Surgical Decision Analysis: Esophagectomy/Esophagogastrectomy With or Without Drainage? Jemi Olak, MD, and Allan Detsky, MD Departments of Surgery and Medicine, University of Toronto, Toronto, Canada
Decision analysis was used to compare three management strategies for patients undergoing esophagogastrectomy for carcinoma of the esophagus or gastric cardia: drain all patients with either pyloromyotomy or pyloroplasty, drain no patient, or perform a test that stratifies patients into high-risk and low-risk groups for development of gastric outlet obstruction and drain the high-risk group. Results indicate that a ”drain all” approach is
appropriate in clinical settings where the risk of gastric outlet obstruction is greater than lo%, as long as the drainage procedure is 95% effective. If a test were developed to stratify patients, it would have to have a sensitivity of 80% when its specificity was loo%, and would require a higher sensitivity as the specificity fell below 100%. (Ann Thoruc Surg 1992;53:493-7)
I
Material and Methods Quest ion
n 1948, Dragstedt and Camp [l]observed that symptoms of gastric stasis developed in 25% of patients who had undergone truncal vagotomy for duodenal ulcer disease. They thus advised the addition of a drainage procedure to truncal vagotomy for the treatment of duodenal ulcer disease. Whether patients with an intrathoracic vagotomized stomach after esophagectomy or esophagogastrectomy for carcinoma are at similar risk for development of symptoms of gastric stasis remains controversial. Some surgeons believe that drainage is as necessary in this setting as after truncal vagotomy for duodenal ulcer [2-71. Other surgeons drain selectively [8, 91 or not at all [lo-121, believing that the intrathoracic stomach empties well and that the risk of harmful bile reflux exceeds the possible benefit or need for a drainage procedure. The controversy continues because the few For editorial comment, see page 373. studies that have addressed the issue include too few patients, short follow-up times, different types of operation, and different drainage procedures [2]. This article will apply decision analysis to the dilemma using a computer software program to create a model that explores all of the treatment options taking into account their attendant risks and benefits. Sensitivity analysis of the variables is undertaken to define thresholds to aid the clinician in choosing a therapeutic approach. To determine the best alternative for an individual patient, the risks and benefits of the drainage procedure and the accuracy of predicting an individual’s risk must be considered. Accepted for publication Oct 9, 1991. Address reprint requests to Dr Olak, Division of Cardiothoracic Surgery, Department of Surgery, Medical College of Virginia, MCV Station, PO Box 68, Richmond, VA 23298-0068.
0 1992 by The
Society of Thoracic Surgeons
Should all, some, or no patients undergoing esophagectomy or esophagogastrectomy with esophagogastrostomy for carcinoma of the esophagus or gastric cardia have a drainage procedure performed at the time of initial operation?
Model DECISION TREE. The decision tree is illustrated in Figure 1. The model compares three strategic choices: treat all patients, treat no patient, or test patients and treat only those at high risk. The goal is to minimize the proportion of patients in whom any complication develops either as a result of gastric outlet obstruction or as a result of the drainage procedure. If a clinician chooses the first branch of the tree, all patients will have a drainage procedure performed. This will expose all patients to the risk of a drainage-associated complication. If the procedure is not 100% efficacious, some patients will also develop gastric outlet obstruction. The signs and symptoms of gastric outlet obstruction and its sequelae are as follows: gastroesophageal reflux, regurgitation, aspiration pneumonitis, pneumonia, hoarseness, bezoar formation, stasis gastritis, gastric ulceration or bleeding, early satiety, bloating or eructation, and inability to maintaidregain weight. If a clinician chooses the drain none branch of the tree, the outcome is determined by the risk of gastric outlet obstruction alone. If a clinician chooses the test branch, patients will be placed into high-risk and low-risk groups for development of gastric stasis based on their test result. Those considered to be at high risk will have a drainage procedure. Possible outcomes for these patients are similar to those in the drain all group, although the risk may be greater. Those considered to be at low risk will have a
0003-4975/92/$5.00
494
Ann Thorac Surg 1992;53:49>7
OLAK AND DETSKY ESOI'HAGECTOMY WITH OR WITHOUT DRAINAGE?
Fig 1 . Decision free of three approaches to gastric outlet drainage in patients undergoing operation for carcinoma of the esophagus or gastric cardia. (UDAC = utility of drainage-associated complication; UGOC = utility of gastric outlet complication; UNC = utility of no complication.)
I
OVTLErOBsmm
UQOC 1
Y
UNC 8 UGOC 9
1 NOOBSTRUCTION less than or equal chance of development of gastric outlet obstruction compared with patients in the drain none branch providing the test is valid. At the present time there exists no valid, reliable method of distinguishing between patients at high risk for development of gastric outlet obstruction and those at low risk. In this article, therefore, we propose a test based on clinical characteristics. Because the sensitivity and specificity of the test are unknown at this time, we will use sensitivity analysis to determine how accurate it would have to be to be useful. The questions in the test are largely based upon intuitive assessment (Appendix 1). There is no risk associated with the test. TEST.
DRAINAGE PROCEDURE. Patients will undergo FredetRamstedt pyloromyotomy or Heineke-Mickulicz pyloroplasty. The pyloromyotomy involves dividing the serosal and muscular layers over the anterior surface of the pylorus for a distance of approximately 4 cm and can be completed in 5 to 10 minutes. A pyloroplasty, on the other hand, involves making a full-thickness longitudinal incision through the anterior wall of the pylorus and proximal duodenum and reapproximating the edges in a transverse fashion using a one- or two-layer closure. The latter procedure takes 15 to 20 minutes to perform. Complications associated with the procedure may be technical (ie, leakage from the drainage site, abscess) or mechanical (ie, dumping syndrome, duodenogastric reflux).
Table 1 summarizes the variables and estimates of their values that will be used in the decision analysis; these values are based on critical review of the available literature. The incidence of symptomatic gastric outlet obstruction after esophagectomy or esophagogastrectomy has been variously estimated to be VARIABLES AND PROBABILITIES.
UNC 10
between 10% and 24% [2, 121. We have chosen an incidence of gastric outlet obstruction of 14%.A randomized clinical trial of drainage versus no drainage by Cheung and associates [2] found the incidence of gastric outlet obstruction without drainage to be 14% (3/21). The effectiveness of the drainage procedure in preventing symptomatic gastric outlet obstruction was 100% in the same randomized clinical trial [2]. For the decision analysis, we will consider the drainage procedure to be 95% effective. Complications associated with the addition of a drainage procedure to esophagectomy or esophagogastrectomy can be categorized as technical or mechanical, as already mentioned. The randomized trial by Cheung and associates [2] found the technical complication rate of pyloroplasty to be zero but did not objectively assess the incidence of duodenogastric reflux. Cheung and associates stated that the incidence of dumping was 10%in their series. We will use a drainage-associated complication rate of 10% in the analysis. Estimates of the test's sensitivity (0.75) and specificity (0.50) are conservative because the test is unproved. A Table 1 . Variables in Decision Tree Variable Pretest likelihood of gastric outlet obstruction (C) Effectiveness of drainage procedure (e) Probability of drainage-associated complication (c) Sensitivity of the test (sens) Specificity of the test (spec)
Estimate (%)
14 95 10
75 50
Ann Thorac Surg
OLAK AND DETSKY ESOPHAGECTOMY WITH OR WITHOUT DRAINAGE?
1992;53493-7
two-way sensitivity threshold analysis approach will be used to ascertain the sensitivity and specificity required of a test to make it a worthwhile strategy upon which to base treatment. To quantify the value that individuals place on each of these health states, we use the concept of utility. This numerical method captures the preferences of individuals for these health states, based on utility theory. This theory bases rational decision making on a number of axioms that describe how individuals ought to behave when faced with decision making under uncertainty [13]. Each of the health states at the terminal nodes are thus assigned a utility value that quantifies preferences for these states. In the context of the present decision tree, the relative utility of each outcome is compared with the optimal outcome (no complication), which is assigned a utility of 1.0. The utilities used in the analysis are presented in Table 2. The utility of a patient having gastric outlet obstruction has been assigned a value of 0.0 because this is the complication that surgeons are trying to eliminate by performing a drainage procedure at the time of initial esophageal resection. The utility of a patient experiencing a complication related to the drainage procedure has been somewhat arbitrarily assigned a value of 0.7 as the complication rate and severity of a complication when it does occur is less morbid. UTILITIES.
Analytic Strategies Using the decision analysis computer software ”SMLTREE,” analyses were conducted using baseline values listed in Table 1. The expected utility of a particular branch is a weighted average of the utility associated with all possible outcomes of that branch. For example, the drain none branch in Figure 1 is associated with two possible outcomes, gastric outlet obstruction and no gastric outlet obstruction. The expected utility of gastric outlet obstruction is determined by multiplying the rate times the utility (0.14 x 0 = O.O), and the expected utility of no obstruction is determined by multiplying the rate times its utility (0.86 x 1.0 = 0.86). Using SMLTREE we can determine the strategy that will yield the greatest expected utility by a procedure referred to as folding back. BASELINE ANALYSIS.
SENSITIVITY “THRESHOLD” ANALYSIS. The effect that changes in baseline variables have on the expected utility of each branch is studied by conducting a series of sensitivity analyses. The threshold point in clinical decision making is defined as the point where two different approaches to managing a patient appear to have the same potential value (eg, are expected to result in equal complication rates) and
Table 2. Utilities in the Decision Tree Utility Name No complication Drainage-associated complication Gastric outlet complication
Utility 1.0 0.7 0.0
495
Table 3. Result of Decision Tree Expected Utility
Value
Drain all Drain none Test
0.96321 0.86000 0.94386
Expected utility (choose) Choose: drain all
0.96321
therefore the clinician is indifferent about choosing either of the two strategies [14]. There are two possible thresholds: a “testing threshold’ where the decision is made between draining no one or draining only high-risk patients, and a “test-drain all threshold” where the decision is made between draining only high-risk patients or draining all patients. To determine the threshold values of pretest likelihood of gastric outlet obstruction (C)and effectivenessof the drainage procedure (e) (the two variables that we believe will vary according to clinical circumstances)that will make clinicians indifferent to the choice of strategies, we plug the baseline values assumed for all of the variables (except C and e) into the three expressions for the expected total hospital complicationrate associated with each strategy (see Fig 1).We then set two of the three expressions equal to each other (ie, drain none = test, test = drain all), which results in an equation and two variables: C and e. This is referred to as a two-way sensitivity or threshold analysis. It shows us how sensitive the choice of strategy is to simultaneous variations in the two variables (C and e).
Results The result of the decision analysis is presented in Table 3. A drain all policy has the highest expected utility (0.96321) using the baseline variables in Table 1 and is thus the preferred approach. One-way sensitivity analysis (Fig 2) indicates that a drain all policy is appropriate if the incidence of gastric outlet obstruction is greater than lo%, keeping all other variables constant. Figure 3 graphs a one-way sensitivity analysis, illustrating that as long as the effectiveness of the drainage procedure exceeds 40%, keeping all other variables constant, a drain all policy is once again favored. Figure 4 plots the test threshold and test-drain all threshold allowing the incidence of gastric outlet obstruction and the effectiveness of the drainage procedure to vary from 0% to 100%. Clearly, for clinicians facing an incidence of gastric outlet obstruction greater than lo%, a drain all policy is appropriate if the drainage procedure works 95% of the time. A test would be useful if the drainage procedure were only 20% to 60% effective in averting gastric outlet obstruction in the setting where its anticipated incidence was between 10% and 20%. Figure 5 plots the test-drain all sensitivity threshold across the entire range of specificities. To be helpful, the test would need to remain at least 80% sensitive even if it were 100%specific. If the sensitivity was 0.6, for example, a drain all policy would be appropriate through the entire
496
OLAK AND DETSKY ESOPHAGECTOMY WITH OR WITHOUT DRAINAGE?
Ann Thorac Surg 1992;53493-7
1
E T
H
R E
0.8
8
H
0 0.0 1 0
U 0.4
T I
ddn
0
0 0.4 C
T Y
R I I
I
I
1
I
I
I
8 0.2
1
0.2 0.9 0.4 0.1 0.8 0.7 0.8 0.0 RISK OF QASTRIC OUTLET OBSTRUCTION
0
DRAIN ALL
+DRAIN NONE
1
K
t
0
*TEST
0
0.4 0.8 EFFECTIVENESS
0.8
1
Fig 4 . Two-way sensitivity analysis varying the efectiveness of the drainage procedure against the incidence of gastric outlet obstruction
Fig 2 . One-way sensitivity analysis of the incidence of gastric outlet obstruction versus its expected utility for each treatment strategy.
(GOC).
range of possible specificities. To date, no test has been assessed with regard to its ability to predict risk of postoperative gastric outlet obstruction. The wide range of clinical management of this problem and the absence of reasonable criteria for adopting a policy on gastric drainage suggest that the likelihood of such a test being developed is small.
Comment This decision analysis represents an attempt to assess the relative merits of draining the intrathoracic stomach in patients undergoing esophagectomy. Although a randomized clinical trial [2] has addressed this question in the past, the number of patients randomized was too small for any definitive conclusion to be made. Despite its
1 ,
0.2
methodological shortcomings, the trial [2] did advocate a drain all policy. A multicentered randomized clinical trial approach would be necessary to answer this question conclusively, however, because no one center could be expected to accumulate enough experience within a reasonable time period. Another approach to the question involves decision analysis. The analysis presented in this article demonstrates that with an incidence of gastric outlet obstruction of 14% and a drainage procedure that is 95% effective, a drain all policy is the most appropriate. The two-way sensitivity analysis presented in Figure 4 demonstrates that a drain all policy should apply in most clinical settings. Allowing the incidence of gastric outlet obstruction and the effectiveness of the drainage proce-
1
T H
R
1
0.8
0.9-
T Y
0.0
!
I
I
0
0.2
I
I
I
I
0.4 0.8 0.8 EFFECTIVENESS OF DRAINAGE
-DRAINALL
+TEST
*DRAINNONE
1
I
Fig 3 . One-way sensitivity analysis of the effctiveness of drainage versus its expected utility for each treatment strategy.
0
0.2
0.4 0.8 SPECIFICITY
0.8
Fig 5. Two-way sensitivity analysis of the specificity versus sensitivity for the test. The line represents the test-drain all threshold.
1
OLAK AND DETSKY ESOPHAGECTOMY WITH OR WITHOUT DRAINAGE?
Ann Thorac Surg 1992:53:49>7
dure to vary between 0% and 100% permits clinicians to decide which approach fits their situation. A test would be considered appropriate for clinical situations where both the incidence of gastric outlet obstruction is less tha10% and the effectiveness of the drainage procedure ranges between 20% and 60%. This is represented by the area entitled test in Figure 4. It can be seen from this figure that the area entitled drain all encompasses the majority of clinical scenarios likely to be encountered in clinical practice. The two-way sensitivity analysis presented in Figure 5 demonstrates that given an incidence of obstruction of 14%, an effectiveness of drainage of 95%, and a drainageassociated complication rate of lo%, if a test were to be developed with the aim of placing patients into high- and low-risk groups it would require a sensitivity of at least 80% even if it was 100% specific. It is unlikely that the test could be this accurate. Although a test seems rational, the analysis presented here indicates that it would be very difficult to devise a test that would be useful for patient stratification. If the probabilities used in this analysis reflect those at your institution, then consideration should be given to performing a drainage procedure in all patients undergoing esophagogastrectomy. We thank Drs Martin F. McKneally and Andrew S. Wechsler for reviewing the manuscript.
References 1. Dragstedt LR, Camp EH. Follow-up of gastric vagotomy alone in the treatment of peptic ulcer. Gastroenterology 1948;11:460-5. 2. Cheung HC, Siu KF, Wong J. Is pyloroplasty necessary in esophageal replacement by stomach? A prospective, randomized controlled trial. Surgery 1987;102:19-24. 3. Akiyama H,Tsurumaru M, Kawamura T, et al. Principles of surgical treatment for carcinoma of the esophagus. Ann Surg 1981;194438-46. 4. Belsey RHR. Palliative management of esophageal carcinoma. Am J Surg 1980;139:789-94. 5. Ong GB,Kwong KH. The Lewis-Tanner operation for cancer of the oesophagus. J R Coll Surg Edinb 1969;143-19. 6. Ellis FH Jr. Cancer of the esophagus and cardia-role of surgery in palliation. Postgrad Med 1984;75:139-48. 7. Hopkins RA, Alexander JC, Postlethwait RW. Stapled esophagogastric anastomosis. Am J Surg 1984;147283-7.
497
8. McKeown KC. Carcinoma of the oesophagus. In: Taylor S,
ed. Recent advances in surgery, vol8. Edinburgh: Churchill Livingstone, 1973:133-65. 9. McKeown KC. Trends in oesophageal resection for carcinoma. With special reference to total oesophagectomy. Ann R Coll Surg Engl 1972;51:21%39. 10. Hinder RA. The effect of posture on the emptying of the intrathoracic vagotomized stomach. Br J Surg 1976;63:5814. 11. Huang GI, Wu YK. Operative technique for carcinoma of the esophagus. In: Juang GJ, Wu YK, ed. Carcinoma of the esophagus and gastric cardia. Berlin: Springer-Verlag, 1984. 31M. 12. Ujiki GT, Pearl GJ, Poticha S, Sisson GA Sr, Shields TW. Mortality and morbidity of gastric ‘pull-up’ for replacement of the pharyngoesophagus. Arch Surg 1987;12644-7. 13. Torrance GW, Feeny D. Utility and quality adjusted lifeyears. Int J Techno1 Assess Health Care 1989;5:559-75. 14. Paulker SG, Kassler JP. The threshold approach to clinical decision making. N Engl J Med 1980;302:1109-17.
Appendix 1. Test for Risk of Postoperative Gastric Outlet Obstruction” History, physical examination Peptic ulcer disease Suspected Proven (roentgenogram, endoscopy) Symptoms of early satiety or bloating Coexisting disease Diabetes Connective tissue disease UGI series: demonstration of Duodenal bulb deformity Duodenal ulcer Pyloric channel narrowing Gastric dilatation Retained food Radionudide study: delayed gastric emptying Endoscopy: demonstration of Duodenal scamng/fibrosis Peptic ulcer Tight pylorus Gastritislduodenitis Intraoperative: demonstration of Duodenal scarring Pyloric hypertrophy
+1 +2 +1 +1 +I +2 +2 +2 +1 +1 +2 +2 +2 +2
+1 +2
+2
Scoring of the test: any +2 or combination of two +I’s is considered a positive test a
Decision analysis was used to compare three management strategies for patients undergoing esophagogastrectomy for carcinoma of the esophagus or gastric cardia: drain all patients with either pyloromyotomy or pyloroplasty, drain no patient, or perform a test that stratifies patients into high-risk and low-risk groups for development of gastric outlet obstruction and drain the high-risk group. Results indicate that a ”drain all” approach is
appropriate in clinical settings where the risk of gastric outlet obstruction is greater than lo%, as long as the drainage procedure is 95% effective. If a test were developed to stratify patients, it would have to have a sensitivity of 80% when its specificity was loo%, and would require a higher sensitivity as the specificity fell below 100%. (Ann Thoruc Surg 1992;53:493-7)
I
Material and Methods Quest ion
n 1948, Dragstedt and Camp [l]observed that symptoms of gastric stasis developed in 25% of patients who had undergone truncal vagotomy for duodenal ulcer disease. They thus advised the addition of a drainage procedure to truncal vagotomy for the treatment of duodenal ulcer disease. Whether patients with an intrathoracic vagotomized stomach after esophagectomy or esophagogastrectomy for carcinoma are at similar risk for development of symptoms of gastric stasis remains controversial. Some surgeons believe that drainage is as necessary in this setting as after truncal vagotomy for duodenal ulcer [2-71. Other surgeons drain selectively [8, 91 or not at all [lo-121, believing that the intrathoracic stomach empties well and that the risk of harmful bile reflux exceeds the possible benefit or need for a drainage procedure. The controversy continues because the few For editorial comment, see page 373. studies that have addressed the issue include too few patients, short follow-up times, different types of operation, and different drainage procedures [2]. This article will apply decision analysis to the dilemma using a computer software program to create a model that explores all of the treatment options taking into account their attendant risks and benefits. Sensitivity analysis of the variables is undertaken to define thresholds to aid the clinician in choosing a therapeutic approach. To determine the best alternative for an individual patient, the risks and benefits of the drainage procedure and the accuracy of predicting an individual’s risk must be considered. Accepted for publication Oct 9, 1991. Address reprint requests to Dr Olak, Division of Cardiothoracic Surgery, Department of Surgery, Medical College of Virginia, MCV Station, PO Box 68, Richmond, VA 23298-0068.
0 1992 by The
Society of Thoracic Surgeons
Should all, some, or no patients undergoing esophagectomy or esophagogastrectomy with esophagogastrostomy for carcinoma of the esophagus or gastric cardia have a drainage procedure performed at the time of initial operation?
Model DECISION TREE. The decision tree is illustrated in Figure 1. The model compares three strategic choices: treat all patients, treat no patient, or test patients and treat only those at high risk. The goal is to minimize the proportion of patients in whom any complication develops either as a result of gastric outlet obstruction or as a result of the drainage procedure. If a clinician chooses the first branch of the tree, all patients will have a drainage procedure performed. This will expose all patients to the risk of a drainage-associated complication. If the procedure is not 100% efficacious, some patients will also develop gastric outlet obstruction. The signs and symptoms of gastric outlet obstruction and its sequelae are as follows: gastroesophageal reflux, regurgitation, aspiration pneumonitis, pneumonia, hoarseness, bezoar formation, stasis gastritis, gastric ulceration or bleeding, early satiety, bloating or eructation, and inability to maintaidregain weight. If a clinician chooses the drain none branch of the tree, the outcome is determined by the risk of gastric outlet obstruction alone. If a clinician chooses the test branch, patients will be placed into high-risk and low-risk groups for development of gastric stasis based on their test result. Those considered to be at high risk will have a drainage procedure. Possible outcomes for these patients are similar to those in the drain all group, although the risk may be greater. Those considered to be at low risk will have a
0003-4975/92/$5.00
494
Ann Thorac Surg 1992;53:49>7
OLAK AND DETSKY ESOI'HAGECTOMY WITH OR WITHOUT DRAINAGE?
Fig 1 . Decision free of three approaches to gastric outlet drainage in patients undergoing operation for carcinoma of the esophagus or gastric cardia. (UDAC = utility of drainage-associated complication; UGOC = utility of gastric outlet complication; UNC = utility of no complication.)
I
OVTLErOBsmm
UQOC 1
Y
UNC 8 UGOC 9
1 NOOBSTRUCTION less than or equal chance of development of gastric outlet obstruction compared with patients in the drain none branch providing the test is valid. At the present time there exists no valid, reliable method of distinguishing between patients at high risk for development of gastric outlet obstruction and those at low risk. In this article, therefore, we propose a test based on clinical characteristics. Because the sensitivity and specificity of the test are unknown at this time, we will use sensitivity analysis to determine how accurate it would have to be to be useful. The questions in the test are largely based upon intuitive assessment (Appendix 1). There is no risk associated with the test. TEST.
DRAINAGE PROCEDURE. Patients will undergo FredetRamstedt pyloromyotomy or Heineke-Mickulicz pyloroplasty. The pyloromyotomy involves dividing the serosal and muscular layers over the anterior surface of the pylorus for a distance of approximately 4 cm and can be completed in 5 to 10 minutes. A pyloroplasty, on the other hand, involves making a full-thickness longitudinal incision through the anterior wall of the pylorus and proximal duodenum and reapproximating the edges in a transverse fashion using a one- or two-layer closure. The latter procedure takes 15 to 20 minutes to perform. Complications associated with the procedure may be technical (ie, leakage from the drainage site, abscess) or mechanical (ie, dumping syndrome, duodenogastric reflux).
Table 1 summarizes the variables and estimates of their values that will be used in the decision analysis; these values are based on critical review of the available literature. The incidence of symptomatic gastric outlet obstruction after esophagectomy or esophagogastrectomy has been variously estimated to be VARIABLES AND PROBABILITIES.
UNC 10
between 10% and 24% [2, 121. We have chosen an incidence of gastric outlet obstruction of 14%.A randomized clinical trial of drainage versus no drainage by Cheung and associates [2] found the incidence of gastric outlet obstruction without drainage to be 14% (3/21). The effectiveness of the drainage procedure in preventing symptomatic gastric outlet obstruction was 100% in the same randomized clinical trial [2]. For the decision analysis, we will consider the drainage procedure to be 95% effective. Complications associated with the addition of a drainage procedure to esophagectomy or esophagogastrectomy can be categorized as technical or mechanical, as already mentioned. The randomized trial by Cheung and associates [2] found the technical complication rate of pyloroplasty to be zero but did not objectively assess the incidence of duodenogastric reflux. Cheung and associates stated that the incidence of dumping was 10%in their series. We will use a drainage-associated complication rate of 10% in the analysis. Estimates of the test's sensitivity (0.75) and specificity (0.50) are conservative because the test is unproved. A Table 1 . Variables in Decision Tree Variable Pretest likelihood of gastric outlet obstruction (C) Effectiveness of drainage procedure (e) Probability of drainage-associated complication (c) Sensitivity of the test (sens) Specificity of the test (spec)
Estimate (%)
14 95 10
75 50
Ann Thorac Surg
OLAK AND DETSKY ESOPHAGECTOMY WITH OR WITHOUT DRAINAGE?
1992;53493-7
two-way sensitivity threshold analysis approach will be used to ascertain the sensitivity and specificity required of a test to make it a worthwhile strategy upon which to base treatment. To quantify the value that individuals place on each of these health states, we use the concept of utility. This numerical method captures the preferences of individuals for these health states, based on utility theory. This theory bases rational decision making on a number of axioms that describe how individuals ought to behave when faced with decision making under uncertainty [13]. Each of the health states at the terminal nodes are thus assigned a utility value that quantifies preferences for these states. In the context of the present decision tree, the relative utility of each outcome is compared with the optimal outcome (no complication), which is assigned a utility of 1.0. The utilities used in the analysis are presented in Table 2. The utility of a patient having gastric outlet obstruction has been assigned a value of 0.0 because this is the complication that surgeons are trying to eliminate by performing a drainage procedure at the time of initial esophageal resection. The utility of a patient experiencing a complication related to the drainage procedure has been somewhat arbitrarily assigned a value of 0.7 as the complication rate and severity of a complication when it does occur is less morbid. UTILITIES.
Analytic Strategies Using the decision analysis computer software ”SMLTREE,” analyses were conducted using baseline values listed in Table 1. The expected utility of a particular branch is a weighted average of the utility associated with all possible outcomes of that branch. For example, the drain none branch in Figure 1 is associated with two possible outcomes, gastric outlet obstruction and no gastric outlet obstruction. The expected utility of gastric outlet obstruction is determined by multiplying the rate times the utility (0.14 x 0 = O.O), and the expected utility of no obstruction is determined by multiplying the rate times its utility (0.86 x 1.0 = 0.86). Using SMLTREE we can determine the strategy that will yield the greatest expected utility by a procedure referred to as folding back. BASELINE ANALYSIS.
SENSITIVITY “THRESHOLD” ANALYSIS. The effect that changes in baseline variables have on the expected utility of each branch is studied by conducting a series of sensitivity analyses. The threshold point in clinical decision making is defined as the point where two different approaches to managing a patient appear to have the same potential value (eg, are expected to result in equal complication rates) and
Table 2. Utilities in the Decision Tree Utility Name No complication Drainage-associated complication Gastric outlet complication
Utility 1.0 0.7 0.0
495
Table 3. Result of Decision Tree Expected Utility
Value
Drain all Drain none Test
0.96321 0.86000 0.94386
Expected utility (choose) Choose: drain all
0.96321
therefore the clinician is indifferent about choosing either of the two strategies [14]. There are two possible thresholds: a “testing threshold’ where the decision is made between draining no one or draining only high-risk patients, and a “test-drain all threshold” where the decision is made between draining only high-risk patients or draining all patients. To determine the threshold values of pretest likelihood of gastric outlet obstruction (C)and effectivenessof the drainage procedure (e) (the two variables that we believe will vary according to clinical circumstances)that will make clinicians indifferent to the choice of strategies, we plug the baseline values assumed for all of the variables (except C and e) into the three expressions for the expected total hospital complicationrate associated with each strategy (see Fig 1).We then set two of the three expressions equal to each other (ie, drain none = test, test = drain all), which results in an equation and two variables: C and e. This is referred to as a two-way sensitivity or threshold analysis. It shows us how sensitive the choice of strategy is to simultaneous variations in the two variables (C and e).
Results The result of the decision analysis is presented in Table 3. A drain all policy has the highest expected utility (0.96321) using the baseline variables in Table 1 and is thus the preferred approach. One-way sensitivity analysis (Fig 2) indicates that a drain all policy is appropriate if the incidence of gastric outlet obstruction is greater than lo%, keeping all other variables constant. Figure 3 graphs a one-way sensitivity analysis, illustrating that as long as the effectiveness of the drainage procedure exceeds 40%, keeping all other variables constant, a drain all policy is once again favored. Figure 4 plots the test threshold and test-drain all threshold allowing the incidence of gastric outlet obstruction and the effectiveness of the drainage procedure to vary from 0% to 100%. Clearly, for clinicians facing an incidence of gastric outlet obstruction greater than lo%, a drain all policy is appropriate if the drainage procedure works 95% of the time. A test would be useful if the drainage procedure were only 20% to 60% effective in averting gastric outlet obstruction in the setting where its anticipated incidence was between 10% and 20%. Figure 5 plots the test-drain all sensitivity threshold across the entire range of specificities. To be helpful, the test would need to remain at least 80% sensitive even if it were 100%specific. If the sensitivity was 0.6, for example, a drain all policy would be appropriate through the entire
496
OLAK AND DETSKY ESOPHAGECTOMY WITH OR WITHOUT DRAINAGE?
Ann Thorac Surg 1992;53493-7
1
E T
H
R E
0.8
8
H
0 0.0 1 0
U 0.4
T I
ddn
0
0 0.4 C
T Y
R I I
I
I
1
I
I
I
8 0.2
1
0.2 0.9 0.4 0.1 0.8 0.7 0.8 0.0 RISK OF QASTRIC OUTLET OBSTRUCTION
0
DRAIN ALL
+DRAIN NONE
1
K
t
0
*TEST
0
0.4 0.8 EFFECTIVENESS
0.8
1
Fig 4 . Two-way sensitivity analysis varying the efectiveness of the drainage procedure against the incidence of gastric outlet obstruction
Fig 2 . One-way sensitivity analysis of the incidence of gastric outlet obstruction versus its expected utility for each treatment strategy.
(GOC).
range of possible specificities. To date, no test has been assessed with regard to its ability to predict risk of postoperative gastric outlet obstruction. The wide range of clinical management of this problem and the absence of reasonable criteria for adopting a policy on gastric drainage suggest that the likelihood of such a test being developed is small.
Comment This decision analysis represents an attempt to assess the relative merits of draining the intrathoracic stomach in patients undergoing esophagectomy. Although a randomized clinical trial [2] has addressed this question in the past, the number of patients randomized was too small for any definitive conclusion to be made. Despite its
1 ,
0.2
methodological shortcomings, the trial [2] did advocate a drain all policy. A multicentered randomized clinical trial approach would be necessary to answer this question conclusively, however, because no one center could be expected to accumulate enough experience within a reasonable time period. Another approach to the question involves decision analysis. The analysis presented in this article demonstrates that with an incidence of gastric outlet obstruction of 14% and a drainage procedure that is 95% effective, a drain all policy is the most appropriate. The two-way sensitivity analysis presented in Figure 4 demonstrates that a drain all policy should apply in most clinical settings. Allowing the incidence of gastric outlet obstruction and the effectiveness of the drainage proce-
1
T H
R
1
0.8
0.9-
T Y
0.0
!
I
I
0
0.2
I
I
I
I
0.4 0.8 0.8 EFFECTIVENESS OF DRAINAGE
-DRAINALL
+TEST
*DRAINNONE
1
I
Fig 3 . One-way sensitivity analysis of the effctiveness of drainage versus its expected utility for each treatment strategy.
0
0.2
0.4 0.8 SPECIFICITY
0.8
Fig 5. Two-way sensitivity analysis of the specificity versus sensitivity for the test. The line represents the test-drain all threshold.
1
OLAK AND DETSKY ESOPHAGECTOMY WITH OR WITHOUT DRAINAGE?
Ann Thorac Surg 1992:53:49>7
dure to vary between 0% and 100% permits clinicians to decide which approach fits their situation. A test would be considered appropriate for clinical situations where both the incidence of gastric outlet obstruction is less tha10% and the effectiveness of the drainage procedure ranges between 20% and 60%. This is represented by the area entitled test in Figure 4. It can be seen from this figure that the area entitled drain all encompasses the majority of clinical scenarios likely to be encountered in clinical practice. The two-way sensitivity analysis presented in Figure 5 demonstrates that given an incidence of obstruction of 14%, an effectiveness of drainage of 95%, and a drainageassociated complication rate of lo%, if a test were to be developed with the aim of placing patients into high- and low-risk groups it would require a sensitivity of at least 80% even if it was 100% specific. It is unlikely that the test could be this accurate. Although a test seems rational, the analysis presented here indicates that it would be very difficult to devise a test that would be useful for patient stratification. If the probabilities used in this analysis reflect those at your institution, then consideration should be given to performing a drainage procedure in all patients undergoing esophagogastrectomy. We thank Drs Martin F. McKneally and Andrew S. Wechsler for reviewing the manuscript.
References 1. Dragstedt LR, Camp EH. Follow-up of gastric vagotomy alone in the treatment of peptic ulcer. Gastroenterology 1948;11:460-5. 2. Cheung HC, Siu KF, Wong J. Is pyloroplasty necessary in esophageal replacement by stomach? A prospective, randomized controlled trial. Surgery 1987;102:19-24. 3. Akiyama H,Tsurumaru M, Kawamura T, et al. Principles of surgical treatment for carcinoma of the esophagus. Ann Surg 1981;194438-46. 4. Belsey RHR. Palliative management of esophageal carcinoma. Am J Surg 1980;139:789-94. 5. Ong GB,Kwong KH. The Lewis-Tanner operation for cancer of the oesophagus. J R Coll Surg Edinb 1969;143-19. 6. Ellis FH Jr. Cancer of the esophagus and cardia-role of surgery in palliation. Postgrad Med 1984;75:139-48. 7. Hopkins RA, Alexander JC, Postlethwait RW. Stapled esophagogastric anastomosis. Am J Surg 1984;147283-7.
497
8. McKeown KC. Carcinoma of the oesophagus. In: Taylor S,
ed. Recent advances in surgery, vol8. Edinburgh: Churchill Livingstone, 1973:133-65. 9. McKeown KC. Trends in oesophageal resection for carcinoma. With special reference to total oesophagectomy. Ann R Coll Surg Engl 1972;51:21%39. 10. Hinder RA. The effect of posture on the emptying of the intrathoracic vagotomized stomach. Br J Surg 1976;63:5814. 11. Huang GI, Wu YK. Operative technique for carcinoma of the esophagus. In: Juang GJ, Wu YK, ed. Carcinoma of the esophagus and gastric cardia. Berlin: Springer-Verlag, 1984. 31M. 12. Ujiki GT, Pearl GJ, Poticha S, Sisson GA Sr, Shields TW. Mortality and morbidity of gastric ‘pull-up’ for replacement of the pharyngoesophagus. Arch Surg 1987;12644-7. 13. Torrance GW, Feeny D. Utility and quality adjusted lifeyears. Int J Techno1 Assess Health Care 1989;5:559-75. 14. Paulker SG, Kassler JP. The threshold approach to clinical decision making. N Engl J Med 1980;302:1109-17.
Appendix 1. Test for Risk of Postoperative Gastric Outlet Obstruction” History, physical examination Peptic ulcer disease Suspected Proven (roentgenogram, endoscopy) Symptoms of early satiety or bloating Coexisting disease Diabetes Connective tissue disease UGI series: demonstration of Duodenal bulb deformity Duodenal ulcer Pyloric channel narrowing Gastric dilatation Retained food Radionudide study: delayed gastric emptying Endoscopy: demonstration of Duodenal scamng/fibrosis Peptic ulcer Tight pylorus Gastritislduodenitis Intraoperative: demonstration of Duodenal scarring Pyloric hypertrophy
+1 +2 +1 +1 +I +2 +2 +2 +1 +1 +2 +2 +2 +2
+1 +2
+2
Scoring of the test: any +2 or combination of two +I’s is considered a positive test a
