ORIGINAL ARTICLE

Most of the variation in length of stay in emergency general

surgery is not related to clinical factors of patient care

Aditya Achanta, BS, Ask Nordestgaard, BA, Napaporn Kongkaewpaisan, MD, Kelsey Han, BA,
April Mendoza, MD, MPH, Noelle Saillant, MD, Martin Rosenthal, MD, Peter Fagenholz, MD,
George Velmahos, MD, PhD, and Haytham M.A. Kaafarani, MD, MPH, Boston, Massachusetts

BACKGROUND: Hospital length of stay (LOS) is currently recognized as a key quality indicator. We sought to investigate how much of the LOS
variation in the high-risk group of patients undergoing Emergency general surgery could be explained by clinical versus nonclin-
ical factors.
METHODS: Using the 2007 to 2015 American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) database,
we included all patients who underwent an emergency appendectomy, cholecystectomy, colectomy, small intestine resection,
enterolysis, or hernia repair. American College of Surgeons National Surgical Quality Improvement Program defines emergency
surgery as one that is performed no later than 12 hours after admission or symptom onset. Using all the ACS-NSQIP demo-
graphic, preoperative (comorbidities, laboratory variables), intraoperative (e.g., duration of surgery, wound classification), and
postoperative variables (i.e., complications), we created multivariable linear regression models to predict LOS. LOS was treated
as a continuous variable, and the degree to which the models could explain the variation in LOS for each type of surgery was
measured using the coefficient of determination (R2).
RESULTS: A total of 215,724 patients were included. The mean age was 47.1 years; 52.0% were female. In summary, the median LOS
ranged between 1 day for appendectomies (n = 124, 426) and cholecystectomies (n = 21,699) and 8 days for colectomies
(n = 19,557) and small intestine resections (n = 7,782). The R2 for all clinical factors ranged between 0.28 for cholecystectomy
and 0.44 for hernia repair, suggesting that 56% to 72% of the LOS variation for each of the six procedures studied cannot be ex-
plained by the wide range of clinical factors included in ACS-NSQIP.
CONCLUSION: Most of the LOS variation is not explained by clinical factors and may be explained by nonclinical factors (e.g., logistical delays,
insurance type). Further studies should evaluate these nonclinical factors to identify target areas for quality improvement.
(J Trauma Acute Care Surg. 2019;87: 408–412. Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.)
LEVELS OF EVIDENCE: Epidemiological study, level III.
KEY WORDS: Length of stay; emergency general surgery; clinical variation; ACS-NSQIP.

T he cost of emergency general surgery (EGS) hospitalizations

in 2010 exceeded 28 billion dollars; these costs are only
projected to increase in the near future.1 It is now well established
in surgical literature that EGS accounts for a disproportionately
high rate of morbidity and mortality compared to non-EGS.2
One of the outcomes often measured and benchmarked in
healthcare in general and in EGS in particular is the hospital length
Submitted: November 8, 2018, Revised: March 6, 2019, Accepted: March 20, 2019,
Published online: April 8, 2019.
From the Division of Trauma, Emergency Surgery and Surgical Critical Care (A.A.,
A.N., N.K., K.H., A.M., N.S., M.R., P.F., G.V., H.M.A.K.); Massachusetts General
Hospital; Harvard Medical School (A.A., A.N., N.K., K.H., A.M., N.S., M.R.,
P.F., G.V., H.M.A.K.), Boston, Massachusetts.
Oral presentation of abstract at 104th Annual Meeting of American College of Sur-
geons Clinical Congress, October 21st to 25th, 2018 in Boston, MA.
Supplemental digital content is available for this article. Direct URL citations appear in
the printed text, and links to the digital files are provided in the HTML text of this
article on the journal’s Web site (www.jtrauma.com).
Address for reprints: Haytham M.A. Kaafarani, MD, Division of Trauma, Emergency
Surgery & Surgical Critical Care, Massachusetts General Hospital, Suite 810, 165
Cambridge St, Boston, MA 02114; email: hkaafarani@mgh.harvard.edu.
DOI: 10.1097/TA.0000000000002279
408
of stay (LOS). Length of stay is affected by clinical factors of pa-
tient care (e.g., comorbidities, degree of illness, complications)
and nonclinical, system-related, factors (e.g., availability of rehabil-
itation centers, logistical delays, insurance type).3 A prolonged
LOS often results in an inefficient use of hospital resources, wastes
patient time, and is a significant driver of unnecessary cost.
In a study of colorectal surgery patients with a very pro-
longed LOS, more than 40% of patients did not have complica-
tions that justified their prolonged LOS.4 Similarly, Hwabejire
et al.5 in 2013 found that clinical deterioration explained less than
20% of the prolonged LOS of trauma surgery patients, and that
system-level issues rather than clinical reasons accounted for
most of these extended hospital stays. For these patients, the au-
thors estimated that identifying and mitigating the system-level
factors can lead to estimated cost reductions of more than five
million dollars at their institution.5 Another recent study found
that a significant portion of the variation in extended LOS was
not related to patient comorbidity and severity in elderly patients
with anterior cervical discectomy.6
In this study, we sought to evaluate what proportion of the
variation in LOS for EGS patients is clinical in nature versus
J Trauma Acute Care Surg
Volume 87, Number 2

Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.

J Trauma Acute Care Surg
Volume 87, Number 2
system-related. We hypothesized that most variation in LOS for
EGS patients is not clinical in nature.
METHODS
To study the LOS of EGS patients, the American College
of Surgeons National Surgical Quality Improvement Program
(ACS-NSQIP) database between 2007 and 2015 was used.14
Patient Population
Figure 1 illustrates our patient population selection algo-
rithm. First, all emergency surgery patients were identified using
the ACS-NSQIP “EMERGENCY” variable, defined as “per-
formed within a short interval of time between patient diagnosis
or the onset of related preoperative symptomatology” in the da-
tabase. Then, the six most commonly performed EGS surgical
procedures, identified using their corresponding CPT codes, were
identified as colectomies, cholecystectomies, appendectomies,
small intestine resections, enterolysis, and hernia repairs. Pa-
tients with missing key variables (e.g., race, sex), those who
died, those transferred from another hospital, and those not
discharged home or to rehabilitation centers were excluded. To
better represent the majority of the patients in our models and
exclude significant outliers that would skew the model, we ex-
cluded the patients whose LOS was over the 99th percentile for
each type of procedure.
Figure 1. Exclusion criteria for selecting patient population.
© 2019 Wolters Kluwer Health, Inc. All rights reserved.
Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.
Achanta et al.
Study Variables
For each of the six EGS procedures, the mean LOS, median
LOS, standard deviation of LOS, minimum LOS, maximum LOS,
and quartiles for LOS were calculated. A multivariable linear re-
gression was performed with total hospital LOS as the dependent
variable, and all the demographic (e.g., age, sex), preoperative
(comorbidities, laboratory values), intraoperative (duration of sur-
gery and wound classification), and postoperative variables (com-
plications) in ACS-NSQIP as the independent variables. Missing
comorbidities or laboratory values were assumed to be normal.
Study Endpoint
For each linear regression model that utilized all existing
variables, the R2 was calculated. By using the R2, we are able to
understand how much of the variation in LOS is explained by
clinical factors, which are included in the model, and nonclinical
factors. Analyzing LOS as a continuous variable in our linear re-
gression models allowed us to factor in the variation in LOS in all
patients, including those with brief hospital LOS, unlike most
existing literature that focused exclusively on prolonged LOS.
Preoperative, Intraoperative, and Postoperative
Variables Partial Contribution to LOS
We also aimed to separate and compare the relative impact
of demographic, preoperative, intraoperative, and postoperative
variables on LOS, as all of these categories of variables are
highly sequentially related and correlated. To achieve that, we
409
 J Trauma Acute Care Surg
Achanta et al. Volume 87, Number 2

calculated the partial R2 for each category. The partial R2 allows

TABLE 1. Preoperative Demographic, Comorbidity, and
for calculating the individual explanatory impact of each in-
Laboratory Value Data for Patient Population
dependent variable, controlling for all other variables in the
model, with the dependent variable remaining LOS.7 We then N 213,808
summed the individual explanatory power (partial R2) of the Demographic variables
variables per category (demographic, preoperative, intraopera- Age 47.1
tive, and postoperative) to understand the relative impact of each Female, n (%) 111,418 (52%)
category on LOS. White, n (%) 159,921 (75%)
Black, n (%) 20,718 (10%)
Ethical Oversight
Asian, n (%) 10,748 (5%)
An ethical oversight study application was submitted then
Native, n (%) 1,619 (1%)
exempted by our institutional review board.
Comorbid variables
Functional dependence, n (%) 7,907 (4%)
RESULTS Ventilator dependence, n (%) 1,297 (1%)
A total of 213,808 patients who underwent one of the six Low BMI, n (%) 9,136 (4%)
EGS procedures were included. Table 1 lists the preoperative char- Diabetes, n (%) 18,157 (8%)
acteristics of the patient population used in this study. In summary, Dyspnea, n (%) 7,669 (4%)
the mean age was 47.1 years; 52% were female, 75% were white, Smoking history, n (%) 42,115 (20%)
28% had hypertension, and 33% had Systemic Inflammatory COPD, n (%) 6,440 (3%)
Response Syndrome (SIRS) or sepsis preoperatively. Ascites, n (%) 3,157 (2%)
Table 2 shows the number of patients, the mean LOS, the Congestive heart failure, n (%) 1,301 (1%)
standard deviation of the LOS, and the maximum and minimum Hypertension, n (%) 60,120 (28%)
LOS for each EGS procedure. In summary, the median LOS ranged Renal failure, n (%) 1,376 (1%)
between 1 day for appendectomies (n = 124, 426) and cholecys- Dialysis, n (%) 1,714 (1%)
tectomies (n = 21,699) and 8 days for colectomies (n = 19,557) Disseminated cancer, n (%) 2,727 (1%)
and small intestine resections (n = 7,782). Wound infections, n (%) 2,179 (1%)
Table 3 shows the results of the multivariable linear regres- Weight loss, n (%) 2,464 (1%)
sion analyses, with many of the clinical variables that affect patient Steroids, n (%) 5,886 (3%)
care (demographic, preoperative, intraoperative, and postoperative Bleeding disorder, n (%) 9,549 (4%)
factors) as the independent variables and the LOS as the dependent RBC transfusion, n (%) 2,045 (1%)
variable. The R2 for all clinical factors ranged between 0.28 for SIRS/sepsis, n (%) 70,918 (33%)
cholecystectomy and 0.44 for all hernia repair, suggesting that Laboratory values
56% to 72% of the LOS variation for the six included procedures Albumin <3.0, n (%) 12,907 (6%)
is not due to the clinical factors of patient care (Table 3, and Alkaline phosphatase >125, n (%) 15,626 (7%)
Supplemental Digital Content, Appendix A: Table 5, http:// Blood urea nitrogen >40, n (%) 5,637 (3%)
links.lww.com/TA/B368). Creatinine >1.2, n (%) 22,358 (10%)
Table 4 explores the relative contribution of preoperative, International Normalized Ratio > 1.5, n (%) 4,740 (2%)
intraoperative and postoperative factors to the variation in LOS Platelets <150, n (%) 13,699 (6%)
using partial R2 analysis. In summary, within the variation of LOS Serum glutamic-oxaloacetic transaminase >40, n (%) 21,214 (10%)
that could be explained by clinical factors, postoperative factors Sodium >145, n (%) 1,469 (1%)
explained the most variation for colectomies, small intestine re- White blood cell <4.5, n (%) 5,302 (2%)
section, enterolysis, and all hernia repairs (Table 4). For cholecys- White blood cell 15–25, n (%) 53,706 (25%)
tectomies, preoperative variables explained most of the variation, White blood cell > or = 25, n (%) 3,461 (2%)
and for appendectomies, intraoperative variables explained most
of the variation (Table 4).
patient hospital LOS is due to system-related issues such as wait
DISCUSSION times to placement in rehabilitation facilities, operational and lo-
gistical delays hospital processes, and/or health insurance and
In our population-based study of EGS patients, we found payer-related difficulties.4,5,8 One of the advantages of our anal-
that only about 28% to 44% of the variation in LOS is related to yses, in addition to focusing on the high-risk and often under-
the patient's clinical situation, and that the clinically related var- served EGS patient population, is analyzing the LOS variation
iability for certain EGS procedures is often best explained by the as a continuous variable without an artificial break of the variable
postoperative course rather than predetermined preoperative and into “extended” and “not-extended” LOS of prior literature. This
intraoperative factors. To the best of our knowledge, this is the allowed detection of small but cumulatively significant variations
first study to explore the variation of LOS in EGS patients, with- in LOS across the spectrum of EGS patients with different ex-
out specific focus on extended LOS, and to suggest that most pected LOS. For example, even small variations in LOS for stan-
variation is not clinical in etiology but might be system-related. dardized, high-volume procedures such as appendectomies can
Several studies have recently emerged in support of our cumulatively lead to significant increases in cost and can impact
finding and suggested that a large degree of the variation in the overall availability of hospital beds. Therefore, this variation

410 © 2019 Wolters Kluwer Health, Inc. All rights reserved.

Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.

J Trauma Acute Care Surg
Volume 87, Number 2 Achanta et al.

TABLE 2. LOS Descriptive Statistics by EGS Operation Type

Appendectomies Cholecystectomies Colectomies Small Intestine Resections Enterolysis Hernia Repairs
n 124,426 21,699 19,557 7,782 15,032 25,312
LOS mean, d 1.8 1.8 10.6 10.2 9.8 4.3
LOS standard deviation 1.9 1.6 7.5 7.9 9.5 4.2
LOS, minimum 0 0 0 0 0 0
LOS, maximum 13 12 52 58 49 28
LOS P25 1 1 6 5 5 1
LOS median 1 1 8 8 7 3
LOS p75 2 2 13 12 11 6
LOS p99 10 9 42 44 36 21

is still important to capture, understand, and address from the
system's perspective, in addition to examining in depth the pa-
tients with very prolonged LOS.
The fact that more of the variation in LOS is explained by
nonclinical factors is of great importance, but perhaps not sur-
prising to many clinicians. Hwabejire et al.5 previously reported
that clinical deterioration was the cause of an extended LOS in
only 20% of trauma patients. Similarly, Krell et al.4 found that,
for 43% of surgery patients with an extended LOS, there was
no recorded postoperative complications that explain the pro-
longed LOS. In both aforementioned studies, the findings could
theoretically be due to differences in the type of procedures an-
alyzed (e.g., EGS vs. non-EGS) and due to studying LOS as a
binary variable versus a continuous variable.4,5 Still, this key
finding in our study, namely, that only a small percentage of
the variation in LOS is clinical in etiology, highlights two impor-
tant themes: (1) the role that inefficiencies and inequity in the
healthcare system might be playing in LOS variation within hos-
pitals, and more importantly across different hospitals; and (2)
the wisdom or controversy of using LOS as a quality metric
for benchmarking clinical care. Prior studies have suggested that
LOS is too influenced by nonclinical factors, such as insurance
type and discharge destination, to be an appropriate quality mea-
sure.9 By the same rationale, others have argued that it is exactly
that characteristic that makes it a measure of quality perfor-
mance at the system level. We tend to agree that it is important
to look in more depth at the underlying nonclinical drivers of
the variation in LOS before fully endorsing LOS as a quality
measure at least at the individual clinician, department or even
hospital level.9 Because our results suggest that nonclinical
drivers are key to understand LOS, it is essential to explore the
accountability of hospitals versus regions and states before utilizing
LOS as an indicator of hospital or department performance. For
example, the availability of sufficient rehabilitation beds and high
rates of underserved patients with difficult discharge destinations
may be some of the drivers of increased LOS, with both having
more to do with societal support structures at large rather than fac-
tors related to the patient's care. The hospital in question has prob-
ably little influence on these societal support factors, and as such,
LOS might not be reflective of the quality of care provided.
Alternatively, with 56% to 72% of the variation in LOS of
EGS patients being not related to the clinical situation and clin-
ical care, a quality improvement opportunity arises. Some of the
system-related factors might be difficult to change, but still nec-
essary to tackle at the hospital and provider-level, such as creat-
ing social support systems for discharged patients, streamlining
the payer-related negotiations, and optimizing staffing models
for EGS.10 Standardizing the care processes to decrease the
number of complications can also lead to substantial decreases
in LOS variation, as most of the explained clinical LOS variation
is postoperative in nature.
Some policy experts, clinicians and researchers argue that
decreasing the LOS might result in an automatic increase in re-
admissions. However, it is becoming clearer that the two are not
necessarily inversely correlated.11 It is possible to decrease var-
iation in surgical LOS and remove inefficiencies through the

TABLE 3. Linear Regression Results by Operation Type

Appendectomies Cholecystectomy Colectomies Small Intestine Resection Enterolysis All Hernia Repairs
2
R 0.37 0.28 0.42 0.43 0.41 0.44
Adjusted R2 0.37 0.28 0.41 0.42 0.41 0.44
Root MSE 1.49 1.39 5.72 5.99 5.3 3.17
Independent variables that were included in this OLS regression are: preoperative variables: age, sex, race, ASA, functional dependence status, ventilator dependence status, BMI, diabetes,
dyspnea, history of smoking, COPD< ascites, CHF, hypertension, renal failure, dialysis, disseminated cancer, preoperative wound infections, history of weight loss, steroid use, bleeding dis-
order, if they had a preoperative RBC transfusion, SIRS/sepsis, laboratory values (albumin, alkaline phosphatase, BUN, creatinine, INR, platelet count, SGOT, Na + values.
Intraoperative variables: wound classification, length of operation.
Postoperative variables: infections, pneumonia, pulmonary embolism, renal insufficiency, acute renal failure, UTIs, wound disruptions, unplanned intubations, if on ventilator longer than
48 hours, if CPR was indicated, myocardial infarction, C. diff infection, DVT, postoperative SIRS/sepsis, if they were comatose for greater than 24 hours.
Dependent variable: LOS.
CPR = Cardiopulmonary Resuscitation; UTI = Urinary Tract Infection; DVT = Deep Vein Thrombosis; SGOT = Serum Glutamic Oxaloacetic Transaminase; BUN = Blood Urea Nitrogen;
COPD = Chronic Obstructive Pulmonary Disease.

© 2019 Wolters Kluwer Health, Inc. All rights reserved. 411

Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.


Achanta et al.
TABLE 4. Partial R2 Analyses Results by Operation Type
Appendectomies Cholecystectomy
2
Total partial R 0.19 0.15
Demographics—partial R2 0.02 0.01
Preoperative—partial R2 0.03 0.07
Intraoperative—partial R2 0.09 0.04
Postoperative—partial R2 0.05 0.04
implementation of programs that standardize surgical care and
improve recovery times without necessarily increasing readmis-
sion rates. For example, a decreasing reimbursement rate for
knee and hip replacements in the United Kingdom decreased pa-
tient LOS dramatically; the average LOS for a total knee replace-
ment fell from 16 days to 5.4 days over 17 years.12 The decreased
LOS for patients was achieved by making perioperative care more
standardized and by implementing enhanced recovery programs.12
As LOS decreased with the implementation of the enhanced re-
covery programs for total knee arthroplasty and total hip arthroplasty
from 2004 to 2008, there was no change in readmissions.13
Limitations
Our study has several limitations. First, our analyses are
limited to the variables contained in the ACS-NSQIP database
and the accuracy of the values recorded in this database. There
could be other clinical variables that impact LOS that are simply
not captured. For example, the patient demographics and clinical
factors clustering effects due to intrahospital versus interhospital
variation could not be analyzed. Second, we have assumed nor-
mality for the missing laboratory data, which is plausible clini-
cally as clinicians typically order tests for any level of suspicion
of abnormality, but we acknowledge that this might have af-
fected our findings to a certain small but unknown effect. Fur-
thermore, ACS-NSQIP also does not record the nonclinical
factors that affect patient care including socioeconomic factors,
such as insurance status, and hospital system-related factors;
therefore, our study lacks the granularity to identify the nonclin-
ical factors of patient care that explain LOS variation. Finally,
the Affordable Care Act, which may affect the nonclinical fac-
tors of patient care, was introduced during the timeframe of this
retrospective study but the ACA's effects were not measured or
controlled for.
CONCLUSION
Most of the LOS variation in EGS is not explained by com-
monly recorded clinical factors and might be system-related.
Further studies should evaluate these system-related factors to
identify target areas for quality improvement. Interventions that
address multiple aspects of the care of the EGS patients, such as
standardized care pathways, payer processes navigation, and
provision of alternative social support structures for postdischarge
care are most likely to be impactful.
412
Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.
J Trauma Acute Care Surg
Volume 87, Number 2
Colectomies Small Intestine Resection Enterolysis All Hernia Repairs
0.19 0.23 0.20 0.26
0.007 0.002 0.01 0.03
0.03 0.05 0.03 0.05
0.008 0.03 0.001 0.09
0.15 0.16 0.17 0.1
AUTHORSHIP
A.A. participated in the literature search. A.A., A.N., G.V., H.M.A.K. partic-
ipated in the study design. A.A., A.N., N.K. participated in the data collec-
tion. A.A., A.N., H.M.A.K. participated in the data analysis. A.A., H.M.A.K.
participated in the data interpretation. A.A., H.M.A.K. participated in the
writing. A.A., A.N., N.K., K.H., A.M., N.S., M.R., P.F., G.V., H.M.A.K. partic-
ipated in the critical revision.
DISCLOSURES
The authors declare no funding or conflicts of interest.
REFERENCES
1. Ogola GO, Gale SC, Haider A, Shafi S. The financial burden of emergency
general surgery. J Trauma Acute Care Surg. 2015;79:444–448.
2. Havens JM, Peetz AB, Do WS, Cooper Z, Kelly E, Askari R, Reznor G,
Salim A. The excess morbidity and mortality of emergency general surgery.
J Trauma Acute Care Surg. 2015;78:306–311.
3. Shojania KG, Showstack J, Wachter RM. Assessing Hospital Quality: A Re-
view For Clinicians. Available at: http://ecp.acponline.org/marapr01/
shojania.htm. Updated March 2001. Accessed March 01, 2018.
4. Krell RW, Girotti ME, Dimick JB. Extended length of stay after surgery.
JAMA Surg. 2014;149:815.
5. Hwabejire JO, Kaafarani HM, Imam AM, Solis CV, Verge J, Sullivan NM,
DeMoya MA, Alam HB, Velmahos GC. Excessively long hospital stays after
trauma are not related to the severity of illness: let's aim to the right target!
JAMA Surg. 2013;148:956.
6. Adogwa O, Lilly DT, Vuong VD, Desai SA, Ouyang B, Khalid S, Khanna R,
Bagley CA, Cheng J. Extended length of stay in elderly patients after anterior
cervical discectomy and fusion is not attributable to baseline illness severity
or post-operative complications. World Neurosurg. 2018;115:e552–e557.
7. Allen MP. Normal equations and partial regression coefficients. Understand-
ing Regression Analysis. 1997;81–85.
8. Bashkin O, Caspi S, Haligoa R, Mizrahi S, Stalnikowicz R. Organizational
factors affecting length of stay in the emergency department: initial observa-
tional study. Isr J Health Policy Res. 2015;4:38.
9. Brasel KJ. Length of stay. Arch Surg. 2007;142:461.
10. Columbus AB, Morris MA, Lilley EJ, Harlow AF, Haider AH, Salim A,
Havens JM. Critical differences between elective and emergency surgery:
identifying domains for quality improvement in emergency general surgery.
Surgery. 2018;163:832–838.
11. Desharnais S, Hogan AJ, Mcmahon LF Jr., Fleming S. Changes in rates of
unscheduled hospital readmissions and changes in efficiency following the
introduction of the Medicare prospective payment system. An analysis using
risk-adjusted data. Eval Health Prof. 1991;14:228–252.
12. Burn E, Edwards CJ, Murray DW, Silman A, Cooper C, Arden NK,
Pinedo-Villanueva R, Prieto-Alhambra D. Trends and determinants of length
of stay and hospital reimbursement following knee and hip replacement: ev-
idence from linked primary care and NHS hospital records from 1997 to
2014. BMJ Open. 2018;8:e019146.
13. Husted H, Otte KS, Kristensen BB, Orsnes T, Kehlet H. Readmissions after fast-
track hip and knee arthroplasty. Arch Orthop Trauma Surg. 2010;130:1185–1191.
14. [Dataset] American College of Surgeons, National Quality Surgical Im-
provement Program, 2007-2015.
© 2019 Wolters Kluwer Health, Inc. All rights reserved.