Importance of Biostatistics in Biomedical Research Dr R

Importance of Biostatistics in Biomedical Research Dr. R. M. Pandey Prof & Head Department of Biostatistics A. I. I. M. S. , New Delhi rmpandey@yahoo. com

Clinical Research Definition “Clinical research is a component of medical and health research intended to produce knowledge valuable for understanding disease, preventing and treating illness, and promoting health” US National Clinical Research Summit Project, 1998 2

Issues & Questions in Biomedical Research ISSUE(S) QUESTION(S) Normality Is a person sick or well? Abnormality What abnormalities are associated with having a disease? Diagnosis How accurate are diagnostic tests or strategies used to find a disease? Frequency How often does a disease occur? Risk What factors are associated with an increased likelihood of disease? Prognosis What are the consequences of having a disease? Prevention Does intervention on people without disease work? Does early detection and treatment improve the course of disease? . Cause What condition results in a disease? What are the pathogenetic mechanisms of disease?

Evidence Based Alliance The ability to use our clinical skills and past experience to rapidly identify each patient's • unique health state and diagnosis, • their individual risks, • the benefits of potential interventions, and their • personal values and expectations. Clinical Expertise EBM Research Evidence Patient Preferences The unique preferences, concerns and expectations of each patient Clinically relevant, patient centred, research about: • Diagnosis • Prognosis • Interventions EBM. Sackett et al 1996 4

Evidence Based Alliance? Our clinical experience alone may not improve our patient’s lives Clinical Expertise A lot of research is not aimed at improving the lives of patients! EBM Research Evidence Are our patients always able to judge what is best for them? Patient Preferences Who asks them? 5

The Research Question • All studies should start with a research question that addresses what the investigator would like to know • Goal is to find an important research question that can be developed into a feasible and valid study plan 6

Research Questions Primary/ Secondary 1. What is the prevalence of a condition? . 2. What is the average (Mean) of a characteristics? 3. What is the strength of correlation between two quantitative parameters? 4. What is the agreement between methods? 5. What are the diagnostic characteristics of a candidate test (categorical/quantitative) with reference to a “Gold Standard”? . 6. What is incidence of an outcome? . 7. What are the predictors of an outcome? 8. What are the risk factors associated with an outcome? 7 9. Evaluation of a candidate intervention against a control (standard of care)? .

Population vs. Sample 8

Population Sample Parameter Statistic Sampling variation 9

Process of Research Project Drawing Conclusions Infer Designing & Implementing Design Infer Implement 10

11

Overview of Study Designs in Clinical Research 12 12 12

Variables Outcome Exposures Confounder(s) 13

Total variability = True variability + Error Sources of Error in Measurements • Observer • Subject • Instrument 14

Types of Variability Systolic Blood Pressure Random Systematic Time 24 hr 15

Two questions asked at the end • Validity the study results? • Reliability of the study results?

Illustration of the Difference Between Precision and Accuracy 17 Hulley & Cummings, Designing Clinical Research, 1988.

Illustration of the Difference Between Precision and Accuracy Good Precision Poor Accuracy Good Accuracy Poor Accuracy Hulley & Cummings, Designing Clinical Research, 1988. 18

Validity and Reliability of both : Measurements, and Study result • Validity : – Asking - Are we measuring/estimating what we think we are meaning? • Reliability : – Asking - How reproducible is the value/study result ? STATISTICAL METHODS, IF USED PROPERLY, PROVIDES VALID AND RELIABLE RESULTS 19

Statistical Methods BIOSTATISTICS Research Methods in Health Sciences Computing Skills

Statistical Methods I. Design Stage Sample Size Sampling/Randomization Data Management Plan Data Analysis Plan II. Analysis Stage Descriptive Analysis Inferential analysis III. Interpretation & Publication 21

Statistical Methods (Analysis) I. Descriptive Methods : Tables Diagrams Summary : Univariable, Bivariable strength Multivariable II. Inferential Methods : n n n Estimation Point Estimation Mean / proportion etc. Interval Estimation i. e. Confidence interval of point estimate n Hypotheses Testing n Comparison between the treatments n Association n Etc. 22

Five Important Terms Outcome Exposures Bias Confounder(s) Chance factor 23

Variables Functional Relationship Measurement/Analysis Categorical Quantitative Time to an event Outcome √ √ √ Exposure √ √ x √ √ √ x x x Other factors • Confounder/s • Effect modifier/s • Interaction 24

Descriptive Analysis • One variable (uni-variable) – Qualitative – Quantitative – Time to an event • Two variables (Bivariable) – Qualitative vs Qualitative – Qualitative vs Quantitative – Quantitative vs Quantitative • More than two variables (Multivariable) 25

Measures of Associations • Measures of association are mathematical comparisons • Mathematical comparisons can be done in – absolute terms, or – relative terms 26

Absolute and Relative Measures Risk Difference (Absolute Measures) Relative Risk (Risk Ratio) Odds Ratio Rate Ratio Each one gives a different perspective Each one appeals to different constituencies 27

Observed Exposure-Outcome Association : Possibilities The observed statistical association between a certain outcome and the hypothesized exposure could be the result of systematic errors in collection of data (sampling, disease and exposure ascertainment) or its interpretation - role of bias Or it could be due to the effect of additional variables that might be responsible for the observed association - role of confounding variable(s) Or it could be just a matter of chance Or it could be a real association

Confounding • Confounding is a distortion in a measure of effect that may arise because we fail to control for other variables that are previously known risk factors for the health outcome being studied. 29

Confounding • Confounding can lead to the observation of apparent differences between the study groups when they do not truly exist, or conversely, the observation of of no difference when they do exist. 30

Confounding variable • It is an independent risk factor(cause) of outcome) • It is unevenly distributed among exposed and unexposed • It is not on the causal pathway between exposure and outcome 31

THE DIFFERENCE BETWEEN BIAS AND CONFOUNDING § Bias creates an association that is not true, § Confounding describes an association that is true, but potentially misleading. 32

EXAMPLES OF RANDOM ERROR, BIAS, MISCLASSIFICATION AND CONFOUNDING IN THE SAME STUDY: Cohort study: babies of women who bottle feed and women who breast feed are compared, it is found that the incidence of gastroenteritis, as recorded in medical records, is lower in the babies who are breast -fed. 33

EXAMPLE OF RANDOM ERROR By chance, there are more episodes of gastroenteritis in the bottle-fed group in the study sample, producing a type 1 error. (When in truth breast feeding is not protective against gastroenteritis). Or, also by chance, no difference in risk was found, producing a type 2 error (When in truth breast feeding is protective against gastroenteritis). 34

EXAMPLE OF RANDOM MISCLASSIFICATION Lack of good information on feeding history results in some breast-feeding mothers being randomly classified as bottle-feeding, and viceversa. If this happens, the study finding underestimates the true RR, whichever feeding modality is associated with higher disease incidence, producing a type 2 error. 35

EXAMPLE OF BIAS The medical records of bottle-fed babies only are less complete (perhaps bottle fed babies go to the doctor less) than those of breast fed babies, and thus record fewer episodes of gastro-enteritis in them only. This is called bias because the observation itself is in error. 36

EXAMPLE OF CONFOUNDING The mothers of breast-fed babies are of higher social class, and the babies thus have better hygiene, less crowding and perhaps other factors that protect against gastroenteritis. Less crowding and better hygiene are truly protective against gastroenteritis, but we mistakenly attribute their effects to breast feeding. This is called confounding because the observation is correct, but its explanation is wrong. 37

Bivariate Association Between Smoking Status and Risk of Death Bivariate Nonsmokers Former smokers Recent quitters Persistent smokers Relative risk of death 1. 0 (ref. ) 1. 08 (0. 92 -1. 26) 0. 56 (0. 40 -0. 77) 0. 74 (0. 59 -0. 94) Hasdai, D. , et al. “Effect of smoking status on the long-term outcome after successful percutaneous coronary revascularization. ” N. Engl. J. Med. 1997; 336: 755 -61. Various other studies have also found similar results. It is known as Smoker’s paradox 38

Association Between Demographic and Clinical Factors and Smoking Status Nonsmokers Former smokers Recent quitters Persistent smokers Age, year + SD 67 + 11 65 + 10 56 + 10 55 + 11 Duration of angina, month + SD 41 + 66 51 + 72 21 + 46 29 + 55 Diabetes, % 21% 18% 8% 10% Hypertension, % 54% 48% 39% Extent of coronary artery disease, % One vessel Two vessels Three vessels 50% 36% 14% 51% 36% 13% 57% 34% 10% 55% 36% 9% Hasdai, D. , et al. “Effect of smoking status on the long-term outcome after successful percutaneous coronary revascularization. ” N. Engl. J. Med. 1997; 336: 755 -61. 39

Comparison of Bivariate and Multivariable Association Between Smoking Status and Risk of Death Nonsmokers Former smokers Recent quitters Persistent smokers Relative risk of death Bivariate 1. 0 (ref. ) 1. 08 (. 92 -1. 26) 0. 56 (. 40 -. 77) 0. 74 (. 59 -. 94) Relative risk of death Multivariable 1. 0 (ref. ) 1. 34 (1. 14 -1. 57) 1. 21 (. 87 -1. 70) 1. 76 (1. 37 -2. 26) Hasdai, D. , et al. “Effect of smoking status on the long-term outcome after successful percutaneous coronary revascularization. ” N. Engl. J. Med. 1997; 336: 755 -61. 40

Intervening Variable • An intervening variable is on the causal pathway to the outcome Alcohol Consum ption Myocardial Infarction Camargo, C. A, Stampfer, M. J. , et al. “Moderate alcohol consumption and risk of angina pectoris in myocardial infarction in U. S. male physians” Ann. Intern. Med. 1997; 126: 372 -5. 41

What to compute? Objective(s) Study Designs Cross-Sectional Case-Control Primary Burden Clinical Trial Hypothesis generation -Prev, mean Prev of RF’s & Measures of Association Hypothesis generation Prev of RF’s in Case-Control & Measures of Assoc. Cause-Effect Cohort Secondary Hypothesis generation Incidence of outcome(s), Measures of Assoc. Cause-Effect Hypothesis generation Incidence of outcome(s), Measures of Assoc. 42

• Statistical Analysis is a computing problem: Avoid Such a thinking • Prevention is Cost-Effective: Also true for Biostatistics • Preventive measures: 1. Must have knowledge of Principles of Research Methods & Biostatistics. 2. Develop computing skills 43

Thank you Thank You 44