A number of risk factors for deep postoperative infection after knee arthroplasty have been found in clinical case-control and cohort studies1-3. Clinical data allow in-depth analysis of a variety of patient-related factors and, for example, diabetes mellitus, rheumatoid arthritis, and other comorbidities as measured with the American Society of Anesthesiologists score carry an increased probability of septic complications1-4. Although they provide detailed patient-level data, various surgery-related and provider-related factors often cannot be reliably studied in clinical investigations because of the limited number of surgeons, techniques, and prosthesis types. Additionally, because of the contemporary low infection rates, there is a risk of obtaining false-negative results in statistical analyses.
Even though local and national arthroplasty register data have been used to study factors associated with the survival of primary and revision knee replacements5-8, to the best of our knowledge, there have been no recent studies involving arthroplasty register-based data that have evaluated the risk factors for infection. Previous operations on the involved knee are associated with an increased infection rate1,2, but we are aware of no reports on factors affecting the risk of infection after revision knee replacement.
The purpose of the present study was to determine the risk factors for infection following primary and revision knee replacement in a large series of knee arthroplasties from the Finnish Arthroplasty Register.
All primary and revision knee arthroplasties (including isolated patellar resurfacing procedures and exchanges of a part of a knee prosthesis) that were performed in Finland from January 1997 to June 2004 and registered in the Finnish Arthroplasty Register were included in the study. The first-stage procedures of staged exchange arthroplasties (that is, resection arthroplasties) and operations for which the type of procedure was not registered were excluded. Each knee was followed until the end of 2004, resulting in a minimum duration of follow-up of six months. The data of the Finnish Arthroplasty Register are based on mandatory reporting on all joint replacement operations performed in Finland; the register has good coverage, but its data have not yet been scientifically validated9,10. For the present study, the Finnish Arthroplasty Register data were supplemented by data on hospitalization, collected from the Finnish Hospital Discharge Register.
Reoperations
Reoperations performed for the treatment of infection were ascertained by combining data from the Finnish Arthroplasty Register and the Finnish Hospital Discharge Register. The technique for using such a combination of two registers to detect reoperations has been described and discussed elsewhere10. In the Finnish Arthroplasty Register data, any reoperation performed for the treatment of infection was considered to be evidence of a deep knee infection. In the Finnish Hospital Discharge Register, the surgical procedure code11 indicating débridement, removal of the prosthesis, change or addition of any prosthesis component, any revision arthroplasty, arthrodesis, or amputation, together with an ICD-10 (International Classification of Diseases, Tenth Revision) diagnosis code suggestive of infection, were required to indicate a septic end point event. All other reoperations were considered to have been performed for reasons other than infection. If two or more reoperations were recorded for an index operation, the one occurring first was included for analysis. If the operatively treated side was unknown and both knees of a patient were being followed, both were excluded from further follow-up.
Statistical Analyses
All data were analyzed with the knee as the statistical unit of analysis, and primary and revision procedures were analyzed separately. The proportion of knees that had a reoperation because of infection to the total number of operatively treated joints (the infection rate) is described for different subgroups, with the 95% confidence intervals calculated according to the method of Wilson12. For continuous variables, the median and the range are reported. Comparisons of explanatory variables between different operation types were performed with use of the chi-square test or the Fisher test for categorical variables and with use of the Kruskal-Wallis test for continuous variables.
Cox regression analysis with reoperation for the treatment of infection as the end point was used for the analysis of risk factors. Besides reoperation for the treatment of infection, follow-up was considered to have ended at death, any reoperation performed for a reason other than infection, and, for the remaining patients, on December 31, 2004.
The effect of each explanatory variable on the rate of reoperations for the treatment of infection was tested in a univariate model. To ensure that the proportional hazards assumption was not violated, each Cox-predicted survival curve for the type of operation, diagnosis, prosthesis constraint, and the type of antibiotic prophylaxis was plotted on a corresponding Kaplan-Meier survival curve (Figs. 1 through 4), and the congruity of predicted and observed survival curves was graphically assessed. Except for the later years of follow-up in several subgroups within the revision arthroplasty group (patients with other illness, a cementless prosthesis, no antibiotic prophylaxis, or antibiotic cement only), the predicted and observed curves were found to be equal, demonstrating that the Cox model adequately followed the observed survival.
In addition to the univariate analysis, each variable was tested in the Cox model with adjustment for age, sex, and diagnosis as recommended by Robertsson et al.7. Additionally, the type (constraint) of the implanted prosthesis (unicondylar, cruciate-retaining or cruciate-substituting total condylar, constrained total condylar, or hinged knee prosthesis) was included as a proxy for the severity of joint destruction in the adjusted analyses because it had been found in an earlier study to affect prosthetic survival3. Hazard ratios along with 95% confidence intervals calculated with the adjusted Cox model are reported. Finally, all variables with a significant (p < 0.05) association with infection in the adjusted analysis were entered into the adjusted forward stepwise Cox model to analyze their relative importance.
The level of significance was set at p < 0.05. For sensitivity analysis, the adjusted analyses were performed with the inclusion of only the reoperations that were registered in the Finnish Arthroplasty Register.
Source of Funding
External funding from independent nonprofit organizations was received for this study. None of the funding sources played any role in the preparation or performance of this study.
The Finnish Arthroplasty Register provided data on 43,149 operations. The procedures included 40,135 primary total knee arthroplasties (93.0%), 2166 revision total knee arthroplasties (5.0%), and 848 partial revision arthroplasties (including 618 secondary patellar resurfacing procedures, 198 isolated exchanges of a tibial insert, and thirty-two revisions of either the femoral or tibial component alone) (Table I). Corresponding hospitalization data in the Hospital Discharge Register could be matched to Finnish Arthroplasty Register data in 95.7% of the cases.
Table I describes the detailed demographic data on the study population. Operative and perioperative details are presented in the Appendix. Same-day arthroplasty of the contralateral knee was performed in 7.6% of all cases. A patellar component was implanted in association with 30.5% and 44.1% of the primary and revision arthroplasties, respectively. Antibiotic-impregnated cement was used for the fixation of at least one prosthetic component in 84% of the knees. With few exceptions, all patients received antibiotic prophylaxis, for which intravenous cefuroxime was most often used.
The populations of patients undergoing primary and revision arthroplasties were essentially similar with respect to age, sex, and diagnosis, although some qualitatively small but significant differences were observed (Table I) (see Appendix).
Reoperations
In total, 387 reoperations for the treatment of infection (0.90%; 95% confidence interval, 0.81% to 0.99%) were registered during the follow-up period (average, 3.1 years; range, zero to 8.6 years), and 262 (68%) of them were performed within one year after the index procedure.
Partial revision arthroplasties (adjusted hazard ratio, 3.4; 95% confidence interval, 2.2 to 5.5) and revision total knee arthroplasties (adjusted hazard ratio, 4.7; 95% confidence interval, 3.6 to 6.3) were associated with significantly higher risks of septic failure in comparison with primary knee replacement without patellar resurfacing (Fig. 1).
Risk Factors for Infection
Table II presents the results of the adjusted analyses for primary and revision knee arthroplasties. The results of the univariate analyses and the proportions of the knees that had a reoperation because of infection are presented in the Appendix.
Sex, diagnosis, the type of prosthesis, the use of antibiotic cement for prosthetic fixation, and postoperative complications were found to be the strongest factors that showed a significant association with postoperative infection following primary knee replacement in the adjusted analyses (Table II). With regard to revision arthroplasties, the strongest determinants of subsequent repeat revision because of infection were sex, the use of antibiotic cement for prosthetic fixation, postoperative complications, and the reason for revision (infection).
Patient-Related Factors
Male patients had a higher overall risk of reoperation for infection (hazard ratio, 1.6; 95% confidence interval, 1.3 to 2.0) as compared with female patients after both primary and revision knee replacement. The difference was independent of age and also was seen in the subgroup of patients with primary osteoarthritis (data not shown).
There was an increased rate of reoperations because of infection following primary arthroplasties performed for the treatment of arthritic diseases other than primary osteoarthritis (see Appendix). The ICD-10 diagnoses that were associated with an increased risk of reoperations for the treatment of infection, as compared with the diagnosis of primary osteoarthritis, were posttraumatic osteoarthritis (hazard ratio, 2.4; 95% confidence interval, 1.3 to 4.2), unspecified osteoarthritis (ICD code M17.9) (hazard ratio, 2.7; 95% confidence interval, 1.3 to 5.4), seropositive rheumatoid arthritis (hazard ratio, 1.7; 95% confidence interval, 1.1 to 2.6), and fracture around the knee (hazard ratio, 6.3; 95% confidence interval, 2.0 to 20.0). After one year, the number of septic failures in patients with rheumatoid arthritis did not differ significantly from that in patients with primary osteoarthritis (hazard ratio, 1.3; 95% confidence interval, 0.6 to 3.0) (Fig. 2).
Type of Prosthesis
In general, there was a trend showing an increased rate of infections in association with constrained and hinged prostheses in comparison with nonconstrained total knee prostheses (Fig. 3), but in the adjusted analyses the trend was significant only for primary arthroplasties (Table II). Although the rate of infections was the lowest after primary unicondylar knee arthroplasty, unicondylar prostheses did not survive significantly better than nonconstrained total condylar prostheses did in the subgroup of knees with primary osteoarthritis (hazard ratio, 0.6; 95% confidence interval, 0.2 to 1.5).
Antibiotic Prophylaxis
After primary knee arthroplasty, fewer infections were seen when antibiotics were administered both intravenously and impregnated in the cement used for prosthetic fixation (combined antibiotic prophylaxis) than were seen when either method of administration was used alone (Table II). The lack of use of antibiotic-impregnated cement (hazard ratio, 1.42; 95% confidence interval, 1.08 to 1.88) had a more dramatic effect than did the lack of use of intravenous antibiotics (hazard ratio, 1.13; 95% confidence interval, 0.42 to 3.04) in comparison with the use of combined antibiotic prophylaxis. This finding also was seen after revision arthroplasty (corresponding hazard ratios, 2.12 [95% confidence interval, 1.14 to 3.92] and 1.85 [95% confidence interval, 0.45 to 7.65]). Despite clear differences between different types of antibiotic prophylaxis in terms of the infection rates after revision knee arthroplasty (Fig. 4), a significant difference was observed only in the comparison between intravenous antibiotics only and combined antibiotic prophylaxis (Table II).
Postoperative Complications
The occurrence of any postoperative complication reported to the Finnish Arthroplasty Register (n = 486 complications in 459 knees) was associated with an increased risk of reoperation for the treatment of infection (hazard ratio, 7.2; 95% confidence interval, 4.8 to 10.8) in comparison with uncomplicated cases. The highest rates of reoperation for the treatment of infection were related to anesthetic complications (11.8% [two of seventeen]; 95% confidence interval, 3.3% to 34.3%), wound infection (21.9% [fourteen of sixty-four]; 95% confidence interval, 13.5% to 33.4%), wound hematoma (13.9% [five of thirty-six]; 95% confidence interval, 6.1% to 28.7%), and wound necrosis (14.3% [seven of forty-nine]; 95% confidence interval, 7.1% to 26.7%). In uncomplicated cases, infection resulted in reoperation in 0.82% (95% confidence interval, 0.74% to 0.91%) of cases.
Revision Arthroplasties
There were more reoperations for the treatment of infection after revision arthroplasties performed within two years after the preceding operation (hazard ratio, 1.7; 95% confidence interval, 1.1 to 2.9) than after revisions performed later. Infection accounted for 33% of these early revisions (p < 0.001 for the differences in the reasons for revision between early and late revisions) and was strongly associated with the risk of repeat revision for the treatment of infection (hazard ratio, 3.0; 95% confidence interval, 1.5 to 6.0). Other reasons for revision were not associated with reoperation for the treatment of infection.
Note: The authors thank the Finnish Arthroplasty Register and the Finnish Hospital Discharge Register of the National Research and Development Centre for Welfare and Health (Stakes) for providing the materials for this study.