Postarrest Steroid Use May Improve Outcomes of Cardiac Arrest Survivors

Min-Shan Tsai, MD, PhD; Po-Ya Chuang, MHA; Chien-Hua Huang, MD, PhD; Chao-Hsiun Tang, PhD; Ping-Hsun Yu, MD; Wei-Tien Chang, MD, PhD; Wen-Jone Chen, MD, PhD


Crit Care Med. 2019;47(2):167-175. 

In This Article


Study Design and Data Sources

This nationwide retrospective cohort study on the effect of postcardiac arrest steroid use was conducted by analyzing health insurance administrative data retrieved from the Taiwan NHIRD for the interval of January 1, 2003, to December 31, 2012. The claims database, which was released by the National Health Research Institute (NHRI) for research purposes, provides comprehensive information on healthcare utilization and demographics (birthdate, gender, insurance status, area of residence, dates of services provided, primary and minor diagnostic codes, primary and minor procedural codes, and itemized expenditures for each medical service rendered) for greater than 99% of the entire Taiwanese population of 23 million people.[18] The study protocol was approved by the NHRI and Review Board of the National Taiwan University Hospital, and the need for informed consent was waived. All data were deidentified, and confidentiality assurances were addressed in accordance with data regulations of the National Health Insurance (NHI) Administration.

Study Subjects

Our database search identified 174,068 patients who had undergone CPR (procedure code: 47029C) in the emergency department between January 1, 2004, and December 31, 2011. Among those patients, 172,016 received subsequent medical care and were identified as possible candidates for our study. Patients who were less than 18 years old or older at date of presentation (n = 4,041) and those involved in trauma (n = 12,698) were excluded. To identify patients in cardiac arrest during acute hospital care, patients not triaged as level 1 (resuscitation/emergent) (n = 2,436), and patients with an emergency department observation time greater than 6 hours (n = 7,197) were also excluded. After identifying 22,768 nontraumatic adult cardiac arrest survivors, we further excluded patients who received steroid during CPR (n = 1,180), patients who received steroid for greater than 1 month after ROSC (n = 2,398), and patients given prednisolone equivalent dose greater than 1,250 mg/d (equal to methylprednisolone dose of 1 g/d as pulse therapy) (n = 23). This left 19,229 patients available for inclusion in our study (the first sample). Based on whether the patient received steroid during hospitalization, the 19,229 patients were then divided into the steroid group (n = 5,477) and nonsteroid group (n = 13,752). Among the first sample, 6,955 patients with a history of steroid use within 1 year (oral, IV, and intraarticular administration of steroid) prior to cardiac arrest were further selected as the secondary sample (steroid group, n = 2,227; nonsteroid group, n = 4,728). A total of 12,274 patients without a history of steroid within 1 year prior to cardiac arrest constituted the third sample (steroid group, n = 3,250; nonsteroid group, n = 9,024) (Figure 1).

Figure 1.

Process of selecting the study subjects. CPR = cardiopulmonary resuscitation, ED = emergency department.

Definition of Variables

The primary study outcome was defined as survival to discharge, and the secondary outcome was 1-year survival. A death event was identified if the discharge status of index admission was "death," the patient had a death record in the catastrophic illness registry, or the patient was disenrolled from the NHI program. The main focus of the study was the effect of steroid use during hospitalization following cardiac arrest, and the relevant information was collected from claims data generated by hospital care. The type and quantity of administered steroid during the entire hospitalization after ROSC were obtained from the NHIRD. The allowable steroidal supplements were hydrocortisone, methylprednisolone, prednisolone, triamcinolone, dexamethasone, and betamethasone. The controlled variables were age, gender, presenting complaint, shockable rhythm, epinephrine dosage, total shocks delivered during CPR, and cardiac catheterization. The presenting complaint was recognized from the International Classification of Diseases, 9th Edition code of primary diagnosis at admission. The steroid and nonsteroid groups had two patients with missing urbanization level of residence and geographic distribution, respectively. They were excluded from analysis of propensity score (PS). To clarify whether the steroid dose was associated with outcomes, we attempted to standardize the prednisolone equivalent dose (prednisolone equivalent dose/d), which was calculated as the total dose of prednisolone equivalent given during hospitalization (when hospitalization duration was < 1 mo) or the first month (when hospitalization duration was ≥ 1 mo), divided by the corresponding number of days (either hospitalization duration or 30 d). We then further divided the steroid group into four different strata using quartiles of standardized prednisolone equivalent dose.

Computation and Matching of PS

Propensity scoring was used to reduce selection bias between the steroid and nonsteroid groups. The PS, defined as the probability of receiving steroid, was first estimated by modeling a logistic regression drawn from pertinent characteristics.[19] The Hosmer-Lemeshow goodness-of-fit test was then used to assess propensity models' performance. Results of the logistic regression model used for PS matching are presented in Supplemental Table 1 (Supplemental Digital Content 1,, Supplemental Table 2 (Supplemental Digital Content 2,, and Supplemental Table 3 (Supplemental Digital Content 3,

The steroid and nonsteroid groups were matched by PS using 8 to 1 digit-based greedy matching algorithm with the nearest available pair matching method at 1:1 ratio (cases:controls) without replacement, as proposed by Parsons.[20] The final PS-matched sample consisted of 5,445 patients in each group among total patients, 2,188 patients among patients with prior steroid use, and 3,248 patients among patients without prior steroid use, respectively.

Statistical Analyses

The demographic and clinical characteristics of study subjects were summarized using the mean ± SD for continuous variables and frequencies and percentages for categorical variables. The standardized difference as effect size was calculated to compare group differences and assess the balance of the baseline characteristic of study subjects before and after PS matching.[21] The impact of steroid use and the steroid dose on survival to discharge and 1-year survival was determined by the Cox proportional hazard model. A bootstrapping approach by resampling for 100 replications with replacement from the original samples was used to perform internal model validation in assessing survival to discharge.[22] The adjusted 1-year survival curves were also plotted via the Cox proportional hazard model. When performing the dose analysis, a receiver operating characteristic (ROC) curve was plotted to determine the optimal steroid dose cut-off value. Subgroup analyses were performed to examine relationships between steroid use, clinical characteristics, and survival to discharge. All computations were performed using standard software (SAS/Stat v9.3 for Windows; SAS Institute, Cary, NC). The components of retrospective review were checked by using the Strengthening the Reporting of Observational Studies in Epidemiology checklist for a cohort study (Supplemental Table 4, Supplemental Digital Content 4,