Advertisement

Abstract

Background and Purpose. Decreasing hospital stays for patients with total knee arthroplasties (TKAs) have a direct effect on rehabilitation. The identification of modifiable determinants of postsurgical functional status would help physical therapists plan for discharge from hospitals. The purpose of this study was to identify preoperative determinants of functional status after a TKA. Participants. Using a community-based, prospective cohort study, data were collected from 276 patients who received a primary TKA in a Canadian health care region. Data were collected in the month before surgery and 6 months after surgery. Methods. Function was measured using the function subscale of a disease-specific measure—the Western Ontario and McMaster Universities (WOMAC) Osteoarthritis Index—and a generic health status measure—the Medical Outcomes Study 36-Item Short-Form Health Survey (SF-36). Independent variables examined included demographic variables (eg, age, sex), medical variables (eg, diagnosis, number of comorbid conditions, ambulatory status), surgical variables (eg, type of implant, number of complications), and knee range of motion. Results. At 6 months after surgery, the average WOMAC physical function score was 70.5 (SD=18.2) and the average SF-36 physical function score was 44.8 (SD=25.3). Using multiple regression analyses, baseline function, walking device, walking distance, and comorbid conditions predicted 6-month function (WOMAC: R2=.20; SF-36 physical function: R2=.27). Discussion and Conclusion. Patients who have lower preoperative function may require more intensive physical therapy intervention because they are less likely to achieve functional outcomes similar to those of patients who have less preoperative dysfunction.

The utilization rates of elective total knee arthroplasties (TKAs) are steadily increasing with an aging population.1 Moreover, the trend toward earlier hospital discharge after TKA has meant that patients are returning home during a more acute phase of recovery. These 2 factors have had direct implications for the rehabilitation of patients with TKA.

Elective TKA is, more often than not, the last effort in managing joint pain and dysfunction caused by arthritis. Extensive evidence indicates that the majority of patients who have had a TKA report improvement in pain and function.24 Eighty-five percent to 90% of patients with TKA report pain relief after surgery, and 70% to 80% report functional improvement.4,5 The greatest amount of improvement is seen within 3 to 6 months after surgery, with more gradual improvements occurring up to 2 years after surgery.68 A meta-analysis of 130 studies4 indicated that these favorable results continue over time. This meta-analysis showed that 89.3% of patients reported good to excellent results at an average follow-up period of 4.1 years. The mean improvement in range of motion in those studies in which preoperative and postoperative range of motion of the knee was measured was 8 degrees.4

Although the improvements following TKA can be dramatic, the gains are typically less than the changes reported by patients who have had a total hip arthroplasty.5,911 Long-term “technical failures” requiring revision of the prosthesis (eg, loosening, fracture, or infection) are low (less than 10% over 10 years),4,12 yet the lack of improvement is usually related to continuing pain and poor function. Approximately 15% to 30% of patients receiving TKA report little or no improvement after surgery or are unsatisfied with the results after a few months.5,13,14

For the physical therapist, rehabilitation of patients with TKA is often a challenge. One of the primary issues in treating patients with TKA is identifying those patients who may require extensive rehabilitation. For those high-risk patients, early rehabilitation is thought to provide a benefit.15 Although much of the published clinical work has focused on recovery, little evidence exists on determinants of recovery from TKA. One group of researchers3 concluded that baseline pain and function (ie, pain and function on date of decision to proceed with surgery) were the single best predictors of functional recovery at 6 months. Fortin and colleagues3 surmised that patients who reported greater pain and dysfunction prior to surgery were more likely to have more pain and dysfunction after surgery than patients who had less pain and dysfunction. In a prospective cohort study,16 psychosocial factors such as motivation and social function were more influential than medical factors or initial function in predicting 3-month function after TKA, accounting for 15% of the variance. To date, no clear predictors of functional recovery have been consistently reported in the literature.

Given the shortened length of stay in acute care hospitals for patients with TKA, we believe that it is important for the physical therapist to identify those patient-related factors that will affect functional independence. If modifiable determinants of function could then be identified, patients who require additional interventions during their recovery could be readily identified. The primary objective of our study was to identify those demographic, medical, and clinical factors available to physical therapists that predict function at 6 months after surgery. A 6-month follow-up time was selected because studies68 have shown that the greatest change in pain and function occurs during the first 3 to 6 months after surgery. Moreover, we contend that shortterm evaluation can provide useful information on patient recovery and may highlight the need for further therapy to augment recovery. This study was part of a larger study that examined the effect of waiting times for hip and knee arthroplasties on the subsequent health-related quality of life (HRQL) after this surgery.5,17

Method

Participants

Our study was a prospective, longitudinal study of an inception cohort of surgical candidates who received TKA in a Canadian health care region, Capital Health. A health care region is a geographical area administered by a regional health authority. Patients in this study were selected based on time of placement on the regional joint arthroplasty waiting list rather than on the time of surgery. Waiting time for a TKA ranged from 7 to 487 days, with a median wait of 78 days. All patients had surgery between February 1996 and February 1998. Patients were eligible for this study if they: (1) were scheduled for elective primary TKA, (2) were placed on the joint arthroplasty waiting list at least 7 days before surgery (which would help to ensure that emergency surgeries were excluded), (3) resided in the health region, (4) were 40 years of age or older, and (5) spoke English. Exclusion criteria included hemiarthroplasties and revision and emergency arthroplasties.

Patients who resided in long-term care institutions before being placed in the joint replacement waiting list also were excluded. Rarely is any elective joint arthroplasty performed in patients from long-term care facilities. We felt that patients from long-term facilities represent a small unique group of this patient population and are atypical of patients who receive elective knee arthroplasty. After meeting the selection criteria and agreeing to participate, each patient signed a consent form before participating in the study.

Of the 377 patients eligible to participate in the study, 53 (14%) refused to participate, and 18 (5%) were lost to follow-up. Another 30 patients (8%) had completed their preoperative assessments but had their surgeries cancelled for either medical reasons or personal choice. Of those patients who had their surgeries, the participation rate was 79.5%. There were no differences between participants and nonparticipants with respect to age or sex.

Patient characteristics are shown in Table 1. Of the 276 patients in our study, the majority of patients tended to be elderly women with osteoarthritis. Sixty-seven percent of patients (n=186) reported unilateral joint involvement. Hypertension (39%) and back pain (26%) were the 2 most commonly reported comorbid conditions.

Table 1.

Participant Characteristics

Procedure

When the orthopedic surgeon and patient agreed that a TKA was necessary, the patient's name was placed on the health care region's joint arthroplasty waiting list. Names were retrieved from the joint arthroplasty waiting list on a weekly basis, and patients were contacted to request participation in the study. When the patient agreed to participate, in-person interviews were completed within 31 days before surgery and 6 months after surgery. The initial interview consisted of questions regarding demographic information, joint pain, function and stiffness, HRQL, comorbid conditions, medical status, and ambulatory status. During the interview, passive range of motion for the operated knee was measured with a large standard universal goniometer with the patient in a supine position. The hip was placed in a comfortable flexed position (degree of hip flexion varied among the patients), and the maximum amount of knee movement, as tolerated by patient, was measured. The reliability and validity of goniometric measurements of the knee have been reported by others.18,19 Rothstein and colleagues19 reported the intrarater reliability (r) of knee goniometric measurements in the clinical setting to be .91 to .99, and interrater reliability of knee flexion was slightly lower (r=.88–.97). Assessments were completed by 1 of 3 health care professionals (a physical therapist and 2 nurses) who were trained using a standardized study protocol and were not involved in the care of any participants. We did not examine the reliability of their goniometric measurements.

All patients received a primary TKA and were managed using a clinical pathway for TKA in an effort to ensure standardized treatment of medical, pharmaceutical, and rehabilitation care over the 5- to 7-day hospital stay. An important part of the clinical pathway was early mobilization. The protocol for physical therapy intervention consisted of commencing basic activities of daily living with assistance on postoperative day 1. Active-assisted range-of-motion exercises were started on postoperative day 2, after removal of the hemovac. Ambulation, assisted by a physical therapist, was started after postoperative day 1, with weight bearing as tolerated unless otherwise stated. The discharge goal related to mobility was independent and safe ambulation with assistive walking devices on a level surface between postoperative days 5 and 7. Patients were discharged home with an exercise program and referral for community therapy as required. Only 10 patients (4%) were not seen by a physical therapist during their hospital stay, and 257 patients (93%) were seen by postoperative day 2. No participants had simultaneous bilateral knee arthroplasties.

Standardized medical chart reviews were completed by 2 health care professionals. The following surgical and perioperative data were extracted from the medical charts: implant fixation (cemented, hybrid, or cementless), number and type of in-hospital complications (wound infection, dislocation, manipulation under anesthesia, cardiovascular/pulmonary complications, peripheral/central nervous system involvement, urinary infection, acute confusion, blood loss requiring transfusion after surgery), medical information (diagnosis, height, weight), and preoperative ambulatory status (walking distance and use of assistive walking devices). Rehabilitation received within the community was retrieved from administrative databases and treated as a dichotomous variable.

Measures

The interview included a disease-specific questionnaire, the Western Ontario and McMaster Universities (WOMAC) Osteoarthritis Index,20 which is a self-administered health questionnaire designed to measure disability of the osteoarthritic hip and knee. The WOMAC provides an aggregate score for each of the 3 subscales: joint pain (5 items), physical joint function (17 items), and joint stiffness (2 items). The 5-point Likert version of this measure was used in our study. In the calculation of each of the 3 subscale scores, the range of the subscale score was transformed to a range from 0 to 100 points, with a score of 100 indicating no pain or dysfunction. This type of transformation has been used by others to allow an easier comparison between the WOMAC and the Medical Outcomes Study 36-Item Short-Form Health Survey (SF-36).21 The WOMAC is a responsive instrument that yields reliable and valid measurements and that has been extensively used to evaluate this patient population.20,21

A multidimensional generic health measure, the SF-36,2224 was used to measure HRQL. The SF-36 examines 8 health dimensions: physical function, role limitation (physical), bodily pain, mental health, emotional role function, social functioning, vitality, and general health perception. Scoring for each dimension ranges from 0 to 100, with higher scores representing better health. There is no global score; however, 2 component summary measures—physical component summary (PCS) and the mental component summary (MCS)—have been derived from the 8 dimensions and standardized using norm-based methods. Summary measures describe the overall changes in HRQL, but do not capture the smaller changes within the specific dimensions. Reliability and validity have been extensively evaluated in a variety of patient populations, including people with total hip and knee arthroplasties and community-dwelling elderly people.21,2528

The types of comorbid conditions were recorded by the patient or reported on the medical chart. Comorbidities were defined as differing from complications, in that coexisting medical conditions are chronic conditions that exist before surgical intervention or hospital admission. Only those medical conditions identified at the time of admission to the hospital were recorded. The list of 23 comorbid conditions identified by the Charlson Comorbidity Index29 was used. The weighting of severity used with this index was not used in our study because the weights were not derived from function. The number of comorbid conditions was treated as a summative score.

Data Analysis

The dependent variables, the 6-month function scores of the WOMAC and SF-36, were examined as continuous variables given the normal distributions. Functional improvement from the baseline value was defined as a gain of at least 60% of the baseline standard deviation and was considered a moderate effect.30 This equated to approximately a 10-point gain (Tab. 2). This definition posed a potential problem for patients with preoperative scores of 80 or greater because the WOMAC may have a ceiling effect. Because the improvement at 6 months was expected to be large, the net difference preoperatively and postoperatively may be artificially low for those patients with higher preoperative scores. To compensate for this effect, we arbitrarily defined those patients with preoperative scores of 80 or more who maintained a 6-month score of at least 80 as having improved. If the 6-month score dropped below 80 for those patients, it was considered as no improvement.

Table 2.

Preoperative and 6-Month Health Statusa

Independent variables consisted of: (1) demographic variables (age, sex), (2) baseline medical variables (diagnosis, body mass index (BMI), number of comorbid conditions, previous joint arthroplasty, preoperative quality of life as measured by the SF-36, preoperative joint function and pain as measured by the WOMAC, preoperative passive range of motion for the knee, preoperative ambulatory status, type of residence and living arrangements), and (3) perioperative variables (the number of in-hospital complications, type of implant fixation, waiting times, and length of stay). Rehabilitation received during the 6 months after surgery within the community may have had a potential confounding effect and was examined.

Univariate linear regression analyses for each of these variables were examined on the dependent variables. All independent variables that met an initial statistical level of less than .25 or were considered to be clinically meaningful were examined in the multivariate analysis.

Multiple linear regression using stepwise entry with separate models was developed to examine those significant variables associated with function of the knee and overall function. Both joint function—as measured by the WOMAC—and overall function—as measured by the SF-36 physical function dimension—were examined because these measures examined slightly different aspects of function. The SF-36 physical function examined the overall function that could be influenced by other problems, whereas the WOMAC physical joint function measurement specifically examined how the knee affected function.

Stepwise forward model selection techniques were used to obtain the final models. In addition, because age and sex were considered to be potential confounding variables, they were forced into the final models. Model diagnostics, such as residual plots, were inspected to verify that the model assumptions of linearity were not violated. Finally, multicollinearity was assessed by an examination of correlation matrixes of all independent variables.

All statistical testing was performed with 2-tailed tests and at a .05 level of significance unless otherwise stated. Statistical analyses were performed using the SPSS software version 11.01 for Windows.*

Results

The median length of stay in the acute care hospitals was 7 days (range=3–20). All procedures for TKA used a medial peripatellar exposure with a midline skin incision. Of the TKA procedures, 157 (58%) were hybrid, 73 (27%) were cemented, and 42 (15%) were cementless. The hybrid prosthesis routinely involved a porous coated femoral component and a cemented tibial component. Twenty-nine percent of the patients (n=77) received patellar components. Thirty percent of the patellae (n=79) were resurfaced. All patellar components were cemented, all-polyethylene (non–metal-backed) components.

Sixty-seven percent of the patients (n=183) did not have in-hospital complications; however, the primary types of complications were urinary tract infection (n=18) and deep venous thrombi or emboli (n=13). There were 2 deaths due to pulmonary embolism within a month of discharge and another death at 3 months that was unrelated to the knee arthroplasty.

More than half of the patients (n=156 [57%]) were discharged directly home, and all patients returned to the community within 6 months after surgery. Those patients who were discharged directly home tended to be younger (mean age=66.2 years, SD=9.0) than those patients who were transferred to another facility (mean age=73.3 years, SD=7.9) (P < .001). Patients discharged directly home also had better preoperative WOMAC function scores (X̄=45.3, SD=18.0) than the patients who were transferred to another facility (X̄=39.4, SD=16.4) (P=.006). A higher proportion of women (53%) than men (27%) were transferred to a rehabilitation facility (P<.001); however, more women (32%) than men (13%) lived alone (P<.001). Within the community, 129 patients (47%) received community rehabilitation over the 6 months after their surgery. Forty-six percent of the patients (n=125) walked without any assistive devices 6 months after surgery. The mean passive knee range of motion at 6 months was 99 degrees (SD=14).

Functional Status

WOMAC.

The preoperative and 6 month scores of the WOMAC and SF-36 are shown in Table 2. The mean preoperative physical joint function score reported was 42.8 (SD=17.4); however, the 6-month score improved 28% to 70.5 (SD=18.2). Despite the improvement, 53 (20%) patients did not report an improvement from their preoperative scores; that is, they did not report at least a 10-point gain. In particular, questions that concerned domestic duties and stairs were rated difficult at 6 months. Sixty-four percent of the patients (n=165) reported “moderate” to “extreme” difficulty for heavy domestic duties (eg, vacuuming), and 60% (n=160) reported moderate to extreme difficulty descending stairs.

SF-36 physical function.

Overall function as measured by the SF-36 physical function subscale showed less improvement—24%. The mean preoperative score, 21.0 (SD=18.1), improved to 44.8 (SD=25.3) at the 6-month follow-up; however, 77 patients (28%) did not report at least a 10-point improvement from their preoperative scores. When matched for age and sex to the general US population, the 6-month score was significantly less than the mean score reported for the general population—67.6 (SD=7.5) (P<.002).31 The overall physical component is derived from the physical function, bodily pain, role–physical, and health perception dimensions and is standardized using norm-based methods. The physical component score improved almost one standard deviation (9 points) from 25.9 (SD=7.5) to 34.6 (SD=10.1).

Multivariate Regression Models

The unadjusted regression coefficients of preoperative variables that were not included in the final multivariate models are seen in Table 3. While many domains of the SF-36, BMI, and a diagnosis were significant in the univariate analysis, they were not significant when adjusted in the final model. A higher preoperative score of the SF-36 (bodily pain, role-physical, social function, mental health, vitality, and health perception), a lower BMI, and a diagnosis of osteoarthritis rather than a systemic arthritis had an association of higher function scores (WOMAC and SF-36 physical function).

Table 3.

Unadjusted Regression Coefficients Relating Preoperative Variables to 6-Month Physical Function for Both the Western Ontario and McMaster Universities (WOMAC) Osteoarthritis Index and Medical Outcomes Study 36-Item Short-Form Health Survey (SF-36) Physical Function Scores

The results of the multiple linear modeling for predictors of 6-month function are presented in Tables 4 and 5. No strong correlations (r>.50) were noted between independent variables; therefore, multicollearity did not affect the regression analyses. Of the variables that met the level of significance in the univariate analyses, 3 variables met the level of significance and were included in the final multivariate models (Tabs. 4 and 5).

Table 4.

Multiple Linear Regression: Western Ontario and McMaster Universities (WOMAC) Osteoarthritis Index Function at 6 Months

Table 5.

Multiple Linear Regression: Medical Outcomes Study 36-Item Short-Form Health Survey (SF-36) Physical Function at 6 Months

To control for confounding effects, age and sex were force entered into both final models of joint function and overall function. The amount of postoperative rehabilitation may have had potential confounding effects, but this was not significant either in the preliminary univariate analyses or in the multiple linear regression model. Therefore, rehabilitation after surgery was not included in the final model.

Preoperative joint function was a predictor of joint function (WOMAC) and overall function (SF-36 physical function). This finding can be interpreted by the unstandardized coefficient; a 10-point increase in preoperative WOMAC physical joint function scores was associated with a 3.0-point increase in WOMAC physical joint function scores at 6 months (Tab. 4) and with a 3.9-point increase in SF-36 physical function scores (Tab. 5). The standardized beta coefficient indicated that preoperative joint function was the most influential variable in predicting both joint function (as determined by WOMAC joint function scores) and overall function (as determined by SF-36 physical function scores) at 6 months.

The type of walking devices used before surgery was also associated with 6-month function. For instance, a patient who ambulates independently will have a WOMAC 6-month score approximately 12 points higher than that of a patient who ambulates with a walker before surgery.

Preoperative walking distance was predictive of overall function as determined by SF-36 physical function scores (ie, patients who were able to walk longer distances before surgery were more likely to have better overall function at 6 months after surgery). Patients who report that they are able to walk more than 10 blocks before surgery are likely to have a score, that is, 26 points higher than patients who are unable to ambulate.

Twenty percent of the variance in the 6-month WOMAC joint function scores was explained by age, sex, preoperative joint function (WOMAC), comorbid conditions, and preoperative walking devices. Age, sex, preoperative walking devices, walking distance, and joint function (WOMAC) explained 27% of the variance in the SF-36 physical function scores.

Discussion

Our results indicate that preoperative joint function is a predictor of function at 6 months after TKA. Those patients who had lower preoperative functional status related to knee arthritis functioned at a lower level at 6 months than patients with a higher preoperative functional status. These findings concur with those of Fortin and colleagues,3 who reported not only that worse preoperative function resulted in a worse postoperative functional status, but that these differences were more pronounced in patients with TKAs than in patients with total hip arthroplasties.

The variables in the final models accounted for 20% and 27% of the variance seen in the 6-month WOMAC and SF-36 physical function scores, respectively. These variances are comparable to those seen in other studies of TKA3,16 as well as other studies that have examined risk factors of total hip arthroplasties.32 We believe that the variances seen in this study's models are not unreasonable given the dependent and independent variables.

We believe the relationship between baseline function and functional outcome has implications related to the issue of waiting times for TKA. Very few studies have examined the effect of waiting time on function,3335 yet it is of interest in the present context. Earlier findings of this cohort reported minimal functional deterioration with longer waiting times.33 In light of the effect of preoperative function, one goal of rehabilitation would be to maximize function while patients wait for surgery. A preoperative exercise program may help so that deterioration of function might be minimized while waiting for surgery. Little quantitative evidence exists regarding the effect of preoperative exercise programs for knee arthroplasties3638; however, other researchers39 have reported that exercise programs can produce pain relief in patients with knee osteoarthritis. Further investigation may be warranted given the implications of preoperative functional status on functional outcome, particularly for those patients with poor preoperative function.

The relationship between initial function and functional outcome following TKA also has implications for identifying those patients who might require further inpatient rehabilitation. With the current trend toward early discharge, not all patients are suitable candidates for early discharge directly home. Munin and colleagues15 reported that older age, living alone, a greater number of comorbid conditions, and function were predictive of inpatient rehabilitation after a total joint arthroplasty. Patients who have lower levels of preoperative function will likely need further rehabilitation in addition to the therapy received in the acute care setting. Although limited research has compared different models of delivery for rehabilitation of joint arthroplasty,40 further evidence is needed regarding the specific treatment protocols and the most appropriate settings to achieve these treatment goals for patients with high-risk characteristics.

Although we did not specifically address effectiveness of rehabilitation for people with TKAs, we believe a more proactive treatment plan for patients with poor preoperative function should be planned before surgery. A treatment plan may include more intensive physical therapy interventions during the 6 months after surgery regardless of whether it is in a rehabilitation setting or a community setting.

Preoperative knee flexion was not a strong predictor for 6-month function as may have been expected. Our findings, however, suggest that preoperative joint function, comorbid conditions, preoperative walking distance, and walking devices were more predictive of function at 6 months than preoperative knee flexion. Thirteen percent of the patients (n=33) in our cohort had less than 90 degrees of knee flexion prior to surgery. A minimum of 90 degrees of knee flexion is typically required for activities of daily living.41 We believe that our cohort was representative of patients with TKA and reflected the preoperative knee range of motion seen in this patient population because it was a community-based cohort, not restricted to one surgeon or center. Although these results did not show a significant relationship between preoperative knee flexion and 6-month functional status, we believe the measurement of knee flexion may be more informative to the therapist postoperatively than preoperatively.

The 6-month follow-up used in this study could be seen as a limitation. We feel that the 6-month follow-up was appropriate, given the objective of our study and supporting evidence from previous literature of pain and functional recovery after total joint arthroplasty. The greatest change with pain and function occurs during the first 3 to 6 months after surgery,9,42,43 with more gradual improvement occurring over 2 years.9,43 A longer follow-up would provide information about the success of the prosthesis, but we believe it most likely would not change the functional outcomes we observed in our study. From a clinical perspective, evaluation over the 6 months after surgery provides valuable practical information to assist the therapists with management of the patient during the recovery phase.

Another limitation of our study concerns the accuracy of self-report measurement of function. Both joint function and overall function were evaluated with self-report assessments. No performance-based functional measures were used. Some authors44 have reported discrepancies between self-report and performance-based measures of activities of daily living during hospitalizations when functional status was changing. We feel that information gained from self-report assessment of function for our study was valid because function was assessed during stable times (ie, within a month before surgery and 6 months after surgery).

Conclusion

Despite these limitations, findings from this study, along with others,3 present persuasive evidence that patients with greater dysfunction prior to surgery will not attain comparable functional outcomes as those patients with less preoperative dysfunction. Those patients who have low preoperative function may require supplemental rehabilitation while waiting for surgery and further rehabilitation after discharge from the acute care setting.

Footnotes

  • All authors provided concept/research design, writing, and data collection. Dr Jones provided data analysis. Dr Voaklander and Dr Suarez-Almazor provided project management, fund procurement, institutional liaisons, and consultation (including review of manuscript before submission). Dr Suarez-Almazor provided facilities/equipment and clerical support. The authors thank Dr Karen Kelly and Sue Barrett for their assistance throughout the study, as well as Lauren Beaupré and Dr DWC Johnston for their clinical expertise. They also are grateful to Dr Lynn Redfern and Gordon Kramer for instigation of this project.

    Ethics approval was obtained from the Health Research Ethics Board (University of Alberta Sciences Faculties, Capital Health Authority, and the Caritas Health Group).

    This research was supported by grants from the Capital Health Authority Research and Grant Fund and the Edmonton Orthopaedic Research Trust. Dr Suarez-Almazor was supported by The Arthritis Society of Canada and the Alberta Heritage Foundation for Medical Research. Dr Jones was supported, in part, by the Canadian Physiotherapy Foundation, the Royal Canadian Legion, and the Alberta Heritage Foundation for Medical Research.

  • * SPSS Inc, 233 S Wacker Dr, Chicago, IL 60606-6307.

  • Received July 5, 2002.
  • Accepted March 24, 2003.

References

View Abstract