Abstract
Purpose
Assess the clinical utility and accuracy of routine surveillance head and neck magnetic resonance imaging (HN-MRI) for the detection of locoregional recurrence in patients with a history of oral cavity squamous cell carcinoma (OCSCC) without concurrent suspicious symptoms or signs 6 months or more after treatment.
Materials and methods
For OCSCC patients who underwent routine (defined as: without concurrent suspicious symptoms or signs) surveillance HN-MRI at 6 months or more after treatment completion, we retrospectively determined the detection rate of locoregional disease and false positive rate.
Results
Out of an original cohort of 533 OCSCC patients, 46 patients, who were disease-free 6 months after treatment, had undergone 108 routine HN-MRIs from 6 to 48 months after surgery without the presence of concurrent suspicious symptoms or signs and had 6 months of subsequent follow up. 1 out of 46 (2.2%) had a true positive regional recurrence. 10 out of 46 (21.7%) patients experienced a false positive locoregional finding.
Conclusions
Routine HN-MRI for locoregional surveillance of OCSCC, when used in patients without concurrent suspicious symptoms or exam findings over 6 months since treatment, may be unnecessary and costly given the very low rate of recurrence and high false positive rate. Our study supports the National Comprehensive Cancer Network guideline of limiting imaging after 6 months of primary treatment completion to patients with suspicious clinical findings. Nonetheless, managing physicians should continue to be empowered to use surveillance imaging based on risk profiles and unique circumstances for each patient.
1
Introduction
The utility of routine locoregional surveillance imaging for oral cavity squamous cell carcinoma (OCSCC) in patients without suspicious symptoms or signs of recurrence over 6 months since treatment is not well supported by specific evidence. In general, accurate early detection of locoregional disease recurrence or second primary carcinoma after primary treatment of head and neck squamous cell carcinoma (HNSCC) patients is potentially important in that salvage treatment is less successful once disease advances to late stages . However, as with all imaging studies, distinguishing true disease from false positives is important to prevent unnecessary psychological harm, morbidity from diagnostic and therapeutic procedures, and the costs of further investigations which includes the risks of exposure to radiation in the case of computed tomography (CT) scanning. Recent studies on HNSCC as a whole or on other subsites besides OCSCC have shown limited utility in routine surveillance imaging in identifying locoregional disease earlier and improving survival. In four recent studies, the use of routine or serial (non-suspicion-based) PET/CT for surveillance of HNSCC (as a whole) was found to have limited utility in identifying disease and improving survival . Because of these and other similar types of evidence, the National Comprehensive Cancer Network (NCCN) has issued guidelines that a baseline imaging study for future reference is recommended within the first 6 months after primary treatment and that subsequent imaging is based on a clinical indication .
To our knowledge, no studies exist evaluating the question of the utility of routine locoregional surveillance imaging for oral cavity squamous cell carcinoma (OCSCC) as an independent disease subsite entity. In OCSCC, as with much of HNSCC, recurrences appear most often within the first 24 months from treatment . An imaging study within the first 6 months of follow up could thus miss a proportion of recurrences that develop after 6 months from treatment, potentially justifying the use of routine periodic scans. However, recurrence in OCSCC is likely easier to detect on clinical examination than in other HNSCC sites, given the ease of visibility and accessibility to palpation of the anatomic region . Recurrences in the oral cavity are also more likely to develop at locoregional sites than at distant sites .
With these factors in mind, our study aimed to analyze the clinical utility and accuracy of routine surveillance head and neck-MRI (HN-MRI) for locoregional recurrence or second primary carcinoma at 6 months or more after primary treatment of OCSCC, in a cohort of patients without concurrent suspicious signs or symptoms . We analyzed the use of HN-MRI in particular because this has been the most frequently used imaging modality at our institution for periodic routine scanning. (With this study, we did not seek to analyze the value of any modes of surveillance of distant metastasis or lung second primary cancers.) As a secondary objective, we aimed to assess the survival outcomes of patients who met our inclusion criteria compared to the subsets of patients who did not have routine imaging or had scans that were provoked by suspicious symptoms or signs.
2
Materials and methods
Approval for this study was obtained from the Committee on Human Research (CHR), the institutional review board at the University of California at San Francisco (UCSF).
2.1
Patient cohort
Study patients were identified retrospectively from the UCSF cancer registry database using an ICD-9 code search, and included patients with OCSCC who were seen at UCSF from 2002 to 2012. Patients’ data were obtained through review of electronic medical records, including clinic visits, operation notes, imaging reports, pathology reports and death information.
Our inclusion criteria were as follows: (1) adult patients with a new diagnosis of OCSCC who underwent primary treatment (primary site resection/reconstruction, with or without neck dissection, with or without post-operative radiation, with or without chemotherapy) for curative intent; (2) patients who had one or more routine surveillance HN-MRIs at 6 months or more after the completion of treatment (either the surgery date if the patient had no adjuvant treatment following surgery or the last day of radiation if the patient had adjuvant treatment), with at least 6 more months of clinical follow up after imaging. For this study, a routine HN-MRI was defined as a HN-MRI study performed in absence of any suspicious symptoms or exam findings for a locoregional recurrence or a second primary carcinoma at a date within the month preceding the image acquisition date. A long-term routine HN-MRI was defined as any routine HN-MRI performed after 6 months from the completion date of treatment. No strict standardized long-term (over 6 months) imaging surveillance protocol exists or appears to have existed in the past across the head and neck oncologic physicians at our institution over the decade period involved in this study; thus, the timing of surveillance HN-MRI varied across practitioners over time.
Our exclusion criteria were as follows: (1) patients who had recurrence or distant metastasis within 6 months of primary treatment completion; (2) patients who had persistent locoregional disease at the end of treatment; (3) patients who had distant metastases at time of primary site treatment; (4) patients undergoing palliative surgery or other palliative treatments; (5) patients without any routine (not associated with suspicious symptoms or signs) surveillance scans; (6) patients without at least 6 months of subsequent follow up to the routine surveillance scan; (7) patients with surveillance scans other than HN-MRIs; (8) patients who underwent HN-MRIs ordered subsequent to the development of suspicious symptoms or signs; (9) subsequent HN-MRIs for patients following a confirmed positive scan.
2.2
Imaging
Each routine surveillance HN-MRI was classified into positive, negative or equivocal categories. A positive read was recorded when new, progressive or persistent areas of growth or abnormal signal intensity were seen in locoregional areas (defined as the original tumor bed, neck, and cervical lymph nodes). A negative read was recorded when no evidence of active disease or recurrence was seen. An equivocal read was recorded when no definitive evidence of recurrence or second primary was seen but could not be ruled out. We found that all equivocal HN-MRI scans in the study group were associated with subsequent investigations (biopsies and/or subsequent imaging studies); thus, for this study, equivocal scans were analyzed as positive scans since the read had provoked further management out of concern for recurrence. Distant metastases (apex of lungs) that were incidentally found on routine surveillance HN-MRI were reported separately; these cases were analyzed as true negative scans for locoregional disease otherwise.
HN-MRI findings were confirmed by at least one of the following: (1) histopathologic verification; (2) obvious progression of locoregional disease on subsequent imaging; (3) death due to locoregional progression; (4) at least 6 months of follow up without any suspicion for recurrence in case of negative findings.
We determined the survival outcomes (locoregional disease free survival (LRDFS), disease specific survival (DSS) and overall survival (OS)) of patients who met our inclusion criteria compared to the subset of patients who were excluded from the main analysis of our study because of lack of scans or because they had scans that were provoked by suspicious symptoms or signs.
2.3
Statistical analysis
To assess the clinical benefits of routine surveillance HN-MRI, the detection rate of locoregional disease for surveillance HN-MRIs was calculated. Drawbacks of surveillance HN-MRIs were assessed by determining false positive rate and estimating the costs of surveillance imaging. True positives (TP), true negatives (TN), false positives (FP) and false negatives (FN) were determined for routine surveillance HN-MRIs. Statistical measures of sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were then calculated both as per-patient and as per-scan. Exact binomial confidence intervals were then calculated. When calculating the confidence intervals for specificity and false positive rate in the per-scan analysis, population-averaged logistic regression models were used to account for the potential lack of independence.
The effect of having a false positive HN-MRI on the rate of surveillance scanning was determined by calculating the hazard ratio comparing the rate of scans per person year during the total surveillance time not following a false positive scan to the rate of scans per person year during the total surveillance time following a false positive scan. This was analyzed using maximum likelihood estimation for parametric exponential survival-time models. The risk of having a false positive finding on routine HN-MRI was analyzed using generalized estimating equations in binomial population-averaged models that included the following potential risk factors: being treated with post-operative radiation, being treated with post-operative chemotherapy, overall tumor stage (early: stages 1 and 2 or advanced: stages 3 and 4), tumor grade (well to moderate or poor), and the total number of routine surveillance HN-MRIs each patient had undergone. When analyzing the effect of post-operative chemotherapy on the risk of having a false positive routine HN-MRI, a likelihood ratio test was used to determine the statistical significance, as none of the patients who had a false positive routine HN-MRI was found to have been treated with post-operative chemotherapy.
When estimating the accuracy of routine HN-MRI by classifying scans rather than patients, dependence between the results of different scans from the same patient was addressed when calculating FP rate and specificity only. For other rates, there were no events or only one event, so there was no information about potential lack of independence to factor into their estimation.
For the survival analyses, LRDFS, DSS, and OS were estimated from the time of primary treatment completion using Kaplan–Meier survival estimates and analyzed using log-rank test. Two-tailed t-test and Fisher’s Exact test were used to determine the p-values for quantitative and categorical variables between patients who were included in the main analysis of our study and the subset of patients who were excluded from the main analysis of our study because they lacked surveillance HN-MRI performed over 6 months from treatment or had undergone HN-MRI provoked by suspicious symptoms or signs.
For all analysis, statistical significance was indicated at a p-value of less than 0.05. Statistical analysis was conducted using the Stata statistical software package (version 12; StataCorp, College Station, TX).
2
Materials and methods
Approval for this study was obtained from the Committee on Human Research (CHR), the institutional review board at the University of California at San Francisco (UCSF).
2.1
Patient cohort
Study patients were identified retrospectively from the UCSF cancer registry database using an ICD-9 code search, and included patients with OCSCC who were seen at UCSF from 2002 to 2012. Patients’ data were obtained through review of electronic medical records, including clinic visits, operation notes, imaging reports, pathology reports and death information.
Our inclusion criteria were as follows: (1) adult patients with a new diagnosis of OCSCC who underwent primary treatment (primary site resection/reconstruction, with or without neck dissection, with or without post-operative radiation, with or without chemotherapy) for curative intent; (2) patients who had one or more routine surveillance HN-MRIs at 6 months or more after the completion of treatment (either the surgery date if the patient had no adjuvant treatment following surgery or the last day of radiation if the patient had adjuvant treatment), with at least 6 more months of clinical follow up after imaging. For this study, a routine HN-MRI was defined as a HN-MRI study performed in absence of any suspicious symptoms or exam findings for a locoregional recurrence or a second primary carcinoma at a date within the month preceding the image acquisition date. A long-term routine HN-MRI was defined as any routine HN-MRI performed after 6 months from the completion date of treatment. No strict standardized long-term (over 6 months) imaging surveillance protocol exists or appears to have existed in the past across the head and neck oncologic physicians at our institution over the decade period involved in this study; thus, the timing of surveillance HN-MRI varied across practitioners over time.
Our exclusion criteria were as follows: (1) patients who had recurrence or distant metastasis within 6 months of primary treatment completion; (2) patients who had persistent locoregional disease at the end of treatment; (3) patients who had distant metastases at time of primary site treatment; (4) patients undergoing palliative surgery or other palliative treatments; (5) patients without any routine (not associated with suspicious symptoms or signs) surveillance scans; (6) patients without at least 6 months of subsequent follow up to the routine surveillance scan; (7) patients with surveillance scans other than HN-MRIs; (8) patients who underwent HN-MRIs ordered subsequent to the development of suspicious symptoms or signs; (9) subsequent HN-MRIs for patients following a confirmed positive scan.
2.2
Imaging
Each routine surveillance HN-MRI was classified into positive, negative or equivocal categories. A positive read was recorded when new, progressive or persistent areas of growth or abnormal signal intensity were seen in locoregional areas (defined as the original tumor bed, neck, and cervical lymph nodes). A negative read was recorded when no evidence of active disease or recurrence was seen. An equivocal read was recorded when no definitive evidence of recurrence or second primary was seen but could not be ruled out. We found that all equivocal HN-MRI scans in the study group were associated with subsequent investigations (biopsies and/or subsequent imaging studies); thus, for this study, equivocal scans were analyzed as positive scans since the read had provoked further management out of concern for recurrence. Distant metastases (apex of lungs) that were incidentally found on routine surveillance HN-MRI were reported separately; these cases were analyzed as true negative scans for locoregional disease otherwise.
HN-MRI findings were confirmed by at least one of the following: (1) histopathologic verification; (2) obvious progression of locoregional disease on subsequent imaging; (3) death due to locoregional progression; (4) at least 6 months of follow up without any suspicion for recurrence in case of negative findings.
We determined the survival outcomes (locoregional disease free survival (LRDFS), disease specific survival (DSS) and overall survival (OS)) of patients who met our inclusion criteria compared to the subset of patients who were excluded from the main analysis of our study because of lack of scans or because they had scans that were provoked by suspicious symptoms or signs.
2.3
Statistical analysis
To assess the clinical benefits of routine surveillance HN-MRI, the detection rate of locoregional disease for surveillance HN-MRIs was calculated. Drawbacks of surveillance HN-MRIs were assessed by determining false positive rate and estimating the costs of surveillance imaging. True positives (TP), true negatives (TN), false positives (FP) and false negatives (FN) were determined for routine surveillance HN-MRIs. Statistical measures of sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were then calculated both as per-patient and as per-scan. Exact binomial confidence intervals were then calculated. When calculating the confidence intervals for specificity and false positive rate in the per-scan analysis, population-averaged logistic regression models were used to account for the potential lack of independence.
The effect of having a false positive HN-MRI on the rate of surveillance scanning was determined by calculating the hazard ratio comparing the rate of scans per person year during the total surveillance time not following a false positive scan to the rate of scans per person year during the total surveillance time following a false positive scan. This was analyzed using maximum likelihood estimation for parametric exponential survival-time models. The risk of having a false positive finding on routine HN-MRI was analyzed using generalized estimating equations in binomial population-averaged models that included the following potential risk factors: being treated with post-operative radiation, being treated with post-operative chemotherapy, overall tumor stage (early: stages 1 and 2 or advanced: stages 3 and 4), tumor grade (well to moderate or poor), and the total number of routine surveillance HN-MRIs each patient had undergone. When analyzing the effect of post-operative chemotherapy on the risk of having a false positive routine HN-MRI, a likelihood ratio test was used to determine the statistical significance, as none of the patients who had a false positive routine HN-MRI was found to have been treated with post-operative chemotherapy.
When estimating the accuracy of routine HN-MRI by classifying scans rather than patients, dependence between the results of different scans from the same patient was addressed when calculating FP rate and specificity only. For other rates, there were no events or only one event, so there was no information about potential lack of independence to factor into their estimation.
For the survival analyses, LRDFS, DSS, and OS were estimated from the time of primary treatment completion using Kaplan–Meier survival estimates and analyzed using log-rank test. Two-tailed t-test and Fisher’s Exact test were used to determine the p-values for quantitative and categorical variables between patients who were included in the main analysis of our study and the subset of patients who were excluded from the main analysis of our study because they lacked surveillance HN-MRI performed over 6 months from treatment or had undergone HN-MRI provoked by suspicious symptoms or signs.
For all analysis, statistical significance was indicated at a p-value of less than 0.05. Statistical analysis was conducted using the Stata statistical software package (version 12; StataCorp, College Station, TX).
3
Results
A total of 46 patients met our study inclusion criteria out of an original 533 potential patient candidates with OCSCC who were treated surgically at UCSF from 2002 to 2012 ( Fig. 1 ). These 46 patients underwent a total of 108 routine HN-MRIs without the presence of concurrent suspicious symptoms or physical exam findings from 6 to 48 months after the completion of primary treatment. 9 out of the 108 routine HN-MRIs (for 6 patients) were performed and interpreted outside of our institution. 37 out of 46 (80.4%) patients had a negative baseline post-treatment HN-MRI within 6 months of the completion of curative treatment, while 9 out of 46 (19.6%) patients did not have a baseline imaging study within 6 months after treatment. The average follow up time was 49 months (range: 13 to 124 months) after the completion of primary treatment.