Biostatistics in Nursing Research 101409.ppt - Biostatistics in ...

42510011 0010 1010 1101 0001 0100 1011

Biostatistics in Nursing Science

Alexandra L. Hanlon, PhD

University of Pennsylvania

October 14, 2009

4251

0011 0010 1010 1101 0001 0100 1011

Topics

• Introduction

• Areas of Contribution

• Study Design

• Working with Doctoral Students

• Requesting Biostatistics Support

• Computer Labs

4251

0011 0010 1010 1101 0001 0100 1011

Alexandra L. Hanlon, PhDOffice of Nursing Research

Claire M. Fagin HallRoom 479L

[email protected]

4251

0011 0010 1010 1101 0001 0100 1011

Areas of Contribution

• Facilitate Research– Pre-award support– Post-award analyses– Unfunded data analyses– Abstract manuscript preparation*– Consultation

*ask for second authorship if I run analyses, write stats, and results portion of paper

4251

0011 0010 1010 1101 0001 0100 1011

Areas of Contribution, continued

• Teaching– Guest lectures (power, significance, multi-level

modeling)– Independent study– Research Residency– Mentorship– Dissertation Advising

4251

0011 0010 1010 1101 0001 0100 1011

Pre-Award Support

• Study design

• Study subjects

• Types of variables– Scales of measurement– Types of data– Methods of data collection– Validity and reliability

4251

0011 0010 1010 1101 0001 0100 1011

Pre-Award Support, continued

• Follow-up period

• Statistical methods

• Sample size calculations/power analysis

4251

0011 0010 1010 1101 0001 0100 1011

The Well-Designed Study

• Double blind

• Random assignment

• Control, Placebo

• Minimizes extraneous variables

• Adequately powered for statistical significance

4251

0011 0010 1010 1101 0001 0100 1011

Blinding• Single-blind: The patient or investigator does

not know which intervention s/he is receiving• Double blind: Both patient and investigator do

not know the intervention assignment• Triple blind: The statistician is also masked to

the intervention assignment.• The evaluator may be a different person, and

blinding of this person is crucial

4251

0011 0010 1010 1101 0001 0100 1011

Random Assignment

• What is it?

• Why do we do it?

• How do we do it?

• Block or stratified randomization

4251

0011 0010 1010 1101 0001 0100 1011

Intent-to-Treat Principle

• The ITT analysis of an experimental study uses the intervention groups as originally randomized, irrespective of the actual intervention received

• The resulting analysis may lead to a conservative estimate of the intervention effect

• Any other analysis can lead to bias of unknown magnitude and direction

4251

0011 0010 1010 1101 0001 0100 1011

Description of Statistical Methods

• Terminology and level of detail

• Appropriate methodology

• Assumptions– Transformations

• Collapsing variables

• Estimation and confidence intervals

4251

0011 0010 1010 1101 0001 0100 1011

Description of Statistical Methods, continued

• Adjustment for confounding

• Cluster randomized trials

• Multiple comparisons

• Intention-to-treat principle

4251

0011 0010 1010 1101 0001 0100 1011

Multiplicity

• Multiple endpoints, multiple comparisons, subgroup analyses

• To guard against dangers of type I errors:– Adjust p-values for multiple

testing

– Use cross-validation to confirm results/confirm with new studies

– Use sophisticated comprehensive analyses

Number of independent

testsType I error*

1 .05

2 .10

5 .23

10 .40

*detecting significance by chance

4251

0011 0010 1010 1101 0001 0100 1011

The Process

Methods

Power

Design

Aims

4251

0011 0010 1010 1101 0001 0100 1011

The Basics

• Comparing two groups– Continuous DV: two-sample t-test– Categorical DV: chi-square test/Fisher’s Exact

test

• Comparing multiple groups– Continuous DV: one-way ANOVA– Categorical DV: chi-square test/Fisher’s Exact

test

4251

0011 0010 1010 1101 0001 0100 1011

The Basics, continued

• Comparing groups with adjustment for other factors– Continuous DV: ANCOVA or General Linear

Modeling

– Dichotomous DV: Logistic regression

– Count DV: Poisson regression

– Survival time DV: Cox regression

• Account for correlated data (repeated measures, clustering by clinics, etc)

4251

0011 0010 1010 1101 0001 0100 1011

Sample Size Calculations

• When do you need one?

• Consistent with study aims and statistical analysis

• Account for attrition

4251

0011 0010 1010 1101 0001 0100 1011

Sample Size Calculations, continued

• Why is it an important consideration?– Underpowered studies discard useful

treatments/interventions

– Overpowered studies waste resources

• Continuous outcomes require smaller sample sizes than dichotomous outcomes

• Calculate the sample size required to detect a smaller effect than has been reported in pilot studies, because the non-significant pilot studies were likely never published

4251

0011 0010 1010 1101 0001 0100 1011

Statistical versus Clinical Significance

• An over-powered study may yield a significant effect that is quite small– Statistical significance, clinical insignificance

• For example, you observe a p-value of 0.001 with a large sample size, say n=1000, but a difference in blood pressure of 1 mmHg

4251

0011 0010 1010 1101 0001 0100 1011

Selecting an Outcome Measure

• Continuous measures are the most powerful measures

• Ordinal measures and survival time measures are next

• Dichotomous measures are the least powerful

4251

0011 0010 1010 1101 0001 0100 1011

Survey Instruments

• Use validated instruments

Examples:– General Health: SF-36 and SF-12– Depression: CES-D or BDI– Pain: Brief Pain Inventory (BPI)– Cancer Specific Quality of Life: FACT-G– Anxiety: Beck Anxiety Inventory–Primary

Care (BAI–PC)

4251

0011 0010 1010 1101 0001 0100 1011

Dissertation Research:Additional Study Needed?

• Statistics is a wide area of research• Statistical methods have advanced rapidly

in recent years• Early in your research considerations,

discuss with your advisor/statistician whether additional coursework or independent study is necessary to support your level of statistical needs

4251

0011 0010 1010 1101 0001 0100 1011

Dissertation Research: Areas of Assistance

• Power Analysis

• Database design

• Review statistics plan for appropriate methodology

• Interpretation of software output

• Appropriate presentation of results

4251

0011 0010 1010 1101 0001 0100 1011

Dissertation Advising: What to Expect

• Schedule a meeting to walk me through your dissertation topic– Have solid research question(s) – Have mature draft aims– Possibly bring data, blank surveys, etc

• Be prepared to describe the level of measurement for study variables (income, continuous or categorized?)

4251

0011 0010 1010 1101 0001 0100 1011

Dissertation Advising: What to Expect

• Be open to alternative suggestions that might strengthen your study

• Take the time to understand the concepts behind the statistics

• Ask questions, study the literature, and meet with me as often as it takes to understand the concepts

4251

0011 0010 1010 1101 0001 0100 1011

Limitations

• Research Position – 10% max teaching load• Four hours/week for student meetings• Should research require >3 meetings, consider

independent study, inclusion of statistician on dissertation committee, offer funding

• Four guest lectures per academic year• Directing two students at any given time in

research residency/independent study/active committee work

4251

0011 0010 1010 1101 0001 0100 1011

Dissertation Advising: The Simple Rules

• The analysis is your responsibility• Seek advise early on• No question is silly

– Why are we using a particular procedure?

– How does it work to answer my research question?

– What results do I need to extract from the computer output?

• Back up your data!

4251

0011 0010 1010 1101 0001 0100 1011

Request for Biostatistics Support

• Link for Request Form:

https://share.nursing.upenn.edu/sites/onr/stats/Lists/Request%20for%20Biostatistics%20Support/AllItems.aspx

• Intranet Departments ONR Request for Biostatistics Support

4251

0011 0010 1010 1101 0001 0100 1011

4251

0011 0010 1010 1101 0001 0100 1011

4251

0011 0010 1010 1101 0001 0100 1011

4251

0011 0010 1010 1101 0001 0100 1011

Data Transfer

• Contact me via email to set up folder for file sharing

• A folder will be created in your name on H:\Secured Folders\Research Statistics

• Please DO NOT email data for privacy/security reasons

4251

0011 0010 1010 1101 0001 0100 1011

Data Transfer, continued

4251

0011 0010 1010 1101 0001 0100 1011

Computer Lab, Room 210

• 24 PC workstations, each equipped with a CD Burner, Floppy Drive, easily accessible USB ports

• Microsoft Office (Word, PowerPoint, Excel, Access, Front Page)

• SPSS Version 16 is available on workstations 1-6• Various Nursing Specific Applications• 1 Media Station with Scanner• 2 Black & White Laser Printers

http://www.nursing.upenn.edu/otis/services/computerlab/default.asp

4251

0011 0010 1010 1101 0001 0100 1011

PhD Computer Lab, Room 316

• 10 PC workstations, each equipped with a CD Burner, Floppy Drive, easily accessible USB ports

• All computers have MS Office 2007 (Word, Excel, PowerPoint, Access)

• Statistics software: SAS, SPSS, STATA• Anticipated: PASS• One scanner• One black & white laser printer

4251

0011 0010 1010 1101 0001 0100 1011

PhD Computer Lab, Room 316

4251

0011 0010 1010 1101 0001 0100 1011

THANK YOU!Alexandra L. Hanlon, PhD

Claire M. Fagin HallRoom 479L

[email protected]