Analysis of the relationship between surgeon procedure volume and complications after total knee arthroplasty using a propensity-matched cohort study

Study sample

We used health administrative databases from Ontario, Canada. The main data sources were hospital discharge abstracts from the Canadian Institute for Health Information Discharge Abstract Database, physician claims from the Ontario Health Insurance Plan (OHIP), and demographic information on each physician from the Ontario Physician Human Resources Data Centre and OHIP Corporate Provider Database. Using specific procedure and diagnostic codes from the Canadian version of the 10th revision of the International Statistical Classification of Diseases and the Canadian Classification of Health Interventions (ICD-10-CA/CCI), we defined a cohort of patients who received their first primary elective TKA for osteoarthritis between April 1, 2002 and March 31, 2016. Patients or the public were not involved in the design, or conduct, or reporting, or dissemination plans of our research.

Outcomes of interest: revision within three years of surgery

The primary outcome of this study was the occurrence of a revision arthroplasty (for any cause) within 3 years of surgery. Revision arthroplasty was defined placement of new components following failure of some or all of a failed primary implant. The secondary outcome was the occurrence of deep infection requiring surgery within 3 years. Primary TKAs that involved the use of computer navigation, patient-specific instrumentation or robotics were also excluded. Revision procedures (for any cause) were identified using ICD-10-CA/CCI procedure codes accompanied by the supplementary status attribute ‘R’. Deep infections requiring surgery were identified using two methods: (1) occurrence of an ICD-10-CA diagnostic code for intra-articular infection, with a confirmatory code for an irrigation and debridement and/or (2) occurrence of an OHIP code for a spacer insertion. Deep infections that led to a revision (irrigation and debridement as a first stage), as well as irrigation and debridement with implant retention were also included.

Covariates of interest

We measured several patient and provider covariates that have been previously shown to affect the risk of occurrence of complications following knee replacement. Patient age and sex was obtained from the OHIP Registered Persons Databases.4–6 Comorbidities listed on hospital discharge abstracts in the 3 years before the index admission were categorized according to an adaptation of the Charlson Comorbidity Index.7 The Johns Hopkins Adjusted Clinical Groups (ACG) ‘frailty’ indicator, based on diagnosis codes from hospitalizations and physician visits in the 2 years before the index admission for TKA, was used to determine the presence of ‘frailty’ (yes/no) at the time of the surgery.8 We identified patients with a history of pre-existing coronary artery disease (CAD), diabetes, hypertension, chronic obstructive pulmonary disease (COPD), and dementia using validated algorithms (see online supplemental appendix 19–17). This study used administrative data available at the Institute for Clinical Evaluative Sciences (ICES). As Ontario, Canada has a single-payer healthcare system, ICES is able to capture every interaction a patient has within this system. Several algorithms comprised of ICD-10 codes and provider billing codes have been used to create cohorts of patients with various comorbidities. The algorithms used in this study have been validated with chart abstraction to determine their sensitivity, specificity and positive predictive value (online supplemental appendix 19–17). We also identified patients that had been previously counseled on smoking cessation through appropriate physician billing codes.

Neighborhood income quintile was used as a surrogate for socioeconomic status. This measure categorizes small geographic areas into five roughly equal population groups, with the lowest quintile referring to the least affluent neighborhoods.18 19 The Ontario marginalization index was also assessed. This comprises four elements: ethnic concentration, residential instability, dependency, and deprivation.20 Each element is sorted into fifths, arranged from least (lowest fifth) to most marginalized (highest fifth).20 The index has been shown to be stable across time periods and across different geographic areas and to be associated with health outcomes including depression, smoking, alcohol consumption, and body mass index.20 For each TKA, hospital volume was defined as the number of primary knee arthroplasty procedures (both primary and revision) performed at the hospital where the surgery was performed in the 365 days prior to the index procedure. Unicompartmental knee arthroplasties (UKA) were excluded, as was any revision of a UKA to a TKA. Surgeon experience was operationalized as the number of years since completion of residency/fellowship orthopedic training, as evidenced by the start of Royal College accreditation and start of independent practice, for the senior surgeon at the time of the index surgery. This includes foreign surgeons moving to Canada with substantial experience, but recent Royal College accreditation.

Main exposure variable: surgeon volume

For each TKA, surgeon volume was defined as the number of TKA procedures (both primary and revision) performed by the senior surgeon in the 365 days prior to the index procedure. This definition allowed for an individual surgeon’s volume to change dynamically over time. Only revisions of primary TKAs were included with revisions of UKAs to TKAs excluded.

Statistical analyses

Baseline cohort characteristics were calculated using proportions and medians as appropriate, and were compared between groups using Wilcoxon rank sum tests for continuous variables and χ² tests for categorical variables. A propensity score for receipt of a TKA from a surgeon with the determined threshold value was calculated using a logistic regression model.21 22 The covariates entered into the propensity score were sociodemographics (age, sex, income quintile, Ontario marginalization index), health status (Charlson score, frailty, hypertension, Chronic Obstructive Pulmonary Disease (COPD), Congestive Heart Failure (CHF), diabetes, Coronary artery disease (CAD), smoking status), provider characteristics (annual hospital volume), and surgeon experience.

Restricted cubic splines with four knots23 were used to model the relationship between surgeon volume and the occurrence of revision and infection after controlling for patient age, sex, income quintile, comorbidities, hospital volume and surgeon experience. Restricted cubic splines are a flexible tool to model complex, non-linear relationships between a continuous variable and an outcome.23 Normal regression analysis assumes a linear relationship between the predictor and outcome variables. This would suggest that the impact of an increase in surgeon volume would be the same if the increase in volume was from 10 to 30 cases/year, or from 210 to 230 cases/year. A spline, on the other hand, does not make any assumptions of linearity. It divides the relationship into smaller ‘chunks’, allowing for portions that are not linear. In the current study, use of a spline also allowed us to identify a threshold of volume at which the greatest benefit of surgeon volume is obtained. The non-linear relationship between surgeon volume and the risk of revision was examined to identify an inflection point, if any, which could be used to dichotomize annual volume into categories in a clinically meaningful way. If an area of inflection was observed, multivariable logistic regression was used to determine the area under the curve (AUC) for the models relating various surgeon volume cut-points to the risk of revision. The surgeon volume with the maximum AUC was selected as the cut-point to dichotomize surgeon volume.

TKA recipients from a surgeon who had operated below the selected threshold value in the preceding year were matched to those from a surgeon with more than the threshold number of cases on the logit of the propensity score using calipers of width equal to 0.2 of the SD of the logit of the propensity score.24 A matching ratio of 1:1 was used.25

We estimated standardized differences for all covariates after matching, with a standardized difference of 10% or more considered indicative of imbalance.25 The occurrence of complications were also compared between the two groups after matching, using methods appropriate for the analysis of matched data in estimating the treatment effect and its statistical significance. Generalized estimating equations were used to determine the increased risk (if any) in patients with surgeons whose volumes were lower than the selected threshold, after taking pair-matching into account.23 All analyses were performed at the Institute for Clinical Evaluative Sciences (http://www.ices.ca) using SAS V.9.3 for UNIX (SAS Institute, Cary, North Carolina, USA). The type I error probability was set to 0.05 for all analyses.

Patient and characteristics

Between April 1, 2002 and March 31, 2016, there were 169 713 eligible TKA recipients (table 1), with a median age of 68, and 62.5% were of female gender. The median annual surgeon volume was 89 cases (IQR 59–127), and the median number of years in practice was 15 years (IQR 8–24).

Regression splines describing the relationship between surgeon volume and risk of complications

The restricted cubic splines relating annual surgeon volume to revision and infection had similar shapes—both were negatively sloped with inflection points at approximately 70 cases/year, after which the rates of complications continued to decrease with increased surgeon volume, but at a lower rate (figures 1 and 2). The AUC was determined for various cut-points of surgeon volume (60, 65, 70, 75, 80, 85, 90). Subsequently, surgeon volume was dichotomized at 70 cases (<70 or ≥70 cases in the 365 days prior to the procedure). A total of 56 265 procedures (33%) were performed by ‘low-volume’ surgeons over the study period.

Request reuse permission

Request reuse permission

Matching

A total of 51 658 subjects (92%) who received a TKA from a surgeon with <70 cases in the 365 days prior to the surgery were successfully matched to a TKA recipient from a surgeon with ≥70 cases in the 365 days prior to the surgery (table 2). There were no exclusions to the propensity-score match. Patients that did not complete follow-up to 3 years postoperatively because of death (n=4374 (2.57%)) were not excluded from the match. A patient flow diagram is illustrated in figure 3. After matching, the absolute standardized differences were less than 10% for all variables entered into the propensity score, indicating an adequate match (table 2).

Request reuse permission

Outcomes after matching

Knee replacement recipients from a surgeon with low annual volumes (<70 cases in the year prior) had a higher rate of revision within 3 years (2.23% (95% CI 1.39 to 3.07) vs 1.70% (95% CI 0.85 to 2.55); HR 1.33, 95% CI 1.21 to 1.47, p<0.0001) and infection requiring surgery within 3 years (1.29% (95% CI 0.44 to 2.14) vs 1.09% (95% CI 0.24 to 1.94); HR 1.33, 95% CI 1.17 to 1.51, p<0.0001) (table 3). The number needed to harm for revision was 189 persons (95% CI 143 to 278), which translates to an absolute risk increase of 0.53% (95%CI 0.37% to 0.71%).