Days alive and out of hospital for adult female and male cardiac surgery patients: a population-based cohort study

This study is reported according to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines and the Reporting of Studies Conducted Using Observational Routinely Collected Health Data (RECORD) statement. We conducted a retrospective cohort study using population-based administrative healthcare databases in Ontario, Canada. The use of data in this project was authorized under Sect. 45 of Ontario’s Personal Health Information Protection Act, which does not require review by a research ethics board. We used CorHealth Ontario to identify cardiac surgeries and specific cardiac health information. The Registered Persons Database (RPDB), Vital Statistics and Ontario census data were used to extract demographics, socioeconomic status and mortality. The Canadian Institute of Health Information Discharge Abstract Database (CIHI-DAD) captures all acute care hospital admissions and provided information additional information during primary hospitalization, and any readmissions. We used the Ontario Health Insurance Plan (OHIP) database to capture all physician service claim data. The Continuing Care Reporting System (CCRS) database was used to capture patients needing long-term care support. Specialized databases (Ontario Diabetes Database, Asthma Database, Chronic Obstructive Pulmonary Disease Database, Ontario Hypertension Database) were used to identify specific comorbidities. Laboratory values were obtained from the Ontario Laboratory Information System (OLIS). Data were linked through unique anonymized patient identifier numbers. Key variables and codes used are summarized in Supplemental Digital Content Tables S1 [11, 18, 19].

We identified adults (≥ 18 years) patients who underwent common elective and emergency cardiac surgeries between 2009 – 2019 in Ontario hospitals. The procedures included coronary bypass surgery (CABG), aneurysectomy, valve repair/replacement, and aortic surgery. We excluded less commonly performed surgeries such as adult congenital heart procedures, heart transplantation and ventricular assist device implantation. The surgical procedures were placed into 3 clinically sensible groups based on operative complexity and commonly described description: (i) isolated CABG, (ii) single non-CABG procedures (e.g., single valve repair/replacement), and (iii) combined (2 or more) procedures (e.g., valve and CABG surgery, multiple valve surgery, aortic and CABG/valve surgery) [5, 7]. We excluded intraoperative deaths (n = 219) in order to assess DAH in the postoperative period, patients with missing unique identifier number or death date (n = 75), patient procedure dates and institution identified in DAD but not verified in CorHealth data (n = 5,350). For patients with multiple surgeries during the study period, we excluded all but the first procedure (n = 5,673).

The primary outcome was DAH at 30 days after surgery (referred to as DAH₃₀). This was calculated using mortality, hospital length of stay, and readmissions between the date of the index surgery and the 30th postoperative day using validated sources from CIHI-DAD (supplemental figure S1) [20]. The approach to calculating DAH has been previously described. In brief, the duration a patient stays in hospital is subtracted from the measured time frame, for example, a patient who survived and was discharged 20 days after the indexed surgery had a DAH₃₀ of 10 days. Patients who died at any time during this 30-day period were assigned a DAH₃₀ of 0 days. The secondary outcomes were DAH at 90 days (DAH₉₀) and 180 days (DAH₁₈₀), which were determined using similar calculations.

Demographics (age, sex) were identified from the RPDB. Comorbidities (coronary artery disease, diabetes, hypertension, chronic obstructive pulmonary disease, atrial fibrillation, asthma, body mass index, stroke, chronic liver disease, smoking status, anemia, left ventricular ejection fraction, components of the Charlson comorbidity index score) were extracted from CorHealth, OLIS, CIHI-DAD (using ICD-10 codes from hospital admissions) and specialized validated Ontario databases within 3 years before the index surgery [21‐24]. Severity of preoperative kidney dysfunction was classified into one of 5 stages of KDIGO renal function based on the patient’s estimated glomerular function prior to surgery [25]. To capture level of patient acuity and sickness before surgery, we recorded those patients needing preoperative intensive level care and DAH in the 3 months preceding the day of indexed surgery. Frailty was estimated using the hospital frailty risk score [26]. Socioeconomic status was based on neighborhood income quintile (1 is lowest, 5 is highest) and extracted from StatsCan. Major (grade 3–4 Clavien-Dindo) complications within 30-days after surgery needing ICU readmission, reoperation, rehospitalization, or another advanced intervention (e.g., pacemaker, angioplasty, dialysis, tracheostomy, prolonged ventilation, prolonged ICU admission, balloon pump, endoscopic procedure, extra-corporeal support) were extracted from CorHealth, CIHI-DAD, OLIS and OHIP [27‐29]. Surgery variables include procedure type, duration, and urgency. Hospital variables included surgical volume, teaching status and bed number [20].

Descriptive statistics were used to initially compare male and female patients using frequency (proportion) for categorical variables, median (interquartile range, IQR) for continuous variables and standardized difference. Subsequently, men and women were studied and reported separately. For each sex-group, descriptive statistics were estimated for each of the three surgical groups (single CABG, single non-CABG, combined surgeries). Construct validity describes how DAH responds to patient and surgical risk factors. We expect DAH to show convergent validity with lower DAH values in patients undergoing more complex procedures or display higher burden of chronic diseases. Hence, to evaluate construct validity of patient level factors and type of surgery, and how this differs for female and male patients, we summarized the unadjusted effect of common patient comorbidities (e.g., diabetes, atrial fibrillation, stroke, body mass index, chronic obstructive pulmonary disease, chronic kidney and liver disease, smoking status, socioeconomic status, and surgery type) on DAH at 30-days using median (IQR). To study the adjusted association of patient, surgery and hospital factors with DAH, we used a multivariable median regression model to model the association of covariates (with the median DAH [30]. This approach has been previously used to manage the skewed nature of our data [9, 16]. The model incorporated hospital-specific random effects to account for within-hospital clustering. Separate models were developed for male and female patients for DAH at 30, 90 and 180 days using the above covariates. A sensitivity analysis was performed after removal of complications from the model. To formally evaluate whether there was an interaction between patient sex and modifiable risk factors (e.g., hospital teaching status, procedure group), we fit the above model to the entire sample of men and women combined and included an interaction term between patient sex and the given risk factor. This was done sequentially for one risk factor at a time. Risk adjusted models were performed on a cohort of 87,826 patients during the study time frame after removal of missing variables (rural 0.1%, income quintile 0.3%, surgery duration 0.2%, left ventricular function 3%, body mass index 5%, smoking status 2%, kidney function 14%). No imputation of data was performed.

We explored the prognostic implications of our sickest patients with the fewest number of DAH at 30 days. After removing patients who died during this 30-day period, patients were ranked based on their value of DAH at 30 days, and then placed into 2 groups—those in the lowest 10th percentile and those above this percentile. The 10th percentile cut-off has been previously used for non-cardiac surgery to capture those patients with the poorest number of DAH, and is appropriate given the left skewness of the data distribution [11]. Patient, surgical and hospital characteristics of those below and above the 10th percentile were quantified using median (IQR), frequency (percentage) and standardized differences. We subsequently determined the proportion of patients in below and above the 10th percentile at 30 days that remained within these group at 90 and 180 days.

The trajectory and morbidity of individual cardiac surgeries subtly vary. For example, in the single non-CABG group, outcomes may be different for men and women undergoing mitral valve versus tricuspid valve surgery. To further explore this, we performed pre-specified sub-group analyses within the isolated non-CABG and combined procedure surgical groups to study these differences in operations between patient sex. This was performed using the same above risk adjusted model for the outcomes of DAH at 30 days.

All analyses were conducted using Microsoft Excel (v.2010, Redmond, WA), SAS version 9.4 (SAS Institute, Cary, US) and R statistical software [31‐33]. Two-sided p-values < 0.05 were considered statistically significant. No statistical power calculation was performed prior to conducting this study and the sample size was based on the available data meeting the above eligibility requirements. This sample was based on our previous experience in conducting health services research using this patient population and research design [19, 34].