Adverse Events and Factors Associated With Potentially Avoidable Use of General Anesthesia in Cesarean Deliveries

Jean Guglielminotti, M.D., Ph.D.; Ruth Landau, M.D.; Guohua Li, M.D., Dr.PH.


Anesthesiology. 2019;130(6):912-922. 

In This Article

Materials and Methods

The study protocol was reviewed by the Institutional Review Board of Columbia University Medical Center, New York, New York, and was granted exemption under 45 Code of Federal Regulation 46 (not human subjects research). The Strengthening The Reporting of OBservational studies in Epidemiology (STROBE) and the Reporting of studies Conducted using Observational Routinely-collected health Data (RECORD) statements were followed.

The initial study protocol was not publicly registered. The currently presented analysis plan was based upon the initial plan combined with peer review process requested changes.

Study Samples and Definition of Exposure

The study sample included all records of discharges after cesarean delivery performed in New York State hospitals between January 1, 2003, and December 31, 2014, without a recorded clinical indication for general anesthesia. Clinical indications for general anesthesia were categorized into three groups (Table 1 and Supplemental Digital Content Table 1, obstetrical indications (e.g., placenta accreta), maternal indications (e.g., pulmonary hypertension), and contraindications to neuraxial techniques (e.g., coagulation factor deficit). Cesarean delivery cases without a recorded clinical indication for general anesthesia may indicate situations where general anesthesia was potentially avoidable.

Hospital discharge records of the State Inpatient Database for New York were analyzed. State Inpatient Databases are part of the Healthcare Cost and Utilization Project sponsored by the Agency for Healthcare Research and Quality (Rockville, Maryland). They capture all inpatient discharges from nonfederal acute care community hospitals, including tertiary and academic centers. For each discharge, the New York State Inpatient Database indicates the type of anesthesia provided, one hospital identifier, patients characteristics, and procedures performed using the International Classification of Diseases, Ninth Revision–Clinical Modification (ICD-9-CM). Hospital characteristics were calculated from the State Inpatient Database or obtained from the American Hospital Association Annual Survey Database.

Cesarean deliveries were identified with a combination of ICD-9-CM diagnosis and procedure codes as previously described.[12] Discharges were excluded if information on the type of anesthesia provided was missing, the hospital identifier was missing, or a clinical indication for general anesthesia was recorded.

The New York State Inpatient Database is the only Healthcare Cost and Utilization Project participating state providing information on anesthesia care. Anesthesia type is reported as a categorical variable with values corresponding to general, regional, other, local, none, and missing. Each discharge record contains a maximum of one value for anesthesia type. For the purpose of the study, the variable was categorized as general anesthesia, regional (neuraxial) anesthesia, and missing. In this data set, anesthetics are coded hierarchically. For example, a woman who received general anesthesia for cesarean delivery because of a failed epidural catheter would be coded as general anesthesia. To take into account the experience of the anesthesia providers within each hospital in performing and managing neuraxial analgesia/anesthesia to obstetric patients, the annual proportion of women delivering with neuraxial analgesia/anesthesia during labor and vaginal deliveries (the labor epidural analgesia rate) was calculated for each hospital using State Inpatient Database data.

Adverse Events

Five adverse events were analyzed: (1) the composite outcome of death or cardiac arrest, (2) anesthesia-related complications, (3) severe anesthesia-related complications, (4) surgical site infections, and (5) venous thromboembolic events (Supplemental Digital Content Table 2, Death was directly recorded from the State Inpatient Database.

Anesthesia-related complications were divided into three groups: (1) systemic complications, (2) complications related to neuraxial techniques, and (3) complications related to anesthetic drugs. Severe anesthesia-related complications were defined as complications associated with death, cardiac arrest, severe organ dysfunction, or hospital stay greater than the ninety-ninth percentile (7 days; Supplemental Digital Content Table 3, Organ dysfunction variables used to define severe complications reflect concurrent coding in individual cases, and do not establish a causal relationship between anesthesia and organ dysfunctions. Venous thromboembolic events included deep venous thrombosis and pulmonary embolism. Analysis of adverse events was limited to the index hospitalization and did not analyze readmissions.

Patient- and Hospital-level Factors

The following patient-level factors were recorded directly from the State Inpatient Database: age, race and ethnicity, insurance type, admission type (elective or nonelective), and admission for delivery during weekend. In the State Inpatient Databases, Hispanic ethnicity is considered as a distinct racial group. The comorbidity index for obstetric patients and the Charlson comorbidity index were calculated using previously described ICD-9-CM algorithms.[13–15] Preexisting maternal conditions and pregnancy-associated conditions were identified with ICD-9-CM diagnostic codes (Supplemental Digital Content Table 4,

The following hospital-level factors were calculated from the State Inpatient Database: annual proportion of neuraxial techniques during labor and vaginal deliveries, annual volume of delivery, annual proportion of admission during weekend, annual proportion of high-risk pregnancy, and annual intensity of coding. High-risk pregnancies were defined as a comorbidity index for obstetrics patients at or above 2.[15] Intensity of coding was the mean number of diagnosis and procedure codes reported per discharge.[3,16]

The following hospital-level factors were obtained from the American Hospital Association Annual Survey Database: hospital location (rural or urban), teaching status, and neonatal level-of-care designation (1, 2, or 3). Rural hospital location included micropolitan or rural areas based on the Core-Based Statistical Areas. A micropolitan area corresponds to at least one urban cluster that has a population of at least 10,000 but less than 50,000. A teaching hospital had an affiliation to a medical school or residency training accreditation. Neonatal level-of-care 1 hospitals provide basic neonatal level of care, level 2 specialty neonatal care (e.g., care of preterm infants with birth weight at or above 1,500 g), and level 3 subspecialty neonatal intensive care (e.g., mechanical ventilation of 24 h or more).

In the final study sample, patient- and hospital-level factors with a count less than 10 in the general anesthesia group or in the neuraxial anesthesia group were excluded from the analysis.

Statistical Analysis

Statistical analysis was performed with R version 3.4.1 (R Foundation for Statistical Computing, Austria) and specific packages (mice for multiple imputations and lme4 for mixed-effect models). Results are expressed as mean ± SD or count (percent or per 10,000).

Comparison of continuous variables used Student's t test and comparison of categorical variables chi-square test. Missing values were estimated using multiple imputations (Supplemental Digital Content Table 5,

Risk of Serious Adverse Events Associated with General Anesthesia. Unadjusted odds ratios for the five adverse events associated with general anesthesia were calculated using univariate logistic regression. Adjusted odds ratios were calculated using an inverse probability of treatment weighting approach.[17,18] Because we examined five adverse events, we used a Bonferroni correction with a P value threshold for statistical significance of 0.05/5 = 0.01.

The probability of treatment (i.e., general anesthesia) was calculated using a mixed-effect logistic regression model. In this model, the random effect was the hospital identifier (normally distributed intercept and constant slope); the fixed-effects were the year of delivery and patient and hospital characteristics described in the Supplemental Digital Content Table 6, Both the fixed and random effects were used to calculate the individual probability of receiving general anesthesia (i.e., the propensity score).

Inverse probability weights were calculated using the propensity score. Using weights aims to create a synthetic sample in which the distribution of measured baseline covariates is independent of treatment assignment (i.e., general anesthesia). This approach is similar to the use of survey sampling weights that are used to weight survey samples so that they are representative of specific populations. Inverse weights were stabilized and truncated at 1 and 99%. Inverse stabilized weights for women who received general anesthesia were calculated as P(Z = 1)/P(Z = 1|X), where P(Z = 1) is the probability of general anesthesia in the study sample (i.e., prevalence) and P(Z = 1|X) the individual probability of general anesthesia conditional of the set of predictors (i.e., propensity score). Inverse stabilized weights for women who did not receive general anesthesia were calculated as 1−P(Z = 1)/(1−P(Z = 1|X)).

The likelihood of serious adverse event associated with general anesthesia was quantified with the odds ratio from a mixed-effect logistic regression. In this model, the outcome was the examined adverse event, the random effect was the hospital identifier, the fixed effect was the exposure to general anesthesia, and the weight was the inverse stabilized weight.

We reestimated the adjusted odds ratio for the five adverse outcomes examined in two sensitivity analyses after limitation of the study sample to (1) women with a Charlson comorbidity index at or above 1 and a comorbidity index for obstetric patients at or below 1 (with all other inclusion criteria unchanged) and (2) women with a Charlson comorbidity index equals 0 and a comorbidity index for obstetric patients equals 0 (with all other inclusion criteria unchanged).

Temporal Trends in the Use of General Anesthesia. General anesthesia rate was calculated for each 2-year interval of the 12-year study period. The percent change for rates over the study period was calculated as the difference between the rate in 2013 to 2014 and the rate in 2003 to 2004 divided by the rate in 2003 to 2004. The 95% CI for the percent change was calculated. The Cochran–Armitage test for trends was used to assess the statistical significance of changes of rate over time. Trends were assessed overall, according to three hospital characteristics (rural or urban location, teaching status, and annual volume of deliveries) and according to two patients characteristics (Medicaid/Medicare status and race).

Risk Factors for General Anesthesia. To take into consideration the nested nature of patients within hospitals, identification of patient- and hospital-level factors associated with general anesthesia used a mixed-effect logistic regression. In this model, the patient- and hospital-level factors with a P value less than 0.2 in the univariate analysis were included as fixed effects, along with the year of delivery and the intensity of coding. The random effect was the hospital identifier with the assumption of a normally distributed intercept and a constant slope. The selection of variables used a backward procedure with a P threshold of 0.05 for exclusion of variables. Discrimination of the model was assessed with the C index.

The relationship between continuous variables and the probability of general anesthesia use was examined using the relationship between the continuous variable and the logarithm of the probability of receiving general anesthesia. Continuous predictors with a nonlinear relationship with the outcome were categorized; predictors with linear relationship were kept continuous or categorized according to clinically relevant thresholds.