DOC PREVIEW
Primer on Medical Decision Analysis

This preview shows page 1-2-3 out of 10 pages.

Save
View full document
View full document
Premium Document
Do you want full access? Go Premium and unlock all 10 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 10 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 10 pages.
Access to all documents
Download any document
Ad free experience
Premium Document
Do you want full access? Go Premium and unlock all 10 pages.
Access to all documents
Download any document
Ad free experience

Unformatted text preview:

Primer on MedicalDecisionAnalysis:Part 2-Building a TreeALLAN S. DETSKY, MD, PhD, GARY NAGLIE, MD,MURRAY D. KRAHN, MD, MSc, DONALD A. REDELMEIER,DAVID NAIMARK, MDMD, MS(HSR),This part of a five-part series covering practical issues in the performance of decisionanalysis outlines the basic strategies for building decision trees. The authors offer sixrecommendations for building and programming decision trees. Following these sixrecommendations will facilitate performance of the sensitivity analyses required toachieve two goals. The first is to find modeling or programming errors, a process knownas “debugging” the tree. The second is to determine the robustness of the qualitativeconclusions drawn from the analysis. Key words: decision analysis; expected value;utility; sensitivity analysis; decision trees; probability. (Med Decis Making 1997;17:126-135)SoftwareWe teach students to build their trees usingSMLTREE, a DOS-based software package (SMLTREE.Hollenberg JP. Version 2.9. Roslyn, NY). Thoughmany other software packages (e.g., DECISION MAKER,Pratt Medical Group, Boston, MA) are available,SMLTREE is widely used by practitioners as well asstudents of decision analysis and comes with an ex-cellent tutorial teaching the students the nuts andbolts of programming a tree. Because this series isintended to be a practical “how to” guide, some ofthe discussion, particularly the discussion related todebugging the tree, is not directly applicable to solv-ing decision problems using spreadsheets, influencediagrams, or other software packages.Decision Analysis Example: Giant CellArterltisWe use one clinical scenario throughout the restof the series: the choice of management strategiesReceived November 27, 1995, from the University of TorontoProgramme in Clinical Epidemiology and Health Care Research(The Toronto Hospital and The Sunnybrook Health Science Cen-tre Units) and the Departments of Health Administration, Medi-cine, and Clinical Biochemistry, Toronto, Ontario, Canada. Re-vision accepted for publication December 12, 1996. Drs. Detskyand Redelmeier are partially supported by Career Awards fromthe National Health Research and Development Programme andthe Ontario Ministry of Health, respectively. Drs. Naglie andKrahn are partially supported by Arthur Bond Fellowships fromthe Physicians’ Services Incorporated Foundation.Address correspondence and reprint requests to Dr. Detsky:EN G-246, General Division, The Toronto Hospital, 200 ElizabethStreet, Toronto, ON M5G 2C4, Canada. e-mail: ([email protected]).126for patients presenting with clinical features thatsuggest giant cell arteritis (GCA). This example is amodification of data found elsewhere.’Giant cell arteritis is a vasculitis that affects largeand medium-sized vessels, mostly in elderly pa-tients. Patients’ symptoms may include headache,fever, and fatigue. When confronted with such a pa-tient, clinicians are faced with diagnostic and treat-ment options. Giant cell arteritis can lead to a verysevere complication of blindness. Steroids are saidto decrease the risk of blindness but come with therisk of side effects such as hypertension, fluid reten-tion, and avascular necrosis of bone. There is a testfor GCA, i.e., biopsy of the temporal artery can revealvasculitis in the specimen. However, the sensitivityof that test is not ideal.We focus on three of the strategies compared inthe paper by Buchbinder and Detsky1 treating nopatient with steroids, treating all patients with ste-roids, and performing a temporal artery biopsy andtreating positive cases only.The Six RecommendationsIn the following section we illustrate six recom-mendations or tips for building a decision model.We have developed these tips for our students andfind that if they are followed, it is much easier tobuild a tree that “functions” appropriately whenperforming sensitivity analyses. The use of sensitivityanalyses to “debug” the tree and determine the ro-bustness of the conclusion is discussed in Part 4 ofthis series.’ For most of the recommendations weshow examples of “mistakes” and “correct” ways ofmodeling the tree. We assume that the reader is fa-VOL 17/NO 2, APR-JUN 1997 Primer on Medical Decision Analysis-2 l 127miliar with the usual methods of pictorial display ofdecision trees.3,4RECOMMENDATION 1The tree must have balance. Real clinical prob-lems represent tradeoffs between risks and benefits.The structure of outcomes in a decision analysismust reflect such a tradeoff. If one of the strategicoptions in the model carries all of the risks andnone of the benefits, or, alternatively, all of the ben-efits and none of the risks, then either the tree isnot a valid model of the clinical problem or the clin-ical problem does not require a decision analysis.Figure 1A shows an example of a model without bal-ance.Imagine that we are comparing two strategies: 1)treating patients with a specific disease (e.g., GGA) toavoid an adverse outcome (called a “bad outcome”;in this case, blindness1 and 2) not treating the pa-tient. The structure of outcomes in both cases in-cludes the possibility of the bad outcome or a goodoutcome. This is represented after the first proba-bility node in both the upper and theBad Outcome80, RxbF&PeoRxrGO, RxlmFIGURE IA. Decision tree without balance.pBOBx = probability of bad outcome with treat-mentpGOBx = probability of good outcome with treat-mentPBO= probability of bad outcome withouttreatmentPGO= probability of good outcome withouttreatmentUBO = utility of bad outcomeUGO = utility of good outcomeThe bad outcome is blindness. The good outcomeis no blindness.lowerbranches of figure 1A. The expression underneaththe line represents the probability of the occurrenceof that event. If the tree is modeled such that treatedpatients have a smaller chance of a bad outcomethan untreated patients and if the treatment has nodown side (e.g., risk of a side effect, inconvenienceof compliance with medication), then this tree hasno balance. The upper branch clearly dominates thelower branch because it contains all of the benefitsand none of the risks.Figure 1B models the same clinical problem usinga tree with balance. First, the therapy arm is asso-ciated with a new outcome, the possibility of a majorside effect from steroids, such as avascular necrosisof bone. Second, the utilities now reflect not onlygood and bad outcomes but also the presence orabsence of a minor side effect such as fluid reten-tion and a new term


Primer on Medical Decision Analysis

Download Primer on Medical Decision Analysis
Our administrator received your request to download this document. We will send you the file to your email shortly.
Loading Unlocking...
Login

Join to view Primer on Medical Decision Analysis and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view Primer on Medical Decision Analysis 2 2 and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?