Association between alcohol use disorder and hospital outcomes in colectomy patients - A retrospective cohort study

Abstract

Study Objective:

In the United States, alcohol use disorder adversely affects 5.6% of all adults. Excessive alcohol consumption adversely affects organ functions critical for adaptation to stress induced by surgery. Colorectal resection is one of the most common major surgeries in patients at risk for alcohol use disorder. The objective of this study was to assess the impact of alcohol use disorder on hospital outcomes after colectomy using a population-based discharge database.

Setting:

A population-based discharge database.

Patients:

The Premier Healthcare Database was queried for the 603,730 adult patients who underwent colectomy from 2016-2019.

Interventions:

None.

Measurements:

Multiple logistic regressions estimated the associations between in-hospital mortality, length of stay, and hospitalization cost with alcohol use disorder as the primary predictor, adjusting for other substance use disorders, psychotic disorders, depression, other Elixhauser comorbidities, age, payor, race, gender, non-elective surgery, and other unbalanced variables.

Main Results:

A discharge code for alcohol use disorder was identified in 2.9% of colectomy patients and the overall in-hospital mortality rate in all sampled these colectomy patients was 1.4%. Alcohol use disorder was associated with a significantly increased risk of in-hospital mortality after adjusting for other factors (AOR 1.36, 95% CI 1.24-1.48, p < 0.0001). Alcohol use disorder was also significantly associated with long length of stay (AOR 1.45, 95% CI 1.39-1.52, p < 0.0001) and high hospitalization costs (AOR 1.63, 95% CI 1.56-1.70, p < 0.0001).

Conclusions:

Alcohol use disorder is associated with an increased risk of in-hospital mortality in patients undergoing colectomy, one of the most common major surgeries. Future research should examine if enhanced efforts to identify patients with alcohol use disorder could enable anesthesiologists to provide worthwhile perioperative interventions for this high-risk population.

1. Introduction

Alcohol consumption adversely affects 26-28% of hospitalized patients and contributes to a loss of 132.6 million disability-adjusted life years and 5.3% of worldwide deaths each year [1, 2]. As alcohol use has substantial health and economic impact, much attention has been directed towards the numerous clinical complications in patients with unhealthy alcohol use. Alcohol use disorder (AUD) is a DSM-V medical diagnosis that integrates the two former diagnoses of alcohol abuse and alcohol dependence [3]. AUD itself is a common and modifiable risk factor for surgery. It is well documented that increasing alcohol consumption adversely affects organ functions critical to mitigating perioperative stress [4–7]. Still, AUD is frequently overlooked in the design of perioperative care plans despite its significant clinical impact [8, 9].

Colorectal resection is one of the most common yet most expensive surgeries in the United States (US), with almost 300,000 cases in 2018 performed at individual costs of up to $25,000 [10, 11]. Further, colorectal surgery is usually major surgery, and as such, is associated with higher morbidity and mortality when compared to many other surgical subspecialties [12]. The postoperative mortality of colorectal surgery is extensively reported in the literature but ranges widely from 1-16%, likely due to the various diagnoses, urgency status, and associated comorbidities [13, 14]. Apart from being one of the most expensive and frequent major surgeries performed, colectomy is often performed in the setting of malignancy [15]. Colorectal cancer is the third most common cancer world-wide, accounts for the fourth most frequent cause of cancer-related death in the United States and has been extensively linked to alcohol consumption [16, 17].

Prior work regarding the association between alcohol and clinical outcomes following colorectal surgery is conflicting [18–22]. With increasing prevalence of AUD and the anesthesiologist’s role expanding into perioperative medicine [2, 23], early identification and mitigation of risks associated with substance use disorders (SUDs) is a promising opportunity for our specialty. Hence, the objective of this study was to contribute to perioperative optimization of colectomy patients by assessing the association of AUD with clinically meaningful outcomes, specifically, in-hospital mortality after colectomy via analysis of a large, US hospital-based dataset [24]. We hypothesized that AUD would be associated with increased in-hospital mortality for patients undergoing colectomy when adjusted for other relevant comorbidities and demographic characteristics. If proven as a persistent risk factor for adverse outcomes, the case for future research to develop perioperative interventions targeted to patients with AUD would be strengthened.

2. Materials and Methods

2.1. Study Design

The Institutional Review Board determined this study exempt as it was based on a fully de-identified dataset and, therefore, does not constitute human subject research from a regulatory perspective. Discharge claims-based outcomes and associated variables for colectomy patients were extracted from the Premier Healthcare database (Fig. 1). The study is reported as a retrospective cohort study according to Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [25].

Figure 1.

Flowchart summarizing data extraction and inclusion and exclusion criteria for the study.

2.2. Data Source and Study Sample

The Premier Healthcare database is a prospective compilation of de-identified data from approximately 25% of all US hospital discharges and contains patient data from standard hospital discharge files, including demographics, diseases, medications, laboratory tests, and other diagnostic and therapeutic services. It also includes hospital characteristics such as geographic location, bed size, and teaching status [24]. Retrospective de-identified cohort data were obtained for hospitalizations between October 1, 2016 and December 31, 2019. Data from participants at least 18 years or older who underwent colectomy were extracted by the Premier Applied Sciences research team. Because the Premier Healthcare Database consists of data on operational charges, missing charges are not available. However, for most demographic and diagnostic elements, less than 0.01% have missing data [24]. Discharge entries with duplicate hospital encounters were reduced to a singular event by date (Fig. 1). Disease states and comorbidities were classified by the International Classification of Diseases, Tenth Revision (ICD-10), whereas therapeutic and diagnostic procedures were defined by Current Procedural Terminology or Healthcare Common Procedure Coding System.

2.3. Outcomes and Independent Variables

In-hospital mortality was the primary outcome compared between patients with and without a discharge code for AUD. Secondary outcomes included length of stay (LOS) and the total cost of the index hospitalization. LOS was defined as “long LOS” if it exceeded the median two-day stay. Cost of hospitalization was defined as “high cost” if greater than the median of $5,988.

Comorbidities were categorized based on the Elixhauser Comorbidity Index and defined according to ICD-10 codes found in the administrative data. The Elixhauser Comorbidity Index is a well-established measurement tool originally developed to improve comorbidity categorization and predict in-hospital mortality, LOS, and hospital charges [26, 27]. The weighted Elixhauser comorbidity system synthesizes the 31 binary comorbidity variables into a cumulative score. In addition to our primary independent variable AUD, secondary independent variables of interest were other SUDs, depression, and psychotic disorders ascertained via the Elixhauser Index. Given that co-occurring substance use and mental health disorders are common [28, 29], the four aforementioned variables were reported separately from the Elixhauser Index score for greater discriminative ability, and the remaining 27 cumulative comorbidities were designated Elixhauser, reduced. ICD-10 and procedural codes used in this study are provided in Appendix Table A.1. Within each comorbidity, having one or multiple of these codes could have indicated a positive.

2.4. Statistical Analyses

Descriptive statistics including means, medians, standard deviations, and percentages, were reported for variables describing the study sample. We evaluated separate associations between in-hospital mortality and each characteristic using independent t-tests and chi-square tests as appropriate. Elixhauser score and weighted Elixhauser score deviated appreciably from normality and therefore, mean ranks were compared by in-hospital mortality (yes/no) via Kruskal Wallis.

Subsequently, we used multiple logistic regression models to estimate the adjusted odds ratios and 95% confidence intervals of in-hospital mortality, comparing those with AUD discharge diagnosis codes and those without. A priori we planned to adjust analyses for other SUDs, psychotic disorders, depression as well as age, non-elective surgery, and the reduced Elixhauser score as in Elixhauser et al. [26]. Additionally, other characteristics reported in Table 1 that differed by in-hospital mortality were considered for inclusion as covariates. Significance level was specified a priori at p<0.05. Clinical significance was determined post-hoc as the adjusted odds for mortality to change by at least 10% based on a given risk factor. Similarly, multiple logistic regression models were generated for long LOS (yes/no) and high total hospital cost (yes/no) as secondary outcomes of interest. Using the methodology proposed by VanderWeele and Ding for sensitivity analyses of unmeasured confounding in observational studies [30], the magnitude of confounder associations that could produce confounding bias equal to the observed outcome-AUD association was estimated for each model [31].

Table 1.

Patient and procedural characteristics overall and stratified by in-hospital mortality

Characteristic	All, N = 603730 (%)	Alive, N = 595404 (%)	Dead, N = 8326 (%)	Test (DF)	P value*
Age (SD), years	62.8 (13.6)	62.7 (13.6)	70.5 (13.1)	−52.0 (603728)+	<0.001
Gender				9.6 (1)¶	0.002
Woman	311196 (51.5)	307046 (51.6)	4150 (49.8)
Race				20.1 (3)¶	<0.0001
White	476827 (79.0)	470328 (79.0)	6499 (78.1)
Black	59532 (9.9)	58628 (9.8)	904 (10.9)
Other	47517 (7.9)	46911 (7.9)	606 (7.3)
Unable to determine	19854 (3.3)	19537 (3.3)	317 (3.8)
Hispanic	38267 (6.3)	37816 (6.4)	451 (5.4)	10.0 (3)¶	0.002
Payor				2312.2 (6)¶	<0.0001
Medicare	300363 (49.8)	294250 (49.4)	6113 (73.4)
Medicaid	52657 (8.7)	51943 (8.7)	714 (8.6)
Managed Care	170072 (28.2)	169360 (28.4)	712 (8.6)
Commercial Indemnity	49859 (8.3)	49530 (8.3)	329 (4.0)
Charity/Indigent	2979 (0.49)	2949 (0.50)	30 (0.36)
Self-Pay	10990 (1.8)	10791 (1.8)	199 (2.4)
Other	16810 (2.8)	16581 (2.8)	229 (2.8)
Location				34.4 (1)¶	<0.0001
Rural	82647 (13.7)	81690 (13.7)	957 (11.5)
Urban	521083 (86.3)	513714 (86.3)	7369 (88.5)
Teaching Hospital	253236 (41.9)	248773 (41.8)	4463 (53.6)	471.2 (1)¶	<0.0001
Procedure				10769.9 (3)¶	<0.0001
Emergent	167395 (27.7)	161388 (27.1)	6007 (72.1)
Urgent	35524 (5.9)	34291 (5.8)	1233 (14.8)
Elective	373950 (61.9)	372931 (62.6)	1019 (12.2)
Unknown	26861 (4.4)	26794 (4.5)	67 (0.80)
Comorbidity
Alcohol Use Disorder	17308 (2.9)	16669 (2.8)	639 (7.7)	700.8 (1)¶	<0.0001
Substance Use Disorder	11562 (1.9)	11292 (1.9)	270 (3.2)	79.2 (1)¶	<0.0001
Psychoses	3908 (0.65)	3774 (0.63)	134 (1.6)	121.5 (1)¶	<0.0001
Depression	60208 (10.0)	59020 (9.9)	1188 (14.3)	173.5 (1)¶	<0.0001
Elixhauser Score
Mean (SD) Median	2.5 (2.3) 2.0	2.5 (2.3) 2.0	6.3 (2.6) 6.0	13427.1 (1)#	<0.0001
Weighted Elixhauser
Mean (SD) Median	4.7 (8.1) 0	4.5 (7.8) 0	20.5 (10.8) 20.0	16085.2 (1)#	<0.0001

^*Significance at p<0.05.

⁺Independent t-test assuming unequal variances.

^¶Chi-square test.

^#Kruskal-Wallis test.

Abbreviations: DF, degrees of freedom; SD, standard deviation.

3. Results

Baseline demographic characteristics of the entire cohort of colectomy patients are summarized in Table 1. During the study period, 603,730 cases of colectomy were identified, of which 17,308 (2.9%) had a discharge diagnosis code for AUD. The overall in-hospital mortality rate among all sampled colectomy patients was 1.4% and univariate analyses of relevant covariates are reported in Table 1 (analyses of all individual Elixhauser comorbidities are listed in the Appendix, Table A.2).

The number and proportion of patients who experienced in-hospital mortality, longer-than-median LOS, or higher-than-median total cost in patients with and without a diagnosis of AUD are depicted in Table 2. Among all surgical colectomy hospitalizations, those with a discharge diagnosis code of AUD were more likely to experience in-hospital mortality, long LOS, and high total cost than those without (Table 2). Results from multiple logistic regressions of each outcome after adjusting for other conditions of interest (other substance use disorders, psychotic disorders, and depression) as well as other patient characteristics are summarized in Fig. 2 and Table A.3. In multivariate analyses, a diagnosis code of AUD was associated with a significantly increased risk of in-hospital mortality after adjustment for other variables (AOR 1.36, 95% CI 1.24-1.48, p < 0.0001).

Table 2.

Descriptive statistics for primary and secondary outcomes overall and stratified by presence of alcohol use disorder

Characteristic	All, N = 603730 (%)	Alcohol Use Disorder, N = 17,308 (%)	No Alcohol Use Disorder, N = 586,422 (%)	K-W test (DF)	P value*
In-Hospital Mortality	8326 (1.4)	639 (3.7)	7687 (1.3)	700.8 (1)	<0.0001
High LOS	277,581 (46)	13,639 (78.8)	263,942 (45)	7729.4 (1)	<0.0001
High Total Cost	301,863 (50)	14,196 (82)	287,667 (49.1)	7307.8 (1)	<0.0001

^*Significance at p<0.05.

Abbreviations: AUD, alcohol use disorder; DF, degrees of freedom; K-W, Kruskal-Wallis; LOS, length of stay.

High LOS defined as greater than the median two-day stay. High Total Cost defined as greater than the median total hospital cost of $5,988.

Figure 2. Adjusted odds ratio plot for the associations between independent variables and in-hospital mortality.

Numeric values for adjusted odds ratios (OR) expressed in Table A.3. Elixhauser (reduced) is an unweighted cumulative Elixhauser score excluding the separately reported alcohol use disorder, drug use disorder, psychotic disorder, and depression variables. Race, black is defined as black versus the referent white. Race, other is defined as all other races except black versus the referent white. Payor presented as all others versus the referent Medicaid, charity, and self-pay categories.

AUD was further associated with a significantly increased adjusted odds of long LOS (AOR 1.45, 95% CI 1.39-1.52, p < 0.0001) and high hospitalization cost (AOR 1.63, 95% CI 1.56-1.70, p < 0.0001). Other SUDs, psychotic disorders, and depression were also independently predictive of both secondary outcomes. Findings remained consistent when secondary outcomes (LOS and total cost) were assessed based on a binary cut-off at the 75^th quartile (data not shown). Potential multicollinearity was considered by estimating the correlation matrices among parameter estimates in the models for the primary (Table A.4) and secondary outcomes (data now shown) [32]. The highest correlation coefficient detected was between payor and age at 0.36, thereby not meeting the commonly cited threshold of 0.8-0.9 that would indicate multicollinearity being a serious problem [33]. In reduced models sequentially eliminating variables with the highest and next highest correlations, the predictive strength of alcohol use disorder for each outcome remained similar to that in the corresponding full model (Table A.3).

3.1. Sensitivity Analysis

The observed risk ratio of 1.36 could be explained away by an unmeasured hypothetical confounder that was associated with both AUD and mortality by a risk ratio of 2.05-fold or greater, above and beyond the 13 other covariates in the model. The observed risk ratio of 1.45 could be explained away by an unmeasured confounder that was associated with AUD and long LOS by a risk ratio of 2.26-fold or greater, above and beyond the 13 other covariates in the model. The observed risk ratio of 1.63 could be explained away by an unmeasured confounder that was associated with both AUD and high cost by a risk ratio of 2.65-fold or greater, above and beyond the 13 other covariates in the model.

4. Discussion

Using claims data from a large cohort of US hospitals, we compared in-hospital mortality rates after colectomy between patients with and without a diagnosis code for AUD. Our findings indicate that a diagnosis of AUD is associated with a statistically and clinically significant elevated risk for adverse perioperative outcomes, thereby confirming our hypothesis. We found that patients with a discharge diagnosis code for AUD are 1.4-times more likely to experience in-hospital death after colectomy, 1.5-times more likely to have a hospital stay that is longer than the sample median of two days, and 1.6-times more likely to accumulate a total hospital cost greater than the median. Similarly, other SUDs, psychotic disorders, and depression were positively associated with increased LOS and higher total hospital costs.

In the present study, the prevalence of AUD was 2.9%. The diagnosis of AUD is based on the 2013 DSM-V definition, which integrated the two distinct disorders of alcohol abuse and alcohol dependence that previously co-existed in the DSM-IV. AUD is characterized by persistent alcohol use despite adverse consequences and is categorized as mild, moderate, or severe based on the number of satisfied criteria [3, 34]. Based on this definition, 5.6% of adults age 18 years or older were identified with AUD in 2019 [35]. The lower 2.9%-estimate of AUD prevalence in this study is likely due to the definition employed and the nature of the claims-based data. However, the reproducibility of similar AUD rates in different subsets of queried patients from claims-based databases is important to address [6]. Gupta et al. extracted 137,496 patients from the Premier Healthcare Database with community-acquired pneumonia and 3.5% had a diagnosis of AUD [36]. Masoomi et al, used the Nationwide Inpatient Sample database and reported the frequency of AUD in patients who underwent colon and rectal resection to be 2.0% [20]. Among 21,203 chronic non-cancer pain patients in the MarketScan medical claims database, 3.5% had an AUD diagnosis code [37]. Similar alcohol use rates and patterns were described in other observational data from single center studies [19, 21]. However, it is likely that the prevalence of AUD is understimated, both in the context of observational datasets and also in clinical practice.

Some previous work has shown no association between AUD and postoperative complications [18, 22, 38], while others demonstrate an association based on complication type [4, 39, 40]. The few studies that have used multivariate logistic regression analyses to determine independent risk factors for relevant postoperative outcomes including mortality, costs, and LOS are equally as inconsistent. In an older retrospective database study of 975,825 patients from the Nationwide Inpatient Sample, multivariate regression analysis identified AUD as a significant risk factor for in-hospital mortality after colon and rectal resection [20]. Similarly, Schmelzer et al. found that AUD was predictive of prolonged hospital LOS following colonic resection in 899 consecutive patients from 1998 to 2005 [21]. However, Klima et al. reported that AUD had no effect on survival in patients undergoing colectomy and in a retrospective cohort study of 181 patients undergoing colectomy for colorectal cancer, Alexander et al. determined that alcohol consumption had no significant associations with survival [18, 19]. These studies are limited by small samples and dated evidence.

The societal costs of AUD are significant. Anesthesiologists are well positioned to positively impact the outcomes of patients with AUD by engaging in their extended perioperative care before, during, and after admission. Early identification of patients who are affected by AUD is paramount to addressing the issue. Anesthesiologists in preoperative assessment clinics who evaluate patients days or weeks before a scheduled elective procedure may then be better equipped to develop a comprehensive preoperative cessation and optimization plan that includes both pharmacologic and non-pharmacologic therapies. This begins with the application of a risk scoring system such as the Alcohol Use Disorders Identification Test Consumption (AUDIT-C), a validated screening tool that, if positive, would prompt further scrutiny of hazardous alcohol consumption [41]. Given that higher scores have been shown to correlate with increased postoperative complications, the score itself may inform the intensity and breadth of intervention required [8]. Laboratory tests such as aspartate aminotransferase, alanine aminotransferase, mean corpuscular volume, and carbohydrate-deficient transferrin can be pursued for further stratification of risk. Serum biomarkers are relatively inexpensive, readily available, and have been shown to improve the sensitivity for detection of chronic excessive alcohol use in conjunction with a thorough history [42, 43]. The altered hematologic and biochemical abnormalities associated with excessive alcohol use may serve as an index of altered drug kinetics or potential for withdrawal in the acute setting [44]. Knowledge about the level of dysfunction may inform intraoperative management, as the anesthetic dose requirement is increased, the hypothalamic-pituitary-adrenal axis response is exaggerated, and the risk of prolonged respiratory depression is magnified in patients with AUD [45].

The periods immediately before and after a surgical episode may also provide anesthesiologists with a unique opportunity to initiate interventions targeted towards treating AUDs, especially in populations that are unlikely to pursue routine medical care. Prior research has shown that hospitalized patients with AUD in the hospital are more inclined to modify their behavior than the general population [46]. In one Cochrane review, evidence from three separate studies showed that intensive perioperative alcohol cessation programming increased the number of patients who successfully abstained and correlated with reduced postoperative complications [47]. Two randomized clinical trials have found that intensive alcohol cessation interventions 6-8 weeks prior to surgery are effective in reducing postoperative complications [48, 49]. Current guidelines suggest that four weeks of abstinence is adequate to recover cardiac function in patients with subclinical dysfunction, reverse prolonged bleeding time, and recover the ability to mount an adequate stress response [45, 50]. As the coronavirus pandemic has changed the healthcare landscape, digital application of preoperative cessation programs are important to consider. There is emerging evidence that digital intervention programs are as effective as face-to-face interventions in regards to lowering alcohol consumption [51]. Granted, optimization efforts are limited by patient willingness, the urgency of the surgical procedure, and time required to provide the support that equips cessation. However, integrated screening for mental health and SUDs such as the AUDIT-C within routine preoperative evaluations is an emerging area of interest that may enable anesthesiologists to identify psychiatric risk factors and refer patients for targeted therapy [52, 53]. Our work supports continued efforts in that vein.

4.1. Limitations

Our study has limitations. Inherent to the retrospective design is the potential for unaccounted confounding factors. Correction for unobserved bias was attempted using statistical techniques that adjusted for identifiable comorbidities [30]. We defined clinical significance post-hoc as the adjusted odds for mortality to change by at least 10% based on a given risk factor. Second, our outcomes of interest were defined based on ICD-10 codes reported in a large national database, which inherently has the potential for reporting bias and coding errors. Yet, reporting should not vary substantially over time, especially within a span of three years. Nonetheless, true rates of AUD are likely to be underestimated, as administrative databases contain information recorded by providers who frequently underdiagnose and underreport AUD in the inpatient setting [7, 9]. This may be due to lack of patient awareness, perceived lack of relevance, or inability to elicit history. Caution should therefore be taken when comparing studies with different AUD and SUD definitions [54]. Likewise, available data did not allow for stratification between a history of AUD and current, uncontrolled AUD, leading to potential ambiguity. However, the Elixhauser Comorbidity Index itself was specifically designed for databases lacking “present-on-admission” coding and was validated in patients undergoing major abdominal surgical resections [26]. It has been shown that the enhanced Elixhauser algorithm using ICD-10 codes performs better in predicting in-hospital mortality than the prior model based on ICD-9 codes due to the more specific and encompassing coding system [27]. Last, although these data provide evidence of a strong association between alcohol use disorder and post-colectomy in-hospital mortality, causality cannot be inferred in the context of an observational study.

4.2. Conclusions

We found that after adjusting for relevant covariates, AUD was associated with a significantly increased risk of in-hospital mortality in patients undergoing colectomy. In addition, LOS and total hospitalization costs were significantly higher in patients with AUD. Appreciation of the magnitude of perioperative risk conferred by AUD may catalyze enhanced efforts to identify patients with AUD that could enable anesthesiologists to provide more comprehensive perioperative care.