Difficult Decisions in Thoracic Surgery Mark K Ferguson, Ed Difficult Decisions in Thoracic Surgery An Evidence-Based Approach Mark K Ferguson, MD Professor, Department of Surgery The University of Chicago Head, Thoracic Surgery Service The University of Chicago Hospitals Chicago, IL, USA British Library Cataloguing in Publication Data Difficult decisions in thoracic surgery Chest — Surgery — Decision making Chest — surgery I Ferguson, Mark K 617.5′4 ISBN-13: 9781846283840 ISBN-10: 1846283841 Library of Congress Control Number: 2006926462 ISBN-10: 1-84628-384-1 ISBN-13: 978-1-84628-384-0 e-ISBN 1-84628-470-0 Printed on acid-free paper © Springer-Verlag London Limited 2007 Apart from any fair dealing for the purposes of research or private study, or criticism or review, as permitted under the Copyright, Designs and Patents Act 1988, this publication may only be reproduced, stored or transmitted, in any form or by any means, with the prior permission in writing of the publishers, or in the case of reprographic reproduction in accordance with the terms of licences issued by the Copyright Licensing Agency Enquiries concerning reproduction outside those terms should be sent to the publishers The use of registered names, trademarks, etc in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant laws and regulations and therefore free for general use Product liability: The publisher can give no guarantee for information about drug dosage and application thereof contained in this book In every individual case the respective user must check its accuracy by consulting other pharmaceutical literature springer.com To Phyllis, a decision that has withstood the test of time Preface Why thoracic surgeons need training in decision making? Many of us who have weathered harrowing residencies in surgery feel that, after such experiences, decision making is a natural extension of our selves While this is no doubt true, correct decision making is something that many of us have yet to master The impetus to develop a text on evidence-based decision making in thoracic surgery was stimulated by a conference for cardiothoracic surgical trainees developed in 2004 and sponsored by the American College of Chest Physicians During that conference it became clear that we as thoracic surgeons are operating from a very limited fund of true evidence-based information What was also clear was the fact that many of the decisions we make in our everyday practices are not only uninformed by evidence-based medicine, but often are contradictory to existing guidelines or evidence-based recommendations The objectives of this book are to explain the process of decision making, both on the part of the physician and on the part of the patient, and to discuss specific clinical problems in thoracic surgery and provide recommendations regarding their management using evidence-based methodology Producing a text that will purportedly guide experienced, practicing surgeons in the decision-making process that they are accustomed to observe on a daily basis is a daunting task To accomplish this it was necessary to assemble a veritable army of authors who are widely considered to be experts in their fields They were given the unusual (to many of them) task of critically evaluating evidence on a well-defined topic and provide two opinions regarding appropriate management of their topic: one based solely on the existing evidence, and another based on their prevailing practice, clinical experience, and teaching Most authors found this to be an excellent learning experience It is hoped that the readers of this book will be similarly enlightened by its contents How should a practicing surgeon use this text? As is mentioned in the book, wholesale adoption of the stated recommendations will serve neither physician nor patient well The reader is asked to critically examine the material presented, assess it in the light of his or her own practice, and integrate the recommendations that are appropriate The reader must have the understanding that surgery is a complex, individualized, and rapidly evolving specialty Recommendations made today for one patient may not be appropriate for that same patient in the same situation several years hence Similarly, one recommendation will not serve all patients well The surgeon must use judgment and experience to adequately utilize the guidelines and recommendations presented herein To produce a text with timely recommendations about clinical situations in a world of rapidly evolving technology and information requires that the editor, authors, and vii viii Preface publisher work in concert to provide a work that is relevant and up-to-date To this end I am grateful to the authors for producing their chapters in an extraordinarily timely fashion My special thanks go to Melissa Morton, Senior Editor at Springer, for her rapid processing and approval of the request to develop this book, and to Eva Senior, Senior Editorial Assistant at Springer, for her tireless work in keeping us all on schedule My thanks go to Kevin Roggin, MD, for sharing the T.S Eliot lines and the addendum to them Finally, the residents with whom I have had the opportunity and privilege to work during the past two decades continually reinforce the conviction that quality information is the key to improved patient care and outcomes Mark K Ferguson, MD Contents Preface Contributors Part vii xv Background Introduction Mark K Ferguson Evidence-Based Medicine: Levels of Evidence and Grades of Recommendation Andrew J Graham and Sean C Grondin 13 Decision Analytic Techniques Anirban Basu and Amy G Lehman 21 Nonclinical Components of Surgical Decision Making Jo Ann Broeckel Elrod, Farhood Farjah, and David R Flum 36 How Patients Make Decisions with Their Surgeons: The Role of Counseling and Patient Decision Aids Annette M O’Connor, France Légaré, and Dawn Stacey Part 44 Lung Radiographic Staging of Lung Cancer: Computed Tomography and Positron Emission Tomography Frank C Detterbeck 59 Routine Mediastinoscopy for Clinical Stage I Lung Cancer Karl Fabian L Uy and Thomas K Waddell 68 Management of Unexpected N2 Disease Discovered at Thoracotomy Hyde M Russell and Mark K Ferguson 75 Induction Therapy for Clinical Stage I Lung Cancer David C White and Thomas A D’Amico 82 10 Induction Therapy for Stage IIIA (N2) Lung Cancer Shari L Meyerson and David H Harpole, Jr 88 ix x 11 12 13 14 15 Contents Adjuvant Postoperative Therapy for Completely Resected Stage I Lung Cancer Thomas A D’Amato and Rodney J Landreneau 94 Sleeve Lobectomy Versus Pneumonectomy for Lung Cancer Patients with Good Pulmonary Function Lisa Spiguel and Mark K Ferguson 103 Lesser Resection Versus Lobectomy for Stage I Lung Cancer in Patients with Good Pulmonary Function Anthony W Kim and William H Warren 110 Lesser Resection Versus Radiotherapy for Patients with Compromised Lung Function and Stage I Lung Cancer Jeffrey A Bogart and Leslie J Kohman 119 Resection for Patients Initially Diagnosed with N3 Lung Cancer after Response to Induction Therapy Antonio D’Andrilli, Federico Venuta, and Erino A Rendina 128 16 Video-Assisted Thorascopic Surgery Major Lung Resections Raja M Flores and Naveed Z Alam 140 17 Surgery for Non-Small Cell Lung Cancer with Solitary M1 Disease Robert J Downey 147 18 Thoracoscopy Versus the Open Approach for Resection of Solitary Pulmonary Metastases Keith S Naunheim 151 Unilateral or Bilateral Approach for Unilateral Pulmonary Metastatic Disease Ashish Patel and Malcolm M DeCamp, Jr 158 19 20 Surgery for Bronchoalveolar Lung Cancer Subrato J Deb and Claude Deschamps 21 Lung Volume Reduction Surgery in the Candidate for Lung Transplantation Christine L Lau and Bryan F Meyers 22 Pleural Sclerosis for the Management of Initial Pneumothorax Richard W Light 165 175 186 Part Esophagus 23 Staging for Esophageal Cancer: Positron Emission Tomography, Endoscopic Ultrasonography Jarmo A Salo 24 Induction Therapy for Resectable Esophageal Cancer Sarah E Greer, Philip P Goodney, and John E Sutton 25 Transthoracic Versus Transhiatal Resection for Carcinoma of the Esophagus Jan B.F Hulscher and J Jan B van Lanschot 195 200 208 Contents xi 26 Minimally Invasive Versus Open Esophagectomy for Cancer Ara Ketchedjian and Hiran Fernando 218 27 Lymph Node Dissection for Carcinoma of the Esophagus Nasser K Altorki 225 28 Intrathoracic Versus Cervical Anastomosis in Esophageal Replacement Christian A Gutschow and Jean-Marie Collard 234 29 Jejunostomy after Esophagectomy Lindsey A Clemson, Christine Fisher, Terrell A Singleton, and Joseph B Zwischenberger 242 30 Gastric Emptying Procedures after Esophagectomy Jeffrey A Hagen and Christian G Peyre 250 31 Posterior Mediastinal or Retrosternal Reconstruction Following Esophagectomy for Cancer Lara J Williams and Alan G Casson 258 Postoperative Adjuvant Therapy for Completely Resected Esophageal Cancer Nobutoshi Ando 265 32 33 Celiac Lymph Nodes and Esophageal Cancer Thomas W Rice and Daniel J Boffa 34 Partial or Total Fundoplication for Gastroesophagael Reflux Disease in the Presence of Impaired Esophageal Motility Jedediah A Kaufman and Brant K Oelschlager 271 279 35 Botox, Balloon, or Myotomy: Optimal Treatment for Achalasia Lee L Swanstrom and Michelle D Taylor 285 36 Fundoplication after Laparoscopic Myotomy for Achalasia Fernando A Herbella and Marco G Patti 292 37 Primary Repair for Delayed Recognition of Esophageal Perforation Cameron D Wright 298 38 Lengthening Gastroplasty for Managing Gastroesophagael Reflux Disease and Stricture Sandro Mattioli and Maria Luisa Lugaresi 305 Lengthening Gastroplasty for Managing Giant Paraesophageal Hernia Kalpaj R Parekh and Mark D Iannettoni 318 Management of Zenker’s Diverticulum: Open Versus Transoral Approaches Douglas E Paull and Alex G Little 323 Management of Minimally Symptomatic Pulsion Diverticula of the Esophagus Giovanni Zaninotto and Giuseppe Portale 332 39 40 41 22 Although almost all DAM perform under the auspices of this simple theoretical intuition, any DAM that attempts to model a practical clinical situation can become complicated very quickly Each level of success at the end depends on a sequence of chance outcomes and various intermediate decisions that, in turn, may depend on further chances and decisions Therefore, understanding, defining, and structuring the decision process is essential for clinical decision analysis In this pursuit, a fundamental decision tool is a decision tree that helps the clinician to systematically display the temporal and logical structure of the decision problem and to thus carry out the analysis In order to facilitate explanation of each part of this process in more detail, we begin with a stylized example of clinical decision making in thoracic surgery 3.2 A Motivating Example Let us begin with a common problem in thoracic surgery – the solitary pulmonary nodule – that can illustrate the basic principles of decision analysis Suppose we have a patient referred to a thoracic surgery clinic She is a 65-year-old woman who presents with an incidentally found solitary pulmonary nodule discovered on chest X ray taken in an emergency room after a minor motor vehicle collision She has a 35 pack-year smoking history, but quit smoking years ago She has well-controlled hypertension, no heart disease, and no diabetes She has no previous chest X ray with which to compare to this new abnormal one The patient underwent a computed tomography (CT) scan scheduled by her primary care doctor, and this scan shows an 8mm peripheral nodule with no enlarged lymph nodes Let us further specify that due to the size of her lesion, positron emission tomography (PET) scan is nondiagnostic, and that due to her particular anatomy, she is not a candidate for a video-assisted thorascopic resection (VATS) biopsy How shall we advise this patient? We are now faced with a choice in recommendations: watchful waiting, or perform open diagnostic thoracotomy There are several considerations that inform our decision that have been discussed in detail elsewhere9; namely, we must consider A Basu and A.G Lehman the pretest probability of cancer, the risk of surgical complications, and whether the appearance of the nodule on CT suggest benign or malignant disease Finally, what are the patient’s preferences given the possibility of adverse outcomes of open surgery? A DAM can help in the systematic integration of all this information in order for the surgeon to make an individualized and informed decision A simple, stylized decision tree to address this clinical decision is illustrated in Figure 3.1 We now discuss in detail each part that goes into making up this decision tree and how one would arrive at the final result 3.3 Elements of the Decision Analytic Approach 3.3.1 Identify and Bound the Decision Problem The first step in decision analysis is to understand the decision problem at hand and the particular issues associated with making that decision To this, one must define the set of alternative actions under consideration and the primary outcome measure based on which decision will be made, determine the perspective and the time frame of the analysis, and consider several factors, such as the clinical characteristics of the patient, which may influence the primary outcome measure These considerations can be broadly classified into the following: Define the set of alternative actions Decision analysis always presumes that there is more than one action for the same decision problem If this were not the case, there would be no decision to make Note that one of these alternative actions may, and often does, include the option to nothing In our stylized example, the alternatives are watchful waiting versus thoracotomy Perspective The most common perspective taken in clinical decision making is that of the patient, whose welfare is the fundamental outcome for the decision at stake This is also the perspective we take in our example Chapter (this book) explores the implications of the patient perspective in greater detail Nevertheless, it is easy to foresee that a clinical decision based on a patient’s perspective will be integrally Decision Analytic Techniques 23 Dead (1%) Dead (3% + 12%) Stage I (54%) Comorbidities (Th) (19%) Thoracotomy Lobectomy Payoff (QOL) 0.5*q1 with Comorbidities (Lb) (25%) 0.5*(q1+q2) Alive Cancer (37%) with no Comorbidities (Lb) Stage II Alive Lobectomy Dead (3% + 12%) Alive 0.5*(q1+q1) 0.5*q1 with Comorbidities (Lb) (25%) 0.5*(q1+q2) with no Comorbidities (Lb) 0.5*(q1+q1) Dead (12%) No Comorbidities (Th) 0.5*q1 Alive No Cancer q1 Same as Comorbidities (Th) branch (Outcomes depend on only comorbidities due to Lobectomy) q3 0.5*q3 Alive q3 Dead (12%) 0.5*q3 Alive Watchful Waiting 0.5*q3 Dead (36%) Cancer (37%) Dead (18%) Alive Stage I (54%) q3 Stage II No Cancer FIGURE 3.1 A decision tree to model the clinical decision between thoracotomy and watchful waiting tied with the preferences of that individual patient Comprehending an individual patient’s preferences and incorporating them in the decision process is essential for optimal decision making However, in certain acute conditions, where the patient’s preferences are unknown, alternative preferences of family members or sometimes those of clinicians may be used as proxies More broadly, depending on the types of questions asked, perspectives of different stakeholders – for example, the hospital, the health insurance company, and even society at large – may become relevant in the analyses.10 For example, from the hospital’s perspective, reducing inpatient mortality may be more important than a patient’s potentially diminished quality of life due to side effects of treatments Outcomes such as costs and cost effectiveness may be more relevant to health insurance companies than to individual patients Clinical conditions and demographics of patient(s) Several risk factors affect patient outcomes and therefore are important to consider when choosing between alternative therapies A patient’s clinical condition constitutes the fundamental source of information for appropriate medical care For example, surgeons require clinical information on the stage, and often grade, of a cancer before performing a surgical resection Often, information on comorbidities, such as genetic susceptibilities to malignancy, becomes critical for prescribing appropriate treatment Demographics also play a key role in determining outcomes For example, pretest probability of cancer would depend on a patient’s gender, smoking history, age, and possibly many other factors Time frame of analysis The time frame of analysis should reflect the time over which the consequences of the clinician’s choices have the potential to influence the patient’s survival and 24 quality of life In some acute conditions, the time frame may be the time until the patient is sent home from inpatient care In chronic conditions, the time frame may extend up to the patient’s remaining life expectancy In the clinical choice problem that we illustrate, the ideal time frame should be the lifetime of the patient because both the disease process as well as the potential complications of surgery influence the patient’s quality of life over her entire lifetime However, because our analysis is a mere illustration of the concepts and not a substantive analysis, we will use a time frame of year Primary outcome of interest The optimal clinical decision may vary depending on what type of benefits the patients and/or the clinicians want to maximize Identifying the primary outcome on which the final decision is made is perhaps the most crucial issue in decision analysis Is the primary concern the survival of the patient over the next few months, or is the overall quality of life for the patient over his/her remaining life expectancy the most relevant measure to dwell on? Answering this question often involves incorporating the patient’s preference and perspective, as well as determining the overall goal of medical treatment It also uniquely determines what type of analysis the clinician is interested in The types of analyses can be broadly classified into four categories based on the types of outcomes being evaluated: (1) clinical outcomes; (2) patient’s values about the clinical outcomes; (3) costs, and (4) both costs and outcomes.11 Outcomes Analysis: In such analysis, neither costs nor patient preferences are used to choose the optimal decision Instead, the focus is entirely on one of the clinical outcomes (e.g., survival or length of hospital stay) that are used as the primary outcome of interest The optimal treatment is selected based on the most beneficial clinical outcome (e.g., lowest mortality or shortest length of hospital stay) Utility Analysis: In these analyses, costs are not incorporated in the decision-making process; however, a patient’s preferences are included, in conjunction with the relevant clinical outcomes Most clinical decisions have an effect not only on life and death, but also on the quality of life of patients, mediated through a variety of health states In utility analyses, the value of any par- A Basu and A.G Lehman ticular health state to the patient, popularly known as utility or quality of life (QOL) weight, is measured using either a time-tradeoff or standard gamble method.10,12,13 The utility for any health state is constructed to lie between (representing death) and (representing perfect health) In time-tradeoff methods, patients are asked to trade-off a longer time in a particular health state for a shorter time in perfect health In standard gamble methods, patients are asked to choose between living with a particular health state and a gamble between perfect health and death The utility for the health state under consideration is obtained at the point of indifference between the choices in either method These utilities, multiplied with the duration of time that the person is in that health state under a specific decision choice, form the quality-adjusted lifeyears (QALYs) corresponding to that decision The decision producing the maximum QALYs is chosen to be the optimal one The advantage of utility analyses over outcomes analyses is that a variety of outcomes that influence the patient’s overall quality of life can be summarized using one generic measure such as the QALYs, and therefore the effects of a decision on multiple outcomes can be simultaneously determined and compared to other decisions We use QALYs as the primary outcome of interest in our analysis Cost Analysis: In these analyses, only the costs of alternative treatments form the primary outcome of interest and the least costly treatment is recognized as the optimal choice.14 Such analyses are carried out when the clinical outcomes of the alternative treatments are not a contentious issue – a situation that is becoming increasingly less common in clinical practice Cost-effectiveness Analysis: Cost-effectiveness analysis (CEA) compare both the resources used (costs) and the health benefits achieved (e.g., QALYs or simply life years) among alternative treatments, making these trade-offs explicit to both the clinician and to the patient so that together they can make the optimal decision for the patient.15,16 The practice of cost-effectiveness analysis when comparing two interventions, for example, a new treatment versus standard care, can be summarized as follows: the first step is to calculate the mean costs incurred and the mean Decision Analytic Techniques benefits produced by each intervention; next, an incremental cost-effectiveness ratio (ICER) is formed by dividing the difference in the mean costs over the difference in the mean benefits between the new and standard interventions The ICER represents the additional costs required by the new intervention in producing one extra unit of benefit over that produced by the standard intervention The ICER is then compared with the threshold value that represents the maximum a decision maker is willing to pay for an additional unit of benefit.17 If the ICER is lower than this threshold value then the new intervention is deemed to be cost effective Because, in most cases, a substantial portion of the costs of health care is often borne by health insurance and not the patient, several interesting normative issues arise when an attempt is made to compare both costs and benefits simultaneously A widely debated question revolves around whose perspective is most appropriate to consider when the burden of costs is distributed amongst patients, healthcare providers, and third-party payers Furthermore, obtaining a threshold value that represents the maximum willingness-to-pay for an additional unit of benefit is difficult to ascertain at the individual patient level However, in order to preserve some notion of fairness, one can uniformly apply a societal threshold to all patients Such discussions are beyond the scope of this chapter but interested readers are encouraged to explore this important literature 5,6 Other considerations Several other considerations may influence treatment choices They include information on how much weight patients place on outcomes that will arise in the future compared to those at present time (this is popularly summarized by the concept of the discount rate),10,11 whether consideration of the effects of patients outcomes on the family members are important,18 and how patient demographic characteristics, health insurance status, and out-ofpocket payments influence the primary outcome of interest.19,20 3.3.2 Structure the Decision Problem over Time Although choosing between alternative actions is the primary goal of a decision analysis, often the 25 decision problem will involve a temporal sequence of choices that inevitably influence the fi nal choice of action Moreover, these choices may themselves depend on certain chance outcomes that may or may not be controlled by previous decisions Therefore, the second step in decision analysis is to identify the components of the decision problem To this, one defines a structure for the temporal and logical sequence of choices, chances, and outcomes, as well as their interactions, which would lead to the final outcome of interest as defined in the previous step A decision tree helps to structure this temporal decision problem in a systematic format An excellent primer for building decision trees is given by Detsky and colleagues.21 Figure 3.1 illustrates the decision tree for our simple example on the choice between proceeding to open thoracotomy and watchful waiting Each point in the decision tree that leads to multiple outcomes or decisions is called a node There are two types of nodes: (1) a decision node is indicated by a square box, and (2) a chance node is indicated by a circle A decision node represents the alternative choices that are available to the clinician A chance node represents the alternative outcomes and patient’s responses that are possible Each chance node is associated with a probability with which a specific outcome is realized, thereby accounting for the inherent uncertainties in such processes The rightmost column of the decision tree illustrates the final payoffs or outcomes associated with each possible branch of the decision tree Each branch can be visualized as a level of success given the initial choice of treatment, and is defined by the sequence of events starting with the initial treatment choice and leading up to the final outcome In our example, we assume that the clinician makes a decision aiming to maximize the patient’s QALYs over the next 1-year period after considering the potential risks and benefits of each choice If the patient undergoes open thoracotomy, the benefits will include defi nitive diagnosis (cancer or no cancer), as well as pathological staging and a potential curative resection (lobectomy) with or without adjuvant therapy if the nodule is indeed malignant The risks include possible morbidity and mortality from both thoracotomy and 26 lobectomy (if needed) These outcomes generally depend on a number of issues, including the experience of the surgeon and the patient’s other medical comorbidities, as well as the risk of development of other chronic problems such as a post-thoracotomy chronic pain syndrome that would affect the patient’s quality of life If the patient undergoes watchful waiting, she avoids the comorbidities associated with thoracotomy However, the patient is then subjected to a higher mortality risk if the nodule is indeed malignant; she may also suffer from the anxiety of not knowing whether she has cancer, and therefore a reduced quality of life Please note that the time frame of year does not permit the increased risk of mortality associated with a cancer diagnosis and the choice of watchful waiting to become fully realized in the real world We have arbitrarily chosen this time frame to simplify our tree, and facilitate the model Therefore, the reader should be cautioned to not use this model for actual clinical decision making, as it is not clinically accurate The non-cancer yearly mortality risk applies irrespective of the choice The associate payoffs include the quality of life weights of the patient Death from thoracotomy happens at the beginning of the period and so is assigned a QOL value of Death from lobectomy or natural death is assumed to occur at the middle of the year; so half year of life is weighted by the patient’s QOL during that time Under the thoracotomy arm, if the patient stays alive for the year, her QOL depends on the presence or absence of comorbidities Under watchful waiting, death from cancer or natural death is also assumed to occur at the middle of the year If the patient stays alive, then her quality of life is determined by her level of anxiety about cancer Note again, that the model in Figure 3.1 is illustrative and uses a very simplified and stylized version of an actual decision-making process Several additional factors, such as the sensitivity and specificity of detecting cancer through thoracotomy, differences between clinical staging, and the true pathological stage and recurrence of cancer post-lobectomy, to name a few, must be considered for developing a comprehensive model that can appropriately represent this clinical situation A Basu and A.G Lehman 3.3.3 Characterize the Information to Fill in the Structure Once the clinician has identified the sequence of choices to be made and the sequence of chance nodes and their associated outcomes determined by these choices, the next step is to obtain information on these chances and outcomes Choices at each step of the decision process can be made based on this information This is the critical part of the decision model because the quality of information that goes into these chance nodes entirely determines the credibility of the decision model and the decision it generates Bayes’ formula provides the key theoretical insight for determining what types of information are required.22,23 The classic example for application of the Bayes’ formula lies in the interpretation of results from a diagnostic test For example, given that a diagnostic test correctly detects a clinical problem 90% of the time (specificity – a measure of prior belief ) and correctly detects the absence of a clinical problem 85% of the time (sensitivity – also a measure of prior belief ), what is the probability that the patient has the problem (posterior belief about success in diagnosis) given a positive or negative test result (evidence)? This question is readily answered using the Bayes’ formula We now provide a more intuitive discussion on the Bayes’ formula A clinician is often interested in knowing about the probability of success (at any chance node) associated with a treatment decision for a patient, based on the current evidence on success rates with that decision However, the clinician can only observe, from published research and his or her own experience, the likelihood of evidence given some underlying prior belief about success rates Bayes’ formula, as shown below, helps the clinician to go from the latter quantity to the former one Pr(Success Evidence) Pr( Evidence Success) × Pr(Success) = Pr( Evidence) (3.2) where Pr(x|y) represents probability of x conditional on or given y Here, Pr(Evidence|Success) indicates the likelihood (or probability) of observing the data that clinicians observe in practice or in clinical trials given a specific Decision Analytic Techniques hypothesis about the prior belief on success rates Often, there will be uncertainty regarding the true success rates and there may be more than one prior belief However, once new evidence (e.g., observed data in clinical trials) is revealed, the likelihood of that evidence under a variety of prior beliefs becomes known Consequently, one can feel more confident to restrict the beliefs on success rates to those ranges that correspond to the highest likelihood for the evidence These new updated beliefs then form the evidencebased posterior beliefs based on which that particular chance node in the model can be informed Note that although this exposition of Bayes’ formula suggests a prospective process of updating beliefs and information about specific parameters in the model, it also can be readily applied to generate a current estimate of success rates that seem most likely given all the evidence that has accumulated to date Thus, the role of evidence is important because more evidence would tend to strengthen the posterior beliefs and dispel a larger part of uncertainty associated with prior beliefs Incorporating all relevant evidence into the model is accomplished by a detailed search on the published and possibly unpublished research literature that point to evidence for the specific parameters in the model Posterior estimates of parameters are generally weighted by the quality of evidence as defi ned by sample size, 27 study design, and use of robust analytical methods Information gathering for decision analysis almost always uses one or more of the following: literature review, meta-analysis, primary data collection, and consultation with experts.5,6 The relevant information for our model and their sources are outlined in Table 3.1 Because we are using a stylized example, we will use point estimates for our model parameters that reasonably lie within the widely disparate ranges that are reported in the literature concerning morbidity and mortality from thoracic procedures In a more formal treatment of this problem, one has to pay considerable attention to pooled and metaanalyzed available evidence so as to obtain estimates that more closely reflect true values and also properly account for uncertainties from multiple sources We discuss these sorts of issues further in section 3.5 Attention should also be paid to the timeliness of information For example, estimates given in Table 3.1 not account for the fact that published literature shows a consistent trend over time toward lower morbidity and mortality from thoracic procedures, probably resulting from a combination of improved technique, improved anesthesia, and improved performance by dedicated thoracic surgeons, as well as the vast improvements in adjuvant chemotherapy therapy now available to patients TABLE 3.1 Information on parameters for the choice between thoracotomy and watchful waiting Description Pr(death due to thoracotomy) Pr(cancer|patient characteristics) Pr(stage I|cancer, patient characteristics) Pr(death|lobectomy, stage I cancer) a Pr(co-morbidities|lobectomy, no death) Pr(death|lobectomy, stage II cancer) a Pr(comorbidities|thoracotomy, no death) Pr(death|watchful waiting, stage I cancer) Pr(death|watchful waiting, stage II cancer) Pr(death|watchful waiting, no cancer, patient characteristics) QOL(alive with no comorbidity & no cancer) QOL(comorbidities due to lobectomy) = q2 QOL(comorbidities due to thoracotomy) = q1 QOL(alive with the anxiety of knowing that cancer might be present) = q3 QOL(death) Value References 1% 37% 54% 3% + 12% 25% 5% + 12% 19% 18% 36% 12% = 0.9*q1 0.90 0.80 24 25 26 27 28 27 29 30, 31 30, 31 32, 33, 34 Standard Assumed Assumed Internal data Standard We model this probability as inclusive of the probability of natural death given no cancer, i.e., Pr(death|lobectomy, stage I cancer) = Pr(death due to lobectomy) + Pr(death|watchful waiting, no cancer, patient characteristics) a 28 3.3.4 Apply Decision Analysis The final step is to synthesize the information gathered and the choices made in the process of decision making This will allow us to obtain estimates of the primary outcome of interest that can be directly compared, in order to choose among alternative actions This is accomplished by working backwards, that is, from right to left, on our decision tree for which we have fi lled in probability values for nodes At decision nodes, we roll back (alternatively, the term fold back is also used5) along the best choices – in effect, the choices that maximize gain or minimize harm At chance nodes, we average out along all branches, yielding an expected value, such as QALYs, expected number of days of hospital stay, or expected risk of postoperative survival, etc Consider the expected QALYs of lobectomy for the patient after she survives thoracotomy with comorbidities and is detected with stage I cancer and undergoes lobectomy Let the q1 and q2 be utility weights for comorbidities due to thoracotomy and lobectomy, respectively As mentioned previously, we also assume that death occurs on average at the middle of the year The levels of outcomes possible are (1) death from lobectomy (payoff = 0.5*q1); (2) survive with comorbidities from lobectomy (payoff = 0.5*q1 + 0.5*q2); and (3) survive without comorbidities from lobectomy (payoff = q1) The overall expected payoff is then calculated using the formula in Equation 3.1, by multiplying the corresponding probability of each level of outcome (Table 3.1) with it corresponding payoffs This process is repeated for each possible level of outcome under thoracotomy and under watchful waiting Based on the parameter estimates in Table 3.1, thoracotomy produces an expected QALY of 0.899 while watchful waiting produces an expected QALY of 0.731 This is our baseline result, which reveals that thorocotomy may be the optimal decision for this patient 3.3.5 Sensitivity Analysis Clinical decision making using a DAM is usually followed by an assessment of the strength of the A Basu and A.G Lehman final decision that can be accomplished by quantitatively assessing the sensitivity of the final decision to the structural assumptions of the models and feasible alternative information sets This is done by substituting a range of values for those parameters that are believed to be the most variable in practice If the conclusions of the model are robust over a range of values for each node, then one can feel very comfortable with the decision that the model suggests If the decision changes as values are varied, then one must be pay closer attention to those parameters and try to get a better sense of the parameters for the case at hand This exercise will produce a threshold value where two decisions stand at equipoise In this latter case, the strength of the data used to generate the parameter estimates becomes exceedingly important In our case, we vary the QOL of the patient anxiety and that of the comorbidities of thoracotomy to see how our baseline result change Figure 3.2 shows the phase diagram for this sensitivity analysis In the diagram, the y-axis represents different levels of QOL for anxiety, while the x-axis represents different levels of QOL for thoracotomy-related comorbidities The area in the graph identifies the regions where, conditional of the respective QOL weights, either thoracotomy or watchful waiting is more beneficial As evident from Figure 3.2, a patient with high levels of anxiety [i.e., low QOL(Anxiety)] would benefit from thoracotomy A patient with low anxiety [i.e., high QOL(Anxiety)] but low QOL(complications of thoracotomy) would benefit from watchful waiting Such phase diagrams can help clinicians identify where their patient lie in this graph so that they can make an informed decision about treatment choices Phase diagrams, as in Figure 3.2, can be produced based on several other parameters in the model However, multiple phase diagrams can easily make the decision as complicated as it was to begin with, and hence lose the utility of DAMs Such scenarios, where there are considerable uncertainties and heterogeneity in several parameters in the model, can be addressed using probabilistic analysis, which we discuss in section 3.5 Decision Analytic Techniques 29 FIGURE 3.2 Sensitivity analysis of the baseline QALY result with respect to QOL weights for patient’s anxiety and of thoracotomy-related comorbidities The shaded and the blank areas in the graph identify the values of QOL weights for which thoracotomy or watchful waiting produces the maximum expected QALY, respectively 3.4 Clinical Decision Analysis in Thoracic Surgery Heretofore, DAMs have been used to address a number of issues and/or problems in thoracic surgery Many of these issues and controversies will be more thoroughly explored in the context of this book Some examples of models that have been published in the past include those that evaluate different surgical techniques, 35 whether or not to proceed to more invasive and/or aggressive treatments given a common problem in thoracic surgery, 36,37 and appropriate management of certain types of metastases, 38 to name a few The perspectives that have been used include: the patient’s perspective (as we have, and is most common in clinical DAMs); the treating hospital’s perspective; and the government’s perspective, via Medicare and Medicaid programs Many DAMs use clinical conditions of patients that have been somewhat simplified, although this may change as older, sicker patients undergo thoracic procedures, and new data are generated about their outcomes Time frames tend to be dominated by traditional markers of success or failure in surgery; that is, 30-day morbidity and mortality or 1- to 5-year survival There are centers, 39 however, that have published long-term data Such information will become increasing relevant and important as outcomes of interest expand to include measures like QALYs and costeffectiveness ratios that usually require long time frames for appropriate conclusions to be drawn These longer term time horizons are more commonly found when a DAM incorporates data from sources like the Surveillance, Epidemiology, and End Results database or the Veterans Administration Hospitals extensive database, where patients are followed for many years Several different primary outcomes of interest have been measured, including mortality, particular morbidities for certain operations, and length of hospital stay, as well as QOL, cost effectiveness, and patient satisfaction The chief problem with many of the published DAMs revolves around the fact that almost all point estimates are generated from a few retrospective studies or blinded, prospective, randomized trials, mostly comprised of a handful of patients The robustness of these estimates can be greatly improved by meta-analyzing the data across studies and properly accounting for the uncertainties arising out of different sources For example, one DAM compares three choices concerning optimal management for patients undergoing esophagectomy for carcinoma of the esophagus40 and bases all probabilities on a single, randomized, prospective trial that compared the use of pyloroplasty with non-use.41 However, in the 72 patients enrolled, there was no statistically significant difference demonstrated between 30 A Basu and A.G Lehman populations In addition, there was little exploration of the potential negative side effects of both procedure non-use and use from the patient’s perspective; that is, which might decrease quality of life more: delayed gastric emptying or dumping syndrome? Though the model is very insightful, it is also quite hypothetical – just as our example is in this chapter Therefore, to overcome these sorts of limitations, one can employ more advanced techniques that have been developed by practitioners of decision analysis 3.5 Advanced Issues in Decision Analytic Techniques The discussions below are an attempt to provide a general understanding of the advanced issues in modeling and analysis It does not, in any way, serve to provide a comprehensive review of these topics More importantly, we recommend that clinicians interested in taking advantage of these advanced techniques consult with professionals who are well versed with the nuances of these methods 3.5.1 Markov Models Although decision trees provide intuitive representation of the disease process and the consequences of choice over a short time period, they can get extremely complicated when extended over long periods of time To address this situation, one can use a Markov process that is a modeling technique based on matrix algebra The fundamental idea behind these models is that, instead of considering health state transitions over a short period of time, as in decision trees, a Markov process is concerned with transitions during a series of short time intervals or cycles For example, consider the watchful waiting arm of the decision tree in Figure 3.1 One can revise the model to include a bi-annual follow-up for detecting whether the lesion is growing or not The probability that the clinician would find an advanced lesion in the next follow-up will be based on the natural progression of cancer over time In order to model the progression of cancer, the cancer can be delineated into mutually exclusive health states based on its stage and grade An example of such a model, borrowed from our colleague David Meltzer’s work in prostate cancer, is shown in Figure 3.3, also known as a bubble diagram Each health state, defined by a combination of stage and grade, is represented by a bubble The figure represents progression of cancer by stage and grade leading up to cancerrelated death The patient begins in one of these bubbles and in every cycle, with some transition probabilities, either stays in that bubble or moves to another bubble representing an advanced health state The sum of all transition probabilities in a cycle must add up to one because, by definition, a patient has to be in one of the mutually exclusive health states The transition probabilities can vary with time (or number of cycles) and also with patient characteristics such as age of diagnosis The transition probabilities determine the progression of the disease over Stage I Low Grade Stage I Mod Grade Stage I High Grade Stage II Low Grade Stage II Mod Grade Stage II High Grade Metastatic Low Grade Metastatic Mod Grade Death from Cancer Metastatic High Grade FIGURE 3.3 Structure of a Markov model for cancer progression Decision Analytic Techniques time, and therefore are critical in evaluating the costs and outcomes of any intervention at a particular point in time We refer interested readers to some of the applications of Markovian processes in decision models available in the literature.42–46 3.5.2 Probabilistic Analysis Uncertainty remains an integral part of any analysis Although a part of this uncertainty is reflected in the probability estimates used in decision models (e.g., such as those in Table 3.1), they not reflect the overall uncertainty for an outcome For example, if we say that Mrs X has a 40% chance of having cancer, we make a statement about a population comprised of millions of Mrs X clones in which 40% of them will have cancer and 60% will not When the clinician is faced with one Mrs X from that population, her true cancer status is not known with certainty But by incorporating the point estimate of 40%, the clinician can characterize the uncertainty that she might fall in the part of the population who has cancer This type of model, as the one we illustrate above, is called a deterministic model as one uses predetermined point estimates for probabilities to reflect the uncertainty However, we seldom know how many patients in that population truly have cancer; instead, we rely on a random sample from the population to determine how many patients in that sample have cancer If we find 40% of patients in the sample have cancer, it gives us a reasonable estimate for the population, but not the exact estimate That is, 35% of patients in the population may truly have cancer, while in the random sample we have chosen, 40% show up with cancer The bigger the sample size, the closer will be the sample estimate to the population estimate Hence, there remains a degree of uncertainty about the true value of the population parameter because we almost always infer based on a fraction of that population The traditional way to deal with uncertainty is to a perform sensitivity analysis by varying the parameter of interest across a range of values and observing the changes in outcomes associated with it, as we have shown in our stylized example However, there are several limitations to this 31 method One-way or two-way sensitivity analysis may severely undermine the effect of uncertainty in a multiparameter model Multiple-way sensitivity analysis can easily become extremely complicated as to disallow straightforward interpretation Even in the absence of complicated analysis, traditional sensitivity analysis can identify the optimal decision if only the true values of the parameters are known, as, for example, in the case of patient preferences that can be directly measured This poses problems for the clinicians who may often fi nd it hard to use sensitivity analysis results in clinical decision making due to lack of guidance and uncertainty about the true value of many clinical parameters Probabilistic analyses help to characterize these uncertainties in the model parameters and to arrive at the final decision by simultaneously averaging out the uncertainties in multiple parameters In these models, instead of specific point estimates of different parameters in the model, one specifies distributions for these input parameters where these distributions might center on the point estimates Using Monte Carlo techniques, a patient is then propagated through the model several times (iterations), each time with random values of a parameter drawn from its respective distribution The costs and outcomes are then averaged over all iterations and compared across treatment options A good example of a probabilistic analysis can be found Andrew Brigg’s work on gastroesophageal reflux disease.47 3.5.3 Bayesian Meta-analysis Bayesian meta-analysis is a sophisticated hierarchical modeling approach that utilizes the concept of Bayes’ theorem discussed earlier It summarizes and integrates the fi ndings of research studies in a particular area One can obtain the posterior distribution of Pr(Success| Evidence), that is the current distribution of belief about success rates given all the past evidence, and directly use it as an input in a probabilistic analysis Such an approach provides a combined analysis of studies that indicate the overall strength of evidence for a success while properly accounting for multiple sources of uncertainty 32 For example, suppose there are four studies with varying sample sizes indicating that a woman with a smoking history similar to that in our example has a pretest probability of cancer of 60%, 45%, 30%, and 75%, respectively Perhaps we also know, from even bigger studies, that women are 1.5 times more likely than men to have a lung cancer unrelated to their smoking, while a person with a significant smoking history is times more likely to have cancer than a neversmoker A Bayesian meta-analysis can combine this prior information on the effects of gender and smoking history with the data from the four samples to produce the posterior distribution of the pretest probability of cancer for this female patient given all relevant evidence With the advent of the Society of Thoracic Surgeons General Thoracic Surgery Database, these methods will be most relevant for the field of thoracic surgery in order to synthesize information arising out of a large number of prospectively collected data Several books on Bayesian estimation and meta-analysis have been written and several examples of Bayesian meta-analysis can be found in the literature Interested readers are encouraged to explore Peter Congdon’s book on this topic.48 Of interest to the readers of this book may be the work by Tweedie and colleagues, who perform a Bayesian meta-analysis of the published literature to determine the association between incidence of lung cancer in female never smokers and exposure to environmental tobacco smoke.49 Bayesian meta-analysis can be most conveniently implemented using freely available software called the WinBUGS (http://www.mrcbsu.cam.ac.uk/bugs) 3.5.4 Value of Information Analyses As we discussed earlier, probabilistic models often can provide a more efficient way to address uncertainty in decision models It provides the clinician with a sense of the strength of evidence for an optimal decision after accounting for multiple sources of uncertainty It also provides the basis for conducting value of information analysis on specific parameters In these analyses, uncertainties in parameter estimates that trans- A Basu and A.G Lehman late into uncertainties surrounding the outcome of interest can be used to establish the value of acquiring additional information by conducting further research Information is valuable because it reduces the expected costs of uncertainty surrounding a clinical decision The expected costs of uncertainty are determined by the probability that a treatment decision based on existing information will be wrong and by the consequences (or costs) if the wrong decision is made The expected costs of uncertainty can also be interpreted as the expected value of perfect information (EVPI), because perfect information (an infinite sample) can eliminate the possibility of making a wrong decision It is also the maximum a decision maker should be willing to pay for additional evidence to inform this decision in the future Such analyses can identify research priorities by focusing on those parameters where more precise estimates would be most valuable.50–52 Additionally, an analogous type of calculation may be done on heterogeneous parameters such as quality of life weights and patient’s preferences and attitudes towards treatment Knowing this information can help clinicians make individualized decisions by incorporating patients’ values and preferences into the process of making treatment decisions Concretely, clinicians might pursue this approach using a variety of decision aids that are designed to facilitate transmission of information on patients’ values and preferences to physicians These decision aids are often expensive in terms of program development and implementation as well as in time costs for patients and physicians.53 Therefore, in order to decide how to best allocate limited resources towards decision aids and similar approaches to implementing individualized care, it is important to have information on the potential social value of such endeavors and which dimensions of patient preference are most valuable to elicit This is accomplished by the Expected Value of Individualized Care (EVIC) analysis.54 EVIC represents the expected costs of ignorance of patientlevel heterogeneity It is the potential value of research, as compared to optimal population level decision making, to elicit information on heterogeneous parameters so that individualized information about each patient can be conveyed to the physician Decision Analytic Techniques 3.6 Conclusions Decision analytic modeling is a systematic approach to integrating information about a variety of different aspects of clinical decision making It helps the clinician to make an objective decision by incorporating both the population-level evidence on outcomes and also the individual-level preferences in the decisionmaking process It is worthwhile to issue a reminder at this point that the sole purpose of DAM is to provide information in a systematic way so as to facilitate decision making based on best and comprehensive evidence Decision analytic modeling, although viewed by some clinician and researchers as prescriptive, only prescribes the consideration of a decision to be optimal and should not be viewed as the final prescription for the patient Decision analytic modeling in no way serves to make a decision for the clinician, who must utilize his/her own expertise to interpret the evidence placed in front him/ her and determine the relevance of this evidence in the context of an individual patient’s case The tension inherent in any DAM lies between the desire to create a model that is conceptually rich – to more accurately reflect the complexity of real-world decision making – that is also somewhat data poor, versus creating a more simple model which utilizes few, highly reliable estimates However, if real-world decision making does demand a richer model, we believe that it is more appropriate to develop such a model, even if subjective decisions and estimates inform some parts of this model We say this because often in real-world clinical practice, such subjective decisions and estimates play a part in clinical decision making Decision analytic modeling helps to translate these subjective estimates into a formal statement about uncertainty and help clinicians understand the implications of these uncertainties in the decision process It also helps clinicians to identify priorities in research areas and patient preferences that are most critical for patient outcomes Most DAMs in thoracic surgery have focused on information from clinical data Such information is certainly crucial and forms the backbone of a DAM However, any attempt to translate efficacy information available from clinical trials 33 and other studies into effectiveness for a given patient requires a careful examination of patient preferences and behavior Therefore, creating models that accurately reflect the choices at hand requires a fusion of several types of information This information can be crudely divided into two categories – information that is derived from biological science, and that derived from social science As we learn more about the genetic behavior and cell biology of tumors, as well as genetic information about individual patients, clinical decision making will most certainly be impacted But this sort of information must be contextualized within the study of actual human behaviors – information that is captured by fields as diverse as behavioral and classical economics to statistics to political science to epidemiology Consequently, it is imperative that surgeons and practitioners in these other fields collaborate if DAMs are to function as nuanced, effective tools to improve the clinical outcomes, and indeed the lives, of patients Acknowledgments We are grateful to Caleb G Alexander, William Dale, and Mark Ferguson at the University of Chicago for helpful comments and suggestions References Meltzer D Can medical cost-effectiveness analysis identify the value of research? In: Murphy KM, Topel RH, eds Measuring the Gains from Medical Research: An Economic Approach Chicago: The University of Chicago Press; 2003:206–247 Sackett DL, Rosenberg WM, Gray JA Evidence based medicine: what it is, what it isn’t BMJ 1996;312:71–72 Coutler A Partnerships with patients: the pros and cons of shared clinical decision-making J Health Serv Res 1997;2:112–121 Pauker SG, Kassirer JP Decision analysis N Engl J Med 1987;316:250–257 Weinstein MC, Fineberg HV Clinical Decision Making Philadelphia: Saunders; 1980 Petitti DB Meta-analysis, Decision Analysis and Cost-Effectiveness Analysis New York: Oxford University Press; 1994 Detsky AS, Naglie G, Krahn MD, et al Primer on medical decision analysis: getting started Med Decis Making 1997;17:123–125 34 Eddy D Clinical Decision Making: From Theory to Practice Boston: Jones & Bartlett; 1996 Ost D, Fein A Management strategies for the solitary pulmonary nodule Curr Opin Pulm Med 2004;10:272–278 10 Gold MR, Siegel JE, Russell LB, Weinstein MC Cost-effectiveness in Health and Medicine New York: Oxford University Press; 1996 11 Drummond M, Stoddard G, Torrence G Methods of Economic Evaluation of Health Care Programmes Oxford: Oxford University Press; 1987:6– 26 12 Torrence GW Utility approach to measuring health-related quality of life J Chron Dis 1987;40: 593–600 13 Torrence GW Measurement of health state utilities for economic appraisal: a review J Health Econ 1986;5:1–30 14 Lowson KV, Drummond MF, Bishop JM Costing new services: long-term domiciliary oxygen therapy Lancet 1981; ii:1146–1149 15 Weinstein MC, Stason WB Foundation of costeffectiveness analysis for health and medical practices N Eng J Med 1977;296:716–721 16 Garber A, Phelps C Economic foundations of costeffectiveness analysis J Health Econ 1997;16:1–32 17 Weinstein M, Zeckhauser R Critical ratios and efficient allocation J Pubic Econ 1973;2:147–158 18 Basu A, Meltzer D Implications of spillover effects within the family for medical cost-effectiveness analysis J Health Econ 2005;24:751–773 19 Alexander GC, Casalino LP, Meltzer DO Patientphysician communication about out-of-pocket costs JAMA 2003;290:953–958 20 Alexander GC, Casalino LP, Meltzer DO Physician strategies to reduce patients’ out-of –pocket prescription costs Arch Intern Med 2005;165:633– 636 21 Detsky AS, Naglie G, Krahn MD, et al Primer on medical decision analysis: building a tree Med Decis Making 1997;17:126–135 22 Bayes T An essay towards solving a problem in the doctrine of chances Philos Trans R Soc 1763;330– 418 Taken from: Biometrika 1958;45:293–315 23 Pratt JW Bayesian interpretation of standard inference statements (with discussions) J R Stat Soc B 1965;27:169–203 24 Dominioni L, Imperatore A, Rovera F, et al Stage I non-small cell lung carcinoma, analysis of survival and implications for screening Cancer 2000;89:2334–2344 25 Swensen SJ, Silverstein MD, Ilstrup DM, et al The probability of malignancy in solitary pulmonary nodules Application to small radiologically inde- A Basu and A.G Lehman 26 27 28 29 30 31 32 33 34 35 36 37 38 terminate nodules Arch Intern Med 1997;157:849– 855 Wisnivesky JP, Yankelevitz D, Henschke CI Stage of lung cancer in relation to its size Chest 2005;127:1136–1139 Ginsberg RJ, Hill LD, Eagan RT, et al Modern thirty-day operative mortality for surgical resections in lung cancer J Thorac Cardiovasc Surg 1983;86:654–658 Stéphan F, Boucheseiche S, Hollande J, et al Pulmonary complications following lung resection: a comprehensive analysis of incidence and possible risk factors Chest 2000;118:1263–1270 Barrera R, Shi W, Amar D, et al Smoking and timing of cessation: impact on pulmonary complications after thorocatomy Chest 2005;127:1977– 1983 Chadha AS, Ganti AK, Sohi JS, et al Survival in untreated early stage non-small cell lung cancer Anticancer Res 2005;25:3517–3520 McGarry RC, Song G, des Rosiers P, et al Observation-only management of early stage, medically inoperable lung cancer: poor outcome Chest 2002;121:1155–1158 U.S Department of Health and Human Services The Health Consequences of Smoking: A Report of the Surgeon General Washington, DC: U.S Department of Health and Human Services, Centers for Disease Control and Prevention, National Center for Chronic Disease Prevention and Health Promotion, Office on Smoking and Health; 2004 Arias E United States Life Tables, 2002 National Vital Statistics Reports Hyattsville, MD: National Center for Health Statistics; 2004:53 English DR, Holman CDJ, Milne E, et al The Quantification of Drug Caused Morbidity and Mortality in Australia Canberra: Commonwealth Department of Human Services and Health; 1995 Ferguson MK, Lehman AG Sleeve lobectomy or pneumonectomy: optimal management strategy using decision analysis techniques Ann Thorac Surg 2003;76:1782–1788 Morimoto T, Tsuguya F, Koyama H, et al Optimal strategy for the fi rst episode of primary spontaneous pneumothorax in young men J Gen Intern Med 2002;17:193–202 Falcoz PE, Binquet C, Clement F, et al Management of the second episode of spontaneous pneumothorax: a decision analysis Ann Thorac Surg 2003;76:1843–1848 Porter G, Cantor S, Walsh G, et al Costeffectiveness of pulmonary resection and systemic Decision Analytic Techniques 39 40 41 42 43 44 45 46 chemotherapy in the management of metastatic soft tissue sarcoma: a combined analysis from the University of Texas M.D Anderson and Memorial Sloan-Kettering Cancer Centers J Thorac Cardiovasc Surg 2004;127:1366–1372 Martini N, Rusch V, Bains, MS et al Factors influencing ten-year survival in resected stages I to III a non-small cell lung cancer J Thorac Cardiovasc Surg 1999;117:32–36 Olak J, Detsky A Surgical decision analysis: esophagectomy/esophagogastrectomy with or without drainage? Ann Thorac Surg 1992;53:493–497 Cheung HC, Siu KF, Wong J Is pyloroplasty necessary in esophageal replacement by stomach? A prospective, randomized controlled trial Surgery 1987;102:19–24 Goldie SJ, Kuntz KM, Weinstein MC, et al Costeffectiveness of screening for anal squamous intraepithelial lesions and anal cancer in human immunodeficiency virus-negative homosexual and bisexual men Am J Med 2000;108:634–641 Goldie SJ, Weinstein MC, Kuntz KM, et al The costs, clinical benefits, and cost-effectiveness of screening for cervical cancer in HIV-infected women Ann Intern Med 1999;130:97–107 Goldie SJ, Kuhn L, Denny L, et al Policy analysis of cervical cancer screening strategies in lowresource settings: clinical benefits and costeffectiveness JAMA 2001;285:3107–3115 Zeliadt SB, Etzioni RD, Penson DF, et al Lifetime implications and cost-effectiveness of using fi nasteride to prevent prostate cancer Am J Med 2005;118:850–857 Manser R, Dalton A, Carter R, et al Costeffectiveness analysis of screening for lung cancer 35 47 48 49 50 51 52 53 54 with low dose spiral CT (computed tomography) in the Australian setting Lung Cancer 2005;48:171–185 Briggs AH, Goeree R, Blackhouse G, et al Probabilistic analysis of cost-effectiveness models: choosing between treatment strategies for gastroesophageal reflux disease Med Decis Making 2002;22:290–308 Congdon P Bayesian Statistical Modeling New York: Wiley; 2001 Tweedie RL, Scott DJ, Biggerstaff BJ, et al Bayesian meta-analysis, with application to studies of ETS and lung cancer Lung Cancer 1996;14(suppl 1):S171–S194 Meltzer D Addressing uncertainty in medical cost-effectiveness analysis: implications of expected utility maximization for methods to perform sensitivity analysis and the use of costeffectiveness analysis to set priorities for medical research J Health Econ 2001;20:109–129 Claxton K The irrelevance of inference: a decision making approach to the stochastic evaluation of health care technologies J Health Econ 1999;18: 341–364 Claxon K, Neumann PJ, Araki S, et al Bayesian value of information analysis: An application to a policy model of Alzheimer’s disease Int J Tech Assoc Health Care 2001;17:38–55 Kennedy AD, Sculpher MJ, Coulter A, et al Effects of decision aids for menorrhagia on treatment choices, health outcomes, and costs: a randomized controlled trial JAMA 2002;288:2701–2708 Basu A Value of information on preference heterogeneity and individualized care In: Three Essays in Cost-Effectiveness Analysis [master’s thesis] Chicago: University of Chicago; 2004 Nonclinical Components of Surgical Decision Making Jo Ann Broeckel Elrod, Farhood Farjah, and David R Flum Examining surgical trends before the National Emphysema Treatment Trial (NETT) demonstrates the importance of nonclinical determinants of care The number of lung volume reduction surgery (LVRS) claims increased dramatically after 1994 despite the fact that there was considerable uncertainty in the available evidence base.1 Favorable media reports and testimonials from patient advocacy groups may have influenced both patient and surgeon attitudes about LVRS.2 Some surgeons felt that investigations prior to the NETT demonstrated clear and dramatic improvements in quality of life, sufficient to justify Medicare reimbursement for the procedure.3 Accordingly, they believed the NETT was a form of coercion because patients who refused to enroll in the study would not have financial coverage of their LVRS or receive the operation from a NETT surgeon Furthermore, even if patients enrolled, the study deprived half of them a procedure with “established” benefits Surgeons less comfortable with this level of scientific uncertainty may have decided against performing the procedure Nonsurgeon observers proposed that surgeons were motivated by financial gains, as the procedure was relatively inexpensive and reimbursement was generous.2 In addition to potential patient and surgeon influence, third-party coverage had an effect on decision making as evidenced by the dramatic decrease in the number of operations upon suspension of Medicare reimbursement in December 1995.4 Because many third-party payers base their coverage plans on Centers for Medicare and Medicaid Services (CMS) guidelines, this policy likely affected many non-Medicare patients and provid- 36 ers as well Whether surgeons stopped performing the operation because of lack of reimbursement or as an acknowledgement of scientific uncertainty is unclear It is clear, however, that the sharp decline in the number of procedures was temporally related to CMS intervention Subsequent CMS policy partly limited surgical decision making because reimbursement was limited to eligible patients and surgeons Professional organizations can also play a role in decision making by effectively regulating surgeon-directed clinical practice in the setting of clinical uncertainty For example, through educational and advisory statements, the American Society of Colorectal Surgeons strongly influenced its membership to avoid performing laparoscopic procedures for colorectal cancer despite the use of these interventions by many surgeons for benign disease There was no similar “prohibition” by thoracic surgical professional organizations in 1994, and these circumstances may have “permitted” surgeon-level, nonevidence–based decision making to flourish with LVRS The case of LVRS reveals that many nonclinical factors involving patients, surgeons, and the practice environment can influence surgical decision making 4.1 Methodology for Evaluating Nonclinical Factors of Decision Making Previous investigations of nonclinical factors influencing clinical decision making have used qualitative or semiquantitative research meth- ... making Chest — surgery I Ferguson, Mark K 617 .5′4 ISBN -1 3 : 97 818 46283840 ISBN -1 0 : 18 462838 41 Library of Congress Control Number: 2006926462 ISBN -1 0 : 1- 8 462 8-3 8 4 -1 ISBN -1 3 : 97 8 -1 -8 462 8-3 8 4-0 e-ISBN... subsequent clinical application of insulin in 19 22, leading to the awarding of a Nobel prize to Banting and Macleod in 19 23 Fleming’s rediscovery of the antibacterial properties of penicillin in 19 28.. .Difficult Decisions in Thoracic Surgery Mark K Ferguson, Ed Difficult Decisions in Thoracic Surgery An Evidence-Based Approach Mark K Ferguson, MD Professor, Department of Surgery The