Association of Hospital Length of Stay and Complications With Readmission After Open Pancreaticoduodenectomy

Abstract

This study uses data from the American College of Surgeons National Quality Improvement Program databases to evaluate associations between hospital length of stay and postoperative complications with hospital readmission among patients who underwent open pancreaticoduodenectomy.

New reimbursement policies under the Affordable Care Act Hospital Readmission Reduction Program place hospitals under increasing pressure to reduce costs by accelerating care and discharge after surgery.¹ However, premature discharge may be associated with subsequent complications and readmission.² We used a national surgical database to investigate possible predictive factors for and the association of postsurgery hospitalization duration with readmission after open pancreaticoduodenectomy.

Methods

The American College of Surgeons National Quality Improvement Program and targeted hepatopancreaticobiliary databases were queried to identify patients who underwent open pancreaticoduodenectomy (Current Procedural Terminology codes 48150 and 48153) between January 1, 2014, and December 31, 2015. Patients with a hospital length of stay (LOS) longer than 14 days, who constituted 1052 (16.9%) of the original cohort of 6226, were excluded to overcome underestimation of readmission rates caused by immortal time bias in the database’s documentation of this variable.³ Data were collected and analyzed between August 1, 2017, and April 28, 2018. The University of California, Los Angeles Investigational Review Board deemed this study exempt from review because the data were obtained from a publicly available deidentified database organized by the American College of Surgeons.

Readmission was modeled using forward stepwise multivariate logistic regression. Interaction variables with centering around the median were incorporated to assess for interaction effect. Propensity score matching with an inverse probability of treatment weighted approach was used to evaluate associations with LOS and adjust for cofounders.⁴ Groups were stratified into long and short LOS groups by using the median as the divider. Statistical analyses were performed using R, version 3.4.3 (R Foundation). Tests of significance were 2-sided, and P < .05 was considered significant.

Results

Within our cohort of 5174 patients, 930 were readmitted. Of those 930 patients, the median (IQR) age was 65 (56-72) years and 518 (55.7%) were male. Data on race/ethnicity were not collected. The median (interquartile range [IQR]) LOS was 8 (7-10) days and the median time from discharge to readmission was 6 (4-12) days. Demographic characteristics, comorbidities, intraoperative factors, and index complications were compared between readmitted and nonreadmitted groups (Table). Readmitted patients were more likely to be male and to have a higher body mass index and index stays complicated by fistula or delayed gastric emptying (DGE). Under multivariate modeling, index grade A fistula (defined as prolonged drainage of amylase-rich fluid) (odds ratio [OR], 1.65; 95% CI, 1.27-2.02; P < .001), grade B or C fistula (defined as fistula requiring percutaneous drainage or reoperation) (OR, 2.00; 95% CI, 1.13-2.86; P = .02), and DGE (OR, 2.77; 95% CI, 2.10 to 3.43; P < .001) were the only factors independently associated with readmission.

Table. Patient Characteristics, Perioperative Features, and Index Complication Rates of Readmitted and Nonreadmitted Patients After Open Pancreaticoduodenectomy^a.

Characteristic	Not Readmitted (n = 4244)	Readmitted (n = 930)	P Value
Demographic Features
Age, median (IQR), y	65 (57-72)	65 (56-72)	.65
Male	2198 (51.8)	518 (55.7)	.03
Comorbidities
BMI, median (IQR)	26 (23.2-29.9)	27 (23.4,-30.8)	.004
Preoperative albumin, median (IQR), g/dL	3.9 (3.4-4.2)	3.9 (3.4-4.2)	.37
Diabetes	1060 (25.0)	245 (26.3)	.38
Smoking history	825 (19.4)	190 (20.4)	.49
Disseminated cancer	172 (4.1)	45 (4.8)	.28
Weight loss	761 (17.9)	151 (16.2)	.22
Corticosteroid use	94 (2.2)	20 (2.2)	.90
Perioperative Factors
Neoadjuvant radiotherapy	304 (7.2)	83 (9.0)	.07
Neoadjuvant chemotherapy	684 (16.2)	141 (15.3)	.47
Vascular resection	740 (17.7)	186 (20.3)	.07
Operative time, median (IQR), min	348 (272.5-431.5)	355 (275-432)	.25
ASA physical status, median (IQR), scoreb	3 (3-3)	3 (3-3)	.77
LOS, median (IQR), d	8 (6-10)	8 (7-10)	.20
Index Complicationsc
Fistulad
Any	403 (9.6)	146 (15.8)	<.001
Grade A	333 (7.9)	120 (13.0)	<.001
Grade B or C	70 (1.7)	26 (2.8)	.03
Delayed gastric emptying	175 (4.2)	64 (7.1)	<.001
Surgical site infection
Superficial	171 (4.1)	39 (4.2)	.82
Deep incisional	17 (0.4)	3 (0.3)	>.99
Organ space	104 (2.5)	20 (2.2)	.59
Wound dehiscence	7 (0.2)	2 (0.2)	.74
Sepsis	95 (2.2)	21 (2.3)	.97
Renal insufficiency or failure	7 (0.2)	2 (0.2)	.74
Reoperation	44 (1.0)	10 (1.1)	.86
Pneumonia	53 (1.2)	17 (1.8)	.17
Thromboembolism	40 (0.9)	13 (1.4)	.21
Urinary tract infection	69 (1.6)	12 (1.3)	.56
Myocardial infarction	17 (0.4)	5 (0.5)	.58

Abbreviations: ASA, American Society of Anesthesiologists; BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); IQR, interquartile range; LOS, length of stay.

SI conversion factor: To convert albumin to grams per liter, multiply by 10.

^aData are expressed as number (percentage) unless otherwise specified.

^bThe ASA score ranges from 1 to 6, with higher scores indicating greater morbidity.

^cPercutaneous drainage was not included in this analysis because the American College of Surgeons National Quality Improvement Program does not document whether procedures were performed during the index hospitalization period.

^dFistulas are graded by the database as A (prolonged drainage of amylase-rich fluid), B (clinically significant fistula requiring percutaneous drainage), or C (clinically significant fistula requiring reoperation).

Although an LOS of 8 days or longer was associated with higher rates of index complications compared with an LOS less than 8 days (76% vs 24%; P < .001), propensity score matching demonstrated no difference in the probability of readmission (OR, 0.97; 95% CI, 0.82-1.12; P = .63) or time to readmission (mean difference, 0.01 days; 95% CI, −0.40 to 0.42; P = .93). Similar findings were observed in the subset of patients with uncomplicated index stays (probability of readmission: OR, 1.06; 95% CI, 0.87 to 1.24; P = .70; time to readmission: mean difference, 0.16 days; 95% CI, −0.32 to 0.63; P = .50).

Interaction analysis was performed to assess the association of LOS with fistula and DGE as factors associated with readmission. Each 1-day increase in LOS decreased the odds of readmission after a grade B or C fistula by a factor of 0.78 (95% CI, 0.66-0.93; P = .004) and after DGE by a factor of 0.70 (95% CI, 0.61-0.80; P < .001). Moreover, within these subsets of patients, those who stayed 8 days or less were readmitted at higher rates than those who stayed longer (grade B or C fistula: 48% vs 21%; P < .001; DGE: 73% vs 30%; P < .001).

Discussion

This study draws on the large, high-quality National Quality Improvement Program database to examine the factors associated with readmission after open pancreaticoduodenectomy. On propensity score matching analysis, LOS alone was not directly associated with readmission, neither in all patients nor in the subset who experienced no index complications. However, patients with an index grade B or C fistula or DGE, previously shown to be associated with readmission,^5,6 were more likely to be readmitted if their subsequent hospital stay was shorter. These findings support the general rationale for enhanced recovery after surgery after open pancreaticoduodenectomy. In addition, they further suggest that “expedient” discharge of patients who experienced clinically significant complications may lead to increased costs owing to a higher rate of readmission. Limitations to our study include limited information regarding long-term follow-up, cost, and stratification based on institution type or hospital volume. There are also constraints inherent in the database we used such as potential variability in the interpretation of patient data by those who compile the database. These findings emphasize that proper discharge planning and medical support after discharge are of utmost importance.