AO Trauma - Statistics and Data Management E-Book

Dirk Stengel

• Herausgeber: Thieme
• Kategorie: Fachliteratur
• Sprache: Englisch

Seitenzahl: 145

Veröffentlichungsjahr: 2009

Beschreibung

Applying new technologies, presenting your experience, and, of course, thinking of your current practice in the light of new evidence, needs some understanding of clinical research methodology.

This book is not a textbook, but a brief guidance for clinical research practice. It aims at bridging the different points of view of statisticians and clinicians, but does not replace personal meetings and discussions between both professions at the earliest step of a clinical study. We hope that you find this book enjoyable, easy to read and understand, and helpful for improving your research skills.

"Me, the editors, and the authors hope that this book will help you to sort and focus your ideas when setting up a clinical study, and to understand why certain information should be expressed in this or that fashion, how data should be compiled, analyzed, and presented. This book is neither exhaustive nor complete; it just fits better into your daily business." - David L Helfet

Leseprobe

Layout and typesetting: nougat GmbH, CH-4056 Basel

Library of Congress Cataloging-in-Publication Data is available from the publisher.

Hazards

Great care has been taken to maintain the accuracy of the information contained in this publication. However, the publisher, and/or the distributor, and/or the editors, and/or the authors cannot be held responsible for errors or any consequences arising from the use of the information contained in this publication. Contributions published under the name of individual authors are statements and opinions solely of said authors and not of the publisher, and/or the distributor, and/or the AO Group.

The products, procedures, and therapies described in this work are hazardous and are therefore only to be applied by certified and trained medical professionals in environments specially designed for such procedures. No suggested test or procedure should be carried out unless, in the user’s professional judgment, its risk is justified. Whoever applies products, procedures, and therapies shown or described in this work will do this at their own risk. Because of rapid advances in the medical sciences, AO recommends that independent verification of diagnosis, therapies, drugs, dosages, and operation methods should be made before any action is taken.

Although all advertising material which may be inserted into the work is expected to conform to ethical (medical) standards, inclusion in this publication does not constitute a guarantee or endorsement by the publisher regarding quality or value of such product or of the claims made of it by its manufacturer.

Legal restrictions

This work was produced by AO Publishing, Davos, Switzerland. All rights reserved by AO Publishing. This publication, including all parts thereof, is legally protected by copyright. Any use, exploitation or commercialization outside the narrow limits set forth by copyright legislation and the restrictions on use laid out below, without the publisher’s consent, is illegal and liable to prosecution. This applies in particular to photostat reproduction, copying, scanning, or duplication of any kind, translation, preparation of microfilms, electronic data processing, and storage such as making this publication available on Intranet or Internet.

Some of the products, names, instruments, treatments, logos, designs, etc. referred to in this publication are also protected by patents and trademarks or by other intellectual property protection laws (eg, “AO”, “ASIF”, “AO/ASIF”, “TRIANGLE/GLOBE Logo” are registered trademarks) even though specific reference to this fact is not always made in the text. Therefore, the appearance of a name, instrument, etc. without designation as proprietary is not to be construed as a representation by the publisher that it is in the public domain.

Restrictions on use: The rightful owner of an authorized copy of this work may use it for educational and research purposes only. Single images or illustrations may be copied for research or educational purposes only. The images or illustrations may not be altered in any way and need to carry the following statement of origin “Copyright by AO Publishing, Switzerland”.

Copyright © 2009 by AO Publishing, Switzerland, Clavadelerstrasse, CH-7270 Davos Platz Distribution by Georg Thieme Verlag, Rüdigerstrasse 14, DE-70469 Stuttgart and Thieme New York, 333 Seventh Avenue, US-New York, NY 10001

Printed in Switzerland.ISBN 978-3-13-152881-0

Table of contents

I    Introduction

II   Foreword

III  Contributors

1    About numbers

2    Errors and uncertainty

3    Outcome selection

4    The perfect database

5    How to analyze your data

6    Present your data

7    Glossary

I Introduction

Dear reader

We, the editors, authors, and all who contributed to this book, appreciate that, in addition to your daily commitment to patient care, you decided to spend extra time and efforts for clinical research. This will definitely make health care a little better, and this book may assist you in the difficult balancing act.

Applying new technologies, presenting your experience, and, of course, thinking of your current practice in the light of new evidence, needs some understanding of clinical research methodology. Your job is to save lives and limbs, and you are doing that perfectly—nobody wants you to become a statistician for a reasonable and clear evaluation of your results. But we need to share a common language, and reach consensus on basic scientific principles.

In fact, studies usually do not fail because of too little use of inference tests, but sloppy planning and inappropriate use of statistics. Data are vulnerable and need attention and diligent care. Spending more time in the beginning will spare time in the long run. Think of the variables you are really interested in, how they can be gathered, stored, processed, and analyzed. Regard statistics as a vehicle to generate and communicate information.

This is not a textbook, but a brief guidance for clinical research practice. It aims at bridging the different points of view of statisticians and clinicians, but does not replace personal meetings and discussions between both professions at the earliest step of a clinical study. Cooperation is vital. Talk. Argue, if necessary. Share opinions, and let your counterpart benefit from your specific expertise.

We hope that you find this book enjoyable, easy to read and understand, and helpful for improving your research skills. Let us know if we got the point.

Dirk StengelMohit BhandariBeate Hanson

II Foreword

Sapere aude! (dare to know)Immanuel Kant, 1784

Whether you love or hate statistics, you need it for clinical decision making, for counseling patients and their relatives, and to argue with those who decide which health care interventions will appear or remain on the market, or even in your hospital. You need statistical knowledge to make your way through the immense and ever-growing body of scientific literature, and, of course, to plan and conduct your own research. Research is an integral part of being a doctor—historically, today, and, far more important, tomorrow. As an orthopaedic or trauma surgeon, you offer a precious good—your skills and your commitment to your patients and the society. Sharing both your expertise and skepticism with the clinical and scientific community is important to bring this discipline forward. Take the helm, and participate in research actively.

The “Handbook of Statistics and Data Management” is obviously not another textbook about statistics. The authors, experts from both a methodological and a clinical point of view, wanted to be brief and concise, spoon-feeding you with the essential knowledge about numerical information, study designs, data storage, and analysis. This book is neither exhaustive nor complete; it just fits better into your daily business. You will probably agree that a book like this was not available to orthopaedic and trauma surgeons before.

Me, the editors, and the authors hope that it will help you to sort and focus your ideas when setting up a clinical study, and to understand why certain information should be expressed in this or that fashion, how data should be compiled, analyzed, and presented. It will help you to negotiate with your statistician—one of the most important persons you have to contact early during study planning. He will probably be amazed that you can express your specific problem in the common language of science—in numbers. And your colleagues will definitely congratulate you that you are able to retranslate numbers in a more important language—the clinical impact of research findings, and the benefit to our patients.

David L Helfet

III Contributors

Editors

Dirk Stengel, MD, PhD, MScHead of the Center for Clinical ResearchDepartment of Trauma and OrthopaedicsUnfallkrankenhaus BerlinWarener Strasse 712683 Berlin, Germany

Mohit Bhandari, MD, MSc, FRCSMcMasters UniversityEpidemiology and Orthopaedics1200 Main Street WestHamilton, Ontario, L8N 3Z5, Canada

Beate Hanson, MD, MPHDirector of AO Clinical Investigation and DocumentationAO Clinical Investigation and DocumentationStettbachstrasse 68600 Dübendorf, Switzerland

Authors

Laurent Audigé, PD Dr (DVM, PhD)Group leader MethodologyAO Clinical Investigation and DocumentationStettbachstrasse 68600 Dübendorf, Switzerland

Kai Bauwens, MDSenior Consultant SurgeonUnfallkrankenhaus BerlinDepartment of Trauma and OrthopaedicsWarener Strasse 712683 Berlin, Germany

Mohit Bhandari, MD, MSc, FRCSMcMasters UniversityEpidemiology and Orthopaedics1200 Main Street WestHamilton, Ontario, L8N 3Z5, Canada

Richard E Buckley, MD, FRCSCHead Division of Orthopaedic TraumaUniversity of CalgaryFoothills Medical CenterDepartment of SurgeryAC 144A1403 29th Street NWCalgary, Alberta, T2N 2T9, Canada

Axel Ekkernkamp, MD, PhDDirectorUnfallkrankenhaus BerlinDepartment of Trauma and OrthopaedicsWarener Strasse 712683 Berlin, GermanyProfessor of SurgeryDepartment of Trauma and OrthopaedicsUniversity Hospital of GreifswaldSauerbruchstrasse17475 Greifswald, Germany

Norbert P Haas, Prof Dr medCharité Universitätsmedizin BerlinCentrum für Muskuloskeletale ChirurgieCampus Virchow-KlinikumAugustenburger Platz 113353 Berlin, Germany

David L Helfet, MD, MBCHBProfessor of Orthopaedic SurgeryCornell University Medical College535 East 70th StreetNew York, 10021, USA

Thomas Kohlmann, PhDProfessor and DirectorInstitut für Community MedicineAbteilung Methoden der Community MedicineWalter Rathenau Strasse 4817487 Greifswald, Germany

Peter Martus, PhDProfessor and DirectorCharité Universitätsmedizin BerlinCampus Benjamin FranklinInstitut für Medizinische Informatik,Biometrie und EpidemiologieHindenburgdamm 3012200 Berlin, Germany

Jörn Moock, PhDInstitut für Community MedicineAbteilung Methoden der Community MedicineWalter Rathenau Strasse 4817487 Greifswald, Germany

Dirk Stengel, MD, PhD, MScHead of the Center for Clinical ResearchDepartment of Trauma and OrthopaedicsUnfallkrankenhaus BerlinWarener Strasse 712683 Berlin, Germany

Michael Suk, MD, ID, MPHAssistant ProfessorUniversity of FloridaDirector, Orthopaedic Trauma ServiceCollege of Medicine Jacksonville655 West Eight Street, 2nd Floor ACCJacksonville, FL 32209, USA

1 About numbers

1 About numbers

1 Introduction

2 Numbers to describe individual patient characteristics

3 Numbers to describe the attributes of a group of patients

3.1 Patient listing versus summary statistics

3.2 Simplification of data

4 Mean versus median

5 Proportions, rates, odds, risks, and ratios

6 Risk difference and number needed to treat (NNT)

7 Summary

1 About numbers

1 Introduction

Numbers can be our friends or foes. They can express information very precisely, or may sometimes put us on the wrong track. Without some knowledge of the anatomy and physiology of numbers, it is almost impossible to conduct meaningful research. The aim of this chapter, therefore, is to introduce you to the proper selection and interpretation of numbers required to transport and distribute your ideas.

To ease communication with your colleagues, you have surely already acquired a personal dictionary of acronyms, synonyms, and abbreviations. As a surgeon dealing with musculoskeletal injuries and diseases you are familiar with terms like ED, OR, CT, MRI, ExFix, or ORIF (emergency department, operating room, computed tomography, magnetic resonance imaging, external fixator, open reduction and internal fixation). Correct use of these terms facilitates communication and forms an important element of your professionalism. You may, however, meet with problems when doing business elsewhere without adapting your vocabulary. Medical language and terminology can be confusing, and similar terms may have very different meanings.

Numbers have a fascinating attribute—they are unequivocally recognized as such by clinicians and researchers, other healthcare professionals, your patients, and everybody, regardless of their background, affiliation, or nationality. The language of numbers is global—so it is the perfect language of science. You may use numbers to encrypt the tons of information you collect about your patients in daily practice, to describe their demographic profile and individual risks, and the results of your treatment. However, using the correct code and choosing the appropriate numbers is essential to compile, handle, and process clinical information.

The key to a successful research project is to translate distinct clinical information into the correct numerical vehicles. The key to evidence-based practice is to retranslate the information encoded in numbers into clinical language.

2 Numbers to describe individual patient characteristics

Information about patient characteristics can be expressed by numbers in one of four major classes of data:

• Binary (or dichotomous)

• Categorical

• Ordinal

• Continuous

Binary (dichotomous) data

The simplest type of information imaginable may be stored in the form of data variables having only two possible categories, such as yes or no, one or zero, male or female, left or right, the presence or absence of a disease or an injury. Such variables are called binary or dichotomous. Although categories may be expressed in words, the data may be stored as numbers or binary information (Fig 1-1). Chapter 5 “How to analyze your data”, chapter 6 “Present your data”, and cross-tables will focus on the utility of binary information.

Fig 1-1a–b

a Example of categories expressed in numbers.

b Example of categories expressed in words.

Categorical data

A fracture of the radial bone may occur in its proximal, mid-, and distal third, which has implications on the treatment, but not necessarily on the outcome. The anatomical classification is value-free, which is the key characteristic of categorical data (Fig 1-2). Another typical example is the pattern of blood types (A, B, AB, and 0). Numbers attached to these categories do not have an intrinsic value and are only used to help store the data and run analysis.

Fig 1-2a–b

a Characteristic of the categorical data is that it is value-free.

b Categorical data can be numbered according to the requirements.

Ordinal data

There are, however, categories which can be placed in distinct order (ie, category B is worse than category A). This type of data is called ordinal data (Fig 1-3). Within the Müller AO Classification of Fractures in Long Bones, a complex, intraarticular fracture with multiple fragments and alteration of the cartilage layer (type C) has a worse functional prognosis than an extraarticular type A fracture. Other examples are the American Society of Anesthesiologists (ASA) risk classification scheme (ASA I–V) or the Gustilo-Anderson grading of open fractures.

Ordinal data variables very often have a limited number of possible categories such as in the later clinical grading systems. These variables are also said to be noninterval because the intervals between adjacent categories (often expressing prognostic information) may not be equal. The difference between type C and type B fractures may not be the same as that between type B and type A fractures.

Fig 1-3 Ordinal data are categorized in a distinct order.

Continuous data

Finally, data variables may be used to store information from counts or measures that, in principle, can take infinity of values within clinically plausible ranges. If you are interested in the treatment of osteoporotic fractures, the T-Score obtained by a dual energy x-ray absorptiometry (DEXA) is a good example of a continuous measure with obvious prognostic impact (Fig 1-4).

Fig 1-4 The T-score obtained by a DEXA is a good example of continuous measure with obvious prognostic impact.

Counts are integers

Some continuous data can simply be counted.

ExampleYour patient…

Counts are said to be integers, which means that they are expressed as number of units (nobody has 1.4 children or presents with 2.7 fractures). In practice, integers have often only a limited number of plausible values to choose from (eg, a person has only a limited number of children or fractures in a life time).

Most basic characteristics of an individual are best described by integer values, eg, age, height, weight, etc:

• You are stabilizing the grade I open femoral shaft fracture of a 49-year-old male using an intramedullary nail.

• During operation, he requires 2 units of packed red blood.

• It takes you 60 minutes to complete the procedure.

Measures are intrinsically nonintegers

Some continuous data require measuring.

ExampleYour patient…

Measures are intrinsically nonintegers because they can take infinity of values expressed by the use of decimals. They can be expressed as integers if they are directly recorded or rounded to their unit (eg, when age is expressed as 49 years instead of 49.2 years).

Data variables stored as numbers thus contain information of different complexity. The complexity of information increases from binary to continuous values. They may describe:

• A certain fact with valuation (someone has 2 or 5 children)

• A certain fact without valuation (someone wears a blue, white, or red jacket)

• A scenario with prognostic impact (someone needs 2 or 5 units of packed red blood)

• Or distinguish between two different clinical situations (a patient with a femoral fracture has a blood pressure of 60/35 mm Hg or 130/80 mm Hg)

3 Numbers to describe the attributes of a group of patients

3.1 Patient listing versus summary statistics

According to Albert Einstein’s famous quote, everything should be made as simple as possible, but not simpler. You spend much time collecting data of varying complexity so, in conducting your analysis, do not hastily rip them to pieces, nor squeeze them into rough classes or categories. In doing so, you may miss subtle, but important associations.

Whenever possible, utilize the full range of information provided by your data.

In a scientific article, and with a small sample size of 20 patients, you may have two different ways (Table 1-1 and Table 1-2) of presenting the demographics of your patients.

Tabulate individual patient data—you will mostly use integer values (Table 1-1).

Strengths Tabulating individual patient data provides the most comprehensive overview of the studied population. It allows presentation of extreme cases (eg, the 84-year-old female), a view of associations between variables, and to recalculate summary statistics.

Limitations Tabulating individual cases may only be possible with small sample sizes and a limited list of measured items. As a rule of thumb, 20–30 patients represent the upper limit.

Tabulate summary statistics—you will often come up with nonintegers(Table 1-2).

Table 1-2 Summary statistics with a group of patients resulting in nonintegers.

Characteristic

Gender (n)

Male

14

Female

6

Mean age (years)

52.7

Mean duration of surgery (minutes)

68.2

Mean number of units of packed red blood

1.3