operationalise variables • Data abstraction process: most important • Inclusion and exclusion criteria • Observer reliability • Pilot test For a good RCR…