Health Technology Assessment (Apr 2025)
Evaluation of prognostic models to improve prediction of metastasis in patients following potentially curative treatment for primary colorectal cancer: the PROSPECT trial
Abstract
Background Despite apparently curative treatment, many patients with colorectal cancer develop subsequent metastatic disease. Current prognostic models are criticised because they are based on standard staging and omit novel biomarkers. Improved prognostication is an unmet need. Objectives To improve prognostication for colorectal cancer by developing a baseline multivariable model of standard clinicopathological predictors, and to then improve prediction via addition of promising novel imaging, genetic and immunohistochemical biomarkers. Design Prospective multicentre cohort. Setting Thirteen National Health Service hospitals. Participants Consecutive adult patients with colorectal cancer. Interventions Collection of prespecified standard clinicopathological variables and more novel imaging, genetic and immunohistochemical biomarkers, followed by 3-year follow-up to identify postoperative metastasis. Main outcome Best multivariable prognostic model including perfusion computed tomography compared with tumour/node staging. Secondary outcomes: Additive benefit of perfusion computed tomography and other biomarkers to best baseline model comprising standard clinicopathological predictors; measurement variability between local and central review; biological relationships between perfusion computed tomography and pathology variables. Results Between 2011 and 2016, 448 participants were recruited; 122 (27%) were withdrawn, leaving 326 (226 male, 100 female; mean ± standard deviation 66 ± 10.7 years); 183 (56%) had rectal cancer. Most cancers were locally advanced [≥ T3 stage, 227 (70%)]; 151 (46%) were node-positive (≥ N1 stage); 306 (94%) had surgery; 79 (24%) had neoadjuvant therapy. The resection margin was positive in 15 (5%); 93 (28%) had venous invasion; 125 (38%) had postoperative adjuvant chemotherapy; 81 (25%, 57 male) developed recurrent disease. Prediction of recurrent disease by the baseline clinicopathological time-to-event Weibull multivariable model (age, sex, tumour/node stage, tumour size and location, treatment, venous invasion) was superior to tumour/node staging: sensitivity: 0.57 (95% confidence interval 0.45 to 0.68), specificity 0.74 (95% confidence interval 0.68 to 0.79) versus sensitivity 0.56 (95% confidence interval 0.44 to 0.67), specificity 0.58 (95% confidence interval 0.51 to 0.64), respectively. Addition of perfusion computed tomography variables did not improve prediction significantly: c-statistic: 0.77 (95% confidence interval 0.71 to 0.83) versus 0.76 (95% confidence interval 0.70 to 0.82). Perfusion computed tomography parameters did not differ significantly between patients with and without recurrence (e.g. mean ± standard deviation blood flow of 60.3 ± 24.2 vs. 61.7 ± 34.2 ml/minute/100 ml). Furthermore, baseline model prediction was not improved significantly by the addition of any novel genetic or immunohistochemical biomarkers. We observed variation between local and central computed tomography measurements but neither improved model prediction significantly. We found no clear association between perfusion computed tomography variables and any immunohistochemical measurement or genetic expression. Limitations The number of patients developing metastasis was lower than expected from historical data. Our findings should not be overinterpreted. While the baseline model was superior to tumour/node staging, any clinical utility needs definition in daily practice. Conclusions A prognostic model of standard clinicopathological variables outperformed tumour/node staging, but novel biomarkers did not improve prediction significantly. Biomarkers that appear promising in small single-centre studies may contribute nothing substantial to prognostication when evaluated rigorously. Future work It would be desirable for other researchers to externally evaluate the baseline model. Trial registration This trial is registered as ISRCTN95037515. Funding This award was funded by the National Institute for Health and Care Research (NIHR) Health Technology Assessment programme (NIHR award ref: 09/22/49) and is published in full in Health Technology Assessment; Vol. 29, No. 8. See the NIHR Funding and Awards website for further award information. Plain language summary Bowel cancer is one the most common United Kingdom cancers and a leading cause of death. Despite apparently curative treatment, up to half of patients ultimately die from their disease because the tumour subsequently spreads around the body, known as ‘metastasis’. Patients are given chemotherapy upfront to prevent this spread, but predicting who will and will not develop metastasis is challenging, so it is difficult to know who to treat. Prediction is based on cancer ‘stage’, which describes how advanced the tumour is on imaging and under the microscope. A ‘multivariable prognostic model’ may improve prediction and is a combination of multiple factors known about the patient and their tumour that provides a score for the chance of future disease. However, multivariable models are not commonly used to predict recurrence for colorectal cancer and are criticised because they omit the latest ‘cutting-edge’ measurements (e.g. from scanning and genetic testing). To improve prediction of outcomes after bowel cancer, we performed a study in 13 National Health Service hospitals, where we collected both basic and more novel measurements from patients at the time of their diagnosis. We then followed patients for 3 years to determine who did and did not develop metastasis. From 2011 to 2016, we recruited 448 patients and used data from 326 to develop a multivariable model to predict metastasis. Our baseline model used a combination of basic factors, such as age, sex, tumour size and location, and treatment. This model predicted future disease significantly better than simple measurement of tumour stage. However, we found that the model did not improve when we added cutting-edge measurements. This suggests that these newer measurements are not useful to predict the chance of future disease. Our results suggest that researchers investigating prediction would be best served by concentrating on basic rather than more novel measurements. Scientific summary Background Colorectal cancer accounts for 12% of all new UK cancers, with over 42,000 new patients diagnosed each year. Despite treatment with curative intent, up to 50% of colorectal cancer patients will develop subsequent recurrent disease, normally metastasis. Chemotherapy aims to combat metastasis but identification of who will and will not develop subsequent metastasis (i.e. who does and does not merit chemotherapy) is difficult. Currently, ‘at-risk’ patients are identified by tumour and nodal (TN) staging from diagnosis and surgery (when performed) but more accurate prognostication remains an unmet need. Multivariable models promise to improve prediction by combining multiple weighted predictor factors measured from the patient in question but are not used widely. A frequent criticism is that such models ignore ‘cutting-edge’ promising biomarkers, which are currently the subject of intense research and which appear to offer an opportunity to improve risk stratification at diagnosis. Also, the move in recent years from offering chemotherapy in the postoperative (adjuvant) to preoperative (neoadjuvant) setting has shifted the need for identification of high-risk patients from the post-surgery setting (i.e. by using pathological samples from the resected specimen) to the preoperative setting (which depends on imaging and biopsy samples of the primary tumour). Objectives Our primary objective was to improve prediction of outcomes from colorectal cancer by developing a multivariable prognostic model of disease-free survival. We aimed to develop a best baseline model using standard clinicopathological variables and to then improve its prediction significantly by incorporating cutting-edge, novel imaging [perfusion computed tomography (CT)], immunohistochemical and genetic biomarkers. Our primary outcome was prediction of the baseline model incorporating CT perfusion when compared with standard TN staging. Secondary outcomes included baseline model prediction when incorporating immunohistochemical or genetic biomarkers; assessment of measurement variability between local sites and central review; and to investigate the biological relationships between perfusion CT and pathology variables. Methods We conducted a prospective multicentre cohort trial. Participants were recruited from 13 representative NHS teaching and district general hospitals in England and Scotland. Participants were eligible if they had histologically proven or suspected primary colorectal cancer (mass suspicious on endoscopy or imaging). Exclusions included polyp cancers, unequivocal metastases at staging, patients aged < 18 years, contraindications to intravenous contrast, pregnancy, and final diagnosis not being cancer. All participants gave written informed consent. Consecutive, unselected patients underwent perfusion CT in addition to standard staging CT. Standard investigations were interpreted locally by the usual clinical care team. Perfusion CT measurements were obtained locally by 26 radiologists. Central review was performed by three radiologists, who were blinded to all standard staging investigations. Treatment decisions were undertaken by the local multidisciplinary team as per usual practice. In patients undergoing surgery, central pathological review of the resected tumour was undertaken by two pathologists who performed additional pathological analysis, including immunohistochemistry for angiogenesis and hypoxia; microsatellite instability, mismatch repair (MMR), and somatic mutation analysis – Kirsten rat sarcoma viral oncogene homolog, neuroblastoma RAS viral oncogene homolog, v-raf murine sarcoma viral oncogene homolog B1 (BRAF). Participants were followed for 3 years and patients with recurrence were identified. A best baseline multivariable prognostic model was developed from prespecified standard clinical (age, sex, treatment) and pathological (tumour location, size, presence of venous invasion) variables. All model variables were prespecified, based on existing literature and expert opinion; that is, univariable significance in the study data set was not used to select any variable, either standard or novel. For our primary outcome, perfusion CT variables were added to the standard model as a composite (principal components) score. Prediction of this model was then compared with standard TN staging. The additive benefit (if any) of CT perfusion variables to the baseline model was calculated. For secondary outcomes, the additive benefit (if any) of immunohistochemical markers of angiogenesis, hypoxia and somatic mutation analysis was also calculated. We calculated the extent of any variation between local and central perfusion CT measurements. We calculated correlations between perfusion CT measurements and histopathological variables to determine biological significance. We estimated a sample size of 320 patients with 80 events (i.e. metastasis) would have 80% power to detect a 15% difference in correct risk classification by the model, allowing for loss to follow-up. We reported our results according to Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis guidelines. Results Between 2011 and 2016, we recruited 448 participants; 122 (27%) were withdrawn (mostly due to additional cancer), leaving 326 for analysis [226 male, 100 female; mean ± standard deviation (SD) 66 ± 10.7 years]; a total of 183 (56%) had rectal cancer. Most cancers were locally advanced [≥ T3 stage, 227 (70%); 151 (46%) were node-positive (≥ N1 stage)]. Surgery was performed in 306 (94%). The resection margin was positive in 15 (5%). Venous invasion was present in 93 (28%). Neoadjuvant therapy was undertaken in 79 (24%) and adjuvant therapy in 125 (38%) participants. Eighty-one (25%, 57 male) developed recurrent disease over the 3-year follow-up period. Perfusion CT measurements were available from local sites in 303 (93%) participants. Perfusion CT parameters did not differ between patients with and without positive local nodes (e.g. mean ± SD blood flow: 64.5 ± 25.2 vs. 75.0 ± 44.1 ml/minute/100 ml) or with and without recurrence (e.g. mean ± SD blood flow: 60.3 ± 24.2 vs. 61.7 ± 34.2 ml/minute/100 ml). Central review was undertaken in 291 (96%). Variability assessed by Bland–Altman plots was considerable between many local and central review perfusion CT measurements, most evident for permeability surface area product, where disagreement was greatest at higher permeability values. Although there were differences regarding where the region of interest was placed when local and central reviews were compared, this was not a major contributor to disagreement for vascular parameter values. Similarly, the individual CT scanner manufacturer did not impact substantially on disagreement, because all common manufacturers displayed large differences over all vascular parameters. There was no clear relationship between perfusion CT variables and immunohistochemical markers of angiogenesis (CD105, vascular endothelial growth factor) or hypoxia (hypoxia-inducible factor-1, glucose transporter-1) in the primary tumour, suggesting that CT does not reflect angiogenesis precisely. There was no difference between perfusion CT variables and MMR deficient/MMR proficient tumours. Prediction for the baseline clinicopathological model improved over standard TN staging due to significantly improved specificity: sensitivity 0.57 [95% confidence interval (CI) 0.45 to 0.68] and specificity 0.74 (95% CI 0.68 to 0.79) versus sensitivity 0.56 (95% CI 0.44 to 0.67) and specificity 0.58 (95% CI 0.51 to 0.64), respectively. The addition of perfusion CT variables to the baseline clinicopathological model did not improve prediction significantly: c-statistic 0.77 (95% CI 0.71 to 0.83) versus 0.76 (95% CI 0.70 to 0.82), respectively. Similarly, the addition of more novel histopathological variables (i.e. markers of angiogenesis, hypoxia, rat sarcoma virus, BRAF and MMR mutation status) to the baseline clinicopathological model did not improve model prediction significantly: c-statistic: 0.78 (95% CI 0.72 to 0.84) versus 0.76 (95% CI 0.70 to 0.82), respectively. Limitations The number of exclusions/withdrawals was higher than anticipated, mostly due to a higher prevalence of metastasis at baseline (possibly due to additional scans and multiple readers interpreting them). While prediction using our best baseline clinicopathologic model was significantly better than current practice, it may still be suboptimal for adoption in day-to-day clinical practice and its clinical utility needs assessment. External evaluation (validation) of the model in an NHS setting was not performed. The number of patients undergoing additional histopathological analysis was relatively small, as the study was not specifically powered to detect an effect for these variables, but if a beneficial effect on prediction exists, it is likely to be small. Conclusions We developed a prognostic model to predict development of metastatic disease following apparently curative treatment for colorectal cancer. The best baseline model comprising prospectively collected prespecified clinicopathological variables improved over standard TN staging prediction significantly. However, the addition of perfusion CT, immunohistochemical or genetic variables was not able to improve prediction significantly. Implications for health care In the NHS setting, applying a prognostic model comprising standard clinicopathological variables achieves significantly greater specificity for predicting subsequent metastasis than does current TN staging, without any diminished sensitivity. If similar prediction is sustained in external validation, application of this model in clinical practice may have immediate beneficial implications for the care of patients presenting with apparently localised colorectal cancer. Recommendations for future research Model prediction should be externally evaluated in an NHS setting, preferably by authors unrelated to model development. In addition to an external evaluation of its predictive accuracy, an evaluation should be made of the clinical utility to clinicians of the model in an NHS setting, including within neoadjuvant chemotherapy trials. Venous invasion on pathological evaluation was a strong prognostic factor within the standard model; further research into preoperative imaging assessment of venous invasion on CT for colon cancer and magnetic resonance imaging for rectal cancer is warranted. The fact that CT, immunohistochemistry and genetic markers of angiogenesis did not improve model prediction suggests that prior small, single-centre, retrospective studies including a benefit to these biomarkers are overoptimistic. This finding should be considered when contemplating funding future studies of such markers. Rather, our data suggest that future prognostic research should focus on standard clinicopathological variables. Changes to protocol Following interim presentation of trial data, the data monitoring committee increased recruitment from 370 to 448 patients, driven by a higher-than-expected baseline dropout rate due to metastasis at staging. The trial then continued recruitment until the original target of 80 evaluable participants with an event was achieved. Intended model analysis adjusted by clustering by study site was removed due to methodological advances in the interim, showing that this adjustment can cause statistical model instability. The proposed discrete choice study was not performed, so results for secondary outcome 7 were expressed as number of true-positive and false-positive patients, without a combined net benefit outcome (that would combine these metrics into a single measure). Trial registration This trial is registered as ISRCTN95037515. Funding This award was funded by the National Institute for Health and Care Research (NIHR) Health Technology Assessment programme (NIHR award ref: 09/22/49) and is published in full in Health Technology Assessment; Vol. 29, No. 8. See the NIHR Funding and Awards website for further award information.
Keywords
