Physician Underestimation of the Risk of Gastrointestinal Stromal Tumor Recurrence After Resection

Importance  Patients with resectable gastrointestinal stromal tumors (GISTs) might not receive the recommended duration of adjuvant therapy if their risk of recurrence is underestimated, which can have an impact on their recurrence-free survival (RFS).

Objective  To determine the extent of physician underestimation of risk of recurrence after complete primary GIST resection, the impact of underestimation on planned adjuvant treatment duration, and the association among high-risk patients of planned adjuvant treatment duration and RFS.

Design, Setting, and Participants  This was a retrospective observational medical record review reported by participating oncologists in 2013. US patients with complete primary GIST resection after 2010 were grouped as underestimated or not if their oncologists’ charted risk assessments were lower than assessments based on the Revised National Institutes of Health Consensus Criteria or not. Patients were followed by general community oncologists until death or the end of follow-up.

Main Outcomes and Measures  Fisher exact tests compared planned adjuvant treatment duration between groups. Cox proportional-hazards models estimated the impact of planned adjuvant treatment duration on RFS.

Results  A total of 109 oncologists reported information on 506 patients with GIST after primary resection (65.8% were high-risk and 8.7% were intermediate-risk). Physicians underestimated risk for 190 patients (37.5%); 30.1% of tumors with an intermediate-level mitotic count (6-10 per 50 high-powered fields) and an intermediate tumor size (6-10 cm) were correctly recognized as high-risk, as were 7.5% of nongastric tumors with an intermediate-level mitotic count and a tumor size of 2 to 5 cm. A smaller proportion of high-risk patients in the underestimated vs not-underestimated groups had at least 3 years of planned adjuvant therapy (36.1% vs 65.9%; P < .001). Planned adjuvant treatment of at least 3 years vs less than 3 years among high-risk patients conferred a lower hazard of recurrence and/or death (adjusted hazard ratio, 0.29; P < .001; 95% CI, 0.14-0.59).

Conclusions and Relevance  Overall, physicians tended to underestimate the risk of recurrence for many patients with GIST, especially for patients with tumors of intermediate size, intermediate-level mitotic count, and nongastric location, which had an impact on planned adjuvant therapy duration. Patients with at least 3 years of planned adjuvant treatment had longer RFS. Improved education on postresection risk assessment and risk reduction is needed.

Gastrointestinal stromal tumors (GISTs) are the most common primary mesenchymal tumors of the gastrointestinal tract, most often occurring in the stomach and small intestines.¹ Treatment varies by disease stage. In patients with primary resectable GISTs, tumor resection is the mainstay of treatment.¹ However, even after complete resection, recurrence is common, and more than 50% of patients at high risk of recurrence develop recurrent and/or metastatic disease within 5 years without adjuvant therapy.^2-6

The tyrosine kinase inhibitor imatinib, when used as adjuvant therapy following primary GIST resection, has been shown to help prevent the recurrence of GIST and improve recurrence-free survival (RFS) among patients at high risk of GIST recurrence. High-risk patients who were given 1 year of adjuvant imatinib in the American College of Surgeons Oncology Group (ACOSOG) 9000 trial⁷ had 1-, 3-, and 5-year RFS rates of 96%, 60%, and 40%, respectively. Furthermore, the Scandinavian Sarcoma Group (SSG) and the German Working Group on Medical Oncology (AIO) showed in the SSG XVIII/AIO trial⁸ that high-risk patients had higher 5-year RFS after receiving 36 months vs 12 months of adjuvant imatinib therapy (5-year RFS, 65.6% vs 47.9%; P < .001). Although there are no specific guidelines regarding the optimal duration of adjuvant treatment, the National Comprehensive Cancer Network (NCCN)⁹ recommends adjuvant imatinib for at least 36 months following GIST resection in patients with a high risk of recurrence.

Competing definitions of which patients are at high risk of GIST recurrence have been proposed. Most studies on risk assessment prior to 1998 were not specific to GISTs but presented general guidelines for stromal and/or smooth muscle neoplasms.¹⁰ In 2002, a consensus-based classification system from the National Institutes of Health (NIH) based the risk of GIST recurrence on tumor size and mitotic count.¹¹ Another set of criteria from the Armed Forces Institute of Pathology (AFIP)¹⁰ presented by Miettinen and Lasota¹² included tumor location as an additional tumor characteristic to consider in estimating recurrence risk. In 2008, a revision of the original NIH consensus classification system,⁸ now known as the Revised NIH Consensus Criteria,¹³ incorporated tumor location and presence of tumor rupture during surgery and are now widely used by many clinicians. Later, Gold et al¹⁴ proposed the Memorial Sloan Kettering Cancer Center (MSKCC) nomogram to calculate recurrence risk.

However, with multiple risk assessment tools available and no standardized approach, a recent study¹⁵ has reported considerable variation in how physicians assess the risk of recurrence; physicians have reported using one of the risk assessment tools above (most frequently, the Revised NIH Consensus Criteria¹³), using no tools at all, or considering other factors such as the presence of specific c-KIT mutations and comorbidities when assessing the risk profile of patients with GIST. Because of this variation, there is a need to better understand how physicians estimate the risk of GIST recurrence in clinical practice, and how their risk estimation affects treatment decisions and long-term outcomes.

The objectives of this study were (1) to evaluate how well physicians assess the risk of tumor recurrence; (2) to evaluate the impact of risk underestimation on planned adjuvant therapy duration; and (3) to evaluate the association between planned adjuvant therapy duration and RFS in high-risk patients with GIST after primary resection.

Box Section Ref ID

Of the 3 objectives of this study, the one requiring the largest sample size was the one to evaluate the association between planned adjuvant therapy duration and RFS among high-risk patients with GIST after primary resection. We estimated the sample size required to achieve 80% power using Cox proportional hazards regression with a 2-sided significance level of .05. First, we defined 2 groups of interest among high-risk patients based on the planned adjuvant treatment duration received: group 1, at least 3 years, and group 2, less than 3 years. We assumed that these 2 groups would have RFS similar to that of the patients in the 2 arms of the SSG XVIII/AIO trial⁸ who received 3 years vs 1 year of adjuvant imatinib. Based on the trial results, we assumed a hazard ratio (HR) of 0.46 in favor of group 1, and we assumed an event probability at 3 years (the maximum follow-up anticipated in our study) of 15% in group 1 and 40% in group 2. Under these assumptions, the study would need 105 in each group. We also assumed, similarly to the findings of Joensuu,¹³ that about 44% of patients enrolled in the study would have tumors that would qualify as high risk under the Revised NIH Consensus Criteria,¹³ so we estimated that 477 patients in all would be needed to enroll the required number of high-risk patients. Given these calculations, we targeted a total sample size of 500 patients. Sample size calculations were performed using the open-source powerSurvEpi R package, version 0.0.6.

A medical record review study was conducted to collect information on adult patients with primary resectable c-KIT-positive GIST using an online medical record-extraction tool administered to participating oncologists in the United States. More than 12 500 US physicians from the oncology community were invited by email to participate in the study, and all invited physicians had an equal opportunity to participate until the target number of 500 patient medical records had been collected. Each participating physician could provide deidentified information on a maximum of 10 eligible patients from their practice; eligible patients were selected in alphabetical order by last name. Data collected from the physicians did not include any patient-identifying information, and the study was exempted from full review by the New England institutional review board.

Patients were eligible if they (1) had received a diagnosis of primary resectable c-KIT (CD117) positive GIST; (2) had undergone tumor resection as primary treatment on or after January 1, 2010; (3) were at least 18 years of age at the time of the tumor resection; (4) had not received any tyrosine kinase inhibitory therapy (imatinib or sunitinib) prior to primary tumor resection; (5) had had no evidence of metastatic GIST prior to or at primary tumor resection and no evidence of residual macroscopic disease after primary tumor resection; and (6) had medical records accessible to their treating oncologist (participating in this study) with all of their GIST-related care for at least 6 months after primary tumor resection, or until death if they died within 6 months of primary tumor resection. In particular, patients whose medical records did not document the tumor size or mitotic count or did not document the recurrence risk as assessed by the physician at the time of resection were not eligible for the study.

First, physicians reported information on their practice, including duration, setting, size, and location. Second, for each selected patient, physicians reviewed each patient’s medical records to provide detailed patient demographic (including race, using categories defined by the US Office of Management and Budget,¹⁶ for adjustment in multivariate models as a potential confounder) and clinical information from the date of the primary GIST diagnosis to the date of the last follow-up recorded in the medical record. Specific tumor characteristics were collected, including location, size, mitotic count, and presence of tumor rupture during surgery. Information on planned duration of any adjuvant therapy was also collected. If a patient died or had disease recurrence, the dates of these events were also collected. Finally, for each patient, physicians provided their assessment of the patient’s risk of recurrence as noted in the medical record after primary resection.

To calculate the patient’s risk of recurrence based on the Revised NIH Consensus Criteria,¹³ which has been previously reported as the most commonly used risk assessment tool, information was collected on the patient’s tumor characteristics at the time of primary resection, including tumor size, mitotic count, tumor location, and presence of tumor rupture during surgery.¹³

The physician’s risk assessment as noted in the patient medical record after tumor resection was compared with the calculated risk assessment based on the Revised NIH Consensus Criteria.¹³ Then, each patient’s risk was classified as (1) underestimated if the physician’s risk assessment was lower than the calculated assessment, (2) consistent if they were the same, or (3) overestimated if the physician’s risk assessment was higher than the calculated assessment. The proportion of patients in each category was reported. The level of agreement between the charted risk assessment and the calculated risk based on the Revised NIH Consensus Criteria¹³ was assessed using weighted κ-coefficients, which can be used to classify the level of agreement as poor (<0.20), fair (0.21-0.40), moderate (0.41-0.60), good (0.61-0.80), and very good (0.81-1.00). The sign test assessed the difference between paired (charted and calculated) risk assessments. For the remainder of the analyses, patients were grouped as those whose risk was underestimated and those whose risk was not underestimated (consistent or overestimated).

Patients’ demographic and clinical characteristics were reported and compared between the 2 patient groups using Fisher exact tests for binary variables, χ² tests for categorical variables with 3 or more levels, and Wilcoxon rank-sum tests for continuous variables. The significance value was adjusted to account for these multiple tests using the Bonferroni correction. For categorical variables with 3 or more levels whose comparison attained statistical significance, the proportion of patients at each level from the 2 patient groups were compared using Fisher exact tests to offer further insight into the association.

To evaluate the impact of risk underestimation on the physicians’ treatment plans following primary GIST resection, each patient was categorized by the duration of adjuvant treatment planned: no adjuvant therapy, less than 1 year, 1 year, more than 1 year but less than 3 years, or 3 years or more. The proportion of patients in each category was compared between both groups using Fisher exact tests after stratifying the patients in both groups by the calculated risk assessment based on the Revised NIH Consensus Criteria.¹³

The impact of planned adjuvant treatment duration on RFS among the high-risk patients was analyzed using Kaplan-Meier survival analyses. The RFS rates were reported at 1 year, 2 years, and 3 years following primary GIST resection and compared using log rank tests between the high-risk patients who did or did not receive an initial treatment plan of at least 3 years of adjuvant therapy. A Cox proportional-hazards regression model assessed the impact of planned adjuvant treatment duration on RFS, while adjusting for potential confounding factors at the patient level (age, sex, race, and Charlson Comorbidity Index [CCI]) and physician level [number of years in practice, practice size, and use of specific risk assessment tools]). Less than 3% of patients had missing values for any of these covariates, including planned adjuvant treatment duration; these patients were dropped from the time-to-event analyses so that a complete case analysis could be conducted for the Kaplan-Meier analysis and the Cox regression modeling. The proportional hazards assumption was assessed using the methods of Lin et al.¹⁷

As a sensitivity analysis, the risk of recurrence of tumors with different characteristics was assessed under the Revised NIH Consensus Criteria,¹³ as well as the AFIP Criteria¹² and the MSKCC nomogram.¹⁴ In addition, the number of patients with each combination of tumor characteristics was reported, as well as the proportion scored as high risk by their physicians. Alternative high-risk categories were defined based on the AFIP criteria,¹² and separately based on the MSKCC nomogram¹⁴ (where high risk was defined as a likelihood of less than 50% to achieve a 5-year RFS). Using each system to define alternative high-risk categories, we regrouped the alternatively defined high-risk patients as underestimated or not underestimated.

The analyses of the impact of the physicians’ treatment plans on RFS were replicated among patients alternatively defined as high-risk using the alternative underestimation grouping. All analyses were conducted using SAS statistical software (version 9.3; SAS Institute Inc).

A total of 109 oncologists participated in this study. Most were experienced oncologists with 5 to 10 years (32%) or more than 10 years (56%) in practice (eTable in the Supplement). They were mainly in private practice (63%), with most in practice sizes of 2 to 9 oncologists (63%) or 10 or more (32%) from a diversity of geographic regions around the United States, mostly in urban settings (84%). The oncologists had treated a median of 30 patients with GIST over the preceding 10 years (interquartile range [IQR], 12-45 patients).

Patients who met the inclusion criteria (N = 506) were followed for a median of 15 months (IQR, 10-25 months) after primary GIST resection. The median age was 59 years, and slightly more than half were male (54.9%) and white (53.2%). At primary GIST resection, most patients (54.2%) had a CCI of 3 to 5. A GIST resection was performed a median of 12 days after diagnosis, most patients (78.5%) had no evidence of microscopic disease at the margins, and 42.3% of patients had a gastric GIST. Overall, 17.0% of patients had a tumor size of 2 cm or smaller, 31.4% had a tumor size of 3 to 5 cm, 35.0% had a tumor size of 6 to 10 cm, and 16.6% had a tumor size greater than 10 cm. Mitotic counts ranged from 5 or less per 50 high-power fields (HPFs) in 31.8% of patients, to 6 to 10 per 50 HPFs in 38.5% of patients, and more than 10 per 50 HPFs in 29.6% of patients. Most patients (81.8%) did not have tumor rupture during surgery, and tumor rupture status was unknown for 6.5% of patients (Table 1).

Based on the risk calculated using the Revised NIH Consensus Criteria,¹³ 65.8% were at high risk of recurrence, 8.7% were at intermediate risk, 10.5% were at low risk, and 15.0% were at very low risk (Table 2). Compared with this calculated risk, physicians’ assessments as noted in the patients’ medical records underestimated risk for 190 patients (37.5%), consistently estimated risk in 270 patients (53.4%) and overestimated risk in 46 patients (9.1%; sign test P < .001). Overall, the level of agreement between the physicians’ risk assessment and the calculated risk based on the Revised NIH Consensus Criteria¹³ was moderate (weighted κ = 0.53; Table 2).

Compared with patients whose risk of recurrence was not underestimated, patients with underestimated risk were of similar age, sex, and race/ethnicity. The 2 groups were similar in terms of CCI, time between primary GIST diagnosis and primary resection, and type of resection. Fewer patients with underestimated risk had a gastric tumor (30.0% vs 49.7%; P < .001), while more underestimated patients had a tumor size of 6 to 10 cm (46.8% vs 27.8%; P < .001) and a mitotic count of 6 to 10 per 50 HPFs (64.7% vs 22.8%; P < .001) compared with those whose risk was not underestimated. Both groups had similar proportions of patients with GIST rupture during surgery (Table 1).

The treatment plan for adjuvant therapy was categorized for each patient based on the planned duration. Table 3 summarizes the percentage of patients whose risk was underestimated and not underestimated who fall into each category of planned adjuvant treatment duration, stratified by the calculated risk of recurrence based on the Revised NIH Consensus Criteria.¹³ Among the low-risk patients, there was no statistically significant difference between the percentages of patients from each group for each category of planned adjuvant treatment duration. Among the intermediate-risk patients, more patients in the underestimated group had a treatment plan with no adjuvant therapy compared with those in the not-underestimated group (81.8% vs 27.3%; P = .003). Among high-risk patients, fewer patients in the underestimated group had a treatment plan of 3 or more years of adjuvant therapy (36.1% vs 65.9%; P < .001), while more patients had a treatment plan of 1 to 3 years of adjuvant therapy (20.7% vs 11.0%; P = .02) or no adjuvant therapy (21.3% vs 8.5%; P = .001) compared with those in the not-underestimated group.

There were 324 high-risk patients with complete data who were analyzed; the remaining 9 of 333 high-risk patients with missing data were excluded. Of these, 58 patients had recurrences or deaths observed, including 15 of the 167 patients (9.0%) who had received an initial treatment plan of at least 3 years of adjuvant therapy, and 43 of the 157 patients (27.4%) who had received an initial treatment plan of less than 3 years. High-risk patients who had a planned adjuvant treatment duration of less than 3 years had a significantly shorter RFS (log rank P < .001) with a median RFS of 2.6 years after the primary resection of GIST compared with high-risk patients who had a planned adjuvant treatment duration of at least 3 years with a median RFS that had not been reached by the end of the study period (Figure). At the end of 3 years of follow-up, RFS rates were 38.8% for patients with underestimated risk compared with 71.2% for patients whose risk was not underestimated. After adjusting for potential physician- and patient-level confounders, patients with planned adjuvant treatment duration of at least 3 years vs less than 3 years had a significantly lower hazard of recurrence and/or death (HR, 0.29; P < .001; 95% CI, 0.14-0.59).

For the sensitivity analyses, high risk was defined using the AFIP criteria¹² and using a calculated likelihood to achieve 5-year RFS of less than 50% under the MSKCC nomogram¹⁴ instead of using the Revised NIH Consensus Criteria.¹³ All 3 scoring systems demonstrated broad agreement; 441 patients had their risk consistently scored by all 3 systems (302 were high-risk and 139 not high-risk). Tumors from 15 patients were scored as high-risk by only 2 of the 3 systems, and tumors from 50 patients were scored as high-risk by only 1 of the 3 systems. Among tumors scored as high-risk by all 3 systems, only 52.6% were categorized as high risk by the responding physicians. In particular, for tumors with intermediate-level mitotic count (6-10 per 50 HPFs) and intermediate-level tumor size (6-10 cm), physicians correctly categorized 25 of 83 tumors (30.1%) as high-risk, and for nongastric tumors of intermediate-level mitotic count and tumor size of 2 to 5 cm, only 4 of 53 tumors (7.5%) were correctly categorized as high-risk (Table 4).

Using the alternative definitions of high risk under different scoring systems, the association between planned adjuvant treatment duration and RFS was replicated with nearly identical results (eFigures A-C in the Supplement).

Recommendations for the duration of adjuvant therapy after GIST resection vary based on the level of risk of disease recurrence; consequently, accurate assessments are essential to optimize disease management. Several tools have recently been developed to evaluate the risk of recurrence in patients with GIST, but without a standardized approach, a recent study¹⁵ has reported considerable variation in how physicians assess this risk. In this study,¹⁵ US oncologists provided information on their patients who underwent complete primary GIST resection, including their assessment of recurrence risk, treatment plan for adjuvant therapy, and clinical outcomes. This study¹⁵ estimated the extent of risk underestimation in clinical practice, identified tumor characteristics associated with underestimation, evaluated the impact of underestimation on the duration of adjuvant therapy planned by the physician, and evaluated the association of underestimation on RFS.

Participating physicians were recruited from a large panel of more than 12 500 US physicians from the oncology community. These oncology and hematology-oncology specialists were similar in terms of geographic distribution, practice size, and practice setting compared with physicians from the US oncology community in national surveys.^18,19 One notable difference is in experience with GIST, for participants reported treating a median of 30 patients in the previous 10 years, or about 3 patients a year. This volume is high enough to give these physicians more experience with patients with GIST than the average physician in the US oncology community, but perhaps not enough volume to accurately discern for every patient the key tumor characteristics that place a patient at high risk of GIST recurrence. Findings suggest a tendency among these physicians to underestimate patients’ risk of recurrence. For example, tumors with intermediate tumor size and intermediate-level mitotic count were classified as high risk by all 3 risk scoring systems^10,12,13 considered in our analysis, yet physicians frequently underestimated the risk, which suggests that borderline measures were more difficult to categorize. In addition, some physicians might not adequately take into account tumor location, which was incorporated into more recent risk assessment tools as larger data sets became available. In particular, under the Revised NIH Consensus Criteria,¹³ intermediate tumor size or intermediate-level mitotic count individually were updated from intermediate to high risk in nongastric tumors but remained intermediate risk in gastric tumors. These nuances might be underappreciated among physicians assessing recurrence risk.

This study also found that among high-risk patients, a greater proportion of patients with underestimated risk had no planned adjuvant therapy after tumor resection. In addition, among high-risk patients, a significantly larger proportion of those with underestimated risk were not given treatment plans for the recommended 3 years of adjuvant therapy, suggesting that more of these patients would have received adjuvant therapy if their high risk had been recognized. Even among the high-risk patients whose risk was not underestimated, physicians planned to give the recommended 3 years of adjuvant therapy to only about two-thirds of them. This may be partially explained by the fact that the randomized study⁸ that showed a survival benefit of 36 months vs 12 months of imatinib adjuvant therapy had only been published in 2012, whereas the study period of the current study had started for some patients in 2010 and 2011.

This study is subject to common limitations inherent in retrospective studies using data collected through patient medical chart reviews. First, generalizability could be limited by the size of the sample and the retrospective design of the study. Second, the data collection could have introduced possible errors and omissions in documentation. More specific to this study, we used the Revised NIH Consensus Criteria¹³ as the gold standard for assessing the risk of recurrence, but these criteria are themselves imperfect and could have led to possible misclassification bias. Nevertheless, replication of our analyses using the AFIP criteria and the MSKCC nomogram led to nearly identical results, and our findings suggest that more accurate risk assessment is essential to appropriately select adjuvant treatment duration and optimally treat patients with GIST. Third, planned adjuvant duration may differ from the actual treatment duration. Finally, patients’ data for this study were available for a median of 15 months after primary GIST resection. Further research would be warranted to confirm our findings over a longer period of time after primary GIST resection.

To our knowledge, this is the first study to assess the extent of risk underestimation and its impact on treatment and clinical outcomes. Findings suggest that oncologists tend to underestimate the risk of GIST recurrence after initial tumor resection. High-risk tumors of intermediate-level tumor size, intermediate-level mitotic count, and nongastric GIST location were frequently underestimated. High-risk patients were more likely to have a planned adjuvant therapy duration of less than 3 years after resection if their recurrence risk was underestimated compared with patients whose risk was not underestimated. Planned adjuvant therapy duration of less than 3 years was associated with reduced RFS. These findings suggest a need for improved education for physicians on risk assessment and risk reduction.

Corresponding Author: Annie Guérin, MSc, Analysis Group Inc, 1000 de la Gauchetière W, Ste 1200, Montreal, QC H3B 4W5, Canada (annie.guerin@analysisgroup.com).

Accepted for Publication: June 3, 2015.

Published Online: July 23, 2015. doi:10.1001/jamaoncol.2015.2407.

Author Contributions: Dr Macalalad had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Guérin, Sasane, Gauthier, Macalalad, Wu, Conley.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: All authors.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Guérin, Sasane, Gauthier, Macalalad, Wu.

Obtained funding: Sasane.

Administrative, technical, or material support: Guérin, Gauthier, Macalalad, Wu.

Study supervision: Guérin, Sasane, Keir, Gauthier, Conley.

Conflict of Interest Disclosures: Drs Sasane and Keir are employees of Novartis Pharmaceuticals Corp and hold Novartis stock. Mss Guérin and Gauthier and Drs Wu and Macalalad are employees of Analysis Group Inc, which has received consultancy fees from Novartis Pharmaceuticals. Dr Conley, an employee of the Department of Sarcoma Medical Oncology at the University of Texas MD Anderson Cancer Center, is an unpaid consultant for Novartis Pharmaceuticals Corp. No other disclosures are reported.

Funding/Support: This study was sponsored by Novartis Pharmaceuticals Corp.

Role of the Funder/Sponsor: The coauthors of this study who are also employees of Novartis (Dr Sasane and Dr Keir) were fully involved with all aspects of this research, including the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and the decision to submit the manuscript for publication.

Additional Contributions: We thank Qing Huang, PhD, and Jinlin Song, PhD, employees of Analysis Group Inc, who contributed to the analysis of the data; and Evangeline McDonald, MPH, an employee of Analysis Group Inc, and Jun Yan, PharmD, an employee of Jun Yan Medical Writing, who contributed to the preparation of this manuscript. Drs Huang and Song and Ms McDonald did not receive compensation for their assistance beyond that received in the normal of course of their employment. Dr Yan received compensation for her assistance.

Correction: The figure was corrected online September 11, 2015.