Overestimation of the risk of revision with Kaplan-Meier presenting the long-term outcome of total hip replacement in older patients

Abstract

Introduction:

The purpose of the present study was (i) to review the long-term outcome of cemented Charnley total hip replacements (THRs) performed by 1 surgeon (GH), 20 to 42 years ago, in patients ≥60 years, using both the Kaplan-Meier (KM) and the cumulative incidence (CI) methods, and (ii) to compare the estimations of the 2 statistical methods.

Methods:

We evaluated the outcome of 306 consecutive primary cemented THRs that were performed in 265 patients. The final clinical, radiographic assessment and satisfaction of living patients were also included. The survivorship was estimated with the use of KM and CI methods and the relative difference between their estimations was calculated.

Results:

Living patients’ final clinical results were significantly improved in comparison with respective preoperative ones, and all the acetabular and 91% of femoral components considered as well fixed. 95% of these patients reported satisfaction. The risk of revision at 25 years, with revision for aseptic loosening for 1 or both components as the endpoint, with 21 hips at risk, assessed with KM analysis was 6.9% and with CI approach was 3.9%. The relative difference between KM and CI estimations was increasing during follow-up, reaching up to 76.8% at 25 years.

Conclusions:

We concluded that fixation of implants with cement in older patients had satisfactory long-term results and can serve as a benchmark with which to compare newer fixation methods (hybrid and uncemented) and materials. However, KM method, in studies that include older population with long-term follow-up, may significantly overestimate the risk of revision and clinicians could consider using besides the cumulative incidence of competing risk method.

Keywords

Competing risk Cumulative incidence Kaplan-Meier Low friction arthroplasty Survival analysis Total hip replacement

Introduction

The method of implant fixation in total hip replacement (THR) continues to be a topic of controversy among orthopaedic surgeons. For many years, the cemented Charnley low friction arthroplasty (LFA) was considered as the gold standard (1, 2). Later, the introduction of the theory of the “cement disease” (3), although never proved, and the hypothesis that the proportion of good results with cemented THR decreased with time, led the orthopaedic community to use uncemented devices (4 -13). However, it seems that there is a stronger consensus that in older patients cemented prostheses are more successful (14 -18).

Analysis of survival of a THR can be performed with 2 main methods: Kaplan-Meier (KM) and cumulative incidence (CI). The KM method is preferred when the exact time of revision is known, and an understandable and more clinically meaningful survival estimate is desired (19), however in studies with a large sample of censored cases, usually due to death, leads to overestimation of the risk of revision (19, 20). These studies included mainly older population with a long-term follow-up (19 -21). This was the reason, in studies with heavy censoring, the CI approach has been proposed as an alternative to a more accurate analysis, accounting the censored cases in the failure risk (19).

The purpose of the present study was to review (i) the long-term outcome of 287 cemented Charnley LFAs performed by 1 surgeon (GH), 20 to 42 years ago, in patients 60 years of age and older, using both the KM and the CI methods, and (ii) to compare the estimations of the 2 statistical methods. Part of the material (41% of the hips) has been previously published (2).

Methods

In January 1973, the cemented Charnley’s LFA was introduced at the Orthopaedic Department of University of Athens, KAT Hospital. Between January 1973 to December 1995, 530 consecutive primary Charnley’s LFAs performed by the senior author and entered into our database. 306 THRs performed in 265 patients, 60 years of age and older. 17 patients (19 hips) were lost to follow-up at a mean of 7 years (range 1-15 years) after the index operation with the original components in place. Thus, the population of the study consisted of 248 patients (287 hips) with a mean age of 68 years (range 60-86 years) at the time of surgery. They were 209 women and 39 men. The primary diagnosis was idiopathic osteoarthritis in 182 hips, congenital hip disease in 51 (graded as dysplasia in 21, low dislocation in 14 and high dislocation in 16) (22, 23), trauma in 9, inflammatory arthritis in 22, avascular necrosis of the femoral head in 21 and other diagnoses in 2. The study was approved by our institutional review board.

Standard Charnley technique and implants (Thackray; now DePuy) (1, 24), by osteotomising the greater trochanter, were implanted in all cases. Beginning in 1982, a 2^nd-generation cementing technique was used for 195 hips (25). Prophylactic antibiotics were not given in 42 hips (15%) that underwent THR until December 1977. From that time prophylactic antibiotics were administered intraoperatively and during the first 2 days postoperatively.

Patients were routinely followed, clinically and radiographically, at 3 and 12 months after surgery and at 1- to 3-year intervals thereafter during their lifetime or at least 20 years after the index operation. A small number of very old patients or with severe health problems were contacted by telephone. The living patients at the preparation of the study, with both components in place, in 1 or both hips, were included in the final clinical and radiographic assessment. Pre- and postoperative clinical evaluation, regarding pain and range of motion, was performed with the Merle d’Aubigné and Postel score as modified by Charnley (24). Function was not evaluated because walking ability had declined with age in this group of patients. Instead, patient’s reported satisfaction was examined and graded as “satisfaction”, “partial satisfaction” and “dissatisfaction”. Radiographs were re-examined by 2 observers (KLA and GH), with 95% inter- and intra-observer agreement on the radiographic measurements for acetabular and femoral osteolysis. Any osteolysis around the acetabular and femoral components was assessed on the latest anteroposterior radiograph. An acetabular or femoral component was considered well fixed when a radiolucent line (linear osteolysis) of less than 1 mm and ≤50% of the cement-bone interface, possibly loose when an at least 1 mm radiolucent line and >50% and <100% cement–bone interface and probably loose when a continuous linear osteolysis without migration were recorded (25 -27).

Statistical analysis

The improvements in the clinical outcome were assessed with Wilcoxon’s signed ranks test. The KM survival curves with 95% confidence intervals were generated, with revision for aseptic loosening of one or both components as the endpoint (event of interest). Using KM analysis, patients alive without having undergone revision surgery at the last follow-up, lost to follow-up and experiencing a competing event were considered censored in the same manner. As a competing event was considered death or revision for other reason than aseptic loosening and may preclude the onset of the event of interest (aseptic loosening), or modify the probability of the onset of the event of interest. The CI analysis, accounting for competing risks, was performed in 2 steps, with: (i) estimation of the overall survival probability of experiencing any event, both event of interest and competing risk event; and (ii) estimation of the cumulative probability of encountering the event of interest (28, 29). The relative difference (diff) of the KM and CI values was calculated (30). Additionally, our recordings are presented as KM and CI curves with endpoint: revision for any reason and resection arthroplasty (Girdlestone procedure), all operations including fracture internal fixation, revision for aseptic loosening of acetabular and femoral components, and revision for aseptic loosening in OA cases in order to facilitate comparisons with other studies regardless the selected endpoint.

In order to perform comparisons between the 2 methods and for the homogeneity of the data presentation, survival rate of KM was expressed as risk of revision (1-KM) (29). All the statistical tests were performed using SPSS 20 statistical software (SPSS Inc.) and Excel 2008 (version 12.2.7, Microsoft Corporation). A value of p<0.05 was selected to indicate statistical significance.

Results

At the time of preparation of the study, 194 patients (223 hips) had died retaining 1 or both components in either hip, at a mean age of 84 years (range 62-99 years), and 40 patients (47 hips) were alive with 1 or both components in place in either hip, for a minimum of 20 years follow-up, at a mean age of 89 years (range 82-96 years) (Figure 1 and 2).

During patients’ lifetime or until the last follow-up examination, 9 (3%) acetabular and 9 (3%) femoral components had been revised for aseptic loosening, at an average of 14 years (range 8-20 years) and 14 years (range 10-20 years), respectively.

Clinical rating was obtained from 37 living patients (43 hips) retaining both components for at least 20 years (mean 24 years; range 20-29 years). At the final follow-up, all patients were free of pain or had slight pain on starting to walk (5 and 6 points). 27 hips had more than 160° total range of motion (5 and 6 points), 14 hips had 100° to 160° total range of motion (4 points), and two hips had less than 100° total range of motion (3 points). All final results were significantly improved in comparison with respective preoperative ones (p<0.005). 35 patients (95%) reported satisfaction, as their preoperative expectations were completely realised and 2 patients (5%) were partially satisfied. Radiographic evaluation of 43 hips of the 37 living patients retaining both components pres ented all the acetabular (100%) and 39 femoral components (91%) as well fixed and 4 femoral components (9%) as possibly loose.

Other complications: 11 hips of 287 studied (3.8%) were infected and converted to resection arthroplasty at an average of 8 years (range 0-18 years) postoperatively. 2 more stems (0.7%) were revised because of breakage at 22 and 25 years. 1 patient (1 hip) presented peroneal and femoral nerve palsy, fully resolved within 6 months. Dislocations presented in 2 hips 10 months and 7 years after surgery and were treated with closed reduction. Periprosthetic fracture of the femur occurred in 1 hip, 2 months after surgery and was treated with internal osteosynthesis. 3 patients died postoperatively because of thromboembolic events.

The risk of revision in KM (1-KM) and CI analysis, for aseptic loosening of 1 or both components as the endpoint, is presented in Table I and Figure 3. The estimation of the overall survival probability of experiencing any event, both event of interest and competing risk event, is presented in Figure 4. In reporting the risk of revision using the 2 different methods, the relative difference increased from 33.3% at 10 years to 63.8% and 76.8% at 20 and 25 years, respectively (Table I, Figure 3). The competing events were 75% (231 of 306). Figures 5 and 6 present additional estimations in KM and CI curves with endpoint: revision for any reason and resection arthroplasty (Girdlestone procedure), all operations including internal fixation, revision for aseptic loosening of acetabular and femoral components, and revision for aseptic loosening in OA cases (Table I).

Table I

Revision rates of 306 THRs, estimated with the use of KM and CI methods with different endpoints at 10, 20 and 25 years and the differences between the estimations

Endpoint	At 10 years			At 20 years			At 25 years
	Risk of revision (%)		Diff (%)/Hips at risk	Risk of revision (%)		Diff (%)/Hips at risk	Risk of revision (%)		Diff (%)/Hips at risk
	1-KM (95% CI)	CI (95% CI)	Diff (%)/Hips at risk	1-KM (95% CI)	CI (95% CI)	Diff (%)/Hips at risk	1-KM (95% CI)	CI (95% CI)	Diff (%)/Hips at risk
Revision for aseptic loosening of one or both components	0.4 (0-1.1)	0.3 (0-1)	33.3/231	5.9 (2.4-9.4)	3.6 (1.4-5.7)	63.8/95	6.9 (3-10.8)	3.9 (1.6-6.1)	76.8/21
Revision for any reason and resection arthroplasty	4 (1.8-6.1)	3.9 (1.6-6.1)	2.5/231	9.4 (5.4-13.3)	7.1 (4.1-10)	32.3/95	13.5 (7.2-19.7)	8.1 (4.9-11.2)	66.6/21
Re-operation	4.4 (2-6.7)	4.2 (1.8-6.5)	4.7/230	9.7 (5.7-13.6)	7.5 (4.4-10.5)	29.3/94	13.9 (7.6-20.1)	8.4 (5.1-11.6)	65.4/21
Revision for aseptic loosening of acetabular components	0.4 (0-1.1)	0.3 (0-1)	33.3/231	4.1 (1.1-7)	2.6 (0.7-4.4)	57.6/97	5.1 (1.7-8.4)	2.9 (0.9-4.8)	75.8/21
Revision for aseptic loosening of femoral components	0	0	0/232	4.4 (1.2-7.5)	2.6 (0.7-4.4)	69.2/98	5.3 (1.7-8.8)	2.9 (0.9-4.8)	82.7/22
Revision for aseptic loosening in OA cases	0	0	0/145	5 (0.7-9.8)	2.6 (0.2-4.8)	92.3/53	5 (0.7-9.8)	2.6 (0.2-4.8)	92.3/10

CI = cumulative incidence; diff = relative difference; KM = Kaplan-Meier; OA = osteoarthritis.

Discussion

In our series, of the 287 followed LFAs in patients 60 years of age and older, 273 (95.1%) retained both of the original implants for 20 or more years, until patients death (219 hips) for an average of 15 years (range 1-32 years) or at the last follow-up evaluation (43 hips) for an average of 24 years (range 20-29 years). The most common reason of failure was infection (3.8%). Of note, the infection rate was 9.5% (4 of 42 hips), when prophylactic antibiotics were not used, and decreased to 2.8% (7 of 245 hips) after their introduction in our protocol. The risk of revision at 25 years, with revision for aseptic loosening as the endpoint, when 21 hips were at risk, assessed with KM analysis was 6.9% and with CI approach was 3.9%. The relative difference between KM and CI estimations was increasing during follow-up, reaching up to 76.8% at 25 years. Thus, similarly to other reports, we found that the KM analysis overestimated significantly the risk of revision in comparison to the CI method (28 -31). If this observation has an important clinical significance, remains to be examined.

As limitations of the present study can be considered its retrospective nature, the limited sample size and the fact that it has been based on 1 surgeon’s database. Furthermore, only 16% of patients were males, there had been used 2 cementation techniques during the 23-year record of this registry and prophylactic antibiotics were not given in the first 15% of hips which had been operated before December 1977. On the contrary, the longest follow-up on this topic to our knowledge, the assessment of homogenous data and estimation of results with the use of 2 different methods of analysis, can be considered as strengths of the study.

After its introduction in 1960 by J. Charnley, THR gradually became 1 of the most successful orthopaedic procedures (32, 33). However, the method of implant fixation, cemented, uncemented or hybrid, remains the topic of controversy between orthopaedic surgeons. Although satisfactory results have been reported, even in young patients, with the use of cemented THRs (34), it seems that there is strong consensus that in older patients the method of cemented fixation is more successful. According to data obtained from a multinational registry of Sweden, Norway, Denmark and Finland, Mäkelä et al (16) studied 347,899 THRs (cemented, uncemented, hybrid and reverse hybrid), performed during 1995-2011 and concluded that the survival of cemented implants was higher than that of uncemented in patients aged 65 years or older and that the increased use of uncemented implants in this age group is not supported by those data. Similarly, from a multinational comprehensive evaluation of the fixation method used in hip replacement that included 239,442 patients with osteoarthritis operated from 2001-2010, Stea et al (17) concluded that cementless fixation should be avoided in patients 75 years of age or older. However, as reported by Troelsen et al (14), existing registries of all countries show increase of the use of uncemented implants in all ages. This phenomenon is in conflict with evidence reporting better results of cemented fixation at least in older patients and called by the authors as paradoxical.

Regarding cost, although it has been reported that cemented prostheses are the cheapest option for THR in all age groups (35), there are arguments as for the true cost of a THR. As stated by Kallala et al (36), the economic analysis of a single surgical procedure should include all costs: the cost of implants, theatre time, inpatient stay, rehabilitation and postoperative complications. The authors concluded that direct comparison of costs without full consideration of all outcomes is an oversimplification (36).

Apart from the method of fixation of a THR, a variety of techniques and implants gain the attention of orthopaedic surgeons. Comparison of these factors that may influence the outcome of a THR is of importance. Arthroplasty registries and long-term studies are called to report the outcomes of THRs, including survival analysis. Comparisons of reliable studies presenting outcomes of different techniques and implants may lead to conclusions in order to improve the clinical practice. It should be stressed that studies with short follow-up cannot possibly be used to forecast the long-term outcome of joint replacement. Only extended follow-up studies of more than 20 years, which currently reach 40 years (2, 37), can provide this information (38). Obviously, these studies include a large proportion of censored cases, mainly due to death, that leads to biased KM survival estimations and comparisons between studies.

The outcome of a THR is assessed with survival analysis using traditionally the KM statistical method. As the endpoint of the analysis is usually considered revision for any reason or revision for aseptic loosening of the components. Patients without having undergone revision surgery at the last follow-up, lost to follow-up and experiencing a competing event were considered censored in the same manner. Also, patients died from unrelated to the procedure reasons, with the implants in place are considered as “drop-out events” at the time interval that the death happened, assuming independence between the endpoint and the event of death (29). Thus, KM method leads to overestimation of revision rate, an observation that is mathematically proven (39).

To overcome this difficulty, the CI method of estimating the risk of revision in the presence of competing risks was developed (40). The rationale was that death is a competing event influencing the risk of revision. It was then suggested that the CI approach provides more accurate results, especially in studies with long-term follow-up, including high number of deaths. This method provides more profound analysis into THR outcomes, allowing comparisons of survival results between different studies (29 -31). These studies followed by a CORR’s Editorial, by Wongworawat et al, suggesting the use of an alternative to KM analysis, when the frequency of competing event is greater than 10%-20% and the follow-up duration reach out 10 years (21).

Footnotes

Disclosures

Financial support: None.

Conflict of interest: None.

References

Charnley

. Low friction arthroplasty of the hip: theory and practice. Berlin: Springer-Verlag; 1979.

Hartofilakidis

Lampropoulou-Adamidou

Stathopoulos

Vlamis

. The Outcome of 241 Charnley Total Hip Arthroplasties Performed by One Surgeon 30 to 40 Years Ago. J Arthroplasty. 2015;30(10):1767-1771.

Jones

Hungerford

. Cement disease. Clin Orthop Relat Res. 1987;(225):192-206.

Belmont

Jr Powers

Beykirch

Hopper

Jr Engh

. Results of the anatomic medullary locking total hip arthroplasty at a minimum of twenty years. A concise follow-up of previous reports. J Bone Joint Surg Am. 2008;90(7):1524-1530.

Della Valle

Mesko

Quigley

Rosenberg

Jacobs

Galante

. Primary total hip arthroplasty with a porous-coated acetabular component. A concise follow-up, at a minimum of twenty years, of previous reports. J Bone Joint Surg Am. 2009;91(5):1130-1135.

Petsatodis

Papadopoulos

Papavasiliou

Hatzokos

Agathangelidis

Christodoulou

. Primary cementless total hip arthroplasty with an alumina ceramic-on-ceramic bearing: results after a minimum of twenty years of follow-up. J Bone Joint Surg Am. 2010;92(3):639-644.

McLaughlin

Lee

. Uncemented total hip arthroplasty with a tapered femoral component: a 22- to 26-year follow-up study. Orthopedics. 2010;33(9):639.

Al Muderis

Bohling

Grittner

Gerdesmeyer

Scholz

. Cementless total hip arthroplasty using the Spongiosa-I fully coated cancellous metal surface: a minimum twenty-year follow-up. J Bone Joint Surg Am. 2011;93(11):1039-1044.

Kolb

Grübl

Schneckener

et al . Cementless total hip arthroplasty with the rectangular titanium Zweymüller stem: a concise follow-up, at a minimum of twenty years, of previous reports. J Bone Joint Surg Am. 2012;94(18):1681-1684.

10.

Loughead

OConnor

Charron

Rorabeck

Bourne

. Twenty-three-year outcome of the porous coated anatomic total hip replacement: a concise follow-up of a previous report. J Bone Joint Surg Am. 2012;94(2):151-155.

11.

Streit

Innmann

Merle

Bruckner

Aldinger

Gotterbarm

. Long-term (20- to 25-year) results of an uncemented tapered titanium femoral component and factors affecting survivorship. Clin Orthop Relat Res. 2013;471(10):3262-3269.

12.

Zügner

Tranberg

Herberts

Romanus

Kärrholm

. Stable fixation but unpredictable bone remodelling around the Lord stem: minimum 23-year follow-up of 66 total hip arthroplasties. J Arthroplasty. 2013;28(4):644-649.

13.

Sandiford

Doctor

Rajaratnam

et al . Primary total hip replacement with a Furlong fully hydroxyapatite-coated titanium alloy femoral component: Results at a minimum follow-up of 20 years. Bone Joint J. 2013;95-B(4):467-471.

14.

Troelsen

Malchau

Sillesen

Malchau

. A review of current fixation use and registry outcomes in total hip arthroplasty: the uncemented paradox. Clin Orthop Relat Res. 2013;471(7):2052-2059.

15.

Hailer

Garellick

Kärrholm

. Uncemented and cemented primary total hip arthroplasty in the Swedish Hip Arthroplasty Register. Acta Orthop. 2010;81(1):34-41.

16.

Mäkelä

Matilainen

Pulkkinen

et al . Failure rate of cemented and uncemented total hip replacements: register study of combined Nordic database of four nations. BMJ. 2014;348(jan13 12):f7592.

17.

Stea

Comfort

Sedrakyan

et al . Multinational comprehensive evaluation of the fixation method used in hip replacement: interaction with age in context. J Bone Joint Surg Am. 2014;96(Suppl 1):42-51.

18.

Hooper

Rothwell

Stringer

Frampton

. Revision following cemented and uncemented primary total hip replacement: a seven-year analysis from the New Zealand Joint Registry. J Bone Joint Surg Br. 2009;91(4):451-458.

19.

Ranstam

Kärrholm

Pulkkinen

et al . NARA study group. Statistical analysis of arthroplasty data. II. Guidelines. Acta Orthop. 2011;82(3):258-267.

20.

Ranstam

Kärrholm

Pulkkinen

et al . NARA study group. Statistical analysis of arthroplasty data. I. Introduction and background. Acta Orthop. 2011;82(3):253-257.

21.

Wongworawat

Dobbs

Gebhardt

et al . Editorial: Estimating survivorship in the face of competing risks. Clin Orthop Relat Res. 2015;473(4):1173-1176.

22.

Hartofilakidis

Stamos

Ioannidis

. Low friction arthroplasty for old untreated congenital dislocation of the hip. J Bone Joint Surg Br. 1988;70(2):182-186.

23.

Hartofilakidis

Stamos

Karachalios

Ioannidis

Zacharakis

. Congenital hip disease in adults. Classification of acetabular deficiencies and operative treatment with acetabuloplasty combined with total hip arthroplasty. J Bone Joint Surg Am. 1996;78(5):683-692.

24.

Charnley

. The long-term results of low-friction arthroplasty of the hip performed as a primary intervention. J Bone Joint Surg Br. 1972;54(1):61-76.

25.

Harris

McCarthy

Jr ONeill

. Femoral component loosening using contemporary techniques of femoral cement fixation. J Bone Joint Surg Am. 1982;64(7):1063-1067.

26.

Zicat

Engh

Gokcen

. Patterns of osteolysis around total hip components inserted with and without cement. J Bone Joint Surg Am. 1995;77(3):432-439.

27.

Hodgkinson

Shelley

Wroblewski

. The correlation between the roentgenographic appearance and operative findings at the bone-cement junction of the socket in Charnley low friction arthroplasties. Clin Orthop Relat Res. 1988;(228):105-109.

28.

Satagopan

Ben-Porat

Berwick

Robson

Kutler

Auerbach

. A note on competing risks in survival data analysis. Br J Cancer. 2004;91(7):1229-1235.

29.

Fennema

Lubsen

. Survival analysis in total joint replacement: an alternative method of accounting for the presence of competing risk. J Bone Joint Surg Br. 2010;92(5):701-706.

30.

Gillam

Ryan

Graves

Miller

de Steiger

Salter

. Competing risks survival analysis applied to data from the Australian Orthopaedic Association National Joint Replacement Registry. Acta Orthop. 2010;81(5):548-555.

31.

Martin

Callaghan

Gao

et al . What Can We Learn From 20-year Followup Studies of Hip Replacement? Clin Orthop Relat Res. 2016;474(2):402-407.

32.

Learmonth

Young

Rorabeck

. The operation of the century: total hip replacement. Lancet. 2007;370(9597):1508-1519.

33.

Hartofilakidis

Babis

Lampropoulou-Adamidou

. Congenital hip disease in adults. Mailand: Springer-Verlag. 2014.

34.

Lampropoulou-Adamidou

Georgiades

Vlamis

Hartofilakidis

. Charnley low-friction arthroplasty in patients 35 years of age or younger. Results at a minimum of 23 years. Bone Joint J. 2013;95-B(8):1052-1056.

35.

Pennington

Grieve

Sekhon

Gregg

Black

van der Meulen

. Cemented, cementless, and hybrid prostheses for total hip replacement: cost effectiveness analysis. BMJ. 2013;346(feb27 1):f1026.

36.

Kallala

Anderson

Morris

Haddad

. The cost analysis of cemented versus cementless total hip replacement operations on the NHS. Bone Joint J. 2013;95-B(7):874-876.

37.

Callaghan

Bracha

Liu

Piyaworakhun

Goetz

Johnston

. Survivorship of a Charnley total hip arthroplasty. A concise follow-up, at a minimum of thirty-five years, of previous reports. J Bone Joint Surg Am. 2009;91(11):2617-2621.

38.

Georgiades

Babis

Hartofilakidis

. Charnley low-friction arthroplasty in young patients with osteoarthritis: outcomes at a minimum of twenty-two years. J Bone Joint Surg Am. 2009;91(12):2846-2851.

39.

Gooley

Leisenring

Crowley

Storer

. Estimation of failure probabilities in the presence of competing risks: new representations of old estimators. Stat Med. 1999;18(6):695-706.

40.

Prentice

Kalbfleisch

Peterson

Jr Flournoy

Farewell

Breslow

. The analysis of failure times in the presence of competing risks. Biometrics. 1978;34(4):541-554.