To estimate the incidence and lifetime risk of diagnosed symptomatic knee osteoarthritis (OA) and the age at diagnosis of knee OA based on self-reports in the US population.
To estimate the incidence and lifetime risk of diagnosed symptomatic knee osteoarthritis (OA) and the age at diagnosis of knee OA based on self-reports in the US population.
We estimated the incidence of diagnosed symptomatic knee OA in the US by combining data on age-, sex-, and obesity-specific prevalence from the 2007–2008 National Health Interview Survey, with disease duration estimates derived from the Osteoarthritis Policy (OAPol) Model, a validated computer simulation model of knee OA. We used the OAPol Model to estimate the mean and median ages at diagnosis and lifetime risk.
The estimated incidence of diagnosed symptomatic knee OA was highest among adults ages 55–64 years, ranging from 0.37% per year for nonobese men to 1.02% per year for obese women. The estimated median age at knee OA diagnosis was 55 years. The estimated lifetime risk was 13.83%, ranging from 9.60% for nonobese men to 23.87% in obese women. Approximately 9.29% of the US population is diagnosed with symptomatic knee OA by age 60 years.
The diagnosis of symptomatic knee OA occurs relatively early in life, suggesting that prevention programs should be offered relatively early in the life course. Further research is needed to understand the future burden of health care utilization resulting from earlier diagnosis of knee OA.
Knee osteoarthritis (OA) is a painful, disabling condition that affects an estimated 9.3 million US adults (1). Because no disease-modifying treatments are available, treatments for symptomatic knee OA focus on symptom relief and functional restoration, including physical therapy, medications, joint injections, and total knee replacement (TKR) (2–5).
For decades, knee OA had been viewed as a disease mostly affecting older persons. However, recent evidence documents increased incidence of 2 key risk factors for knee OA, i.e., traumatic knee injury (6) and obesity (7, 8), particularly in younger persons (9, 10). Evolving data point to a high prevalence of posttraumatic knee OA in younger persons (11–13). This trend and the increasing prevalence of obesity among children are likely to lead to increased rates of OA in young adults (14).
Most population-based data on knee OA prevalence and incidence refer to studies conducted in the mid-1990s (15–17). In fact, the 2 most recent studies reporting the incidence of symptomatic knee OA in the US were published 18 years ago in 1995 (15, 16). Furthermore, although obesity is a known risk factor, no studies have reported age- and sex-specific knee OA incidence separately for obese and nonobese persons. Determining the age at diagnosis of symptomatic knee OA is critical to understanding the trajectory of decrements in quality of life and utilization of health services.
We sought to use national self-reported data to derive current age-, sex-, and obesity-stratified estimates of the incidence of diagnosed symptomatic knee OA, to estimate the mean and median ages at diagnosis, and to estimate the lifetime risk of diagnosis among 25-year-olds representative of the US population.
We provide estimates of the incidence of diagnosed symptomatic knee osteoarthritis (OA) in the US using self-reported population-based national data from 2007-2008.
The estimated mean age at diagnosis of symptomatic knee OA was 53.5 years and the estimated median age at diagnosis was 55 years.
With half of the cases of symptomatic knee OA diagnosed by age 55 years, the burden of future health care utilization for knee OA may be high.
The incidence of diagnosed symptomatic knee OA was derived as a ratio of disease prevalence odds to disease duration (18) (see Technical Appendix, section 2, for a more detailed explanation of incidence calculations, available in the online version of this article at http://onlinelibrary.wiley.com/doi/10.1002/acr.21898/abstract). We used self-reported data from the 2007–2008 National Health Interview Survey (NHIS) to estimate the prevalence of diagnosed symptomatic knee OA (19). Disease duration was derived using the Osteoarthritis Policy (OAPol) Model, a published and validated computer simulation model of knee OA natural history and management (20, 21). Incidence was estimated within 10-year age groups and was further stratified by sex and obesity status. We then used the OAPol Model and our newly derived incidence estimates to determine the mean and median ages at diagnosis of symptomatic knee OA in the US. To model disease duration, cohorts were initialized with knee OA, and the output of interest was life expectancy. To model the age at diagnosis, cohorts were initialized without knee OA, and the output of interest was the development of OA, or cumulative incidence of OA.
The OAPol Model is a validated, state transition Monte Carlo computer simulation model of the natural history of knee OA (20, 21). “State transition” implies that the natural history and clinical management of knee OA are characterized as a series of annual transitions between health states. The health states are characterized by age, sex, obesity, comorbidities, and knee OA disease severity among those who develop the condition. The model uses a set of transition probabilities to determine each individual's sequence of annual transitions among different health states, which may include diagnosis of symptomatic knee OA. Every subject without knee OA is considered to be at risk for knee OA. The risk is governed by age, sex, and obesity (defined by body mass index [BMI] ≥30 kg/m2).
In addition to capturing the risk of knee OA, the model tracks chronic comorbidities that increase mortality either directly (coronary heart disease, cancers, and obesity) and/or by increasing the risk of other chronic diseases (diabetes mellitus and obesity). Each simulated person is followed by the model until death. Death can occur from any health state. All-cause mortality rates were obtained from Centers for Disease Control and Prevention life tables (19, 22–24). Rates specifically attributable to chronic comorbidities (coronary heart disease, cancer, and obesity) were then removed from the mortality of the life tables to estimate the mortality rates for healthy individuals for all age and sex strata. The mortality rates attributable to specific comorbidities were then separately applied to persons with the corresponding comorbidities. Further details of the OAPol Model structure have been published elsewhere (20, 21), and can also be found in Technical Appendix, section 1 (available in the online version of this article at http://onlinelibrary.wiley.com/doi/10.1002/acr.21898/ abstract).
We used self-reported data from the 2007–2008 NHIS to estimate the prevalence of diagnosed symptomatic knee OA (19). The NHIS is a cross-sectional survey that is representative of the civilian noninstitutionalized population residing in the US. The adult sample, which was the basis of our analysis, was comprised of persons ages ≥18 years. For the purposes of our analysis, we included persons that were ages ≥25 years.
Survey participants were considered to have prevalent knee OA if they: 1) answered “yes” to the questions “Have you ever been told by a doctor or other health professional that you have some form of arthritis, rheumatoid arthritis, gout, lupus, or fibromyalgia?” and “During the past 30 days, have you had any symptoms of pain, aching, or stiffness in or around a joint?”; 2) specified the knee as an affected joint; and 3) did not report having rheumatoid arthritis, lupus, fibromyalgia, or gout. We then used a logistic regression model that accounted for the complex survey design to derive age-, sex-, and obesity-stratified population-based prevalence estimates of the prevalence of self-reported diagnosed symptomatic knee OA. The regression model, built using SAS, version 9.2, included age, age2, age3, and age4 and the interaction between obesity and sex.
Similar algorithms for classifying survey respondents as having symptomatic knee OA have shown specificity greater than 90% (25, 26), which indicates that these algorithms perform well in terms of classifying patients without symptomatic knee OA. Since self-reported data may likely overstate the true prevalence of diagnosed symptomatic knee OA, we used published data on the positive predictive value (PPV) of self-reported diagnosed symptomatic knee OA to reduce the prevalence estimates and the corresponding 95% confidence intervals (95% CIs). Specifically, we assumed that the PPV was lower for those ages <60 years (PPV 66.0%) (27) compared with those ages ≥60 years (PPV 80.7%) (26). For instance, for a nonobese man age 44 years with self-reported prevalence of symptomatic knee OA at 0.0374, the PPV-adjusted prevalence was estimated at 0.0247, or 66.0% of the self-reported prevalence. On the other hand, for an obese woman age 64 years, the predicted prevalence would be 80.7% of the self-reported prevalence of 0.2864, or 0.2311. These reduced prevalence data and 95% CIs were used as benchmarks for the calibration of OA incidence.
We estimated the incidence rates of diagnosed symptomatic knee OA as a ratio of the 2007–2008 NHIS, PPV-adjusted prevalence estimates to disease duration (18). Because there is no permanent cure for knee OA, to derive disease duration we used the OAPol Model to simulate life expectancy within each decade of age, further stratified by sex and obesity (for details, see Technical Appendix, section 3, available in the online version of this article at http://onlinelibrary.wiley.com/doi/10.1002/acr.21898/ abstract). Incidence rates within 10-year age groups between 25 and 85 years were then estimated by dividing PPV-adjusted prevalence rates by model-derived disease durations.
Disease duration was calculated as an average 10-year survival of persons affected by OA, diagnosed within each year of the decade, assuming constant rates of diagnosis in every year of a given decade. The analysis was conducted for the cohort with race/sex and obesity distributions of persons affected by knee OA in the US as well as separately for each age/sex/race and obesity cohort with knee OA.
The calculated incidence estimates were calibrated to generate model-based prevalences that fell within the 95% CIs of the prevalence estimates derived from the 2007–2008 NHIS (for details, see Technical Appendix, section 4, available in the online version of this article at http://onlinelibrary.wiley.com/doi/10.1002/acr.21898/abstract). The 95% CIs for incidence estimates were calculated empirically based on model results. Using a normal approximation of the Poisson distribution, empirical incidence rates were calculated as the number of new cases of diagnosed knee OA from model output divided by the time “at risk.” Time at risk was defined as the sum of years alive without knee OA within each decade.
Using the OAPol Model and our calculated incidence rates, we estimated the lifetime risk of diagnosed symptomatic knee OA from age 25 years, defined as the cumulative probability of being diagnosed with symptomatic knee OA over the lifetime for US adults in each of the 5 cohorts described below. The lifetime risk was calculated by dividing the cumulative number of incident cases of diagnosed symptomatic knee OA within each cohort predicted by the OAPol Model by the size of the initial population at risk for the disease reported by US Census data. We used the OAPol Model to estimate the mean and median ages at symptomatic knee OA diagnosis using the derived incidence rates. Mean age was obtained by constructing a distribution by age of all incident cases at diagnosed symptomatic knee OA based on the OAPol Model and calculating a mean of this distribution (for details, see Technical Appendix, section 5, available in the online version of this article at http://onlinelibrary.wiley.com/doi/10.1002/acr.21898/abstract). Median age represented the age at which 50% of those ultimately diagnosed with symptomatic knee OA had been diagnosed (i.e., the 50% mark of the cumulative distribution function of incident cases). Mean age is reported to one-tenth of a year; median age is reported as an integer.
Using the OAPol Model, we simulated 5 cohorts that were followed from age 25 years until death: 1) nonobese women, 2) obese women, 3) nonobese men, 4) obese men, and 5) the general US population. For nonobese cohorts, the mean ± SD BMI at age 25 years was 25.0 ± 0.5 kg/m2. For obese cohorts, the mean ± SD BMI was 34.5 ± 1.5 kg/m2. For the general US population, BMI distributions, stratified by sex and race/ethnicity, were derived from the 2005–2008 National Health and Nutrition Examination Survey (NHANES) and ranged from a mean ± SD BMI of 26.7 ± 6.1 kg/m2 for white, non-Hispanic men ages 25 years to a mean ± SD BMI of 30.2 ± 8.8 kg/m2 for black, non-Hispanic women ages 25 years (28, 29) (Table 1). The distributions of sex and race/ethnicity within each cohort were derived from 2009 US Census data (30). Prevalence rates of comorbid conditions, including cancer, coronary heart disease, and diabetes mellitus, were stratified by age, sex, and race/ethnicity and were derived from the 2005–2008 NHANES (28, 29) (Table 1). Incidence rates for comorbidities were calculated as the ratio of prevalence odds to disease duration. We assumed that chronic comorbid conditions were not cured in the model, so life expectancies of persons with these comorbidities were used as a measure of disease duration. Dividing prevalence odds derived from the 2005– 2008 NHANES data by the estimated disease duration led to estimates of the comorbid incidence rates used in the model, stratified by age, sex, and race/ethnicity (Table 1).
|Male||Female||Prevalence, %†||Incidence, %†||RR of mortality‡||Source|
|Initial age, years||25.0||25.0|
|Race/ethnicity (all cohorts), %||2009 US Census population estimates (30)|
|BMI, mean ± SD kg/m2|
|Nonobese (maximum 29.9)||25.0 ± 0.5||25.0 ± 0.5|
|Obese (minimum 30.0)||34.5 ± 1.5||34.5 ± 1.5|
|Overall US population||NHANES 2005–2008 (28, 29)|
|White, non-Hispanic||26.7 ± 6.1||26.8 ± 7.7|
|Hispanic||27.6 ± 6.0||28.0 ± 6.7|
|Black, non-Hispanic||27.7 ± 6.2||30.2 ± 8.8|
|Comorbid condition||NHANES 2005–2008 (28, 29)|
|Coronary heart disease||0.0–0.3||0.0–4.4||1.0–36.6|
Age-, sex-, and obesity-stratified estimates of the prevalence of diagnosed symptomatic knee OA are shown in Table 2. For nonobese men, the estimated prevalence ranged from 0.74% among those ages 25–34 years to 12.94% in those ages ≥85 years. For nonobese women, the estimated prevalence ranged from 0.88% for the youngest age group (25–34 years) to 14.97% among those ages ≥85 years. For obese men, the estimated prevalence ranged from 1.54% in the youngest group to 23.54% in the oldest group. Obese women had the highest estimated prevalence, ranging from 2.41% in the youngest group to 32.45% in the oldest group.
|Estimated prevalence (using 2007–2008 NHIS data), % (95% CI)||Estimated incidence (annual), % (95% CI)|
|Age 25–34 years||0.74 (0.61–0.89)||0.12 (0.12–0.12)|
|Age 35–44 years||1.74 (1.51–2.00)||0.13 (0.12–0.13)|
|Age 45–54 years||3.61 (3.21–4.06)||0.22 (0.22–0.22)|
|Age 55–64 years||6.70 (6.09–7.37)||0.37 (0.37–0.38)|
|Age 65–74 years||9.83 (9.01–10.71)||0.20 (0.19–0.20)|
|Age 75–84 years||11.64 (10.51–12.88)||0.13 (0.12–0.13)|
|Age ≥85 years||12.94 (10.86–15.34)||0.04 (0.04–0.04)|
|Age 25–34 years||0.88 (0.73–1.05)||0.14 (0.14–0.14)|
|Age 35–44 years||2.06 (1.83–2.31)||0.15 (0.14–0.15)|
|Age 45–54 years||4.26 (3.90–4.64)||0.27 (0.27–0.27)|
|Age 55–64 years||7.85 (7.22–8.52)||0.43 (0.43–0.43)|
|Age 65–74 years||11.44 (10.48–12.47)||0.27 (0.27–0.27)|
|Age 75–84 years||13.50 (12.44–14.65)||0.16 (0.16–0.16)|
|Age ≥85 years||14.97 (13.04–17.12)||0.06 (0.06–0.06)|
|Age 25–34 years||1.54 (1.26–1.87)||0.25 (0.24–0.25)|
|Age 35–44 years||3.58 (3.12–4.11)||0.24 (0.24–0.24)|
|Age 45–54 years||7.25 (6.49–8.09)||0.44 (0.43–0.44)|
|Age 55–64 years||13.00 (11.76–14.33)||0.64 (0.64–0.65)|
|Age 65–74 years||18.48 (16.71–20.39)||0.32 (0.32–0.33)|
|Age 75–84 years||21.48 (19.33–23.78)||0.17 (0.17–0.18)|
|Age ≥85 years||23.54 (20.24–27.19)||0.05 (0.05–0.05)|
|Age 25–34 years||2.41 (2.02–2.88)||0.37 (0.37–0.38)|
|Age 35–44 years||5.53 (4.93–6.21)||0.40 (0.39–0.40)|
|Age 45–54 years||10.93 (10.01–11.92)||0.57 (0.57–0.58)|
|Age 55–64 years||18.94 (17.52–20.44)||1.02 (1.01–1.02)|
|Age 65–74 years||26.20 (24.25–28.23)||0.41 (0.40–0.41)|
|Age 75–84 years||29.94 (27.73–32.23)||0.28 (0.27–0.28)|
|Age ≥85 years||32.45 (28.79–36.31)||0.10 (0.10–0.10)|
For nonobese men, the estimated disease duration within each decade ranged from 5.48 years for the youngest age group to 3.39 years for the age group ≥85 years. For obese men, the estimated disease duration ranged from 5.47 years to 3.39 years for the 25–34-year-olds and ≥85-year-olds, respectively. For nonobese and obese women, the estimated disease duration ranged from 5.49 years among those ages 25–34 years to 3.89 years among those ages ≥85 years.
Estimated annual incidences of diagnosed symptomatic knee OA, stratified by age, sex, and obesity, are shown in Table 2. Estimated incidence ranged from 0.04% per year in nonobese men ages ≥85 years to 1.02% per year in obese women ages 55–64 years. For all sex and obesity status combinations, the incidence peaked at ages 55–64 years and was lowest among those ages ≥85 years. For nonobese men, the incidence ranged from 0.04% to 0.37% per year. For nonobese women, the incidence ranged from 0.06% to 0.43% per year. Among obese men, the incidence ranged from 0.05% to 0.64% per year. For obese women, the incidence ranged from 0.10% to 1.02% per year.
Results of the internal validation analysis using our current incidence estimates are shown in Figure 1. Within each 10-year age group, the model-based prevalence estimates fell within the 95% CI of the NHIS data for each cohort, defined by sex and obesity status.
Using current demographic, obesity, and comorbidity profiles representative of the general population in the US, the estimated mean ± SD age at symptomatic knee OA diagnosis was 53.5 ± 14.4 years. The estimated median age at diagnosis of symptomatic knee OA was 55 years (Figure 2).
The lifetime risk of diagnosed symptomatic knee OA from age 25 years in the US population was estimated at 13.83%, with a 9.29% risk of having diagnosed symptomatic knee OA by age 60 years (Figure 3). In obese persons, the lifetime risk of diagnosed knee OA was estimated at 19.67% compared to 10.85% for nonobese persons. In women, the lifetime risk was estimated at 16.34% compared to 11.42% in men. In sex- and obesity-stratified cohorts, we estimated the lifetime risk to be highest in obese women at 23.87% and lowest in nonobese men at 9.60%.
We used the OAPol Model, a state transition Monte Carlo computer simulation model, combined with national data from the 2007–2008 NHIS on the prevalence of diagnosed symptomatic knee OA, to estimate the annual incidence of diagnosed symptomatic knee OA in the US. We estimated that incidence peaked during ages 55–64 years, was higher among obese persons than nonobese persons and, adjusting for obesity, was higher among women than men. The estimated median age at knee OA diagnosis was 55 years. Our incidence estimates were derived with an approach that relates incidence, prevalence odds, and disease duration (prevalence odds/disease duration = incidence rate). Prevalence was obtained from a national population-based cross-sectional study. To ensure a “stable population,” we considered prevalence to be stable within 10-year age intervals. This method is widely used when longitudinal data are not available (31, 32).
This study adds important insights to research on OA incidence. Until now, a report from the mid-1990s provided the most recently published estimates of the incidence of diagnosed symptomatic knee OA (15–17). Our contemporary estimates are consistent with prior literature in showing a higher incidence of diagnosed knee OA among women (16) and among obese individuals (8, 33, 34).
However, the estimated rates of diagnosed knee OA incidence differ somewhat from prior studies published 10–20 years ago. The study by Oliveria et al, published in 1995 based on data from the Fallon Community Health Plan, used medical records to estimate the annual incidence of diagnosed knee OA. The authors found that the incidence of diagnosed symptomatic knee OA increases with increasing age until age 80 years (16). The key difference in the estimated incidence between our study and the study reported by Oliveria et al in the 1990s lies in distributional shift. In our study, diagnosed OA incidence peaked in an earlier age group (55–64 years), consistent with current trends where use of TKRs occurs earlier in life, with 40% of TKR recipients being ages <65 years (35–38).
Our estimates also provide insight into the differential impact of obesity on diagnosed OA incidence across age and sex strata. A recent study from the Johnston County Osteoarthritis Project in North Carolina found that the lifetime risk of symptomatic knee OA from age 45 years was nearly 45% (39), considerably greater than our estimate of 14%. Several differences in the study design and population may explain this difference. First, the Johnston County study utilized radiographs to define symptomatic knee OA. Unpublished data from the Johnston County Osteoarthritis Project indicate that only approximately two-thirds of those identified as having symptomatic knee OA reported being diagnosed with OA by a health professional. Second, the current study projected lifetime risk from an earlier age (25 years), while the Johnston County study estimated lifetime risk from age 45 years (39). Third, differences between the study populations in Johnston County and the NHIS may also account for some of the differences in estimated lifetime risk. Indeed, when we ran a cohort through the model using the sex, race, and BMI distributions, as well as the incidence, prevalence, and progression of symptomatic knee OA representative of the Johnston County population, the lifetime risk for developing knee OA increased from 14% to 38%, similar to the 45% risk reported in the Johnston County study. The remaining differences may reflect the fact that although the Johnston County Study had good followup, not all study participants returned for the followup assessment. As in many cohorts, those participants who developed OA were more likely to return for followup assessments, increasing the estimates of OA risk.
Our estimates suggest that persons ages 55–64 years today are at highest risk for a new diagnosis of symptomatic knee OA, with the incidence of diagnosed disease tapering off in older ages. If these differences are indicative of secular trends, they may reflect increased likelihood of diagnosis earlier in life rather than earlier onset of biologic disease. An increase in the rate of diagnosed cases could result from increased patient awareness of knee OA, as well as a heightened inclination to diagnose knee OA on the part of physicians. Because rigorous comparisons between our findings and the study by Oliveria and colleagues are limited by methodologic differences and heterogeneity in the study populations, further research is needed to understand secular trends in the incidence and diagnosis of symptomatic knee OA.
Our findings have important implications for disease prevention and health care utilization. The early median age at diagnosis of symptomatic knee OA (55 years) suggests that public health officials should introduce prevention strategies relatively early in the life course. Policymakers should implement prevention strategies aimed at reducing obesity and the risk of knee injury, 2 major risk factors for knee OA. Furthermore, the early age at diagnosis of symptomatic knee OA may yield high levels of lifetime health care utilization and costs. In the last decade, the mean age of persons undergoing TKR has decreased from 69 to 66 years and utilization of TKR has tripled among US adults ages 45–64 years (38). Whether health outcomes improve as a result of early diagnosis of symptomatic knee OA offers a rich area for future research.
The results from this study should be viewed within the context of certain assumptions and limitations in our approach. First, the definition of OA poses challenges (40). The incidence estimates reported here depend on the method used to define diagnosed symptomatic knee OA. To derive population-based national data, we relied on a national, self-reported survey (2007–2008 NHIS). Diagnosed symptomatic knee OA was defined based on responses to several questions asked in the survey. Certain questions pertain more to recent knee pain than to chronic knee pain, which would serve as a better indicator of knee OA prevalence. However, algorithms similar to the one we used have specificity greater than 90% (25, 26), and the sample size and national scope of the NHIS offer advantages over smaller, geographically-limited observational studies. Nevertheless, to account for possible misclassification in the self-reported prevalence of diagnosed symptomatic knee OA from the 2007–2008 NHIS survey data, we applied PPVs from published studies to the younger and older age groups (25–27). We could not identify any published data on negative predictive values in relation to underreporting of diagnosed symptomatic knee OA; therefore, our estimates of lifetime risk of diagnosed knee OA are conservative. Second, we assumed a constant annual incidence of symptomatic knee OA within 10-year age strata. Although a more precise method might have captured a smoother age distribution, the incidence of the disease is unlikely to differ enough across any 10-year age group to have a meaningful effect on our estimates.
Our estimates suggest that diagnosis of symptomatic knee OA occurs early in the life course (median age 55 years), with the incidence peaking between ages 55 and 64 years. Early diagnosis of symptomatic knee OA may yield future increases in health care utilization and medical costs associated with knee OA. Physicians and policymakers can use our findings to direct resources toward preventing risk factors for knee OA. Policymakers and planners can also use our estimates to prepare for the potential future burden on the US health care system resulting from the early age at diagnosis of symptomatic knee OA.
All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Losina had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study conception and design. Losina, Weinstein, Daigle, Katz.
Acquisition of data. Losina, Weinstein, Reichmann, Daigle, Rome, Chen, Jordan.
Analysis and interpretation of data. Losina, Weinstein, Reichmann, Burbine, Solomon, Daigle, Rome, Chen, Hunter, Suter, Katz.
The authors would like to thank Dr. Edward H. Yelin for helpful comments on a previous draft of the manuscript.