Study Design and Study Population
This prospective cohort study is nested within a larger implementation project on TKA/THA service delivery. The aim of the implementation project was to describe current practice and identify deficiencies in delivering an arthroplasty service, and to investigate the association between mobilising early after surgery and acute length of stay at hospital. The implementation project included all patients undergoing elective primary THA or TKA at Fairfield Hospital in Sydney, Australia between August 2018 and May 2019.The current sub-study was designed to describe current investigation practices and analyse the association between complications and the number and cost of imaging and pathology investigations performed during the acute post-operative period following THA and TKA. There were no further exclusion criteria for inclusion in the acute-care analysis. Patients provided informed consent to an investigator to have their data reviewed by research personnel.
Data Sources
For the implementation study, research personnel collected patient demographic, anthropometric, comorbid and procedure information. These data were collected directly from the patient during their pre-admission visit and from the hospital medical record.
Using unique patient identifiers including Medical Record Number (MRN), which is assigned by a hospital or facility, and surgery date, we extracted imaging and pathology data for each hospital admission from the electronic medical record. The imaging data contains information on each imaging test (also called procedure), including the unique procedure identifier, procedure name, time, unique patient identifier and the associated Medicare Benefits Schedule (MBS) code(s), which identify the medical services subsidised by the Australian government including associated fees (11). The pathology data contains the same information without the associated MBS codes. We obtained cost information for pathology tests by matching each test from the record to descriptions in the 2019 MBS Book (11). When descriptions in the extracted data did not accurately match those listed in the MBS, we consulted the department managers of radiology and pathology, as well as the clinical nurse consultant to ensure valid cost estimations. We costed all imaging and pathology tests at 100% of the MBS schedule fees (see Additional files 1 and 2).
Exposure
The exposure variable was the presence of a complication. A complication was defined as any medical, physical or surgical deviation from the normal post-operative course (12, 13). Prior to analyses, major and minor complications were classified as “minor” or “major” based on the invasiveness of intervention required to treat the complication, whether it resulted in a change of functional status, and whether it usually prolongs hospitalisation (12, 13) (see Additional File 3).
The study population was categorised into three groups based on complication status, which describes the presence and/or severity of complications experienced during the acute hospital stay. The first group experienced no complications during their stay; the second group experienced minor complications only; and the third group experienced at least one major complication, with or without minor complications.
Outcome
The primary outcome was the cost of imaging and pathology tests received by each patient, by exposure category. Secondary outcomes included the number and type of tests. The timeframe of measurement was the acute post-operative period, starting on the day of surgery and ending when the patient was discharged from hospital or discharged from the surgical ward to in-hospital rehabilitation.
Data Analysis
For descriptive statistics, we used one-way analysis of variance (ANOVA) to compare continuous variables and chi-squared tests to compare categorical variables by group. After assessing distributional assumptions, we used negative binomial models to estimate the relationship between complication status and the number of tests, as the number of tests received during a stay followed a count distribution with overdispersion (14). We used linear regression models to estimate costs by exposure group. Although the distribution of cost was skewed, we chose to use a linear model without transformation in order to reflect more accurately the effects of outliers on the total cost for the health system (15).
We risk-adjusted all models for known and suspected confounders, including age, sex, body mass index (BMI), procedure, operation time and co-morbidities, including anxiety and depression, cancer (past and current), diabetes mellitus, dementia, hypertension, hyperlipidaemia, hyperthyroidism, osteoporosis, urinary incontinence, chronic urinary tract infection, and autoimmune, cardiac, chronic respiratory, cerebrovascular, central nervous system, liver, renal, gastro-oesophageal reflux disease, and past venous thromboembolism. We used R Version 3.6.0 (www.r-project.org) to conduct all analyses (16).