Predictors of one and two years' mortality in patients with colon cancer: A prospective cohort study

Abstract

Background

Tools to aid in the prognosis assessment of colon cancer patients in terms of risk of mortality are needed. Goals of this study are to develop and validate clinical prediction rules for 1- and 2-year mortality in these patients.

Methods

This is a prospective cohort study of patients diagnosed with colon cancer who underwent surgery at 22 hospitals. The main outcomes were mortality at 1 and 2 years after surgery. Background, clinical parameters, and diagnostic tests findings were evaluated as possible predictors. Multivariable multilevel logistic regression and survival models were used in the analyses to create the clinical prediction rules. Models developed in the derivation sample were validated in another sample of the study.

Results

American Society of Anesthesiologists Physical Status Classification System (ASA), Charlson comorbidity index (> = 4), age (>75 years), residual tumor (R2), TNM stage IV and log of lymph nodes ratio (> = -0.53) were predictors of 1-year mortality (C-index (95% CI): 0.865 (0.792–0.938)). Adjuvant chemotherapy was an additional predictor. Again ASA, Charlson Index (> = 4), age (>75 years), log of lymph nodes ratio (> = -0.53), TNM, and residual tumor were predictors of 2-year mortality (C-index:0.821 (0.766–0.876). Chemotherapy was also an additional predictor.

Conclusions

These clinical prediction rules show very good predictive abilities of one and two years survival and provide clinicians and patients with an easy and quick-to-use decision tool for use in the clinical decision process while the patient is still in the index admission.

About the Authors:

José M. Quintana

Roles Conceptualization, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Supervision, Visualization, Writing – original draft, Writing – review & editing

* E-mail: [email protected]

Affiliations Unidad de Investigación, Hospital Galdakao-Usansolo, Galdakao, Bizkaia, Spain, Red de Investigación en Servicios Sanitarios y Enfermedades Crónicas (REDISSEC), Galdakao, Bizkaia, Spain

ORCID logo http://orcid.org/0000-0003-2170-7876

Ane Antón-Ladislao

Roles Conceptualization, Formal analysis, Investigation, Validation, Visualization, Writing – original draft, Writing – review & editing

Affiliations Unidad de Investigación, Hospital Galdakao-Usansolo, Galdakao, Bizkaia, Spain, Red de Investigación en Servicios Sanitarios y Enfermedades Crónicas (REDISSEC), Galdakao, Bizkaia, Spain

Nerea González

Roles Conceptualization, Data curation, Investigation, Resources, Validation, Visualization, Writing – original draft, Writing – review & editing

Affiliations Unidad de Investigación, Hospital Galdakao-Usansolo, Galdakao, Bizkaia, Spain, Red de Investigación en Servicios Sanitarios y Enfermedades Crónicas (REDISSEC), Galdakao, Bizkaia, Spain

Santiago Lázaro

Roles Conceptualization, Formal analysis, Investigation, Supervision, Visualization, Writing – original draft, Writing – review & editing

Affiliations Red de Investigación en Servicios Sanitarios y Enfermedades Crónicas (REDISSEC), Galdakao, Bizkaia, Spain, Servicio de Cirugía General, Hospital Galdakao-Usansolo, Galdakao, Bizkaia, Spain

Marisa Baré

Roles Conceptualization, Data curation, Investigation, Resources, Visualization, Writing – original draft, Writing – review & editing

Affiliations Red de Investigación en Servicios Sanitarios y Enfermedades Crónicas (REDISSEC), Galdakao, Bizkaia, Spain, Unidad de Epidemiología Clínica, Corporació Parc Taulí, Sabadell, Spain

Nerea Fernández-de-Larrea

Roles Conceptualization, Data curation, Investigation, Resources, Visualization, Writing – original draft, Writing – review & editing

Affiliations Centro Nacional de Epidemiología, Instituto de Salud Carlos III, Madrid, Spain, CIBER Epidemiología y Salud Pública (CIBERESP), Madrid, Spain

Maximino Redondo

Roles Conceptualization, Data curation, Funding acquisition, Investigation, Methodology, Resources, Supervision, Visualization, Writing – original draft, Writing – review & editing

Affiliations Red de Investigación en Servicios Sanitarios y Enfermedades Crónicas (REDISSEC), Galdakao, Bizkaia, Spain, Unidad de Investigación, Hospital Costa del Sol, Málaga, Spain

Eduardo Briones

Roles Conceptualization, Data curation, Funding acquisition, Investigation, Methodology, Resources, Supervision, Visualization, Writing – original draft, Writing – review & editing

Affiliations CIBER Epidemiología y Salud Pública (CIBERESP), Madrid, Spain, UDG Salud Pública, Distrito AP Sevilla, Sevilla, Spain

Antonio Escobar

Roles Conceptualization, Funding acquisition, Investigation, Methodology, Supervision, Visualization, Writing – original draft, Writing – review & editing

Affiliations Red de Investigación en Servicios Sanitarios y Enfermedades Crónicas (REDISSEC), Galdakao, Bizkaia, Spain, Unidad de Investigación, Hospital Universitario Basurto, Bilbao, Bizkaia, Spain

Cristina Sarasqueta

Roles Conceptualization, Data curation, Funding acquisition, Investigation, Methodology, Resources, Supervision, Visualization, Writing – original draft

Affiliations Red de Investigación en Servicios Sanitarios y Enfermedades Crónicas (REDISSEC), Galdakao, Bizkaia, Spain, Unidad de Investigación, Hospital Universitario Donostia/Biodonostia, Donostia-San Sebastian, Gipuzkoa, Spain

Susana García-Gutierrez

Roles Conceptualization, Formal analysis, Investigation, Validation, Visualization, Writing – original draft, Writing – review & editing

Affiliations Unidad de Investigación, Hospital Galdakao-Usansolo, Galdakao, Bizkaia, Spain, Red de Investigación en Servicios Sanitarios y Enfermedades Crónicas (REDISSEC), Galdakao, Bizkaia, Spain

Inmaculada Aróstegui

Roles Conceptualization, Formal analysis, Investigation, Validation, Visualization, Writing – original draft, Writing – review & editing

Affiliations Red de Investigación en Servicios Sanitarios y Enfermedades Crónicas (REDISSEC), Galdakao, Bizkaia, Spain, Departamento de Matemática Aplicada, Estadística e Investigación Operativa, UPV/ EHU, BCAM-Basque Center for Applied Mathematics, Leioa, Bizkaia, Spain

for the REDISSEC-CARESS/CCR group

¶Membership of the REDISSEC-CARESS/CCR group is provided in the Acknowledgments.

Introduction

Colon cancer is one of the most frequently diagnosed cancers and is increasing in prevalence in some countries, partly owing to better screening and diagnosis strategies. Despite this increased prevalence, survival expectancy is improving, owing to earlier detection and better tools for treatment.[1,2] Most patients will undergo surgery and some will also receive chemotherapy. However, mortality remains high and, so, a main outcome of interest. Therefore, there is concern about the prognosis of these patients in terms of identifying predictors of mortality based on patient characteristics.[3] For this assessment, prediction models, decision rules, and risk scores are statistical tools intended to classify patients and guide clinicians in their everyday decision-making. These tools consist of a combination of multiple predictors, such as patient characteristics, diagnosis, and evolution, to estimate the probability of a certain outcome.[4,5] Different researchers have also developed various prediction rules for colon cancer patients, mainly to predict survival after surgery or very shortly afterwards.[6,7] Five-year survival prediction has been the outcome of some other studies.[8–10] There are fewer studies trying to predict survival at a longer period of time, such as 1–2 years.

The goal of this study was to develop clinical prediction rules for 1- and 2-year mortality in colon cancer patients who underwent surgery with the purpose of helping to personalize treatment and assist in the follow-up of these patients.

Methods

This prospective cohort study included patients drawn from 22 hospitals belonging to the Spanish National Health Service, which covers the majority of the Spanish population. All covered residents have free access to their primary care physician and the Emergency Department (ED) of the hospitals. All of the hospitals have similar technological and human resources.

A description of the study protocol is detailed elsewhere.[11] In summary, patients diagnosed with colon cancer presenting at any of these hospitals to undergo surgery between June 2010 and December 2012 were informed of the goals of the study and invited to participate. In order to take part in the study, a patient had to provide a signed informed consent. All information was kept confidential. The Institutional Review Board of the Basque Country approved this project.

Patients were eligible for the study if they were included in the surgical waiting list of one of the participating hospitals with a diagnosis of surgically resectable colon cancer. Colon cancer diagnosis was based on anatomopathological diagnosis after a biopsy by colonoscopy. Inclusion criteria were: 1) having been diagnosed with colon cancer(up to 15 cm above the anal margin), 2) initial application of curative and/or palliative surgery for treatment, and 3) signing the informed consent to participate in the study.[12] Exclusion criteria were: 1) colon carcinoma in situ, 2) unresectable tumor, 3) severe mental or physical conditions that precluded the patient from responding to questionnaires, 4) terminal disease, 5) inability to respond to questionnaires from any cause, and 6) did not give consent to participate in the study. We estimated the sample size needs of the study based on current recommendations for clinical prediction rules based on the expected rates of the outcome of interest (mortality at 1 or 2 years).[5,13]

Data collected upon hospital admission included information on sociodemographic factors, clinical data (including information about onset of symptoms, habits, personal and family background, comorbidities including those of the Charlson Comorbidity Index,[14] diagnostic tests, and pre-intervention treatments), preoperative data (including American Society of Anesthesiologists Physical Status Classification System (ASA) stage,[15] analytical data, tumor markers, and diagnostic tests), outpatient anesthesia data from the surgical intervention, pathological data, and data related to the remainder of the hospital stay (including the presence of complications, need for reoperation, or death). Subsequently, data were collected up to 30 days after surgery (analytical data, diagnostic tests, presence of complications, readmissions, reoperation, or death). Finally, patient information was collected through the first and second postoperative year, including radiation therapy, chemotherapy (treatment schedule, cycles, complications of treatment, and supportive care), laboratory results and diagnostic tests, presence of complications, tumor recurrence, readmission or reoperation, and death.

Outcome measures

The primary outcomes were mortality at 1 and 2 years after the patient was first admitted to the hospital (index admission). Vital status was established by reviewing medical records and examining the hospital database and public death registries. Deaths were considered confirmed if the name, sex, and date of birth on the record matched those of the participant.

Statistical analyses

The unit of analysis was a patient with a diagnosis of colon cancer who underwent operation in one of the participating hospitals. The sample was randomly divided in two subsamples (derivation and validation), each with half of the total population. Randomization of patients was performed automatically by a statistical program. To ease the interpretation of all the models, as well as, the punctuation of the scores derived from the models, all continuous variables were categorized. In the case of the age and Charlson comorbidity index, the optimal categorization of both was performed following the methodology developed by Barrios I et al [16]. In the case of Log Lymph ratio, categorization of the continuous variable was done based on the publication of Persiani R et al [17]. Descriptive statistics for both samples included frequencies and percentages for categorical data and means and standard deviations for continuous variables.

Univariable analysis was first performed in the derivation sample to identify risk factors related to mortality up to 1 and 2 years, using univariable Cox proportional-hazards regression models. Variables that were significant at a level of 0.20 were considered as potential independent variables for the multivariable analysis. Multivariable Cox proportional-hazards regression models were developed for mortality up to 1 and 2 years. Final predictive factors in multivariable analysis were those with a significance level of 0.05. Final models were also adjusted by the treating hospital to see if that affected the results and evaluated in the validation sample. As an internal validation of the models, 200 bootstrap samples with replacement were drawn from the original sample, and using for the selection of the variables stepwise method, 200 different models were developed in each sample. In order to compare the results of the 200 models with the model fitted in the original sample, the times that different predictors were predictors in the final models of each sample were summarized. Additionally, presence of collinearity was evaluated for some variables. Indeed, we developed two severity risk scores (for mortality up to 1 and 2 years) by assigning a weight to each risk factor category based on the β parameter from the multivariable Cox proportional-hazards regression. Then we added the weights of each of the risk factors presented by a patient with a higher score corresponding to a higher likelihood of mortality. Considering the optimal classifier points, three severity categories were created in each score.[16]

The predictive accuracy of each model was determined by the C-index. The predictive capacity of the risk groups derived from the models was also evaluated and validated in both samples by calculating the area under the receiver operating characteristic curve (AUC) and the calibration of the models was tested by Hosmer-Lemeshow test. All effects were considered significant at p < 0.05. All statistical analyses were performed using SAS 9.4 software (SAS Institute Inc., Cary, NC, USA).

Results

We initially contacted 2832 patients. Of these patients, 1979 fulfilled the selection criteria, with 1942 patients having full information at 1 year and 1817 at 2 years of follow-up as displayed in “Fig 1”.

[Figure omitted. See PDF.]

Fig 1. Flow-chart of the recruitment and follow-up process.

https://doi.org/10.1371/journal.pone.0199894.g001

Table 1 shows the descriptive of the derivation, validation and whole samples. Table 2 shows the different chemotherapy strategies employed in our patients. Univariable analysis in the derivation sample of the relationship of different sociodemographic and clinically relevant variables after 30 days from surgery with mortality at 1 (6.58% mortality rate) and 2 years (6.59% mortality rate) is displayed in Table 3.

[Figure omitted. See PDF.]

Table 1. Descriptive sociodemographic and clinical statistics of all samples of the study.

https://doi.org/10.1371/journal.pone.0199894.t001

[Figure omitted. See PDF.]

Table 2. Descriptive statistics of the adjuvant chemotherapy and molecular targeted agents employed in each sample.

https://doi.org/10.1371/journal.pone.0199894.t002

[Figure omitted. See PDF.]

Table 3. Univariable analysis in the derivation sample for mortality up to 1 and 2 years.

https://doi.org/10.1371/journal.pone.0199894.t003

Table 4 shows, based on the multivariable analysis, predictors of 1-year mortality were the ASA IV, a score ≥ 4 in the Charlson comorbidity index (CCI), age older than 75 years, TNM stage IV, lymph nodes ratio (>-0.53) and the residual tumor classification (R2 vs. R0-R1 categories). All the significant variables in the derivation sample were also significant in the validation sample. The C-index showed a good discrimination for both samples (0.865 (0.792–0.938) and 0.808 (0.734–0.882), in the derivation and validation samples, respectively). When including the treating hospital in the model, the C-index rose to 0.881 (0.807–0.955) in the derivation sample and 0.852 (0.776–0.928) in the validation sample. Since the use of adjuvant chemotherapy after 30 days of surgery has a role in survival, this variable was introduced as an interaction term with TNM stage. C-indexes were 0.885 (0.797–0.973) in the derivation and 0.848 (0.762–0.934) in the validation sample and, when including the treating hospital, the C-indexes were 0.914 (0.828–1.000) and 0.898 (0.810–0.986) in the derivation and validation samples, respectively.

[Figure omitted. See PDF.]

Table 4. Predictors of mortality at 1 year in patients with colon cancer.

Multivariate Cox regression analysis.

https://doi.org/10.1371/journal.pone.0199894.t004

Predictors of mortality during the first 2 years after surgery were the same as those that predicted mortality during the first year (Table 5). The C-index showed a good discrimination for both samples (0.808 (0.751–0.865) and 0.796 (0.739–0.853) in the derivation and validation samples, respectively). When including the treating hospital in the model, the C-index rose to 0.832 (0.775–0.888) in the derivation sample and 0.819 (0.762–0.876) in the validation sample. When chemotherapy was included in the model C-indexes were 0.808 (0.749–0.867) in the derivation and 0.817 (0.760–0.874) in the validation sample and, when including the treating hospital, the C-indexes were 0.836 (0.777–0.895) and to 0.834 (0.775–0.893), respectively.

[Figure omitted. See PDF.]

Table 5. Predictors of mortality at 2 year in patients with colon cancer.

Multivariate Cox regression analysis.

https://doi.org/10.1371/journal.pone.0199894.t005

The predictive accuracy of each continuous and categorical risk scores for 1-year mortality and 2-year mortality were measured by the AUCs which are displayed in Table 6. The AUC of the continuous risk score for 1-year mortality was 0.876 (0.828–0.924) in the derivation sample and 0.780 (0.715–0.845) in the validation sample, without considering adjuvant chemotherapy. However, including adjuvant chemotherapy, the AUCs of the continuous risk score were 0.882 (0.826–0.939) and 0.791 (0.718–0.864) in the derivation and validation samples, respectively. A risk score was also developed for 2-year mortality. The AUCs were 0.806 (0.760–0.851) in the derivation sample and 0.807 (0.758–0.856) in the validation sample, without considering adjuvant chemotherapy. Including adjuvant chemotherapy, the AUCs of the continuous risk score were 0.804 (0.758–0.850) and 0.805 (0.756–0.855) in the derivation and validation samples, respectively. From the previous risk score, two categorical risk scores, with three categories each, were developed for both mortality periods (Table 6). The three risk categories of each risk score, for 1- or 2-year mortality, differentiated among the categories. The categorical risk scores developed showed AUCs of 0.854 (0.796–0.912) in the derivation sample and 0.766 (0.692–0.841) in the validation sample for 1-year mortality. For 2-year mortality, the AUCs for the derivation and validation sample were, respectively, 0.776 (0.725–0.826) and 0.779 (0.729–0.829), both including adjuvant chemotherapy. However, without considering adjuvant chemotherapy, the categorical risk scores developed showed AUCs of 0.838 (0.793–0.883) and 0.768 (0.708–0.827) in the derivation and validation samples, respectively, for 1-year mortality. For 2-year mortality, the AUCs for the derivation and validation samples were, respectively, 0.770 (0.722–0.818) and 0.778 (0.727–0.829). All the results were estimated without considering the effect of the treating hospital. As additional information, internal validation of all the models were performed where it can be seen that those variables selected for our 4 models are the ones more often obtained by bootstrap (Table 7).

[Figure omitted. See PDF.]

Table 6. Categorical risk scores prediction of mortality at 1 or 2 years after surgery.

https://doi.org/10.1371/journal.pone.0199894.t006

[Figure omitted. See PDF.]

Table 7. Percentage of times each variable was selected by stepwise selection in internal validation using bootstrap.

https://doi.org/10.1371/journal.pone.0199894.t007

To better display the final results, trees charts derived from previous Tables 4 and 5 and the weights of the categories of the statistically significant variables and based on risk categories of Table 6 were developed. “Figs 2 and 3” presents the display for predicting mortality at one year with and without chemotherapy. “Figs 4 and 5” present similar trees but for two years mortality.

[Figure omitted. See PDF.]

Fig 2. Detail of risk factors associated to one-year mortality risk categories.

Risk factors include: ASA = IV, Charlson index ≥4, age>75 years, results of the surgery = R2 and Log lymph nodes ratio >-0.53.

https://doi.org/10.1371/journal.pone.0199894.g002

[Figure omitted. See PDF.]

Fig 3. Detail of risk factors associated to one-year mortality risk categories, chemotherapy included.

Risk factors include: ASA = IV, Charlson index ≥4, age>75 years, results of the surgery = R2 and Log lymph nodes ratio >-0.53.*Minor risk factor: Charlson index ≥4. †Moderate risk factors: ASA = IV, age>75 years and results of the surgery = R2. ⱡSevere risk factor: Log lymph nodes ratio >-0.53.

https://doi.org/10.1371/journal.pone.0199894.g003

[Figure omitted. See PDF.]

Fig 4. Detail of risk factors associated to two-years mortality risk categories.

Risk factors include: ASA = IV, Charlson index ≥4, age>75 years, results of the surgery = R2 and Log lymph nodes ratio >-0.53.*Minor risk factor: Charlson index ≥4 and ASA = IV. †Moderate risk factors: age>75 years, results of the surgery = R2 and Log lymph nodes ratio >-0.53.

https://doi.org/10.1371/journal.pone.0199894.g004

[Figure omitted. See PDF.]

Fig 5. Detail of risk factors associated to two-year mortality risk categories, chemotherapy included.

Risk factors include: ASA = IV, Charlson index ≥4, age>75 years, results of the surgery = R2 and Log lymph nodes ratio >-0.53.*Minor risk factor: Charlson index ≥4 and ASA = IV. †Moderate risk factors: age>75 years and Log lymph nodes ratio >-0.53. ⱡSevere risk factor: results of the surgery = R2. NA: Not applicable.

https://doi.org/10.1371/journal.pone.0199894.g005

Discussion

This prospective cohort study included a large sample of 1942 patients with colon cancer who underwent curative or palliative surgery. We identified factors before the intervention and up to 30 days and 1 year afterwards that related to mortality at 1 and 2 years in these patients. The categories of those factors were weighted to create continuous scores that, ultimately, were categorized into three categories of risk clinical prediction rules from minor to severe, based on the risk of dying at those points in time. Different prediction rules, though based in the same variables, were created for 1- and 2-year mortality, and, in each case, based on information available before surgery, during the index admission, and up to 30 days after surgery when, we took into account whether adjuvant chemotherapy was administered.[8–10, 18, 19]

We identified as predictors of medium-term mortality parameters such as ASA and comorbidities (based on the CCI). These are parameters related to the general condition of the patient before surgery, and are quite easily available for review. Similarly, older age can be related to higher frailty and need to be institutionalized or were more likely to undergo colon cancer resection during an unscheduled admission, which has been shown to be related to a higher rate of death in this patients. [20] We also identified TNM stage and lymph nodes ratio, which are related to the severity of the current disease, as a predictor of medium-term mortality. Finally, the residual tumor (R) classification, which is also related to the severity of the disease as well as reflects the result of surgery, had an influence on 1-year mortality. On the other hand, around 30 days after surgery is usually when, if it is indicated, the patient will begin chemotherapy for colon cancer. Adjuvant chemotherapy, as a specific treatment of the disease, has a clear relationship with mortality that changes depending on TNM stage. Therefore, a second set of prediction rules added chemotherapy treatment in conjunction with TNM stage to the previous parameters.

All of the predictors for 1-year mortality also appeared as predictors of 2-year mortality. These included those related to the general condition of the patient, such as age, ASA and CCI, but the main influence was from factors related to colon cancer and the patient’s evolution, including TNM stage. As with 1-year mortality, disease treatment, specifically adjuvant chemotherapy in relation to TNM stage, had a clear relationship with 2-year mortality.

Previous studies have found that disease severity, in this case, colon cancer stage as measured by TNM or lymph node status, or number of positive lymph nodes, or depth of primary tumor penetration are predictors of mortality.[21] Additionally, comorbidities, as measured by the Charlson comorbidity index in our case, jointly with age, have also been shown to affect mortality.[8, 20, 22, 23] Some studies have also implicated ASA level in mortality in colon cancer patients.[18, 24] Parameters from the evolution of disease in colon cancer patients, such as complications and reoperations, also were found to be predictive by some authors. Additionally, adjuvant chemotherapy was also a predictive factor in some studies, as in ours.[25] However, we first wanted to know which factors were predictors of mortality without considering chemotherapy. We introduced an interaction term to determine the different effects that chemotherapy had, depending on the TNM grade of the tumor. Additional variables, such as gender (a parameter that arose in some studies) and race, were not predictive in our analysis.[18, 26]

Most studies working on the identification of predictors of mortality in patients with colon cancer have focused on perioperative, in-hospital, or short-term mortality (up to 30 days post-operation). In this study, we explicitly removed this short-term mortality from our outcomes. We wanted to avoid the influence of surgeon ability, complications related to the surgery, and hospital health care quality in our prediction. However, as we have shown in our results, the variable of the treating hospital did have an influence on the 1- and 2-year mortality, and we have adjusted for it in our models. This means that the specific health care location has a role in explaining the causes of mortality, which indicates a possible problem of variability and equity among our centers.[26, 27]

From a practical point of view, our summary trees are developed to have a quick and simultaneous view of the prognosis of a patient at one and two years after surgery just at the time of the index admission-without chemotherapy- and, additionally, at 30 days afterwards, when chemotherapy is added. In general terms, mortality is low and remains similar at 1 an 2 years for TNM 0/I/II with no risk factors or just the Charlson comorbidity index ≥4. On the opposite side, is high for TNM IV or III with additional risk factors based on our rules. Those graphics can help clinicians as well as patients.

This study has several strengths. It was based on a large prospective cohort obtained from 22 hospitals, thus providing variability. The number of variables on the possible predictors of death was quite large. We have also included models with and without chemotherapy. The development of the predictive models followed the structure of current guidelines, such as those included in the TRIPOD statement [28], including an internal and external validation of all models. We identified predictive factors common to various other studies. However, the strength of our study lies in the fact that we have put them together, with just a few variables that are usually recorded and easily accessed, into single scores that may help clinicians easily and quickly classify the prognosis of these patients. As our scores have excellent predictive ability, clinicians will be better able to guide patients in follow-up. These scores may also serve as a reference to patients and families.

However, this study also has limitations. As in any prospective multicenter study with a 2-year follow-up, loss to follow-up was our main source of bias. Though reviewers in each center were trained to collect all of the planned information from all of the patients recruited at baseline, loss of patients and information was inevitable, though contained. We did study the influence of various HRQoL parameters in our outcomes, but, surprisingly, none of the questionnaires studied, even those detailing markers of frailty or health status, showed any predictive influence on mortality. Additionally, none of the tumor biomarkers studied, such as carcinoembryonic antigen levels, showed any relationship with mortality. Nevertheless, as new genetic and biological markers are discovered, they should be considered for addition to our models to increase predictive ability. Even further, improvements in the treatment of these patients, as the addition of new and more effective drugs, may change the predictive ability of our models though more than 50% of our patients did not receive chemotherapy and just those more severe received the new molecular targeted agents. We also did not do in-depth analysis on the influence of the quality, appropriateness, or existence of errors in the health care to the patients, or adherence to various treatments for the disease and comorbidities, when all of these factors could have a clear influence on mortality.

Finally, there are various statistical techniques currently used to develop clinical prediction rules. The ones we used in our study, namely, survival and logistic regression models, are classical but robust and clear for reading comprehension. Other techniques that have been employed in other studies, such as neural networks or a Bayesian belief network, may be alternatives, though some of them have a “black box” effect, wherein the design of the model is obscure to the reader. They are also usually no superior to the classical techniques.[10, 18, 29] As with any clinical prediction rule, proper external validation of our proposed models is the necessary next step, and, from a practical clinical point of view, studies are needed where the use of these rules improve patient health care.[30]

In conclusion, these clinical prediction rules provide clinicians and patients with decision tools that, with a minimal amount of easily obtainable information, have shown good predictive abilities and can be used in the clinical decision process. They provide information before starting the chemotherapy treatment process and are also valuable tools once the chemotherapy treatment has started in those patients where it has been prescribed since we present models for both situations. Lastly, future studies should examine the validity of our models in other settings, as well as apply them in real-life situations to determine their usefulness in the patient care and management process.

Acknowledgments

We are grateful for the support of the 22 participating hospitals, as well as the ED physicians, other clinicians, and staff members of the various services, research, quality units, and medical records sections of these hospitals. We also gratefully acknowledge the patients who participated in the study.

The REDISSEC CARESS-CCR (health services research network in colorectal cancer)- group:

Jose María Quintana López1, Marisa Baré Mañas2, Maximino Redondo Bautista3, Eduardo Briones Pérez de la Blanca4, Nerea Fernández de Larrea Baz5, Cristina Sarasqueta Eizaguirre6, Antonio Escobar Martínez7, Francisco Rivas Ruiz8, Maria M. Morales-Suárez-Varela9, Juan Antonio Blasco Amaro10, Isabel del Cura González11, Inmaculada Arostegui Madariaga12, Amaia Bilbao González7, Nerea González Hernández1, Susana García-Gutiérrez1, Iratxe Lafuente Guerrero1, Urko Aguirre Larracoechea1, Miren Orive Calzada1, Josune Martin Corral1, Ane Antón-Ladislao1, Núria Torà13, Marina Pont13, María Purificación Martínez del Prado14, Alberto Loizate Totorikaguena15, Ignacio Zabalza Estévez16, José Errasti Alustiza17, Antonio Z Gimeno García18, Santiago Lázaro Aramburu19, Mercè Comas Serrano20, Jose María Enríquez Navascues21, Carlos Placer Galán21, Amaia Perales22, Iñaki Urkidi Valmaña23, Jose María Erro Azkárate24, Enrique Cormenzana Lizarribar25, Adelaida Lacasta Muñoa26, Pep Piera Pibernat26, Elena Campano Cuevas27, Ana Isabel Sotelo Gómez28, Segundo Gómez-Abril29, F. Medina-Cano30, Julia Alcaide31, Arturo Del Rey-Moreno32, Manuel Jesús Alcántara33, Rafael Campo34, Alex Casalots35, Carles Pericay36, Maria José Gil37, Miquel Pera37, Pablo Collera38, Josep Alfons Espinàs39, Mercedes Martínez40, Mireia Espallargues41, Caridad Almazán42, Paula Dujovne Lindenbaum43, José María Fernández-Cebrián43, Rocío Anula Fernández44, Julio Ángel Mayol Martínez44, Ramón Cantero Cid45, Héctor Guadalajara Labajo46, María Heras Garceau46, Damián García Olmo46, Mariel Morey Montalvo47.

1Unidad de Investigación. Hospital Galdakao-Usansolo, Galdakao-Bizkaia / Red de Investigación en Servicios de Salud en Enfermedades Crónicas–REDISSEC.

2Epidemiologia Clínica y Cribado de Cancer, Corporació Sanitaria Parc Taulí, Sabadell / Red de Investigación en Servicios de Salud en Enfermedades Crónicas–REDISSEC.

3Servicio de Laboratorio. Hospital Costa del Sol, Málaga / REDISSEC.

4Unidad de Epidemiología. Distrito Sevilla, Servicio Andaluz de Salud.

5Centro Nacional de Epidemiología, Instituto de Salud Carlos III, Madrid / REDISSEC.

6Unidad de Investigación. Hospital Universitario Donostia /Instituto de Investigación Sanitaria Biodonostia, Donostia–REDISSEC.

7Unidad de Investigación. Hospital Universitario Basurto, Bilbao / REDISSEC.

8Servicio de Epidemiología. Hospital Costa del Sol, Málaga–REDISSEC.

9Department of Preventive Medicine and Public Health, Univesity of Valencia / CIBER de Epidemiología y Salud Pública (CIBERESP) / CSISP-FISABIO, Valencia.

10Unidad de Evaluación de Tecnologías Sanitarias, Agencia Laín Entralgo, Madrid.

11Unidad Apoyo a Docencia-Investigación. Dirección Técnica Docencia e Investigación. Gerencia Adjunta Planificación. Gerencia de Atención Primaria de la Consejería de Sanidad de la Comunidad Autónoma de Madrid.

12Departamento de Matemática Aplicada, Estadística e Investigación Operativa, UPV- REDISSEC.

13Epidemiologia Clínica y Cribado de Cancer, Corporació Sanitaria Parc Taulí, Sabadell / REDISSEC.

14Servicio de Oncología. Hospital Universitario Basurto, Bilbao.

15Servicio de Cirugía General. Hospital Universitario Basurto, Bilbao.

16Servicio de Anatomía Patológica. Hospital Galdakao-Usansolo, Galdakao.

17Servicio de Cirugía General. Hospital Universitario Araba, Vitoria-Gasteiz.

18Servicio de Gastroenterología. Hospital Universitario de Canarias, La Laguna.

19Servicio de Cirugía General. Hospital Galdakao-Usansolo, Galdakao.

20IMAS-Hospital del Mar, Barcelona.

21Servicio de Cirugía General y Digestiva. Hospital Universitario Donostia.

22Instituto de Investigación Sanitaria Biodonostia, Donostia.

23Servicio de Cirugía General y Digestiva. Hospital de Mendaro.

24Servicio de Cirugía General y Digestiva. Hospital de Zumárraga.

25Servicio de Cirugía General y Digestiva. Hospital del Bidasoa.

26Servicio de Oncología Médica. Hospital Universitario Donostia.

27Instituto de Biomedicina de Sevilla. Hospital Universitario Virgen del Rocío, Sevilla.

28Servicio de Cirugía. Hospital Universitario Virgen de Valme, Sevilla.

29Servicio de Cirugía General y Aparato Digestivo. Hospital Dr. Pesset, Valencia.

30Servicio de Cirugía General y Aparato Digestivo. Agencia Sanitaria Costa del Sol, Marbella.

31Servicio de Oncología Médica. Agencia Sanitaria Costa del Sol, Marbella.

32Servicio de Cirugía. Hospital de Antequera.

33Coloproctology Unit, General and Digestive Surgery Service, Corporació Sanitaria Parc Taulí, Sabadell.

34Digestive Diseases Department, Corporació Sanitaria Parc Taulí, Sabadell.

35Pathology Service, Corporació Sanitaria Parc Taulí, Sabadell.

36Medical Oncology Department, Corporació Sanitaria Parc Taulí, Sabadell / REDISSEC.

37General and Digestive Surgery Service, Parc de Salut Mar, Barcelona.

38General and Digestive Surgery Service, Althaia- Xarxa Assistencial Universitaria, Manresa.

39Catalonian Cancer Strategy Unit, Department of Health, Institut Català d’Oncología.

40Medical Oncology Department, Institut Català d’Oncología.

41Agency for Health Quality and Assessment of Catalonia–AquAS- and REDISSEC.

42Agency for Health Quality and Assessment of Catalonia–AQuAS-, CIBER de Epidemiología y Salud Pública–CIBERESP.

43Servicio de Cirugía General y del Aparato Digestivo, Hospital Universitario Fundación Alcorcón, Madrid.

44Servicio de Cirugía General y Aparato Digestivo, Hospital Universitario Clínico San Carlos, Madrid.

45Servicio Cirugía General y del Aparato Digestivo, Hospital Universitario Infanta Sofía, San Sebastián de los Reyes, Madrid.

46Servicio de Cirugía General y del Aparato Digestivo, Hospital Universitario La Paz, Madrid.

47REDISSEC. Unidad de Apoyo a la Investigación, Gerencia Asistencial de Atención Primaria de la Comunidad de Madrid, Madrid.

Citation: Quintana JM, Antón-Ladislao A, González N, Lázaro S, Baré M, Fernández-de-Larrea N, et al. (2018) Predictors of one and two years' mortality in patients with colon cancer: A prospective cohort study. PLoS ONE 13(6): e0199894. https://doi.org/10.1371/journal.pone.0199894

References

1. Ait Ouakrim D, Pizot C, Boniol M, Malvezzi M, Boniol M, Negri E, et al. Trends in colorectal cancer mortality in Europe: retrospective analysis of the WHO mortality database. BMJ (Clinical research ed). 2015;351:h4970.

2. Brenner H, Kloor M, Pox CP. Colorectal cancer. Lancet. 2014;383(9927):1490–502. pmid:24225001

3. Burke HB. Outcome prediction and the future of the TNM staging system. J Natl Cancer Inst. 2004;96(19):1408–9. pmid:15467022

4. Adams ST, Leveson SH. Clinical prediction rules. BMJ (Clinical research ed). 2012;344:d8312.

5. Laupacis A, Sekar N, Stiell IG. Clinical prediction rules. A review and suggested modifications of methodological standards. JAMA. 1997;277(6):488–94. pmid:9020274

6. Al-Homoud S, Purkayastha S, Aziz O, Smith JJ, Thompson MD, Darzi AW, et al. Evaluating operative risk in colorectal cancer surgery: ASA and POSSUM-based predictive models. Surg Oncol. 2004;13(2–3):83–92. pmid:15572090

7. Ren L, Upadhyay AM, Wang L, Li L, Lu J, Fu W. Mortality rate prediction by Physiological and Operative Severity Score for the Enumeration of Mortality and Morbidity (POSSUM), Portsmouth POSSUM and Colorectal POSSUM and the development of new scoring systems in Chinese colorectal cancer patients. Am J Surg. 2009;198(1):31–8. pmid:19246025

8. Jorgensen ML, Young JM, Dobbins TA, Solomon MJ. A mortality risk prediction model for older adults with lymph node-positive colon cancer. Eur J Cancer Care (Engl). 2015;24(2):179–88.

9. Jung M, Kim GW, Jung I, Ahn JB, Roh JK, Rha SY, et al. Application of the Western-based adjuvant online model to Korean colon cancer patients; a single institution experience. BMC Cancer. 2012;12:471. pmid:23061542

10. Snow PB, Kerr DJ, Brandt JM, Rodvold DM. Neural network and regression predictions of 5-year survival after colon carcinoma treatment. Cancer. 2001;91(8 Suppl):1673–8.

11. Quintana JM, Gonzalez N, Anton-Ladislao A, Redondo M, Bare M, Fernandez de LN, et al. Colorectal cancer health services research study protocol: the CCR-CARESS observational prospective cohort project. BMC Cancer. 2016;16:435. pmid:27391216

12. Hari DM, Leung AM, Lee JH, Sim MS, Vuong B, Chiu CG, et al. AJCC Cancer Staging Manual 7th edition criteria for colon cancer: do the complex modifications improve prognostic assessment? J Am Coll Surg. 2013;217(2):181–90. pmid:23768788

13. Vittinghoff E, McCulloch CE. Relaxing the rule of ten events per variable in logistic and Cox regression. Am J Epidemiol. 2007;165(6):710–8. pmid:17182981

14. Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis. 1987;40(5):373–83. pmid:3558716

15. Owens WD, Felts JA, Spitznagel EL Jr. ASA physical status classifications: a study of consistency of ratings. Anesthesiology. 1978;49(4):239–43. pmid:697077

16. Barrio I, Arostegui I, Rodriguez-Alvarez MX, Quintana JM. A new approach to categorising continuous variables in prediction models: Proposal and validation. Statistical methods in medical research. 2015.

17. Persiani R, Cananzi FC, Biondi A, Paliani G, Tufo A, Ferrara F, Vigorita V, D’Ugo D (2012) Log odds of positive lymph nodes in colon cancer: a meaningful ratio-based lymph node classification system. World J Surg. 2012 Mar. 36(3):667–74. pmid:22270984

18. Stojadinovic A, Bilchik A, Smith D, Eberhardt JS, Ward EB, Nissan A, et al. Clinical decision support and individualized prediction of survival in colon cancer: bayesian belief network model. Ann Surg Oncol. 2013;20(1):161–74. pmid:22899001

19. van Westreenen HL, Ijpma FF, Wevers KP, Afzali H, Patijn GA. Reoperation after colorectal surgery is an independent predictor of the 1-year mortality rate. Diseases of the colon and rectum. 2011;54(11):1438–42. pmid:21979191

20. Aquina CT, Mohile SG, Tejani MA, Becerra AZ, Xu Z, Hensley BJ, Arsalani-Zadeh R, Boscoe FP, Schymura MJ, Noyes K, Monson JR, Fleming FJ. The impact of age on complications, survival, and cause of death following colon cancer surgery. Br J Cancer. 2017 Jan;116(3):389–397. pmid:28056465

21. Fielding LP, Phillips RK, Fry JS, Hittinger R. Prediction of outcome after curative resection for large bowel cancer. Lancet. 1986;2(8512):904–7. pmid:2876336

22. Ouellette JR, Small DG, Termuhlen PM. Evaluation of Charlson-Age Comorbidity Index as predictor of morbidity and mortality in patients with colorectal carcinoma. J Gastrointest Surg. 2004;8(8):1061–7. pmid:15585394

23. Wu CC, Hsu TW, Chang CM, Yu CH, Lee CC. Age-adjusted Charlson comorbidity index scores as predictor of survival in colorectal cancer patients who underwent surgical resection and chemoradiation. Medicine (Baltimore). 2015;94(2):e431.

24. Manilich EA, Kiran RP, Radivoyevitch T, Lavery I, Fazio VW, Remzi FH. A novel data-driven prognostic model for staging of colorectal cancer. J Am Coll Surg. 2011;213(5):579–88, 88. pmid:21925905

25. Renfro LA, Grothey A, Xue Y, Saltz LB, Andre T, Twelves C, et al. ACCENT-based web calculators to predict recurrence and overall survival in stage III colon cancer. J Natl Cancer Inst. 2014;106(12).

26. Healy MA, Grenda TR, Suwanabol PA, Yin H, Ghaferi AA, Birkmeyer JD, et al. Colon cancer operations at high- and low-mortality hospitals. Surgery. 2016;160(2):359–65. pmid:27316824

27. Simunovic M, Rempel E, Theriault ME, Coates A, Whelan T, Holowaty E, et al. Influence of hospital characteristics on operative death and survival of patients after major cancer surgery in Ontario. Canadian journal of surgery Journal canadien de chirurgie. 2006;49(4):251–8. pmid:16948883

28. Moons KG, Altman DG, Reitsma JB, Ioannidis JP, Macaskill P, Steyerberg EW, et al. Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): explanation and elaboration. Ann Intern Med. 2015;162(1):W1–73. pmid:25560730

29. Steele SR, Bilchik A, Johnson EK, Nissan A, Peoples GE, Eberhardt JS, et al. Time-dependent estimates of recurrence and survival in colon cancer: clinical decision support system tool development for adjuvant therapy and oncological outcome assessment. Am Surg. 2014;80(5):441–53. pmid:24887722

30. Steyerberg EW, Moons KGM, van der Windt DA, Hayden JA, Perel P, Schroter S, et al. Prognosis Research Strategy (PROGRESS) 3: Prognostic Model Research. PLoS Medicine. 2013;10(2).