ORIGINAL CONTRIBUTION

Increased 30-Day Mortality Risk in Patients With

Diabetes Mellitus After Colon Cancer Surgery:
A Mediation Analysis
Mario Schootman, Ph.D.1,2 • Donna B. Jeffe, Ph.D.2,3 • Kendra L. Ratnapradipa, Ph.D.1
Jan M. Eberth, Ph.D.4 • Nicholas O. Davidson, M.D., D.Sc.2,5
1 Department of Epidemiology and Biostatistics, College for Public Health and Social Justice, Saint Louis University, St. Louis,
Missouri
2 Alvin J. Siteman Cancer Center at Barnes-Jewish Hospital and Washington University School of Medicine, St. Louis, Missouri
3 Division of General, Medical Sciences, Department of Medicine, Washington University School of Medicine, St. Louis,
Downloaded from http://journals.lww.com/dcrjournal by BhDMf5ePHKbH4TTImqenVBaqevB2sTM0YR37Qfy9Ay/E/sMxmAMBVbIA+pns+0EG on 02/10/2020

Missouri
4 Department of Epidemiology and Biostatistics, Arnold School of Public Health, University of South Carolina, Columbia,
South Carolina
5 Division of Gastroenterology, Department of Medicine, Washington University School of Medicine, St. Louis, Missouri

BACKGROUND:  Patients with (versus without) diabetes treatment related, comorbidity, health behaviors, surgical
mellitus who develop colon cancer are at increased risk of complications, and biomarkers of underlying disease.
dying within 30 days after surgery. MAIN OUTCOME MEASURES:  We measured all-cause 30-
OBJECTIVE:  The purpose of this study was to identify day mortality.
potential mediators of the effect of diabetes mellitus on all- RESULTS:  Of 26,060 patients, 18.8% (n = 4905) had
cause 30-day mortality risk after surgery for colon cancer. diabetes mellitus that was treated with insulin (n = 1595)
DESIGN:  A retrospective cohort study was conducted or other antidiabetic agents (n = 3340). Patients with
using the 2013–2015 National Surgical Quality diabetes mellitus had a 1.57 (95% CI, 1.23–1.99) higher
Improvement Program data. unadjusted odds of dying within 30 days versus patients
SETTING:  The study was conducted at various hospitals without diabetes mellitus. In the multivariable model,
across the United States (from 435 to 603 hospitals). 76.7% of the association between diabetes mellitus and
30-day mortality was explained; patients with diabetes
PATIENTS:  Patients who underwent resection for colon
mellitus were equally likely to die within 30 days versus
cancer with or without obstruction based on the National
those without diabetes mellitus (OR = 1.05 (95% CI,
Surgical Quality Improvement Program colectomy
0.81–1.35)). Anemia and sepsis explained 33.7% and
module were included. Patients who had ASA physical
15.2% of the effect of diabetes mellitus on 30-day
status classification V or metastatic disease and those
mortality (each p < 0.0001). Treatment-related variables,
who presented emergently were excluded. Patients
cardiovascular disease, surgical complications, and
were classified as “no diabetes,” “diabetes not requiring
biomarkers played limited roles as mediators.
insulin,” or “diabetes requiring insulin.” Potential
reasons for increased risk of dying within 30 days were LIMITATIONS:  The study was limited to larger hospitals,
and limited information about duration and type of
Funding/Support: Dr Davidson was supported in part by National diabetes mellitus was available.
Institutes of Health grants P30DK52574 and R01DK56260. Dr Eberth
was supported in part by MRSG-15-148-01-CPHPS from the American
CONCLUSIONS:  Better management and prevention of
Cancer Society. anemia and sepsis among patients with diabetes mellitus
may reduce their increased risk of death after colon
Financial Disclosure: None reported. cancer resection. See Video Abstract at http://links.lww.
com/DCR/B140.
Correspondence: Mario Schootman, Ph.D., SSM Health, 10101 Woodfield
Lane, St. Louis, MO 63132. E-mail: [email protected]

Dis Colon Rectum 2020; 63: 290–299 AUMENTO DEL RIESGO DE MORTALIDAD A 30 DÍAS EN
DOI: 10.1097/DCR.0000000000001586 PACIENTES DIABETICOS LUEGO DE CIRUGÍA DE CÁNCER
© The ASCRS 2020 DE COLON: ANÁLISIS DE MEDIACIÓN
290 DISEASES OF THE COLON & RECTUM VOLUME 63: 3 (2020)

Copyright © The American Society of Colon & Rectal Surgeons, Inc. Unauthorized reproduction of this article is prohibited.
 DISEASES OF THE COLON & RECTUM VOLUME 63: 3 (2020) 291

ANTECEDENTES:  Los pacientes con (y sin) diabetes que KEY WORDS:  Anemia; Colon and rectal cancer; Diabetes
desarrollan cáncer de colon tienen un mayor riesgo de mellitus; Mediation analysis; Prognosis; Sepsis.
morir dentro de los 30 días posteriores a la cirugía.

C
OBJETIVO:  Identificar los posibles mediadores del efecto olorectal cancer (CRC) is the third most common
de la diabetes sobre el riesgo de mortalidad dentro los 30 cancer in the United States and accounts for 9% of
días, por cualquier causa después de cirugía por cáncer de all cancer deaths.1 Although evidence exists for the
colon. association between diabetes mellitus and increased risk
of CRC incidence,2 the impact of diabetes mellitus on 30-
DISEÑO:  Estudio de cohortes retrospectivo entre 2013- day mortality after CRC resection is less well established.
2015 utilizando los datos del Programa Nacional de Thirty-day mortality is a key quality measure of cancer
Mejoría en Calidad Quirúrgica. care,3 and variation in mortality among CRC patients
AJUSTE:  Entre 435 a 603 hospitales en los Estados Unidos. undergoing surgery is especially pronounced during this
PACIENTES:  Se incluyeron aquellos pacientes sometidos time.4 Patients with diabetes mellitus who develop CRC
a resección por cáncer de colon con o sin obstrucción are more likely to die within 30 days compared with those
según el módulo de colectomía Programa Nacional without diabetes mellitus in some5,6 but not all studies.7,8
de Mejoría en Calidad Quirúrgica. Se excluyeron los However, most studies were conducted outside the United
pacientes estadío V de la clasificación de la Sociedad States or sampled only a few hospitals with relatively few
Estadounidense de Anestesiólogos (ASA), aquellos con diabetic patients. Improved understanding of the associa-
enfermedad metastásica y aquellos operados de urgencia. tion between diabetes mellitus and 30-day mortality risk
Los pacientes se clasificaron como “sin diabetes," “con after CRC resection has important implications for treat-
diabetes que no requiere insulina” o “con diabetes ing CRC patients with diabetes mellitus given the increas-
que requiere insulina." Las posibles razones para un ing prevalence of diabetes mellitus in the United States.
mayor riesgo de morir dentro de los 30 días estuvieron There is a need to understand the mechanisms that
relacionadas con el tratamiento, la comorbilidad, lead to worse prognosis among patients with diabetes mel-
los comportamientos de salud, las complicaciones litus to identify opportunities to reduce their risk of dying
quirúrgicas y los biomarcadores de enfermedad. within 30 days of colon cancer resection.9 Although direct
empirical evidence is lacking, studies have suggested po-
PRINCIPALES RESULTADOS:  Mortalidad de cualquier tential explanatory mechanisms through which diabetes
orígen dentro los 30 días depués de la cirugía. mellitus increases the risk of 30-day mortality after sur-
RESULTADOS:  De 26’060 pacientes, 18.8% (n = 4,905) gery, including surgical site infection and sepsis, anasto-
tenían diabetes tratada con insulina (n = 1,595) u otros motic leakage, myocardial infarction, comorbidity, BMI,
agentes antidiabéticos (n = 3,340). Los pacientes con diabetes and different treatments for colon cancer.5–7,10 In this
tenían 1.57 (IC 95%: 1.23-1.99) mayores probabilidades no study of a national sample of patients with colon cancer,
ajustadas de morir dentro de los 30 días en comparación we examined whether patients with diabetes mellitus were
con los pacientes sin diabetes. En el modelo multivariable, at increased risk of 30-day mortality after resection for
se explicó que el 76,7% de la asociación entre diabetes y colon cancer and the extent to which the association be-
mortalidad a los 30 días; los pacientes con diabetes tenían tween diabetes mellitus and 30-day mortality was medi-
la misma probabilidad de morir dentro de los 30 días que ated (ie, explained) by type of treatment for colon cancer,
aquellos sin diabetes (OR: 1.05; IC 95%: 0.81-1.35). La surgical complications, comorbidity, biomarkers, and be-
anemia y la sepsis explicaron el 33,7% y el 15,2% del efecto havior. Mediation analysis is increasingly used in clinical
de la diabetes en la mortandad a 30 días (p <0,0001). Las research.11,12 It is especially useful in observational studies
variables relacionadas con el tratamiento, las enfermedades like ours, because mediation analysis can provide causal
cardiovasculares, las complicaciones quirúrgicas y los explanations for the association between an exposure var-
biomarcadores jugaron un papel limitado como mediadores. iable and outcome, which in this study is the association
LIMITACIONES:  Estudio limitado a hospitales más between diabetes mellitus and 30-day mortality.
grandes e información limitada sobre la duración y el
tipo de diabetes. MATERIALS AND METHODS
CONCLUSIONES:  Una mejor prevención y manejo de la
Data Source and Patient Selection
anemia y la sepsis en los pacientes con diabetes puede reducir
el mayor riesgo de muerte después de la resección del We used the 2013–2015 American College of Surgeons
cáncer de colon. Consulte Video Resumen en http://links. National Surgical Quality Improvement Program (ACS-
lww.com/DCR/B140. (Traducción—Dr. Xavier Delgadillo) NSQIP) Participant Use File and Targeted Colectomy File.
The ACS-NSQIP measures the quality of surgical care
by collecting patient-level data at participating hospitals.

Copyright © The American Society of Colon & Rectal Surgeons, Inc. Unauthorized reproduction of this article is prohibited.
 292 SCHOOTMAN ET AL: 30-DAY MORTALITY IN COLON CANCER

Additional details about the NSQIP can be found else- Potential mediators:
where https://www.facs.org/quality-programs/acs-nsqip). • Treatment related
The Saint Louis University Institutional Review Board • Comorbidity
• Behavior
considered this study to be exempt from oversight. Pre- • Surgical complications
operative patient characteristics, preoperative laboratory • Biomarkers of underlying disease
results, intraoperative procedure characteristics, surgical
complications, and 30-day postoperative mortality rates
were abstracted by trained reviewers from medical charts
of participating hospitals. Patients who underwent resec- Diabetes mellitus Direct effect 30-day mortality
tion for nonemergent, nonmetastatic colon cancer (with
or without obstruction) were included. Patients whose
ASA physical status was class V (not expected to survive Demographic confounders
without surgery) and those with metastatic colon cancer
were excluded from analysis because of their increased risk FIGURE 1.  Conceptual model of potential mediators and
of 30-day mortality. confounders for the association between diabetes mellitus and risk
of 30-day mortality after colon resection. A double-headed arrow
refers to a nondirectional association; a single-headed arrow refers
Diabetes Status to variables predicting 30-day mortality.
Patient diabetes status was classified in the NSQIP data as
“no diabetes” (no diagnosis of diabetes mellitus or diabe-
no), preoperative antibiotic use (yes, no), chemotherapy
tes mellitus controlled by diet alone), “diabetes not requir-
(yes, no), American Joint Commission on Cancer TNM
ing insulin” (diabetes mellitus requiring therapy with a
stage, and tumor location (left, right, other). Comorbidi-
noninsulin antidiabetic agent, such as oral agents or other
ties included patient functional status (partially dependent
noninsulin agents), or “diabetes requiring insulin” (diabe-
or totally dependent versus independent), hypertension,
tes mellitus requiring daily insulin therapy). Data about
congestive heart failure, renal failure, preoperative loss
the time since first diabetes diagnosis and the type of dia-
of blood necessitating a transfusion, weight loss >10% in
betes were not available in the NSQIP data.
the last 6 months, and chronic obstructive pulmonary di-
sease.10 All were measured within 30 days before surgery.
Patient Outcome Behavior-related variables included smoking status
All-cause 30-day mortality was operationalized as death and BMI. A patient who had smoked cigarettes in the year
within 30 days of resection obtained by NSQIP-trained before admission for surgery was considered a smoker. The
abstractors. patient’s most recent height and weight, documented in
the medical chart within the 30 days before the colectomy
Potential Mediators and Confounders or at the time the patient was being considered a candi-
Potential mediators and confounders were selected based date for surgery, was used to calculate BMI in kilograms
on previous studies of prognostic factors of 30-day mor- per meter squared. Patients were classified as underweight
tality among persons with diabetes mellitus or those (BMI <18.5), normal weight (18.5 ≤ BMI < 25.0), over-
­diagnosed with colon cancer.6,13–16 Mediators are variables weight (25.0 ≤ BMI < 30.0), class 1 to 2 obesity (30.0
that are hypothesized to be in the causal pathway between ≤ BMI < 40.0), or class 3 obesity (BMI ≥40.0).
diabetes mellitus and 30-day mortality and must be asso- Surgical complications included readmission, reoper-
ciated with both diabetes mellitus and 30-day mortality; ation, sepsis, deep vein thrombosis, pulmonary embolism,
importantly, the exposure is presumed to cause the medi- myocardial infarction, stroke, blood transfusion, sum of
ator, and the mediator is presumed to cause the outcome, infections (including surgical site, wound, and deep infec-
and not vice versa.17 The confounders also are associated tions), anastomotic leakage, and prolonged postoperative
with both diabetes mellitus and 30-day mortality but are ileus within 30 days. Presence of anastomotic leakage was
not in the causal pathway and are not amenable to inter- determined if there was chart documentation, regardless
vention (Fig. 1). Potential mediators were categorized as of treatment interventions. Absence of a potential anasto-
treatment-related variables for colon cancer, comorbidi- motic leak was determined if there was no definitive diag-
ties, behavior-related variables, surgical complications, nosis of a leak or related abscess documented in the chart.
and biomarkers of underlying disease. Potential con- Readmission was defined as any readmission for any rea-
founders included patient demographic characteristics son, (to the same or another hospital), within 30 days of
(sex, age group, Hispanic ethnicity, race, and year of sur- the resection. Reoperation was defined as any unplanned
gery), which are not amenable to intervention. return to the operating room for a surgical procedure, for
Treatment-related variables included type of surgery any reason, within 30 days of surgery at any hospital or
(open, laparoscopic), mechanical bowel preparation (yes, surgical facility. A wound infection included superficial or

Copyright © The American Society of Colon & Rectal Surgeons, Inc. Unauthorized reproduction of this article is prohibited.
 DISEASES OF THE COLON & RECTUM VOLUME 63: 3 (2020) 293

deep incisional surgical site infection or any other wound Sobel test to determine whether statistically significant (p
infection. Receipt of transfusion (yes/no) from the start of < 0.05) mediation existed (http://quantpsy.org/sobel/so-
surgery to 72 hours postsurgery was used as a measure of bel.htm)26,27; this test determines whether the reduction
extensive blood loss. in the effect of the diabetes mellitus on 30-day mortality
Biomarkers included the following preoperative lab- risk is significant after adjusting for the mediator, and, if
oratory values with abnormal value cutoffs noted in pa- so, the mediation effect is significant. Sixth, we used the
rentheses: white blood cell count (≤4.5 or ≥11.0 × 103/µL), KHB decomposition method to calculate the mediating
glomerular filtration rate, serum albumin (3.0 mg/dL), percentage for each of the 5 categories of potential media-
serum alkaline phosphatase (>125 mg/dL), serum biliru- tors. Finally, we used a single model with multiple media-
bin (≥1.0 mg/dL), blood urea nitrogen (BUN; ≥40 mg/dL) tors that were statistically significant using the Sobel test
and BUN:creatinine ratio (>20),18 platelet count (150– in single-mediator models. This multiple-mediator model
450 × 103/µL), partial thromboplastin time (>35), and identified variables that significantly mediated the associ-
preoperative hematocrit. Hematocrit values were treated ation of interest while adjusting for other mediators, con-
as a categorical variable with the following cutoffs for ane- founders, and diabetes mellitus. The multiple-mediator
mia: severe (<25%), moderate (25% to <29%), mild (29% approach allows the comparison of the magnitude of dif-
to <37%), and no anemia (≥37%).19,20 Glomerular filtra- ferent mediators of the effect of diabetes mellitus on 30-
tion rate was calculated based on race, sex, age, and creati- day mortality risk.
nine levels with a cutoff glomerular filtration rate >60 mL/
min/1.73 m2.21
RESULTS
Statistical Analysis Data were collected on 66,031 patients with colon resec-
First, we tested the univariate associations between diabe- tions in the 2013–2015 NSQIP. The primary indication for
tes status and each potential mediator and demographic 36,734 of those patients was for nonmetastatic colon can-
confounder using χ2 tests. Second, we identified demo- cer with or without obstruction; of those patients, we ex-
graphic variables that were confounders of the associa- cluded from analysis 10,674 who were designated as ASA
tion between diabetes mellitus and 30-day mortality or physical status class V, presented emergently and required
between a mediator and 30-day mortality for inclusion surgery, or lacked data for type of surgery. In all, 26,060
in the mediator models using logistic regression analy- patients were included in the analysis, 4905 (18.8%) of
sis.11 Third, we determined the interaction between dia- whom had been diagnosed with diabetes mellitus requir-
betes mellitus and the potential mediators.22 Fourth, we ing treatment with insulin (1595 (6.1%)) or other anti-
determined which potential mediators were associated diabetic agents (3340 (12.7%)). Table 1 shows that those
with diabetes status and 30-day mortality, adjusting for with diabetes mellitus were more likely to be male, older,
confounders, reporting OR and 95% CI.22 Fifth, we used Hispanic, black, hospitalized longer, overweight or obese,
a decomposition method (KHB package in Stata/SE 14.2; have comorbid conditions, have elevated adverse bio-
StataCorp, College Station, TX) to determine the extent to markers, and to develop surgical complications. Patients
which each of the variables mediated the association be- with diabetes mellitus were less likely to receive chemo-
tween diabetes mellitus and 30-day mortality in separate therapy, to be diagnosed with cancer located in the left
single-mediator models.23–25 This is done by decompos- colon, or to smoke. The only demographic characteristic
ing the total logit coefficient into its direct and indirect that acted as a confounder was age (Table 1). The other
(mediated) parts and calculating a mediating percentage demographic characteristics were no longer confounders
to assess the relative magnitude of direct and indirect ef- once age was included in the model. Many of the afore-
fects, which can be given a causal interpretation.23–25 The mentioned characteristics increased the risk of 30-day
mediating percentage provides an estimate of the relative mortality when controlling for age. Characteristics that
magnitude of the explanatory effect of each mediator sep- could not be mediators because they were not associated
arately on the association between diabetes mellitus and with diabetes mellitus and/or 30-day mortality included
30-day mortality risk and typically ranges from 0% (no treatment-related variables (type of surgery, receipt of
mediating effect) to 100% (complete mediation). Negative chemotherapy, stage at diagnosis, antibiotic preparation),
mediating percentages mean that the mediator increased, comorbidity (hypertension), biomarkers (white blood
rather than reduced, the OR (further away from the value cell count, albumin, alkaline phosphatase, partial throm-
of 1.0) between diabetes mellitus and 30-day mortality boplastin time), and surgical complications (reoperation,
when included in the model. This could be attributed to deep vein thrombosis, pulmonary embolism, blood trans-
an increase in 30-day mortality with an increase in the fusion, surgical site infection, anastomotic leak, prolonged
mediator or because a negative association exists between ileus). The interaction between diabetes mellitus and none
diabetes mellitus and the potential mediator. We used the of the potential mediators was statistically significant.

Copyright © The American Society of Colon & Rectal Surgeons, Inc. Unauthorized reproduction of this article is prohibited.
 294 SCHOOTMAN ET AL: 30-DAY MORTALITY IN COLON CANCER

TABLE 1.   Study population characteristics in association between each potential mediator and each of diabetes status and 30-day
mortality among patients with nonmetastatic colon cancer after surgical resection
Diabetes Association between Association between
each potential each potential mediator
mediator and DM and 30-day mortality
Yes No adjusted for age group, adjusted for DM and
Characteristic (n = 4905) % (n = 21,154) % OR (95% CI) age group, OR (95% CI)
Demographic confounders
Sex (male)* 56.7 49.8 1.38 (1.29–1.47) 1.21 (0.98–1.49)
Age group, y*
 <45 1.3 6.8 Ref Ref
 45–54 9.1 16.5 2.89 (2.21–3.78) 1.49 (0.32–7.04)
 55–64 22.7 23.1 5.13 (3.97–6.66) 4.69 (1.13–19.45)
 65–74 34.4 24.6 7.31 (5.66–9.44) 7.92 (1.94–32.33)
 ≥75 32.5 29.0 5.88 (4.55–7.60) 22.41 (5.56–90.30)
Hispanic ethnicity* 6.2 4.0 1.79 (1.56–2.05) 0.69 (0.35–1.34)
Race*
 White 67.0 73.2 Ref Ref
 Black  13.7 9.4 1.72 (1.56–1.89) 0.83 (0.56–1.23)
 Other 5.9 4.7 1.46 (1.28–1.68) 0.71 (0.39–1.27)
 Unknown 13.5 12.7 1.16 (1.05–1.27) 0.65 (0.45–0.93)
Treatment-related mediators
Type of surgery: open vs laparoscopic 30.4 28.4 1.06 (0.99–1.14) 2.56 (2.08–3.16)
Mechanical bowel preparation* 57.2 57.2 1.08 (1.01–1.17) 0.44 (0.35–0.55)
Oral antibiotics 31.3 31.9 1.01 (0.94–1.08) 0.52 (0.40–0.68)
Chemotherapy* 7.8 10.4 0.86 (0.77–0.96) 0.79 (0.49–1.28)
TNM stage*
 0 1.8 2.2 Ref Ref
 I 15.7 14.6 1.15 (0.91–1.45) 3.09 (0.42–22.8)
 II 17.3 17.4 0.99 (0.79–1.25) 5.15 (0.71–37.24)
 III 16.4 17.9 0.96 (0.76–1.21) 5.18 (0.71–37.54)
 Unknown 48.9 48.0 1.06 (0.85–1.32) 6.70 (0.94–48.00)
Tumor location*
 Left 37.3 41.7 Ref Ref
 Right 41.1 35.8 1.10 (1.02–1.18) 1.05 (0.81–1.37)
 Transverse 21.6 22.5 1.00 (0.92–1.09) 1.75 (1.34–2.31)
Comorbidity mediators
Functional status* 4.3 2.4 1.59 (1.35–1.87) 3.98 (2.93–5.40)
Hypertension* 81.7 47.3 4.55 (4.20–4.93) 1.15 (0.91–1.46)
Congestive heart failure* 2.6 1.1 2.05 (1.65–2.56) 4.07 (2.72–6.08)
Renal failure* 1.5 0.5 3.13 (2.30–4.26) 2.25 (0.98–5.16)
Preoperative loss of blood 13.3 9.6 1.31 (1.19–1.44) 3.87 (3.10–4.83)
necessitating transfusion*
Weight loss >10%* 4.7 5.4 0.85 (0.74–0.99) 2.12 (1.51–2.98)
COPD* 6.9 5.2 1.19 (1.05–1.35) 2.59 (1.95–3.44)
Behavioral mediators
BMI, kg/m2*
 <18.5 0.9 2.7 1.76 (1.28–2.43) 0.41 (0.26–0.63)
 18.5–24.9 17.7 32.8 Ref Ref
 25.0–29.9 29.9 34.6 2.94 (2.14–4.05) 0.28 (0.18–0.44)
 30.0–39.9 40.6 25.6 5.79 (4.21–7.97) 0.29 (0.18–0.47)
 ≥40.0 11.0 4.3 10.57 (7.57–14.77) 0.47 (0.26–0.87)
Smoking* 10.2 13.8 0.81 (0.73–0.90) 1.85 (1.36–2.51)
Biomarker-related mediators
Abnormally low or high WBC count* 16.8 17.6 0.96 (0.88–1.05) 1.85 (1.46–2.34)
Glomerular filtration rate* 28.5 15.8 1.75 (1.62–1.89) 1.55 (1.24–1.95)
Abnormally high albumin* 7.5 6.7 0.98 (0.87–1.11) 4.54 (3.56–5.79)
Abnormally high alkaline phosphatase 32.9 33.1 1.01 (0.95–1.08) 1.01 (0.81–1.26)
Abnormally high bilirubin* 5.3 6.9 0.73 (0.63–0.83) 1.83 (1.32–2.54)
Abnormally high BUN* 3.0 0.9 2.74 (2.20–3.40) 3.98 (2.62–6.03)
Abnormally high BUN:creatinine ratio* 23.6 19.3 1.39 (1.11–1.74) 1.38 (1.10–1.74)
Abnormally low platelet count* 7.3 5.6 1.24 (1.09–1.40) 1.97 (1.44–2.70)
(Continued)

Copyright © The American Society of Colon & Rectal Surgeons, Inc. Unauthorized reproduction of this article is prohibited.
 DISEASES OF THE COLON & RECTUM VOLUME 63: 3 (2020) 295

TABLE 1.   Continued

Diabetes Association between Association between
each potential each potential mediator
mediator and DM and 30-day mortality
Yes No adjusted for age group, adjusted for DM and
Characteristic (n = 4905) % (n = 21,154) % OR (95% CI) age group, OR (95% CI)
Abnormally high partial 5.3 4.1 1.10 (0.95–1.27) 1.52 (1.03–2.24)
thromboplastin* time
Preoperative low hematocrit*
 Severe 2.3 1.7 1.52 (1.22–1.89) 4.80 (2.97–7.75)
 Moderate 10.4 7.0 1.72 (1.54–1.93) 2.69 (1.91–3.80)
 Mild 43.3 33.4 1.54 (1.44–1.65) 2.18 (1.69–2.81)
 No anemia 41.6 54.3 Ref Ref
 Unknown 2.3 3.6 0.83 (0.68–1.02) 0.34 (0.08–1.40)
Complication-related mediators
 Readmission* 11.5 9.5 1.22 (1.11–1.35) 1.79 (1.35–2.37)
 Reoperation 5.0 4.4 1.15 (0.99–1.33) 7.30 (5.66–9.44)
 Sepsis* 4.6 3.5 1.31 (1.12–1.53) 15.86 (12.54–20.04)
  Deep vein thrombosis 1.3 1.2 1.04 (0.79–1.37) 3.89 (2.37–6.38)
  Pulmonary embolism 1.0 0.7 1.28 (0.92–1.78) 4.81 (2.74–8.45)
  Myocardial infarction* 1.1 0.7 1.28 (0.94–1.75) 11.31 (7.74–16.54)
 Stroke* 0.6 0.3 1.96 (1.24–3.09) 11.01 (6.23–19.46)
  Blood transfusion* 3.7 2.9 1.14 (0.96–1.34) 2.15 (1.47–3.14)
  One or more infections* 7.8 5.6 1.42 (1.26–1.61) 2.06 (1.48–2.86)
  Anastomotic leak 3.8 3.5 1.13 (0.95–1.33) 7.31 (5.49–9.73)
  Prolonged ileus* 15.2 14.3 1.03 (0.94–1.12) 4.78 (3.87–5.90)
BUN = blood urea nitrogen; WBC = white blood cell; COPD = chronic obstructive pulmonary disease; Ref = reference; DM = diabetes mellitus.
*Data show p < 0.05 between patients with and those without diabetes mellitus based on χ2 test.

Overall, 30-day mortality was 1.4%, but was higher a- (95% CI, 1.08–1.74) to 1.38 (95% CI, 1.09–1.75). Signif-
mong patients with diabetes mellitus than those without icant biomarker mediators included hematocrit (26.2%)
diabetes mellitus (96 (2.0%) vs 266 (1.3%); p < 0.001). In and BUN:creatinine ratio (10.4%). These biomarkers
unadjusted analysis, patients with diabetes mellitus were mediated the association between diabetes mellitus and
1.57 times more likely to die within 30 days than patients 30-day mortality for 42.9%.
without diabetes mellitus (95% CI, 1.23–1.99). The dif- Table 3 shows that the significant mediators (shown
ference in 30-day mortality between patients treated with in Table 2) based on the Sobel test explained 76.7% of the
insulin and those treated with other antidiabetic agents total association between diabetes mellitus and 30-day
was not significant (31 (1.9%) vs 65 (2.0%); p = 0.962). mortality. Patients with diabetes mellitus were as likely to
Adjusting for age showed that patients with diabetes mel- die within 30 days versus those without diabetes mellitus
litus were 1.37 times more likely to die within 30 days of (OR = 1.05 (95% CI, 0.81–1.35)) in this multiple-medi-
surgery than patients without diabetes mellitus (95% CI, ator model. Of these variables, anemia explained 33.7%
1.08–1.74). and sepsis explained 15.2% of the association between
Table 2 shows that several variables significantly me- diabetes mellitus and 30-day mortality. Length of hospi-
diated the association between diabetes mellitus and 30- talization, functional status, congestive heart failure, renal
day mortality based on the Sobel test, adjusting for the failure, BUN:creatinine ratio, and stroke explained <10%
age confounder. Of the comorbidity variables, functional of the association between diabetes mellitus and 30-day
status, congestive heart failure, renal failure, and chronic mortality.
obstructive pulmonary disease explained 18.7% of the
association between diabetes mellitus and 30-day mor- DISCUSSION
tality. Because underweight, overweight, and obese pa-
tients were less likely to die within 30 days (Table 1), BMI The purpose of this study was to identify mediators that
increased rather than decreased the association between explain the increased risk of 30-day mortality among
diabetes mellitus and 30-day mortality to 1.45 (95% CI, patients with diabetes mellitus undergoing resection for
1.14–1.86), but the Sobel test was not statistically dif- nonemergent, nonmetastic colon cancer. Patients with di-
ferent (p = 0.08), indicating that BMI was not a media- abetes mellitus had higher odds of dying within 30 days
tor. Smoking mediated the association between diabetes than patients without diabetes mellitus. Although overall
mellitus and 30-day mortality for –4.2% (p=0.006) and 30-day mortality was 1.4%, it was higher among patients
slightly increased the OR for diabetes mellitus from 1.37 with diabetes mellitus (2.0%) than those without d
­ iabetes

Copyright © The American Society of Colon & Rectal Surgeons, Inc. Unauthorized reproduction of this article is prohibited.
 296 SCHOOTMAN ET AL: 30-DAY MORTALITY IN COLON CANCER

TABLE 2.   Percentage of the effect of diabetes mellitus on 30-day mortality after colon resection that is explained by each mediating
variable in single-mediator models and risk of 30-day mortality because of diabetes mellitus controlling for each mediator or blocks of
mediators
Odds of 30-day mortality
Percentage of the effect of associated with diabetes
diabetes mellitus on P value for the Sobel test mellitus, adjusted for
30-d mortality explained of the hypothesis that the each mediating
Potential mediator by the mediator mediated effect is zero factor, OR (95% CI)
Treatment related
Mechanical bowel preparation –1.0 0.2302 1.36 (1.08–1.73)
All treatment mediators combined with Sobel test p < 0.05
Comorbidity
Functional status 7.9 <0.0001 1.29 (1.02–1.64)
Congestive heart failure 6.3 <0.0001 1.30 (1.03–1.66)
Renal failure 2.7 0.0449 1.34 (1.06–1.70)
Preoperative loss of blood necessitating transfusion 1.1 0.8578 1.35 (1.07–1.71)
COPD 3.1 0.0059 1.35 (1.07–1.71)
Weight loss >10% –1.8 0.0588 1.37 (1.08–1.74)
All individual comorbidity mediators combined with 18.7 1.23 (0.97–1.57)
Sobel test p < 0.05
Behavior
BMI, kg/m2 –12.2 0.0745 1.45 (1.14–1.86)
Smoking –4.2 0.0062 1.38 (1.09–1.75)
All behaviors combined with Sobel test p < 0.05 –4.2 1.38 (1.09–1.75)
Biomarkers
Abnormally low platelet count –0.1 0.8494 1.36 (1.07–1.72)
Preoperative hematocrit 26.2 <0.0001 1.26 (0.99–1.60)
Abnormally high bilirubin 1.5 0.0770 1.35 (1.07–1.72)
Glomerular filtration rate 2.3 0.1048 1.34 (1.06–1.71)
BUN 2.8 0.1307 1.36 (1.07–1.72)
BUN:creatinine ratio 10.4 0.0320 1.34 (1.05–1.69)
All biomarkers combined with Sobel test p < 0.05 42.9 1.21 (0.95–1.54)
Complications
Readmission 3.6 0.0042 1.34 (1.06–1.70)
Sepsis 11.8 0.0006 1.24 (0.97–1.59)
Stroke 2.4 0.0030 1.32 (1.04–1.67)
One or more infections 5.1 0.0005 1.34 (1.06–1.70)
All complications combined with Sobel test p < 0.05 17.5 1.20 (0.94–1.54)
Each model controlled for age as a confounder. Variables that were not associated with diabetes status and 30-day mortality in Table 1 were not included because they
cannot be mediators. These variables included: type of surgery, receipt of chemotherapy, TNM stage, tumor location, hypertension, abnormally high or low white blood cell
count, albumin, alkaline phosphatase, partial thromboplastin time, deep vein thrombosis, pulmonary embolism, blood transfusion, anastomotic leak, and prolonged ileus.
BUN = blood urea nitrogen; COPD = chronic obstructive pulmonary disease.

mellitus (1.3%). These findings underscore the impor- Our results confirm the high prevalence of anemia
tance of recognizing and managing diabetes mellitus in all and its prognostic value in colon cancer,28,29 as well as the
hospitalized patients, particularly those undergoing sur- role of anemia as an important mediator of the association
gical interventions. Variables found to be significant me- between diabetes mellitus and 30-day mortality. Although
diators of this association between diabetes mellitus and the presence of anemia preoperatively may reflect blood
30-day mortality risk in single-mediator models together loss, it may be attributed to lack of bone marrow produc-
explained 69.8% of this association in the multiple-medi- tion of red blood cells in the setting of chronic renal fail-
ator model. Among these mediators, anemia (30.8%) and ure.30 The prevalence of anemia in patients with diabetes
sepsis (13.0%) had significantly large explanatory effects mellitus is typically associated with the presence of kid-
on the association between diabetes mellitus and 30-day ney disease.31 Although important among all patients with
mortality. After adjusting for all multiple mediators and colon cancer, appropriate screening, detection, and treat-
the age confounder, patients with (versus without) dia- ment of both the anemia and underlying kidney disease
betes mellitus were no longer at increased risk of dying are essential to improve the clinical outcomes, particularly
within 30 days of colon resection (OR = 1.06 (95% CI, in colon cancer patients with diabetes mellitus.32
0.82–1.38)). Identification of significant mediators can Our results also show that sepsis could explain the
help inform targeted intervention efforts to reduce the el- greater likelihood of 30-day mortality after colon cancer
evated 30-day mortality risk after colon cancer resection resection in patients with diabetes mellitus. Patients with
among patients with diabetes mellitus. diabetes mellitus were more than 15 times more likely to

Copyright © The American Society of Colon & Rectal Surgeons, Inc. Unauthorized reproduction of this article is prohibited.
 DISEASES OF THE COLON & RECTUM VOLUME 63: 3 (2020) 297

TABLE 3.   Percentage of the effect of diabetes mellitus on 30-day they did not differ between patients with and without di-
mortality after colon resection that is explained by each mediating abetes mellitus. As a result, they were not considered to be
variable in a multiple-mediator model adjusted for age group, diabetes-specific mediators. Thus, although implementing
NSQIP 2013-2015
interventions targeting these variables has been shown to
Percentage of the reduce mortality, our findings suggest that the impact is
effect of diabetes P value of the unlikely to be confined to patients with diabetes mellitus
mellitus on 30-day hypothesis
mortality explained that the mediated
after colon cancer resection.
Mediators* by the mediator effect is zero We found that being overweight may provide ben-
efits relative to being within the normal weight category.
Anemia 33.7 0.0001
Sepsis 15.2 0.001 Because data on patient postdiagnosis weight was not
Functional status 7.1 0.0005 available in the NSQIP data, it was impossible to deter-
High BUN:creatinine ratio 6.3 0.0009 mine whether patients subsequently lost weight, which
BUN:creatinine ratio 6.4 0.20 could have a negative impact on patient outcomes. Oth-
unknown
Congestive heart failure 5.6 0.003
ers have suggested that heavier patients, who have greater
COPD 3.6 0.03 muscle and fat mass, may better cope with the metabolic
Stroke 3.0 0.04 demands of tumor progression and treatment.39 However,
Renal failure 1.1 0.66 BMI was not a significant mediator of the association be-
One or more infections 0.2 0.93 tween diabetes mellitus and 30-day mortality (Sobel test,
Readmission –1.5 0.36
Smoking –3.9 0.09
p = 0.0745).
Total mediating effect 76.7 We recognize there are limitations of our study. We
NSQIP = National Surgical Quality Improvement Program; BUN = blood urea nitro-
analyzed existing data available in the NSQIP database
gen; COPD = chronic obstructive pulmonary disease. and therefore were limited to the data that the NSQIP col-
*Data include variables that were mediators of the association between diabetes lects; characteristics about NSQIP participating hospitals
mellitus and 30-day mortality based on the Sobel test from the single-mediator
models in Table 2. and surgeons, and data about other potential mediators
with prognostic significance (eg, depression, inflamma-
tion, hyperglycemia, physical activity, alcohol use), were
die within 30 days if they developed sepsis. Other studies not available. In addition, the NSQIP database typically
showed sepsis to be a prognostic factor after colon cancer includes data from larger hospitals and not from a na-
resection and suggested more restrictive transfusion prac- tionally representative sample, limiting generalizability
tices to reduce the risk of sepsis and improve survival.33 to smaller hospitals and hospitals that are not part of the
However, for patients with anemia, other strategies may
NSQIP. The NSQIP data pertaining to diabetes did not in-
help avoid perioperative transfusions or reduce the risk
clude its duration and type of diabetes, and patients whose
of infection, including preoperative iron supplementation
diabetes was managed without oral medications or insulin
and reducing the storage time of blood.34,35 Postoperative
were classified in the no-diabetes group. Thus, we expect
multidisciplinary care as managed by geriatricians also
that the effect of diabetes status on 30-day mortality may
decreases the rate of surgical complications for older on-
be underestimated, because an estimated 30% of patients
cology patients.36 Mild glycemic control may reduce the
may have undiagnosed diabetes, and these patients were
risk of death after sepsis.37 Our results show that colon
included in the no-diabetes group in the NSQIP database.
cancer patients with diabetes mellitus should be especially
Our study also had strengths, including the identification
targeted.
of actionable targets for why diabetic patients were at in-
Macrovascular disease is a major complication a-
creased 30-day risk of dying after colon cancer surgery
mong persons with diabetes mellitus.5 Contrary to other
and use of the high-quality NSQIP data, which contained
studies,38 we found that stroke and congestive heart fail-
many potential mediators and confounders for a large
ure were also some of the reasons by which patients with
sample of patients with diabetes mellitus.
diabetes mellitus were more likely to die following colon
cancer resection. However, anemia and sepsis played more
important roles. A recent meta-analysis observed that very CONCLUSION
mild glycemic control reduced cardiovascular mortality.37
Despite assertions to the contrary,5,6,10,38 our analysis Oncologists, surgeons, and cancer patients should be aware
also identified variables that were not significant media- of the excess postoperative mortality risk related to diabe-
tors in the multiple-mediator model. Treatment-related tes mellitus. Recognizing and optimizing management of
variables (eg, stage at diagnosis, type of surgical treat- diabetes mellitus in an increasingly obese and comorbid
ment), anastomotic leakage, comorbidity, and myocardial population is thus an important priority for hospitalized
infarction were not mediators. Although these variables patients. Our findings suggest that managing anemia and
were associated with the increased risk of 30-day ­mortality, preventing sepsis among patients with diabetes mellitus

Copyright © The American Society of Colon & Rectal Surgeons, Inc. Unauthorized reproduction of this article is prohibited.
 298 SCHOOTMAN ET AL: 30-DAY MORTALITY IN COLON CANCER

may reduce their increased risk of dying within 30 days 17. MacKinnon DP. Introduction to Statistical Mediation Analysis.
after resection for nonmetastatic colon cancer. New York, NY: Lawrence Erlbaum Associates; 2008.
18. Matsue Y, van der Meer P, Damman K, et al. Blood urea nitro-
gen-to-creatinine ratio in the general population and in pa-
REFERENCES tients with acute heart failure. Heart. 2017;103:407–413.
19. Kwaan MR, Al-Refaie WB, Parsons HM, Chow CJ, Rothen-
1. American Cancer Society. Cancer Facts & Figures 2010. Atlanta,
berger DA, Habermann EB. Are right-sided colectomy out-
GA: American Cancer Society; 2010.
comes different from left-sided colectomy outcomes?: study of
2. Larsson SC, Orsini N, Wolk A. Diabetes mellitus and risk
patients with colon cancer in the ACS NSQIP database. JAMA
of colorectal cancer: a meta-analysis. J Natl Cancer Inst.
Surg. 2013;148:504–510.
2005;97:1679–1687.
20. Sirany AM, Chow CJ, Kunitake H, Madoff RD, Rothenberger
3. Institute of Medicine. Ensuring Quality Cancer Care. Washing-
DA, Kwaan MR. Colorectal surgery outcomes in chronic dial-
ton, DC: National Academy Press, 1999.
ysis patients: an American College of Surgeons National Sur-
4. Møller H, Sandin F, Robinson D, et al. Colorectal cancer sur-
gical Quality Improvement Program study. Dis Colon Rectum.
vival in socioeconomic groups in England: variation is mainly
2016;59:662–669.
in the short term after diagnosis. Eur J Cancer. 2012;48:46–53.
21. Levey AS, Stevens LA, Schmid CH, et al.; CKD-EPI (Chronic
5. Barone BB, Yeh H-C, Snyder CF, Peairs KS, Stein KB, Derr RL, Kidney Disease Epidemiology Collaboration). A new equa-
Wolff AC, Brancati FL. Postoperative mortality in cancer pa- tion to estimate glomerular filtration rate. Ann Intern Med.
tients with preexisting diabetes. Syst Rev. 2010;33:931–939. 2009;150:604–612.
6. Davila JA, Rabeneck L, Berger DH, El-Serag HB. Postopera- 22. Vanderweele TJ, Vansteelandt S. Odds ratios for media-
tive 30-day mortality following surgical resection for colorectal tion analysis for a dichotomous outcome. Am J Epidemiol.
cancer in veterans: changes in the right direction. Dig Dis Sci. 2010;172:1339–1348.
2005;50:1722–1728. 23. Kohler U, Karlson KB, Holm A. Comparing coefficients of nest-
7. Yap R, Wilkins S, Staples M, Oliva K, McMurrick PJ. The effect ed nonlinear probability models. Stata J. 2011;11:420–438.
of diabetes on the perioperative outcomes of colorectal cancer 24. Breen R, Karlson KB, Holm A. Total, direct, and indirect effects
surgery patients. PLoS One. 2016;11:e0167271. in logit and probit models. Sociol Methods Res. 2013;42:164–191.
8. Anand N, Chong CA, Chong RY, Nguyen GC. Impact of diabe- 25. Karlson KB, Holm A, Breen R. Comparing regression coef-
tes on postoperative outcomes following colon cancer surgery. J ficients between same-sample nested models using logit and
Gen Intern Med. 2010;25:809–813. probit. Sociol Methodol. 2012;42:286–313.
9. Meyerhardt JA, Catalano PJ, Haller DG, et al. Impact of diabe- 26. Sobel M. Asymptotic confidence intervals for indirect effects
tes mellitus on outcomes in patients with colon cancer. J Clin in structural equation models. In: Leinhardt S ed. Sociological
Oncol. 2003;21:433–440. Methodology. San Francisco, CA: Jossey-Bass; 1982:290–312.
10. Jullumstrø E, Kollind M, Lydersen S, Edna TH. Diabe- 27. Baron RM, Kenny DA. The moderator-mediator varia-
tes mellitus and outcomes of colorectal cancer. Acta Oncol. ble distinction in social psychological research: conceptual,
2009;48:361–367. strategic, and statistical considerations. J Pers Soc Psychol.
11. Hayes AF, Rockwood NJ. Regression-based statistical medi- 1986;51:1173–1182.
ation and moderation analysis in clinical research: observa- 28. Edna TH, Karlsen V, Jullumstrø E, Lydersen S. Prevalence of a-
tions, recommendations, and implementation. Behav Res Ther. naemia at diagnosis of colorectal cancer: assessment of associ-
2017;98:39–57. ated risk factors. Hepatogastroenterology. 2012;59:713–716.
12. Aleksandrova K, Schlesinger S, Fedirko V, et al. Metabolic me- 29. Wilson MJ, van Haaren M, Harlaar JJ, et al. Long-term prog-
diators of the association between adult weight gain and colo- nostic value of preoperative anemia in patients with colorec-
rectal cancer: data from the European Prospective Investigation tal cancer: a systematic review and meta-analysis. Surg Oncol.
into Cancer and Nutrition (EPIC) cohort. Am J Epidemiol. 2017;26:96–104.
2017;185:751–764. 30. Thomas MC, MacIsaac RJ, Tsalamandris C, Power D, Jerums
13. Bilimoria KY, Liu Y, Paruch JL, et al. Development and evalua- G. Unrecognized anemia in patients with diabetes: a cross-sec-
tion of the universal ACS NSQIP surgical risk calculator: a deci- tional survey. Diabetes Care. 2003;26:1164–1169.
sion aid and informed consent tool for patients and surgeons. J 31. Astor BC, Muntner P, Levin A, Eustace JA, Coresh J. Association
Am Coll Surg. 2013;217:833–42.e1. of kidney function with anemia: the Third National Health and
14. Dolejs SC, Waters JA, Ceppa EP, Zarzaur BL. Laparoscopic ver- Nutrition Examination Survey (1988-1994). Arch Intern Med.
sus robotic colectomy: a national surgical quality improvement 2002;162:1401–1408.
project analysis. Surg Endosc. 2017;31:2387–2396. 32. Sahay M, Kalra S, Badani R, et al. Diabetes and anemia: Inter-
15. Cohen ME, Bilimoria KY, Ko CY, Hall BL. Development of national Diabetes Federation (IDF) - Southeast Asian Region
an American College of Surgeons National Surgery Qual- (SEAR) position statement. Diabetes Metab Syndr. 2017;11(sup-
ity Improvement Program: morbidity and mortality risk pl 2):S685–S695.
calculator for colorectal surgery. J Am Coll Surg. 2009;208: 33. Aquina CT, Blumberg N, Becerra AZ, et al. Association a-
1009–1016. mong blood transfusion, sepsis, and decreased long-term
16. Egenvall M, Mörner M, Martling A, Gunnarsson U. Prediction survival after colon cancer resection. Ann Surg. 2017;266:
of outcome after curative surgery for colorectal cancer: preop- 311–317.
erative haemoglobin, C-reactive protein and albumin. Colorec- 34. Cata JP, Wang H, Gottumukkala V, Reuben J, Sessler DI. In-
tal Dis. 2018;20:26–34. flammatory response, immunosuppression, and cancer

Copyright © The American Society of Colon & Rectal Surgeons, Inc. Unauthorized reproduction of this article is prohibited.
 DISEASES OF THE COLON & RECTUM VOLUME 63: 3 (2020) 299

r­ ecurrence after perioperative blood transfusions. Br J Anaesth. 37. Yamada T, Shojima N, Noma H, Yamauchi T, Kadowaki T.
2013;110:690–701. Glycemic control, mortality, and hypoglycemia in critically ill
35. Lidder PG, Sanders G, Whitehead E, et al. Pre-operative oral patients: a systematic review and network meta-analysis of ran-
iron supplementation reduces blood transfusion in colorectal domized controlled trials. Intensive Care Med. 2017;43:1–15.
surgery: a prospective, randomised, controlled trial. Ann R Coll 38. Fransgaard T, Thygesen LC, Gögenur I. Increased 30-day mor-
Surg Engl. 2007;89:418–421. tality in patients with diabetes undergoing surgery for colorec-
36. Partridge JS, Harari D, Martin FC, Dhesi JK. The impact of pre- tal cancer. Colorectal Dis. 2016;18:O22–O29.
operative comprehensive geriatric assessment on postoperative 39. Kroenke CH, Neugebauer R, Meyerhardt J, et al. Analysis of
outcomes in older patients undergoing scheduled surgery: a body mass index and mortality in patients with colorectal can-
systematic review. Anaesthesia. 2014;69(suppl 1):8–16. cer using causal diagrams. JAMA Oncol. 2016;2:1137–1145.

Bookmark DC&R on Your Phone

Access mobile-optimized content when you’re on the go!

How to Bookmark on an iPhone: How to Bookmark on an Android:

• Open Safari browser • Go to dcrjournal.com
m
on your browser
• Go to dcrjournal.com
• Tap the Action
• Tap the Action
Overflow on the
Menu icon
Action bar
• Scroll up and select
• Select Add to Home
Add to Home Screen
Screen
• Confirm by tapping
• Tap Add on next
Add at the top right
screen
of next screen
• Touch Add Automatically
• All done!
• All done!

Don’t miss DC&R’s vital content:

bookmark dcrjournal.com today!

9-R010

Copyright © The American Society of Colon & Rectal Surgeons, Inc. Unauthorized reproduction of this article is prohibited.