Establishment and validation of a nomogram model for aseptic loosening after tumor prosthetic replacement around the knee: a retrospective analysis

This research was conducted according to the “Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD)” statement [8]. Data on patients who underwent tumor prosthetic replacements between August 2001 and September 2016 were retrospectively collected from two clinical centers. Tumor prosthetic replacements were carried out for management of reconstruction following resection of malignant bone tumors and invasive (stage 3) benign bone tumors. The inclusion criteria were the use of tumor prostheses for primary reconstruction of the mega bone defect after tumor resection. The exclusion criteria were extendable prostheses used for children, uncemented prostheses, allograft-prosthetic composites, and patients who were lost to follow up. This study followed the “Declaration of Helsinki” and was approved by the hospital ethics committee and obtained informed consent from the patients.

The medial or lateral arc incision was selected to expose the knee joint and made an elliptical incision around the biopsy scar to remove the biopsy channel. The medial and lateral collateral ligaments, cruciate ligaments, and meniscus were severed to dislocate the knee joint. The osteotomy plane was determined by where the normal bone marrow signal on T1WI becomes an abnormal signal. Resection of tumor was based on common oncology principles [9]. The marrow cavity of the distal femur and proximal tibia was expanded until it was 2 mm larger than the diameter of prosthetic stem. The trial prosthesis was then placed to adjust the condition of surrounding soft tissue and force line and for activity tracking of the patella. The medullary cavity was washed and injected with bone cement, the proximal and distal prosthetic stems were placed, and additional components were installed. We usually did not perform patella replacement. Reconstruction of knee extension mechanism involved suturing the patellar tendon with non-absorbable suture on the anterior hole of the prosthesis or the gastrocnemius. If the soft tissue envelope was difficult to suture, free skin grafting was considered to reduce tension. We used two kinds of native tumor prostheses (Lidakang, Beijing, China and Wego, Beijing, China). Neither of them possessed a hydroxyapatite (HA) collar design.

Demographic, preoperative, intraoperative, and postoperative data were obtained from medical records and subsequent follow-up. Preoperative data included the tumor site (distal femur vs. proximal tibia), sex (female vs. male), age (> 30 years vs. ≤ 30 years), prosthetic motion mode (fixed vs. rotating hinge), material for prosthetic stem (CoCrMo vs. titanium), type of prosthesis (custom vs. modular), and type of tumor (benign vs. malignant). Intraoperative data included length of prosthetic stem (≥ 14 cm vs. < 14 cm), length of bone resection (≥ 14 cm vs. < 14 cm), extra-cortical grafting (yes vs. no), and diameter of prosthetic stem (≥ 13 mm vs. < 13 mm). Postoperative data included the intensity of activity (high vs. low) and body mass index (BMI) (≥ 25 vs. < 25). These previously described factors were included as explanatory variables in this study. The calculation of resection length at the distal femur and proximal tibia was based on the femoral condyle and tibial plateau, respectively. Patients who were retired at home or were involved in limited leisure activities were described as low-intensity, and those who regularly participated in physical exercise were categorized as high-intensity. In terms of diameter of prosthetic stem, length of prosthetic stem, and length of bone resection, patients were divided into two groups about the mean value for each independent factor. The response variable was aseptic loosening, which was defined as fracture of the cement mantle or progressive radiolucencies between bone-prosthesis interface, without infection.

Descriptive statistics were calculated using means for continuous data and proportions for count data. Survival analysis was performed with Kaplan-Meier curves [10], and patients were censored for any other cause of failure, death, and no problem at latest follow-up. To determine the risk factors associated with aseptic loosening, univariate Cox regression analysis was carried out, and variables significant at the p ≤ 0.15 level were included in the multivariate Cox regression analysis [11]. p values were based on two-tailed tests and at < 0.05 were considered significant. The forest plot was used to display the results of univariate and multivariate analyses. Finally, a nomogram model was constructed using the independent risk factors of the multivariate analysis. The concordance index (C-index), which ranged from 0.5 to 1.0, was applied to validate the discrimination, and then the calibration curve was applied to validate the consistence for each time point. The building, validation, and interpretations of the nomogram model were carried out following the literature published by Iasonos et al. [7]. Statistical analysis was performed using R version 3.5.1 for Windows (R Foundation for Statistical Computing, Vienna, Austria), SPSS 22.0 software (SPSS Inc., Chicago, Illinois, USA), and GraphPad Prism 7 Software (GraphPad Software Inc., San Diego, CA).

Two hundred and twenty-eight prostheses were identified. Eighteen implants were excluded because they were allograft-prosthetic composites or extendable prostheses and 23 prostheses had been lost to follow-up, leaving 177 prostheses for evaluation. There were 61 female patients (34.5%) and 116 male patients (65.5%). The mean age at the time of surgery was 35 years (range, 13–77 years). One hundred nineteen and 58 implants were placed in the distal femur and proximal tibia, respectively. The histologic diagnoses were osteosarcoma in 85 patients, giant cell tumor in 61 patients, malignant fibrous histiocytoma in 14 patients, chondrosarcoma in 9 patients, and other sarcomas in 8 patients. Table 1 displays the baseline information including type of prosthesis, tumor site, and additional details of the patients and prostheses.

Minimum duration of follow-up was 2 years (mean, 92 months; range, 24–199 months). At the most recent follow-up, among the 177 patients with benign or malignant bone tumors, 121 were continuously disease-free, 23 had no disease after treatment of local recurrence, nine were alive with tumor, and 24 were deceased. Of 116 patients with sarcomas, 11 (9.5%) developed a recurrence, and of those, 27 (23.3%) were thought to exhibit metastasis. All patients with giant cell tumor were alive.

The incidence of prosthetic failure in our study was 24.9% (44 of 177). Mechanical failures occurred in 61.4% (27 of 44) of all failures. Only two (4.5%) of all failures were due to problems related to soft tissues, and most treatment measures did not involve the prosthesis itself. Twenty-two failures (50%) were from aseptic loosening, and three failures (6.8%) were due to periprosthetic or prosthetic fractures. Nonmechanical failures accounted for 38.6% (17 of 44) of all failures, which included five failures (11.4%) that were due to tumor progression and 12 failures (34%) that came from deep infection. The absolute risk of every failure mode type 1 through 5 was 1.1%, 12.4%, 1.7%, 6.8%, and 2.8%, respectively.

Aseptic loosening occurred in 13.6% (24 of 177) prostheses at mean of 91 months (range, 15–166 months). The prosthetic survival rate without aseptic loosening was 92.5% at 5 years, 85.0% at 10 years, and 45.5% at 15 years (Fig. 1). The incidence of aseptic loosening was 16.0% (19 of 119) in the distal femoral prostheses and 8.6% (5 of 58) in the proximal tibial prostheses. Rotating hinge prostheses (8.2%, 13 of 158) had a lower aseptic loosening rate than fixed hinge prostheses (57.9%, 11 of 19).

Most patients (22 of 24) with aseptic loosening achieved satisfactory results through revision surgery, and the remaining two patients were waiting for revision. During revision surgery, we reset the original prosthesis or replaced it with another prosthesis. The new prosthesis generally had a longer and thicker prosthetic stem than the original prosthesis.

Univariate Cox regression analysis showed that the length of bone resection (p = 0.046) and prosthetic motion mode (p = 0.001) were risk factors for aseptic loosening. The length of prosthetic stem (p = 0.059) and tumor site (p = 0.150) were shown to be marginally significant for predicting aseptic loosening (Fig. 2). Other factors including the sex (p = 0.496), type of tumor (p = 0.432), age (p = 0.627), extra-cortical grafting (p = 0.826), custom or modular (p = 0.675), diameter of prosthetic stem (p = 0.611), material for prosthetic stem (p = 0.698), intensity of activity (p = 0.347), and BMI (p = 0.823) were not risk factors for aseptic loosening of tumor prosthesis (Table 2; Fig. 3).

Multivariate Cox regression analysis indicated that the tumor site (HR = 3.99, CI 95% 1.21~13.16, p = 0.023), length of prosthetic stem (HR = 2.84, CI 95% 1.13~7.12, p = 0.026), and prosthetic motion mode (HR = 4.11, CI 95% 1.74~9.70, p = 0.001) were independent risk factors for aseptic loosening (Table 2; Fig. 3). The length of bone resection (p = 0.382) was excluded from the model.

A nomogram model was constructed based on the three independent risk factors derived from the multivariate analysis (Fig. 4). Each factor in the model was given a weighted point. The sum of each factor was the total points of the patient. According to the total points, the prosthetic survival rate without aseptic loosening for 5 years, 10 years, and median survival time can be calculated. The greater the number of points, the lower the prosthetic survival rate without aseptic loosening. The C-index was 0.74 (CI 95% 0.65~0.84), which indicated that the nomogram model had acceptable predictive discrimination. The calibration curves for time points of 5 years and 10 years are shown in Fig. 5. There was a good consistency between the predicted and the actual prosthetic survival rates shown for each time point.