Depression and Time to Knee Arthroplasty among Adults who have Knee Osteoarthritis

Abstract

Background:

Depression is common in osteoarthritis (OA) and is associated with poor outcomes following total knee arthroplasty (TKA). Depression can increase pain sensitivity and may be related to an increased likelihood of TKA.

Methods:

Nationally distributed electronic health record data from 2010 to 2018 were used to identify eligible patients (n = 9,466) who had knee OA and were 45 to 80 years of age. Cox proportional hazard models were computed to estimate the association between depression and incident TKA for all patients and by age group (45 to 54, 55 to 64, and 65 to 80 years of age). Confounding was controlled using entropy balancing. Sensitivity analyses determined if the association between depression and TKA differed when depression occurred in the 12 months occurring 90, 60, 30, and 0 days lag time before TKA.

Results:

The mean age of the sample was 63 (range, 45 to 80), 64.0% were women, 83.3% were white race, and approximately 50% resided in the Midwest. There was no association between depression and incident TKA (HR [hazard ratio] = 0.97 [Confidence Interval [CI]=0.81 to 1.16]). Results did not differ in age stratified analyses. Sensitivity analyses revealed a higher percentage of TKA among depressed versus non-depressed patients (24.2 versus 21.6%, P = 0.028) when the patient’s depression diagnosis was established in the 12 months with no lag time before TKA.

Conclusion:

Patients who have knee OA and comorbid depression, compared to those who have only knee OA, do not have an increased likelihood of TKA. The multifactorial, complex decision to obtain TKA does not appear to be influenced by depression, but depression is a common comorbidity.

INTRODUCTION

Osteoarthritis (OA) is a leading cause of chronic pain and disability in the United States (US) and worldwide[1, 2]. Knee OA prevalence is on the rise in the US, with recent trends indicating knee OA is occurring in younger patients[3], resulting in billions of dollars in healthcare costs and time loss due to disability[2, 4].

The patients’ experience with knee OA symptoms is extremely complicated. For instance, radiographic findings on a knee x-ray do not correlate well with experienced knee OA symptoms, with some patients having severe findings on x-ray experiencing minimal pain and others having mild x-ray findings experiencing severe pain[5, 6]. While there are many pharmacologic treatments for knee OA, the definitive management of end-stage disease is total knee arthroplasty (TKA). The decision to pursue TKA is multifactorial, with the most important drivers being decreased function, duration of symptoms, and increased severity of pain[7]. While patient experience of pain is also multifactorial, mental health conditions are frequently comorbid with pain and are associated with worse outcomes.

Depression has long been associated with worse outcomes in pain, such as an increased likelihood of transitioning to chronic pain, increased pain sensitivity, and increased pain catastrophizing[8–11]. Prior studies have observed that depression is associated with a poor response to conservative pain treatment[12–14]. In the case of knee OA pain, depression is associated with worsening knee pain over time[15, 16]. Depression is also a common comorbidity in the knee OA population with roughly twenty percent of patients having depression, and a similar prevalence has been observed in the month prior and up to 3 months after total joint arthroplasty[17, 18]. While depression is known to play a role in post-operative outcomes following TKA, such as increased likelihood of post-operative complications or hospital readmission, much less is known about how depression may impact incident TKA or time to TKA[19]. One analysis investigating the timeliness of TKA found that the presence of depression may be associated with delayed use of TKA[20]. In a study of patients scheduled for TKA, patients who had positive Patient Health Questionnaire-8 (PHQ-8) results, a common screening questionnaire for depression, were significantly more likely to have utilized all non-operative therapies[21]. The authors suggested this may reflect increased pain and, thus, treatment seeking with their healthcare providers. Greater treatment seeking and the use of more non-surgical therapies may result in patients being noted to have symptoms refractory to therapy, and thus promote TKA[22]. Conversely, a study conducted by Bendich et al. found no connection between depressive symptoms measured by the Center for Epidemiologic Studies Depression (CES-D) and willingness to undergo TKA[23].

Given the key role TKA plays in the treatment of end-stage OA and the substantial prevalence of comorbid depression in knee OA, there is a need to determine if depression is associated with probability of TKA. Since depression has been implicated in potentially promoting poorer outcomes following TKA, understanding the role that depression plays in decision to undergo arthroplasty may help inform surgeons’ approach to patients who have depression. If depression accelerates patients toward TKA while simultaneously contributing to poorer outcomes following surgery, identifying these patients for intervention prior to surgery could be critical to supporting good post-operative outcomes.

The research question we attempted to answer with our study was: Is baseline depression diagnosis associated with the likelihood of and time to TKA?

MATERIALS AND METHODS

A random sample of 5 million adult (> 17 years of age) patients was leased from Optum’s de-identified Integrated Claims-Clinical dataset (Optum, Inc, Eden Prairie, Minnesota). Patient data from 2010 to 2018 came from inpatient and/or outpatient health care encounters at academic, non-academic, and safety-net health care settings across the United States. These data included patients who had commercial, federal, or no health insurance.

Design:

This study utilized a retrospective, fixed cohort design. A fixed cohort design was chosen to allow for as much follow-up time for all eligible patients as possible, as knee OA is notorious for its slow change in symptoms[24, 25]. This is a common methodology in retrospective cohorts to maximize follow-up time[26–28]. The index date was set as January 1, 2012, which allowed for a maximum of 7 years of follow-up time for all patients at endpoint of December 31, 2018. We designated a baseline lookback period in the 2 years prior to the index date, during which categorization of exposure status (depression) and removal of patients who had prevalent TKA (washout) were established. Since this lookback/washout period occurs prior to the index date, this period does not contribute to the follow-up time. See Figure 1 for a schematic of the design.

Figure 1.

Schematic of study design with time periods of variable measurement

Note: OA=Osteoarthritis, TKA=Total Knee Arthroplasty

Primary outcome:

A TKA was defined as the presence of a single Current Procedural Terminology (CPT) code (27437, 27438, 27440, 27441, 27442, 27443, 27445, 27446, 27447), International Classification of Diseases (ICD)-9 procedure (81.54), and ICD-10 procedure codes (0SRC*, 0SRD*) in follow-up. Months from the index date to TKA or censoring were defined as the last encounter available.

Exposure:

Depression was defined by ICD-9 and ICD-10 codes, and we required two depression diagnoses on separate outpatient encounters or one inpatient diagnosis within the same 12-month period. This diagnostic algorithm has 99% agreement with manual chart abstraction[29]. Diagnosis was measured in the 2-year lookback period described above (2010 and 2011).

Covariates:

Covariates were measured in the 2-year look-back period and included demographics, geographic region, volume of health services utilization, the osteoarthritis severity index (OASI), and physical and psychiatric comorbidities.

Demographic variables included age, sex, and race. Because rates of TKA and depression vary in the US, we controlled for census regions (Midwest, Northeast, South, and West). Patients who use more health care are more likely to receive a diagnosis or treatment; therefore, we controlled for detection bias by computing the distribution of average monthly clinic encounters. The top 25^th percentile was considered high utilization.

The OASI at baseline was computed using methods previously published[30]. It is the sum of the types of therapies, treatments, or diagnostic assessments received in the lookback period (2010 to 2011). Components include diagnostic imaging (X-ray, Magnetic Resonance Imaging or Computed Tomography scan), intra-articular injections, physical therapy, opioid prescriptions, and Non-Steroidal Anti-Inflammatory Drugs (NSAIDs) prescriptions. The OASI ranges from 0 to 6. Information used to define each component of the index can be found in e-Table 1.

Table 1.

Baseline characteristics of knee osteoarthritis patients 45 to 80 years old (n = 9,466)

Covariates, n(%) or mean(±sd)	Overall (n=9,466)
Sociodemographic-related
Age, mean (±sd)	63.0 (±9.5)
Age category
45–54	2096 (22.1)
55–64	3152 (33.3)
65–80	4218 (44.6)
Women	6059 (64.0)
Race
White	7888 (83.3)
Black	1000 (10.6)
Other/unknown	578 (6.1)
Region
Midwest	4971 (52.5)
Northeast	936 (9.9)
South	2424 (25.6)
West	921 (9.7)
Other/unknown	214 (2.3)
High healthcare utilization	2362 (25.0)
Osteoarthritis Severity Index, mean (±sd)	2.0 (±1.4)
Comorbidities and medication
Charlson index, mean (±sd)	1.3 (±1.9)
Obese	5539 (58.5)
Anxietya	647 (6.8)
Nicotine dependence	1071 (11.3)
Any Substance Use Disorder	219 (2.3)
Antidepressantsb	1570 (16.6)

^aAnxiety disorder = panic disorder, social phobia, Generalized Anxiety Disorder, Anxiety Not Otherwise Specified

^bAntidepressants= sustained use in 2-years prior to index (≥2 prescriptions in a 6-month period)

sd=Standardized Deviation

The Charlson Comorbidity Index was used to control for morbidity and mortality, with higher scores indicating worse physical conditions[31–33]. We also controlled for the presence of obesity. Last, we controlled for comorbid anxiety disorder (composite of panic disorder, social phobia, generalized anxiety disorder, anxiety disorder not otherwise specified), nicotine dependence, any substance use disorder (SUD) (drug or alcohol), and sustained use of antidepressant medication (≥ 2 prescriptions in any 6-month period in the baseline period). Detailed variable definitions are shown in Appendix A.

Eligibility criteria

Eligible patients were 45 to 80 years old at index, had a knee OA diagnosis in 2010 or 2011 (lookback period), and had a yearly clinic encounter in 2010 and 2011. In order to ensure patients were free of the outcome at index, we utilized the lookback period as a wash-out for the TKA outcome. Patients were removed if they had a CPT or ICD-9/10 code indicating the presence or history of TKA prior to or within the lookback period. We then removed patients who had ≤ 90 days of follow-up and excluded 4 patients with missing sex data. After applying all eligibility criteria, 9,466 patients who had knee OA were eligible for analysis. See Figure 2 for eligibility.

Figure 2.

Sampling and flow of inclusion Base sample: 5,000,000 active patients 2010–2018

Note: OA=Osteoarthritis, TKA=Total Knee Arthroplasty

Analytic Approach

To control for confounding, baseline characteristics were balanced between those who did and did not have depression at the index date using entropy balancing (e-balance) [34]. E-balance was chosen over other common methods like propensity scores and inverse probability of exposure weighting because this method is not limited by the correct model specification or iterative checking of balance given a particular model. Thus, better balance can be achieved. E-balance weights the non-depressed group by deriving weights so that covariate moments (mean, variance) are almost equal to those that are depressed. E-balance was conducted so that balance can be achieved between depressed versus non-depressed overall and by age group (45 to 54, 55 to 64, and 65 to 80 years). Balance is evaluated using the standardized mean difference percent (SMD*100 = SMD%), and well-balanced covariates have a SMD < 10%[35]. Stata 16 (StataCorp, College Station, Texas) was used to calculate e-balance weights.

All outcome analyses were conducted using SAS v9.4 (SAS Institute, Cary, North Carolina). Summary descriptive information was expressed as means (± standard deviations) or frequencies and percentages. Bivariate comparisons of depression versus no depression and baseline covariates were performed using Chi-square tests and independent samples t-tests. The SMD% before and after e-balance assessed covariate balance. Cox proportional hazard models before and after e-balance were used to calculate hazard ratios and 95% confidence intervals for the association of depression and time to TKA. Separate models in unweighted and weighted data were computed for the entire cohort and stratified by age groups (45 to 54, 55 to 64, and 65 to 80 years). Effect modification by age group was assessed with an age group by depression status interaction term in an overall cohort model. Weighted models used robust, sandwich-type variance estimators to calculate confidence intervals[35]. The proportional hazard assumption was assessed and met in all models (P > 0.05).

Out of concern for the possibility of a long latency between depression diagnosis and TKA, a series of sensitivity analyses were conducted to assess the association of more proximal depression exposures on the probability of TKA. Lag periods assessed were 90, 60, and 30 days, as well as no lag. For example, using a 90-day lag prior to the study end date, depression was measured 3 to 15 months prior to the study end date. Similarly, if using a 60-day lag prior to the cohort study end date, depression was measured in the 2 to 14 months prior to the study end date. For the 30 day-lag, depression was measured in the 1 to 13 months prior to the study end date. For no lag, depression was measured 0 to 12 months prior to the study end date. Crude, bivariate assessments were assessed via Chi-square tests.

Human Subjects Protections: Because all data were de-identified, the Institutional Review Board reviewed and exempted this study as non-human subjects’ research.

RESULTS

Table 1 displays the demographic information of the cohort. The distributions of covariates by depression status at index are shown in Table 2. The youngest age group was more common among those who had depression (SMD% = 19.3) and the oldest age group was more common among those who did not have depression (SMD% = −19.7). Those who had depression were more likely to be women (SMD% = 32.7). Persons living in the Midwest were more often in the depressed group, while those in the Northeast and West were more prevalent among nondepressed patients. High healthcare utilization was more than twice as common among patients who had depression compared to those who did not have depression (47.7 versus 22.3%, SMD% = 55.3). The mean OASI and mean Charlson comorbidity index were greater among depressed patients compared to non-depressed patients (SMD% = 29.3 and SMD% = 45.9%, respectively). Obesity, anxiety disorder, nicotine dependence, any SUD, and sustained antidepressant use were each substantially more prevalent among patients who have depression compared to those who do not have depression (SMD% range, 30.8 to 96.4%).

Table 2.

Baseline characteristics by depression

Covariates, n(%) or mean(±sd)	No DEP (n=8481)	DEP (n=985)	P-value	SMD%
Age category			<.0001
45–54	1804 (21.3)	292 (29.6)		19.3
55–64	2813 (33.2)	339 (34.4)		2.6
65–80	3864 (45.6)	354 (35.9)		−19.7
Women	5298 (62.5)	761 (77.3)	<.0001	32.7
Race			.018
White	7040 (83.0)	848 (86.1)		8.5
Black	905 (10.7)	95 (9.6)		−3.4
Other/unknown	536 (6.3)	42 (4.3)		−3.4
Region			<.0001
Midwest	4409 (52.0)	562 (57.1)		10.2
Northeast	874 (10.3)	62 (6.3)		−14.6
South	2157 (25.4)	267 (27.1)		3.8
West	852 (10.1)	69 (7.0)		−10.9
Other/unknown	189 (2.2)	25 (2.5)		2.0
High healthcare utilization	1892 (22.3)	470 (47.7)	<.0001	55.3
Osteoarthritis Severity Index, mean (±sd)	1.9 (±1.4)	2.4 (±1.5)	<.0001	29.3
Charlson index, mean (±sd)	1.2 (±1.8)	2.2 (±2.5)	<.0001	45.9
Obese	4834 (57.0)	705 (71.6)	<.0001	30.8
Anxiety	368 (4.3)	279 (28.3)	<.0001	68.6
Nicotine dependence	787 (9.3)	284 (28.8)	<.0001	51.4
Any Substance Use Disorder	124 (1.5)	95 (9.6)	<.0001	36.3
Antidepressants	1047 (12.4)	523 (53.1)	<.0001	96.4

SMD% = standardized mean difference percent (SMD*100), DEP=Depression

As shown in Appendix A, e-table 2, entropy balancing successfully removed differences in the distribution of covariates by depression status of all subjects and within each age subgroup (45 to 54, 55 to 64, 65 to 80 years of age). All SMD% were < 1.0 and near 0.

As shown in Table 3, the TKA cumulative incidence rate was similar in patients who did (20.6%) and did not have depression (22.1%), and the TKA incidence rate per 1000 person years (PY) was similar in depressed and non-depressed patients (43.9/1,000 PY and 45.3/1,000 PY, respectively). This pattern was observed for those 45 to 54 years of age and 55 to 64 years of age. Among patients 65 to 80 years of age, the TKA incidence rate was slightly higher among patients who had depression, however, the cumulative incidence rate was nearly identical in depressed and non-depressed patients from this age group.

Table 3.

Cumulative incidence % and incidence rate per 1,000 person-years (PY) for total knee arthroplasty (TKA)

Age group	Total n	TKA events	Cumulative incidence %	Incidence rate per 1,000PY
All ages
Overall	9,466	2,077	21.9%	45.1/1,000PY
No depression	8,481	1,874	22.1%	45.3/1,000PY
Depression	985	203	20.6%	43.9/1,000PY
Age 45–54
Overall	2,096	318	15.2%	29.4/1,000PY
No depression	1,804	273	15.1%	29.2/1,000PY
Depression	292	45	15.4%	30.9/1,000PY
Age 55–64
Overall	3,152	727	23.1%	47.0/1,000PY
No depression	2,813	657	23.4%	47.5/1,000PY
Depression	339	70	20.6%	42.7/1,000PY
Age 65–80
Overall	4,218	1,032	24.5%	52.3/1,000PY
No depression	3,864	944	24.4%	51.9/1,000PY
Depression	354	88	24.9%	57.4/1,000PY

Note: PY=person-years; TKA=Total Knee Arthroplasty

The median follow-up time from index to end of follow-up (TKA or censoring) was 66 months (IQR [interquartile range]: 42 to 81) among patients who did not have depression and 65 months (IQR: 36 to 79) among those who had depression. As shown in Table 4, depression was not associated with incident TKA (HR [hazard ratio] = 0.97; 95% CI [confidence interval]: 0.81 to 1.16) before and after controlling for confounding in weighted data. There were no significant associations between depression and incident TKA within different age groups.

Table 4.

Results (HR (95%CI)) from Cox proportional hazard models estimating the association of depression and TKA, overall and stratified by age

Age group	Crude HR (95% CI) Depression vs. no depression	Weighted HR (95% CI) Depression vs. no depression
All ages	0.97 (0.84–1.12)	0.97 (0.81–1.16)
Age 45–54	1.06 (0.77–1.46)	0.97 (0.65–1.44)
Age 55–64	0.90 (0.70–1.15)	0.84 (0.62–1.12)
Age 65–80	1.11 (0.89–1.38)	1.11 (0.85–1.45)
Age*Depression interaction term	P=0.451	P=0.398

Note: HR=Hazard Ratio, CI=Confidence Interval, TKA=Total Knee Arthroplasty

Sensitivity analyses assessing depression measurement more proximal to the end of follow-up (TKA or censoring) are shown in e-Table 3. Depression was measured for one-year prior to the start of any lag period. Results showed that there was no relationship between depression and the probability of TKA for 3-month, 2-month, or 1-month lags prior to the end of follow-up. However, if measuring depression with no lag period (from 1-year prior to 1-day prior to the end of follow-up), those who had depression had a slightly higher probability of TKA than those who did not have depression (24.2 versus 21.6%, respectively; P = 0.028).

DISCUSSION

In this large, nationally-distributed cohort of patients who have knee OA, diagnosed depression was not associated with an increased probability of or a shorter time to TKA. We found evidence that depression was associated with a slightly greater probability of TKA when identified in the 12 months prior to the date of surgery. This may suggest depression is a common comorbidity at the time of TKA, potentially identified in pre-operative risk-optimizing appointments. However, it is unlikely that depression contributed directly to the decision to pursue TKA, given there was no relationship between depression and TKA when diagnoses occurred one month or longer before TKA.

There is a large body of literature describing the bi-directional associations between depression and the increased pain sensitivity, which would infer a connection between depression and increased likelihood of a patient who has knee OA obtaining TKA due to increased pain[8–11]. Additionally, previous studies have observed that depression is associated with many factors linked to increased odds of TKA, such as poor response to conservative pain treatment, worsening knee pain, and increased pain-related impaired functioning[7, 15, 16]. However, the complexity of the interplay between depression symptoms and decisions to pursue TKA may offer an explanation for our results, such as anhedonia contributing to delays in committing to an intensive rehabilitation program[12–14, 36]. We also lack measures of patient preference, social support, and trust in the healthcare system. These factors are contributors to decisions to undergo TKA, and the role of depression in TKA may be overwhelmed by these psychosocial factors [36–38]. Our results are further supported by two prior studies of patients who had osteoarthritis, which revealed that depressive symptoms and severity of depression were not associated with willingness to undergo surgery [23, 39]. While we found no connection between depression and TKA, we did demonstrate that depression was associated with higher OASI scores, which likely indicates increased severity of knee OA, and patients undergoing a higher number of non-surgical knee OA treatments, which is constant with prior studies [21]. This suggests patients who have depression underwent more non-surgical knee OA interventions, thus depression may lead to treatment seeking for knee OA in general but has no association with the more consequential decision to undergo TKA.

Although depression is not related to the probability of TKA, it is a well-established risk factor for poor TKA outcomes[19]. Depression has been associated with poorer functional improvement, less pain relief, an increased length of stay, and an increased risk of revision [40, 41]. Since depression does not appear to delay TKA, there is a possibility that many depressed patients are undergoing TKA, which may contribute to poorer post-operative outcomes. Clinicians should consider screening for depression and ensure patients receive appropriate treatment to improve the odds of good surgical outcomes.

Potential Limitations

We lacked measures of depression severity and pain severity, both of which may contribute to seeking TKA. The sample was limited to the United States, and the results may not generalize to other regions. Observational cohort studies are vulnerable to misclassification. If true cases of depression were misclassified as nondepressed at baseline, point estimates would be biased toward the null. Use of a fixed date for cohort entry may result in individuals entering the cohort at various stages of the knee OA trajectory, e.g., severity. We felt the benefit of long follow-up outweighed this risk. While we did not have traditional means to account for severity at baseline, we utilized the validated OASI to account for this. Survivorship bias may impact this analysis, as patients must “survive” the washout/look-back period without the outcome to be included in the analysis. However, given this was roughly 2 years and the change in knee OA symptoms takes on average 6 years, we feel this likely would not have significantly biased the results [24, 25]. We are unable to control for wait times that may vary by institution; however, we argue that wait times are not likely to be systematically different among institutions attended by those who have depression and those who do not. While there are many phenotypes of depression symptoms, we are unable to adjust for specific symptoms, such as catastrophic thinking. We expect such symptoms to be more associated with post-operative outcomes rather than the decision to seek TKA. Unmeasured confounding is a potential source of bias.

Conclusions

Depression is not associated with time to TKA in this study. Surgeons should consider looking for depression and depressive symptoms in their patients, but they should not expect depression to contribute to patients choosing to delay surgery. As a common comorbidity with a potential impact on post-operative outcomes, ensuring patients are getting treatment may be key to facilitating good outcomes. Further research measuring the severity of depression and the contribution of dominant symptoms such as anhedonia is warranted to rule in or rule out a role for depression in the decision to undergo TKA. Additionally, prospective longitudinal studies using smaller cohorts within a particular institution may offer insights into this question.

Appendix A

e-Table 1.

Variable definitions

Variable	Definition
Index/baseline	January 1, 2012
“Baseline period”	2010 and 2011 – measurement of depression and covariates
Follow-up time	Months from index to study end date (incident TKA or censoring). Patients are censored at last encounter.
Knee OA	ICD-9 code: 296.2x, 296.3x, 311; ICD-10 code: F32.0-F32.5, F32.9, F33.0-F33.3, F33.4x, F33.9
Depression	Measured in the baseline period. At least 2 outpatient diagnoses in any 12-month period or 1 inpatient diagnosis. ICD-9 code: 715.16, 715.26, 715.36, 715.96, 716.56, 716.66, 716.86, 716.96; ICD-10 code: M17.x, M12.56x, M12.86x, M13.16x, M13.86x
TKA exclusions at baseline	These codes were used to exclude patients with a history of TKA prior to index/baseline. CPT: 27437, 27438, 27440, 27441, 27442, 27443, 27445, 27446, 27447, 27486, 27487; ICD-9 procedure: 81.54, 81.55, 00.80; ICD-9 diagnosis: V43.65 ICD-10 procedure: – 0SRC*, 0SRD*, 0SPC08*, 0SPC09*, 0SPC0J*, 0SPC48*, 0SPC4J*, 0SPD08*, 0SPD09*, 0SPD0J*, 0SPD48*, 0SPD4J*, 0SWC*, 0SWD* ICD-10 diagnosis: Z96.65x
TKA outcome	At least one occurrence in follow-up. CPT: 27437, 27438, 27440, 27441, 27442, 27443, 27445, 27446, 27447; ICD-9 procedure: 81.54; ICD-10 procedure: – 0SRC*, 0SRD*;
Covariates – measured in the 2-years prior to index (2010 and 2011), unless otherwise specified.
OASI index	Ranges from 0 to 6. Sum of each type of therapy/diagnostic received in 2010–2011. Xray: CPT = 73560, 73580, 73562, 27370, 73564, 73565 MRI/CT: CPT = 73721, 27369, 73722, 73700 Intra-articular injection: CPT=20611, 20610; HCPCS=J7321, J7323 to J7326 PT: CPT=97001, 97002, 97110, 97140, 97124, 97150 Opioid prescription: Oral (unless otherwise specified) for codeine, tramadol, hydrocodone, oxycodone, morphine, methadone, fentanyl patch, dihydrocodeine, hydromorphone, levorphanol, meperidine, oxymorphone, pentazocine, tapentadol NSAID prescription: Oral or topical for celecoxib, diclofenac, diflunisal, etodolac, fenoprofen, flurbiprofen, ibuprofen, indomethacin, ketoprofen, ketorolac, meclofenamate, mefenamic acid, meloxicam, nabumetone, naproxen, oxaprozine, piroxicam, sulindac, tolmetin
Charlson Comorbidity Index	See following references for conditions and weights: Charlson ME, Pompei P, Ales KL, et al. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. Journal of Chronic Diseases 1987; 40:373–383. DOI: 10.1016/0021-9681(87)90171-8 Sharabiani MT, Aylin P, Bottle A. Systematic review of comorbidity indices for administrative data. Medical Care 2012; 50(12):1109–1118. DOI: 10.1097/MLR.0b013e31825f64d0 Quan H, Sundararajan V, Halfon P, Fong A, Burnand B, Luthi JC, et al. Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data. Medical Care 2005; 43(11):1130–1139. DOI: 10.1097/01.mlr.0000182534.19832.83 Lix L, Smith M, Pitz M, Ahmed R, Quon H, Griffith J, Turner D, Hong S, Prior H, Banerjee A, Koseva I, Kulbaba C. Cancer Data Linkage in Manitoba: Expanding the Infrastructure for Research. Winnipeg, MB: Manitoba Centre for Health Policy, 2016. [Summary] [Full Report] (View)
Obesity	ICD-9 code: 278.00, 278.01; ICD-10 code: E66.9, E66.01; If available, last BMI in 2-years prior to index is ≥ 30
Any substance use disorder	ICD-9 code: 303.9x, 305.0x, 304.0x, 304.1x, 304.2x, 304.3x, 304.4x, 304.5x, 304.6x, 304.7x, 304.8x, 304.9x, 305.2x, 305.3x, 305.4x, 305.5x, 305.6x, 305.7x, 305.9x ICD-10 code: F10.x, F11.x, F12.x, F13.x, F14.x, F15.x, F16.x, F18.x, F19.x - Composite of alcohol, sedative, cocaine, cannabis, amphetamine, hallucinogens, ‘other’, opioid, opioid with other SUD, other SUD excluding opioid, unspecified drug abuse/dependence.
Smoking/nicotine dependence	ICD-9 code: V15.82, 305.1; ICD-10 code: Z87.891, Z72.0, F17.20x, F17.21x
Any anxiety disorder	ICD-9 code: 300.00, 300.01, 300.02, 300.23; ICD-10 code: F40.1x, F41.0, F41.1, F41.9 - Composite of panic disorder, anxiety disorder not otherwise specified, social phobia, and generalized anxiety disorder - 2 outpatient occurrences (on different days) in same 12 month period or 1 inpatient occurrence.
Antidepressant medication	Sustained use of any type of antidepressant medication. Sustained use is at least 2 prescriptions in any 6-month period.
High healthcare utilization	- Controls for detection bias related to more healthcare encounters. - Average number of outpatient clinic visits per month in 2-years prior to baseline. The distribution of the mean is dichotomized into high utilizer, >75th percentile vs low utilizer, ≤75th percentile.
Demographic information	Age at index; Sex (men/women); Race (White/Black/Other or unknown); Region (Midwest/Northeast/South/West/Other or unknown)
Abbreviations	OA=Osteoarthritis, TKA=Total Knee Arthroplasty, ICD=International Classification of Diseases, CPT=Current Procedural Terminology, OASI=Osteoarthritis Severity Index, MRI=Magnetic Resonance Imaging, CT=Computed Tomography, PT=Physical Therapy, and NSAID=Non-steroidal anti-inflammatory drugs

e-Table 2.

Baseline characteristics after entropy balancing – overall and within age category

Covariates	Overall Weighted SMD%	45–54 yrs Weighted SMD%	55–64 yrs Weighted SMD%	65–80 yrs Weighted SMD%
Age category
45–54	0.1	---	---	---
55–64	0.1	---	---	---
65–80	0.01	---	---	---
Women	0.2	0.2	0.3	0.2
Race
White	0.1	0.2	−0.01	0.1
Black	−0.1	−0.2	−0.01	−0.1
Other/unknown	−0.1	−0.2	−0.01	−0.1
Region
Midwest	0.1	0.2	0.0	0.1
Northeast	−0.1	−0.01	−0.1	−0.1
South	0.01	0.1	−0.01	−0.1
West	−0.04	−0.3	0.1	0.0
Other/unknown	−0.2	−0.6	0.0	−0.01
High healthcare utilization	0.1	0.2	0.1	0.2
Osteoarthritis Severity Index	0.2	0.2	0.04	0.3
Charlson index	0.2	0.1	0.1	0.3
Obese	0.2	0.3	0.1	0.1
Anxiety	0.1	0.1	0.1	0.1
Nicotine dependence	0.1	0.1	0.04	0.1
Any Substance Use Disorder	0.1	0.04	0.04	0.02
Antidepressants	0.2	0.2	0.2	0.3

SMD% = standardized mean difference percent (SMD*100)

e-Table 3.

Sensitivity analyses using lag time from depression^a diagnoses to study end date (TKA or censor date).

	Group n	% with TKA
3-month lag
No depression	8149	21.9%
Depression	1317	22.3%
P-value		p=.773
2-month lag
No depression	8135	21.8%
Depression	1331	22.5%
P-value		p=.570
1-month lag
No depression	8139	21.9%
Depression	1327	22.0%
P-value		p=.952
No lag
No depression	8082	21.6%
Depression	1384	24.2%
P-value		p=.028

^aDepression is measured in the year prior to the start of the lag period

TKA=Total Knee Arthroplasty