Health-related quality of life in relation to symptomatic and radiographic definitions of knee osteoarthritis: data from Osteoarthritis Initiative (OAI) 4-year follow-up study

Data used in the preparation of this article were obtained from the open access Osteoarthritis Initiative (OAI) study (http://​www.​oai.​ucsf.​edu/​), which is a multi-center, longitudinal cohort study on knee OA [19]. The study incorporates a progression, an incidence, and a reference sub-cohort. Subjects in the progression sub-cohort have symptomatic knee OA at baseline, while subjects in the incidence sub-cohort are at increased risk of developing it. Reference sub-cohort subjects have neither symptomatic knee OA nor eligibility risk factors at baseline. The specific eligibility risk factors and ethical issues are described in detail in Osteoarthritis Initiative Study Protocol [19]. Ethical approval for collecting all subject information was provided by the OAI. Informed consent was obtained from all individual participants included in the study. Applied OAI datasets are listed in additional file (see Additional file 1). Data from baseline and follow-up visits at 12, 24, 36 and 48 months were applied in the study.

A mapping algorithm (https://​www.​sheffield.​ac.​uk/) was used to convert 12-item Short Form Health Survey (SF-12) data from the OAI study into SF-6D utility scores [20]. SF-12 is a measure of general health covering eight health domains. The SF-6D estimates a preference-based single index utility measure from SF-12 using general population values and standard gamble valuation technique. SF-6D scores fall on the scale from 0 (death) to 1 (perfect health). The worst SF-6D score, excluding death, is 0.291 (‘floor effect’). We handled SF-6D utilities as decrement in health utility by subtracting 1.000 from the SF-6D utility score, yielding a negative value (referred as disutility).

Kellgren-Lawrence (K-L) radiographic system classifies knee OA into five grades based on the severity of radiographic findings of joint space narrowing, osteophytes, sclerosis, and bone deformity [21]. K-L grade 0 indicates intact joint without any features of OA, and K-L grade 2 is considered the cut-off point of definite OA, while K-L grade 4 indicates severe OA. However, persons with K-L grade ≥ 2 may be asymptomatic and vice versa. Therefore, symptomatic definition of knee OA is considered more relevant. In prevalence studies, symptomatic knee OA is usually defined as the concurrent presence of radiographic findings (usually K-L grade ≥ 2) and frequent knee pain in the same knee [1, 22, 23].

In order to estimate the effect of different definitions on the HRQoL associated with knee OA, seven definitions of knee OA were studied. Two definitions were based on the symptomatic definition of knee OA (K-L grade ≥ 2 and knee pain, aching or stiffness on more than half the days during past month in the same knee): (a) 2-scale: no/yes (OA present either uni- or bilaterally), or (b) 3-scale: no/unilateral/bilateral knee OA. The remaining five definitions were based solely on the radiographic definition of knee OA: (c) 2-scale: no/yes (K-L grade ≥ 2 either uni- or bilaterally), (d) 3-scale: no/unilateral/bilateral K-L grade ≥ 2, (e) 5-scale unilateral definition: the highest K-L grade of both knees, (f) 9-scale bilateral definition: mean K-L grade of both knees, or (g) 15-scale bilateral definition: combination of K-L grades of both knees ((0;0), (1;0), (1;1) … (4;2), (4;3), (4;4)). Missing data regarding K-L grades was not imputed.

We used 11 covariates as adjusting variables. Demographics (age, gender, education, living status), clinical status (injuries, surgical history, body mass index (BMI), comorbid conditions, smoking status), and physical activity were selected as a standard set of adjusting variables according to Rolfson and co-authors [24]. We also included race into adjusting variables because of its association with pain level and health status [25, 26].

Injury status and surgical history were both dichotomized and based on the OAI query on ever (baseline) or since last visit (follow-up) injuring either knee badly enough to limit ability to walk for at least 2 days, and ever having knee surgery or arthroscopy (answers regarding left and right knee pooled). Self-reported race was dichotomized (White or Caucasian/Other than White) because the frequency in original groups ‘Black or African American’, ‘Asian’ and ‘other Non-White’ was low.

Age, living status, BMI, physical activity and comorbid conditions were categorized because the distribution of the covariates was skewed and in order to allow for nonlinear associations. Age was categorized into three groups (45–54, 55–64, and 65–79 years). Education based on the OAI query refers to the highest grade of school completed, and had three categories: primary/none (less than college), secondary (college graduate or some graduate school), and tertiary level (graduate degree).

Living status based on the OAI query was dichotomized as living alone or living with someone else. BMI based on physical examination was categorized into four groups (< 25, 25 to < 30, 30 to < 35, ≥35 kg/m2). Physical activity was based on the Physical Activity Scale for the Elderly (PASE) score, where higher scores indicate greater physical activity [27], and was divided into quintiles (0–90, 91–134, 135–175, 176–237, 238–526). Smoking status was categorized into three groups (never/former/current smoker) based on OAI questions on smoking pipe, cigars or cigarillos and smoking cigarettes. Comorbid conditions based on Charlson comorbidity index score [28] was dichotomized (none/> 0 comorbid conditions).

OAI data covered living status at baseline and year 3 follow-up visit, comorbid conditions at baseline, year 2 and 4 follow-up visits, and smoking status at baseline and year 4 follow-up visit. Missing data regarding living status, comorbid conditions and smoking status were imputed with the information available from the baseline visit or with the previous follow-up visit (i.e., last observation carried forward).

Descriptive statistics were used to summarize characteristics of OAI study participants. Repeated measures generalized estimating equation (GEE) models [29] were used to evaluate the population average [30] disutility in relation to the symptomatic and radiographic definitions of knee OA in the longitudinal data in order to take into account the repeated measurements of the same individual as observations of a same study subject are correlated. GEE modeling was used because it allows the inclusion of all baseline participants in the analyses rather than only those participants who remained in the study and had data available at all follow-up visits. By utilizing a GEE model, we were able to include data for participants who dropped out of the study, or who did not have radiographic knee examination in all visits. All GEE-models were specified using a normal (Gaussian) distribution, an identity link function and an unstructured correlation matrix. The assumption of normal distribution is justified by the large number of observations, and because by that means we avoided transformation of explanatory variable values equal to zero or one, which is required for example in gamma and beta regressions. The inspection of disutility distributions is presented in the additional file (see Additional file 2). The repeated within-subject time-variable was the visit number (0, 1, 2, 3 or 4). All GEE models were adjusted for the 11 covariates mentioned earlier. The dependent variable was the SF-6D disutility score and the independent variable was a symptomatic or radiographic definition of knee OA. Quasi Likelihood under Independence Model Criterion was used to study models’ goodness of fit. Estimated marginal means and 95% Wald confidence intervals of disutility scores were compared between different subgroups of knee OA definitions. Participants with no knee OA on the basis of different definitions were regarded as the reference group. Results are presented also as minimally important difference (MID) values using a score difference of ≥0.027 as a cut-point for MID [31]. In all statistical analyses, conducted with IBM SPSS for Windows, version 23.0, the level of statistical significance was considered as p < 0.05.

The flowchart of the cohort definition and number of observations eligible for the adjusted GEE analyses is presented in Fig. 1. In order to be eligible for analyses, participants had to have knee radiograph assessment available for both knees and to have answered the SF-12 questionnaire in full at the same study visit at least once. In addition, participants with partial or total knee replacement (KR) were excluded at baseline. Altogether, 4278 participants were eligible at baseline and had data on all adjusting variables. Participants having a partial or total KR at least for one knee during the follow-up were censored once after the replacement. There were 14,161 observations eligible for analyses according to the radiographic definition of knee OA during the 4-year follow-up. In order to be eligible for analyses according to the symptomatic definition of knee OA, participants had to have answered the knee pain query also (14,074 observations eligible for analyses). The main reason for missing data in the present analyses was the missing data on K-L grades for 42, 43 and 44% of participants at study visits 1, 2 and 3, respectively (see Fig. 1 and Additional file 3). Fifty-three percent of participants eligible for the analyses had three or more observations (see Additional file 3).

The baseline characteristics of participants are presented in Table 1. At baseline, the mean age was 61.1 years (standard deviation (SD) 9.2), and 58% of participants were female. The mean BMI was 28.5 (SD 4.8), and 76% of participants had no comorbid conditions at baseline. Fifty-seven and 78% of participants reported no knee injuries and no knee surgical history, respectively, at baseline. The characteristics of participants during follow-up are presented in additional file (see Additional file 3).

The prevalence of knee OA according to different definitions among participants at baseline and among all observations are presented in Table 2. At baseline, 57% of participants had K-L grade ≥ 2 in at least one knee, and 26% had knee OA according to the symptomatic definition of knee OA. At baseline, when using the definition of the highest K-L grade for knee OA, the most prevalent grades were 2 (30%) and 0 (28%), while the most prevalent mean K-L grades were 0.0 (28%) and 2.0 (17%). The most prevalent K-L grade combinations at baseline were (0;0) and (2;2) accounting for 28 and 13% of participants, respectively. The prevalence of knee OA according to different definitions during follow-up are presented in additional file (see Additional file 3).

At baseline, the mean (SD) unadjusted SF-6D utility score of all study participants was 0.801 (0.120) indicating utility loss (disutility) equal to − 0.199 (see Table 1). The mean adjusted disutility scores in relation to different symptomatic and radiological definitions of knee OA are presented in Figs. 2, 3 and 4 and in more detail in additional file (see Additional file 4). The parameter estimates from GEE analyses are presented in additional file (see Additional file 4). Pairwise comparisons are presented in additional file (see Additional file 5).

The estimated disutility score increased with worsening of symptomatic or radiographic status of knee OA. Symptomatic and radiographic knee OA in at least one knee was associated with an increase of 0.026 and 0.006 (corresponding to 1.0 MID and 0.2 MID), respectively, in disutility scores compared with the reference group (Fig. 2a and b). Bilateral or unilateral symptomatic knee OA was associated with 0.030 (1.1 MID) or 0.024 (0.9 MID) higher disutility scores, respectively, compared with the reference group (Fig. 2c). The estimated increase in disutility was 0.013 points (0.5 MID) with bilateral radiographic knee OA compared with the reference group (Fig. 2d).

The estimated disutility scores increased by 0.027 and 0.040 points (1.0 MID and 1.5 MID) from the best to the worst among the highest and mean of K-L grades (Fig. 3a and b), respectively. Compared to a mean K-L grade value of 0.0, the estimated increase in disutility score was statistically significant with mean values of 2.0, 3.0, 3.5, and 4.0 (Fig. 3b). An additional file shows this in more detail (see Additional file 5).

Increase in estimated disutility scores from the best combination of K-L grades (0;0) to the worst (4;4) was the same (0.040 points, corresponding to 1.5 MID) as with mean K-L grade classification due to arithmetical reasons (Fig. 4). The disutility scores in health states (2;2), (3;3), (4;0), (4;2), (4;3), and (4;4) differed significantly from the reference health state (0;0) (Fig. 4). An additional file shows this in more detail (see Additional file 5).

The difference in disutility scores between the worst and best health state according to different definitions of knee OA was the greatest when using combinations or mean values of K-L grades. The differences in the remaining definitions can be ranked as follows: symptomatic OA (3-scale) > highest K-L grade > symptomatic OA (2-scale) > radiographic OA (3-scale) > radiographic OA (2-scale).