The purpose of this article was to evaluate the utility of a pharmacokinetically modeled measure of regional contrast access, based on dynamic contrast-enhanced magnetic resonance imaging (MRI) studies after preoperative chemotherapy, as a predictor of disease free survival in osteosarcoma.
The kinetic parameters of a two-compartment pharmacokinetic model of MRI contrast agent accumulation were analyzed in relation to disease free survival in 31 patients who received protocol-based therapy for nonmetastatic osteosarcoma of the extremities. The modeled exchange rate of contrast between the plasma and the tumor extravascular extracellular fluid space served as a measure of regional contrast access. The prognostic impact of both the clinically accepted standard of histologic evaluation of tumor necrosis and the regional contrast access were analyzed with tumor size as an influential factor.
Although the histologic grade of response was not a statistically significant prognostic factor in these patients (P = 0.884), regional contrast access after preoperative chemotherapy was significantly predictive of disease free survival (P = 0.035) in the Cox proportional hazards model. Lower regional access before surgery and smaller tumor size were associated with a better treatment outcome. Log-rank analyses of Kaplan–Meier curves indicated that the impact of regional access was most pronounced in patients with larger tumors (P = 0.052). Higher regional access at presentation also was associated significantly with greater decreases during therapy.
For chemotherapy to be effective, cytotoxic agents must reach tumor cells in adequate concentrations and with minimal toxicity to normal tissues.1 The resistance of solid tumors to drug penetration likely plays a significant role in treatment failure.2 Once a blood-borne molecule reaches an exchange vessel, transport across the vessel wall is governed by the surface area of exchange and by the transvascular concentration and pressure gradients.3 Vascular permeability varies both spatially and temporally within a given tumor and between various tumor types.4 In animal studies, transient irradiation-induced increases in tumor capillary permeability to cisplatin can be quantified with contrast-enhanced magnetic resonance imaging (MRI).5 We hypothesized that the transfer rate of a low-molecular-weight MRI contrast agent between the vasculature and the tumor extracellular fluid (termed “regional access”) would provide a surrogate measure of drug access. If this is the case, measures of regional access should be related to rates of disease free survival in solid tumors such as osteosarcoma and might provide valuable information for the planning and refinement of treatment regimens.
The response of pediatric osteosarcoma to preoperative chemotherapy is an important prognostic factor.6–8 The extent of chemotherapy-induced tumor necrosis has been strongly associated with effective local control and survival.9, 10 However, identically treated tumors show variable drug sensitivity, as measured by traditional histologic evaluation of viable versus necrotic tissue after en bloc resection.8 More recently, the use of dynamic contrast-enhanced MRI before surgery has proven effective in predicting the histologic response of osteosarcoma to preoperative chemotherapy.11–17 This finding, combined with the fundamental characteristics of dynamic MRI evaluations, suggests that data obtained with this technique may prove useful in assessing the overall likelihood of treatment success.
Dynamic MRI studies consist of rapid sequential imaging of the delivery of a paramagnetic contrast agent (gadopentetate dimeglumine or gadodiamide) into the tumor capillaries and its subsequent diffusion into the extravascular space. Because the low-molecular-weight chelates of the contrast agent do not cross cell membranes, several groups have used dynamic MRI to investigate the pharmacokinetics of contrast agent delivery in brain lesions and, more recently, breast tumors.18–24 However, similar studies have not been performed in pediatric bone tumors, which pose challenges in chemotherapy response assessment and prediction of overall treatment outcome. We performed dynamic MRI studies at presentation and after completion of preoperative chemotherapy in children and young adults with osteosarcoma. These studies originally were analyzed using the dynamic vector magnitude (DVM) method previously shown to correlate with histologic assessment of resected tumor necrosis16 Because the DVM method characterizes the change in the signal intensity of the MRI images as a function of time, it is dependent on field strength, coil, coil loading, imaging sequence, and many more influential factors that make it difficult to compare results between various imaging facilities. Therefore, we retrospectively calculated estimates of regional contrast access by using a two-compartment pharmacokinetic model combined with an MRI signal model of the sequence used to acquire the data. This combination of models allows for comparisons between institutions using various sequences and machines. The ability of this new measure to predict disease free survival was evaluated, with the goal of identifying a noninvasive biologically based outcome predictor that did not require a resected specimen.
PATIENTS AND METHODS
Study participants were drawn from 69 children and young adults enrolled on an institutional protocol for newly diagnosed, histologically confirmed osteosarcoma25 who underwent comparable imaging studies between September 1991 and July 1997. Eligibility criteria for the current study included resectable primary nonmetastatic lesions, no evidence of disease after surgery, and completion of all imaging examinations. Thirty-one patients met these study criteria. Thirty-eight patients were excluded from the study for the following reasons: histologic diagnosis other than high-grade osteosarcoma (malignant fibrous histiocytoma [n = 6] and multipotential sarcoma of bone later confirmed as Ewing sarcoma [n = 1]), unresectable primary lesions (mandible [n = 1], maxilla [n = 1], nasopharynx [n = 1], illium [n = 3], sacrum [n = 2], pubis [n = 1], spine [n = 2]), and metastatic disease at diagnosis (n = 13), refusal of preoperative chemotherapy after the first course (n = 1), distant secondary malignancy (chondrosarcoma [n = 1]), and incomplete imaging studies (n = 5). The patients with other diagnoses were excluded to provide a uniform patient population for the study. Patients with unresectable lesions and metastatic disease were excluded because of the significantly different disease free survival rates in these patients. The patient that refused preoperative chemotherapy after the first dose and the one patient that developed a distant chondrosarcoma during preoperative therapy were excluded because treatment of these patients was modified and was no longer comparable to the study population. These exclusions should not bias the sample but should provide a uniform sample of patients with high-grade osteosarcoma treated with the same preoperative and postoperative chemotherapy regimen and complete surgical resection of the primary lesions. Treatment and imaging protocols were approved by the hospital's Institutional Review Board and written informed consent was obtained from the patient, parent, or guardian, as appropriate.
Treatment and Imaging Procedures
All patients received preoperative chemotherapy with three cycles of intravenous ifosfamide (2.65 g/m2/day for 3 days with mesna uroprotection) and carboplatin (560 mg/m2), given at 3-week intervals over 9 weeks.25 After surgery (en bloc resection), patients received 30 weeks of chemotherapy comprising high-dose methotrexate (12 g/m2 × 9 doses), doxorubicin (75 mg/m2 × 5 doses), and 2 additional cycles of ifosfamide and carboplatin. The response to preoperative chemotherapy was assessed by histologic evaluation after en bloc resection to determine the percentage of tumor necrosis.8 Imaging studies and physical examinations were performed at scheduled intervals before, during, and after treatment to monitor response and, as clinically indicated, to evaluate possible recurrence.
The MRI examinations were performed at presentation and after preoperative chemotherapy, a median of 6 days before en bloc resection (range, 1–31 days). A 1.5-T SP63 Magnetom unit (Siemens Medical Systems, Iselin, NJ) was used, with the standard quadrature body coil as transmitter and receiver. After selection of the longitudinal (usually coronal) section best exhibiting the tumor, 30 sequential fast low angle shot (FLASH) images (TR/TE = 23/10 milliseconds, 40° flip angle, 256 phase encodes, 10-mm slice thickness, 40– 50-cm field of view, 2 acquisitions) were acquired over a 7-minute period before, during, and after a bolus injection of 0.1 mmol/kg gadopentetate dimeglumine, as described previously.16, 26 Dynamic MRI studies were performed concurrently with routine clinical MRI (longitudinal T1-weighted and short time inversion recovery images; transverse T2-weighted and contrast-enhanced T1-weighted images).
After the imaging data were transferred to an off-line work station, a pediatric radiologist (S.C.K.) used an interactive display to select the region of interest encompassing the tumor area identified on routine clinical imaging studies.16, 26 A second radiologist (B.D.F.) reviewed the MRI data for each patient to confirm tumor boundary selection. Tumor size was defined as the cross-sectional area of the tumor (in square centimeters) and was calculated based on the number and size of imaging pixels in the region of interest.
Because MRI contrast does not cross cellular membranes, the contrast agent is present only in the plasma and the extracellular fluid space. Hence, a conventional two-compartment pharmacokinetic model comprising plasma volume and extracellular fluid space volume was used to analyze contrast distribution in each voxel of the dynamic MRI examinations.18, 19, 22 The system equations for the model are functions of contrast agent infusion rate, rate of transfer between the vasculature and extracellular fluid space, and rate of elimination or extraction from the plasma in the vasculature. The modeled exchange rate of contrast between the plasma and the tumor extracellular fluid space (designated as kep) was used as the measure of regional access.
The contrast agent was infused at a constant rate over approximately 5 seconds (less than the time required for one image acquisition). Assumptions of the pharmacokinetic model were as follows: unhindered contrast exchange in both directions, with instantaneous mixing of contrast throughout the blood volume;22 a constant elimination rate from the plasma of 0.06 minute−1 during the dynamic MRI examination (resulting in plasma concentrations approximating the empirically determined biexponential elimination rates reported by Weinmann et al.27); and no interaction between the extracellular extravascular fluid compartment of adjacent voxels during the measurement period. We used a linear relation between the total tissue concentration and the extracellular fluid concentration to simplify the model of signal intensity in the presence of contrast, in accord with the Bloembergen equation.28, 29 This equation assumes that incremental increases in local tissue relaxation rates are directly proportional to the local tissue contrast concentrations. We also assumed a constant T1 relaxation rate of 700 milliseconds in the absence of contrast. The pharmacokinetic model of regional MRI contrast access then was combined with the signal intensity equation for the FLASH imaging sequence to complete the analysis procedure30, 31 (See Fig. 1).
Disease free survival was defined as the interval from complete surgical resection to treatment failure or most recent follow-up. Probabilities of disease free survival were estimated using the method of Kaplan and Meier, and the survival curves were compared with the log-rank test. The mean value of the regional access parameter, kep, in the identified region of interest was determined for each patient. The impact of histologic response8 and regional access on survival were evaluated using the Cox proportional hazards model.32, 33 Potential influential factors of patient gender, age, and tumor size as well as interactions between these variables and nonlinear terms were evaluated in the Cox model. All reported P values are for two-sided chi-squared tests, and an acceptance criterion of 10% was used to assess significance in these exploratory analyses (i.e., P < 0.1 accepted as trending toward significance). The relation between regional access at presentation and change in regional access during preoperative chemotherapy was assessed with a linear regression.
The 31 patients studied (17 female, 14 male) had a median age of 13.0 years at the diagnosis of osteosarcoma (range, 5–24 years). The predominant tumor histology was osteoblastic (n = 16); four tumors were classified as chondroblastic, four as fibrohistiocytic, four as fibroblastic, two as telangiectatic, and one as multipotential. Most tumors were located in the femur (n = 21) or tibia (n = 6); other sites included the humerus (n = 2) and fibula (n = 2). Primary tumor sizes (cross-sectional area) ranged from 11 to 130 cm2, with a median of 56 cm2. Large tumors were defined as those lesions with cross-sectional areas greater than or equal to the median size.
The 2-year estimated disease free survival rate in this patient cohort of 31 patients was 76.1% ± 8.0% (standard error). At the time of analysis, there had been 7 recurrences, which occurred 5 months to 4.5 years after surgery. Follow-up intervals ranged from 5 months to 6.0 years with a median of 2.1 years. Seventeen patients were responders by histologic grading of necrosis, whereas the remaining 14 patients failed to achieve the required 90% necrosis at resection. For the 31 patients, the median measure of regional contrast access (kep), based on pharmacokinetic modeling of the MRI data, was 1.167 minute−1 at completion of preoperative chemotherapy (mean, 1.25 minute−1; standard deviation, 0.40 minute−1; range, 0.34–2.54 minute−1).
Histologic grade of response, based on the degree of necrosis in the resected lesion, was not found to be a prognostic factor of disease free survival in these patients (P = 0.884) using the Cox proportional hazards model (Table 1). Only tumor size was found to be a statistically significant prognostic factor, with smaller tumors being predictive of improved outcome. The additional factors of patient age and gender were not statistically significant. To investigate the impact of histologic grade of response on survival, we stratified patients by histologic grade of response with responders exhibiting greater than 90% necrosis. The Kaplan–Meier curves of disease free survival for the two groups were compared using the log-rank test. As shown in Figure 2, the degree of necrosis at the completion of preoperative chemotherapy was not a statistically significant prognostic factor (P = 0.538).
Table 1. Cox Proportional Hazards Model, Based on Disease Free Survival, for the Entire Seriesa
Regression coefficient (standard error)
Prognostic value of two models were evaluated: 1) histologic assessment of response (responder [>90% necrosis] or nonresponder [<90% necrosis]), and 2) regional access (kep). Both models were evaluated at completion of preoperative chemotherapy and included tumor size as an influential factor. P values are given for a two-sided chi-squared test. For the Regional Access model, the log ratio hazard (LRH) is equal to (0.4635∗kep + 0.0348∗Size). Smaller kep after therapy and smaller tumor size would correspond to smaller LRH and therefore better survival rates. The larger the regression coefficient, the more influence the variable has on the LRH and therefore survival rate.
Using the Cox proportional hazards model, the estimated coefficient for kep at completion of preoperative chemotherapy was significantly correlated to treatment outcome, with lower regional access being predictive of improved outcome (Table 1). The estimated coefficient for kep at presentation and the change in kep during therapy were not significantly correlated with outcome. As before, the factors of patient age and gender were not statistically significant prognostic factors in this group of patients. Size of the tumor was also a strong prognostic factor with smaller tumors being predictive of improved outcome. To investigate the impact of regional access on survival in relation to other factors identified in the Cox model, we stratified patients into subgroups defined by tumor size and kep below or above the medians. The curves of Kaplan–Meier estimates of disease free survival for these subgroups were compared using the log-rank test. As shown in Figure 3, the risk of recurrence was associated with higher regional access at the completion of preoperative chemotherapy in patients with larger tumors (P = 0.052 for larger tumors). Note that the relatively low numbers of patients in specific subgroups limits the power to detect differences. In this series, patients with tumors smaller than the median size exhibited 100% disease free survival up to 4 years after resection regardless of regional access.
Although kep at completion of preoperative chemotherapy was significantly correlated to treatment outcome, it is preferable to assess patient risk at presentation. Because lower regional access after preoperative therapy was predictive of improved outcome, we investigated the relation between regional access at presentation and change in regional access during therapy. Two of the patients were identified as outliers and were excluded to achieve a normal distribution for change in regional access. The linear regression of change in regional access as a function of regional access at presentation had an r2 of 0.85 and is shown in Figure 4.
The rationale for preoperative chemotherapy in the treatment of primary osteosarcoma was proposed by Rosen et al. in 1979,34 with the goal of delaying surgery to provide time to obtain custom-made endoprostheses. Histologic examination of the primary tumor removed at surgery revealed varying degrees of tumor destruction attributable to chemotherapy, with a positive impact of greater tumor response on disease free survival.6, 10, 34, 35 In more recent studies, the percentage of tumor necrosis induced by preoperative chemotherapy has been predictive of disease free survival.9 However, this feature cannot be evaluated at multiple time points during therapy and is applicable only to patients with resectable tumors who have completed preoperative chemotherapy. Preliminary studies indicated that the DVM method of evaluating the dynamic MRI signals provided a noninvasive measure of good response to therapy (>90% tumor necrosis before surgical resection),16 This relation is confirmed in this study, which includes all of the patients from the previous report, with a predicted accuracy of 83% (24 of 29), sensitivity for responders of 81% (13 of 16) and specificity for nonresponders of 85% (11 of 13) by using a DVM threshold of 1.9. Two patients could not be evaluated with the DVM method because they were imaged on different magnets and the DVM method is a function of signal intensity, which could not be easily compared at different field strengths. The evolution from empirical to pharmacokinetic modeling of dynamic contrast-enhanced MRI images to assess microcirculation and the rationale for its application in bone sarcoma have been thoroughly described in recent literature.36–38
We found that a combination of tumor size and a pharmacokinetically modeled measure of regional contrast access (kep) determined at completion of preoperative chemotherapy was significantly associated with disease free survival in pediatric osteosarcoma. In the Cox proportional hazards model, lower kep estimates after 9 weeks of preoperative chemotherapy, predicted a better disease free survival. Of interest, for the 31 patients included in the current study, the histologic grade of response was not predictive of disease free survival. This lack of difference in outcome based on histologic response also was demonstrated in the analysis of the whole patient population treated on the protocol.25 This finding is inconsistent with many previous studies but may be due to study design. Lack of response to the experimental drug pair used in the preoperative window would not necessarily predict a poor outcome when the remainder of the therapy is composed of agents of known efficacy in the treatment of osteosarcoma.
The effective regional access is a function of the microvascular density present in the tumor. A low value for the effective regional access after preoperative chemotherapy may indicate that a high proportion of the tumor tissue either is necrotic or is still viable with a reduced microvasculature that could indicate decreased angiogenesis. These viable regions may have responded to the chemotherapy but were not yet necrotic. The regional access measure produced by our modeling of data from dynamic MRI studies is a function of both flow and permeability of vessels. Tumors with less response to this chemotherapy would have larger viable regions, which are highly angiogenic and therefore more permeable to the contrast agent, resulting in a higher value for regional access.
The prognostic value of the primary tumor size has been established previously,39 but the finding of an interaction between this feature and regional access is novel. This relation is in accord with studies in xenograft models, which have shown that larger solid tumors have increased interstitial pressures, resulting in decreased regional access at or near the core of the tumor.1–4 These findings suggest that higher levels of interstitial pressure decrease the ability of chemotherapeutic agents to permeate the tumor periphery sufficiently to exert therapeutic effects.
Analyses of the impact of regional access, as a function of tumor size was particularly informative. Comparison of Kaplan–Meier curves identified a subset of patients with larger tumors—those with regional access above the median after preoperative chemotherapy—in whom the risk of recurrence tended to be increased. These data indicate that regional access is an important predictor of outcome in children with localized large osteosarcoma treated on a contemporary regimen involving preoperative chemotherapy, complete resection, and multiagent postoperative chemotherapy. Unfortunately, the impact of regional access in the subset of smaller tumors was inconclusive because 100% of patients were disease free up to 4 years after resection regardless of regional access. A longer follow-up period (> 4.5 years) for all patients will be necessary to detect if the significance of regional access in these patients. In future studies, this evaluation might prove useful in patients with unresectable lesions of the axial skeleton, whose responses to preoperative chemotherapy cannot be evaluated directly. Combining information from both the DVM measure (related to percent necrosis) and the regional contrast access (related to microvascular environment) could provide useful information about the response of the unresectable lesion to the therapy.
An additional analysis of the relation between regional access at presentation and change in regional access during therapy demonstrated that greater regional access at presentation corresponded to higher decreases in regional access during therapy. These results are consistent with the hypothesis that a noninvasive measure of regional microcirculatory access of small molecular weight intravenous contrast agent may provide a measure of the access of some chemotherapeutic agents to the tumor. Because regional access at presentation is related to change in regional access during therapy and regional access after therapy is related to disease free survival, there should exist a relation between regional access at presentation and disease free survival. However, with the limited number of patients in this study and the amount of variance inherent between patients, we were unable to establish this relation.
These findings, however, if confirmed in larger prospective trials, suggest that dynamic contrast-enhanced MRI-based evaluations of regional access eventually could contribute to the development of risk-adapted therapy for pediatric osteosarcoma. It may be possible to use a combination of tumor size and regional access in efforts to identify those patients at increased risk of recurrence in whom more intensive or novel therapies are needed. The efficacy of this or other investigational interventions could be evaluated before definitive surgery by repeat dynamic MRI studies. Clearly, clinical investigations based on this measure of regional access must await confirmation of these preliminary findings in a larger series. A new prospective trial at this institution is designed to provide this information.
The authors thank David M. Parham, M.D., and Jesse J. Jenkins III, M.D., for the histologic analyses; Travis A. Miller, B.S.M.E., for his efforts in software development and data analysis; June S. Taylor, Ph.D., for her advice and encouragement throughout this process; and Christy Wright, E.L.S., for editorial consultation.