Overview

We conducted a population-based cohort study of all patients receiving an elective primary hip or knee replacement in Alberta, Canada, a province of ~4 million people, with a single healthcare system, over a 3-year period. Data for all patients was linked to an infection prevention and control (IPC) database which accurately identifies all patients who develop a joint infection within 90 days of arthroplasty and to health administrative data that capture patient comorbidities and healthcare costs.

Data sources

The Alberta Bone and Joint Health Institute (ABJHI) database was used to identify all patients in Alberta who received a primary hip or knee replacement between April 1, 2012, and March 31, 2015. This database captures information on patients undergoing orthopedic surgeries in Alberta. If more than 1 primary arthroplasty was performed on a patient in the study period, only the initial arthroplasty on the first replaced joint was considered when creating the baseline cohort.

To assess for joint infections, we used data from the Alberta Health Services (AHS) IPC group. Their surveillance population includes all adults ≥18 years of age admitted to any public Alberta hospital facility for arthroplasty in urban and nonurban sites. Data for all complex SSIs found within 90 days of arthroplasty from April 2012 to March 2015 were collected. Cases of SSI were detected using several mechanisms: electronic review of microbiology laboratory results, review of patient charts including observation of the incision, physician record and pharmacy data, reoperation records, readmissions, emergency visit records, and clinic visit records. Provincial surveillance is centralized using a web-based data entry platform. An IPC Data Quality Working Group began completing an additional review of provincial admission and disease administrative codes in April 2013 to ensure that no cases of SSI following hip and knee arthroplasty were being missed across the province. We also used this database to capture the causative organism of the infections.

Data from the IPC database and ABJHI database was linked to Discharge Abstract Database (DAD) records to obtain additional information around admissions secondary to infected arthroplasties (eg, length of stay [LOS]) in addition to the surgical procedures themselves. Patient comorbidities, categorized according to the Elixhauser index, were collected using diagnosis codes. This index includes 30 different patient conditions and is a predictor of in-hospital mortality.Reference Fabbian, De Giorgi and Maietti ¹⁷

Microcosting data, described below, were obtained from corporate data for the major urban zones (Calgary and Edmonton). Costs for zones outside the major urban centers were based on gross costing methods, obtained from Alberta Health Services Division of Analytics, Data Integration, Measurement, and Reporting.

Cohort

We created a cohort of adults aged 18 years or older who received elective primary hip or knee replacements in the province of Alberta between April 1, 2012, and March 31, 2015. Age, sex, First Nations status, outmigration, and death were collected from the Alberta Health registry for all patients in the cohort. This cohort was then divided into those who developed a complex SSI within 90 days of arthroplasty (the infected cohort) and the remaining noninfected cohort. Because we were interested in the costs of complex SSI, the small proportion of superficial SSI were included in the noninfected cohort.

Outcomes

The primary outcome was 12-month cumulative healthcare costs, including index admission, hospitalizations and emergency room, day surgery, and day medicine costs. As secondary outcomes, we also determined 24-month cumulative healthcare costs because we hypothesized that the impact of a complex SSI on costs might continue even after the infection had been treated (typically within the first year), and number of days in hospital over the first 12 months in the infected versus the noninfected group. Costs associated with admissions to hospital included costs incurred during the admission (surgical procedures and days in hospital). Hospital costs were measured using microcosting, the gold standard for obtaining costing estimates, where each component of resource use is estimated and a unit cost derived for each.Reference Shrive, Ghali, Donaldson and Manns ¹⁸ The resource use was then tracked to specific patients. Because microcosting data were only available for inpatient hospitalizations within the large urban hospitals in Alberta, we used microcosting data for inpatient hospitalizations for all patients in Calgary and Edmonton. We used gross costing methods based on resource intensity weights provided by the DAD records for the emergency room visits, day medicine, and day surgery visits as well as the inpatient hospitalizations outside of Calgary and Edmonton.Reference Shrive, Ghali, Donaldson and Manns ¹⁸ Gross costing methods were used for ~30% of hospitalizations across Alberta. While these methods are not as accurate as microcosting, they still provide a reasonable overview of patient costs.

Costs related to other physician claims and outpatient antibiotics were not available. We used the perspective of the healthcare payer, therefore, patient-borne costs were not considered. All costs were inflated to 2016 Canadian dollars using the consumer price index. Where US dollars are included, the January, 2016 exchange rate was utilized.

Statistical analyses

As our primary method, we assessed costs for all of the patients’ hospitalization and emergency room, day surgery, and day medicine visits over the subsequent 12 and 24 months. In a complementary analysis, to better estimate the incremental cost of the infection itself, we only included the cost of hospitalizations and emergency room, day surgery, and day medicine visits for admissions with International Classification of Disease, Tenth Revision (ICD-10) codes that were likely to represent infection. These were determined by tracking the infected cohort of patients over time and identifying the 6 most commonly used ICD-10 codes related to infection: T8453 (infection and inflammation reaction due to hip prosthesis), T8454 (infection and inflammation reaction due to knee prosthesis), Z508 (rehabilitation care), T814 (infection following a procedure), T847 (infection and inflammatory reaction due to other internal orthopedic prosthetic devices, implants and grafts), and T8403 (mechanical complication of hip prosthesis).

Once the mean costs were determined for the infected and noninfected cohorts, they were compared to each other using Mann-Whitney U tests. To adjust for differences in patients developing infections (ie, age, gender, and comorbidities), the association between having a complex SSI and 1-year costs was determined using multiple variable linear regression, with the goal of determining an attributable cost of complex joint infection. A linear regression model using ordinary least squares estimation was used based on its interpretability and ability to model the mean costs for each group. While ordinary least squares estimation assumes a normal distribution of the residual values, if the sample size is large enough, then this assumption being violated does not impact the magnitude of coefficients when compared with other regression methods. ¹⁹ When compared to other linear models including log total costs with smearing retransformation, negative binomial (γ), and inverse Gaussian distributions, ordinary least squares performed well on measures of concordance, mean absolute error, root mean squared error, and probability plots. All statistical analyses were conducted using Stata version 14 software (StataCorp, College Station, TX).

The University of Calgary Health Research Ethics Board approved this study.