This study utilized de-identified data from the Institute of Clinical Evaluative Sciences, an independent research institute that houses the population-based health administrative data for the province of Ontario, Canada. The population of Ontario at the time of this study was approximately 12 million. Ethical oversight was provided and approved on a programmatic basis by the Ethical Review Board at Sunnybrook Health Sciences Centre (Toronto, ON, Canada). We conducted a population-based historical cohort study in Ontario where all hospital and physician services are funded through a public healthcare system. We used population-based health administrative data that were collected using standardized disease classification, procedural terminology,28 and abstraction formats.29 All data were linked deterministically using encrypted, patient-specific identifiers. Databases used for the study included the Same-day Surgery (SDS) database, which records all ambulatory surgeries; the Discharge Abstract Database (DAD), which captures all inpatient hospitalizations; the National Ambulatory Care Reporting System (NACRS), which captures all ED visits; the Ontario Health Insurance Plan database, which captures physician service claims; and the Registered Persons Database (RPDB), which captures all death dates for Ontarians. This study conforms to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement.30
We identified all Ontario residents aged ≥ 40 yr who underwent knee surgery, shoulder surgery, inguinal or femoral hernia repair, cholecystectomy, lumpectomy, or transurethral resection of urinary obstruction (see specific procedures and associated codes in Appendix A; available as Electronic Supplementary Material). These surgeries are the most common non-ophthalmic or non-endoscopic ambulatory surgeries performed in the United States.31 Surgeries were identified from the SDS, which specifically records all planned ambulatory surgeries in the province of Ontario (i.e., those in which hospital discharge is planned within 24 hr of surgery).
We conducted a patient-level analysis, studying only the first surgery for each patient in our observation window. Patients < 40 yr of age were excluded to increase outcome rates, since age is a known predictor of adverse outcomes after surgery.32 We chose 2002 as the start year for our study as it coincides with the introduction of the International Classification of Diseases, 10th revision (ICD-10), which we used to identify diagnoses. We used the Canadian Classification of Health Interventions to identify procedures. The latest year in which all datasets were complete was 2012. Patients were excluded if they did not have a valid provincial health insurance number or if their 30-day observation window extended beyond the study end date. Our primary analysis was limited to individuals who were successfully discharged on the day of surgery, defined as discharge within 24 hr of surgery. Our secondary analysis involved individuals who did not achieve successful hospital discharge.
Our primary outcome was the composite of hospital readmission or an ED visit within 30 days of successful discharge following ambulatory surgery. The postoperative disposition status of each patient was derived from the SDS record and was categorized as discharged home or admitted to hospital (including admission to a critical care area or to another operating room, admitted to hospital as an inpatient, transferred to another acute care facility, transferred to another non-acute care facility, or died in the operating room). Unanticipated postoperative admissions immediately following surgery were verified by cross-referencing the DAD to confirm the recording of an inpatient hospitalization on the day of planned ambulatory surgery. Post-discharge hospital admission was defined as the creation of a record in the DAD within 30 days of discharge from hospital. All ED visits were identified from the NACRS within 30 days of discharge from hospital. All-cause mortality was captured from the RPDB, which captures both in- and out-of-hospital deaths. The specific date of each outcome was recorded.
Patient demographics were collected from the RPDB. Validated algorithms were used to determine if patients had congestive heart failure,33 hypertension,34 a history of acute coronary syndromes,35 diabetes mellitus,36 asthma,37 or chronic obstructive pulmonary disease.38 Previously described methods were used to identify the following comorbidities based on ICD-9 and ICD-10 codes from the DAD in the two years preceding surgery39: atrial fibrillation or flutter; dementia; hemiplegia or hemiparesis; cerebrovascular disease; primary (excluding non-melanoma skin cancer) malignancy; metastatic tumours; peripheral vascular disease; renal disease; dialysis; liver disease; peptic ulcer disease; rheumatologic disease; and venous thromboembolism. Emergency department visits and acute care hospitalizations in the year prior to surgery were obtained from the NACRS and the DAD, respectively. Finally, the Mortality Risk Score40 was used to estimate each patient’s risk of death. The Mortality Risk Score is based on the Johns Hopkins Adjusted Clinical Groups (ACG®) System and measures the one-year risk of all-cause death with excellent discrimination (i.e., c-statistic 91.7%) and calibration (i.e., the absolute difference between the observed probability of death and the mean predicted probability of death was less than 0.01 in 97 of 100 centiles).
The day of the week on which surgery took place was determined from the SDS. We identified only those patients who had a record for surgery on a weekday since elective ambulatory surgery is not commonly performed on the weekend. We used a five-level main exposure variable to account for each weekday, using Monday as the reference category.
Outcome and main exposure status were complete for all patients without missing data. Covariate data, if missing for less than 1% of patients, were imputed with the group mean. Covariates missing more than 1% of data were excluded.
Demographic characteristics were compared between each day of the week using Chi square tests for categorical variables, analysis of variance for normally distributed continuous variables, and the Kruskal-Wallis test for continuous variables that were not normally distributed.
The primary analysis consisted of time-to-event analysis using Cox proportional hazards regression. The primary outcome was the first occurrence of a visit to the ED or a hospital readmission following successful discharge after ambulatory surgery (i.e., the patient achieved discharge within 24 hr). Our time window spanned from the day of discharge to 30 days after discharge, at which time all remaining participants were censored. We did not consider death to be a substantive competing risk in our analysis because mortality following ambulatory surgery is rare. We accounted for clustering using a robust estimate of standard error based on the local health network as the unit of clustering. The local health network was chosen because we thought this would best address the nature of readmissions and ED visits after discharge from ambulatory surgery. Such events could happen at the index hospital or another hospital, but patients would be unlikely to leave their local health network. Adherence to the assumption of proportional hazards was tested using the significance of covariate-time interactions, and the hazard functions were also inspected. Since demographic variables, comorbid illnesses, and health system characteristics may confound the true association between surgical day and outcomes, we included age (used as a restricted cubic spline with five knots), sex, hospital type (teaching vs non-teaching), health system usage in the year prior to surgery, patient neighbourhood income quintile, patient residence (urban vs rural), comorbid diseases, the Mortality Risk Score (used as a continuous linear predictor), year of surgery (used as a categorical variable representing each calendar year), and surgery type. An interaction term between outcome and procedure was tested in the model to ensure the appropriateness of reporting a single effect measure for all procedures.
We performed several sensitivity analyses (using adjusted proportional hazards regression and the five-level exposure variable accounting for each weekday). We examined the impact of day of the week for each specific procedure, each time period (2002-2007 and 2008-2012), and each individual component of the primary composite outcome, as well as over a seven-day outcome period (an analysis which was performed post hoc). Finally, for models where the main exposure variable (i.e., surgical day of the week) did not adhere to the proportional hazards assumption (based on covariate time interactions and inspection of the stratified hazard function), we constructed a post hoc proportional hazards regression model that contained a time-surgical day of the week interaction term. We determined the optimal continuous representation of time using fractional polynomials.41 This allowed us to calculate the relative hazard of the primary outcome between Monday vs Friday surgery and to illustrate and explore the possible time-dependent nature of surgical day of the week on ED visits and readmission. As a secondary analysis, we used a multivariable logistic regression model that adjusted for the same covariates as our primary analysis and used the health region as the unit of clustering to measure the association between day of the week and failure to be discharged from hospital following surgery. We also calculated the number and proportion of ED visits and readmissions following successful discharge from ambulatory surgery that occurred on each day of the week (including Saturday and Sunday).