Subject selection
This study utilized data from the Osteoarthritis Initiative (OAI; https://nda.nih.gov/oai) [15], a multi-center, longitudinal study of individuals aged 45–79 years at enrollment. The OAI dataset includes MRI and radiographic knee images of participants over 8 years. The study protocol, amendments, and informed consent documentation were reviewed and approved by the local institutional review boards of all participating centers (University of Maryland School of Medicine, Ohio State University, University of Pittsburgh, Memorial Hospital of Rhode Island). In addition, all methods were performed in accordance with the relevant guidelines and regulations the Human Research Protection Program (HRPP) at UC San Francisco.
The present study analyzed participants enrolled in the OAI with the following inclusion criteria: (i) available data on the Center for Epidemiological Studies Depression Scale at the baseline and 4-year follow-up visit, (ii) a baseline Kellgren Lawrence score (KL) ≤ 3 in the right or left knee, (iii) available body mass index (BMI) data at baseline and (iv) either normal BMI (16.9–24.9 kg/m2) or obese BMI (30–49 kg/m2) at baseline. The overweight group was excluded to better investigate the effects of obesity in comparison to a normal BMI control cohort (16.9–24.9 kg/m2). Participants were excluded if their depression symptoms no longer met the threshold between baseline and 4-year follow-up, or participants became depressed (detailed description below) between baseline and 4-year follow-up. Participants with rheumatoid arthritis were also excluded. Based on these criteria, a total of 1844 participants (mean BMI: 28.8 ± 5.90 kg/m2) were included in this study (Fig. 1) and were categorized into 4 groups: no sustained depressive symptoms (defined below) and normal BMI (16.9–24.9 kg/m2), n = 772; no sustained depressive symptoms and obese BMI (30–49 kg/m2), n = 971; sustained depressive symptoms and normal BMI (16.9–24.9 kg/m2), n = 33; and sustained depressive symptoms and obese BMI (30–49 kg/m2), n = 68.
Depressive symptoms
Depressive symptoms were assessed using the Center for Epidemiological Studies Depression Scale (CES-D) [17] (threshold ≥16) at the baseline and 4-year follow-up visit based on previous studies [13]. The CES-D is a 20-item questionnaire that asks individuals how often they experience symptoms associated with depression. This questionnaire has good sensitivity and specificity as well as a high internal consistency [18]. A threshold of ≥16 is often recommended as a cutoff when for screening for “clinical depression” [13] based on published studies [13, 16]. Participants with sustained high level of depressive symptoms were defined as those a CES-D score of ≥16 at baseline and 4-year follow-up, while participants without sustained depressive symptoms had a CES-D score of < 16 at baseline and 4-year follow-up. Participants with depressive symptoms that were not sustained between baseline (CES-D ≥ 16) and 4-year follow-up (CES-D < 16) or became depressed between baseline (CES-D < 16) and 4-year follow-up (CES-D ≥ 16) were excluded to focus the analysis on participants with or without depressive symptoms at both timepoints.
Additional clinical questionnaires
Knee pain was assessed using the WOMAC (Western Ontario McMaster Universities Osteoarthritis) Index, a standard questionnaire used to evaluate symptoms related to knee OA, including pain, limited function and stiffness [19]. This questionnaire has three subscales (pain (range: 0 to 20), stiffness (range: 0 to 8), and physical function (range: 0 to 68)) and has been utilized in a number of previous OA studies [20, 21]. The current study focuses on the WOMAC pain subscore; higher scores indicate worse pain.
The participants’ physical activity levels were determined using a Physical Activity Scale for the Elderly (PASE) with a range of 0 to 400. This is a well-established, reliable, validated questionnaire that has been used to measure physical activity in individuals of similar age to those investigated in the current study [22,23,24,25]. The areas of assessment are activities of occupation, household, and leisure activities over a 1 week period.
Radiographs
Standardized bilateral standing posterior-anterior fixed flexion knee radiographs were acquired in all participants in the OAI. For eligibility and to assess baseline disease burden, knee Kellgren Lawrence (KL) gradings [26] were performed at baseline with a score ranging from 0 (none) to 4 (severe). A KL grade of 0 represents definite absence of radiographic changes of OA; grade 1 represents: doubtful joint space narrowing (JSN) and possible osteophytic lipping; grade 2 represents definite osteophytes and possible JSN; grade 3 represents moderate multiple osteophytes, definite JSN and some sclerosis and possible deformity of bone ends; grade 4 represents: large osteophytes, marked JSN, severe sclerosis and definite deformity of bone ends. In addition, JSN (maximum score of the medial and lateral joint sides in each knee) was assessed longitudinally from baseline to 2- and 4-year follow-up [27] based on the OARSI grading scale.
MR imaging acquisition and analyzed parameters
MR imaging acquisition
MR imaging was performed using 3 T MRI scanners (Trio, Siemens, Erlangen, Germany) at four centers (Ohio State University in Columbus, University of Maryland in Baltimore, University of Pittsburgh and Brown University in Rhode Island) as part of the imaging OAI protocol. The following sequence of the right knee were analyzed in this study: sagittal 2D multi-echo (ME) spin-echo (SE) sequences for T2 quantification. The imaging parameters for the MESE T2 mapping sequence were: TR = 2700 ms, 7 TEs = 10, 20, 30, 40, 50, 60 and 70 ms, in-plane spatial resolution of 0.313 mm × 0.446 mm (0.313 mm × 0.313 mm after reconstruction), slice thickness of 3.0 mm, and 0.5 mm gap. These scanning parameters were optimized based on the OAI MR imaging protocol; additional details on image acquisition parameters have been previously published [15].
Cartilage T2
MRI cartilage T2 measurements quantify the composition of the cartilage extracellular matrix, which includes collagen integrity and orientation, as well as water content. Cartilage T2 measurements of the right knees were quantified at baseline, 2, and 4 years in six regions (medial and lateral tibia, medial and lateral femur, trochlea, and patella). A deep learning-based algorithm with 2D U-Net convolutional neural networks, with high efficacy and precision, was utilized for automatic cartilage segmentation and T2 quantification as previously described [28, 29]. Briefly, the dataset was randomly split to training, validation, and test sets (65:25:10) and 3D V-Net architecture was used for segmentation. Although the OAI dataset provided images with 7 echoes (TE = 10, 20, 30, 40, 50, 60, 70 ms) for T2 quantification, the first echo (TE = 10 ms) was not included in the T2 fitting procedure in order to reduce potential errors resulting from stimulated echoes, and a noise-corrected algorithm was implemented [30, 31]. Average T2 values for each region were computed and analyzed in this study.
Statistical analysis
Descriptive statistics were performed using a SAS Studio (version 3.8, SAS Institute Inc., Cary, NC, USA) macro program called “Tablen” [32]. Differences in continuous parameters between groups (i.e., age, BMI) were assessed using Kruskal Wallis tests, and differences in categorical parameters between groups (i.e., sex and race) were assessed using Chi-squared tests.
The primary statistical analyses were performed using STATA version 16 software (StataCorp LP, College Station, TX, USA) with significance set to p < 0.05. Two types of mixed effects models were performed (described below).
The first set of mixed models were interaction analyses to assess whether having both sustained depressive symptoms and obese BMI had a greater effect on knee outcomes (JSN, cartilage T2, and knee pain) over and above the additive effects of each predictor. The mixed models included a test for statistical interaction between BMI (normal/obese) and sustained depressive symptoms over 4-years (yes/no). All outcomes were treated as continuous variables. First, a model with a triple interaction was coded (interaction between BMI (normal/obese), depression (yes/no) and by time (years), in order to capture BMI-depressive symptoms interactions in the change in the outcome over time. If this interaction was not significant, the model was further simplified by including three double interactions (depression-BMI, depression-time, BMI-time). The interactions reported in this study are between BMI (normal/obese) and sustained depression (yes/no) as none of the interactions for longitudinal change were statistically significant (p > 0.05). JSN and cartilage T2 outcomes were analyzed at baseline, 2 and 4 years, while knee pain outcomes were analyzed annually over 4 years. A random effect for both person and knee were modelled for all outcomes except cartilage T2. A random effect for only person was modelled for cartilage T2 outcomes since cartilage T2 measurements were only obtained in the right knee in the OAI, and thus accounting for two knees was not needed.
The second set of mixed models (that did not include an estimate for and test for an interaction) were group-based analyses that investigated the overall differences in outcomes (JSN, cartilage T2, and knee pain) over all timepoints between participants subdivided into four groups based on baseline BMI (normal/obese) and sustained depression over 4 years (yes/no). The four groups were: no sustained depression and normal BMI (16.9–24.9 kg/m2), no sustained depression and obese BMI (30–49 kg/m2), sustained depression and normal BMI (16.9–24.9 kg/m2), sustained depression and obese BMI (30–49 kg/m2). The coefficients (which represent the difference in outcomes between each group and the reference group averaged over all timepoints) and p-values were derived from these model outputs. These analyses are described as “group-based” in the results section.
All mixed effects models were adjusted for age, sex, race, and PASE score. All assumptions for linear mixed models including a normal distribution and independent errors were met.
The outcome variables were designated as primary or exploratory to address potential issues stemming from multiple testing [33]. For cartilage T2, the primary analyses focused on the average of all regions (medial and lateral tibia, medial and lateral femur, trochlea, and patella). For JSN, the maximum score of the medial and lateral joint sides in each knee was assessed. For the WOMAC score, only the pain subscale was assessed. The remaining outcomes were designated as exploratory.
As a sensitivity analysis, an interaction between BMI-depression-sex was added to each model to assess whether the effects of BMI and depressive symptoms on outcomes differed by sex. Another sensitivity analysis was performed in participants with KL 0 or 1 in both knees to assess participants without radiographic evidence of OA in either knee. The first sensitivity analysis was included to assess whether the results of the main analyses differed by sex; the second sensitivity analysis was included to assess whether the results held true in participants without radiographic OA.