Cost Effectiveness of Chest Computed Tomography After Lung Cancer Resection: A Decision Analysis Model

Michael S. Kent, MD, Peter Korn, MD, Jeffrey L. Port, MD, Paul C. Lee, MD, Nasser K. Altorki, MD, and Robert J. Korst, MD Department of Cardiothoracic Surgery, Division of Thoracic Surgery, Weill Medical College of Cornell University, New York, New York

Background. Postoperative surveillance with chest computed tomography (CT) is often performed in patients who have undergone resection of non-small cell lung cancer (NSCLC), despite lack of supporting data. This study involves the creation of a decision analysis model to predict the cost effectiveness of postoperative surveillance CT. Methods. A decision analysis model was created in which a hypothetical cohort of patients underwent annual chest CT after resection of a stage IA NSCLC. The incidence of second primary lung cancer (SPLC), sensitivity and specificity of CT, as well as survival after resection of initial primary and SPLC were derived from published literature. The cost of CT and other procedures prompted by a positive finding on CT was calculated from Medicare reimbursement schedules. Cost effectiveness was defined as a cost of less than $60,000 per

quality-adjusted life-year gained in the cohort under surveillance compared with controls under no surveillance. Results. In the initial (base case) analysis, the cost of surveillance CT was $47,676 per quality-adjusted lifeyear gained, implying cost effectiveness. However, factors that rendered surveillance CT cost ineffective were (1) age at entry into the surveillance program greater than 65 years, (2) cost of CT greater than $700, (3) incidence of SPLC of less than 1.6% per patient per year of follow-up, and (4) a false positive rate of surveillance CT greater than 14%. Conclusions. Surveillance with postoperative CT may be a cost-effective intervention to detect SPLC in selected patients with previously resected stage IA NSCLC. (Ann Thorac Surg 2005;80:1215–23) © 2005 by The Society of Thoracic Surgeons

T

tive, the potential benefit of such a surveillance program can only be expressed within the context of its associated costs. These costs include not only the expense of the CT scans themselves, but also the cost of false positive scans and the potential harm to patients from invasive and unnecessary diagnostic procedures. The ideal method to determine the cost effectiveness of a postoperative surveillance program would be by a prospective, randomized study in which patients are followed up longitudinally, and the costs associated with surveillance and treatment are carefully recorded. Such a trial, however, is unlikely to be completed in the near future. As an example, in the National Lung Screening Trial, designed to determine the role of CT scans to screen smokers for primary lung cancer, more than 50,000 patients are necessary for the study to have adequate power, the results are not expected until the next decade, and the trial is being performed at considerable cost [8]. In the absence of compelling data from a randomized trial, decision analysis may serve as a useful tool to guide clinical practice regarding a large group of patients. Similar models have been used to investigate cost effectiveness of screening for Barrett’s esophagus [9], repair of abdominal aortic aneurysms [10], and a multitude of other clinical decisions. Decision analysis has also been

he risk of developing second primary lung cancer (SPLC) after complete resection of an initial nonsmall cell lung cancer (NSCLC) is 2% per patient, per year of follow-up, accumulating over time [1]. Despite this risk, the standard of care regarding follow-up for these patients is the subject of considerable debate, stemming from the lack of prospective studies to address this issue. Published reports on the subject represent single-institution, retrospective series with limited follow-up [2– 4]. As a consequence, most national organizations, including the American Society of Clinical Oncology [5], the American College of Chest Physicians [6], and the National Comprehensive Cancer Network [7], do not endorse routine chest computed tomography (CT) for the surveillance of lung cancer patients who have undergone resection. Despite these guidelines, many practicing physicians follow up patients with completely resected NSCLC using surveillance chest CT. From the societal perspecAccepted for publication April 1, 2005. Presented at the Forty-first Annual Meeting of The Society of Thoracic Surgeons, Tampa, FL, Jan 24 –26, 2005. Address reprint requests to Dr Korst, Department of Cardiothoracic Surgery, Suite M404, Weill Medical College of Cornell University, 525 East 68th St, New York, NY 10021; e-mail: [email protected].

© 2005 by The Society of Thoracic Surgeons Published by Elsevier Inc

0003-4975/05/$30.00 doi:10.1016/j.athoracsur.2005.04.006

GENERAL THORACIC

Cost Effectiveness of Chest Computed Tomography After Lung Cancer Resection: A Decision Analysis Model

GENERAL THORACIC

1216

KENT ET AL COST EFFECTIVENESS OF SURVEILLANCE CT

Ann Thorac Surg 2005;80:1215–23

utilized by government agencies such as the Health Care Financing Administration and the Food and Drug Administration to inform public policies [11]. In this study, we used a decision analysis model to predict the cost effectiveness of surveillance CT.

Material and Methods Overview of the Decision Analysis Model We evaluated the cost effectiveness of a surveillance program in which a hypothetical cohort of patients who had undergone resection of a stage IA NSCLC underwent annual CT of the chest. The survival of this cohort of patients was compared with a similar group of patients who did not undergo annual surveillance with CT (control group). The decision tree was based on a Markov model [12], in which patients transition from one health state to another on an annual basis. The probability of transitioning from one health state to another was derived from the available literature, as described below. The model was created using 1998 DATA 3.5.4 (Treeage Software, Williamstown, Massachusetts). One of the authors (P.K.), with extensive experience in designing decision analysis models [10, 13–16], was responsible for entering data into the software program according to the specifications of the thoracic surgeons (M.S.K., J.L.P., P.C.L., N.K.A., and R.J.K.). Four variables that would affect the cost effectiveness of surveillance CT were considered. These included the age of the patient at the initiation of surveillance, the false positive rate of surveillance CT for detecting SPLC, the cost of surveillance CT, and the annual risk of developing SPLC. The likelihood that patients who develop SPLC would be candidates for curative resection, as well as the stage distribution of these patients and their anticipated survival after treatment were all obtained from the existing literature (as described in the subsequent paragraphs) and inserted into the model. Figure 1 demonstrates the basic model design with the various health states that patients move between. Effectiveness of the surveillance program was measured by the quality-adjusted life-years (QALYs) gained in the surveillance group compared with the control group. In the Markov model, time spent in a specific state of health is adjusted by the utility of the health state, a variable between 0 and 1. As an example, perfect health was assigned a utility of 1, and death 0. Other health states such as alive with disease were assigned intermediate values, as derived from the literature [17]. Procedures that negatively affected a patient’s quality of life, such as recovery from surgery, treatment with chemotherapy, or invasive diagnostic procedures, were also assigned a disutility [17]. A QALY consequently represents the overall time spent in a health state multiplied by the utility assigned to that state. In the Markov model, all patients are followed up until death, and the total number of QALYs accrued by patients in each group is then calculated. Costs associated with surveillance, such as the cost of

Fig 1. Flow diagram representing the basic design of the Markov model for computed tomography (CT) surveillance. Patients in the surveillance program transition annually between health states shown in the shaded region. Numbers in parentheses represent the probability of each designated outcome. 1DOC ⫽ dead from causes other than lung cancer. 2NED ⫽ disease-free from lung cancer; these patients continue in surveillance. 3CT-detected SPLC ⫽ second primary lung cancer detected using surveillance CT. 4Interval SPLC ⫽ second primary lung cancer presenting independent of surveillance CT. 5DOD ⫽ dead of lung cancer. 6SCLC ⫽ small-cell lung cancer.

CT scans, surgery, and diagnostic procedures were derived from Medicare fee schedules for year 2004 in the State of New York [18]. Cost effectiveness was defined as a cost of less than $60,000 per QALY gained in the cohort under surveillance compared with controls under no surveillance [10, 13–16]. Within the model, the value of each variable (eg, the annual incidence of SPLC, or the probability of a false positive CT) could be altered, and its effect on the cost effectiveness of surveillance CT determined. In the base case analysis, the value of each variable was set at the level thought to be the most realistic as determined from the literature. Table 1 displays the values of these four variables used in the base case analysis. Additionally, in one-way sensitivity analysis, the value of a single variable was varied to determine its impact on cost effectiveness.

Control Cohort The survival of patients who have undergone resection of stage IA NSCLC was derived from the literature [19]. Those who survive beyond 5 years faced an age-based annual mortality, derived from US census data, to which an excess mortality of 1% was added to reflect increased cardiovascular risk from smoking [20]. The probability of developing SPLC in these patients was 2% per year, cumulative over time [1]. Patients who developed SPLC were considered to fall into one of four groups: (1) localized, resectable disease, (2) unresectable disease owing to advanced stage, (3) unresectable owing to insufficient pulmonary reserve, or (4) unresectable because of other reasons (eg, patient refusal, small-cell histology). The distribution of patients

KENT ET AL COST EFFECTIVENESS OF SURVEILLANCE CT

Table 1. Values for Variables Utilized in the Decision Analysis Model Base Case Value [Reference]

Variable Age at entrya Cost of surveillance CTa Specificity of surveillance CTa Annual incidence of SPLCa Lead time bias Five-year survival figures Initial stage IA nonsmall-cell lung cancer (resection) SPLC, stage IA (resection) SPLC, stage IB or higher (resection) Stage IA, definitive radiation Sensitivity of surveillance CT

60 $454 [[18] 90.5% [30] 2% [1] 1 year [27, 28] 67% [19] 40% [21] 18% [21] 16% [22] 85% [23–26]

a

Values for these parameters were varied in the one-way sensitivity analysis. CT ⫽ computed tomography;

SPLC ⫽ second primary lung cancer.

among these groups was derived from a summation of the literature on multiple primary lung cancer, in which no specific surveillance protocol was described (Tables 2 and 3). Survival after resection for SPLC was derived from the largest series on the subject [21]. In that report of 127 patients, the operative mortality was 5%, the 5-year survival after resection of stage IA disease was 40%, and survival after resection of stage IB disease or higher was 18%. For the purposes of the present analysis, patients with early-stage disease who were unresectable owing to pulmonary insufficiency were considered to undergo definitive radiotherapy, with a 5-year survival of 16% [22]. Patients with advanced disease were assigned a median survival of 12 months with no survivors at 5 years.

Table 2. Historical Resectability Rates for Second Primary Lung Cancer (SPLC) Author [Reference]

Year

Total SPLC

SPLC Resected

Razzuk [44] Smith [46] van Bodegom [43] Deschamps [41] Rosengart [42] Fleischer [47] Saito [38] Verhagen [39] Ribet [40] Antakli [45] Van Meerbeeck [48] Lamont [2] Rice [1] Total

1974 1976 1989 1990 1991 1991 1994 1994 1995 1995 1996 2002 2003

29 45 89 73 78 19 13 40 51 34 23 19 49 562

12 11 45 44 57 9 6 33 17 21 12 14 31 312 (55%)

1217

Surveillance Cohort Patients in the surveillance cohort underwent annual CT of the chest. These patients faced the same risk of death from their primary lung cancer and development of SPLC as those in the control cohort. Since the clinical characteristics as well as stage distribution of the patients in a surveillance CT cohort who develop SPLC is virtually unknown, two assumptions were made. First, the proportion of patients considered unresectable due to pulmonary insufficiency or other reasons was assumed to be equal to that in the control group. Second, for those patients with resectable disease, the proportion of stage IA versus higher stage cancers was extrapolated from prospective screening trials of smokers for primary lung cancer, in which the stage distribution for cancers detected on incidence scans (as opposed to prevalence scans) was provided (Tables 4 and 5) [23–26]. In addition to SPLC detected using surveillance CT, patients in the surveillance cohort could also present with a symptomatic SPLC in between annual CT scans. The proportion of patients developing interval SPLC was derived from the sensitivity of CT as published in primary screening trials [23–26]. These patients were assumed to have advanced, unresectable disease and were treated with palliative therapy. It was also assumed in the model that surveillance did not impact on the outcome (survival) of patients who developed a clear recurrence of their initial lung cancer. Those who recurred, regardless of whether recurrence was detected by surveillance CT or the development of symptoms, were considered to have a median survival of 1 year with a decline in their quality of life associated with recurrent disease. Interpretation of outcomes after detection of cancers within a surveillance protocol is confounded by the biases of overdiagnosis and lead-time. Overdiagnosis is possible when a screen detects an indolent tumor that may never cause symptomatic disease within the lifetime of the patient. Overdiagnosis was not incorporated into the model, because SPLC detected on surveillance CT is, by definition, progressive cancer that was not present on earlier scans. Lead-time bias occurs when diagnosis occurs earlier in the natural history of the disease in the screened group, although the overall survival is equivaTable 3. Stage Distribution of Patients With Resected Second Primary Lung Cancer (SPLC)

Author [Reference]

Year

SPLC Resected

Deschamps [41] Ribet [40] Lamont [2] van Rens [21] Rice [1] Battafarano [49] Total

1990 1995 2002 2002 2003 2004

44 17 14 127 31 69 302

Stage IA SPLC Resected 29 11 13 50 19 34 156 (52%)

GENERAL THORACIC

Ann Thorac Surg 2005;80:1215–23

GENERAL THORACIC

1218

KENT ET AL COST EFFECTIVENESS OF SURVEILLANCE CT

Ann Thorac Surg 2005;80:1215–23

Table 4. Stage Distribution of Lung Cancer Detected Using Repeat (Incidence) Screening Computed Tomography Study [Reference]

Total Cancers

Stage IA NSCLC

Stage IB– IIIA NSCLC

Stage IIIB– IV NSCLC

SCLC

Unresected NSCLC

ELCAPa [23] Mayo [26] ALCAb [25] Sone [24] Total

7 11 22 40 80

5 5 18 32 60

1 5 2 3 11

0 0 2 2 4

1 1 0 3 5

2 NS 5 NS —

a

Early Lung Cancer Action Project.

NS ⫽ not stated;

b

Anti-lung Cancer Association Project.

NSCLC ⫽ nonsmall-cell lung cancer;

SCLC ⫽ small-cell lung cancer.

lent in both groups. A lead-time bias of 1-year was incorporated in the model, based on published lead-time biases from randomized trials of screening using chest radiographs [27] and estimated tumor doubling times [28].

Cost Analysis The costs of surgery and noninvasive and invasive diagnostic procedures were obtained from Medicare fee schedules for the State of New York for the year 2004 (Table 6) [18]. These costs included the associated surgeon, anesthesia, pathology and facility fees, and represent Medicare reimbursement rates. While this does not represent the actual, realized cost of these procedures that patients and the health care facility incur for these workups, these figures can be viewed as how much these patients’ care costs the Medicare system, and are useful for comparative purposes. Patients who were treated with palliative chemotherapy for advanced disease were assigned an overall cost of care of $55,185 (in 2004 dollars), based upon the economic analysis of a recent, randomized clinical trial for advanced-stage lung cancer [29]. This trial was chosen because the costs of nonprotocol care, including radiotherapy and supportive care

utilized until the patient’s death, were reported. Both costs and utilities were discounted 3% annually. A critical issue is a determination of the excess cost associated with CT surveillance. This cost is in large measure attributable to the expense of unnecessary diagnostic tests driven by false positive CT scans. Data for this calculation (Table 6) was taken from a retrospective review of postoperative CT scans performed at our institution [30]. In that study, 16 of 168 surveillance CT scans performed was falsely positive (9.5%), which led to a number of additional diagnostic tests, including surgery, for a total cost of $78,427. Thus, the cost associated with a single false positive scan was calculated to be $4,901. It was also assumed in the model that the cost of working up a patient with advanced cancer is identical to that of a patient with an early-stage cancer.

Results The 5-year survival among the surveillance patients in whom an SPLC developed was calculated by the model Table 6. Cost of Diagnostic Testing/Procedures/Treatmentsa Procedure

Table 5. Distribution of Patients in Surveillance and Control Cohorts in the Base Case Analysisa

Resected stage IA Resected stage IB–IIIA Not resected, advanced Not resected, pulmonary insufficiency Not resected, small-cell lung cancer or other

Control

Surveillance

29% 26% 22% 18% 5%

61% 11% 5% 18% 5%

a The method for determining the distribution of patients in the surveillance cohort was generated in the following manner. The percentage of those unresectable because of pulmonary insufficiency or other reasons was assumed to be the same as for the control cohort [31]. The percentage of patients who presented with unresectable disease owing to advanced stage in prospective studies of primary screening for lung cancer was 4 of 80, or 5% (Table 3). Therefore, the total percentage of patients in the surveillance cohort resectable for cure was calculated to be 72%. In the studies of primary screening, 60 of 71 potentially resectable patients (85%) were stage IA, and 15% were more advanced (Table 3). Consequently, 61% (0.85 ⫻ 0.71) of the surveillance cohort would be predicted to undergo resection for stage IA disease, and 11% (0.15 ⫻ 0.71) would be predicted to undergo resection for stage IB or higher disease.

Chest computed tomography scan Bone scan Brain magnetic resonance imaging Position emission tomography Thoracentesis Transthoracic fine-needle aspiration Fiberoptic bronchoscopy with biopsy Cervical mediastinoscopy Thoracoscopy and biopsy Thoracoscopy and wedge resection Thoracotomy and wedge resection Thoracotomy and segmentectomy Thoracotomy and lobectomy Thoracotomy and completion pneumonectomy Definitive radiotherapy Treatment unresectable stage III and IV nonsmall-cell lung cancer a

See reference 30 for detailed cost breakdown.

Cost (Nearest Dollar) 454 282 830 1761 159 1695 1479 2259 29,598 30,238 32,015 31,968 32,106 32,340 10,348 55,185

KENT ET AL COST EFFECTIVENESS OF SURVEILLANCE CT

1219

Fig 2. Sensitivity analysis of four variables affecting the cost effectiveness of surveillance computed tomography (CT). In each graph, the solid line represents the base case analysis, and the dashed line represents the cost effectiveness threshold of $60,000/quality-adjusted life-year (QALY). (A) Patient age upon entry into surveillance. (B) The cost of a surveillance CT scan. (C) The probability of a false-positive CT scan. (D) The annual incidence of small-cell lung cancer (SPLC).

to be 29%, compared with 19% in the control cohort. In the base case analysis, surveillance increased overall survival by 0.16 QALY at an incremental cost of $7,716. Thus, annual CT scans were determined to be a costeffective intervention at an overall cost of $47,676 per QALY gained.

Age at Entry Into the Surveillance Program The age at which patients were entered into surveillance was a significant determinant of cost effectiveness. In the base case analysis, surveillance CT was cost effective at an entry age of nearly 65 (Figure 2A) at $61,775 per QALY (at age 65). However, surveillance of an older cohort of patients was not a cost-effective intervention. For patients aged 70, the cost of surveillance was $84,781 per QALY.

Cost of Surveillance CT Scans The base case estimate for the cost of a full-dose, chest CT scan was $454. In the sensitivity analysis, the use of surveillance CT for detecting SPLC was cost effective over a wide range of estimated costs (Figure 2B). However, the model determined that surveillance CT became cost ineffective when the cost of CT scans reached $700 or greater.

Frequency of False Positive Surveillance CT Scans (Specificity) As expected, the frequency of false positive CT scans significantly impacted on the cost effectiveness of surveillance CT (Figure 2C). That is because of the expense and morbidity of additional and unnecessary procedures prompted by a suspicious surveillance CT scan. These procedures not only include noninvasive (although expensive) studies such as positron emission tomography (PET), but also invasive procedures such as thoracoscopy

and even thoracotomy performed for diagnostic purposes (Table 6) [30]. In the base case analysis, it was estimated that 9.5% of surveillance CT scans would represent false positive studies. This figure was based on a retrospective review of the use of surveillance CT conducted at our institution [30]. However, an increase in this estimate of false positivity (to 14%) rendered surveillance CT cost ineffective.

Incidence of Second Primary Lung Cancer The most influential determinant of cost effectiveness was the annual risk of developing SPLC (Figure 2D). Earlier studies had estimated this risk to be between 1% and 4% per patient-year [31]. A recent prospective study has more precisely defined this risk to be 2% per patientyear of follow-up [1]. Further, in this latter study, the smoking status of the patient at the time of initial surgery significantly impacted this risk. For example, SPLC did not develop in patients who had never smoked. Current smokers faced a risk of 2.7% per year compared with a risk of 1.8% per year for former smokers. In the present study, surveillance CT was cost effective if the annual incidence of SPLC was greater than 1.6%. The cost effectiveness of surveillance CT with an annual incidence of 1.8% was $53,473 per QALY.

Comment In the United States, it has been estimated that approximately 35,000 patients undergo surgery for lung cancer each year [32]. Debate regarding the benefit of surveillance of these patients with CT scans stems from the lack of prospective data, the potential harm associated with

GENERAL THORACIC

Ann Thorac Surg 2005;80:1215–23

GENERAL THORACIC

1220

KENT ET AL COST EFFECTIVENESS OF SURVEILLANCE CT

false positive scans, and pessimism regarding the overall prognosis of patients with metachronous lung cancer. From a strictly clinical standpoint, the rationale for offering patients surveillance CT is twofold. First, patients who undergo resection face a high rate of disease recurrence. Surveillance CT may detect recurrence at an asymptomatic, and perhaps more treatable, stage. However, it seems unlikely that surveillance will detect recurrence for which potentially curative treatments, such as re-resection, are possible. In the M.D. Anderson experience, re-resection was offered to only 3% of patients undergoing strict follow-up [33]. A second, and more important justification for performing surveillance CT is the recognition that patients who survive surgical resection of NSCLC face a significant risk of developing another primary lung cancer. In a prospective trial of chemoprevention of second lung cancers, this risk was shown to be 2% per patient-year of follow-up [1]. Furthermore, the risk of developing SPLC accumulates over the lifetime of the patient. Eight years after resection, the risk of developing a SPLC was nearly 25% [1]. To place this figure into perspective, the estimated risk of developing primary lung cancer in current heavy smokers, the population for which screening with chest CT scans has been extensively studied [23–26] is only 0.5% per patientyear [17]. In addition to the clinical rationale, however, surveys of thoracic surgeons have demonstrated that the motivation to provide postoperative follow-up is less to offer patients curative interventions for disease recurrence than it is to “please patients, avoid malpractice suits and improve patient quality of life” [34]. Despite this pessimism, surveillance may realize a significant survival benefit if SPLC can be detected at an early, and perhaps more curable stage.

Assumptions of the Decision Analysis Model Conclusions drawn from a decision analysis model are only relevant to clinical practice if the many assumptions used to create the model are specifically stated, and agreed upon. In the present study, given the lack of prospective data concerning the use of postoperative surveillance CT for detecting SPLC, the following assumptions were necessary. First, a critical assumption of the model was that surveillance CT would not impact on the outcome of patients who developed a recurrence of their initial lung cancer. This was assumed because no clear evidence exists to support that early detection of recurrences affects long-term survival of patients with NSCLC. Retrospective studies on postoperative surveillance have failed to demonstrate a significant difference in survival among those patients who were in a “strict” or “symptom driven” surveillance program [4] or between those whose recurrences were asymptomatic or symptomatic [33]. Furthermore, it is unlikely that a rigorous surveillance program would in fact detect recurrences at an asymptomatic and localized stage. In the prospective study by Westeel and coworkers [35], a scheduled chest CT scan diagnosed only 10 recurrences among 136 patients. In-

Ann Thorac Surg 2005;80:1215–23

stead, the majority of patients presented with a symptomatic, distant recurrence independent of the surveillance protocol. Additional rationale for this assumption was to bias the model against a benefit of surveillance CT, making a conclusion that surveillance CT is cost effective as conservative as possible. Second, it was assumed that data from prospective trials of primary screening for lung cancer could be extrapolated to a model of secondary surveillance. From these studies, the sensitivity of CT for detecting SPLC and the stage distribution of patients in a hypothetical cohort under surveillance was generated [23–26]. Given the absence of prospective data regarding the use of surveillance CT for the detection of SPLC, this assumption appears reasonable. In addition, only data from primary screening studies that reported on patients who had undergone incidence scans (eg, those who had a normal baseline scan and were followed up by serial scans) was utilized. This population seemed most comparable to patients in the present hypothetical cohort who had been resected, were free of disease initially postoperatively, and were followed up with annual CT scans. Third, it was assumed that there would be 100% compliance with the surveillance program. In actual practice, compliance with a surveillance program would most likely be less than 100%. However, compliance with an intensive surveillance program that included annual CT as well as flexible bronchoscopy was reported to be 90% in a recently reported prospective study that spanned 10 years, implying that compliance may indeed be relatively high [35]. Fourth, it was assumed that the use of surveillance CT would not lower the proportion of patients who had insufficient pulmonary reserve to tolerate resection. Although it is conceivable that surveillance CT could detect tumors at an earlier stage for which patients with poor pulmonary reserve may tolerate a more limited resection, this was not incorporated into the model as no data are available to quantify this effect. In addition to the above assumptions, limitations of the present study include that the decision analysis model does not consider the psychological impact of surveillance on patients. Indeterminate findings on surveillance scans may lead to anxiety and a decline in quality of life that was not modeled. Conversely, close surveillance may lead to an earlier detection of health issues unrelated to lung cancer and a greater sense of well-being in a subset of patients.

Cost Effectiveness of Surveillance CT It must be emphasized that the cost-effectiveness threshold of $60,000 per QALY is an arbitrary figure. Clearly, our society supports the use of many health care interventions with cost-effectiveness ratios of greater than $60,000 per QALY. As an example, lung volume reduction surgery has been shown to reduce mortality in a subset of patients with severe emphysema, and the cost of surgery is now reimbursed by Medicare. However, the cost-effectiveness ratio of this procedure greatly exceeds

$60,000 per QALY [36]. We chose $60,000 per QALY for use in the present analysis because it lies close to the “median” value used in other publications [10, 13–16, 37]. Clearly, this figure is meant more as a guideline as opposed to a distinct cutoff value. In this study, using a cost-effectiveness threshold of $60,000 per QALY, a decision analysis model has suggested that annual surveillance CT is a cost-effective intervention after curative resection of stage IA NSCLC provided several criteria are met: (1) patients are enrolled into a surveillance program at an age of less than 65 years, (2) the cost of a surveillance CT scan is less than $700, (3) the rate of false positive CT scans is less than 14%, and (4) the annual risk of a second primary lung cancer is 1.6% per year or greater. Given these criteria, it appears that postoperative surveillance CT is not likely to be cost effective for every patient who has undergone resection of an early-stage lung cancer. As an example, the model suggests that surveillance CT is not cost effective for patients over the age of 65. That may not be due to increased costs, but perhaps because surveillance CT may only be beneficial if it is carried out over a lengthy period of time, requiring patients to live longer to reap the benefits of surveillance. Older patients have a shorter life expectancy and face a higher annual mortality from cardiovascular disease and other malignancies that would not be reduced by surveillance CT. As a consequence, these patients may not live long enough for surveillance CT to be cost effective. In addition, the model also suggests that surveillance CT is not cost effective if the rate of false positive scans is 14% or greater. Although our group has reported a false positive rate of 9.5% at our institution [30], this could be higher at centers with less experience in the follow-up care of lung cancer patients. Further, a distinction needs to be made between a “finding” on a scan (eg, a nodule), and a “positive” scan, about which the clinician is suspicious enough to perform additional testing. Clearly, not all findings will be suspicious. Finally, surveillance CT does not appear to be cost effective if the annual rate of developing SPLC is less than 1.6% per year. In the largest prospective report to quantify this risk [1], patients who had quit smoking at the time of their initial resection faced a risk of 1.8% per year compared with a risk of 2.7% per year for those who continued to smoke. As a result, surveillance CT for both current and former smokers appears to be cost effective. In conclusion, the decision to enroll patients in a surveillance CT program must be made on an individual basis. Older patients, or those with limited life expectancy from other diseases, may not benefit from a surveillance CT program. Conversely, younger patients, including former smokers and (especially) current smokers, may derive the most benefit from surveillance CT. Prospective data collection is necessary to confirm these findings, as well as to address additional variables including the use of adjuvant therapy and the utility of early detection of metastatic disease using surveillance CT.

KENT ET AL COST EFFECTIVENESS OF SURVEILLANCE CT

1221

References 1. Rice D, Kim HW, Sabichi A, et al. The risk of second primary tumors after resection of stage I nonsmall cell lung cancer. Ann Thorac Surg 2003;76:1001– 8. 2. Lamont JP, Kakuda JT, Smith D, Wagman LD, Grannis FW Jr. Systematic postoperative radiologic follow-up in patients with non-small cell lung cancer for detecting second primary lung cancer in stage IA. Arch Surg 2002;137:935–9. 3. Chiu CH, Chern MS, Wu MH, et al. Usefulness of low-dose spiral CT scan of the chest in regular follow-up of postoperative non-small cell lung cancer patients: preliminary report. J Thorac Cardiovasc Surg 2003;125:1300 –5. 4. Younes RN, Gross JL, Deheinzelin D. Follow-up in lung cancer: how often and for what purpose? Chest 1999;115: 1494 –9. 5. American Society of Clinical Oncology. Clinical practice guidelines for the treatment of unresectable non-small cell lung cancer: adopted on May 16, 1997 by the American Society of Clinical Oncology. J Clin Oncol 1997;15:2996 –3018. 6. Colice GL, Rubins J, Unger M, and the American College of Chest Physicians. Follow-up and surveillance of the lung cancer patient following curative-intent therapy. Chest 2003; 123(Suppl):272S– 83S. 7. National Comprehensive Cancer Network. Practice guidelines for non-small cell lung cancer. Rockledge, PA: National Comprehensive Cancer Network, 2000. 8. Gohagan J, Marcus P, Fagerstrom R, Pinsky P, Kramer B, Prorok P, for the Lung Screening Study Research Group. Baseline findings of a randomized feasibility trial of lung cancer screening with spiral CT scan versus chest radiograph: the lung screening study of the National Cancer Institute. Chest 2004;126:114 –21. 9. Inadomi JM, Sampliner R, Lagergren J, Lieberman D, Fendrick AM, Vakil N. Screening and surveillance for Barrett esophagus in high-risk groups: a cost-utility analysis. Ann Intern Med 2003;138:176 – 86. 10. Patel ST, Korn P, Haser PB, Bush HL Jr, Kent KC. The cost-effectiveness of repairing ruptured abdominal aortic aneurysms. J Vasc Surg 2000;32:247–57. 11. Russell LB, Gold MR, Siegel JE, Daniels N, Weinstein MC. The role of cost-effectiveness analysis in health and medicine: panel on cost-effectiveness in health and medicine. JAMA 1996;276:1172–7. 12. Birkmeyer JD, Liu JY. Decision analysis models: opening the black box. Surgery 2003;33:1– 4. 13. Patel S, Haser P, Korn P, Bush H, Deitch J, Kent C. Is carotid endarterectomy cost-effective in symptomatic patients with moderate (50% to 69%) stenosis? J Vasc Surg 1999;30:1024 – 33. 14. Kilaru S, Korn P, Kasirajan K, et al. Is carotid angioplasty and stenting more cost effective than carotid endarterectomy? J Vasc Surg 2003;37:331–9. 15. Lee TY, Korn P, Heller JA, et al. The cost-effectiveness of a “quick-screen” program for abdominal aortic aneurysms. Surgery 2002;132:399 – 407. 16. Korn P, Patel ST, Heller JA, et al. Why insurers should reimburse for compression stockings in patients with chronic venous stasis. J Vasc Surg 2002;35:950 –7. 17. Mahadevia PJ, Fleisher LA, Frick KD, Eng J, Goodman SN, Powe NR. Lung cancer screening with helical computed tomography in older adult smokers: a decision and costeffectiveness analysis. JAMA 2003;289:313–22. 18. 2004 Fee schedule. Empire Medicare Services New York Area 01. Available at: http://www.empiremedicare.com/ fees/ny/NY04PFS01.pdf. Accessed April 15, 2004. 19. Mountain CF. Revisions in the International System for Staging Lung Cancer. Chest 1997;111:1710 –7. 20. Cigarette-smoking related morbidity—United States. Morbidity and Mortality Weekly Report, September 5, 2003. Available at: http://www.cdc.gov/mmwr/PDF/wk/ mm5235.pdf. Accessed April 15, 2004.

GENERAL THORACIC

Ann Thorac Surg 2005;80:1215–23

GENERAL THORACIC

1222

KENT ET AL COST EFFECTIVENESS OF SURVEILLANCE CT

21. van Rens MT, Zanen P, de la Riviere AB, Elbers HR, van Swieten HA, van den Bosch JM. Survival after resection of metachronous non-small cell lung cancer in 127 patients. Ann Thorac Surg 2001;71:309 –13. 22. Cheung PC, Mackillop WJ, Dixon P, Brundage MD, Youssef YM, Zhou S. Involved-field radiotherapy alone for earlystage non-small-cell lung cancer. Int J Radiat Oncol Biol Phys 2000;48:703–10. 23. Henschke CI, Naidich DP, Yankelevitz DF, et al. Early lung cancer action project: initial findings on repeat screening. Cancer 2001;92:153–9. 24. Sone S, Li F, Gang ZG, et al. Results of three-year mass screening programme for lung cancer using mobile lowdose spiral computed tomography scanner. Br J Cancer 2001;84:25–32. 25. Sobue T, Moriyama N, Kaneko M, et al. Screening for lung cancer with low-dose helical computed tomography: antilung cancer association project. J Clin Oncol 2002;20:911–20. 26. Swensen SJ, Jett JR, Hartman TE, et al. Lung cancer screening with CT: Mayo Clinic experience. Radiology 2003;226: 756 – 61. 27. Walter SD, Kubik A, Parkin DM, Reissigova J, Adamec M, Khlat M. The natural history of lung cancer estimated from the results of a randomized trial of screening. Cancer Causes Conrol 1992;3:115–23. 28. Henschke CI, McCauley DI, Yankelevitz DF, et al. Early lung cancer action project: overall design and findings from baseline surveillance. Lancet 1999;354:99 –105. 29. Ramsey SD, Moinpour CM, Lovato LC, et al. Economic analysis of vinorelbine plus cisplatin versus paclitaxel plus carboplatin for advanced non-small cell lung cancer. J Natl Cancer Inst 2002;94:291–7. 30. Korst RJ, Gold HT, Kent MS, Port JL, Lee PC, Altorki NK. Surveillance computed tomography following complete resection for non-small cell lung cancer: results and costs. J Thorac Cardiovasc Surg 2005;129:652– 60. 31. Johnson BE. Second lung cancers in patients after treatment for an initial lung cancer. J Natl Cancer Inst 1998;90:1335– 45. 32. Virgo KS, Johnson FE, Naunheim KS. Followup of patients with thoracic malignancies. Surg Oncol Clin North Am 1999;8:355– 69. 33. Walsh GL, O’Connor M, Willis KM, et al. Is follow-up of lung cancer patients following resection medically indicated and effective? Ann Thorac Surg 1995;60:1563–72. 34. Virgo KS, Naunheim KS, Coplin MA, Johnson FE. Lung cancer patient follow-up: motivation of thoracic surgeons. Chest 1998;114:1519 –34.

Ann Thorac Surg 2005;80:1215–23

35. Westeel V, Choma D, Clement F, et al. Relevance of an intensive postoperative follow-up after surgery for nonsmall cell lung cancer. Ann Thorac Surg 2000;70:1185–90. 36. Ramsey SD, Berry K, Etzioni R, Kaplan RM, Sullivan SD, Wood DE, for the National Emphysema Treatment Trial Research Group. Cost effectiveness of lung volume reduction surgery for patients with severe emphysema. N Engl J Med 2003;348:2092–102. 37. Eichler H-G, Kong SX, Gerth WC, Mavros P, Jonsson B. Use of cost-effectiveness analysis in health-care resource allocation decision-making: how are cost-effectiveness thresholds expected to emerge? Value Health 2004;7:518 –28. 38. Saito Y, Sato M, Sagawa M, et al. Multicentricity in resected occult bronchogenic squamous cell carcinoma. Ann Thorac Surg 1994;57:1200 –5. 39. Verhagen AF, Tavilla G, van de Wal HJ, Cox AL, Lacquet LK. Multiple primary lung cancers. Thorac Cardiovasc Surg 1994;42:40 – 4. 40. Ribet M, Dambron P. Multiple primary lung cancers. Eur J Cardiothorac Surg 1995;9:231– 6. 41. Deschamps C, Pairolero PC, Trastek VF, Payne WS. Multiple primary lung cancers. Results of surgical treatment. J Thorac Cardiovasc Surg 1990;99:769 –78. 42. Rosengart TK, Martini N, Ghosn P, Burt M. Multiple primary lung carcinomas: prognosis and treatment. Ann Thorac Surg 1991;52:773–9. 43. van Bodegom PC, Wagenaar SS, Corrin B, Baak JP, Berkel J, Vanderschueren RG. Second primary lung cancer: importance of long-term follow-up. Thorax 1989;44:788 –93. 44. Razzuk MA, Pockey M, Urschel HC Jr, Paulson DL. Dual primary bronchogenic carcinoma. Ann Thorac Surg 1974;17: 425–33. 45. Antakli T, Schaefer RF, Rutherford JE, Read RC. Second primary lung cancer. Ann Thorac Surg 1995;59:863–7. 46. Smith RA, Nigam BK, Thompson JM. Second primary lung cancer. Thorax 1976;31:507–16. 47. Fleisher AG, McElvaney G, Robinson CL. Multiple primary bronchogenic carcinomas: treatment and follow-up. Ann Thorac Surg 1991;51:48 –51. 48. Van Meerbeeck J, Weyler J, Thibaut A, et al. Second primary lung cancer in Flanders: frequency, clinical presentation, treatment and prognosis. Lung Cancer 1996;15:281–95. 49. Battafarano RJ, Force SD, Meyers BF, et al. Benefits of resection for metachronous lung cancer. J Thorac Cardiovasc Surg 2004;127:836 – 42.

DISCUSSION DR SANDRO MATTIOLI (Bologna, Italy): I think it’s a very important issue, how to follow up the patient who had surgery for lung cancer. You addressed in the study very early stage IA, which is correct, because the possibility of local recurrence is lower, the lowest maybe in all lung cancers. However, I see here the key point of the study—when the percent of false positive scans is below 14%. So the problem is, if I understood well, you said in the not yet published paper by Korst in the JTCVS, which is a specificity of 90%, you addressed, again, if I understood well, that you have a 20% rate of recurrence from literature studies. So the problem is, are you sure that with this kind of statistical method you weighted in the proper manner the cases of local recurrence, which, as you stated, are not helped by early detection? I think this is the problem, how to distinguish between recurrence of the tumor and a second primary even after 2 years from operation. Thank you. DR KENT: Thank you for your comments. I would like to address your points one by one. Number one, we did only look in this model at stage IA

patients, and the reason we did that was because we believed that these patients would be best served by a potential screening program. We did not look at patients who were staged IB or higher, although our feeling would be that, for example, patients with stage IIIA disease who have a very high rate of disease recurrence and metastatic disease would probably not benefit from a screening program, although we did not look at that. In terms of defining what the rate of a false positive scan would be, I want to take a minute to talk about that. In our study we showed that the false positive rate was 9.5%, and that number was a combination of both the inherent ability of a scan to detect a primary lung cancer and also what the surgeon did with that information. When we looked at our series of patients who underwent surveillance CT scan, about two thirds of those patients had an abnormal scan, either suspicious for recurrent disease or a new primary as read by the radiologist. When we said that our rate of false positive scans was 9.5%, what we meant was that 9.5% of those scans when looked at by the surgeon who initiated further workup was found to be benign

disease. So that was a figure that we found at our institution, although clearly that figure would be different based upon the practice philosophy of people following up these patients. The final issue that you brought up is a very important one, looking at the distinction between recurrence and a new primary. In the model, clearly we did not need to make that distinction. We assumed that patients who developed a recurrence of their disease would go on to die and that patients who developed a new primary would be treated in the manner that we discussed. But clearly, I think in some cases it is very important to make that distinction, although we have all been in scenarios where a patient undergoes resection of a stage IA tumor, a year or two later has what appears to be a new primary, and it can be a difficult distinction between a new primary and a local recurrence. DR LEWIS WETSTEIN (Freehold, NJ): Congratulations on a nice presentation. I find the results, however, somewhat disconcerting: the impression that it’s either all or none. We continuously reevaluate the way we follow patients, and over 25 years have developed an extremely productive protocol. Our protocol includes an initial baseline CT scan of the chest taken at 3 months after resection. We subsequently utilize chest roentgenograms and basic blood work quarterly for 2 years, semiannually for another 2 years, and then yearly. We only resort to repeating the CT scan if we notice either a change on the chest roentgenogram, some new symptomatology that we’re confused about, or the blood work has changed. After this protocol, I can’t remember a recurrence we’ve missed. Moreover, I’m confident you will agree that our protocol is even more cost effective. DR KENT: I think you have raised some very good points. The first point is where do we set that threshold for cost effectiveness and does it matter if a test is $65,000 per quality-adjusted life-year or $40,000. And there’s no question that at a very important level, that threshold that we set is an arbitrary one, although it is one which is used really in the majority of publications that are looking at cost effectiveness for other interventions. I think that the decision to offer a patient a surveillance program is really not one that is solely based on cost effectiveness, and that a decision to initiate a health care program is at the one level perhaps a political one and a societal one, and also I think between the individual patient and surgeon

KENT ET AL COST EFFECTIVENESS OF SURVEILLANCE CT

1223

or pulmonologist or oncologist following that patient. I think what’s important about these studies is that we do have a way to at least determine whether we’re in the ballpark of cost effectiveness, and also we can see what factors impact on cost effectiveness. So I think that we did show, for example, that the false positive rate is a significant determinant of cost effectiveness, and clearly if we find that we’re pursuing a lot of studies with invasive workups for benign disease, we’re probably not helping these patients, or at the very minimum, we’re adding cost and certainly adding potential morbidity. DR H. CHRISHANTHA FERNANDO (Boston, MA): Most local recurrences occur within the first 2 to 3 years, so perhaps you should be doing CT scans more frequently in the first 2 to 3 years. Why in the model did you choose annual CT scans rather than more frequent scans earlier on? DR KENT: I think that’s another good point. The purpose of the model was to look at developing or the detection of a second primary lung cancer, and we really were not able to model whether screening would impact on the survival of patients who developed a recurrence. For simplicity’s sake, we assumed that screening would not impact on patients who developed a recurrence of their primary lung cancer, and we did that for two reasons. Number one, there are not a lot of good data to show that screening would impact on the treatment of patients who recur from their primary lung cancer, and those studies that did show a benefit, that benefit may very well be due simply to lead-time bias. The second issue is that we really wanted to bias the model against screening, so that if we were to conclude that screening were cost effective, it would be a conservative estimate. I think if you were to assume that screening may impact on patients who develop a recurrence, then screening would be even more cost effective. DR DANIEL L. MILLER (Atlanta, GA): Just a real quick poll. How many of you in the audience are involved in the surveillance of your patients for lung cancer? [A show of hands.] How many of you get CT scans? [A show of hands.] About half. Actually, there’s discussion now within the American College of Surgeons Oncology Group about this very study of looking at CT surveillance and so forth. So you’ll be hearing about that.

GENERAL THORACIC

Ann Thorac Surg 2005;80:1215–23