Statistical Thinking in Biomedical Research

The goal of the course is to present the biostatistical concepts that researchers need in order to evaluate the results of published clinical studies, to design and analyze studies, and to collaborate effectively with methodologists. The course will illustrate how biostatistical and epidemiologic principles affect all phases of the study, including choosing an outcome measure, developing procedures for collecting accurate and complete data and summarizing study results. Participants will be introduced to fundamental concepts in statistical inference and will learn to recognize the problems of bias and imprecision that can lead to misleading or uninformative results. Participants will also be introduced to the value as well as the limitations of statistical models. Participants will learn how to avoid misleading graphics and will see examples of the latest graphical methods for summarizing data accurately. The emphasis in the course is on the underlying concepts, not on computations.

Syllabus

Section 1: Introduction to Statistical Concepts.

Instructor: Frank Harrell

Statistical inference is the process of drawing general conclusions from specific studies. This section presents the fundamental concepts that provide the basis for statistical inference and examines the effect of endpoint definition, patient selection, and study design (especially randomized trials versus observational studies) on the validity of the general conclusions. This section also introduces descriptive statistics and measures of change, as investigators frequently summarize change from baseline or from control using the difference or percent change without using objective criteria for choosing the best measure of change. This section will show how graphical methods can assist in choosing between differences, percent change, log ratios, and other measures so that the resulting measure is as independent of baseline as possible. Section 1 will also present an overview of statistical resources available at Vanderbilt.

Learning Objectives:

1. To illustrate the main statistical concepts related to study design.
2. To discuss with the basic method of using statistics to draw inferences about treatments and other factors.
3. To illustrate the levels of evidence provided by different studies.
4. To identify when to use different types of descriptive statistics.
5. To learn what constitutes a good measure of change.
6. To know where to find biostatistical resources in the School of Medicine at Vanderbilt

Recommended reading: Kaiser, Stat in Med 8:1183; 1989

Section 2: Tools for Formal Statistical Inference.

Instructor:

The statistical inferences needed to answer many of the scientific questions addressed in biomedical research take the form of formal hypothesis tests and estimation of effects. For example, a clinical trial can address the question of whether a new therapy is superior to a standard treatment and can provide an assessment of the magnitude of the benefit of the new therapy. This section describes the concepts involved in formal hypothesis testing and estimation, including the interpretation of P-values and confidence intervals.

Learning Objectives:

1. To illustrate the basic logic behind hypothesis testing using a simple example.
2. To discuss the notion of a null and alternative hypothesis.
3. To discuss the notion of a P-value and statistical power.
4. To demonstrate how sample size can influence statistical power.
5. To illustrate the use of a confidence interval.

Section 3: Statistical Modeling

Instructor:

In many studies, multivariable statistical models are used to estimate an effect after controlling for a number of extraneous factors. For example, a statistical model might be used to estimate the effect of an intervention on 30-day mortality probabilities, adjusting for the severity of the patient s condition at the time of the intervention. Statistical models are also used for predicting outcomes, such as final diagnosis or in-hospital mortality. This section describes the uses, interpretation and limitations of statistical models.

Learning Objectives:

1. To discuss simple linear regression.
2. To discuss the use of regression models for prediction and inference.
3. To discuss the use of multivariable techniques for statistical adjustment and improvements in inference.
4. To fit nonlinear patterns in data through the use of piecewise polynomial regression models.

Recommended reading: Spanos, Harrell, and Durack, JAMA 262: 2700; 1989.

Optional reading: Harrell, Lee, and Pollock, J Nat Cancer Institute 80: 1198; 1988.

Section 4: Experimental Design and Analyzing Serial Measurements

Instructor:

This section introduces the principles of experimental design as applied to biomedical research. A list of requirements for a good experiment are described along with the issues that determine the choice of an experimental design that will meet these requirements. Examples from published biomedical studies are used to illustrate the advantages and disadvantages of a number of commonly used experimental designs. Biomedical studies often involve multiple measurements taken from the same subject. Two common examples are crossover designs and longitudinal studies. This section will describe the advantages and disadvantages of conducting studies in this way and describe the special methods of analysis that are often useful for these studies. These methods include the use of summary measures, such as within-patient rates of change or area under the curve, for analyzing dose-response and time-response data with repeated measurements. This section will also describe the need for modeling within-subject correlation patterns and will describe the effects of subject dropout or noncompliance.

Learning Objectives:

1. To discuss the principles of experimental design as applied to biomedical research.
2. To describe the issues involved in choosing an appropriate experimental design.
3. To illustrate commonly used designs with examples drawn from current biomedical journal articles.
4. To describe the advantages and disadvantages of conducting studies with repeated measurements.
5. To describe the methods of analysis required by these designs.
6. To discuss the issues in handling dropout or noncompliance,

Recommended reading: Matthews, Altman, Campbell, and Royston, British Medical J 300:230; 1990

Section 5: Measuring of Effect and Evidence in Clinical Research

Instructor:

This section introduces commonly used measures of frequency and effect in clinical studies, and their calculation, interpretation, appropriate use, and relative merits. Results of clinical trials and other clinical studies are often reported in terms of the benefit of therapy in the treated group(s), relative to the experience of a control group which receives usual care or no care/placebo. Ability to select, calculate, and interpret measures of effect is important in reporting your own work honestly and convincingly. Ability to interpret common measures of effect, along with estimates of variation around those estimates, in the work of others is crucial in deciding whether reported findings are likely to be valid and important. This section will review and contrast the variety of basic experimental and non-experimental study designs used in clinical research. The appropriate use of each type of design depends on the state-of-the-art in the particular branch of research, the availability of appropriate study participants or data, the time and resources available, and the nature of the question to be answered. Understanding the relative merits of clinical research designs can both assist you with developing your own research program as well as aid you in judging the relative merits of studies in the biomedical literature. Similarly, Understanding and identifying potential sources of bias can aid both your own work and the interpretation of others work.

Learning Objectives:

1. To calculate and interpret common measures of frequency used in clinical research, describe their relative merits, and know when they are appropriately used.
2. To calculate and interpret common measures of effect used in clinical studies, describe their relative merits, and know when they are appropriately used.
3. To interpret commonly reported confidence intervals associated with estimates of effect in clinical research.
4. To understand the relative merits and appropriate use of experimental and non-experimental research designs, and the strength of evidence from each.
5. Understand common sources of potential bias in clinical research, and common techniques for their minimization.

Recommended reading: Any or all of the JAMA Users Guides series publishes since 1993. For example:

1. Levine M, Walter S, Lee H, Haines T, Holbrook A, Moyer V. Users guides to the medical literature. IV. How to use an article about harm. JAMA 1994; 271: 1615-9
2. Guyatt GH, Sackett DL, Cook DJ. Users guides to the medical literature.11. How to use an article about therapy or prevention.A. Are the results of the study valid? JAMA 1993; 270: 2598-601

Section 6: Graphical Methods

Instructor: Frank Harrell

Graphical methods can provide simple and effective summaries of study results. This is particularly true for the results of complex statistical modeling procedures. Graphical methods may also uncover problems in the data or deficiencies in the model selected. This section describes graphical methods for presenting the results of a study, and methods for avoiding misleading or confusing graphs.

Learning Objectives:

1. To describe some of the most frequent mistakes made in medical graphics.
2. To discuss methods to better illustrate the point that poorly generated graphics were trying to make.
3. To discuss methods to illustrate multiple variables.

Required reading: Handout on graphical methods.

Recommended reading: Singer and Feinstein, J Clin Epi 46:231; 1993.

Section 7: Wrap-up and Interactive Demonstration of Statistical Software and Data Analysis

Instructor: Frank Harrell

This session consists of an interactive demonstration of R software for initial graphical inspection of data as well as final summarization of results, using a variety of graphical methods. The limited graphical capabilities of other software such as Microsoft Excel will be described.

In this session there will also be time for participants to ask questions about material covered in any previous sessions.

Learning Objectives:

1. To discuss capabilities of statistical graphics software and how it can be used to inspect, explore, and summarize data.
2. To become introduced to some ways that a statistician begins to examine data in an exploratory fashion.
3. To clarify any remaining issues from the entire course.