Introduction

Level I evidence is considered the gold standard for clinical decision-making. However, when evaluating the long-term durability of hip arthroplasty designs, Level I prospective clinical trials are impractical. Thus, to date, large registry databases and longitudinal followup studies have provided the best available evidence regarding the implant design characteristics most likely to provide lasting durability and satisfactory function [2, 4–9, 11, 13, 15, 18, 19, 21–24, 26–29]. Most importantly, because hip arthroplasty was traditionally only performed in older patients, the cohorts from these studies tended to be elderly and thus had low patient survivorship at final followup. High rates of patient attrition introduced bias into these studies, and some authors have rightfully questioned the statistical validity of implant survivorship analyses in these elderly cohorts [1, 3, 10, 14].

The majority of long-term followup studies, including our own, used Kaplan-Meier (KM) [20] survivorship curves to report implant revision rates. A KM survivorship curve estimates the time to a single event of interest and assumes that the event of interest occurs independently from other possible competing events [14]. In the case of joint arthroplasty, the event of interest is typically the occurrence of a revision surgery. However, other events may take place that would compete with or even preclude the possibility of revision surgery. In particular, a patient death is a competing event because patients who die cannot be revised later. The KM approach treats those who died (with no chance of revision) similarly to those who are lost to followup (who could still undergo revision). Prior studies have shown that this biases KM analyses toward an overestimation of event rates [14, 16, 17].

More recent authors have argued for reporting implant revision rates using the cumulative incidence of competing risks (cumulative incidence) methodology. Although arthroplasty surgeons may still be somewhat unfamiliar with reporting revision rates using the cumulative incidence methodology, its use in both our field and others has been reported for some time [1, 3, 30]. The cumulative incidence method reports the probability of failure as a result of the event of interest in the presence of competing risks [14]. If a large number of patients die during followup, late implant failure becomes less likely and survivorship is increased accordingly. The current authors have previously evaluated the long-term followup of TKAs using similar methods [10]. However, the degree to which the differences between KM and competing incidence estimators may be clinically relevant has varied across the few studies on the topic in orthopaedic surgery, and so we wished to further characterize it in a population of patients who have been followed into the third decade after THA. A review of these methods should be useful in informing the design of future long-term followup studies.

Patients and Methods

This study received an institutional review board exemption and was HIPPA-compliant. Our institution has maintained long-term followup records for three separate series of Charnley total hip cohorts. The methodology for each cohort has been previously published [6, 25, 28]. Each series is a prospectively followed, consecutive, nonselected cohort from the time period specified. Followup evaluations were performed by a single surgeon (DDG) not involved in the initial surgical care of the patients. Radiographs were evaluated by two independent observers (JJC, DDG) with agreement by consensus at each followup interval. One observer reviewed all radiographs at each followup interval of all cohorts (JJC). We retrospectively reviewed these records for the current analysis.

The first cohort consisted of 330 hip arthroplasties in 262 patients performed between July 1970 and April 1972 using first-generation cement techniques [6]. Details from this cohort have been published at regular intervals out to 35 years of followup [5, 6, 12, 18, 19, 27]. The second cohort consisted of 357 hip arthroplasties in 320 patients performed between July 1976 and June 1978 and used modern second-generation cementing techniques [26]. Details from this cohort have been published at regular intervals out to 20 years of followup [25, 26]. The final cohort consisted of 93 hip arthroplasties performed in 69 patients, all of whom were aged 50 years or younger at the time of surgery, performed between January 1970 and December 1976. Details from this cohort have also been published at regular intervals out to 35 years of followup [7, 24, 28]. All patients were followed for a minimum of 20 years or until death. Notably, a total of 21% (23 of 109) of the cohort who were younger than 50 years at the time of THA died during the 20-year followup period compared with 72% (467 of 649) who were older than 50 years at the time of surgery (p < 0.001) (Fig. 1).

Fig. 1
figure 1

KM analysis shows the difference in patient survivorship out to final followup between patients younger than or older than 50 years of age at the time of their index procedure. The younger cohort was significantly more likely to survive the duration of the study period (79% versus 28%, p < 0.001).

Full size image

The surgical approach and prosthesis implantations were done in a uniform fashion across all three cohorts. All operations were done by a single surgeon (RCJ). All patients were implanted with a Charnley hip prosthesis (Thackray, Leeds, UK, or Zimmer, Warsaw, IN, USA) with a polished stainless steel stem and a 22-mm diameter nonmodular femoral head. The acetabular component was made of ultrahigh-molecular-weight polyethylene with an outer diameter of either 40 or 44 mm. Both the femoral and acetabular components were inserted with the use of Simplex P cement (Northill Plastics, London, UK, or Howmedica, Rutherford, NJ, USA). All surgeries were performed through a transtrochanteric approach, and no antibiotics were used perioperatively.

The femoral cementing technique used by the surgeon changed over time. In the hips done between July 1970 and December 1976, cement was inserted using finger-packing [7, 27]. In the hips done between July 1976 and June 1978 [26], a contemporary cement technique was used in which the bone was meticulously dried and all loose cancellous bone was removed. A distal cement plug was then placed and the cement was inserted using a cement gun under pressure. A previous analysis by our group compared the two cement techniques and found no difference in implant survivorship between them [25]; thus, we feel it is reasonable to consider them together as a single group in this analysis.

For our statistical analysis, all three cohorts were combined into a single group. Of note, the third cohort of patients, specifically younger than 50 years of age, included some overlap with the other studies. Thus, after excluding 22 duplicate procedures, we were left with a total of 758 unique Charnley total hip procedures in 635 patients from all three cohorts. The average age was 64 years (range, 18–91 years). Three hundred forty hips were in men (45%). The most common diagnosis was osteoarthritis in 518 hips (68%) followed by posttraumatic osteoarthritis in 83 hips (11%), developmental dysplasia of the hip in 62 hips (8%), rheumatoid arthritis in 32 hips (4%), slipped capital femoral epiphysis in 16 hips (2%), postseptic arthritis in 14 hips (2%), avascular necrosis in 11 hips (1%), Legg-Calvé-Perthes disease in five hips (< 1%), and other diagnoses in 17 hips (2%).

The patients were stratified by age with 109 hips implanted in patients aged 50 years or younger and 649 hips implanted in patients older than 50 years of age. We then compared implant survivorship between the patients 50 years or younger against patients older than 50 years according to cumulative incidence methods [14]. The details of this calculation method have been previously well described [1, 3, 14]. The primary endpoint was revision for aseptic implant failure (pain or radiographic loosening of the implants). Patients who died or who had a revision for an infection or fracture were considered to have had a competing event. Additionally, the risk of revision was also calculated using KM methods using this same cohort. In KM methodology, patients with a competing event are censored and are assigned a risk of revision equal to that of the remaining cohort [20]. Patients with a death or a revision resulting from infection or fracture were censored in this analysis. All curves were truncated at 20 years in each analysis for similar comparison across cohorts. Statistical analysis was performed using SPSS 13.0 software (SPSS Inc, Chicago, IL, USA).

Results

A larger proportion of the younger patients in this report underwent revision for aseptic causes during the surveillance period than did the older patients. Specifically, 21 of 109 (19%) Charnley hip arthroplasties implanted in patients younger than 50 years of age underwent a revision of either the femoral or acetabular component for aseptic causes within 20 years of their index procedure as compared with 33 of 649 (5%) hips in patients older than 50 years of age (p < 0.001, chi square analysis).

In reporting these incidences, the two analysis methods produced differing results. The cumulative incidence function reported very similar percentages to the actual revision rates noted with an estimated 19% revision rate in the younger than 50 years cohort (95% confidence interval [CI], 13%–27%) and an estimated 5% revision rate in the older than 50 years cohort (95% CI, 3%–7%). However, in the KM analysis, the risk of revision for the younger than age 50 years cohort was reported as 23% (95% CI, 15%–32%) and for the older than age 50 years cohort, the risk of revision was reported as 8% (95% CI, 7%–11%) (p < 0.001) (Fig. 2). This represents a 22% and 66% relative increase, respectively. Patient death represents the primary source of bias in the KM analysis, and thus the relative magnitude of difference between the KM and cumulative incidence methods increased as the patients aged over time during the study period (Fig. 3).

Fig. 2
figure 2

A comparison is shown of implant revision rates using the cumulative incidence methodology and the KM methodology (for the KM method, incidence = 1 − KM survivorship). The incidence of revision was higher for the older than 50 years of age cohort compared with the younger than 50 years of age cohort for both KM and cumulative incidence methods (p < 0.001 for each).

Full size image
Fig. 3
figure 3

Trend lines are plotted, showing the relative difference between the KM and cumulative incidence methods over time. The magnitude of the relative difference between the two methods increased over time as the incidence of patient deaths increased throughout the study period, particularly in the older than 50 years of age cohort.

Full size image

Discussion

Patients undergoing a THA want to know how long their implant is likely to last. Historically, long-term followup studies of specific implant designs have been one of the few available ways to acquire this information, and most studies have reported implant survivorship using KM methods. However, patient deaths during the study period violate the assumptions of the KM model. Previously, the degree to which this was clinically relevant had not been well established. Thus, we sought to compare the KM and cumulative incidence estimators from a large cohort of patients followed for a minimum of 20 years or until patient death. Overall, we found that KM methodology substantially overestimated revision rates, particularly in elderly cohorts. We feel that the results of this study will be useful in the planning and design of future long-term followup analyses.

This study does have several limitations. First, the patients operated on in our study period from the 1960s and 1970s are likely of different demographics and life expectancy than modern patient populations, and thus the patient survivorship data presented here may not completely correlate with future studies. Second, although implant revision rates are a commonly accepted measure of implant performance, the occurrence of a revision is not an ideal outcome measure because patients may have pain, radiological changes associated with loosening, major medical comorbidities, or be dissatisfied without requesting or undergoing a revision surgery. Third, we have included two cementing techniques in a single cohort for our statistical analysis. However, all of the surgeries were performed by a single surgeon using a polished flat-backed Charnley, and prior studies have shown that there is no difference in the long-term durability despite the difference in cementing technique [4, 25]. Thus, we feel it is appropriate to include them in a single analysis.

Overall, we found that the KM method overestimated the risk of revision by 66% in the older than 50 years of age cohort and by 22% in the younger than 50 years of age cohort. In contrast, the cumulative incidence method more accurately reported the revision risk. The reason for this discrepancy is straightforward. A patient who dies cannot possibly be revised, and this is taken into account in the cumulative incidence methodology [14, 17]. In contrast, in the KM analysis, patients with a competing event (death) are assigned a risk of revision equal to that of the remaining cohort [16, 17], which provides the risk of implant revision assuming that no patient ever dies. Clearly this is an unrealistic scenario, and KM analysis tends to overestimate the risk of revision for this reason [14, 17]. Therefore, the cumulative incidence method is a more appropriate statistical tool for evaluating implant survivorship and we would encourage authors of future long-term followup studies to implement it in favor of the widely used but inappropriately applied KM methodology.

In addition to introducing bias into the KM analysis, the high rate of patient deaths in the older than 50 years of age cohort highlights a second important point. Specifically, only 28% of the older than 50 years cohort survived the duration of the study period, and only 5% required revision for aseptic causes. Thus, simply as a result of the high rates of patient mortality, the older patients are unlikely to require revision for aseptic reasons at any time in their remaining years. Therefore, comparisons of performance across different implant designs would be very difficult. We suggest that clinicians focus their efforts on ensuring regular followup among their younger patients. Younger patients are much more likely to survive to final followup and thus provide a more accurate estimate of implant durability and performance over time.

In summary, our study found that high rates of patient deaths introduced substantial bias into the analysis of long-term followup studies when the results were reported using KM methodology. Because patient death is a competing risk with revision, the use of a KM curve to report revision rates is inappropriate. Future investigators conducting long-term followup studies of hip arthroplasty implants should use patient survivorship curves that account for competing risks. For the investigators designing future long-term followup studies, the patient survivorship curves we provided should be useful for determining the necessary composition of patients, both in terms of patient age and numbers of patients needed, to have adequate numbers for statistically valid comparisons. Furthermore, if we wish to be able to report clinically relevant long-term results of hip arthroplasty designs, it seems likely that multicenter or joint registry studies will be necessary to acquire robust patient numbers of younger patients.