Chart review and quantification of baseline comorbidity
We conducted a retrospective observational cohort analysis of all patients who underwent surgery for hip fracture repair in the Mater Misericordiae University Hospital, Dublin, Ireland between January and June 2017 inclusively. Patients were identified for inclusion in the study through the Hospital In-Patient Enquiry health information system, which generates a basic summary of administrative, clinical and demographic data whenever a patient is discharged from hospital or dies in hospital in Ireland. Patients were therefore included whether they presented through the emergency department from home, or were in-patients with other diagnoses who sustained hip fractures in hospital, as well as those transferred from nursing homes or long term care facilities. Some patients were transferred from other healthcare facilities without a trauma orthopaedic service. All patient clinical data were collected and stored on secured hospital computers with individual password protection for each investigator. Institutional review board approval was obtained for this retrospective analysis.
Clinical records were retrieved for the identified patients, and were analysed for demographic and clinical data including age, gender, preoperative comorbidities, ASA status, length of stay in hospital, comorbidities during the first 30 postoperative days, use of general or spinal anaesthesia, and use of non-invasive or invasive blood pressure monitoring. We described the cumulative preoperative comorbidity of our cohort using the Charlson Comorbidity Index; this internationally validated weighted scoring system stratifies patients according to expected in-hospital mortality based on their comorbid condition, assigning expected one-year mortality rates accounting for number and severity of comorbidities.13 Data were then correlated to intraoperative events stored electronically and accessed through Centricity High Acuity Anesthesia (GE Healthcare, Chicago, Illinois, United States), which is the anaesthesia information management system (AIMS) used in our institution to automatically record intraoperative events and measurements of patient vital signs throughout anaesthesia and recovery.
Assessing duration of hypotension
Blood pressure measurements were retrieved and analysed for each case from the time of induction of anaesthesia until the time the patient was transferred to the post-anaesthesia care unit. We applied several definitions for intraoperative hypotension:
- cumulative time during which SAP was ≤80% of the initial measurement taken when the patient was clerked into hospital with a diagnosis of hip fracture;
- cumulative time during which SAP was ≤80% of the last SAP measurement recorded prior to induction of anaesthesia;
- cumulative time during which mean arterial pressure (MAP) was <75 mmHg; and
- cumulative time during which MAP was <55 mmHg.
Where invasive arterial pressure monitoring was used, recordings had been logged by the AIMS every minute, and where non-invasive monitoring was used, recordings had been logged as per the clinical judgement of the anaesthesiologist caring for the patient, usually every three to five minutes. For each of our definitions for intraoperative hypotension, we defined duration of hypotension as the time between successive blood pressure measurements during anaesthesia where the patient’s most recent recorded blood pressure met the respective criteria.
Quantification of postoperative morbidity
The patients’ clinical records were analysed to ascertain the incidence of complications postoperatively during the first 30 postoperative days, or until discharge from hospital or death if this occurred sooner. We used a postoperative complications survey modified from that described by Bennett-Guerrero et al.14 Complications of all major organ systems were included: cardiac arrhythmia, myocardial infarction, sepsis, respiratory compromise, thromboembolic events, cerebrovascular events, new cognitive impairment, confusion, coma, renal impairment, liver dysfunction, need for transfusion, need for intensive care, wound dehiscence, pain limiting early post-operative mobilisation, and death.
Postoperative morbidity data was expressed in terms of the Clavien-Dindo classification and Comprehensive Complication Index. The Clavien-Dindo classification qualitatively ranks the severity of a postoperative surgical complication according to the extent of corrective therapy required.15 We recorded the highest Clavien-Dindo classification associated with each patient based on the single most severe complication they encountered during the first 30 postoperative days. The Comprehensive Complication Index is based on the Clavien-Dindo classification, and allows a patient’s overall morbidity to be quantified on a scale from 0 to 100 by integrating multiple weighted Clavien-Dindo classifications for different complications into a composite number.16
Statistical analysis
The dependent variables chosen for the prediction models were the Comprehensive Complication Index, which was treated as a continuous outcome variable, and the Clavien-Dindo index, which was treated as an ordinal outcome variable. Because of the limited number of cases per level, particularly at the higher end of the Clavien-Dindo index, the levels 3a, 3b, 4a, 4b and 5 were collapsed to a new level “≥3”.
Univariate summary statistics were calculated for all independent and dependent variables.
The association between hypotension and post-operative complications were analysed by multivariable regression. As dependent variables the collapsed Clavien-Dindo classification and the Comprehensive Complication Index were chosen. The four previously mentioned cumulative time of hypotension variables of interest were included in analysis as continuous independent variables. Additional independent variables were also included due to their previously established association with post-operative complications:
- age at surgery;15,16
- gender;16,17
- preoperative morbidities as composed in the Charlson comorbidity index or the ASA status.16,18
To analyse the association between independent variables and post-operative complications a proportional odds models were chosen for the Clavien-Dindo index. The proportional odds assumption of the models was tested using a visual method described by Harrell and the likelihood ratio statistic.38,39 Linear models were chosen for the Comprehensive Complication Index. Relative parsimony between base and experimental models was determined using the corrected Akaike’s Information Criterion (AICC).19,20 Likelihood ratio tests were used to compare goodness-of-fit between models. P value of 0.05 or less was taken to indicate statistical significance for hypothesis testing. Statistical analyses were performed using the statistical software package R (R Development Core Team, Vienna, Austria).21