1. Biostatistics and research Methods in Drug Therapy
Amber Olek, PharmD, BCPS
December 12th, 2017

2. Disclosure
Nothing to Disclose

3. Objectives
At the conclusion of this activity, participants will be able to:
Describe key features of the primary study designs
Understand key components of statistical analysis and their use in biomedical literature
Apply the above principles to interpret and determine the quality of a study
Formulate a study question and method for your own biomedical research

4. Research Methods

5. Where to Begin: A Study Outline
Abstract
Introduction
Methodology
Study Design
Bias
Results
Validity and Reliability
Discussion
Conclusion

6. Study Designs Descriptive Analytical Explanatory Retrospective
Observational
Prospective
Experimental
Case
Qualitative
Controlled
Quantitative
Exploratory

7. Study Designs Descriptive Explanatory Observational Case-Control
Follow-Up
Cross-Sectional
Experimental
Parallel
Crossover

8. Descriptive Studies Record events, observations, or activities
Don’t provide evidence of efficacy or show causation
Examples
Case reports
Clinical series
Population medicine
Program or course

9. Explanatory Observational Experimental
Seek to link outcomes with risk factors and vice versa.
Examples
Case Control
Follow-up
Cross-Sectional
Primarily evaluate efficacy
Active interventions
Examples
Parallel or Independent
Cross-Over or Related

10. Study Designs Study Design Key Features Descriptive
No comparative group, no interventions
Observational: Careful Watching
Case-Control
Identify patients with outcome; identify control; look BACK in time for risk factors
Follow-Up
Identify a cohort; classify based on risk factor
Cross-Sectional
Identify a cohort; classify based on outcome
Experimental: Active Intervention
Parallel
Randomized to intervention groups
Crossover
Randomized to sequence; each patient gets each intervention

11. Observational: Case-Control
Cases
People with disease/outcome
Risk Factor (present)
Risk Factor (absent)
Controls
People without disease/outcome
Risk Factor (present)
Risk Factor (absent)
Past
Present
Gehlbach, SH. Interpreting the medical literature. 5th ed. New York: McGraw Hill Medical Publishing. 2006

12. Observational: Follow-Up
Study Population
(Free of Disease)
Risk Factor (present)
Disease/Outcome (Present)
Disease/Outcome (Absent)
Risk Factor (absent)
Present
Future
Gehlbach, SH. Interpreting the medical literature. 5th ed. New York: McGraw Hill Medical Publishing. 2006

13. Observational: Cross-Sectional
Study Population
Free of Disease/Outcome
Risk Factor (Present
Risk Factor (Absent)
Disease/Outcome Absent
Risk Factor (Present)
Present
Gehlbach, SH. Interpreting the medical literature. 5th ed. New York: McGraw Hill Medical Publishing. 2006

14. Experimental Parallel Crossover
Patients are assigned to a single treatment group
Must consider interpatient variability
Requires more patients/participants
Each patient receives each treatment, one followed by the other
Randomized to sequence of therapy
Patients serve as their own control  less interpatient variability

15. Clinical Trials Superiority Trial Hypothesis: “No difference”
Equivalence Trial
Hypothesis: “Groups are NOT equal”
Non-Inferiority Trial
Hypothesis: “Not Non-Inferior”

16. Validity and Reliability
Does the measurement represent a true value?
Reliability
Are the measurements reproducible?

17. Bias: All the Kinds Selection Bias Measurement Bias
Classification Bias
Confounding Bias
Allocation Bias
Attrition Bias
Observer Bias
Compliance Bias

18. Types of Bias Selection Inclusion and Exclusion criteria
Is the study population adequately described
Clear definitions—ex: Sepsis
Classification
How are classifications made—Severity of Sepsis

19. Types of Bias Allocation Most difficult to assess
How were patients assigned to their groups?
Randomization—was it truly random?
Observer and Measurement Bias
Confounding Bias
Attributing the outcome to a risk factor not assoc. with the outcome
Exclusion criteria

20. Types of Bias Attrition Bias Differences in losses between groups
Compliance Bias
Patient compliance with therapies

21. Statistical Analysis

22. Statistical Tests Which test do we use? Type of data Number of groups
Independent (parallel) or related (crossover)

23. Types of Data Nominal Yes or No response Ordinal Ranked Continuous
Numerical Data

24. Nominal Data Examples Response rates Adverse event or not
Alive or dead
Pregnant or not
Race
“Traps”
Percentages
Always go back to the data origin

25. Ordinal Data Examples Likert scales (strongly agree—strongly disagree)
Age at diagnosis
Years receiving therapy
“Traps”
Data often presented numerically
Can behave and present like continuous data

26. Continuous Data Examples Age, height, weight
Time to disease progression
Changes in hemoglobin
Interval vs Ratio
“Traps”
Data presented as % is probably not continuous
Can be made to look like nominal or ordinal data
Visual Analog Scales

27. Statistical Tests Type of Data Parallel Groups Cross-Over Groups
Three or more Groups
Nominal Data
Chi-squared (>40 pts)
Fishers Exact
McNemar Test
Chi-Squared for n-independent samples
Ordinal Data
Mann-Whitney U test
Wilcoxon Rank Sum test
Sign Test
Kruskal Wallis ANOVA
Continuous Data
Student’s T-test
Mann Whitney U test
Paired T-test
ANOVA
Tyler, LS. Research Design, Evidence-Based Medicine and Statistical Analysis. Oral presentation for ASHP 2016 Pharmacotherapy Specialty Review Course.

28. Survival Analysis
Can be used in both experimental and observational trials
Looks at follow-up period or time frame from intervention/exposure to a discrete event
Most often death; stroke, MI, etc.
Gives us information regarding both survival and hazard
Censored data

29. Survival Analysis Commonly presented on a Kaplan-Meier Curve or Plot
Cox Proportional Hazard Model
Gives us the Hazard Ratio
Log Rank Test

30. Hazard Ratio
“Ratio of the hazard rates corresponding to the conditions described by two levels of an explanatory variable”
Difference in rate of an event or outcome between two groups
Survival Analysis
Calculus based

31. Hazard Ratio
HR > 1: Increased risk associated with a given variable
HR < 1: Reduced risk

32. Hypothesis Testing Research vs. Null Hypothesis
Research hypothesis is the more natural question/statement
Ex): Incidence of QT-prolongation is greater in patients receiving ondansetron vs promethazine
Null Hypothesis represents the “negative” alternative
Ex) There is no difference in incidence of QT-prolongation between patients receiving ondansetron vs promethazine
The Null Hypothesis is necessary for statistical testing
Probability (P) Value

33. P-Value
Probability (P) Value: “the probability of obtaining a result equal to or more extreme than what was actually observed.”
Level of significance should be set at the beginning
p< 0.05… “Statistically Significant!”
Pitfalls….must also consider clinical significance of the results

34. Interpreting P-Value Superiority Trial Hypothesis: “No difference”
Goal: To demonstrate there is a difference between groups (p<0.05)
Non-Inferiority Trial
Hypothesis: “Not Non-Inferior”
Goal: To demonstrate there is NO difference between groups (p>0.05)

35. Errors in Hypothesis Testing
Type I
Alpha error
Proposed difference does not exist
Type II
Beta error
Hypothesis: No difference, Results: Yes, different
Need more patients
Anticipated
Conclusion
Truth
Difference Exists
NO Difference
Correct
Type I or alpha error
Type II or beta error
Gehlbach, SH. Interpreting the medical literature. 5th ed. New York: McGraw Hill Medical Publishing. 2006

36. Confidence Interval 95% CI Gives more meaning to the p-value
Size of the interval matters
Narrow CI reflects a large sample size, gives strength to result
Wide or large CI reflects a smaller sample size; potential for uncertainty in the results

37. Intention to Treat vs Per Protocol Analysis
Includes ALL patients who were randomized
Maintains comparability of groups and reduces bias
“Real Life”
Effectiveness
Per Protocol Analysis
Includes ONLY patients who followed the protocol
More likely to show an exaggerated effect
Stricter efficacy

38. Assessing Risk Evaluating Outcome Results Prevalence vs Incidence
Prevalence: Number of established cases at a given time
Incidence: Number of new cases occurring during a given time
2x2 Table

39. Assessing Risk Odds Ratio Relative Risk Based on Prevalence
Used in case-controlled and cross-sectional
No denominator
Based on Incidence
Used in follow-up and experimental
Denominator

40. Outcome (i.e. disease, mortality, etc.)
2 X 2 Table
Risk Factor/ Intervention
Outcome (i.e. disease, mortality, etc.)
Present
Absent A B
A+B C D
C+D

41. Odds Ratio
OR = AD/BC

42. Relative Risk RR = (a/a+b) ÷ (c/c+d) Interpretation
RR<1: Intervention decreased risk of outcome
RR=1: No difference in risk
RR>1: Intervention increased risk of outcome

43. Relative and Absolute Risk Reduction
RRR = 1-RR
Expressed as a percentage reduction
Demonstrates proportion of events
ARR = [ A/(A+B)] – [ C/(C+D)]
Gives the actual frequency of the results
Needed to calculate NNT

44. Number Needed to Treat NNT = 1/ARR
More meaningful value than RRR, ARR, etc.

45. New Drug A in Pancreatic Cancer
Study of New Drug A vs standard treatment for pancreatic cancer.
Double-blind, randomized, multicenter trial
Outcome: Mortality at 90 days
1589 patients enrolled
798 received new drug A
791 received standard chemotherapy

46. Mortality at 90 days in Pancreatic Cancer
Intervention
Outcome = Mortality at 90 days
Died
Survived
New Drug A
589
209
798
Chemotherapy (Standard)
624
167
791

47. New Drug A in Pancreatic Cancer
Relative Risk =
Relative Risk Reduction =
Absolute Risk Reduction =
Number Needed to Treat =

48. Forming a Study Question

49. Where to begin? Topic Selection
What is relevant to your practice site?
Patient population
Study Design
Observational vs Experimental
Avoid the “Retrospective Review”

50. Barriers Time Never enough Resources Sample size
How many charts reviewed vs patients included

51. Question Development FINER Criteria Feasible? Interesting? Novel?
Ethical?
Relevant?
PICO
Patient/Population
Intervention
Comparator
Outcome

52. Question Development Consider problematic aspects of your practice
Availability of resources
Shortages
Concerning outcomes occurring?
“Why do we do it that way?”
Lipowski, EE. Developing great research questions. AJHP. 2008; 65(17): 

53. PICO Method
Topic: Reduction in missed doses of meropenem after switching from IVPB to IV push
Patient/Pop: Patients who received meropenem
Intervention: Meropenem administration changed from IVPB to IV push
Comparator: IVPB control group; retrospective data collection prior to intervention
Outcome: Reduction in missed doses

54. Study Sample
Important to consider how you’re going to obtain your patient sample
Sample size
Too many vs. too few

55. Resources
Patanwala, AE. A practical guide to conducting and writing medical record review studies.
AJHP November 15th 2017
References the AJHP “Tool Kit” for practice-based research

56. References
Gehlbach, SH. Interpreting the medical literature. 5th ed. New York: McGraw Hill Medical Publishing. 2006
Tyler, LS. Research Design, Evidence-Based Medicine and Statistical Analysis. Oral presentation for ASHP 2016 Pharmacotherapy Specialty Review Course.
Patanwala AE. A practical guide to conducting and writing medical record review studies. AJHP. 2017; 74(22):
Farrugia P, Petrisor BA, Farrokhyar F, Bhandari M. Research questions, hypotheses and objectives. Canadian Journal of Surgery. 2010;53(4):
Developing great research questions
Earlene E. Lipowski
Lipowski, EE. Developing great research questions. AJHP. 2008; 65(17): 

57. References
Ranganathan P, Pramesh CS, Aggarwal R. Common pitfalls in statistical analysis: Intention-to-treat versus per-protocol analysis. Perspectives in Clinical Research ;7(3): doi: /
Flynn, R. Survival Analysis. Journal of Clinical Nursing. 2012; 21:
Ranganathan, P, Pramesh, CS, Aggarwal, R. Common pitfalls in statistical analysis: Intention-to-treat versus per-protocol analysis. Perspect Clin Res Jul-Sep; 7(3): 144–146.