Determinants of self-reported health status in a population-based sample of persons with radiographic knee osteoarthritis

Authors


Abstract

Objective

Knee osteoarthritis (OA) is highly prevalent and disabling. Patients with radiographic knee OA may experience pain and functional impairment, which can diminish their health status. Our objective was to determine factors associated with self-reported health status in a national population-based sample with radiographic knee OA.

Methods

Our sample included all of the Third National Health and Nutrition Examination Survey (NHANES-III) participants who underwent a knee radiograph and were found to have radiographic OA (defined as Kellgren/Lawrence grade 2 or higher). Self-reported health status was determined by asking the participant to rate their overall health as excellent, very good, good, fair, or poor. Self-reported health status was analyzed as an ordinal variable using cumulative logit regression, as a dichotomous variable (excellent/very good/good versus fair/poor) using logistic regression, and as a continuous variable after transformation using linear regression.

Results

A total of 1,021 (42%) of NHANES-III participants with a knee radiograph were included in this analysis. The multivariable analyses were performed on 1,009 (99%) of the eligible participants with complete data. We found that nonwhite race, lower income, more comorbidities, and greater functional limitation were associated with worse self-reported health status in all 3 multivariable analyses.

Conclusion

This study has quantified the role of clinical, radiographic, and socioeconomic factors associated with self-reported health status in a population-based sample of patients with knee OA. Self-reported health status in patients with knee OA was associated with functional status and comorbidity.

INTRODUCTION

Knee osteoarthritis (OA) is painful, disabling, and costly, and diminishes a person's health status dramatically. Over 10 million people in the US have knee OA (1) and OA of the knee is one of the 5 leading causes of disability among the elderly (2).

Health status is a multidimensional domain comprising biologic, physical, and emotional functioning (3, 4). Self-reported health status is a measure of how one perceives and reports one's own well-being. For example, participants may rate their health as excellent, very good, good, fair, or poor. This sort of self-reported information provides an important indicator of a person's health status (4). A variety of methods have been proposed to express self-reported health status (5). One method is to classify the responses into 2 groups (excellent, very good, or good versus fair or poor) (5). Another method is to analyze the responses as an ordinal variable (5). Lastly, a particularly transparent approach is to express self-reported health status on a scale from 0 to 1, where 0 corresponds to the worst possible health and 1 corresponds to perfect health. Such scales are easy to interpret, and the unitless 0 to 1 response scale permits comparisons across medical conditions (3).

Many studies have documented the effect of knee OA, as well as other specific conditions, on the self-reported health status of patients' lives (6, 7). Few studies, however, have attempted to characterize or explain variability in self-reported health status within specific conditions. For example, the Beaver Dam Health Outcomes Study examined the influence of dozens of conditions on self-reported health status, but only controlled for age. The study did not examine the contribution of comorbidities, the severity of disease symptoms, or other patient characteristics on self-reported health status (6). Another study, conducted among subjects enrolled in Pennsylvania's Pharmaceutical Assistance Contract for the Elderly (PACE), examined the effect of OA and rheumatoid arthritis on self-reported health status compared with no arthritis. This study controlled for important factors, including age, sex, race, income, and comorbidities. However, the diagnosis was not specific to knees, and radiographs were not used to determine the presence of OA (7).

Our objective was to determine the factors associated with self-reported health status in a national population-based sample of persons with radiographic knee OA. We chose participants affected by radiographic knee OA so that we could examine the impact of symptomatic OA on a person's self-reported health status. We used data from the Third National Health and Nutrition Examination Survey (NHANES-III), which is the most recent NHANES to perform radiographs of the knee, in order to identify factors that correlate with self-reported health status in individuals in a population-based sample with radiographic knee OA.

MATERIALS AND METHODS

Sample.

The NHANES-III is a national population-based survey that was conducted from 1988–1994 by the National Centers for Health Statistics of the Centers for Disease Control and Prevention. The survey was conducted in 2 phases. Phase I took place from 1988–1991, and phase II from 1991–1994. The NHANES-III survey data were collected during a household interview. All participants were then asked to schedule an appointment at a medical examination center where additional data would be collected. Additional details about patient recruitment and selection for the NHANES-III survey have been previously described (8).

Radiographic assessment.

The knee radiographs were performed using a non–weight-bearing anteroposterior approach according to the NHANES-III protocol (9). Radiographs were performed on both knees of all participants surveyed between 1991 and 1994 who were ≥60 years of age and could transfer themselves to the radiograph table under their own power. Severity of radiographic knee OA was defined by Kellgren/Lawrence (K/L) grade, where 0 = no knee OA, 1 = a questionable osteophyte (doubtful knee OA), 2 = a definite osteophyte but no joint space narrowing (mild knee OA), 3 = moderate narrowing of the joint space (moderate knee OA), and 4 = severe narrowing of the joint space (severe knee OA) (10). K/L grades were computed for each knee by a trained radiologist. For our analysis we used the greater of the right and left K/L grades, and we defined radiographic knee OA as having a K/L grade of ≥2. This definition served as the inclusion criterion for this analysis.

Outcome: self-reported health status.

Self-reported health status was determined by asking the participant to rate their overall health as excellent, very good, good, fair, or poor. Self-reported health status was analyzed using 3 techniques. First, self-reported health status was dichotomized as excellent, very good, or good versus fair or poor. Next, we considered self-reported health as an ordinal variable. Finally, we assigned each possible response a rating. To address the issue of transforming ordinal variables into a rating scale using equal-length intervals, we used the rating scale proposed by Diehr and colleagues (11). Values of 0.95, 0.90, 0.80, 0.30, and 0.15 were assigned to the 5 possible responses, respectively (11). The rating scale was then transformed to an estimate of standard gamble utilities (0 to 1, representing the worst to the best possible health) using the power transformation suggested by Torrance and colleagues to create a continuous self-reported health status score, which is often called a utility score in economic evaluations (12).

Correlates of self-reported health status.

Sociodemographic characteristics.

Sociodemographic factors such as age, sex, race/ethnicity, and income were hypothesized to be associated with self-reported health status. Age was classified into 5 categories: 60–64 years old, 65–69 years old, 70–74 years old, 75–79 years old, and ≥80 years old. Race/ethnicity was classified into 3 categories: Hispanic, non-Hispanic black, and non-Hispanic white. Income was also classified into 4 categories: <$20,000, $20,000–$34,999, ≥$35,000, and missing. The missing category was added so that participants who did not report their income could be included in the multivariable regression models.

Comorbidities.

The NHANES-III included several closed-ended questions about the participant's medical problems at the time of the survey or in the past. A comorbidity index was computed by counting the total number of self-reported medical problems. These medical problems included asthma, chronic bronchitis, emphysema, congestive heart failure, myocardial infarction, stroke, elevated cholesterol, hypertension, diabetes mellitus, cancer (including skin cancer), fractures of the hip, wrist, or spine, gout, lupus, osteoporosis, back pain most days for ≥1 month, goiter, urinary tract infection, pneumonia in the last 12 months, blindness, cataracts, and thyroid disease. We then categorized the total number of comorbidities into 3 levels: 0–1 comorbidities, 2–3 comorbidities, and ≥4 comorbidities. Obesity status was considered as a separate factor and was defined as nonobese (body mass index [BMI] <30 kg/m2), obese (BMI ≥30 kg/m2), and missing (13). The missing category for obesity status was added so that participants with missing BMI values could be included in the multivariable regression model.

Radiographic severity, knee pain, and functional limitation.

Severity of radiographic knee OA was classified using K/L grades. We analyzed K/L grade as a categorical variable (grades 2, 3, and 4). Knee pain was defined as having knee pain on most days for at least 6 weeks. A composite functional limitation index was defined from 2 questions in the household interview: difficulty walking a quarter of a mile, and difficulty stooping, crouching, or kneeling. The participants were assumed to have reported zero limitations if they answered “No” to both questions, 1 limitation if they answered “Yes” to one of the questions, and 2 limitations if they answered “Yes” to both questions. We performed internal validation of the index by examining its association with radiographic severity and the use of a walking aid, such as a cane or walker.

Statistical analysis.

The interview and examination data sets were merged to produce a single data set that contained all of the relevant information. Three analyses were performed, one for each of the 3 techniques of examining the outcome of self-reported health status.

For self-reported health status as a dichotomous variable, multivariable logistic regression was performed on our dichotomous outcome to obtain odds ratios (ORs) of being in fair or poor health.

For self-reported health status as an ordinal variable, multivariable cumulative logit regression was performed on our ordinal outcome to obtain cumulative ORs. Cumulative ORs that were >1 were more indicative of worse self-reported health status.

For self-reported health status as a continuous variable, we used 2-sample independent t-tests and one-way analysis of variance to compare unadjusted mean estimates of our self-reported health status score for our hypothesized correlates. We employed multivariable linear regression to obtain adjusted mean estimates of the self-reported health status score. Scheffe's procedure was used to adjust for multiple comparisons where appropriate (14).

We tested for interactions to identify specific subgroups that reported a worse self-reported health status for all 3 variants of the outcome. The interaction between radiographic severity and knee pain was of particular interest, but we also examined interactions involving functional limitation with comorbidity, comorbidity with obesity, and sex with radiographic severity, comorbidity, and obesity. P values less than 0.05 were considered significant. All statistical analyses were performed using SAS statistical software, version 9.1 (SAS, Cary, NC).

We employed survey regression methods as a sensitivity analysis to take into account the survey design. We did not use the sampling weights in this analysis because the variables used to determine the weights (age, sex, and race/ethnicity) were also predictors of interest (15). The methods used in our data analyses were consistent with a more detailed discussion of the use of sampling weights for the NHANES analyses (15, 16). The results that took into account the possible correlations between individuals within primary sampling units and strata were analogous to the standard multiple linear regression and are not presented here because the results would be redundant.

RESULTS

Study sample.

A total of 2,586 participants completed the NHANES-III household questionnaire and underwent the physical examination. Of the 2,586 participants, 2,412 (93%) had a K/L score evaluated on at least 1 knee. Of these, 1,021 (42%) had radiographic knee OA (K/L grade ≥2), which comprised the sample for this analysis.

The baseline demographic features of the sample are presented in Table 1. The average age was 73 years and the sample primarily consisted of women (61%). A substantial portion of the sample consisted of racial/ethnic minorities (46%), and 55% of the participants reported annual incomes <$20,000. Seventy-six percent had ≥2 comorbid conditions, and 25% were obese. Seventy-one percent had a K/L grade of 2 and 63% reported ≥1 functional limitation.

Table 1. Demographic features of NHANES-III participants with radiographic knee OA*
FactorNo. (%)
  • *

    NHANES-III = Third National Health and Nutrition Examination Survey; OA = osteoarthritis; K/L = Kellgren/Lawrence.

  • Includes asthma, chronic bronchitis, emphysema, congestive heart failure, myocardial infarction, stroke, elevated cholesterol, hypertension, diabetes mellitus, cancer (including skin cancer), fractures of the hip, wrist, or spine, gout, lupus, osteoporosis, back pain most days for ≥1 month, goiter, urinary tract infection, pneumonia in the last 12 months, blindness, cataracts, and thyroid disease.

  • Obesity defined as a body mass index ≥30 kg/m2.

  • §

    Defined as knee pain on most days for the past 6 weeks.

Age, years 
 60–64181 (17.7)
 65–69200 (19.6)
 70–74216 (21.2)
 75–79159 (15.6)
 ≥80265 (26.0)
Sex 
 Women621 (60.8)
 Men400 (39.2)
Race/ethnicity 
 Non-Hispanic white544 (53.9)
 Non-Hispanic black240 (23.8)
 Hispanic225 (22.3)
Income 
 <$20,000563 (55.1)
 $20,000–$34,999211 (20.7)
 ≥$35,000150 (14.7)
 Missing97 (9.5)
Comorbidities 
 0–1246 (24.1)
 2–3426 (41.7)
 ≥4349 (34.2)
Obesity status 
 Nonobese690 (67.6)
 Obese255 (25.0)
 Missing76 (7.4)
K/L grade 
 2724 (70.9)
 3219 (21.5)
 478 (7.6)
Knee pain§ 
 No647 (63.4)
 Yes374 (36.6)
Number of functional limitations 
 0374 (36.6)
 1317 (31.1)
 2330 (32.3)

Internal validation of the functional limitation index.

The functional limitation index was highly correlated with radiographic severity and the self-reported use of a walking aid. Seventeen percent of participants who reported no functional limitations had a K/L grade of 4, whereas 51% of participants who reported 2 functional limitations had a K/L grade of 4. Also, 6% of participants who reported zero functional limitations used a walking aid, compared with 74% who reported 2 functional limitations (Figure 1).

Figure 1.

Distribution of the functional limitation index by Kellgren/Lawrence (K/L) grade and use of a walking aid. Open bars represent 0 functional limitations, shaded bars represent 1 functional limitation, and solid bars represent 2 functional limitations.

Correlates of self-reported health status.

Dichotomous outcome.

Results of the multivariable logistic regression are shown in Table 2. We found that clinical (functional limitation and comorbidity) and socioeconomic (race/ethnicity and income) factors were associated with self- reported health status, but not radiographic factors. Those with 1 functional limitation (OR 1.68, 95% confidence interval [95% CI] 1.13–2.48) or 2 functional limitations (OR 4.42, 95% CI 2.95–6.64) were more likely to be in fair or poor health than those with zero functional limitations. Those with 2 or 3 comorbidities (OR 2.68, 95% CI 1.76–4.10) or ≥4 comorbidities (OR 5.49, 95% CI 3.48–8.66) were more likely to be in fair or poor health than those reporting zero or 1 comorbidities. Non-Hispanic blacks (OR 2.32, 95% CI 1.56–3.43) and Hispanics (OR 2.94, 95% CI 1.97–4.39) were more likely to be in fair or poor health than non-Hispanic whites. Those making >$35,000 (OR 0.44, 95% CI 0.27–0.72) were less likely to be in fair or poor health than those making <$20,000. We did not find a statistically significant association between age, sex, obesity status, K/L grade, or knee pain status and self-reported health status in this model.

Table 2. Odds ratios (ORs) and cumulative ORs of correlates of self-reported health status from logistic and cumulative logit regression models for NHANES-III participants with radiographic knee OA*
 Correlates of fair/poor health statusCorrelates of worse health status
OR95% CICumulative OR95% CI
  • *

    ORs come from a multivariable logistic regression model where the probability of fair or poor health is being modeled. ORs >1 are associated with being in fair or poor health. Cumulative ORs come from the multivariable cumulative logit regression model. Cumulative ORs >1 are associated with worse self-reported health. NHANES-III = Third National Health and Nutrition Examination Survey; OA = osteoarthritis; 95% CI = 95% confidence interval; K/L = Kellgren/Lawrence.

  • Includes asthma, chronic bronchitis, emphysema, congestive heart failure, myocardial infarction, stroke, elevated cholesterol, hypertension, diabetes mellitus, cancer (including skin cancer), fractures of the hip, wrist, or spine, gout, lupus, osteoporosis, back pain most days for ≥1 month, goiter, urinary tract infection, pneumonia in the last 12 months, blindness, cataracts, and thyroid disease.

  • Obesity defined as a body mass index ≥30 kg/m2.

  • §

    Defined as knee pain on most days for the past 6 weeks.

Age, years    
 60–641.001.00
 65–691.130.70–1.830.980.67–1.43
 70–740.790.48–1.300.800.55–1.17
 75–790.710.41–1.210.680.45–1.03
 ≥800.750.45–1.240.630.42–0.93
Sex    
 Women1.001.00
 Men1.350.99–1.851.321.03–1.68
Race/ethnicity    
 Non-Hispanic white1.001.00
 Non-Hispanic black2.321.56–3.432.151.58–2.93
 Hispanic2.941.97–4.392.701.96–3.71
Income    
 <$20,0001.001.00
 $20,000–$34,9990.690.47–1.020.690.51–0.93
 ≥$35,0000.440.27–0.720.420.30–0.60
 Missing0.730.44–1.220.820.55–1.23
Comorbidities    
 0–11.001.00
 2–32.691.76–4.102.231.65–3.01
 ≥45.493.48–8.664.112.94–5.74
Obesity status    
 Nonobese1.001.00
 Obese1.040.73–1.481.140.86–1.51
 Missing1.250.71–2.201.701.08–2.70
K/L grade    
 21.001.00
 30.890.61–1.290.970.72–1.29
 41.410.80–2.471.330.83–2.11
Knee pain§    
 No1.001.00
 Yes0.940.67–1.311.140.88–1.49
Number of functional  limitations    
 01.001.00
 11.681.13–2.481.371.02–1.84
 24.422.95–6.643.242.34–4.47

Ordinal outcome.

Results of the multivariable cumulative logit regression yielded similar results as the multivariable logistic regression and are also presented in Table 2. Functional limitation, comorbidity, race/ethnicity, and income remained associated with self-reported health status. The multivariable cumulative logit regression model also showed that age and sex were associated with self- reported health status. NHANES-III participants who were ≥80 years old (cumulative OR 0.63, 95% CI 0.42–0.93) were less likely to have a lower self-reported health status than those who were between the ages of 60–64 years. Also, men (cumulative OR 1.32, 95% CI 1.03–1.68) were more likely to have a lower self-reported health status than women.

Continuous health status score.

The mean ± SD self-reported health status score for the entire sample was 0.81 ± 0.24. The unadjusted and adjusted mean self- reported health status scores are shown in Table 3. K/L grade and knee pain both had a substantial influence on self-reported health status scores in the unadjusted analysis (P = 0.01 and P < 0.01, respectively). However, these 2 variables were not associated with self-reported health status scores in the multivariable regression (P = 0.21 and P = 0.71, respectively). Participants with increased functional limitation experienced significantly lower self- reported health status scores (P < 0.01 for unadjusted and adjusted analyses) (Table 3). In the adjusted analysis, those reporting zero functional limitations had a mean self- reported health status score of 0.85 (95% CI 0.82–0.89), whereas those reporting 2 functional limitations had a mean self-reported health status score of 0.70 (95% CI 0.67–0.73).

Table 3. Unadjusted and adjusted mean self-reported health scores (0–1 scale where 0 = worst and 1 = best) for NHANES-III participants with radiographic knee OA*
 UnadjustedAdjusted
Mean ± SD95% CIPMean95% CIP
  • *

    Unadjusted mean estimates of the self-reported health status score came from independent sample t-tests when the variable had 2 levels and from one-way analysis of variance when the variable had ≥3 levels. Adjusted mean estimates of the self-reported health status score came from the multivariable linear regression model. The 95% CIs are adjusted for multiple comparisons using Scheffe's procedure for both the unadjusted and adjusted analyses. See Table 2 for definitions.

  • Includes asthma, chronic bronchitis, emphysema, congestive heart failure, myocardial infarction, stroke, elevated cholesterol, hypertension, diabetes mellitus, cancer (including skin cancer), fractures of the hip, wrist, or spine, gout, lupus, osteoporosis, back pain most days for ≥1 month, goiter, urinary tract infection, pneumonia in the last 12 months, blindness, cataracts, and thyroid disease.

  • Obesity defined as a body mass index ≥30 kg/m2.

  • §

    Defined as knee pain on most days for the past 6 weeks.

Age, years  0.66  0.27
 60–640.81 ± 0.250.75–0.87 0.770.73–0.81 
 65–690.80 ± 0.240.74–0.85 0.770.73–0.80 
 70–740.82 ± 0.240.77–0.88 0.800.76–0.84 
 75–790.80 ± 0.250.74–0.87 0.810.77–0.85 
 ≥800.82 ± 0.240.77–0.87 0.800.77–0.84 
Sex  0.55  0.03
 Women0.81 ± 0.250.79–0.83 0.810.78–0.83 
 Men0.82 ± 0.240.79–0.84 0.770.74–0.81 
Race/ethnicity  < 0.01  < 0.01
 Non-Hispanic white0.86 ± 0.220.83–0.89 0.860.82–0.89 
 Non-Hispanic black0.77 ± 0.260.73–0.81 0.770.73–0.80 
 Hispanic0.74 ± 0.260.70–0.79 0.750.71–0.78 
Income  < 0.01  < 0.01
 <$20,0000.77 ± 0.260.74–0.80 0.750.73–0.78 
 $20,000–$34,9990.85 ± 0.220.80–0.90 0.800.76–0.83 
 ≥$35,0000.91 ± 0.170.85–0.97 0.830.79–0.87 
 Missing0.80 ± 0.250.73–0.88 0.780.73–0.83 
Comorbidities  < 0.01  < 0.01
 0–10.90 ± 0.180.86–0.95 0.870.84–0.91 
 2–30.82 ± 0.240.79–0.86 0.790.76–0.82 
 ≥40.73 ± 0.260.70–0.77 0.710.68–0.75 
Obesity status  < 0.01  0.35
 Nonobese0.83 ± 0.230.81–0.86 0.800.78–0.83 
 Obese0.78 ± 0.250.74–0.83 0.800.77–0.84 
 Missing0.72 ± 0.280.65–0.80 0.760.71–0.82 
K/L grade  0.01  0.21
 20.82 ± 0.240.80–0.85 0.800.77–0.82 
 30.81 ± 0.240.76–0.86 0.810.78–0.85 
 40.73 ± 0.270.66–0.81 0.760.71–0.81 
Knee pain§  < 0.01  0.71
 No0.84 ± 0.230.82–0.86 0.790.76–0.82 
 Yes0.76 ± 0.260.73–0.79 0.790.76–0.82 
Number of functional  limitations  < 0.01  < 0.01
 00.89 ± 0.190.86–0.93 0.850.82–0.89 
 10.84 ± 0.220.80–0.87 0.820.78–0.85 
 20.70 ± 0.270.66–0.73 0.700.67–0.73 

Comorbidities greatly influenced self-reported health status scores in this sample. The relationship between comorbidities and the self-reported health status score stratified by functional limitation is shown in Figure 2. Although the interaction between comorbidities and functional limitation with respect to the relationship with self-reported health status scores was not statistically significant, there was a trend in decrement in self-reported health status due to worsening functional status and increasing comorbidity (Figure 2).

Figure 2.

Adjusted mean estimates of self-reported health status scores stratified by comorbidities and functional limitation. The error bars represent 95% confidence intervals for the adjusted mean, and the width is adjusted using Scheffe's procedure.

Race/ethnicity was among the factors influencing self-reported health status scores. Results of the multivariable analysis revealed that non-Hispanic whites had a significantly higher mean self-reported health status score (mean 0.86; 95% CI 0.82–0.89) than both non-Hispanic blacks (mean 0.77; 95% CI 0.73–0.80) and Hispanics (mean 0.75; 95% CI 0.71–0.78).

Income was also associated with self-reported health status in both the unadjusted and the adjusted analyses (P < 0.01 for both analyses). The multivariable analysis showed an association between sex and self-reported health status (P = 0.03), but not between age and self-reported health status (P = 0.27). Obesity was associated with lower self-reported health status in the unadjusted analysis (P < 0.01), but not in the adjusted analysis (P = 0.35).

Interactions.

We did not find evidence that K/L grade modifies the relationship between functional limitation and self-reported health status; the interaction between K/L grade and knee pain was not statistically significant in any of the multivariable analyses. In fact, none of the interactions that we tested were statistically significant in any of the multivariable regression models.

DISCUSSION

We conducted an analysis to determine correlates of self-reported health status in individuals with radiographic knee OA among NHANES-III participants. Although we found clear evidence that a greater number of comorbidities and a higher degree of functional limitation were associated with worse self-reported health status, neither radiographic severity of OA nor knee pain influenced self-reported health status in analyses that adjusted for functional status. Furthermore, we did not find evidence of modification of the effect of knee pain on self-reported health status by K/L grade. That is, although radiographs provide the most widely acknowledged indicator of OA severity, this parameter did not appear to influence self-reported health status in our sample. Although the fact that we found an association between functional limitation and self-reported health status was not surprising, the magnitude of the decrement in self-reported health status for those with 2 functional limitations was quite striking. Additionally, we found that demographic and socioeconomic factors including nonwhite race and lower income were also associated with lower self-reported health status in all 3 multivariable regression models.

Obese participants had, on average, lower self-reported health status scores than nonobese participants in the unadjusted analysis (0.83 versus 0.78; P < 0.01). This relationship was not seen when the other factors in the model were adjusted for (P = 0.35). This is most likely due to the fact that obese participants also experience some of the comorbidities that we accounted for in our summative comorbidity index, and the fact that obesity is correlated with other factors in the model (e.g., functional status), leaving no additional variance in the self-reported health status score to be explained by obesity after adjusting for the variables that were included in the multiple regression model. Finally, we examined interactions involving functional limitation, knee pain, radiographic severity, obesity, comorbidity, and sex, but we were unable to find any that further explained the variability in self-reported health status.

To the best of our knowledge, there are no prior studies evaluating correlates of self-reported health status in a population-based sample of individuals with evidence of radiographic knee OA. The Beaver Dam Health Outcomes Study derived age-adjusted mean estimates of a true health utility using the time trade-off technique in patients with arthritis, but not specifically radiographic knee OA (6). In their age-adjusted analysis, which adjusted for the mean age of 64.1 years in their sample, they found that participants with arthritis had a health utility of 0.82 (95% CI 0.80–0.83) (6). This result was consistent with the overall mean self-reported health status score (0.81) that we described in this report, even though the mean age in our sample was 73 years. This is due to the fact that age was not associated with self-reported health status in 2 of our analyses. The association seen between age and our ordinal self-reported health status outcome is likely due to a healthy survivor effect because patients age ≥80 years were more likely to report being in better health than those who were between the ages of 60–65 years. The advantage of our study is that it provides more detailed information about factors that are associated with self-reported health status in patients with knee OA, which allows further discrimination of self-reported health status among persons with knee OA.

Our results are consistent with other studies that examined the impact of race and income on self-reported health status (7, 17–19). The study using subjects enrolled in Pennsylvania's PACE demonstrated findings similar to what we found with regard to race, income, and comorbidities, but not age. The authors analyzed the general health status question as a dichotomous variable (excellent, very good, or good versus fair or poor) as well (7). Subjects age ≥85 years were more likely to respond as having fair or poor health than those between the ages of 65–74 years, but they were less likely to respond as having fair or poor health than those between the ages of 75–84 years (7). The authors also found that nonwhite race, lower income, and more comorbidities were associated with worse self-reported health status (7), which is concordant with our results. However, our sample is a national sample, whereas PACE is restricted to residents in Pennsylvania. Also, our additional use of a continuous measure allows for further discrimination and estimates that can be used in economic evaluations.

There are several limitations to our study, including that the estimates provided in this manuscript are not population-based estimates. One is that NHANES-III used non–weight-bearing knee radiographs. This may lead to underestimation of the presence and severity of radiographic knee OA (20), and could potentially reduce the influence of radiographic severity on self-reported health status. Also, patellofemoral OA is not assessed formally in the K/L grading system. Since patellofemoral OA may be quite painful, this may adversely influence self-reported health status. We acknowledge that the cumulative logit regression model may not meet the proportional odds assumption in this case. However, because the test for proportionality of odds can be overly sensitive in the presence of large samples, we used it as a secondary analysis that confirmed many of the same relationships as ordinary logistic regression.

Another limitation of this study is that we were not able to determine causality or observe how self-reported health status changes over time due to the cross-sectional design of the study. As with any observational study, we acknowledge the possibility of not accounting for other confounders, which may explain additional variability in self-reported health status. This includes OA in other joints, which were not formally ascertained in NHANES-III. It would be important to study the effect of changes in disease severity, symptoms, and comorbidities on changes in self-reported health status over time. We recognize that estimating the effect of missing obesity status and income on self-reported health status may lead to biased estimates with respect to these 2 variables if the data is not missing completely at random. We acknowledge that our continuous outcome (the transformed rating scale) is an estimate and not a directly elicited standard gamble utility (12). Further research should be done in samples in which utilities would be elicited using direct assessment. Lastly, we recognize that the various comorbidities included in our summative index may not be equal contributions to one's self-reported health status. However, creating a single comorbidity index with equal weights seemed to provide adequate adjustment in our model while accounting for the fact that each comorbidity was relatively rare. This simplification, which has been done in other studies looking at the impact of comorbidity on self-reported health status (4), would result in a conservative bias, making it more difficult to detect the influence of comorbidities on self-reported health status.

We found that more comorbidities, a higher degree of functional limitation, nonwhite race, and lower income were associated with a worse self-reported health status in this sample of participants with radiographic knee OA. These results point to subpopulations of patients with knee OA at risk for experiencing worse health. From a clinical point of view, health-enhancing interventions could be aimed especially at these subpopulations. From a research standpoint, these patient characteristics should be noted as potential confounders in studies of self- reported health status in knee OA. Further research needs to be done to evaluate how self-reported health status changes in patients as their disease severity and symptoms progress or recede.

AUTHOR CONTRIBUTIONS

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 submitted for publication. Mr. Reichmann 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. Reichmann, Katz, Kessler, Losina.

Acquisition of data. Reichmann, Losina.

Analysis and interpretation of data. Reichmann, Katz, Kessler, Jordan, Losina.

Ancillary