Discharge Against Medical Advice and Risk of Readmission After Sepsis Hospitalization

Background: To describe characteristics of sepsis patients who discharged against medical advice (AMA), identify factors associated with AMA discharges in the patients, and evaluate the association of AMA discharge with 30-day unplanned readmission and outcomes of readmission. Methods: Using the National Readmission Database of the United States, we identied inpatients with sepsis who discharged AMA or discharged home between 2010 and 2017. The baseline characteristics were compared between the two groups. Multivariable models were used to identify factors related to AMA discharge, evaluate the association between AMA discharge and 30-day unplanned readmission, and elucidate the relationship between the AMA discharges and in-hospital outcomes. Results: AMA discharges accounted for 2.29% of all the hospitalized sepsis patients. The prevalence of AMA discharge in sepsis patients increased from 1.99% in 2010 to 2.55% in 2014 (p for trend < 0.001).The unplanned 30-day readmission rates of sepsis patients who discharged AMA and who discharged home are 25.51% and 12.26%, respectively. AMA discharge is statistically signicantly associated with 30-day [odds ratio (OR), 2.24; 95% condence interval (CI), 2.15–2.33], 60-day (OR, 2.07; 95% CI, 1.99–2.15), and 90-day (OR, 1.97; 95% CI, 1.90–2.05) readmission. AMA discharge is also associated with longer length of stay in 30 days (0.44 day, 95% CI, 0.12 days-0.76 days, p=0.007), whereas there was no statistically signicant difference in hospitalization costs and in-hospital mortality for patients discharged AMA versus those discharged home. Conclusions: Due to the high risk of readmission, vulnerable patients should be early identied. Medical institutions should conduct post-discharge interventions for patients with AMA discharge, such as follow-up visits and psychological counseling, to reduce readmission.


Introduction
More than 970,000 sepsis cases are diagnosed in the United States (U.S.) each year, and the number is increasing. [1] Although the in-hospital mortality rate decreased from 24.1-14.8%,the proportion of hospitalized patients with sepsis increased from 3.9% in 2010 to 9.4% in 2017. [2] Reducing readmissions has been addressed in order to improve patient outcomes and control health care costs. The Centers for Medicare & Medicaid Services (CMS) of the U.S. use the 30-day readmission rate as a measure of hospital care quality. Because heart failure (HF), pneumonia, and acute myocardial infarction (AMI) account for a large proportion of readmissions, CMS reports readmissions for these conditions. [3] In fact, initial hospitalization costs for sepsis only account for 30% of total costs and are related to severity and length of stay (LOS), [4] but the hospitalization costs and LOS of readmission in sepsis patients are higher than patients with AMI, HF, and pneumonia. [5,6] Discharge against medical advice (AMA) is a global public health problem. In the U.S., about 1 to 2 percent of discharges are AMA. [7] Compared with patients who discharged routinely, patients who discharged AMA are more likely to discharge with inadequate care, and increased risk of readmission and death. [8] A study of AMI patients found that those who left AMA were more than twice as likely to be readmitted, and resulted in a mortality 57% higher. [9] In patients who underwent PCI, AMA discharge was a strong predictor for readmission, and associated with greater mortality. [10] However, little has been known about the impact of AMA discharge in sepsis patients. Using the National Readmission Database (NRD) of the U.S., Shruti et al. evaluated the factors associated with 30-day readmission in patients with sepsis, while they classi ed patients who left AMA into other discharge group. [6] In this study, we aimed to describe characteristics of patients who left AMA, identify factors associated with AMA discharge in patients with sepsis, and evaluate the association of AMA discharge with 30-day unplanned readmission, and outcomes of readmission.

Institutional approval and data availability
The NRD is publicly available, therefore this study was exempt from the formal institutional review board approval.

Data source
The NRD is maintained by the Healthcare Cost and Utilization Project (HCUP) sponsored by the Agency for Healthcare Research and Quality (AHRQ). It is one of the largest publicly available all-payer inpatient databases in the U.S., representing approximately half of total resident population and hospitalizations in the country. [11] The large database may support various types of analyses related to speci c diseases or procedures. The current register-based cohort study used data from the NRD covering the period from 2010 to 2017.

Study population
The study population involved patients 18 years or older with a diagnosis of sepsis who were subsequently discharged home or discharged AMA between 2010 and 2017. The International Classi cation of Diseases, 9th Revision, Clinical Modi cation (ICD-9-CM) and ICD-10-CM diagnosis codes, which have been proved effectiveness in identifying sepsis patients, were used to identify sepsis patients in our study. [12,13] The NRD variable 'DISPUNIFORM' was used to identify patients who discharged AMA. We excluded patients who died during the rst index admission or discharged in December of each year because of incomplete 30-day follow-up data. We further excluded patients with elective readmissions, and those who were not discharged home or not discharged AMA. The details regarding inclusion and exclusion are showed in Figure 1.

Covariates
Information related to demographic characteristics (age, sex, weekend admission, income by postal code, insurance type), hospital characteristics (bed size and teaching status), the degree of loss of function, and the risk of mortality were extracted from the NRD database. Elixhauser Comorbidity Index (ECI) was calculated to account for the burden of 29 common comorbidities. [14] The comorbidities related to sepsis, such as acute kidney injury, chronic kidney disease, shock, vasopressor use, acute cardiorespiratory failure, ventilator use, gastrointestinal bleeding, endocarditis, meningitis, and urinary tract infection or pyelonephritis were also identi ed using ICD-9-CM and ICD-10-CM codes (e- Table 1).

Primary and secondary outcomes
The primary outcome was 30-day unplanned readmission. An unplanned readmission was de ned as the rst readmission of a sepsis patient within 30 days of discharge that was not elective. The secondary outcomes were (1) the temporal trend of AMA discharge from 2010 to 2014; (2) the temporal trend of 30day unplanned readmissions; (3) 60-day and 90-day unplanned readmissions; (4) outcomes of readmission.

Statistical analysis
The NRD was based on a complex sampling design and we utilized survey estimation commands according to the HCUP guideline to obtain national estimates. Descriptive statistics were presented as means and standard errors for continuous variables, and as frequencies and percentages for categorical variables. Chi-squared test was used to compare differences between categorical variables and t-test was used for the comparison between continuous variables. We calculated weighted annual prevalence of AMA discharge for the overall study population as well as subgroups categorized by age, gender, insurance type and income. The trends of annual AMA discharge prevalence over the study period were evaluated using the Cochran-Armitage trend test.
Multivariable logistic regression analysis was used to identify statistically signi cant factors associated with AMA discharge, adjusted for demographic characteristics, hospital factors, severity indices, and treatment variables. Both univariate model (model 1) and multivariable-adjusted models (models 2 and 3) were constructed to evaluate the associations between AMA discharge and 30-day, 60-day, or 90-day unplanned readmissions in the sepsis patients. Model 2 was adjusted for age, weekend admission, sex, elective admission status, insurance, income, hospital bed size, and teaching status. Model 3 was additionally adjusted for individual comorbidities based on model 2. We also performed a multivariable regression analysis to elucidate the relationship between AMA discharge and in-hospital outcomes (hospitalization costs, LOS, and in-hospital mortality) in readmission, adjusted for demographic characteristics, hospital factors, severity indices, and treatment variables.
Statistical signi cance was de ned as two-sided p values ≤ 0.05. All statistical analyses were conducted in SAS software version 9.4 (SAS Institute, Cary, NC).
Similar trends were also observed for shock, acute cardiorespiratory failure, ventilator use, and endocarditis. Furthermore, we also noticed that patients who discharged AMA had shorter LOS (2.09 days vs. 3.05 days, p < 0.001) during their index admission.

Time trends of annual AMA discharge
The annual prevalence of AMA discharge in sepsis patients increased from 2.00% in 2010 to 3.89% in 2017 (p for trend < 0.001) ( Figure 2). E- Figure 1 shows the temporal trends of annual AMA discharge prevalence in the subgroups categorized by age, sex, types of insurance, and income. The prevalence was highest in patients younger than 50 years of age. The prevalence of AMA discharge was higher in males and increased from 2.72% in 2010 to 4.36% in 2017 (p for trend < 0.001). In addition, higher prevalence of AMA discharge AMA was observed among patients with self-pay and lower income.
Causes of 30-day unplanned readmissions Table 6 shows the top 10 causes of 30-day unplanned readmissions. The most common cause for readmission is infections, while the proportion was higher in patients who left AMA (41.51%) than that in patients who discharged home (34.30%). The top 15 causes of readmission accounted for more than 85 percent of all readmissions.

Discussion
Our results reveal that although the proportion of patients with AMA discharge is not high, they were more likely to be readmitted compared to those discharged home. We observed that the unplanned 30-day readmission rate of patients who left AMA was 21.64%, twice as the rate of patients who discharged home, and comparable to that for patients with severe sepsis. [15] In addition, we found that 60-day and 90-day readmission rates for patients who left AMA were also signi cantly higher than that for patients who discharged home. Our ndings support the need for a deeper understanding of AMA discharge in sepsis patients to further develop e cient interventions.
In our study, AMA discharge was associated with younger age, male, Medicaid, lower income, higher risk of mortality, more comorbidities, psychoses, drug abuse, and alcohol abuse. We found that patients with a high risk of mortality were more likely to discharge AMA, which may be related to palliative care. [16] Hospital-level characteristics, such as location and teaching status were also found associated with AMA discharge. The high prevalence of AMA discharge in urban nonteaching hospitals may be related to the limited attention of service providers. Patients with alcohol and drug abuse were more likely to discharge AMA, consistent with previous ndings. [10,17,18] Both alcohol and drugs can impair cognition and lead to impulsive or risky behavior, prompting the patient's subsequent decision to leave AMA. [19] In addition, patients who are dependent on drugs and/or alcohol may be more likely to leave the hospital toobtain dependent substances, especially if withdrawal symptoms are not properly treated. [20] The risk of AMA discharge was lower in patients with obesity, acute cardiorespiratory failure, metastatic cancer, solid tumor without metastasis, meningitis, and urinary tract infection or pyelonephritis. We speculate that the associated more severe comorbidities may place a greater burden on sepsis patients and thus in uence their discharge decisions. We also found that the patients admitted electively in the index hospitalization are less likely to discharge AMA, which suggests that there might be selection bias in the included population. The patients who were admitted at an elective hospital were likely to have better compliance than those who were not. In addition, it also may be related to the expectations of the patients and capacity to plan. The LOS is usually controlled and communicated with family members prior to admission, so patients generally do not decide to discharge AMA.
It is essential to increase our understanding of the factors associated with AMA discharge. It helps us identify high-risk groups for AMA discharge and intervene early to reduce subsequent readmission and additional health care costs. Previous studies have shown that although there may be many reasons for patients to leave AMA, communication disorders are often the main reason. [21,22] Known factors affecting communication quality include patient's age, gender, race, and income. Insurance status, hospital type, hospital size, and severity of illness were also identi ed as associated factors of AMA discharge. Although these factors are immutable, they can be used as indicators to identify vulnerable patients who may need to improve communication to prevent AMA charge. [7] Providing a prescription at the time of discharge to patients who are about to leave AMA may reduce readmission and mortality rates in infected patients who need to complete a course of antibiotics. If the patient persists in leaving AMA, appropriate drug intervention and follow-up may play an important role after discharge. Establishing a discharge plan that bring the least harm to patients may increase patients' adherence to the plan and eliminate disadvantageous factors at discharge. [23] In this analysis, AMA discharge was a factor associated with 30-day readmission, which is consistent with several previous studies. In a survey of trauma patients by Olufajo et al., the risk of 30-day readmission was three times higher in patients who discharged AMA and two times higher in patients with multiple readmissions compared to those discharged upon medical advice. [24] A retrospective cohort study in California showed that patients with AMI who discharged AMA had a 60% higher risk of hospital readmission with AMI or unstable angina within two years of discharge than those who were discharged routinely. [9] In our study, the effect of AMA discharge on readmission after the rst discharge of the sepsis patients lasted for 3 months.
Our analysis showed that patients who left AMA had a higher rate of readmission for infection. A retrospective cohort study showed that half of sepsis patients readmitted for infection were due to recurrent or unresolved infections. [25] The initial treatment strategy for infection is to eliminate the causative organism, and the duration of treatment is critical. Patients who discharged AMA are likely to be discharged from the hospital before the course of treatment is completed. The patients who leave the hospital before reaching clinical stability will have their conditions continue to deteriorate, leading to an increased risk of readmission.
We found that AMA discharge was associated with higher in-hospital mortality in 30-day, 60-day, and 90day readmission, which is consist with previous studies. A retrospective cohort study from a Veterans Administration hospital showed that AMA patients had a higher mortality rate than those who discharged home for all admissions. [8] In another retrospective cohort study involving adults in Manitoba from 1990 to 2009, elevated mortality associated with AMA discharge was manifest within one week and continued to 180 day after discharge.
[26] We have the following assumptions. First, behaviors of the patients with AMA discharge after discharge from hospital may lead to higher mortality, such as substance abuse, rather than AMA discharge itself. Secondly, these patients often failed to complete the relevant examinations, which limits clinicians to make accurate diagnosis. Third, the patients with AMA discharge have poor compliance, such as taking prescription drugs on time.
There are several limitations in our study. First, as in other administrative databases, reporting biases, data loss, and coding errors are also existing in the NRD. Second, sociodemographic data, such as race, marriage and other characteristics, are associated with the AMA dischagre [7], however they are not available in the NRD. Third, the yearly datasets contained in the NRD could not be linked across the years, and the patients admitted in December were also excluded, so the prevalence of AMA discharge in this study may be underestimated. Fourth, as a retrospective study, we only evaluated the associations. In order to make causal inference between AMA discharge and the interested outcomes, more re ned study design such as propensity score matching, and powerful adjustment such as instrumental variable technique may be considered in the future.

Conclusions
This study identi ed several characteristics associated with AMA discharge AMA in sepsis patients and revealed the associations between AMA discharge and unplanned readmissions. Due to the high risk of readmission, vulnerable patients should be early identi ed, and medical institutions should conduct postdischarge interventions for patients with AMA discharge, such as follow-up visits and psychological counseling, to reduce readmission. Our ndings are helpful to raise awareness among clinicians and policy makers about the patient population, thereby use the health care resources e ciently. Further studies on the effects of AMA discharge on the outcomes of readmission are needed in the future.

Declarations
Ethics approval and consent to participate The NRD is publicly available, so this study was exempt from formal institutional review board approval.

Consent for publication Not applicable.
Availability of data and materials The datasets generated and/or analysed during the current study are available in the https://www.hcup-us.ahrq.gov/nisoverview.jsp.
Competing interests The authors declare that they have no competing interests.     Figure 1 Selection ow diagram of target population.

Figure 2
The prevalence of Discharging against Medical Advice in sepsis. Figure 3