If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Centre for Statistics in Medicine, Nuffield Department of Orthopaedics, Rheumatology and Musculoskeletal Sciences, University of Oxford, UKGREMPAL Research Group, Idiap Jordi Gol and CIBERFes, Universitat Autonoma de Barcelona and Instituto de Salud Carlos III, Barcelona, Spain
The aim of this study was to estimate lifetime risk of knee and hip replacement following a GP diagnosis of osteoarthritis and assess how this risk varies with patient characteristics.
Routinely collected data from Catalonia, Spain, covering 2006 to 2015, were used. Study participants had a newly recorded GP diagnosis of knee or hip osteoarthritis. Parametric survival models were specified for risk of knee/hip replacement and death following diagnosis. Survival models were combined using a Markov model and lifetime risk estimated for the average patient profile. The effects of age at diagnosis, sex, comorbidities, socioeconomic status, body mass index (BMI), and smoking on risk were assessed.
48,311 individuals diagnosed with knee osteoarthritis were included, of whom 2,561 underwent knee replacement. 15,105 individuals diagnosed with hip osteoarthritis were included, of whom 1,247 underwent hip replacement. The average participant's lifetime risk for knee replacement was 30% (95% CI: 25–36%) and for hip replacement was 14% (10–19%). Notable patient characteristics influencing lifetime risk were age at diagnosis for knee and hip replacement, sex for hip replacement, and BMI for knee replacement. BMI increasing from 25 to 35 was associated with lifetime risk of knee replacement increasing from 24% (20–28%) to 32% (26–37%) for otherwise average patients.
Knee and hip replacement are not inevitable after an osteoarthritis diagnosis, with average lifetime risks of less than a third and a sixth, respectively. Patient characteristics, most notably BMI, influence lifetime risks.
. The hip and knee are the principal large joints affected by osteoarthritis. Following diagnosis of knee or hip osteoarthritis, patient care is primarily managed in primary care. A broad range of non-pharmacological and pharmacological interventions are available as initial treatments after diagnosis
. Further progression of osteoarthritis also does not appear to be inevitable. In two community-based studies, for example, 39% of knees with radiographic knee osteoarthritis had not worsened after 15 years
. Individuals with osteoarthritis experience different symptom trajectories. In one study, disability progressively worsened over time for only 24% of patients. The remaining patients were stable, had short-term fluctuations, or steadily improved over time.
. The lifetime risks of knee and hip replacement would satisfy these requirements and give patients and healthcare providers a better idea of an individual's future healthcare needs. Understanding the effect of modifiable and non-modifiable patient characteristics on lifetime risk can help instigate self-management and patient-driven treatments following diagnosis.
In this study, we estimated the lifetime risks of knee and hip replacement following a GP diagnosis of knee and hip osteoarthritis, respectively. We then assessed the effect of patient characteristics on lifetime risk.
This analysis was based on actual practice data from patients from Catalonia, Spain. Individual-level primary care data were extracted from the Sistema d’Informació pel Desenvolupament de la Investigació a l’Atenció Primària (SIDIAP). SIDIAP (www.sidiap.org) is a database containing patient records of 80% of the Catalan population and is highly representative of the population in terms of geographical, age, and sex distributions
. About 30% of the contributing practices are also linked to the regional Conjunt Mínim Bàsic de Dades (CMBD), a database containing details of admissions to every hospital of Catalonia. The linked records gave us a total source population of 1.7 million active subjects for the period 1 January 2006 to 31 December 2015.
Individuals were eligible for inclusion into the study if they had an incident diagnosis of knee or hip osteoarthritis. Diagnostic coding in SIDIAP is based on ICD-10 codes. The recording of knee and hip osteoarthritis have previously been validated, with a sensitivity of 94% and specificity of 71% when compared to self-reported physician diagnosed osteoarthritis
. Individuals were excluded if they were younger than 50 at their date of diagnosis, had less than 1year of observation time prior to their diagnosis, had a prior diagnosis of inflammatory arthritis, or had a knee or hip replacement recorded before or on their date of diagnosis. The identification of knee and hip replacement is summarised in the outcomes section below.
Participant characteristics at diagnosis
Participant characteristics at diagnosis were extracted from the primary care records. Age and sex were extracted. Participants’ comorbidities were summarised using the Charlson score, with possible scores of 0, 1, 2, or 3+, and a higher score indicating a higher degree of comorbidity
. All observation time prior to the index diagnosis of osteoarthritis were used to identify generate this score. Socioeconomic status was summarised using MEDEA, a census-based indicator that has a distinct category for those living in rural areas
. Participants were grouped by MEDEA quintile (first indicating least deprived and fifth most deprived), or as living in a rural area.
We extracted the most recent record of body mass index (BMI) and smoking status (non-smoker, ex-smoker, or current smoker) before the osteoarthritis diagnosis and kept only those recorded within a year of diagnosis. For the purposes of summarising the observed outcomes, continuous variables were categorised. BMI was categorised following the World Health Organisation categories: normal or underweight (BMI <25), overweight (≥25 and < 30), obese class I (≥30 and < 35), obese class II (≥35 and < 40), and obese class III (≥40). Age was categorised by splitting the data into quartiles (with first indicating the youngest and fourth the oldest).
Instances of total knee and hip replacement (ICD-9 procedure codes 8154 and 8151) were identified using linked hospital records, with the database used collecting details of admissions to every public and private hospital of Catalonia
. All knee and hip replacements were included in the analysis, and so procedures were not necessarily attributable to the prior osteoarthritis diagnosis. Deaths were identified based on recorded date of death in SIDIAP. In the absence of a knee or hip replacement or death during the study period, subjects were censored at the date of exit from a GP practice linked to SIDIAP (e.g., due to moving out of Catalonia or changing practice within Catalonia to one not linked to SIDIAP) or the end of the study (31 December 2015).
Comparison of cumulative incidence of knee/hip replacement
Observed knee and hip replacement over 9 years following diagnosis of knee or hip osteoarthritis were summarised by their cumulative incidence. Cumulative incidence allows the incidence of an event to be estimated while taking competing risk into account
. We summarised cumulative incidence for the study populations as a whole and compared cumulative incidence stratified by participant characteristics of interest (age, sex, Charlson score, MEDEA quintile or living rurally, BMI, and smoking status).
Estimating parametric survival models for risks of knee/hip replacement and mortality
Parametric survival models were estimated for cause-specific risks of knee/ hip replacement and death following diagnosis of knee or hip osteoarthritis, respectively. Such models require an assumption of the underlying distribution of the events of interest. Alternative distributions were compared and chosen on the basis of their fit to the observed data and the plausibility of extrapolation
. Further details on the choice of distributions is provided in Appendix Section 3, with the chosen distributions shown in Appendix Fig. A2-A4.
Univariable and multivariable survival models were estimated for each of the participant characteristics of interest. Non-linearity in continuous variables was incorporated through the use of polynomials, if their inclusion improved fit relative to specifying a linear relationship. Fit was assessed by comparing Akaike information criteria (AIC). Non-proportionality in hazards was assessed using a visual inspection of a log-log plot and by testing the weighted residuals
. As there was evidence of non-proportionality in age for risks of knee and hip replacement, the models were estimated separately for each of the age groups, with age included as an explanatory variable within each of the stratified models.
Missing data in MEDEA, BMI, and smoking status was addressed using multiple imputation, with 50 imputed datasets generated. Both explanatory variables and outcomes were used as predictors for missing data, with predictive mean matching used for BMI and multinomial logit models used for MEDEA and smoking status. Hazard ratios and corresponding 95% confidence intervals (CIs) were calculated using Rubin's rules. Estimates from models based on complete case data were also provided for comparison.
Estimating lifetime risk
To estimate lifetime risk, parametric survival models for knee replacement and death or hip replacement and death were combined using a state-based cohort Markov model. For lifetime risk of knee replacement, a cohort of individuals began as being diagnosed with knee osteoarthritis and then they remained in the diagnosis state or progressed to either the knee replacement or death state as time progressed in yearly cycles. An equivalent structure was used for lifetime risk of hip replacement following a diagnosis of hip osteoarthritis.
Transition probabilities for knee and hip replacement were based exclusively on the parametric survival models estimated for these events. These models were extrapolated beyond the 9 years of observed data to participants’ total remaining lifetime. Transition probabilities for death were based on the relevant parametric models for the first 9 years of the model, after which they were assumed to revert to estimates based on age- and sex-specific lifetables for Spain
. Estimated hazard ratios were similar across multiply imputed datasets and so to reduce computational time, lifetime risks were estimated using the survival models estimated on the first imputed dataset. Parameter uncertainty was incorporated using 1,000 bootstrapped models.
The models were first run for cohorts of individuals with average characteristics (median for continuous variables and mode for categorical ones) for those diagnosed with knee replacement and hip replacement, separately. The partial effect of explanatory factors on lifetime risk was assessed by re-running the models with participant profiles varying in the explanatory factor of interest, while holding other characteristics constant at their average. For the partial effect of continuous variables, a smoothed line was fitted across the lifetime risk estimates for the different simulated values of the variable.
48,311 and 15,105 individuals were included in the knee and hip osteoarthritis cohorts, respectively. A study inclusion flow chart is provided in Appendix Fig. A1. The characteristics of these individuals are summarised in Table I. The median prior observation time over which prior diagnoses and events could be observed was 5 years (with an interquartile range of 3–8 years). A comparison of those with and without missing data in socioeconomic status (MEDEA), BMI, or smoking status is given in Appendix Table A1, and combinations of missing data are summarised in Appendix Figs. A2 and A3. Individuals with missing data were generally younger and had fewer comorbidities than those with complete data. A comparison of observed and imputed values for BMI, the one continuous variable that was imputed, are also summarised in Appendix Figs. A4 and A5.
Table IParticipant characteristics at the time of a knee or hip osteoarthritis diagnosis
Age groups based on quartiles for knee osteoarthritis cohort: 50 to 62, 62 to 69, 69 to 77, 77 to 105, and quartiles for hip osteoarthritis: 50 to 62, 62 to 70, 70 to 78, 78 to 103. MEDEA is a measure of socioeconomic status developed for Catalonia (Spain), with those living in rural areas having a distinct category. BMI: body mass index; IQR: interquartile range.
Gender: male (%)
Charlson score (%)
BMI (median [IQR])
30.4 [27.5, 33.9]
29.1 [26.5, 32.3]
BMI group (%)
Normal or underweight (<25)
Overweight (≥25 and < 30)
Obese class I (≥30 and < 35)
Obese class II (≥35 and < 40)
Obese class III (≥40)
MEDEA quintile or rural (%)
1st (least deprived)
5th (most deprived)
Smoking status (%)
∗ Age groups based on quartiles for knee osteoarthritis cohort: 50 to 62, 62 to 69, 69 to 77, 77 to 105, and quartiles for hip osteoarthritis: 50 to 62, 62 to 70, 70 to 78, 78 to 103. MEDEA is a measure of socioeconomic status developed for Catalonia (Spain), with those living in rural areas having a distinct category. BMI: body mass index; IQR: interquartile range.
Observed risks of knee/hip replacement and mortality
Cumulative incidence of knee and hip replacement in the 9-year period following primary care diagnosis was 9.4% and 11.6% respectively. Cumulative incidences stratified by participant characteristics of interest are summarised in Table II. Those patients in the oldest age quartile had a substantially lower cumulative incidence of both knee and hip replacement than younger patients, males had a substantially higher cumulative incidence of hip replacement than females, and cumulative incidence of knee replacement was greater for those with a higher BMI. The hazard ratios estimated for explanatory factors in each of the cause-specific survival models for knee replacement and death or hip replacement and death are detailed in Appendix Tables A2 and A3. Estimates for models based on complete case data were similar, Appendix Tables A4 and A5.
Table IICumulative incidence of knee or hip replacement at 9 years follow-up
Age groups based on quartiles for knee osteoarthritis cohort: 50 to 62, 62 to 69, 69 to 77, 77 to 105, and quartiles for hip osteoarthritis cohort: 50 to 62, 62 to 70, 70 to 78, 78 to 103. MEDEA is a measure of socioeconomic status developed for Catalonia (Spain), with those living in rural areas having a distinct category. BMI: body mass index; CI: confidence interval; HR: hip replacement; KR: knee replacement; OA: osteoarthritis; PY: person years.
BMI group (%)
Normal or underweight (<25)
Overweight (≥25 and < 30)
Obese class I (≥30 and < 35)
Obese class II (≥35 and < 40)
Obese class III (≥40)
MEDEA quintile or rural (%)
1st (least deprived)
5th (most deprived)
Cumulative incidence of total knee or hip replacement 9 years after a diagnosis of knee or hip osteoarthritis.
∗ Age groups based on quartiles for knee osteoarthritis cohort: 50 to 62, 62 to 69, 69 to 77, 77 to 105, and quartiles for hip osteoarthritis cohort: 50 to 62, 62 to 70, 70 to 78, 78 to 103. MEDEA is a measure of socioeconomic status developed for Catalonia (Spain), with those living in rural areas having a distinct category. BMI: body mass index; CI: confidence interval; HR: hip replacement; KR: knee replacement; OA: osteoarthritis; PY: person years.
Average lifetime risks of knee and hip replacement
At diagnosis, the average participant with knee osteoarthritis was a non-smoking 69-year-old woman in the fourth MEDEA quintile, with a Charlson score of 0 and a BMI of 30. At diagnosis, the average participant with hip osteoarthritis was a non-smoking 70-year-old woman living rurally, with a Charlson score of 0 and a BMI of 29. For participants with these characteristics at diagnosis, the parametric models indicated that the risks of knee and hip replacement peaked in the second year after diagnosis, then fell over time (Fig. 1). After accounting for the competing risk of mortality, these translated into a lifetime risk of knee replacement following a diagnosis of knee osteoarthritis of 30% (95% CI: 25–36%) and a lifetime risk of hip replacement following a diagnosis of hip osteoarthritis of 14% (10–19%).
Participant characteristics and lifetime risks of knee/hip replacement
Lifetime risk of knee and hip replacement following a diagnosis of knee or hip osteoarthritis generally fell as age at diagnosis increased (Fig. 2). Younger women generally had a slightly higher lifetime risk of knee replacement than younger men. For example, a 60-year-old woman had a 37% (27–50%) lifetime risk of knee replacement, while a 60-year-old man had a 30% (22–46%) risk. However, men had a substantially higher lifetime risk of hip replacement than women at younger ages. An average 60-year-old man, for example, had a 30% (25–36%) lifetime risk of hip replacement after a diagnosis of hip osteoarthritis, while a 60-year-old woman had a 17% (12–24%) lifetime risk.
A higher BMI was associated with a substantially higher lifetime risk of knee replacement, but relatively little difference in lifetime risk of hip replacement (see Fig. 3 for the partial effect of BMI on transition probabilities, and Fig. 4 for the partial effect on lifetime risks). Holding other explanatory variables fixed at their average, lifetime risk of knee replacement after a diagnosis of knee osteoarthritis was 24% (20–28%) for a BMI of 25 which increased to 32% (26–37%) for a BMI of 35. Meanwhile, the lifetime risk of hip replacement after a diagnosis of hip osteoarthritis was 12% (9–17%) for a BMI of 25% and 15% (11–19%) for a BMI of 35. Differences in comorbidities, smoking status, and socioeconomic status and rurality had relatively little effect on lifetime risk of knee or hip replacement (Appendix Figures A9-A11).
This is to our knowledge the first study to assess the lifetime risks of knee or hip replacement for a patient diagnosed with knee or hip osteoarthritis in a primary care setting. Despite the prevailing belief that further deterioration is inevitable following diagnosis, we find that the lifetime risks of knee and hip replacement are less than one-third and less than one-sixth following a diagnosis of knee osteoarthritis and hip osteoarthritis, respectively. The risk of undergoing a knee or hip replacement peaks in the second year after diagnosis, then steadily falls.
Older participants generally had lower lifetime risks of both knee and hip replacement than younger participants. Young women had a lower lifetime risk of hip replacement than young men. The lifetime risk of knee replacement increased as BMI increased.
Study findings in context
We found that joint replacement was not inevitable following a GP diagnosis of osteoarthritis. This finding is consistent with previous research into the structural and symptomatic progression of osteoarthritis
. This latter study estimated that 13.5% of the cohort would develop osteoarthritis, implying a 45% lifetime risk of knee replacement for those diagnosed. Although above the estimate for average lifetime risk found in our study, it is not dramatically so.
Risk of knee and hip replacement appears to be highest in the second year for those at average age at time of osteoarthritis diagnosis. This implies that there is a sizeable proportion of patients whom referral to surgery is made shortly following diagnosis. This may be explained to some degree by rapid progression of osteoarthritis for some patients
. It is likely though to be in large part because of a proportion of patients being diagnosed at a late-stage in the disease process, at which point knee or hip replacement was already merited.
We identified participant characteristics associated with differences in lifetime risk. These differences could be due to variation in need, disease progression, time at risk (i.e., risk of mortality), or access to care.
Age at diagnosis had a substantial effect on lifetime risks of knee and hip replacement. With mortality as a competing risk, younger age at diagnosis was associated with a longer time at risk due to greater life expectancy. All else being equal, lifetime risks of knee and hip replacement can be expected to be higher for younger patients. Age also appeared to influence cause-specific risks of knee and hip replacement, with those older at diagnosis generally having a reduced risk. This finding is consistent with previous research that found those over 82 to have less than half the risk of knee and hip replacement than younger patients, even after controlling for severity of osteoarthritis symptoms
, as reflected in our study cohorts with both being majority women. However, our findings suggest that once diagnosed, young men have a substantially higher lifetime risk of hip replacement than young women. Previous research has found that the effect of sex on risk of surgery is mediated by willingness to undergo surgery
Higher BMI was associated in this study with a substantially increased lifetime risk of knee replacement following an osteoarthritis diagnosis. Higher BMI has previously been associated with an increased risk of developing knee osteoarthritis
. In contrast, higher BMI at time of diagnosis of hip osteoarthritis appeared to have little effect on the risk of hip replacement following that diagnosis. This discordance has previously been found in a large population-based cohort study, and may be explained by differences in biomechanical effects.
This study was based on a large, representative sample from routinely collected data, with lifetime risks of knee and hip replacement estimated from time of GP diagnosis of knee or hip osteoarthritis. The routinely collected data used, however, only covered a window of time and so historical diagnoses and procedures may have been missed, leading to a possible underestimation of comorbidities and a failure to exclude some patients, for example those who had a diagnosis of inflammatory arthritis prior to their observation time. In addition, further research into the generalisability of our findings would be useful. Previous research has shown that risks of joint replacement for the general population vary across countries
, and this will likely also be the case for risks for those diagnosed with osteoarthritis.
The use of parametric survival models with flexible distributions allowed us to estimate lifetime risk, which is an understandable, and possibly the most pertinent, description of risk. By using parametric models instead of lifetable methods, we were able to thoroughly analyse the effect of patient characteristics on the risk of knee or hip replacement and mortality. However, this approach required extrapolation well beyond the end of study follow-up, particularly for younger patients. This extrapolation necessarily had a high degree of uncertainty, as reflected in the wide CIs around estimates for younger patients.
When analysing the relationship between patient characteristics and lifetime risks of knee and hip replacement, we were limited to those factors available in routinely collected data. A wide range of other factors are likely to influence lifetime risk, such as willingness to undergo knee or hip replacement and disease severity
. In addition, this analysis was limited to risk of first knee or hip replacement following a diagnosis of osteoarthritis with no data available on laterality at diagnosis or at knee or hip replacement. If data on laterality were available for future research, analyses incorporating this information could provide a more detailed assessment of prognosis.
Knee and hip replacement are not inevitable following a GP diagnosis of knee or hip osteoarthritis. Those with knee osteoarthritis have a lifetime risk of less than a third for knee replacement, and those with hip osteoarthritis have a lifetime risk of less than a sixth for hip replacement. These findings provide a clear indication of prognosis for doctors and patients, which should help to inform treatment choices after diagnosis.
Risk of knee and hip replacement generally peaked in the second year following diagnosis in this study. This is likely because of a late diagnosis for a proportion of the study participants, with diagnosis made at a point where knee or hip replacement was already merited. This finding underscores the importance of timely diagnosis, following which non-operative treatments can be pursued.
Lifetime risks of knee and hip replacement vary depending on patient characteristics at diagnosis. In particular, higher BMI is associated with an increased risk of knee replacement. Effective weight loss interventions provided at the time of a knee osteoarthritis diagnosis would therefore likely lead to substantial health benefits for patients and cost savings for the health system.
EB, DWM, RPV, and DPA made substantial contributions to the conception and design of the study. EB, DWM, GH, RPV, and DPA made substantial contributions to the interpretation of the data for the work. EB, RPV, and DPA undertook the statistical analysis. EB, RPV, and DPA drafted the manuscript, with GH and DWM revising it for important intellectual content. All authors read and approved the final manuscript.
Conflict of interest
All authors have completed the ICMJE uniform disclosure form at www.icmje.org/coi_disclosure.pdf and declare: DPA reports grants from Amgen, Servier and UCB Biopharma, and non-financial support from Amgen, all outside the submitted work. RPV reports consultancy fees from Kyowa Kirin, UCB, and Mereo, all outside the submitted work. DWM reports grants and personal fees from Zimmer Biomet. In addition, DWM has various patents related to Unicompartmental Knee Replacement (Zimmer Biomet) with royalties paid, all outside the submitted work.
DPA is funded by a National Institute for Health Research Clinician Scientist award (CS-2013-13-012). This article presents independent research funded by the National Institute for Health Research (NIHR). The views expressed are those of the authors and not necessarily those of the NHS, the NIHR, or the Department of Health. This work was supported by the NIHR Biomedical Research Centre, Oxford. GH receives salary support as the Sir John and Lady Eaton Professor and Chair of Medicine at the University of Toronto.
Approval for all observational research using SIDIAP data is obtained from a local ethics committee (Clinical Research Ethics Committee of the IDIAP Jordi Gol).
Data were provided under a licence that does not permit sharing. Data are obtainable from the SIDIAP subject to a full application.
The senior and corresponding authors (DPA and RPV) affirm that this manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned have been explained.
The authors would like to thank Miss Susan Thwaite (National Rheumatoid Arthritis Society) for her role as the patient and public representative and her role on the study steering committee. We also thank Dr Jennifer A. de Beyer of the Centre for Statistics in Medicine, University of Oxford, for English language editing.
Appendix A. Supplementary data
The following is the supplementary data to this article: