We used the Texas Inpatient Public Use Data File (TIPUDF), an administrative data set maintained by the Texas Department of State Health Services [14]. The TIPUDF includes approximately 97% of hospital discharges in the state and its characteristics were previously described [15].

We initially extracted data on all patients aged ≥18 years, who were hospitalized between the years 2009 and 2014. We then identified among these hospitalizations those with a diagnosis of cirrhosis and further identified, among the latter, hospitalizations with in-hospital CPR. This latter group formed the primary analytic cohort. A diagnosis of cirrhosis was based on the presence of International Classification of Diseases, Ninth Revision, Clinical Modification codes (ICD-9) 571.2, 571.5, and 571.6 in the principal or secondary diagnosis fields. These codes were used in prior studies of cirrhosis in administrative data [4, 16] and were reported to have a positive predictive value of 90% and negative predictive value of 87% [17]. CPR was identified based on the presence of ICD-9 codes 99.60 or 99.63 [6, 10]. Hospitalizations with a diagnosis of liver transplantation and those with principal diagnoses of cardiac arrest (ICD-9 code 427.5), ventricular fibrillation (ICD-9 code 427.41) or ventricular flutter (ICD-9 code 427.42) were excluded.

In order to identify secular trends not captured through analysis of the primary cohort, we anchored the data to similar-aged hospitalizations during 2009–2014 without a diagnosis of cirrhosis who had in-hospital CPR. In addition, we have examined all hospitalizations aged ≥18 years during the years 2009–2014 with and without a diagnosis of cirrhosis and those who died during hospitalization within each group to estimate the annual rates of in-hospital CPR among all hospitalizations and among hospital decedents.

The co-primary outcomes were rates of in-hospital CPR among all hospitalizations with and without cirrhosis and rates of short-term survival among hospitalizations with in-hospital CPR. We defined short-term survival as that of hospitalizations without in-hospital death or discharge to hospice.

The secondary outcomes included a) overall and over time rates of in-hospital CPR among all hospitalizations with and without cirrhosis who died in the hospital and b) overall and over time short-term survival among those with and without shockable rhythm, and hospital disposition of those without in-hospital death or discharge to hospice, both among hospitalizations with and without cirrhosis who have undergone in-hospital CPR.

Study variables were selected based on clinical plausibility and prior reports [4, 6, 10]. The abstracted variables included: a) demographics (age, gender, race/ethnicity, and health insurance) b) comorbid conditions, based on the Deyo modification of the Charlson Comorbidity Index [18] c) type of admission (elective vs. non-elective) d) admission during the weekend (vs. during weekdays) e) presence of a shockable rhythm (ICD-9 codes 427.41 or 427.42) [10, 19]; these codes were reported to have positive predictive values between 74% [20] and 100% [21] in administrative data f) number of failing organs [22] g) cirrhosis-related complications (ascites, hepatic encephalopathy, hepatorenal syndrome, hepatocellular cancer, and variceal bleeding) [4] h) sepsis g) hospitals’ teaching status h) hospital disposition and i) year of admission.

We summarized categorical variables as numbers and percentages, while continuous variables were reported as means (standard deviation [SD]). Group comparisons were performed using chi-square test for categorical variables and t-test or Mann-Whitney test for continuous variables. We used hospitalizations as the unit of analysis because the TIPUDF does not identify individual patients and thus does not capture repeated hospitalizations.

In order to provide further anchoring context to the trends of outcome trajectories between hospitalizations with and without cirrhosis who had in-hospital CPR, we examined their respective temporal trends of the burden of chronic illness and illness severity using the Deyo comorbidity index and the number of failing organs, respectively, as proxy measures.

Temporal trends of the primary and secondary outcomes were examined using weighted least square regression, with calendar year as predictor. Modeled findings are reported as average annual percent change (AAPC) and 95% confidence intervals (95% CI). All temporal analyses were carried out separately for hospitalizations with and without cirrhosis. Comparison of regression slopes was performed using the methods described by Armitage et al [23].

We estimated the risk-adjusted of short-term survival following in-hospital CPR among hospitalizations with and without cirrhosis using empirical Bayesian posterior estimates from multivariate logistic regression models for each (see modeling approach below) and then derived the corresponding annual rates of short-term survival.

A multivariate logistic regression model was fitted to estimate the association between cirrhosis as independent predictor and short-term survival following in-hospital cardiopulmonary resuscitation as dependent variable. Predictors with p < 0.1 on univariate logistic regression were considered for multivariate analysis, following examination for multicollinearity. Candidate predictor covariates were then entered into the model using backward stepwise selection. Results of the multivariate models were expressed as adjusted odds ratios (aOR) and their 95% CI.

The potential predictors of short-term survival following in-hospital cardiopulmonary resuscitation among hospitalizations with cirrhosis were examined using multivariate logistic regression modeling, employing the approach described above for the whole cohort.

In order to examine the robustness of our findings and to facilitate comparison to prior studies we re-examined both temporal trends and logistic regression models by substituting the outcome of short-term survival, as defined earlier, with hospital survival, defined as hospitalization without in-hospital death, among hospitalizations with in-hospital CPR, with and without cirrhosis, as well as among the corresponding strata with and without a shockable rhythm and those aged ≥ 65 years. Because the number of independent predictors of short-term survival among hospitalizations with cirrhosis was unexpectedly substantially more limited than among those without cirrhosis, we have repeated the multivariate logistic regression model without the cirrhosis-related complications to examine whether the findings of the original model were driven by the latter complications.

Data analyses were carried out on MedCalc version 17.5.5 (MedCalc Software, Ostend, Belgium). A 2-sided p value < 0.05 was considered statistically significant.

There were 54,280 hospitalizations with in-hospital CPR during the study period, of which 2,511 were with a diagnosis of cirrhosis and 51,969 without cirrhosis. The characteristics of these hospitalizations are detailed in Table 1. In-hospital CPR was reported among hospitalizations with and without cirrhosis in 7.6 and 4.0 per 1,000 hospitalizations, respectively (p < 0.0001) and in 10.7% vs. 13.4%, respectively, among hospital decedents (p < 0.0001).

The short-term survival following in-hospital CPR was markedly lower among hospitalizations with cirrhosis than among those without cirrhosis (14.9% vs. 27.3%, respectively; p < 0.0001). When stratified by the presence or absence of a shockable rhythm, short-term survival was similar between the two groups among hospitalizations with cirrhosis (14.6% vs. 15.0%, respectively; p = 0.8929), while being markedly higher among those with shockable rhythm among hospitalizations without cirrhosis (36.7% vs. 26.0%, respectively; p < 0.0001).

The temporal patterns of the rates of in-hospital CPR, presence of a shockable rhythm, and related short-term survival among hospitalizations with and without cirrhosis are described in Table 2. The details of the annual volumes of corresponding strata are provided in Table A in S1 File. The temporal trends of risk-adjusted short-term survival following in-hospital CPR among hospitalizations with and without cirrhosis are presented in Fig 1.

The annual rates of in-hospital CPR remained unchanged among hospitalizations with cirrhosis, while rising on average by 2.6%/year among those without cirrhosis. Similarly, the annual rates of in-hospital CPR among hospital decedents remained unchanged among hospitalizations with cirrhosis, while rising slightly, but statistically significantly, among those without cirrhosis.

There was no significant change over time in the crude short-term survival following in-hospital CPR among hospitalizations with or without cirrhosis. However, while the risk-adjusted short-term survival following in-hospital CPR remained unchanged among hospitalizations with cirrhosis, it increased over time among those without cirrhosis (Fig 1). The Deyo comorbidity index and the number of organ failures rose progressively over the study period among hospitalizations with and without cirrhosis who have undergone in-hospital CPR. However, there was no difference between the slopes of either measure between hospitalizations with and without cirrhosis (Table B in S1 File).

The annual rates of discharge to hospice following in-hospital cardiopulmonary resuscitation remained unchanged among hospitalizations with cirrhosis (AAPC = -3.1 [95% CI -12.4 to +18.6]; p = 0.6079), while rising 6.7%/year among those without cirrhosis (95% CI 3.9 to 9.5; p = 0.0026). Transfers to another hospital following in-hospital cardiopulmonary resuscitation did not change over time for hospitalizations with (p = 0.9901) or without cirrhosis (p = 0.9910). Discharges to a nursing facility remained unchanged overtime among hospitalizations with cirrhosis (p = 0.2004), while rising 8.2%/year among those without cirrhosis (95% CI 4.2 to 12.1; p = 0.0045).

The predictors of short-term survival following in-hospital CPR for the whole cohort and among hospitalizations with cirrhosis are described in Tables 3 and 4, respectively.

Cirrhosis was associated with 45% lower odds of short-term survival following in-hospital CPR on adjusted analysis. Among hospitalizations with cirrhosis who had in-hospital CPR, hepatocellular cancer (aOR = 0.35 [95% CI 0.16–0.76]) and lack of health insurance (aOR = 0.47 [95% CI 0.33–0.67]) were associated with the lowest odds of short-term survival. The predictors of short-term survival following in-hospital CPR among hospitalizations without cirrhosis are detailed in Table C in S1 File.

When hospital survival following in-hospital CPR was used as the outcome instead of short-term survival, the former remained significantly lower among hospitalizations with cirrhosis than among those without cirrhosis (17.9% vs. 30.6%, respectively; p < 0.0001). The hospital survival following in-hospital CPR among those aged ≥65 years with and without cirrhosis was 20.1% vs. 28.0%, respectively (p < 0.0001). On trend analysis (Table D in S1 File), hospital survival following in-hospital CPR remained unchanged among those with cirrhosis, while the upward trend in short-term survival noted among hospitalizations without cirrhosis became statistically significant for hospital survival. When re-analyzed without cirrhosis-related complications, there were no additional independent predictors of short-term survival and the effect size of the remainder independent predictors did not change significantly (Table E in S1 File).

Using hospital survival as outcome of interest, cirrhosis remained associated independently with worse outcome (OR = 0.55 [95% CI 0.50–0.62]; p < 0.0001) (Table F in S1 File), and among hospitalizations with cirrhosis the statistically significant predictors of adverse outcome remained unchanged (Table G in S1 File).

The comparative rates of in-hospital CPR between patients with cirrhosis and contemporaneous non-cirrhotic patients were not previously reported, to our knowledge. Our finding of higher rates of in-hospital CPR among cirrhotic patients likely reflects their markedly higher burden of chronic illness and acute severity of illness, associated with the documented higher rate of in-hospital death. On the other hand, in-hospital CPR was used more selectively among cirrhosis hospitalizations, evidenced by its persistently lower use among hospital decedents with cirrhosis, as compared with those without cirrhosis.

An earlier population-based study of Medicare patients aged ≥67 years by Stapleton and colleagues showed markedly lower hospital survival among cirrhotic patients between 1994 and 2005, as compared to those without cirrhosis [6], and a preliminary report described similar findings in all adults during 2000–2009 [7]. Our study is consistent with these observations in a contemporaneous cohort.

Despite the lower hospital survival following in-hospital CRP among cirrhotic patients in the present study, as compared to noncirrhosis patients, survival rates appear to rise in among the former. Thus, hospital survival among those aged ≥65 years has been nearly 2-fold higher in the present study, as compared to that reported by Stapleton et al (20.1% vs. 12.6% [6], respectively), and increased more slowly among all adults from 13% during 2000–2009 [7] to 17.9% in the present study.

There have been no population-level data, to our knowledge, on the epidemiology of shockable vs. nonshockable rhythm among cirrhotic patients undergoing in-hospital CPR, though no difference was found in the rate of a shockable rhythm between patients with and without cirrhosis in a small, single-center cohort [8]. Our finding of markedly lower rates of a shockable rhythm among hospitalizations with cirrhosis who have undergone in-hospital CPR than among noncirrhosis hospitalizations may reflect in part differences in the proximate causes of cardiac arrest and patients’ location at the time. However, administrative data preclude more detailed analysis.

When examined in isolation, our findings of unchanged adjusted short-term survival over time following in-hospital CPR among cirrhosis patients, coupled with their progressive rise in the burden of chronic illness and severity of illness, may be interpreted as reflecting improving care in an increasingly sick population. However, such postulate is not supported by the contemporaneously improving adjusted short-term survival among those without cirrhosis, despite similar upward trajectories of measures of chronic illness and severity of illness. Nevertheless, the severity of comorbidities and organ failures may have diverged over time between patients with and without cirrhosis, exceeding that among the latter, though these data are not available in administrative data sets.

Alternatively, the lack of improvement in short-term survival among cirrhosis hospitalizations, as compared to those without cirrhosis, may have been in part to the progressively decreasing rates of shockable rhythm among the former. However, a key finding of the present study is the similar short-term survival following in-hospital CPR among cirrhosis patients regardless of presence or absence of a shockable rhythm. This finding, not previously reported to our knowledge, contrasts the well-documented prognostic advantage of a shockable rhythm in the general population [10, 11]. The underlying sources of the observed divergent prognostic association of a shockable rhythm between patients with and without cirrhosis are unclear and require further study.

The factors underlying the worse risk-adjusted short-term outcomes following in-hospital CPR among patients with cirrhosis remain incompletely understood. A recent report by Roedl et al documented markedly worse profile of organ failure following CPR among cirrhotic patients, as compared to those without the disease [13], which may have contributed to the lower survival among the former. In addition, as noted earlier, although we controlled our modeling for the burden of chronic illness and the number of failing organs, the severity of neither could be captured. Finally, patients’ location, the proximate precipitant of cardiac arrest, and care during and following cardiac arrest may have differed between cirrhosis and non-cirrhosis hospitalizations. Further studies of the abovementioned data gaps are needed to enhance understanding of the observed outcome patterns in the present study and inform future preventive and intervention strategies.

We found that with the exception of hepatocellular cancer, lack of health insurance was associated with the worst odds of short-term survival among cirrhotic hospitalizations undergoing in-hospital CPR. Lack of health insurance has likely adversely affected patients’ access to primary and specialty care, and thus may have led to worse control of cirrhosis and other comorbidities. In addition, lack of health insurance was shown to be associated with higher mortality among critically ill patients in the general population [24].

Several comorbid conditions, including myocardial infarction, cerebrovascular disease, and diabetes in the whole cohort, and cerebrovascular disease and diabetes and among hospitalizations with cirrhosis were found, unexpectedly, to be associated with increased odds of short-term survival following in-hospital CPR. The factors underlying these findings are unclear. However, each of these comorbidities is associated with increased risk of acute cardiac events and prior studies found that coronary artery disease is associated with increased survival following cardiac arrest [10, 12, 19]. In addition, primary cardiac diagnosis for cardiac arrest was found to be associated with 70% higher odds of survival following in-hospital CPR, as compared to a respiratory diagnosis [25]. It may be hypothesized that the abovementioned comorbidities serve as proxies for a subgroup of patients who are more likely than others to develop in-hospital cardiac arrest due to an acute cardiac event, rather than due to the more common occurrence of cardiac arrest due to progression of another life-limiting condition (e.g., cancer) or due to other acute illness. Thus, patients with these comorbidities may have better odds of survival following in-hospital cardiac arrest only in comparison with those without these comorbidities. Additional studies, using more granular data, are needed to examine the factors preceding in-hospital cardiac arrest across the comorbidity and acute illness spectrum to evaluate the validity of our postulate.

A striking finding of the present study was the markedly limited number of factors that were independently associated with short-term survival following in-hospital CPR among hospitalizations with cirrhosis, in contrast to their non-cirrhotic counterparts, and to the markedly broader range of predictors of hospital mortality among cirrhosis hospitalizations in general [4]. Specifically, commonly impactful sociodemographic factors, the burden of other comorbidities or the circumstances of hospital admission (e.g., weekend admission) no longer affected cirrhotic patients’ short-term survival following CPR for in-hospital cardiac arrest. The factors underlying these latter observations, likely substantially driven by peri-arrest pathobiology of cirrhotic patients, warrant further study.

Given the gradual rise of hospital survival over the past decades among patients with cirrhosis undergoing in-hospital CPR, our findings suggest its use has a potential to benefit select patients in this population. Indeed, our study demonstrates that in-hospital CPR continued to be used selectively among cirrhotic hospitalizations, despite the rising hospitalization burden in this population [4]. However, the characteristics of those most likely to benefit have not been systematically examined. In addition, studies are needed to better understand the proximate causes of cardiac arrest among cirrhotic patients and their pathobiology following CPR, in order to inform future preventive and interventional efforts, and to examine the neurological and other functional outcomes among survivors.

Our finding of temporally divergent trajectories of discharge to hospice among hospitalizations with and without cirrhosis who had in-hospital CPR underscores the need for caution when interpreting the longitudinal patterns of hospital mortality-based outcomes in the general population and in comparative population studies, given the dynamic changes in illness complexity and discharge practices.

Last, our findings on the adverse prognostic factors following in-hospital CPR in cirrhotic patients underscore the importance of assuring adequate health insurance, early recognition and timely care of infection and sepsis and, when applicable, interventions to limit evolvement of organ failure.

Our study has several important limitations, in addition to those noted earlier, related predominantly to its retrospective design and use of administrative data. First, use of administrative data may have led to misclassification of some of the examined hospitalizations, though similar approach was used in other epidemiological studies [4, 6, 10, 16]. Nevertheless, there are limited data on the validity of the ICD-9 procedure codes for CPR in administrative data. A recent registry-based study found considerable underestimation of in-hospital CPR events by these ICD codes, with slightly lower survival than in the reference data [26]. However, there is no evidence for systematic differences in coding for CPR between the general population and patients with cirrhosis. Thus, the low sensitivity of the ICD codes for CPR would not explain the gaps in short-term survival following in-hospital CPR between hospitalizations with and without cirrhosis, nor the divergent temporal trajectories of short-term survival between these groups. Second, although we adjusted for a broad array of patient- and hospital-level covariates, residual confounding cannot be excluded. Last, it is unknown whether our observations reflect those in other states or nationally.