This was a single-center retrospective cohort study conducted in Singapore General Hospital (SGH). SGH is an 1800-bedded tertiary academic hospital in Singapore. The study was granted an exemption by SingHealth’s Centralized Institutional Review Board (CIRB Reference number 2021/2496), with a waiver of patient consent and waiver for approval due to the use of anonymized routinely collected clinical data. All research was performed in accordance with relevant institutional guidelines and regulations.
Our study cohort was extracted from the Singapore General Hospital’s Perioperative and Anesthesia Subject Area (PASA), a curated electronic medical records database containing the clinical records of all operative procedures performed under anesthetic care in the institution since 2013. [9] (Chiew et al. 2020) Information on patient demographics, anthropometric parameters, and preoperative comorbidities had been assessed by the attending anesthetist as part of the structured clinical notes for routine clinical care and were included. All laboratory results for 30 days prior to the procedure and 7 days after were also available. Laboratory investigations were conducted in our institution’s College of American Pathologists accredited clinical laboratory. Hematological parameters were processed with the Sysmex XN Automated Hematology Analyzer (Sysmex Corporation, Kobe, Japan) and ADVIA 2120i Hematology System (Siemens Healthcare Diagnostics Inc, Malvern, PA, USA) while biochemical parameters were processed using the Roche Cobas c501 analyzer (Roche Diagnostics). Intraoperative data were collected as part of the routine electronic Anesthesia Information Management System (Mortality data in the system was synchronized with the National Electronic Health Record, which includes mandatory registration of all deaths), thus ensuring near-complete all-cause mortality follow up). A full list of the extracted variables can be found in Appendix 1.
Inclusion Criteria
Adult patients aged 18 and above undergoing elective abdominal surgeries (gastrointestinal, colorectal, hepatopancreaticobiliary, urological, gynecological) under general anesthesia between 2013 and 2020 in the PASA dataset were included in this study. No patients with COVID-19 were included.
Exposure
The primary independent variable was the use of remifentanil intraoperatively. The induction and maintenance of general anesthesia in our institute are anesthetist-dependent and not performed based on a standardized protocol. The decision to use remifentanil too was made by the individual anesthetic specialist without any standardized dosing protocol. However, popular indications for remifentanil use include the presence of renal impairment, obstructive sleep apnea and other comorbidities leading to concerns with opioid accumulation, and expected large swings in haemodynamics requiring precise titration of ongoing anaesthetic agents. We included remifentanil infusions in mcg/kg/min as well as ng/ml by target controlled infusion (Minto model).
Missing data were cross-checked and corrected. Missing data that could not be corrected were coded as a dummy missing variable. To minimize the risk of false associations being drawn between remifentanil treatment and the outcomes investigated in our retrospective observational study, we used propensity score matching (PSM) to reduce the distribution of measured baseline covariates between our treatment and control subjects, simulating attributes of a randomized controlled trial.
Outcome Measures
The primary outcomes were length of stay in hospital, high dependency (HD)/ intermediate care area (ICA) and intensive care unit (ICU). Secondary outcomes included all-cause death at 90-days and 2 years post-surgery.
In our institution, HD wards nurse patients who are seriously ill but do not require intensive care. Our ICA wards are a step above the HD wards, and nurse critically ill patients who do not yet need invasive mechanical ventilation.
Statistical Analysis
We aimed to investigate the effect of remifentanil on the following outcomes:
1. Average length of stay, the average number of days that patients spend in hospital post-abdominal surgery
2. Length of stay in the High Dependency and Intermediate Care Area, wards that manage patients with complex conditions requiring more medical attention than a general ward
3. Intensive care unit (ICU) admission rates and hours spent in the ICU
4. Death at 3 months and 2 years post-surgery
Propensity Score Matching and Association Testing
To model the propensity score, we identified 12 potential confounders and performed the matching of treatment and control groups based on a generalized linear model. Matched pairs were formed by nearest-neighbor matching, with 1:1 matching without replacement and a caliper width of 0.1. The level of balance between the treatment of control groups was evaluated based on a comparison of pre-matching and post-matching statistical data including visual analysis and evaluating standardized differences.
Using the matched data, we tested outcome differences between the treatment and control groups with the Poisson regression model for continuous outcomes and chi-square test for categorical data. In view of the multiple comparisons being performed, the Bonferroni correction was applied to decrease the risk of type 1 error.
All analysis, statistical computing and visualizations were carried out in the R environment version 4.3.1 (The R Foundation for Statistical Computing, Vienna, Austria). The Love Plot (Figure 2) and Propensity-adjusted density plot (Figure 3) were performed with the “Cobalt” R library package [10] (Greifer 2021).