Abstract
Introduction
Treatment of metastatic colon cancer may be driven as much by practice patterns as by features of disease. To optimize management, there is a need to better understand what is determining primary site resection use.
Methods
We evaluated all patients with Stage IV cancers in the National Cancer Data Base between 2002 and 2012 (50,791 patients, 1,230 hospitals). We first identified patient characteristics associated with primary tumor resection. Then, we assessed nationwide variation in hospital resection rates.
Results
Overall, 27,387 (53.9%) patients underwent primary site resection. Factors associated with resection included younger age, having <2 major comorbidities and white race (P<0.001). Nationwide, hospital adjusted primary tumor resection rates ranged from 26.0% to 87.8% with broad differences across geographical areas and hospital accreditation types.
Conclusions
There is statistically significant variation in hospital rates of primary site resection. This demonstrates inconsistent adherence to guidelines in the presence of conflicting evidence regarding resection benefit.
Keywords: Colon neoplasms, colon surgery, neoplasm metastasis, hospitals, clinical decision-making
Synopsis
Conflicting evidence to direct surgical decision-making in metastatic colon cancer result in widely varying treatment patterns. We find that the decision to undergo resection of the primary tumor site is influenced not only by patient-specific clinical factors, but also by hospital type and geography.
Introduction
Of the 137,000 new cases of colon cancer diagnosed per year in the United States, approximately 20% present with metastatic disease.(1) Compared to other stages of this disease, these patients have substantially worse long-term survival and fewer well-established practice guidelines to direct treatment choices. Historically, primary site resection was common in the setting of metastatic disease, with 74.5% of cases undergoing primary tumor resection in 1988.(2) However, with the use of newer systemic therapies, primary site resection can often be avoided.(3)
Despite these more recent developments, more than two-thirds of patients with metastatic disease continue to undergo primary site resection.(4) Primary site resection in metastatic colon cancers is morbid and its oncologic benefit remains an area of significant controversy, but can be lifesaving in cases of bleeding, perforation or obstruction.(5, 6) The extent to which hospitals’ primary site resection rates reflect patient disease characteristics or their culture and habits is also unclear. Other treatment patterns such as chemotherapy, radiation and metastectomy for metastatic colon cancer are highly variable across the country and largely influenced by hospital practices rather than patient disease.(7) Similarly, the same trends may apply specifically to primary site resection in addition to patient and tumor-specific factors such as bleeding or obstruction. The current national landscape of primary site resection practices has not been well described.
In the present study, we evaluated patient and hospital characteristics associated with primary site resection in advanced colon cancers. We hypothesized that patients who undergo primary site resection would have more favorable comorbidity profiles and socio-demographic characteristics. We also hypothesized that regional and hospital-level variation in resection practices would be present but have a weaker influence than patient disease factors.
Material and methods
Data Source and Study Population
The National Cancer Data Base (NCDB) Participant User File (PUF) is a joint project of the American College of Surgeons Commission on Cancer (CoC) and the American Cancer Society. Data represent nearly 70% of all newly diagnosed patients with cancer from greater than 1,500 CoC-accredited hospitals in the United States. Data are collected using standardized coding schemata and have been described previously.(8, 9) The NCDB records both clinical and pathologic stage. We sought to identify patients who were known to be Stage IV prior to any surgical intervention, and thus we chose to use clinical stage for our cohort selection. All patients ≥ 18 years were identified with a clinical diagnosis of Stage IV cancer of the colon, rectosigmoid, and rectum from 1998 to 2011 using the International Classification of Diseases for Oncology 3rd Edition code for adenocarcinoma (8140) and primary site codes for the colon (C180; C182–189). Excluded patients included those without ultimate adenocarcinoma histology, those with more than one primary malignancy, patients with unknown primary tumor sites, and those who did not have complete treatment status information, follow-up or metastatic disease assessment (eAppendix Figure 1). We also excluded patients from very low volume hospitals (<10 patients/hospital) from our analysis, as we felt that these hospitals did not have sufficient data for reliable hospital-level analyses.
Statistical Analysis
We first identified clinical and demographic characteristics associated with primary tumor resection. Initially tested covariates included age, sex, race, comorbidities as defined by the Charlson-Deyo comorbidity index (10), patient distance to treating hospital, details regarding median income, high school diploma status, rurality and urban influence, primary tumor site, and whether the patient underwent distant site surgery, any (neoadjuvant or adjuvant) radiation therapy or any chemotherapy. Hospital factors included geographic region of facility as well as facility type (Community vs. Academic/Research vs. Other). We also included quartile of hospital volume, our closest available proxy to hospital size, as a covariate in our analyses. First, we compared characteristics of patients using χ2 tests for categorical variables and Student’s t-tests for continuous variables. Then, we performed multiple bivariate regressions using primary site resection as a dependent variable and each of the above covariates as independent variables. Each independent variable that was a significant predictor of primary site resection with P<0.1 was retained in a multivariable model. All factors except for sex were significant predictors and were retained in the model. Our multivariable logistic regression model was created using robust standard errors, to account for hospital-level clustering.
There are several variables tracked by the NCDB, which warrant further description. A brief explanation of these follows and more details may be found on the NCDB data dictionary website.(11) Median household income for the year 2000 and the percentage of people in a given zip code without a high school diploma were both estimated matching patient ZIP codes against 2000 U.S. Census Bureau data. Similar later data regarding median household income and percent with no high school diploma for 2008–2012 were tracked using American Community Survey data for 2008–2012 from the U.S. Census Bureau. Rurality and urban influence are tracked by matching the state/county Federal Information Processing Standard (FIPS) code for a patient against 2013 files from the U.S. Department of Agriculture (USDA) Economic Research Service. Geographic hospital location was based on U.S. Census Division of Reporting facility data. Facility type was based on classification of American College of Surgeons Commission on Cancer Accreditation programs.
Following initial regression analyses, patient-level data was collapsed to the hospital level and reliability adjusted outcomes were then calculated. Reliability adjustment is useful to account for hospitals with small numbers of cases whose crude rates of resection may be skewed due to statistical “noise.”(12–14) Facilities were then ranked by risk- and reliability-adjusted probability of resection and divided into quintiles of probability of resection, with the first quintile including hospitals with the least likely probability of performing primary site resection and the fifth quintile including those hospitals with the highest average probability of performing primary site resections. All statistical analyses were conducted using Stata version 13 (StataCorp LP, College Station, TX). All tests are 2-sided with significance set at a P value of less than 0.05.
Results
Patient Characteristics and Comorbidities
There were 50,791 patients from 1,230 hospitals eligible for analysis, of whom 27,387 patients (53.9%) underwent primary site resection. As outlined in Table 1, patients undergoing primary site resection were younger and more likely to be white. There was a similar distribution of patient sex between groups. Patients undergoing primary site resection were less likely to have Charlson-Deyo comorbidity index ≥2 and slightly more likely to have an index of zero or one.
Table 1.
Characteristics of patients and hospitals in the study are shown, with stratification into those undergoing primary site resection and those not undergoing primary site resection.
| Whole cohort (n=50,791) |
Primary site resection (n=27,387) |
No primary site resection (n=23,404) |
P Value | |
|---|---|---|---|---|
| Age | <0.001 | |||
| <=40 | 2,125 (4.2) | 1,299 (4.7) | 826 (3.5) | |
| 41 – 60 | 17,264 (34.0) | 10,150 (37.1) | 7,114 (30.4) | |
| 61 – 80 | 23,567 (46.4) | 12,526 (45.7) | 11,041 (47.2) | |
| >80 | 7,835 (15.4) | 3,412 (12.5) | 4,423 (18.9) | |
| Sex | 0.354 | |||
| Female | 24,983 (49.2) | 13,419 (49.0) | 11,564 (49.4) | |
| Male | 25,808 (50.8) | 13,968 (51.0) | 11,840 (50.6) | |
| Race | <0.001 | |||
| White | 39,549 (77.9) | 21,818 (79.7) | 17,731 (75.8) | |
| Black | 8,806 (17.3) | 4,223 (15.4) | 4,583 (19.6) | |
| Other | 2,436 (4.8) | 1,346 (4.9) | 1,090 (4.7) | |
| Charlson-Deyo index | <0.001 | |||
| 0 | 37,517 (73.9) | 20,520 (74.9) | 16,997 (72.6) | |
| 1 | 9,748 (19.2) | 5,276 (19.3) | 4,472 (19.1) | |
| ≥2 | 3,526 (6.9) | 1,591 (5.8) | 1,935 (8.3) | |
| Received any chemotherapy | 30,854 (60.8) | 18,283 (66.8) | 12,571 (53.7) | <0.001 |
| Received any radiation therapy | 2,256 (4.4) | 935 (3.4) | 1,321 (5.6) | <0.001 |
| Distant site resection | 6,408 (12.6) | 5,701 (20.8) | 707 (3.0) | <0.001 |
| Median distance to facility, miles (IQR) | 7.8 (15.0) | 8.3 (16.4) | 7.2 (13.4) | <0.001 |
| % no HS diploma ’00** | <0.001 | |||
| 29% or more | 9,795 (19.3) | 4,941 (18.0) | 4,854 (20.7) | |
| 20% – 28.9% | 11,928 (23.5) | 6,420 (23.4) | 5,508 (23.5) | |
| 14% – 19.9% | 11,130 (21.9) | 6,086 (22.2) | 5,044 (21.6) | |
| Less than 14% | 15,549 (30.6) | 8,629 (31.5) | 6,920 (29.6) | |
| Unknown | 2,389 (4.7) | 1,311 (4.8) | 1,078 (4.6) | |
| % no HS diploma ’08-‘12*** | <0.001 | |||
| 21% or more | 10,066 (19.8) | 5,093 (18.6) | 4,973 (21.3) | |
| 13% – 20.9% | 13,848 (27.3) | 7,511 (27.4) | 6,337 (27.1) | |
| 7% – 12.9% | 15,279 (30.1) | 8,307 (30.3) | 6,972 (29.8) | |
| Less than 7% | 10,101 (19.9) | 5,695 (20.8) | 4,406 (18.8) | |
| Unknown | 1,497 (3.0) | 781 (2.9) | 716 (3.1) | |
| Rurality/Urban influence ‘03**** | <0.001 | |||
| Metro | 40,401 (79.5) | 21,510 (78.5) | 18,891 (80.7) | |
| Urban | 7,172 (14.1) | 4,110 (15.0) | 3,062 (13.1) | |
| Rural | 921 (1.8) | 527 (1.9) | 394 (1.7) | |
| Unknown | 2,297 (4.5) | 1,240 (4.5) | 1,057 (4.5) | |
| Rurality/Urban influence ‘13**** | <0.001 | |||
| Metro | 40,963 (80.7) | 21,790 (79.6) | 19,173 (81.9) | |
| Urban | 6,649 (13.1) | 3,844 (14.0) | 2,805 (12.0) | |
| Rural | 881 (1.7) | 512 (1.9) | 369 (1.6) | |
| Unknown | 2,298 (4.5) | 1,241 (4.5) | 1,057 (4.5) | |
| Median household income ‘00** | <0.001 | |||
| Less than $30,000 | 7,886 (15.5) | 4,007 (14.6) | 3,879 (16.6) | |
| $30,000 – $34,999 | 9,127 (18.0) | 4,959 (18.1) | 4,168 (17.8) | |
| $35,000 – $45,999 | 13,370 (26.3) | 7,123 (26.0) | 6,247 (26.7) | |
| $46,000 + | 18,024 (35.5) | 9,991 (36.5) | 8,033 (34.3) | |
| Unknown | 2,384 (4.7) | 1,307 (4.8) | 1,077 (4.6) | |
| Median household income ’08-‘12*** | <0.001 | |||
| Less than $38,000 | 10,547 (20.8) | 5,379 (19.6) | 5,158 (22.0) | |
| $38,000–$47,999 | 12,040 (23.7) | 6,571 (24.0) | 5,469 (23.4) | |
| $48,000–$62,999 | 12,537 (24.7) | 6,740 (24.6) | 5,797 (24.8) | |
| $63,000+ | 14,150 (27.9) | 7,899 (28.8) | 6,251 (26.7) | |
| Unknown | 1,527 (3.0) | 798 (2.9) | 729 (3.1) | |
| Geographic Location***** | <0.001 | |||
| Atlantic | 19,104 (37.6) | 9,910 (36.2) | 9,194 (39.3) | |
| Northeast | 2,657 (5.2) | 1,299 (4.7) | 1,358 (5.8) | |
| Great Lakes/Midwest | 13,349 (26.3) | 7,186 (26.2) | 6,163 (26.3) | |
| Mountain/West/Pacific | 7,513 (14.8) | 4,225 (15.4) | 3,288 (14.1) | |
| South/Southeast | 8,168 (16.1) | 4,767 (17.4) | 3,401 (14.5) | |
| Facility Type****** | <0.001 | |||
| Community/Comprehensive Community | 34,216 (67.4) | 19,160 (70.0) | 15,056 (64.3) | |
| Academic/Research | 16,526 (32.5) | 8,207 (30.0) | 8,319 (35.6) | |
| Other Cancer Program | 49 (0.1) | 20 (0.1) | 29 (0.1) | |
| Hospital volume, cases | 0.001 | |||
| <= 32 | 12,852 (25.3) | 7,116 (26.0) | 5,736 (24.5) | |
| 33 – 51 | 12,604 (24.8) | 6,774 (24.7) | 5,830 (24.9) | |
| 52 – 80 | 13,082 (25.8) | 7,008 (25.6) | 6,074 (26.0) | |
| >80 | 12,253 (24.1) | 6,489 (23.7) | 5,764 (24.6) |
SD=standard deviation; HS=high school;
Calculated using Pearson’s X2 test for categorical variables, two-tailed Student’s t test for continuous variables and a two-sample Wilcoxon rank-sum (Mann-Whitney) test when medians are reported for continuous variables.
Estimated by matching ZIP code of patient against 2000 US Census data.
American Community Survey data for 2008–2012.
Estimated by matching state/county Federal Information Processing Standard (FIPS) code of patient against 2013 files from U.S. Department of Agriculture (USDA) Economic Research Service (http://www.ers.usda.gov/data-products/rural-urban-continuum-codes).
US Census Division of reporting facility.
Classification by the Commission on Cancer Accreditation program.
Treatment Characteristics
Patients who underwent primary tumor site resection were more likely to receive either neoadjuvant or adjuvant chemotherapy and less likely to receive either neoadjuvant or adjuvant radiation therapy. Patients undergoing primary site resection were also far more likely to undergo distant site resection.
Socio-demographic Characteristics
Patients undergoing primary site resection were more likely to live in communities with higher high school graduation rates and slightly higher median household income on average than those not undergoing primary site resection. Median distance to treating facility was statistically greater (8.3 vs. 7.2 miles; P<0.001) for those undergoing primary site resection.
Univariate and Multivariable Analyses
Results of univariate and multivariable analyses are displayed in Table 2. Controlling for other patient and facility factors, significant patient-level factors associated with not undergoing resection included age 61–80 (OR:0.84; 95% CI: 0.76 – 0.93; P=0.001) or >80 (OR: 0.70; 95% CI: 0.63 – 0.79; P<0.001) black race (OR:0.85; 95% CI: 0.80–0.90; P<0.001), receipt of any radiation (OR:0.52; 95% CI: 0.47–0.57; P<0.001), and Charlson-Deyo comorbidity score ≥ 2 (OR:0.84; 95% CI: 0.78–0.91; P<0.001]). Associations were not statistically significant for high school graduation rates, rural or urban influence, median household income and hospital volume in the multivariable analysis.
Table 2.
Predictors of primary site resection in the whole cohort are displayed. Results represent odds of undergoing primary site resection with 95% confidence intervals in parentheses.
| Factors | Univariate Analysis | Multivariable Analysis |
|---|---|---|
| Age | ||
| <=40 | Reference | Reference |
| 41 – 60 | 0.91 (0.83 – 1.00) | 0.96 (0.86 – 1.06) |
| 61 – 80 | 0.72 (0.66 – 0.79) | 0.84 (0.76 – 0.93) |
| >80 | 0.49 (0.44 – 0.54) | 0.70 (0.63 – 0.79) |
| Sex | ||
| Male | Reference | Not retained in model |
| Female | 0.98 (0.95 – 1.02) | |
| Race | ||
| White | Reference | Reference |
| Black | 0.75 (0.71 – 0.78) | 0.85 (0.80 – 0.90) |
| Other | 1.00 (0.92 – 1.09) | 1.03 (0.94 – 1.13) |
| Charlson-Deyo index | ||
| 0 | Reference | Reference |
| 1 | 0.98 (0.93 – 1.02) | 1.06 (1.01 – 1.11) |
| ≥ 2 | 0.68 (0.64 – 0.73) | 0.84 (0.78 – 0.91) |
| Received any chemotherapy | 1.86 (1.79 – 1.93) | 1.63 (1.56 – 1.70) |
| Received any radiation***** | 0.59 (0.55 – 0.65) | 0.52 (0.47 – 0.57) |
| Distant site resection | 9.21 (8.50 – 9.98) | 9.31 (8.56 – 10.13) |
| Distance to facility | 1.00 (1.00 – 1.00) | 1.00 (1.00 – 1.00) |
| % no HS diploma ’00** | ||
| 29% or more | Reference | Reference |
| 20% – 28.9% | 1.15 (1.09 – 1.21) | 1.06 (0.98 – 1.15) |
| 14% – 19.9% | 1.19 (1.12 – 1.25) | 1.08 (0.97 – 1.19) |
| Less than 14% | 1.23 (1.16 – 1.29) | 1.10 (0.98 – 1.23) |
| % no HS diploma ’08-‘12*** | ||
| 21% or more | Reference | Reference |
| 13% – 20.9% | 1.16 (1.10 – 1.22) | 1.08 (1.00 – 1.16) |
| 7% – 12.9% | 1.16 (1.11 – 1.22) | 1.04 (0.94 – 1.15) |
| Less than 7% | 1.26 (1.19 – 1.33) | 1.08 (0.96 – 1.21) |
| Rurality/Urban influence ‘03**** | ||
| Metro | Reference | Reference |
| Urban | 1.18 (1.12 – 1.24) | 0.94 (0.82 – 1.08) |
| Rural | 1.17 (1.03 – 1.34) | 0.89 (0.67 – 1.19) |
| Rurality/Urban influence ‘13**** | ||
| Metro | Reference | Reference |
| Urban | 1.21 (1.14 – 1.27) | 1.19 (1.03 – 1.37) |
| Rural | 1.22 (1.07 – 1.40) | 1.24 (0.93 – 1.67) |
| Median household income ‘00*** | ||
| Less than $30,000 | Reference | Reference |
| $30,000 – $34,999 | 1.15 (1.08 – 1.22) | 1.08 (1.00 – 1.17) |
| $35,000 – $45,999 | 1.10 (1.04 – 1.17) | 1.06 (0.96 – 1.16) |
| $46,000 + | 1.20 (1.14 – 1.27) | 1.14 (1.01 – 1.28) |
| Median household income ’08-‘12** | ||
| Less than $38,000 | Reference | Reference |
| $38,000–$47,999 | 1.15 (1.09 – 1.21) | 1.01 (0.94 – 1.09) |
| $48,000–$62,999 | 1.11 (1.06 – 1.17) | 0.96 (0.87 – 1.05) |
| $63,000+ | 1.21 (1.15 – 1.27) | 0.96 (0.86 – 1.08) |
| Hospital volume, cases | ||
| <= 32 | Reference | Reference |
| 33 – 51 | 0.94 (0.89 – 0.98) | 0.97 (0.92 – 1.03) |
| 52 – 80 | 0.93 (0.89 – 0.98) | 0.99 (0.94 – 1.05) |
| >80 | 0.91 (0.86 – 0.95) | 1.03 (0.97 – 1.10) |
HS=high school;
Estimated by matching ZIP code of patient against 2000 US Census data.
American Community Survey data for 2008–2012.
Estimated by matching state/county Federal Information Processing Standard (FIPS) code of patient against 2013 files from U.S. Department of Agriculture (USDA) Economic Research Service (http://www.ers.usda.gov/data-products/rural-urban-continuum-codes).
Interaction terms, odds ratios should not be interpreted separately.
Following calculation of risk- and reliability-adjusted hospital-level resection rates, quintiles of hospital resection probability were created with the first quintile representing hospitals with the least likely probability of performing primary site resection and the fifth quintile representing the highest probability, and this is portrayed in Figure 1. The mean for each quintile ranged from 42.7% (95% CI: 35.0% – 50.4%) to 64.4% (95% CI: 57.2% – 71.6%). Significant hospital-level variation in use of primary site resection was present, with adjusted rates at the hospital level ranging from 26.0% (95% CI: 18.5% – 35.2%) to 87.8% (95% CI: 83.0% – 91.4%) for individual hospitals.
Figure 1.
Hospital-level variation in risk- and reliability-adjusted primary site resection rates is displayed.
Variation in patient odds of resection by facility type, after controlling for patient factors and geography, is shown in Figure 2. Compared to patients at academic/research facilities patients at community or comprehensive community cancer programs were statistically significantly more likely to undergo primary site resection (OR: 1.50; 95% CI: 1.42 – 1.57; P<0.001). Patients at other designated cancer programs were not statistically any more or less likely than those at academic/research facilities to undergo resection (OR: 0.74; 95% CI: 0.40 – 1.36).
Figure 2.
Adjusted odds of primary site resection by facility type. Results are from model controlling for age, sex, race, comorbidities as defined by the Charlson-Deyo comorbidity index(10), patient distance to treating hospital, details regarding median income, high school diploma status, rurality and urban influence, primary tumor site, whether the patient underwent distant site surgery, any radiation therapy (including interaction between rectal tumor site and radiation receipt) or any chemotherapy, facility geographic location, and hospital volume.
Variation in patient odds of resection by facility geographic location, after controlling for patient factors and facility type, is shown in Table 3. Compared to patients in the Atlantic United States, patients in the Northeast were statistically significantly less likely to undergo primary site resection (OR: 0.87; 95% CI: 0.79 – 0.96; P<0.005). Patients in the Great Lakes/Midwest were equally as likely as Atlantic patients to get resection (OR: 1.04; 95% CI: 0.99 – 1.09; P=0.160). Patients in the Mountain/West/Pacific region were more likely to undergo primary site resection (OR 1.20; 95% CI: 1.13 – 1.28; P<0.001). Patients in the South/Southeast were the most likely to undergo primary site resection (OR 1.39; 95% CI: 1.31 – 1.48; P<0.001).
Table 3.
Adjusted odds of primary site resection by geographic location. Results are from model controlling for age, sex, race, comorbidities as defined by the Charlson-Deyo comorbidity index(10), patient distance to treating hospital, details regarding median income, high school diploma status, rurality and urban influence, primary tumor site, whether the patient underwent distant site surgery, any radiation therapy (including interaction between rectal tumor site and radiation receipt) or any chemotherapy, facility type, and hospital volume.
| Region* | Adjusted Odds Ratio (95% Confidence Interval) | P |
|---|---|---|
| Atlantic | Reference | |
| Northeast | 0.87 (0.79 – 0.96) | <0.005 |
| Great Lakes/Midwest | 1.04 (0.99 – 1.09) | 0.160 |
| Mountain/West/Pacific | 1.20 (1.13 – 1.28) | <0.001 |
| South/Southeast | 1.39 (1.31 – 1.48) | <0.001 |
Based on US Census Division of reporting facility
Discussion
We have shown that 1) statistically significant variation exists in hospital-based utilization of primary site resection in advanced colon cancers; 2) drivers of this appear to include facility type and geographic location, with more patients at community hospitals and at hospitals in the South/Southeastern United States undergoing resection. 3) patient predictors of primary site resection include favorable comorbidity status and white race. This study demonstrates broad overall variation in practice patterns in primary site resection for advanced colon cancers.
This large nationally representative cohort provides a picture of recent practice patterns. The variation that we found most likely reflects the lack of consistent data regarding benefits of primary site resection in Stage IV patients. Current NCCN guidelines are to avoid primary site resection in asymptomatic patients, unless the goal is for cure. However, this is in contrast to past practices in which resection was nearly always part of first-line therapy. Furthermore, these guidelines are in the presence of mixed data regarding efficacy of primary site resection. Work in other cancers has shown that adherence to guidelines depends largely on the quality of supporting data and the availability of institutional resources.(15) As these advanced colon cancer guidelines are in the presence of mixed data, our finding of correspondingly mixed adherence is consistent with this pattern. But this variation is perhaps not clinically warranted.
Our results show substantial variation in practice patterns that is not supported in one clear direction by the current body of literature. One previous similar study of factors associated with resection found that at the patient level, gender, geographic region, insurance status, tumor location, grade and CEA were independent predictors of tumor resection.(16) This study used the Surveillance, Epidemiology and End Results (SEER) Database, which captures approximately 28% of all cancer cases in the U.S. The outcome in their study should be reliable because of the fairly granular data that SEER provides over a large number of years. Our study used the NCDB, which captures over 70% of incident cancer cases in the U.S., a broader nationally representative sample, and our methodology accounts for an array of patient, hospital and regional covariates that are captured in the NCDB, allowing for the control of many potential confounders. Our findings were similar to the aforementioned study with regard to geographic variation, but we did not find differences based on gender. Our study also shows variation based on facility type, with more patients at community hospitals undergoing resection.
There are many possible explanations for our findings, specifically regarding variation in practice by region and facility type. First, the findings could be reflective of patient access to surgical resection. It does appear that patients undergoing resection traveled farther, and perhaps only patients with the means to seek out complex surgical care at tertiary treatment centers could have done this. Second, we found that patients at community cancer centers were more likely to undergo primary site resection. This might speak to differences in operative management patterns but should not be limited to patient factors, as many were adjusted for in our models. Perhaps patients treated at academic centers did not present for appropriate follow-up or lacked the resources to absorb the personal financial burden of complex cancer surgery, as previous work has shown that cancer patients may substantially alter their care in order to decrease out-of-pocket expenses.(17) Specifically, in patients with colorectal cancer, patients with higher personal financial burdens were found to have higher rates of complications from surgery, which led to more financial stress, with a potential impact on adherence to treatment recommendations.(18) These are all possibilities which could be assessed at academic and community hospitals using a qualitative or mixed methods approach, to better understand these barriers to resection.
Our study has several limitations. First, this is an observational study using registry-based data. Inherent in this is a lack of granularity as to factors that may guide decisions beyond those that were measured in our data. It is not possible to know if these findings represent overuse of resection in asymptomatic patients, or appropriate use in symptomatic patients (e.g. obstructing or bleeding lesions). However, the large cohort provides a broad and representative view of overarching trends that should be reliable. Given that over 27,000 patients (53.9% of all patients with metastatic/stage IV disease) in our cohort underwent primary resection, there were certainly many patients that would fall into each of these categories. Additionally, we were able to control for some markers of patient socioeconomic status and rural locations, both of which might be expected to increase symptomatic presentations. Second, this study relies only on data collected from accredited American College of Surgeons Commission on Cancer (CoC) hospitals. While this means that not every hospital in the U.S. is accounted for, over 1,500 accredited cancer care programs report data to the NCDB.(8) Finally, there is the likelihood that certain time-varying covariates existed over the ten years of our cohort. We chose to treat all covariates equally over time. Trends in resection have been previously studied,(2) and grouping our entire cohort together provided the best statistical power to detect drivers of variation.
We have shown that there is substantial variation in practice patterns for primary site resection in a large nationally representative study of patients with advanced colon cancers in the United States. This study highlights the need for better long-term data to guide decision-making. Additionally, it underscores a likely deviation from current practice guidelines that appears to vary by hospital and be driven by geography and facility type.
Supplementary Material
Acknowledgments
Funding: Dr. Healy is supported by NIH T32CA009672-24.
Footnotes
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Conflict of Interest Disclosures: The authors have no conflicts to report.
References
- 1.Siegel R, Desantis C, Jemal A. Colorectal cancer statistics, 2014. CA: a cancer journal for clinicians. 2014 Mar-Apr;64(2):104–117. doi: 10.3322/caac.21220. PubMed PMID: 24639052. Epub 2014/03/19. eng. [DOI] [PubMed] [Google Scholar]
- 2.Hu CY, Bailey CE, You YN, et al. Time trend analysis of primary tumor resection for stage IV colorectal cancer: less surgery, improved survival. JAMA Surg. 2015 Mar 1;150(3):245–251. doi: 10.1001/jamasurg.2014.2253. PubMed PMID: 25588105. Epub 2015/01/15. eng. [DOI] [PubMed] [Google Scholar]
- 3.McCahill LE, Yothers G, Sharif S, et al. Primary mFOLFOX6 plus bevacizumab without resection of the primary tumor for patients presenting with surgically unresectable metastatic colon cancer and an intact asymptomatic colon cancer: definitive analysis of NSABP trial C-10. Journal of clinical oncology : official journal of the American Society of Clinical Oncology. 2012 Sep 10;30(26):3223–3228. doi: 10.1200/JCO.2012.42.4044. PubMed PMID: 22869888. Pubmed Central PMCID: PMC3434981. Epub 2012/08/08. eng. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Cook AD, Single R, McCahill LE. Surgical resection of primary tumors in patients who present with stage IV colorectal cancer: an analysis of surveillance, epidemiology, and end results data, 1988 to 2000. Annals of surgical oncology. 2005 Aug;12(8):637–645. doi: 10.1245/ASO.2005.06.012. PubMed PMID: 15965730. Epub 2005/06/21. eng. [DOI] [PubMed] [Google Scholar]
- 5.Cirocchi R, Trastulli S, Abraha I, et al. Non-resection versus resection for an asymptomatic primary tumour in patients with unresectable stage IV colorectal cancer. The Cochrane database of systematic reviews. 2012;8:Cd008997. doi: 10.1002/14651858.CD008997.pub2. PubMed PMID: 22895981. Epub 2012/08/17. eng. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Stillwell AP, Buettner PG, Ho YH. Meta-analysis of survival of patients with stage IV colorectal cancer managed with surgical resection versus chemotherapy alone. World journal of surgery. 2010 Apr;34(4):797–807. doi: 10.1007/s00268-009-0366-y. PubMed PMID: 20054541. Epub 2010/01/08. eng. [DOI] [PubMed] [Google Scholar]
- 7.Krell RW, Regenbogen SE, Wong SL. Variation in hospital treatment patterns for metastatic colorectal cancer. Cancer. 2015 Jun 1;121(11):1755–1761. doi: 10.1002/cncr.29253. PubMed PMID: 25640016. Epub 2015/02/03. eng. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Bilimoria KY, Stewart AK, Winchester DP, Ko CY. The National Cancer Data Base: a powerful initiative to improve cancer care in the United States. Ann Surg Oncol. 2008 Mar;15(3):683–690. doi: 10.1245/s10434-007-9747-3. PubMed PMID: 18183467. Pubmed Central PMCID: Pmc2234447. Epub 2008/01/10. eng. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Winchester DP, Stewart AK, Phillips JL, Ward EE. The national cancer data base: past, present, and future. Ann Surg Oncol. 2010 Jan;17(1):4–7. doi: 10.1245/s10434-009-0771-3. PubMed PMID: 19847564. Pubmed Central PMCID: Pmc2805801. Epub 2009/10/23. eng. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases. J Clin Epidemiol. 1992 Jun;45(6):613–619. doi: 10.1016/0895-4356(92)90133-8. PubMed PMID: 1607900. Epub 1992/06/01. eng. [DOI] [PubMed] [Google Scholar]
- 11. [cited 2016 January 1];National Cancer Data Base Participant User File Data Dictionary. Available from: http://ncdbpuf.facs.org/node/259?q=print-pdf-all.
- 12.Dimick JB, Ghaferi AA, Osborne NH, et al. Reliability adjustment for reporting hospital outcomes with surgery. Ann Surg. 2012 Apr;255(4):703–707. doi: 10.1097/SLA.0b013e31824b46ff. PubMed PMID: 22388108. Epub 2012/03/06. eng. [DOI] [PubMed] [Google Scholar]
- 13.Dimick JB, Welch HG, Birkmeyer JD. Surgical mortality as an indicator of hospital quality: the problem with small sample size. Jama. 2004 Aug 18;292(7):847–851. doi: 10.1001/jama.292.7.847. PubMed PMID: 15315999. Epub 2004/08/19. eng. [DOI] [PubMed] [Google Scholar]
- 14.Pradarelli JC, Varban OA, Ghaferi AA, et al. Hospital variation in perioperative complications for laparoscopic sleeve gastrectomy in Michigan. Surgery. 2015 Oct 21; doi: 10.1016/j.surg.2015.08.044. PubMed PMID: 26506567. Epub 2015/10/28. Eng. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Greenberg CC, Lipsitz SR, Hughes ME, et al. Institutional variation in the surgical treatment of breast cancer: a study of the NCCN. Ann Surg. 2011 Aug;254(2):339–345. doi: 10.1097/SLA.0b013e3182263bb0. PubMed PMID: 21725233. Pubmed Central PMCID: Pmc3428030. Epub 2011/07/05. eng. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Shapiro M, Rashid NU, Whang EE, et al. Trends and predictors of resection of the primary tumor for patients with stage IV colorectal cancer. Journal of surgical oncology. 2015 Jun;111(7):911–916. doi: 10.1002/jso.23906. PubMed PMID: 25919984. Epub 2015/04/29. eng. [DOI] [PubMed] [Google Scholar]
- 17.Zafar SY, Peppercorn JM, Schrag D, et al. The financial toxicity of cancer treatment: a pilot study assessing out-of-pocket expenses and the insured cancer patient's experience. The oncologist. 2013;18(4):381–390. doi: 10.1634/theoncologist.2012-0279. PubMed PMID: 23442307. Pubmed Central PMCID: PMC3639525. Epub 2013/02/28. eng. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Regenbogen SE, Veenstra CM, Hawley ST, et al. The personal financial burden of complications after colorectal cancer surgery. Cancer. 2014 Oct 1;120(19):3074–3081. doi: 10.1002/cncr.28812. PubMed PMID: 24889014. Epub 2014/06/04. eng. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.