Survival analysis of joint implants in a regional registry
Kaplan-Meier, Cox models and Fine-Gray competing risks to assess implant durability over more than a decade of population follow-up
Assessing the long-term durability of joint implants in a population registry, with mortality treated as a competing event
A regional health technology assessment agency in Spain needed to update the results report of its joint implant registry, bringing in advanced survival methodology and producing both an institutional report and an interactive platform for professionals to consult.
The registry covered every joint implant procedure performed in the region's public health system over more than a decade, with data on the primary procedure, revisions and patient follow-up. The volume of data was substantial: multiple joint types, different fixation techniques (cemented, uncemented, hybrid and reverse hybrid), a range of implant models, and longitudinal follow-up that called for appropriate handling of censoring and competing events.
The methodological challenge had two dimensions. The first was to estimate the cumulative risk of implant revision while taking into account that the patient's death competes with revision as a terminal event — ignoring this competing risk overestimates the probability of revision, especially in older patients and those with comorbidities. The second was to produce results stratified by joint type, fixation technique, implant model, sex, age and comorbidities, in a format that would serve both an institutional report for managers and a web platform that surgeons could consult.
10+ years
of population follow-up linked to the mortality registry
3 methods
of survival analysis: Kaplan-Meier, multivariable Cox and Fine-Gray
4 techniques
of fixation assessed (cemented, uncemented, hybrid and reverse hybrid)
A survival pipeline with three levels of complexity: descriptive Kaplan-Meier, multivariable Cox and Fine-Gray competing risks
Stratified descriptive survival analysis
Kaplan-Meier survival curves estimated at 1, 3, 5, 10 and 15 years for each combination of joint and fixation technique. Tables of cumulative revision risk with confidence intervals. Stratification by sex, age group and specific implant model, allowing direct comparison between surgical alternatives.
Multivariable Cox models adjusted for comorbidity
Hazard ratios estimated with adjustment for sex, age and number of comorbidities (Elixhauser index) to quantify the revision risk associated with each factor. Separate analyses by joint type and fixation technique. Identification of the factors associated with a higher risk of revision, providing evidence for the comparative assessment of implants.
Competing risks: cumulative incidence and a Fine-Gray model
Estimation of the cumulative incidence function of revision, treating the patient's death as a competing event, and a Fine-Gray model to estimate which factors are associated with revision when death competes with it. Comparison between the Kaplan-Meier estimate and the competing-risks cumulative incidence function to quantify the overestimation produced by treating death as simple censoring. The analysis showed that the overestimation grows with follow-up time and becomes relevant from the fifth postoperative year onwards — a finding with direct implications for interpreting any implant registry with long-term follow-up.
Evidence for health technology assessment and clinical decision-making on implants
Quantifying competing-risk bias in long-term implant registries
A systematic comparison of estimates with and without adjustment for mortality showed that treating death as censoring can overestimate the probability of revision, with a divergence that widens progressively with follow-up time. This methodological finding applies to any implant or device registry with long follow-up in an ageing population.
An updated institutional report built on reference methodology
A Word deliverable with a parallel structure for each joint type, following the reporting standard of national arthroplasty registries. Results stratified by period, fixation technique, implant model, demographics and comorbidities, with tables and figures ready for institutional publication.
An interactive web platform for professionals
A website generated with Quarto that lets surgeons and managers look up survival results by joint type, technique and model. Sidebar navigation gives direct access to each section, removing the need to work through a report of hundreds of pages to find the data relevant to a specific clinical decision.
Project carried out under a confidentiality agreement. The client's identity and the specific registry data have been omitted.
What this case shows about the way we work
Competing risks built in as standard, not as a post hoc analysis
Fine-Gray estimation with death as a competing event was part of the analytical design from the outset, not a supplementary sensitivity analysis. In implant registries with an ageing population, ignoring competing mortality can distort the estimates used to assess and compare implants.
A dual institutional and clinical deliverable from a single pipeline
The same R code generates a Word report for managers and a web platform for clinicians, removing duplicated work and ensuring consistency between the two products. A multi-profile Quarto architecture means both deliverables can be updated at the same time when new data are added to the registry.
A replicable structure for device and implant registries
The analytical pipeline is designed to be reusable: any implant registry (joint, cardiovascular, ophthalmic) with a similar structure can apply the same stratified survival methodology with competing risks and automated report production.
Familiarity with the ecosystem of institutional health registries
Demonstrated experience working with data from regional population registries, with their particular features: linkage between databases, procedure coding, handling of administrative censoring and institutional reporting requirements.
Capabilities deployed in this project
Shall we talk?
Tell us what you are working on and we will give you a straight answer on how we can help.