Factors Predicting Complication Rates Following Total Knee Replacement

SooHoo, Nelson F. MD; Lieberman, Jay R. MD; Ko, Clifford Y. MD, MS, MSHS; Zingmond, David S. MD, PhD

Journal of Bone & Joint Surgery - American Volume: March 2006 - Volume 88 - Issue 3 - p 480–485
doi: 10.2106/JBJS.E.00629
Scientific Articles

Background: The purpose of this investigation was to expand on previous studies by more fully examining the role of a variety of patient and hospital characteristics in determining adverse outcomes following total knee replacement.

Methods: With use of data from all hospital admissions in California from 1991 through 2001, multiple logistic regression was performed on the information regarding patients treated with total knee replacement. Rates of mortality and readmission due to infection and pulmonary embolism during the first ninety days after discharge were regressed against a variety of independent variables, including demographic factors (age, gender, race, ethnicity, and insurance type), burden of comorbid disease (Charlson comorbidity index), and provider variables (hospital size, teaching status, and surgical volume). A separate baseline probability analysis was then performed to compare the relative importance of all predictor variables.

Results: The sample size for this analysis was 222,684. A total of 1176 deaths (rate, 0.53%), 1586 infections (0.71%), and 914 pulmonary emboli (0.41%) occurred within the first ninety days after discharge. The average age of the patients at the time of surgery was sixty-nine years. Sixty-two percent of the patients were women, and 32% had a Charlson comorbidity index of >0. The significant predictors for complications (p < 0.05) included age, gender, race/ethnicity, Charlson comorbidity index, insurance type, and hospital volume. A baseline probability analysis was performed with the base case considered to be a white woman who was over the age of sixty-five years, had a Charlson comorbidity index of 0, had Medicare insurance, and was treated at a high-volume, non-teaching hospital. For a patient with the baseline case characteristics, the probability of death was 31/10,000, the probability of infection was 59/10,000, and the probability of pulmonary embolism was 41/10,000 in the first ninety days after discharge. Altering the base case by assuming that care was received at a low-volume hospital increased the expected mortality rate by a factor of 26%. Increasing the Charlson comorbidity index to 1 increased the mortality rate by 170%, whereas decreasing the age to younger than sixty-five years lowered the mortality rate by 73%. Hospital volume, comorbidity, and age had similar effects on the expected rates of readmission due to infection and pulmonary embolism.

Conclusions: The effects of age and the Charlson comorbidity index on the baseline probability of adverse outcomes following total knee replacement were shown to be similar to or greater than the effect of hospital volume. This study elucidates and compares the relative importance of the effects of several different factors on outcome. This information is important when considering the conclusions and implications of this type of policy-relevant outcomes research.

Level of Evidence: Prognostic Level II. See Instructions to Authors for a complete description of levels of evidence.

1 Department of Orthopaedic Surgery, University of California at Los Angeles, 10945 Le Conte Avenue, PVUB #3355, Los Angeles, CA 90095. E-mail address for N.F. SooHoo: nsoohoo@mednet.ucla.edu

2 Department of Surgery, University of California at Los Angeles, School of Medicine, 10833 Le Conte Avenue, Room 72-215, Los Angeles, CA 90095

3 Department of Internal Medicine, University of California at Los Angeles, School of Medicine, 911 Broxton Plaza, Los Angeles, CA 90095

Article Outline

Total knee replacement has been demonstrated to be a highly effective treatment for arthritis that is refractory to conservative management with anti-inflammatory medications, activity modification, and weight loss1. Documenting and improving the quality of care and outcomes following total knee replacement remain priorities2,3. One ongoing controversy in this regard is the relationship between surgical volume and outcome—i.e., whether total knee replacements done at hospitals and by surgeons performing a high volume of the procedures are associated with lower rates of complications4-7. While our research group and others previously reported an association between hospital volume and the rate of complications, there is uncertainty regarding whether other factors have equal or greater importance in predicting outcomes4-7. The purpose of this study was to expand on previous studies by more fully examining the role of a variety of patient and hospital characteristics in determining adverse outcomes following total knee replacement.

The primary patient-based predictors of interest examined in this study were the Charlson comorbidity index, age, race/ethnicity, gender, and insurance status. The hospital characteristics that were analyzed as potential predictors included the volume of total knee replacements performed (hospital volume), teaching status, and size. We examined the relative importance of these factors as predictors of the most common major complications after total knee replacement, including mortality, infection, and pulmonary embolism1.

Back to Top | Article Outline

Materials and Methods

Data Sources

Data for all hospitalizations in California for the years 1991 through 2001 were obtained from California's Office of Statewide Health Planning and Development (OSHPD). The OSHPD Patient Discharge Database is compiled annually and includes discharge abstracts from all licensed nonfederal hospitals in California8. Each discharge abstract contains demographic information including age, gender, insurance type, and the race or ethnicity of the patient. In addition, International Classification of Diseases, Ninth Revision (ICD-9) codes are included for up to twenty inpatient procedures and twenty-four diagnoses per hospitalization. The hospital identifier for the facility where each patient underwent surgery is recorded as well. Using the OSHPD annual hospital financial reports and Council of Teaching Hospitals membership status, we also identified characteristics of the hospitals where total knee replacements were performed. Reported hospital characteristics include the hospital size (number of licensed acutecare beds) and whether it is a teaching hospital.

The hospital discharge data from OSHPD were linked to the California State Death Statistical Master File to identify rates of mortality due to all causes9. This allowed the identification of deaths occurring after discharge and the time elapsed prior to death of patients who underwent primary total knee replacement. This file is a database of death certificates for all individuals who died in California and for California residents who died outside of California's borders but within the United States9.

Back to Top | Article Outline

Sample

Patients who had undergone primary total knee replacement during the study period were identified on the basis of the ICD-9 procedure code for primary total knee arthroplasty (81.54). A previously published algorithm was used to exclude patients with infection, pathologic fracture, or codes suggestive of a previous arthroplasty4. This algorithm was modified to account for coding changes made during the study period. Specifically, ICD-9 codes that were modified during the observation period were identified and the changes were incorporated into the coding algorithm10.

Back to Top | Article Outline

Patient and Provider Characteristics (Independent Variables)

The primary patient-based predictors of interest were age, race/ethnicity, gender, insurance status, and comorbidity as reported in the OSHPD database. Comorbidity was assigned a Charlson comorbidity index, which is based on assessments of seventeen comorbid conditions. The Charlson comorbidity index has been validated for use in administrative database studies11,12. Hospital characteristics that were analyzed as potential predictors included volume of total knee replacements performed (hospital volume), teaching status, and size. Hospital size was categorized in one of four groups based on the number of beds (<100, 100 to 199, 200 to 299, or >299).

Hospital volume was defined as the average number of primary total knee replacements performed yearly during the study period. Hospitals were categorized as low-volume if their annual volume was in the lowest 40th percentile among hospitals at which total knee replacements were performed. They were categorized as intermediate-volume if the annual volume was in the next 40th percentile and as high-volume if it was in the highest 20th percentile. These definitions of low, intermediate, and high-volume were used in a previous study, which demonstrated significant differences in outcomes among patients on the basis of these categories6.

Back to Top | Article Outline

Identification of Outcomes (Dependent Variables)

The clinical outcomes examined were the rates of mortality, readmission due to infection, and readmission due to pulmonary embolism. These outcomes were tracked for up to one year after discharge by analyzing hospital admission records and state death records. Previously published algorithms were adapted to detect codes consistent with readmission due to infection at the site of the prosthesis or pulmonary embolism4. These algorithms were modified to correct for coding changes made during the study period10. Mortality was identified by linking the California State Death Master Statistical File to the OSHPD database9.

Back to Top | Article Outline

Statistical Analysis

Descriptive Statistics

All statistical analyses were conducted with use of Stata/SE 8.0 software (Stata, College Station, Texas, 2003). The patient sample was analyzed with use of descriptive statistics of patient and provider characteristics and of patient outcomes. The patient characteristics were age, gender, Charlson comorbidity index, insurance type, and race/ethnicity. The hospital characteristics were volume of total knee replacements, teaching status, and size.

Back to Top | Article Outline

Logistic Regression Analysis

The outcomes that were analyzed as dependent variables were the mortality rate and the rates of readmission due to the specific complications of infection and pulmonary embolism during the first ninety days after discharge. Multiple-variable logistic models were used to examine the relationship between each patient and provider variable and each outcome. Separate regression models were used for each dependent variable, and each model included the patient characteristics (race/ethnicity, age, gender, insurance type, and Charlson comorbidity index) and provider characteristics (hospital size, surgical volume, and teaching status). The strength of the association between the risk of complications and the patient and provider characteristics is reported as the odds ratio in relation to a reference group. The threshold for significance was a p value of <0.05. All statistical analyses were conducted with use of Stata/SE 8.0 software.

Back to Top | Article Outline

Baseline Probability Analysis

A baseline probability analysis was performed with use of the equation: 1/1 + e^ (b0 + b1 × 1 + b2 × 2 +... + bp × p)13. In this analysis, the relative importance of different variables in predicting complications was evaluated by using probability and risk difference instead of odds ratios. The analysis involves defining a base-case scenario and subsequently adding individual predictor variables into the analysis. The relative effect of each added variable is assessed by calculating the marginal risk associated with each added variable. The predicted mortality per 10,000 is also reported.

The base-case values were selected to be consistent with the most common values for the patient and hospital characteristics in the overall sample. Specifically, the baseline values used for the predictor variables were an age of greater than sixty-five years, female gender, a Charlson comorbidity index of 0, white race, Medicare insurance, treatment at a high-volume hospital, and treatment at a non-teaching hospital. Subsequent separate analyses were conducted to calculate the change in the risk of mortality and complications per 10,000 cases when age was assumed to be less than sixty-five years. Similar analyses were conducted to examine the effect of altering the variables of gender (to male), Charlson comorbidity index (to >0, 1, 2, or >2), race/ethnicity (to black, Hispanic, or Asian), insurance type (to private or Medicaid), hospital volume (to intermediate or low), and hospital status (to teaching hospital). For each variable, the risk was calculated in comparison with the baseline probability.

Back to Top | Article Outline

Results

Patient Demographics

We identified 222,684 patients who had had a total knee replacement during the observation period. A total of 1176 deaths (a rate of 0.53%), 1586 infections (0.71%), and 914 pulmonary emboli (0.41%) occurred within the first ninety days after discharge. The average age of the patients at the time of surgery was sixty-nine years. Sixty-two percent of the patients were women, and 32% had a Charlson comorbidity index of >0. A total of 154,563 patients were over the age of sixty-five years. The majority (65%) of patients had Medicare insurance and were white (80%). A summary of the demographic characteristics of the sample is presented in the Appendix.

Back to Top | Article Outline

Hospital Characteristics

The majority (58%) of patients were treated at hospitals that performed a high volume of total knee replacements. Of a total of 413 hospitals, eighty-three were designated as high-volume, 165 were designated as intermediate-volume, and 165 were designated as low-volume. The mean number of total knee replacements performed annually (and standard deviation) was 147 ± 47 at the high-volume hospitals, 50 ± 15 at the intermediate-volume hospitals, and 13 ± 5 at the low-volume hospitals. The largest proportion of knee replacements (42%) was performed at large hospitals with more than 299 beds. The teaching status of the treating hospital was recorded for 222,662 of the patients in the sample. There were thirty-one teaching hospitals, at which 21,829 (10%) of the knee replacements were performed during the observation period. A large majority of patients (200,833; 90%) were treated at non-teaching hospitals.

Back to Top | Article Outline

Multivariate Logistic Regression

The final regression model showed the most consistent associations with complications to be age, Charlson comorbidity index, and hospital volume. There were also significant associations between race and the likelihood of readmission due to pulmonary embolism or infection, but the patterns were not as consistent. Specifically, black race was associated with a higher odds ratio for pulmonary embolism within ninety days after discharge, and Hispanic ethnicity was associated with a higher odds ratio for infection. In addition, Medicaid insurance was associated with a higher odds ratio for infection but showed no consistent association with mortality or pulmonary embolism. Similarly, males had a higher odds ratio for mortality but a lower odds ratio for infection. There was no significant relationship between gender and readmission for pulmonary embolism (Table I).

Back to Top | Article Outline

Baseline Probability Analysis

In the base case, the procedure was assumed to be performed at a non-teaching high-volume hospital, in a white female patient over the age of sixty-five who had no comorbidities (Charlson comorbidity index of 0) and was insured by Medicare. These base-case parameters were selected to be consistent with the most common values for the patient and hospital characteristics in the overall sample. The results of the base-case analysis are summarized in Table II. Baseline probability analysis showed the probability of mortality for the base case to be 0.31%, which is consistent with the 0.53% mortality rate observed for the total cohort. The probability of infection for the base case was 0.59%, compared with an overall probability of 0.71%, and the probability of pulmonary embolism for the base case was 0.41%, compared with 0.41% overall.

Particular variables were selectively added to the analysis to determine the incremental probabilities of death that would occur. For example, increasing the Charlson comorbidity index from 0 to 1 resulted in a 161% increase in the probability of mortality, from 0.31% for the base case to 0.81%. An increase in the Charlson comorbidity index to >2 increased the probability of mortality by 519%, to 1.92%. Similar results were seen for the rate of infection, which increased from the base-case value of 0.59% to 1.18% when the Charlson comorbidity index was assumed to be >2. Decreasing age to less than the base-case assumption of sixty-five years or older lowered the probability of mortality by a factor of 73% and the probability of pulmonary embolism by 34% but increased the probability of infection by 46% during the ninety-day observation period.

Decreasing the hospital volume also resulted in a sizeable change in risk compared with that for the base case. Changing the base-case assumption of a high-volume hospital to that of a low-volume hospital increased the rate of mortality (from 0.31% for the base case to 0.39%), infection (from 0.59% to 0.86%), and pulmonary embolism (from 0.41% to 0.56%) (Table II).

Back to Top | Article Outline

Discussion

Total knee replacement is widely accepted as a highly cost-effective procedure that improves the quality of life of patients with arthritis1. Despite this consensus, there is ongoing controversy with regard to the practical implications of the known role of hospital volume in predicting outcomes. Several authors have used administrative databases to establish the relationship between hospital volume and the outcomes of orthopaedic surgical procedures, including total knee arthroplasty. Katz et al. reported an association between hospital and surgeon volume and rates of mortality and complications during the first ninety days following total knee replacement in Medicare patients5. In a prior study of data from California over an eleven-year period, we also linked hospital volume to mortality and readmission rates after total knee arthroplasty6. Similar results were shown by Hervey et al., in a nationwide sample of patients4. Kreder et al. demonstrated an association between low provider volume and an increased rate of revision arthroplasty7. The present study expands on these prior findings by directly addressing the relative importance of hospital volume in relation to other patient and hospital characteristics in predicting outcomes following total knee replacement.

In the current study, age and the Charlson comorbidity index showed significant associations with the rates of mortality and readmission due to infection. Age was also associated with the rate of readmission for pulmonary embolism. Both the odds ratios and the effects on the baseline probability of these variables were similar to or greater than the odds ratios and the effect on the baseline probability of hospital volume. Our findings also indicate an association between black race and the rate of pulmonary embolism as well as between Hispanic ethnicity and the rate of infection. The overall magnitude of the race effect was smaller than the effect of age, comorbidity, and hospital volume. Previous studies have also demonstrated differences among racial and ethnic groups with regard to the risk of adverse outcomes following medical treatments14-17. These results are useful in determining which patient populations are at increased risk for specific complications. Identifying these different risks is a necessary first step to improving the quality of care and the outcomes in diverse patient populations.

The results of this study indicate that regionalizing total knee replacement to high-volume centers may decrease mortality from approximately thirty-nine cases per 10,000 to thirty-one cases per 10,000 and generate similar theoretical changes to the rates of infection and pulmonary embolism. The cost-effectiveness of this approach requires further study. In addition, an Institute of Medicine approach has argued that the more important underlying issue in the volume-outcome debate is quality of care. More specifically, high volume may be a proxy for high-quality processes of care. Thus, an alternative approach could be to focus on identifying and disseminating the practices that lead to lower complication rates at high-volume hospitals. These may include such factors as hospital resources, clinical pathways, and other specific processes used to deliver care at these high-volume institutions18,19.

A major limitation of this study is that the use of readmission and death records may result in underestimation of morbidity and mortality rates if complications are not coded properly or do not require hospitalization. However, the validity of this approach is supported by the fact that the complication rates reported in this study are consistent with those found both in previous clinical series and in other studies in which administrative databases were used1,4-7. Another limitation of our study is that the OSHPD statewide database does not include information on long-term functional outcomes. As a result, we were unable to evaluate the relationship between the predictor variables and functional outcome. Similarly, we did not examine the relationship between these predictors and long-term rates of revision. Kreder et al. demonstrated a link between provider volume and the rate of revision arthroplasty for up to three years postoperatively7. Future clinical studies are required to better understand the role of patient and hospital characteristics in predicting both functional outcomes and rates of revision. Despite these limitations, the use of administrative databases can provide novel insights, at the population level, regarding outcomes. Such observations will be important for benchmarking, improving quality and safety of care, and generating hypotheses for future intervention-related studies focusing on the factors that were shown to be associated with better outcomes.

There were also limits on our ability to identify confounding variables such as surgeon volume and training. Information on surgeon volume was not available and could not be evaluated separately from hospital volume. The studies by Katz et al.5, Hervey et al.4, and Kreder et al.7 suggest that both surgeon volume and hospital volume are independently associated with outcomes following total knee replacement. These previous studies suggested that the importance of surgeon volume as a predictor of outcome is similar to that of hospital volume. However, we were not able to directly address the issue of the relative importance of hospital and surgeon volume.

We reported on the roles of a variety of patient and hospital characteristics in predicting rates of mortality and readmissions due to infection and pulmonary embolism following total knee replacement. The results of this study can be used to improve quality of care by identifying patient and provider factors that increase the risk of these complications. The effects of age and the Charlson comorbidity index on the baseline probability of these outcomes were found to be similar in magnitude to the effect of hospital volume. Race was also shown to have a role as a predictor of complication rates. These results suggest the need for further study to determine the relative importance and underlying causes of these differences among populations. Future studies of these predictive factors would be facilitated by more complete data sources that include long-term functional outcomes.

Back to Top | Article Outline

Appendix

A table showing patient demographics and summary statistics for the patients and hospitals studied is available with the electronic versions of this article, on our web site at jbjs.org (go to the article citation and click on “Supplementary Material”) and on our quarterly CD-ROM (call our subscription department, at 781-449-9780, to order the CD-ROM). ▪

A commentary is available with the electronic versions of this article, on our web site () and on our quarterly CD-ROM (call our subscription department, at 781-449-9780, to order the CD-ROM).

The authors did not receive grants or outside funding in support of their research for or preparation of this manuscript. They did not receive payments or other benefits or a commitment or agreement to provide such benefits from a commercial entity. No commercial entity paid or directed, or agreed to pay or direct, any benefits to any research fund, foundation, educational institution, or other charitable or nonprofit organization with which the authors are affiliated or associated.

Investigation performed at the Department of Orthopaedic Surgery, University of California at Los Angeles, Los Angeles, California

1. , White RE Jr; Council of Musculoskeletal Specialty Societies, American Academy of Orthopaedic Surgeons. What's new in adult reconstructive knee surgery. J Bone Joint Surg Am. 2004;86: 1839-49.
2. , Maloney WJ, Wright JG. An AOA critical issue. The outcome of the outcomes movement. J Bone Joint Surg Am. 2004;86: 633-40.
3. . National joint replacement registries: has the time come? J Bone Joint Surg Am. 2001;83: 1582-5.
4. , Purves HR, Guller U, Toth AP, Vail TP, Pietrobon R. Provider volume of total knee arthroplasties and patient outcomes in the HCUP-nationwide inpatient sample. J Bone Joint Surg Am. 2003;85: 1775-83.
5. , Barrett J, Mahomed NN, Baron JA, Wright RJ, Losina E. Association between hospital and surgeon procedure volume and the outcomes of total knee replacement. J Bone Joint Surg Am. 2004;86: 1909-16.
6. , Zingmond DS, Lieberman JR, Ko CY. Primary total knee replacement in California 1991-2001: does hospital volume affect outcomes? J Arthroplasty. In press.
7. , Grosso P, Williams JI, Jaglal S, Axcell T, Wal EK, Stephen DJ. Provider volume and other predictors of outcome after total knee arthroplasty: a population study in Ontario. Can J Surg. 2003;46: 15-22.
8. 1996 Discharge Data File Format Documentation—Confidential Layout. Sacramento, CA: Office of Statewide Health Planning and Development; 1997.
9. , Ye Z, Ettner SL, Liu H. Linking hospital discharge and death records—accuracy and sources of bias. J Clin Epidemiol. 2004;57: 21-9.
10. Centers for Disease Control and Prevention, National Center for Health Statistics. National Hospital Discharge Survey: 2002 Public Use Data File Documentation. Hyattsville, MD: United States Department of Heath and Human Services; 2004. p 19-48.
11. , Parsons GA, Ghali WA. Validity of information on comorbidity derived from ICD-9-CCM administrative data. Med Care. 2002;40: 675-85.
12. , Cherkin DC, Ciol MA. Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases. J Clin Epidemiol. 1992; 45: 613-9.
13. , Chang JT, Chaudhry S, Kominski G. Are high-volume surgeons and hospitals the most important predictors of in-hospital outcome for colon cancer resection? Surgery. 2002;132: 268-73.
14. , Henderson MC. Risk factors for venous thromboembolism after total hip and knee replacement surgery. Curr Opin Pulm Med. 2002; 8: 365-71.
15. , Romano PS, Zhou H, Rodrigo J, Bargar W. Incidence and time course of thromboembolic outcomes following total hip or knee arthroplasty. Arch Intern Med. 1998;158: 1525-31.
16. , Hull RD, Patel KC, Olson RE, Ghali WA, Alshab AK, Meyers FA. Venous thromboembolic disease: comparison of the diagnostic process in blacks and whites. Arch Intern Med. 2003;163: 1843-8.
17. , Beardmore TD. Risk factors for early wound complications after orthopedic surgery for rheumatoid arthritis. J Rheumatol. 1995;22: 1844-51.
18. , for the Committee on Quality of Health Care in America and the National Cancer Policy Board. Interpreting the volume-outcome relationship in the context of health care quality: workshop summary. Washington, DC: Institute of Medicine; 2000.
19. . Explorations in quality assessment and monitoring. Volume 1, The definition of quality and approaches to its assessment. Ann Arbor, MI: Health Administration Press; 1980.
Copyright 2006 by The Journal of Bone and Joint Surgery, Incorporated