The use of positron emission tomography (PET) has been increasing since the mid 1980s [1
]. PET is primarily used in oncology, cardiology, and neuropsychiatry. Oncologic uses of PET include the evaluation of lung, breast, brain, head and neck, and colorectal cancer; melanoma; and lymphoma (Hodgkin's and non-Hodgkin's) [2
]. Although a variety of radionuclides can be used to perform PET, FDG is the most commonly used radionuclide for applications in oncology. In cardiology, the use of PET is focused on the assessment of coronary artery disease to differentiate viable from nonviable ischemic myocardial tissue [6
]. Patients with neuropsychiatric disorders, such as epilepsy and dementia, are also examined using PET [6
]. PET is also potentially useful in evaluating the response of cancer patients to chemotherapy and radiation treatments [4
The Centers for Medicare & Medicaid Services have approved Medicare reimbursement for PET scans obtained for a variety of indications. These indications include the assessment of myocardial viability; presurgical evaluation of refractory seizures; diagnosis of solitary pulmonary nodules; and evaluation of patients with non–small cell lung cancer, colorectal cancer, esophageal cancer, breast cancer, lymphoma, and melanoma. The Centers for Medicare & Medicaid Services are reviewing Medicare reimbursement of PET for the diagnosis of Alzheimer's disease, dementia, soft-tissue sarcoma, and thyroid cancer [8
The accuracy of FDG PET in characterizing solitary pulmonary nodules [9
] and in staging non–small cell lung cancer is well documented [3
]. In addition, several groups have evaluated the cost effectiveness of PET in the management of lung cancer [2
]. Most of these studies used decision analysis models to evaluate the projected cost effectiveness of PET for pulmonary nodule diagnosis [18
] or staging in patients with non–small cell lung cancer [16
]. The models suggest that more accurate diagnosis or staging of non–small cell lung cancer will lead to more effective treatment and to improved patient outcomes. Prospective evidence to support this view is emerging [23
]. A recently concluded randomized controlled trial found that PET not only is more accurate than traditional methods for staging non–small cell lung cancer, but also results in improved patient outcomes in the clinical setting [24
]. The costs of PET for these models were estimated using Medicare reimbursement rates [21
], hospital charges [16
], insurance reimbursement rates [2
], or German national reimbursement rates [17
]. Additional cost-effectiveness studies using accurate estimates of costs, performed in conjunction with prospective trials, are needed to assess the true benefit and determine the appropriate role of PET in clinical applications.
Operating a PET center is expensive. In 1983, the annual operating expense of a PET and cyclotron center was $2.1 million (adjusted to 2000 dollars) [25
]. Scan volume is reported to be the most critical factor in determining scanner profitability [25
]. Most previously published cost estimates were based on hospital charges, Medicare reimbursement rates, or estimates found in the literature. One study modeled the cost of PET and projected patient volume on the basis of interviews with key informants such as PET center managers and nuclear medicine physicians, evidence from prior studies, and financial information from one PET center. In that study, the estimated annual operating cost of a PET and cyclotron center was between $0.9 and $2.6 million (in 2000 dollars) per year over a 7-year period [26
]. In 1998, ECRI (formerly the Emergency Care Research Institute) used previously published charge and reimbursement estimates to calculate that the cost of a PET scan was between $1901 and $2535 (in 2000 dollars) [19
]. A 1983 analysis of financial information from two PET centers over a 5-year period estimated that PET costs ranged from $1059 to $4787 (adjusted to 2000 dollars) per scan [25
We present a survey-based analysis that includes direct, indirect, and capital costs from eight PET centers participating in a clinical trial. This convenience sample is not necessarily representative of all PET centers; however, it allowed us to capitalize on the fact that the directors of the centers were willing to participate in research and to share potentially sensitive cost data with us. Because these centers are functioning PET centers, our study is an improvement over previous modeling studies and provides more up-to-date information than the prior estimates. The sensitivity of the cost analysis to four different methods of calculating indirect costs is also examined.
Materials and Methods
This analysis was performed as a substudy of VA Cooperative Study 27, “18-F-Fluorodeoxyglucose (FDG) Positron Emission Tomography (PET) Imaging in the Management of Patients with Solitary Pulmonary Nodules.” Therefore, the respondents to our survey are not a random sample; we sent a survey to each of the 10 sites participating in the clinical trial. Sites included eight Veterans Affairs (VA) and two non-VA hospitals. All sites were at teaching hospitals where research is conducted. The two non-VA hospitals in the survey were both academic medical centers, with 450 and 1020 beds, respectively. Two hospitals did not complete the survey. Both are similar in size to the hospitals that did respond.
We determined the annual cost of operating a PET scanner with and without a dedicated FDG facility from the perspective of the hospital. We used information about whole-body, lung, heart, and brain PET scans at eight sites to calculate costs per scan.
We developed a draft survey instrument and tested it at two sites. A revised survey was mailed to the directors of 10 PET centers in November 2000, and responses were requested by December 15, 2000. The final survey consisted of a section about PET and a section about FDG synthesis. Each section was composed of questions about the cost of equipment, personnel, space, supplies, and repairs needed for PET and FDG-compounding facilities. The survey also included questions about the number and type of scans obtained and the duration of scanning. Centers that do not compound FDG were asked to report the price per dose of FDG that they purchase. The response rate was 80%. Responses were received from seven of the eight VA sites and from one of the non-VA hospitals.
Respondents were contacted to clarify incomplete or unclear responses before the cost analysis began. In some instances, however, whether because of nonresponse or insufficient knowledge on the part of the respondent, reasonable estimates and assumptions had to be made; these estimates and assumptions are explained in detail. There are several ways to handle missing data, each with the potential to bias an analysis. Observations with missing values can be discarded; however, this method of handling missing data would have made our small sample less representative. Missing values can be treated as zero, but this method would have understated costs. Instead, we input the mean value of the other respondents for the same item for missing values.
Respondents were asked to list the position title, full-time equivalent, and annual gross salary or the federal general schedule (GS) salary grade and step for each employee at both the PET and FDG-compounding facilities. Employee benefits were assumed to equal 28% of the salaries. When survey respondents provided the GS grade and step for employees (sites 3, 7, and 8), salaries for the year 2000 were obtained from the GS for that locality. When positions were listed without the complete GS and step information, the mean salaries for the same or similar positions at other sites were used. Mean values were calculated for sites 3 and 7. Physician costs for site 2 were estimated using the average per-scan cost of physicians at all other sites and multiplying this mean cost by the number of scans obtained.
Respondents were asked to report the amount of space occupied by their center in square feet. We obtained the rental cost per square foot of medical office space near each medical center from real estate brokers. This value was multiplied by the total square feet occupied by the center to estimate the total cost of renting space for the center. The rental cost represents the opportunity cost of space. This cost includes the cost of construction, building depreciation, and land and the effect of supply and demand on the cost of medical office space.
Survey respondents reported the year of purchase and the acquisition cost of the PET scanner, cyclotron, and other equipment and the expected lifetime of each piece of equipment. We used this information to estimate the annual cost of each item of capital equipment. We calculated this cost as the payments on a loan for the purchase price of the equipment borrowed for the lifetime of the equipment. We assumed continuous compounding and the interest rate of 10-year Federal Treasury certificates for the year of acquisition. Because the annual payments are eroded by the effect of inflation, this method assumes that the real costs of capital are higher in the first years after equipment is acquired. The effect of temporal trends and technology on manufacturing cost is reflected by our use of the purchase price in the year of acquisition.
We used the equipment lifetime specified by survey respondents to calculate the annual cost of the equipment. Equipment lifetimes depend on when equipment is replaced. This decision is made by the manager at each site who must consider whether to replace old equipment or to continue to incur high rates of repair, more maintenance, more operating labor, and cope with the consequences of lower accuracy. Managers are in the best position to know the plans for replacing equipment and equipment life-time. One site had two PET scanners; the older model was still in use, although it had exceeded its expected lifetime. We used the current age of this machine as its useful lifetime.
The acquisition cost of the PET scanner at site 4 was missing and was assumed to be equal to the cost of the same make and model of the scanner purchased 1 year earlier at site 6. This value was our best estimate given our uncertainty about trends in scanner acquisition costs. The cost of scanner repairs and supplies at site 4 was estimated by multiplying the overall average cost per scan for supplies and repairs from other sites by the number of scans obtained at site 4. Information about the manufacturer, original purchase price, and supply and repair costs of the cyclotron at site 4 was also unavailable. These costs were similarly estimated from the average costs of cyclotron facilities at other sites. The acquisition cost of the cyclotron at site 10 was assumed to equal the cost of the same make and model of cyclotron purchased at site 1.
We excluded research scans from our analysis because we were concerned that these scans might have different costs than clinical scans. Center managers were asked to report the percentage of time the scanner was used for research activities. All scanner facility costs were reduced by this percentage. Similarly, at sites that compound FDG, managers were asked to report the number of total doses of FDG compounded by the cyclotron and the number of FDG doses used for research purposes. All FDG costs were reduced by the percentage of research doses. All FDG costs were further reduced by the percentages of staff time, cyclotron time, and supply costs that were used for compounding other radionuclides.
Respondents were asked about the type and number of scans obtained and the duration of scanning. When respondents gave a range of times for the duration of a scan, the midpoint of the range was used. Missing information about the duration of a brain scan at site 1 was estimated as the average reported by the other centers.
Other cost analyses have estimated that indirect costs are approximately 50–100% of direct costs [25
]. Hospital cost reports use a variety of statistical measures to assign indirect costs. In Medicare cost reports, indirect costs are assigned to departments on the basis of the square feet of space used, gross salaries of employees, and total direct costs incurred by each department. The VA Decision Support System uses direct cost as the statistical basis to assign indirect costs.
We used direct cost as the statistical basis for distributing indirect cost for this study. Direct cost is highly correlated with other measures of productivity and can be a superior basis for assigning indirect cost. For example, a surgical suite that occupies the same number of square feet as an outpatient clinic incurs higher direct costs and is also likely to require more housekeeping services.
To estimate indirect costs, we calculated a ratio of indirect to direct costs using financial records from the VA Decision Support System from fiscal year 2000 and the Medicare hospital cost report from fiscal year 1999. The method used to calculate this ratio is described in more detail. To find the indirect costs associated with PET scanner operation and FDG synthesis, we multiplied the direct cost estimate obtained from the survey times this ratio.
The Decision Support System provides information about the cost of patient care at the encounter and departmental level at the VA. However, we were unable to rely exclusively on the Decision Support System data to calculate the costs of PET at the VA for several reasons. Most importantly, the Decision Support System excludes the cost of capital financing, a considerable component of the total cost of the PET and FDG-compounding facilities. Second, the Decision Support System is a new system, so its limitations and level of accuracy are still not well defined. Finally, our cost analysis includes both VA and non-VA sites; the Decision Support System data are not available for non-VA hospitals.
Definitions of direct and indirect costs used in the Decision Support System and Medicare reports differed from what we considered to be direct and indirect costs for our survey. We therefore had to make several adjustments to the indirect and direct cost totals obtained from the Decision Support System and Medicare before calculating the indirect cost rate.
First, capital (but not financing) costs, including the costs of buildings and equipment, were treated as an indirect cost by both the Decision Support System and Medicare. Because we had a more accurate measure of capital costs from our survey, we excluded capital costs from our indirect cost rate. Second, both the Decision Support System and Medicare listed the cost of medical education as an indirect cost in their cost reports. We did not include education in our estimate of the indirect costs of FDG synthesis and PET scanning because education is a separate product of health care facilities. Finally, the Decision Support System included additional indirect costs, such as research, national and regional administration, and administration of veterans' benefits programs. We excluded these costs from our estimates as well. For an explanation of these excluded costs, please refer to the indirect cost estimate technical report [27
We calculated four indirect cost ratios: two based on Decision Support System reports and two based on Medicare cost reports. The two Decision Support System–based ratios were derived from all VA hospital departments and records of encounter-level outpatient PET scans at the VA, adjusted with hospitalwide information. Medicare-based ratios were derived from the cost of all Medicare hospital departments and diagnostic radiology departments in Medicare hospitals, adjusted with hospitalwide information.
We made two assumptions in estimating the indirect cost ratios. First, we assumed that indirect costs are proportionate to direct costs. We also assumed that the ratio of indirect to direct costs of PET scanning and FDG synthesis is the same as the ratio of indirect to direct costs of other hospital services.
We used outpatient PET scan data because no information was available for inpatient PET scans obtained at the VA. Similarly, Medicare costs for diagnostic radiology departments were used as a proxy for the cost of PET centers because Medicare does not require hospitals to identify PET centers as a separate department when reporting costs.
These four ratios were used to calculate indirect costs of the scanner facility and the cyclotron facility at sites that compound FDG. At sites that purchase FDG, the average ratio was used to calculate the indirect costs of the scanner; however, the indirect costs of purchased FDG were calculated using a lower ratio (25% of the per-dose purchase price). We applied a lower indirect cost ratio because fewer indirect costs are incurred in purchasing FDG than in manufacturing it. We based our estimate on the federally approved indirect cost ratio for grants to research hospitals [28
]. We included only the administrative component, as distinguished from the operating component, of this ratio.
We estimated the direct, indirect, and total costs associated with PET. We performed a survey to estimate direct costs and used data from Decision Support System reports and Medicare reports to estimate indirect costs. We found that the mean per-scan costs of PET was $1885 for sites that manufacture FDG and $1898 for PET centers that purchase FDG. These estimates are consistent with the results of previous analyses, which reported a PET scan costs between $1059 and $4787 [19
In contrast to previous analyses, we did not find a cost advantage for facilities that purchase FDG over those that manufacture FDG on-site. This discrepancy may be caused by several factors. The survey included an open-ended question about equipment used to manufacture FDG (other than the cyclotron). Several sites answered this question in great detail, whereas others provided superficial information. Consequently, the costs attributable to FDG manufacturing may have been underestimated at some sites. Additionally, the centers that compound FDG use the cyclotron for activities other than FDG synthesis 43% of the time and use 15% of the synthesized FDG for research. Although these activities were accounted for when calculating total direct costs at each facility, we may have underestimated the extent to which the production of other radionuclides and research activities reduce the cost of FDG synthesis. On the other hand, previous studies may have overestimated economies of scope—that is, the cost savings resulting from an increase in the diversity of goods produced by one enterprise.
Our study is limited by its small sample size, possible nonresponse bias, and the possibility that VA hospitals are not typical of other providers. Nonresponse to individual survey questions may affect the validity of our study. The response rate to individual questions in the survey varied across sites, which required us to make assumptions about some costs. To ensure data quality, we contacted respondents for clarification if survey responses were incomplete or unclear. However, some respondents did not have enough information to provide complete cost data. It is unclear if the large variation in our results stems from incomplete information obtained by the survey and its small sample size or if it stems from a true variation in costs across centers. All hospitals, both respondents and nonrespondents, were teaching hospitals. Nonrespondents were similar in size to respondents. It is possible that the two centers that did not respond had differentially high or low costs. However, center-level nonresponse bias remains a smaller concern than the limited size of our sample.
Our study may also be limited by the pre-dominance of VA hospitals. This factor must be considered when assessing the generalizability of our findings. However, previous assessments of cost comparisons between VA and nonfederal hospitals have shown that there are no significant differences in their costs [29
]. The salaries of VA physicians have in the past been lower than those of physicians working in the non-VA sector. However, a 1991 reform made salaries for VA physicians more competitive with those for physicians working in the private sector in most specialties and regions [30
]. To determine unit cost, an analyst would also need to determine whether VA physicians are as productive as their private sector counterparts, an issue that we have not considered.
suggest that volume has a considerable impact on per-scan and per-dose costs. Both the PET and cyclotron facilities have high fixed costs (capital costs and staff costs account for 79% of the total direct costs of the PET facilities) and 66% of the total direct costs of cyclotron facilities. Increasing volume can decrease unit costs. There may also be a fixed component to staffing; a base level of staff is needed to operate the scanner and cyclotron regardless of patient volume. Moderate increases in patient volume will not automatically necessitate increases in staff. However, large increases in patient volume will lead to an increase in staffing needs. Therefore, increasing patient volume, up to a certain point, will reduce the per-scan costs attributable to staff. Our finding of a possible inverse association between scan volume and scanning cost agrees with other cost analyses, which have also shown that scan volume affects cost [1
]. In our analysis, this relationship is largely defined by the center that produced approximately 2500 scans. Given the limitations of our study noted, this trend should be considered as a hypothesis for further research rather than as a firm result.
In 2000, Medicare reimbursement was $2185–2301 for a PET scan, depending on the indication and body part scanned. This value can be thought of as a breakeven point for PET centers. Figure 1
suggests that centers with higher scan volumes may be able to break even on Medicare reimbursement or perhaps make a profit. Centers with low scan volumes have difficulty keeping costs below Medicare reimbursement rates. From Figure 1
, the break-even volume for a PET center appears to be approximately 700 scans annually, or approximately two scans per day. However, scan volume is not the only important factor in determining the cost per scan, as shown by one site that had costs above the Medicare reimbursement rate despite obtaining approximately 700 scans per year.
Like most technologies, PET is changing over time. Recent studies have examined FDG imaging with a modified, dual-detector gamma camera. It is unclear how the accuracy of coincidence imaging with a gamma camera, which is referred to as co-PET, compares with traditional PET. The main advantage of FDG imaging with a modified gamma camera is the lower acquisition cost. Adding coincident imaging capability to a standard gamma camera costs approximately $500,000. In contrast, according to our analysis, the average purchase price of a dedicated PET scanner is $1.9 million.
Another avenue for possible change is the development of scanners that obtain anatomic–metabolic fusion images with PET and CT, an integrated CT–PET system. The integrated CT–PET system is approximately 20% more expensive to acquire than a dedicated PET scanner [31
]. The future clinical usefulness of the hybrid system is still uncertain. Another delivery option is the use of a mobile PET center. This setup could improve patient access and decrease overhead costs.
It is unclear how many PET centers are operating in the United States and what the age of equipment at these sites is. Thus, it is unclear how representative the sites in this study are of centers nationally. According to the Academy of Molecular Imaging, 173 PET centers are in the United States [32
]. This total is an underestimate because only seven of the 10 centers in our study were listed among the 173 centers. Additionally, many of the centers listed appear to be free-standing PET facilities; these centers may avoid the extra costs of being hospital-based.
This analysis provides an improved estimate of the costs involved in operating a dedicated PET scanner with and without on-site FDG manufacturing. Additional cost analysis and cost-effectiveness studies are needed to determine the proper roles that dedicated PET scanners and cyclotrons are to play in clinical medicine.