All-cause Mortality in Knee and Hip Osteoarthritis and Rheumatoid Arthritis

Abstract

Background: 

While increased mortality in rheumatoid arthritis (RA) is well established, there is conflicting evidence on the association between osteoarthritis (OA) and mortality. Our aim was to estimate all-cause mortality in Swedish patients with RA and OA compared with the general population.

Methods: 

Cohort study of the population of Skåne region, Sweden (1.3 million), based on physicians’ diagnostic codes in a mandatory register covering all health care. We included all subjects aged ≥45 years who between 1998 and 2012 consulted any physician at least once. We identified those who received a diagnosis of RA, knee OA, or hip OA. We followed all subjects until death, relocation outside Skåne region, or end of 2013, and analyzed data using Cox proportional hazard regression with attained age as time scale.

Results: 

We identified 8,067 patients with RA, 51,939 with knee OA and 29,442 with hip OA among 524,136 in the population aged ≥45 years. The mortality rates adjusted for sex, socioeconomic status, and comorbidities were elevated for RA, hazard ratio 1.86 (95% confidence interval = 1.78, 1.94) but not in knee or hip OA compared with the general population seeking health care, hazard ratio 0.87 (0.85, 0.89) and 0.90 (0.87, 0.92), respectively. Extensive sensitivity analyses supported the conclusion of no increased mortality in OA.

Conclusions: 

In Sweden, RA is associated with about doubled mortality rate, but we found no increased mortality in patients with knee and hip OA. Possible selection of those seeking physician care for knee or hip pain and/or OA management in health care are plausible explanations.

Rheumatoid arthritis (RA) and osteoarthritis (OA) are associated with substantial joint pain and reduced function and quality of life in populations worldwide.^1,2 They are both common in middle-aged and elderly people and more common in women than men.^3,4 The major risk factors include smoking and genetic factors for RA and obesity, joint injury, and genetic factors for OA.^5,6

There is ample evidence of increased mortality in patients with RA compared with the general population, where inflammation (and its treatments), and higher levels of infectious, pulmonary, cardiovascular, and renal comorbidities contribute.⁷ In contrast, the relationship between OA and mortality is less clear. The comprehensive systematic review by Hochberg in 2008 concluded in a best evidence synthesis of six studies that there was moderate evidence of slightly increased all-cause mortality among persons with OA compared with the general population, particularly for knee OA and generalized OA.⁸ Since then, Nüesch et al.⁹ have reported that persons with symptomatic knee or hip OA had a standardized mortality ratio of 1.55 as compared with the general population. Other investigators have reported lower mortality rates for OA as compared with other chronic conditions or,¹⁰ for hand OA, with the general population.¹¹ However, as noted by Hochberg,⁸ a concern with the studies based on health-care databases has been potential misclassification of OA resulting in bias toward the null.^10,12 The same studies had usually a relatively short follow-up that may have been too short to capture the late effects of OA on mortality. Furthermore, the other studies were small and/or had a potential for selection bias, as the controls may have not been drawn from the same underlying population as the cases.^9,13,14 In addition, different OA management in different countries and health-care systems may have different consequences on all-cause mortality as mortality in OA has been suggested to be related to the actual physical disability rather than biological consequences of OA.^9,15

Thus, our objective was to determine mortality rates in Swedish patients with physician-diagnosed OA of the knee or the hip as compared with the general population seeking health care using a validated health-care register covering the entire population of the most southern part of Sweden with a follow-up period of 16 years. In conjunction, we estimated mortality rates in patients with physician-diagnosed RA, primarily to ascertain the internal validity of the register data. We performed extensive sensitivity analyses to assess the potential impact of misclassification of the disease on the results.

METHODS

Data Sources

The Skåne region, located in southern Sweden has 1.27 million inhabitants (December 31, 2013). Health care provided on primary, specialist, and inpatient levels by both public and private providers in this region is registered in the Skåne Healthcare Register (SHR). The registration is, among others, used for reimbursement purposes and is compulsory by law. Both private and public providers are reimbursed through the same tax-based financing system and the patient’s co-pay is similar irrespective of provider type. It is common that patients seek care from, or are referred to, both types of providers. Around 30% of doctor visits in primary care are through private providers. The SHR includes data on the health-care provider, the profession (physician, physical therapist, etc.), type of contact (e.g., primary/specialist, in- or out-patient visit, clinic, etc.), and the visit date. Furthermore, the register contains the publicly practicing physicians’ diagnostic codes according to the International Classification of Diseases (ICD) 10 system (the codes from private providers are not forwarded). These codes are assigned at the time of the health-care visit by the doctors themselves and are automatically transferred to the register from the electronic medical records.

A separate Swedish population register contains complete information on births, deaths, and changes in residential addresses for all residents in Sweden and is continuously updated by the Swedish Tax Agency. Statistics Sweden provided data on income, education, and marital status. All data were linked through the coded personal unique identification number that is assigned to all residents in Sweden by the Swedish Tax Agency.

The study was approved by the Lund University Ethics Committee.

Study Design

We retrieved data for all health-care visits between January 1, 1998 and December 31, 2012 and included all persons with at least one health-care visit with any diagnosis registered during this time to minimize potential confounding due to propensity to seek care. We required that the subjects were aged 45 years or older and resident in the Skåne region at the time of the first visit. Using information from the population register, we followed all subjects from the time of first health-care visit until death, relocation outside Skåne region or December 31, 2013, whichever occurred first.

To be considered an RA case we required an ICD-10 diagnostic code M05 or M06 assigned at least twice, with at least one being from a specialist (or a physician under specialty training) in rheumatology or internal medicine. This definition was validated in a previous study and yielded sensitivity of 93%.¹⁶ To be considered an OA case, we required at least one visit with an ICD-10 diagnostic code M17 (knee OA, including codes M172 and M173 denoting posttraumatic knee OA) or M16 (hip OA) assigned. The reason is that OA (in contrast to RA) is a disease that does not necessarily require scheduled follow-ups and the definition has been used previously.¹⁷ The positive predictive value for knee OA in the register has been estimated in a subset of this population (aged 56 to 84) to be 88% with respect to radiographic or symptomatic disease.¹⁸ To avoid immortal time bias, the follow-up time from the first health-care visit until the date at which the RA or OA definition was fulfilled was treated as unexposed, while the follow-up time from the date of fulfilling the RA or OA definition and onwards was treated as exposed (Figure 1).¹⁹

FIGURE 1: Study design.

Covariates

For included subjects, we retrieved information on residential area, income, highest level of achieved education, and marital status in the year of their first health-care visit. Income and marital status were missing for 0.4% of all subjects while education was missing for 5%. Those with missing data were excluded from the adjusted analyses.

We included chronic comorbidities that are main causes of death, i.e., ischemic heart diseases (ICD-10 code: I20–I25), cerebrovascular disease (I60–I69), malignant neoplasms (C, other than C34), and diabetes mellitus (E10–E14); and those associated with smoking, i.e., chronic obstructive pulmonary disease (J44) and malignant neoplasm of bronchus and lung (C34). To identify persons suffering from those comorbidities, we used the SHR diagnoses from years 1998 to 2012, i.e., the same data source and time period as to define the OA and RA. The percentage of persons with registered comorbidities is available in eTable1 (https://links.lww.com/EDE/B35). We assumed that a particular comorbidity is present from the date of the first registration of this comorbidity in SHR and onward (eFigure; https://links.lww.com/EDE/B35).

Statistical Analysis

To estimate the hazard ratio (HR) of death in patients with RA, knee OA, or hip OA as compared with the general population seeking health care, we used the Cox proportional hazard model with attained age as time scale.²⁰ When adjusting for chronic comorbidities, we used the inverse probability weighting approach to be able to estimate the total effect of RA and OA on mortality in the presence of the time-dependent confounding. Those comorbidities may act both as confounders, if they influence the risk to become RA/OA case (or to get the RA/OA registered) and act as independent predictors of survival, and as mediators if they are part of casual pathway from RA/OA to death. Thus, both the unadjusted and the standard survival analysis adjusted for those comorbidities could be biased if the aim was to estimate the total effect of the RA/OA on mortality.²¹ To estimate the HR of death in persons with RA/OA as compared with the general population adjusted for the chronic comorbidities, we have followed the implementation described by Fewell et al.²² The model was fitted in a two-stage process in which, first, we estimated each subjects probability of having his/her own exposure (or censoring) history and used these to derive inverse-probability-of-exposure (or censoring) weights; second, the exposure–death association was estimated in a regression model that was weighted accordingly. We have used separate models to estimate HR of death in subjects with RA, knee OA, and hip OA.

We have discretized the analysis set by attained age in 1-year strata and created the power basis for a natural cubic spline at the 1st, 25th, 50th, 75th, and 99th percentile of the age distribution. We have used the logistic regression model with RA/OA status as outcome to calculate the stabilized inverse-probability-of-exposure weights. We have used the logistic regression with censoring status as outcome to calculate the stabilized inverse-probability-of-censoring weights. In the models, we have included baseline covariates (sex, income, highest level of achieved education, marital status, residential area, and year of first health-care visit), the cubic splines for attained age, the attained age, the current comorbidity status, the comorbidity status in the previous year, and the comorbidity status 2 years before. Inclusion of more years in the model did not change the results (eTable2; https://links.lww.com/EDE/B35). We have calculated the final weights by multiplying the stabilized inverse-probability-of-exposure weights with the stabilized inverse-probability-of-censoring weights (eCode 1; https://links.lww.com/EDE/B35). We have truncated the weights at 1st and 99th percentile to reduce variance.²³ The weights had means close to one (eTable2, eFigure 2; https://links.lww.com/EDE/B35). We estimated the HRs of death and its 95% confidence intervals using a weighted logistic regression model with vital status (dead or alive) as dependent variable and the following independent variables: RA/knee OA/hip OA, respectively, the cubic splines for the attained age, attained age, and baseline covariates. We used robust standard errors to account for clustering within individuals.²⁴

Sensitivity Analyses

We conducted a series of sensitivity analyses to investigate and quantify the potential impact of three factors on our findings: unmeasured confounding, misclassification of OA case status, and the choice of comparator population.

First, using a rule-out approach,²⁵ we estimated how common among OA cases and how strongly associated with mortality an unmeasured binary confounder would need to be to explain our results if physician diagnosed knee and hip OA truly had an increased mortality risk (HR = 1.5) or had no association with mortality (HR = 1).⁹ We assumed a confounder as prevalent as physical activity “sufficient for sustaining health” (equivalent to approximately 3000 metabolic equivalent of task minutes per week) in the general population, i.e., 30%.²⁶

Second, based on previous data on consultation frequency among knee OA patients in Skåne region,²⁷ we assumed an extreme scenario that 40% of all true cases with OA in the target population were misclassified as non-OA in our study cohort, and that those 40% would have a true mortality ratio of 1.5 as compared to the general population. We sampled additional OA patients (separately for knee and hip) from those not fulfilling our OA case definition so that their mortality ratio would be as specified above (i.e., we sampled proportionally more individuals that died during the study period than those who did not) and then repeated the Cox analysis adjusted for sex and other baseline covariates. We performed this procedure 1,000 times and reported estimates with bootstrap confidence intervals.

Third, we repeated the main analysis using the entire population of the Skåne region aged 45 years or older in 1998 (N = 463,900), i.e., without restricting to those with a health-care visit to a physician as an inclusion criterion.

In addition, we estimated the mortality in persons undergoing joint replacement due to knee or hip OA compared with the general population seeking health care (eTable 3; https://links.lww.com/EDE/B35).

RESULTS

We identified 8,067 subjects with RA, 51,939 with knee OA and 29,422 with hip OA among 524,136 persons aged 45 years or older that consulted a physician at least once between 1998 and 2012. The mean (SD) age of the cohort at inclusion was 63.3 (12.2) and 53% were women (Table 1). The mean (SD) age at the time of the diagnosis was 68 (11), 70 (11), and 72 (10) years for persons with RA, knee OA, or hip OA, respectively. Among persons with knee OA, 4.5% were diagnosed with the posttraumatic knee OA at least once during the study period.

TABLE 1: Descriptive Characteristics of the Study Cohort

The mean (range) follow-up time was 10.3 (0–16) years and 157,008 persons (30% of the cohort) died before the end of the study period. Four percent of subjects were censored due to relocation outside the Skåne region.

The HR of death adjusted for sex, baseline covariates, and comorbidities were elevated for RA patients with the estimate of 1.86 (95% confidence interval [CI] = 1.78, 1.94), 1.59 (95% CI = 1.47, 1.72) for men and 2.03 (95% CI = 1.93, 2.13) for women (Table 2). The relative mortality in RA was highest in adults under 75 years old (Table 3).

TABLE 2: Mortality in Physician-diagnosed RA, Knee OA, and Hip OA Estimated with the Cox Proportional Regression Model with Attained Age as Time Scale

TABLE 3: Mortality in Physician-diagnosed RA, Knee OA, and Hip OA, by Age Group

We did not find higher mortality in hip or knee OA as compared with the general population with adjusted HRs of 0.87 (95% CI = 0.85, 0.89) for knee OA and 0.90 (95% CI = 0.87, 0.92) for hip OA. The results were similar for men and women (Table 2) and after excluding persons with traumatic knee OA (data not shown).

Sensitivity Analyses

In the first sensitivity analysis, we found that if the true HR of death in OA was 1.5, an unmeasured binary confounder that was present in 30% of the general population would need to be three times more prevalent in the OA group than in the general population and would need to decrease mortality by a ratio of 0.45 to explain an apparent HR of 0.9. If the true HR of death was 1, an unmeasured binary confounder would need to be twice as prevalent in the OA group and decrease the hazard of death by ratio of 0.7 to explain our results (Figure 2).

FIGURE 2: Required strength of an unmeasured binary confounder needed to explain an apparent hazard ratio of 0.9 with the assumed 30% prevalence of the confounder in the general population. AHR indicates apparent (observed) hazard ratio of OA-death association; PC0, prevalence of the confounder in the population without OA.

Assuming that 40% of all persons with OA were misclassified as non-OA in our cohort and that those would have a higher mortality than the general population (with an HR = 1.5), the HR of death for persons with OA as compared with those without would be 1.11 (95% CI = 1.10, 1.12) for knee OA and 1.12 (95% CI = 1.11, 1.13) for hip OA.

When we included the entire general population, without the requirement of a consultation with a physician, the HR of death, adjusted for sex, socioeconomic factors, and comorbidities was 2.05 (95% CI = 1.97, 2.13) for RA, 0.92 (95% CI = 0.90, 0.94) for knee OA, and 0.95 (95% CI = 0.93, 0.97) for hip OA.

DISCUSSION

In this population-based study in Sweden, we identified individuals with physician diagnosed RA or knee or hip OA. We found about a twofold increased risk of mortality in persons with RA as compared to the general population seeking health care. However, we did not find any increased mortality in persons diagnosed with knee or hip OA.

The higher mortality in persons with RA has been well recognized and our estimates are in line with those previously published.^7,28,29 The relative mortality in RA was higher in women and younger adults with a peak for the ages 65 to 74 years in contrast to the absolute mortality rates that are higher for men and elderly. Both findings are in accordance with previously published evidence.⁷

Importantly, and in contrast to the findings reported by Nüesch et al.⁹ (and Lawrence et al for women), we did not find any increased mortality in those with knee or hip OA.³⁰ However, our results are in line with those reported for OA in any joint, based on health-care data.^10,12 There may be several explanations for the discrepancy. First, those who are prone to consult a physician for OA may systematically differ from other subjects with OA with respect to general health, attitude to health-care systems, and treatments. In England, most of those aged ≥55 and identified as having extremely severe pain and disability reported that they had seen their general practitioner (GP) within the previous year about their joint problems.³¹ Persons with chronic pain or severe pain consult a GP more often, but the median number of comorbid consultations was reported to be similar between those who do and do not consult.^32,33 Those who did not consult a GP had on average lower level of pain and functional limitations than consulters.³⁴ This suggests that the persons with OA who consult a physician would not be expected to be healthier on average than those who do not consult a GP. On the other hand, in a sample from the population of Malmö, two of three with symptomatic knee OA received a knee OA diagnosis from a physician during an 8-year period.²⁷ In England, of those with at least 4 weeks of knee pain in the past year, 16% consulted a GP during a course of 1 year.³⁵ Thus, there may be a potential for misclassification of some of the severe OA cases in a health-care setting. We have conducted a conservative sensitivity analysis that showed that even a major misclassification of the OA status in our cohort (i.e., if we had erroneously classified 40% of all OA as non-OA) would only yield a 10% increased mortality if those misclassified would have a HR of 1.5. In addition, to adjust for a possible effect of “competing diagnoses,” i.e., if a patient consulting for OA might not get the OA diagnosis registered if he or she suffers from other more serious diseases,³⁴ we adjusted the analyses for the chronic comorbidities associated with mortality.

Second, previous studies based on health-care data had a shorter follow-up (mean of 5 years) than studies based on a radiographic OA definition. This could yield lower HRs if the negative effects of OA on mortality appear late in the disease course. Monson and Hall¹³ noted a decrease in relative survival first after 10 years of follow-up. In our study, with a maximum of 16 years follow-up, we did not see such a decline.

Third, another possible explanation could be the underlying health state of the population or the treatment of OA received when consulting a physician. This treatment could possibly modify the intermediate factors, such as physical inactivity, weight gain, and use of nonsteroidal anti-inflammatory drugs in pain management, which are associated with increased mortality. In Sweden, the prevalence of obesity is lower,^36,37 while the proportion of individuals involved in physical activity is higher than in most European countries.³⁶ In our study sample, the median number of visits to a physiotherapist per one follow-up year was 0.87 after a knee OA diagnosis versus 0.11 for the general population. This includes exercise training given as a standard OA care within a management program that is accessible for everybody and to a high extent financed by the health-care system. Exercise training has been reported to be effective not only in reducing OA-related pain³⁸ but also in improving the general health and could possibly result even in the reduction of walking disability that was identified as a strong predictor of death in persons with symptomatic OA.^9,15

One-fourth of all with knee OA and half of all hip OA patients underwent joint replacement surgery at some point during the follow-up. Those subjects may be healthier than other persons with OA due to bias by indication. On the other hand, the joint replacement has been recognized as effective in reducing pain and improving cardiovascular fitness³⁹ even if functional limitations and muscle weakness are more common than in control subjects without knee complaints.⁴⁰

Fourth, high level physical activity increases the risk of joint injury and thus of posttraumatic OA. It is unknown what proportion of those with OA diagnosis that had a previous joint injury. It was estimated that about 5% to 12% of the overall prevalence of symptomatic OA may be due to previous injury in the knee, hip, or ankle,^41,42 but this proportion could be expected to be higher in a health-care setting, especially in younger age groups. We have evaluated a possible effect of adjustment for physical activity in the sensitivity analysis. It suggested that a higher level of pre-OA physical activity, not adjusted for in our analysis model, could possibly result in a HR of 0.9 if the true HR was 1.0. A similar scenario would be rather implausible if the true HR was 1.5.

The previous evidence of high mortality in OA comes mainly from a cohort study from Lawrence (higher mortality in women) published in 1989 and from a more recent study of Nüesch et al.^9,30 Since 1989, the scientific evidence for different treatments in OA has increased considerably and the importance of physical activity despite pain has been advocated while the risks associated with use of nonsteroidal anti-inflammatory drugs have been well recognized.³⁸ Mortality due to cardiovascular diseases, believed to be a major cause of death in OA, has decreased considerably over the past 30 years.⁴³ It may be that temporal trends in OA-related mortality during the past decades explain, at least in part, the discordance between our results and those reported by Lawrence et al. The comparison with the study from England is more challenging, because the reported mortality ratios were standardized to the general population and based on wide, 10-year, age stratum while the knee and hip OA groups consisted of highly symptomatic patients willing to participate in the study, resulting in the potential for selection bias. A recent study on mortality in older women with hip OA suggested an increased risk for death due to cardiovascular causes but not cancer or gastrointestinal disease.⁴⁴

We would like to acknowledge some limitations of our study. We were not able to adjust our analyses for smoking or body weight that are important confounders of the association between RA/OA and mortality, respectively. Although we did adjust for chronic lung diseases associated with smoking and cardiovascular comorbidities and diabetes associated with obesity, we could not exclude residual confounding. The involvement of subclinical inflammation and the decrease in physical activity in OA suggest that persons with the disease could be at higher risk of death from cardiovascular causes specifically.^9,14 Unfortunately in this study, we were not able to retrieve causes of death. Finally, the RA and OA cases identified were not necessary incident cases (i.e., could have consulted health care for RA/OA before year 1998) and thus the time before the first occurrence of a RA/OA diagnosis in the register could be misclassified (treated as if the patient did not have the disease). However, as the RA and OA cases constituted a minor percentage of the whole cohort (10% at most), this would have only minor influence on the mortality rates in the general population.

Nonetheless, our study also has a number of strengths. We provide updated and partly novel knowledge of mortality in RA and OA. The large size of the cohort enabled identification of almost 8,000 RA and 80,000 OA patients, much larger numbers than included in previous studies, and we used a long follow-up time, up to 16 years. We were able to adjust the analyses for several relevant confounders, including socioeconomic variables and the most important comorbidities. We have also performed extensive sensitivity analyses that assessed the robustness of our results.

CONCLUSIONS

We confirmed an almost twofold increased mortality in patients diagnosed with RA by a physician as compared to the general population seeking health care. However, we could not find evidence for a higher mortality in 21st century physician diagnosed knee or hip OA in Sweden. We speculate that the underlying health of the population may be a part of the explanation. In addition, the modern health-care management of OA in Sweden with easy and equal access to exercise training and joint replacement might contribute to preventing a part of excess mortality in knee and hip OA. However, we cannot exclude that OA patients diagnosed in a health-care setting differ systematically from the underlying OA population and thus our results may only be generalized to all OA with caution.

ACKNOWLEDGMENTS

We would like to express our sincere appreciation to Charlotte Bergknut from Epidemiology and Register Centre South for extracting the data from SHR and to Susanne Bauer from Lund University for language editing.