Ethics statement
This study was initiated after its protocol was approved by the Institutional Review
Board of Dalin Tzu Chi Hospital, Buddhist Tzu Chi Medical Foundation, Taiwan, and
was conducted in conformity with the Declaration of Helsinki.
Database
The National Health Insurance (NHI) program was implemented in Taiwan in 1995 and
provides compulsory universal health insurance. It enrolls about 99Â % of the Taiwanese
population and contracts with 97 % of all the country’s medical providers 10], 11]. The database contains comprehensive information about all insured subjects, including
sex, date of birth, residential or work area, dates of clinical visits, diagnoses
identified by International Classification of Diseases (Ninth Revision) Clinical Modification
(ICD-9-CM) diagnostic codes, details of prescribed medications and procedures administered,
expenditure amounts, and outcome at hospital discharge (i.e., recovered, died, or
transferred out) 12]. A random sample of 1,000,000 people who received health benefits from the NHI program
was selected based on calendar-year 2005 reimbursement data and was considered representative
of the entire population; according to the Taiwan National Health Research Institute,
the group did not differ statistically from the larger cohort in age, sex, or health
care costs 13], 14]. This sample was used as our study cohort.
Study population
The population sample was followed from 1 January 2003 to 31 December 2012 (a total
of 10Â years). For our study cohort, we first identified individuals who were still
alive in 2005 and were older than 18Â years at the time of OHCA. OHCA was defined by
the ICD-9-CM codes ventricular fibrillation (427.4), cardiac arrest (427.5), and sudden
death (798.0–798.9) in outpatient clinic records. To avoid including patients who
had Do Not Attempt Resuscitation orders, no chance of survival, and coding errors,
we excluded patients on whom no chest compression was attempted. ETCO
2
monitoring was defined by the charge record for continuous capnography during the
visits. For each charge record of ETCO
2
, the institutes could claim for about US$ 14 from the Bureau of National Health Insurance.
After exclusions, we identified 83 patients who received ETCO
2
monitoring and 4958 who did not receive such monitoring. In order to identify sustained
ROSC, each patient was tracked if he or she was hospitalized after OHCA (Fig. 1).
Fig. 1. Flow diagram of the population-based study
Prespecified covariates
In order to investigate a significant influence on survival associated with ETCO
2
-monitoring, we included several covariates in the analysis; age, sex, calendar year,
urbanization level (i.e., urban, suburban, and rural), health care institutes, and
socioeconomic status (SES). Income-related insurance payment amounts were used as
a proxy measure of individual SES at follow-up. People were divided into 3 groups:
(1) low SES: payment lower than US$571 per month (New Taiwan Dollars [NT$] 20,000);
(2) moderate SES: payment between US$571 and US$1141 per month (NT$ 20,000–40,000);
and (3) high SES: payment of US$1142 or more per month (NT$40,001 or more) 12]. The health care institutes visited by patients were classified into 4 levels (medical
centers, regional hospitals, local hospitals, and clinics) based on hospital accreditation.
Two additional covariates that may be related to sustained ROSC following resuscitation,
the CPR duration and attempted defibrillation, were identified based on charge records.
Finally, the prevalence of selected comorbid conditions (i.e., diabetes, hypertension,
coronary artery disease, hyperlipidemia, malignancies, heart failure, atrial fibrillation,
intracranial hemorrhage, ischemic stroke, chronic renal insufficiency, and liver cirrhosis)
and the Charlson Comorbidity Index (CCI) score were determined using discharge diagnoses
either during outpatient clinic visits or hospitalizations before 1 January 2005.
The CCI is a scoring system that assigns weights to important concomitant diseases;
it has been validated for use in studies that employ ICD-9-CM data 14], 15].
Propensity score methods
In this study, the propensity score was the conditional probability for using ETCO
2
monitoring in the presence of possible confounders. The prespecified covariates were
added into a multivariable logistic regression model to predict the probability of
ETCO
2
use. The predicted probability from the model was used as the propensity score for
each patient. We then matched each patient in the ETCO
2
group to 20 patients in the untreated group with the closest propensity score using
a standard greedy-matching algorithm 16] and compared the probability of survival benefits between these groups.
Statistical analysis
The SAS statistical package, version 9.4 (SAS Institute, Inc., Cary, NC, USA) was
used for data analysis. All covariates were taken as categorical variables except
age, calendar year, CPR duration, and propensity score, which were treated as continuous
variables. Categorical variables were compared using Pearson’s chi-square test, and
continuous variables were assessed using the t test to determine baseline heterogeneity in the 2 groups. Simple conditional logistic
regression models were then used to calculate the ORs of sustained ROSC and survival
to hospital discharge for patients with ETCO
2
use in the matched group.
In order to further assess the robustness of our results, we sampled another cohort
by matching each patient in the ETCO
2
group to 4 patients in the untreated group using the same method (Fig. 1). We compared the crude ORs and risk differences of survival benefits among 2 matched
cohorts and the original group to evaluate if the results are similar. We also evaluated
the extent of the effect of a potentially unmeasured confounder in accounting for
the results 17]. A two-tailed P value of 0.05 was considered significant.