Fluorodeoxyglucose Positron Emission Tomography as an Outcome Measure for Castrate Metastatic Prostate Cancer Treated with Antimicrotubule Chemotherapy

Assessing response to treatment is critical both in routine clinical care and clinical trials. For patients with metastatic prostate cancer, such determinations are complicated by the fact that the primary site of metastatic spread is bone. The dimensions of such osseous lesions cannot be measured using standard radiographic imaging techniques. Furthermore, bone scintigraphy is not an ideal modality for assessing treatment effects, as a brisk antitumor response may manifest as a worsening bone scan, a flare phenomenon known as “pseudoprogression.” Changes on bone scintigraphy also often lag behind clinical events, such as increases in pain, or biochemical changes such as an increase in prostate-specific antigen (PSA). Thus, standard imaging studies are usually not done until 2 to 3 months of treatment have passed. Due to the shortcomings of standard imaging techniques, it has long been recognized that prostate cancer clinical trials are poorly served by standard response criteria, which focus on measurable visceral metastases (1–4). Thus, prostate cancer clinical trials have required disease-specific methods of assessing and reporting outcomes (5, 6). One of the most important elements in assessing treatment effects in this disease is the post-treatment PSA. Post-treatment PSA declines of >50% are associated with a survival benefit (7), as are post-treatment alterations in PSA velocity and/or doubling time (8–10). Despite the apparent linkage between post-treatment PSA alterations and clinical benefits, the PSA is not accepted by the Food and Drug Administration as a surrogate for survival and as a basis for drug approval.

Positron emission tomography (PET) is a noninvasive technique that can image bone and soft tissue in a single modality, evaluate high-grade tumors that may not produce PSA, and provide a quantifiable expression of change using the standardized uptake value (SUV). Although PET has the potential to do all of these functions, the study of PET for prostate cancer has a checkered history. Studies of PET as a staging tool for prostate cancer have historically examined heterogeneous groups of patients. These studies have, unsurprisingly, revealed inconsistent results (11–16).

The ability of PET to detect active prostate cancer in bone has been shown when PET was studied in discrete groups of progressing patients who represented a single point in the natural history of the disease (17). In the present trial, this same approach was used to explore PET as an outcome measure. We sought to determine whether treatment produced changes in PET that would parallel clinically judged treatment responses based on PSA, bone scintigraphy, and soft tissue imaging. Patients were strictly controlled for progressive metastatic disease despite castration, and all received chemotherapy that acted by interrupting microtubule trafficking. Finally, all patients received PET scans, PSAs, bone scans, and soft tissue imaging at fixed intervals, and scans were read by reference radiologists.

All patients were imaged as part of a clinical trial (IRB 97-007) approved by the Institutional Review Board of Memorial Sloan-Kettering Cancer Center. Written informed consent was obtained from all patients.

Eligibility. Patients were eligible based on having progressive, histologically proven, metastatic prostate cancer that progressed despite castrate testosterone levels. Patients were required to have at least a ≥50% increase in PSA sustained for a minimum of three observations obtained at least 1 week apart, development of new lesions, or an increase in preexisting lesions on bone scintigraphy or in measurable disease by computerized tomography or magnetic resonance imaging. All chemotherapy was given in the context of therapeutic clinical trials. Patients were coregistered to this study, which was a pure imaging study. If the eligibility criteria of the therapeutic trial were more stringent than that of the imaging study (e.g., in defining the degree of soft tissue disease that qualified as progression), then the eligibility of the therapeutic trial superseded that of the imaging trial. Patients were required to have a Karnofsky performance status of >60%.

Treatment. Patients had progressive metastatic prostate cancer despite castrate testosterone levels and received chemotherapy that targets microtubules. The treatment regimen and schedule were dictated by the therapeutic trial to which they were coregistered. All patients were followed until progression of disease or death. Time to progression was defined by each clinical trial but used a combination of PSA, bone scan data, and soft tissue imaging which is standard in prostate cancer by consensus (6). None of the therapeutic clinical trials employed fluorodeoxyglucose PET (FDG-PET) as a criterion for defining progression or response.

Imaging. Patients received a pretreatment PET, bone scan, soft tissue imaging study with either computerized tomography or magnetic resonance imaging, and PSA assay. At week 4 of treatment, patients received a PET and a PSA; at week 12, the PSA, PET, bone scintigraphy, and soft tissue imaging were repeated.

FDG-PET imaging was done on the ADVANCE (General Electric Medical Systems, Milwaukee, WI) whole-body PET scanner. With the exception of water, patients fasted 6 hours before injection with 10 ± 1 mCi (370 ± 37MBq) of 18 FDG, which was provided by the Memorial Sloan-Kettering Cancer Center Cyclotron Core facility using the method of Hamacher et al. (18). Standard whole-body scans were obtained, with most scans encompassing the body from the clavicle to below mid-thigh. Imaging was done using transmission correction of all fields of view, and a filtered backprojection reconstruction algorithm was used.

Image analyses. Up to five FDG-positive indicator lesions were identified on the baseline PET scan. These were selected based on being anatomically discrete and identifiable. Images created with an iterative reconstruction algorithm as well as a filtered backprojection algorithm were reviewed. Region-of-interest tools available with the commercial release of the camera's software were used, and the maximal SUV within the tumor was recorded (SUV_max). As the study was originally started when only filtered backprojection was available, we chose to continue only with this methodology.

SUV_max was defined as:

The average SUV_max of the indicator lesions was calculated, and each scan was subsequently assigned an “average SUV_max” (SUV_maxavg) value. In addition, comparison was made with the baseline scan, and appearance of a new lesion was considered progression, despite change in SUV_maxavg.

All PET and bone scans were read by a single reference nuclear medicine physician (T.A.). All soft tissue imaging was read by a single reference radiologist (L.S.B.).

Statistical methods. Imaging and biochemical data were analyzed at baseline, 4, and 12 weeks of treatment. For SUV and PSA, post-treatment changes were examined based on whether either a variable increased or declined relative to baseline. We presumed that most clinicians would continue therapy on which patients were not progressing and therefore dichotomized post-treatment changes as either representing progression or nonprogression. We evaluated changes in SUV_maxavg and PSA, starting with any increase over baseline as a true increase. Recognizing that many clinicians would not change treatment for a nominal increase in PSA or SUV, we reanalyzed the data defining progression as >10% or 25% increase, the latter representing the PSA Working Group Consensus Criteria for progressive disease (6). We also inspected the standard clinical response data to see if there might be a better threshold for SUV change. The appearance of new lesions seen on the PET scan was treated in the same manner as an increase in SUV_maxavg of the indicator lesions. Hence, even if the SUV_maxavg declined, the appearance of new lesions on the scan would categorize the patient as progressing.

Estimates of the probability of discordance between standard outcome measures and SUV_maxavg were computed, and a 95% upper confidence bound for the probability of discordance was calculated. The confidence bounds were derived from the binomial distribution.

Patients and treatment. Twenty-three patients with progressive castrate metastatic prostate cancer, per the Memorial Sloan-Kettering Cancer Center clinical states model (19), were treated on four therapeutic clinical trials and coregistered to the PET study. Twenty-four scan sets were done as shown in Table 1. One patient underwent two discrete antimicrotubule treatments and therefore accounts for two scan sets. In total, 68 PET scans were done. Twenty-four PET scans were done at baseline and 4 weeks. Twenty PET scans were done at week 12, as four patients were unable to continue on study due to clinical events related to treatment or progressive disease. Two patients had a prolonged delay between injection and imaging at their baseline PET scans. These two scan sets were therefore considered unevaluable. Thus, there were 22 evaluable PET scans to compare with PSA changes at 4 weeks, and 18 evaluable PET scans to compare with PSA changes and standard imaging studies at 12 weeks. A detailed description of the scan sets may be found in Table 2. Scan sets “K” and “U” were derived from the same patient who completed two separate courses of antimicrotubule therapy on two different clinical trials.

Table 2 summarizes the data describing each scan set in terms of change in SUV_maxavg, PSA, and standard imaging studies. Also summarized are the treating physicians' clinical assessments of the patient at week 12, using standard outcome measures, times to progression from the 12-week assessments, and the number of index lesions followed. In terms of index lesions used to calculate SUV_maxavg, 16 patients had five lesions, two patients had four, three patients had three, and one patient had two. The predominance of osseous index lesions over soft tissue lesions reflects the bone tropism of prostate cancer. Patients A and E had bone scans that showed either an increase in the number of lesions or an increase in the intensity of lesions, in the face of either clinical improvement and/or a declining PSA. Their bone scans were felt to represent radiographic flare rather than disease progression by their treating physicians and were treated as indicators that the patient was responding to treatment. Of note, patient Q underwent standard imaging at week 8 rather than at week 12 to satisfy his therapeutic clinical trial requirements.

Of the patients represented by the 22 evaluable scan sequences, 5 progressed and 17 were stable by week 12 of treatment. Sixteen patients were treated with paclitaxel 100 mg/m² weekly, carboplatin (AUC = 6), every 4 weeks, and oral estramustine 10 mg per kg per day or dose-escalated IV estramustine (TEC and HI-TEC) for 5 days. The details of these regimens may be found elsewhere (20, 21). Two patients were treated with weekly docetaxel 30 mg/m2 with estramustine 10 mg/kg for 5 days each week for 8 weeks, followed by weekly doxorubicin at 20 mg/m² and ketoconazole (TEAK). One patient was treated with docetaxel 70 mg/m² every 3 weeks, carboplatin (AUC = 6) given every 3 weeks, and estramustine 280 mg tid four to five times per week on the week of treatment. Two patients received BMS 247550, an epothilone, at 40 mg/m² with oral estramustine 280 mg thrice a day for 5 days (22).

Correlations among SUV_maxavg, prostate-specific antigen, and clinical status at 4 weeks using “any increase” to define disease progression.Table 3 summarizes the correlation between changes in SUV_max and PSA at 4 weeks, using any increase in PSA or SUV_maxavg to define disease progression. Twenty-two patients were assessable. The PSA and SUV_maxavg changed in parallel in 18 of 22 patients (82%), simultaneously declined in 17 patients, and increased in one patient. Four patients (18%) showed divergent PET and PSA findings at 4 weeks. Three patients had an increase in SUV_maxavg (or new lesions) at 4 weeks but a decline in PSA. For cases K and E, the PET incorrectly suggested disease progression. By contrast, the PET accurately detected early progression of disease that was otherwise not discernible in patient C. The PET also seemed reflected nonprogression in the case of patient Q.

Therefore, the PET accurately reflected the patients' clinical status in two (C and Q) of the four cases in which the PSA and SUV_maxavg did not change in parallel at 4 weeks. Thus, in 20 of 22 cases (91%), the 4-week PET accurately reflected the clinical status of the patient.

Correlations among SUV_maxavg, prostate-specific antigen, standard imaging, and clinical status at 12 weeks using any increase to define disease progression. The relationship between standard imaging studies, SUV_maxavg, and PSA at 12 weeks is described in Table 4. These data are based on a definition of progression as any increase in PSA or SUV. For 13 of 18 patients (72%), the SUV_maxavg, PSA, and standard scans all changed in parallel.

Five patients (N, H, Q, K, and S) had scans, PSA, or SUV measurements that were not in agreement. For patients N and Q, the PET accurately described the patient's clinical status, but PET incorrectly described the status of patients K, S, and H.

In total, therefore, the PET accurately reflected the clinical status of the patients in 15 of 18 cases (83%).

Correlations among SUV_maxavg, prostate-specific antigen, and clinical status using a 10% or 25% increase in prostate-specific antigen or standardardized uptake value to define disease progression. The 4- and 12-week data were reanalyzed using a PSA or SUV_maxavg increase of at least 10% or 25% to define progression, in recognition of the fact that most clinicians would not take a patient off treatment based on a minimally increased value over baseline.

Allowing for a 10% increase resulted in no changes in the data at either 4 or 12 weeks. Table 5 summarizes the data at 4 weeks using a cutoff of 25% to define an increase in PSA or SUV_maxavg. PSA and PET were in agreement in 19 of 22 cases (86%). In 18 of these cases, the PSA and SUV_maxavg both declined. The median PSA decline was −72% (range, −95% to 17%) and the median SUV_maxavg decline was −43.5% (range, −71% to −4%). There was disagreement between the PSA and SUV in 3 of 22 patients (14%). Based on these data, we are 95% confident that the true probability of a discordance between PSA and SUV_max lies below 0.32.

Table 6 summarizes the data when the same 25% rule is applied at 12 weeks. Using these thresholds, PET accurately reflects clinical status in 17 of 18 cases (94%). In 16 of those cases, the PSA, scans, and SUV_maxavg suggested that the patient was not progressing on therapy. The median PSA decline was −88% (range, −100% to 17%) and the median SUV_maxavg decline was −55% (range, −79% to 20%). Disagreement between PSA and SUV_maxavg was observed in 1 of 18 patients (6%). Based on these data, we are 95% confident that the true probability of a disagreement between PSA and SUV_maxavg lies below 0.24. Disagreement at 12 weeks between any two of the three outcome measures was observed in 2 of 18 patients (11%). The 95% upper confidence bound for the probability of discordance between any two modalities is 0.31.

Although applying Consensus Criteria for biochemical progression to PET resulted in a good correlation between PET and clinical status, we asked whether another threshold would work even better. If a >33% increase in SUV_maxavg or the appearance of new lesions was used, then patient K in Table 5 would be reclassified as not demonstrating disease progression, and PET would accurately identify the clinical status of 21 of 22 patients (95%) at 4 weeks, or 17 of 18 patients (94%) at 12 weeks.

Prostate cancer is a disease with a paucity of good outcome measures, as most metastases localize to bone and are therefore not measurable. By consensus, investigators now independently report PSA, bone scan findings, and changes in soft tissue imaging, in recognition that each of these outcome measures tells only part of the story (6). As an imaging modality, PET has the potential to capture all of these data in a single study. The SUV is quantifiable like the PSA, and PET can assess both bone and soft tissue disease simultaneously, including the ability to detect the appearance of new lesions.

This study is the first to systematically examine PET as an outcome measure in prostate cancer and to compare PSA changes, SUV changes, and standard imaging studies in a controlled population. As most clinicians continue therapy on which patients are not progressing, we examined whether PET could accurately distinguish between nonprogressing patients and those with progressive disease as defined by standard outcome measures. We applied a variety of thresholds for PSA and SUV_maxavg to define progression, in recognition of the fact that such cutoffs are arbitrary, made without the support of prospective data, and are not validated.

We first analyzed the data using a stringent cutoff of any increase in PSA or SUV_maxavg as defining a true increase representing progressive disease. Using this “no increase” versus “stable or better” rule, we found that the PET accurately reflected clinical status in 20 of 22 patients (91%) and changed in parallel with the PSA in 18 of 22 patients (82%) at 4 weeks. At 12 weeks, the PET accurately described the clinical status in 15 of 18 cases (83%). In 13 of 18 cases (72%), patients had PET scans, PSA data, and standard imaging studies change in parallel.

Most clinicians would feel uncomfortable using these stringent criteria to define progressive disease, as negligible increases in PSA or SUV_maxavg can be attributed to technical factors and not necessarily disease progression. We therefore reexamined the data using the PSA Working Group Phase II Consensus Criteria's definition of progressive disease, with which most investigators are familiar. The Working Group defined a 25% increase in PSA over baseline as representing progressive disease, unless the patient has achieved a PSA post-treatment PSA decline of >50%, in which case progression is an increase of 25% above nadir (6). As we did not examine the degree of decline of either PSA or SUV_maxavg in this analysis, we simplified the criteria to a 25% increase over baseline for all patients. Using these criteria, PET was highly accurate in reflecting clinical outcome: It accurately described the clinical status in 20 of 22 cases (91%) at 4 weeks and 17 of 18 cases (94%) at 12 weeks. It also seemed that the SUV and PSA more closely mirrored each other; the PSA and SUV were in agreement in 19 of 22 cases (86%) at 4 weeks and 17 of 18 cases (94%) at 12 weeks. These results suggest that allowing a 25% increase over baseline before categorizing either variable as truly “increasing” is more favorable than either of the two other alternatives tested. Raising the threshold to 33% further increased the accuracy of PET, as it correctly identified the clinical status of 21 of 22 patients (95%) at 4 weeks, although the accuracy at 12 weeks remained unchanged. We recognize that a methodologic weakness in this comparison is the fact that PSA is integrally involved in the judgment about whether a clinical response has, in fact, occurred. The fact that FDG-PET actually done as well or better in predicting clinical response is still reassuring for the role of PET imaging.

Although some might argue that the PET is therefore just an expensive PSA, it may substitute not only for the PSA but for bone scintigraphy and soft tissue imaging in terms of indicating treatment effects. Even if this modality does not replace these outcome measures, FDG-PET may well complement standard methods of assessing treatment effects in routine clinical practice. For example, clinicians may obtain routine interval scans even if the PSA is declining to confirm that the PSA decline is not masking early progression of resistant clones, or to verify the absence of an emerging non-PSA producing phenotype. In these scenarios, the PSA and scans may be at odds. FDG-PET may have a role in clarifying whether a patient is responding or progressing in the face of these discordant data. In addition, clinicians who do not routinely perform scans and primarily follow the PSA to guide therapy may now have a reliable method of directly assessing skeletal tumor burden and of detecting early progressive disease. Finally, FDG-PET may substitute for the PSA as an early response indicator in patients whose tumors produce scanty levels of PSA. Of course, further study is needed to establish whether PET scanning can serve in these roles. Because PSA is universally applied for assessing prostate cancer during therapy despite its limitations, and because there is no clear alternative beyond clinical consensus as a gold standard for assessing response, it may prove difficult to actually prove that FDG-PET is better overall than PSA in predicting or monitoring response.

We selected up to five indicator lesions representing bone and soft tissue disease feasible to track and treated any new lesion as an indicator of new disease. An ideal scenario would have been to track every single lesion. The impracticality of tracking every lesion is recognized, and indeed standard outcome assessments are made on the basis of a maximum of five lesions per organ and 10 lesions in total according to the Response Evaluation Criteria in Solid Tumors (23). Whether five lesions represented too few lesions might only be determined by a comparative analysis of outcome using a variety of lesion sets, which is beyond the scope of this paper.

In addition, we averaged the maximum SUV to express changes in these five lesions, which risks homogenizing the data. For example, if soft tissue disease were responding to therapy differently than bone disease, such information would be lost by averaging the SUV of the five index lesions. However, because we did treat any new lesion as progressive disease, presumably progressive disease would be detected despite averaging the maximal SUV of the index lesions.

Finally, the patients in this study all received active antimicrotubule therapy. As the median time to progression on antimicrotubule therapy is ∼4 to 6 months (24, 25), the patients in this study were generally responding to treatment at the time that they were scanned at 4 and 12 weeks. To properly explore FDG-PET as a means of distinguishing the patient who is responding to therapy from one who is progressing, this modality will need to be examined not only at the treatment start but later, at the time of treatment failure, as well.

These methodologic issues underscore the fact that PET warrants further study as an outcome measure. Nonetheless, these preliminary data suggest that PET is an accurate descriptor of treatment effects, reflecting in a single modality the effects now captured by three measures: PSA, bone scintigraphy, and soft tissue imaging.