Abstract
Remote Cognitive Behavioral Therapy-based interventions for chronic pain (CBT-CP-based) have consistently shown modest pain benefits. With remote treatment availability accelerating and increasingly tight healthcare resources, clinical decision-makers need to understand when working with a therapist may be beneficial versus a largely self-directed online program. Using data from a large-scale 3-arm pragmatic trial, we assessed the role of patient session adherence, and potential moderators of intervention pain severity effectiveness, for two remote CBT-CP-based approaches (online program [painTRAINER] and therapist delivered via telephone/video program [health coach]) and usual medical care. Of the pre-specified moderators (demographics [sex, age, race/ethnicity, rurality], social determinants of health, and clinical variables [comorbid depression and/or anxiety, multiple types of chronic pain]), the health coach program was more effective than painTRAINER at 3 months among males and for those who screened positive for depression. No other factors moderated intervention pain severity effects at 3 or 12 months. The health coach program had higher session adherence than painTRAINER (70.4% vs 47.8%). Amongst intervention completers, pain severity outcomes were similar between CBT-CP-based interventions (Adjusted RR [95% CI]: 0.99 [0.85, 1.16] and 0.93 [0.82, 1.05] at 3 and 12 months, respectively), suggesting that both interventions may be helpful for those with a variety of demographic and clinical characteristics if adherence is achieved. Participant engagement is critical in optimizing outcomes for online programs, but these findings suggest flexibility in the specific modality for delivering remote CBT-CP based on patient preference and healthcare system capacity that may enhance scalability and patient access to care.
Trial Registration:
ClinicalTrials.gov identifier: NCT04523714
1. Introduction
Cognitive behavioral therapy-based treatments for chronic pain (CBT-CP) are effective and widely recommended.[50;55;60] Recently, use of remote CBT-CP modalities, such as guided or self-directed online programs, therapist-led telehealth sessions, and automated interactive voice response, has increased markedly, driven by reduced costs and improved accessibility.[23;47;52]
Remote CBT-CP-based interventions have shown modest pain management benefits.[47] However, no evidence indicates any remote modality is superior, with few studies directly comparing approaches.[30;38] Among remote CBT-CP-based treatments, online programs offer scalability and lower cost, making them attractive to healthcare systems. Yet, patient adherence to these largely self-directed CBT-CP programs tends to be lower than with therapist-led interventions.[10;12;24;53]
Patient motivation and engagement are important to achieving CBT-CP benefits.[2;27;49] Some patients succeed with self-directed programs, preferring to engage on their own schedule. For others, working with a therapist in real-time via telephone/video may help them overcome barriers to adopting CBT skills – especially for those with comorbid depression or anxiety, common among individuals with chronic pain.[1] These patients may experience avoidance and deactivation, impacting motivation to engage, thereby making therapist support particularly valuable.[3;5;32;33] Similarly, those having multiple types of chronic pain may benefit from a therapist’s tailoring and guidance. In addition, limited technology skills or access – common among older adults, rural residents, and those from under-resourced households – can hamper engagement with online programs.[31;56] Understanding how these factors influence treatment adherence and response may help connect patients to the most suitable remote CBT-CP services.
The RESOLVE randomized clinical trial (RCT) evaluated the effectiveness of two remote CBT-CP-based interventions for individuals with high impact chronic musculoskeletal pain, 40% of whom were from rural or medical underserved areas of the U.S. The two interventions, compared to each other and usual care in the trial, were analogous: 1) an 8-session, master’s-level behavioral therapist-led, CBT-CP-based program delivered via telephone/video (“health coach”) and 2) an 8-session, online, largely self-directed, CBT-CP-based program (painTRAINER). Both treatments showed statistically significant improvements in pain severity (binary measure of achieving minimal clinically important difference; MCID) compared to usual care at 3 months (post-treatment) with effects sustained through 12 months. The health coach program demonstrated greater improvement in pain severity post-treatment than the painTRAINER program (RR [95% CI]: 1.20 [1.03, 1.40]), but not at 12 months.[17] Patient treatment adherence in the RCT differed: 70% of the health coach group completed at least 6 sessions compared to 43% of the painTRAINER group.
Our objective is to assess if there is differential intervention effectiveness across the two remote CBT-CP approaches in achieving MCID in pain severity from baseline to post-treatment by conducting pre-specified moderator analyses based on demographics (sex; age; race/ethnicity; rurality), social determinants of health [SDoH], and clinical variables (comorbid depression and/or anxiety; multiple types of chronic pain). We hypothesized that for patients experiencing comorbid depression/anxiety, multiple types of chronic pain or higher unmet SDoH-related needs, therapist-delivered CBT-CP would improve adherence and outcomes relative to self-directed online CBT-CP because a therapist can personalize the treatment. In addition, the significant differential patient adherence by intervention observed in the RCT is explored.
2. Methods
The RESOLVE trial protocol[39] and main trial findings[17] have been previously published. It is registered under ClinicalTrials.gov identifier NCT04523714. The trial design is briefly summarized below. See Figure 1 for patient flow throughout the trial.
Figure 1.
Participant flow
Study Summary: Remote CBT-CP approaches beneficial across a variety of demographic and clinical characteristics although patient adherence critical and likely meriting support among those using online self-directed programs.
2.1. Setting and participants
The RESOLVE trial was a phase 3, comparative effectiveness, parallel-group design RCT comparing the effectiveness of two remote CBT-CP-based interventions and usual care on reducing pain severity (primary outcome). Participants (N=2,331) were randomized (1:1:1 ratio) who met these key eligibility criteria at the time of enrollment: had an ICD-10-CM diagnosis for nonmalignant musculoskeletal pain;[40] had High Impact Chronic Pain[57] determined by the Graded Chronic Pain Scale-Revised; had pain score ≥12 on PEG (Pain, Enjoyment, and General activity scale)[35]; were age 18 years or older; spoke and read English; had internet/phone access; and were receiving healthcare in one of the 4 participating healthcare systems: 1) Kaiser Permanente Georgia (KPGA); 2) Kaiser Permanente Northwest (KPNW); 3) Kaiser Permanente Washington (KPWA); and 4) Essentia Health (serving northern Minnesota, eastern North Dakota, and northern Wisconsin). Participants were recruited from January 2021 through February 2023. Follow-up occurred from April 2021 to April 2024.
2.2. Interventions
Study participants in all 3 groups could access pain treatment as usual (available pain management services in the participating healthcare systems). The 8-session CBT-CP-based interventions were developed and refined by Drs. Keefe, DeBar, and Rini in prior studies [15;16;44;45] and were analogous in sequencing and content (see Supplementary Table 1 for details).
2.2.1. Health Coach Program.
The health coach program included 8 sessions of CBT-CP-based skills training provided one-on-one via either telephone or videoconferencing per participant preference. Twelve master’s-level behavioral health practitioners (4 had master’s degrees in social work and 8 had master’s degrees in counseling-related fields), intentionally referred to as “health coaches” within the study to differentiate from “therapists” who are generally associated with mental health services, delivered the intervention and were centrally based at the KPNW and KPWA clinical study sites. All had some experience with CBT prior to their participation in the trial (although not with CBT for chronic pain) and had an average of 8.0 years (SD=6.1 years) of post-training work experience. We purposefully sought out master’s level clinicians to deliver the intervention and did not require prior experience with the treatment of chronic pain to enhance generalizability and scalability of our approach and findings. Sessions lasted ~45–60 minutes with approximately one session per week. Participants were asked to complete all 8 sessions within the 12 weeks post-randomization.
2.2.2. Online Program.
painTRAINER[44;45] is an online, 8-session, CBT-CP-based skills training program that can be accessed free of charge at www.mypaintrainer.org. Each session takes ~30–45 minutes to complete and provides interactive training in one or more evidence-based pain-coping skills. Participants self-completed approximately one session per week, with all 8 intended to be completed within 12 weeks post-randomization. Participants had access to the online program for the full year of study participation and were assisted in registering via an individual onboarding phone call. Research staff also provided technical support and outreach to encourage engagement if session completion differed from the recommended completion schedule in pre-specified ways (e.g., >10 days since the last session was completed or 3 or more sessions completed in 9 days). No guidance on treatment content was provided during these contacts.
2.2.3. Usual Care.
Participants who were randomized to the usual care group received a mailed copy of the 2020 edition of the American Chronic Pain Association Resource Guide to Chronic Pain Management.[21]
2.2.4. Therapeutic Dose and Session Completion
The study protocol prespecified that participants were deemed as receiving a full therapeutic dose of the interventions if at least 75% or 6 of the 8 possible sessions were completed[39], consistent with clinical thresholds specified in other CBT-CP clinical trials.[11;18;19;46;50]
2.3. Measures
Baseline measures, including moderators, covariates and baseline outcomes, were assessed prior to randomization via self-report through the baseline assessment (completed following consent either by phone with research staff or via a REDCap survey) or extracted from the electronic health records (EHR) using data from the 360 days prior to randomization. Outcome assessments at 3, 6, and 12 months were completed via online survey (REDCap), by phone with study staff, or by postal mail, based on participant preference, in the main trial. These secondary analyses assessed outcomes at only the 3- and 12-month follow-up time points.
2.3.1. Outcome Measure:
The primary outcome was attaining a 30% or greater decrease in pain severity score, also known as a minimal clinically important difference (MCID),[20] from baseline to 3 months (post-treatment; primary time point) and 12 months. Pain severity is a composite of pain intensity and pain-related interference and was measured by an 11-item version of the Brief Pain Inventory – Short Form (BPI-SF), which has demonstrated reliability and validity.[13;34;41;51] The score is the calculated mean of all 11 items; the range is 0–10 with a higher score indicating greater pain severity and impact. All 4 items of the pain intensity subscale and at least 4 of the 7 items in the pain-related interference subscale were required to score pain severity.
2.3.2. Potential Moderators of Intervention Effectiveness
The pre-specified potential moderators detailed in the study protocol[39] included 4 demographic variables: sex (Male vs. Female/Other), age (<65 years old vs. ≥ 65 years old), race and ethnicity (White and non-Hispanic, Black or African American and non-Hispanic, Hispanic, vs. Other); and rural and/or medically underserved residency (Rural and/or medically underserved vs. Urban). Race and ethnicity were self-reported at baseline. Rural and/or medically underserved residency was based on geocoding participants’ EHR documented address at time of randomization to the 2010 US Census tracts; those with Rural-Urban Commuting Area (RUCA) Codes 4, 5, 6, 7, 8, 9 or 10 were deemed rural and tracts corresponding to a HRSA-designated health professional shortage area[29] for primary or mental healthcare were considered medically underserved. (A participants’ census tract might be both a rural and medically underserved area or either.)
Social determinants of health (SDoH) (≥1 any vs. none) was also a prespecified potential moderator. SDoH needs were assessed in the following domains using previously validated measures: 1) Financial Resource Strain; 2) Food Insecurity; 3) Transportation Needs; and 4) Housing Instability (see Supplementary Table 2 for more details).[6;8;28;42;43]
Two clinical variables were pre-specified potential moderators: multiple types of chronic pain (EHR-identified ICD-10 diagnosis of 1 chronic pain type vs. >1 pain type) and presence of any mental health mood disorder (EHR-identified ICD-10 diagnosis of depression and/or anxiety vs. neither). Multiple types of chronic pain included the following nonmalignant musculoskeletal pain conditions: back pain; neck pain; limb/extremity pain, joint pain, and arthritic disorders; fibromyalgia; headache; orofacial, ear, and temporomandibular disorder pain; musculoskeletal chest pain; and general pain.[40]
We further included additional exploratory moderators to understand the moderating effects of anxiety and/or depression diagnoses and symptoms. Specifically, we added EHR-identified ICD-10 diagnosis of depression and anxiety as separate moderators and added a positive screen for depressive symptoms at baseline via self-report (Personal Health Questionnaire Depression Scale [PHQ-2] ≥ 3) and anxiety symptoms at baseline via self-report (General Anxiety Disorder Questionnaire 7-item [GAD-7] ≥10, indicating moderate severity). For depression symptoms, we decided to use the 2-item screener with its focus on the core cardinal symptoms of depression (sadness, anhedonia) rather than the PHQ-8 to minimize the confounding in the longer scale between symptoms common to pain and depression (e.g., trouble sleeping, fatigue, difficulty concentrating).
2.3.3. Other Baseline Covariates
To address bias due to missing data (see statistical supplement for details[17]) we further included employment status (employed vs. other), PEG pain score, and indicator of healthcare system clinical site.
2.4. Statistical Analysis
We first present descriptive statistics of baseline demographics, health and pain conditions for the entire trial sample (N=2,331). We then present the number and percent who completed the therapeutic dose of ≥ 6 sessions (“completers”) overall and by each baseline variable amongst: 1) the 2 CBT-CP-based groups combined (N=1,554); and 2) by each CBT-CP-based group (N=776 in painTRAINER and N=778 in health coach). Further, we statistically compare differences by health coach versus painTRAINER program for the binary outcome, intervention completers, as well as differences by those who received any intervention sessions, applying modified Poisson regression models including indicator for intervention group, baseline variable, and interaction between intervention group and baseline variable amongst those randomized to a CBT-CP-based intervention.
We then present a dosing analysis consisting of a series of different levels of intervention comparisons on the primary outcome, MCID in pain severity (≥30% reduction in score from baseline) at 3 and 12 months including the following comparisons: 1) 2 CBT-CP-based groups combined (N=1,554) versus usual care; 2) health coach (N=778) versus painTRAINER (N=776) Intention to Treat (ITT) amongst all those randomized to a CBT-CP-based group; and 3) health coach (N=548) versus painTRAINER (N=371) completer analysis amongst those who were intervention completers. For a given comparison, we fit a modified Poisson regression for binary outcomes using generalized estimating equations (GEE) to model the primary outcome.[37;62] We assumed working independence for correlation and calculated standard errors using the robust sandwich estimator to account for within-health coach and within-participant correlation and the mis-specified mean-variance structure when using Poisson regression for a binary outcome.[61;62] We included indicators for intervention group, time points, and all interactions by intervention and time points. To best reduce potential bias due to missing data, we included 3-, 6-, and 12-month follow-up time points together in all models (see missing data section below). For a given time point, we report the adjusted mean percent and 95% CI by intervention group and the adjusted relative risk (RR) and 95% CI comparing groups with the corresponding two-sided p-value.
Next, we conducted a series of moderator analyses for each follow-up time point (3 months [post-treatment] and 12 months) and moderator separately, to assess for differential intervention effects for the: 1) 2 CBT-CP-based groups combined versus Usual Care and 2) health coach versus painTRAINER. For a given moderator, we fit a similar model as detailed for the dosing analysis except in the modified Poisson regression model we include the indicators for the intervention group, the moderator, and the interaction between intervention and moderator, as well as all time-by-moderator-by-intervention interactions. Within each moderator subgroup, for a given time point, we report the adjusted mean percent and 95% CI by intervention group and the adjusted relative risk (RR) and 95% CI comparing groups with the corresponding two-sided p-value for within subgroup comparison. We further report the interaction p-value, comparing a difference across the subgroup RRs, to determine statistically significant moderation of effect.
Finally, we conducted an exploratory moderator analysis comparing the health coach program versus painTRAINER within the subset of participants who completed ≥6 sessions (“completers”) to assess if patient adherence to intervention explained away any moderation of the effect comparing different CBT-CP-based programs.
Missing Data:
We applied the same missing data approach used in the main study outcomes analyses as detailed in the published manuscript.[17] Briefly, to account for potential bias due to missing data (Figure 1 shows missing data amounts by group) we use the following combination of approaches: 1) pre-specified baseline covariate adjustment (baseline pain severity score, sex, age, clinical site, rural/medically underserved residency, multisite pain, and co-occurring mental health conditions (presence of depression and/or anxiety diagnoses); 2) robust non-ignorable pattern mixture imputation[58] for those with at least one observed follow-up outcome (added additional covariates predictive of 3-month primary outcome response: education (any college vs. HS degree or less), unemployment (unemployed vs. employed [full or part-time]); and 3) non-response weighting for those without any follow-up outcomes (added additional covariates predictive of not having any follow-up outcomes: PEG score, Depression (PHQ-8 score), Anxiety (GAD-7 score), presence of any negative social determinant of health, and education (any college vs. HS degree or less)). We used all outcome time points (3, 6, and 12 months) that were collected for the study to improve missing data imputation, but only report moderator analyses for 3 and 12 months because these were the pre-specified time points of interest. For the completer analyses, given this is a post-randomization subset of the dataset, we further weighted the completers population back to the randomized population using inverse probability of completion non-response multiplied by the inverse probability of outcome non-response.
All analyses were performed using R version 4.4.1 on Windows 10 and all statistical tests and confidence intervals were two-sided with an α=0.05.
3. Results
Patient treatment adherence.
Table 1 presents baseline characteristics for the overall study sample and for those who met the therapeutic dose threshold – completion of at least 6 of 8 sessions (intervention “completers”) for each CBT-CP-based program. Those randomized to the health coach program had a much higher intervention completion percent than did those randomized to painTRAINER (70.4% vs 47.8%). Males (52.6% vs 61.5% females), those who were younger than 65 years old (55.2% vs 65.4% for age ≥65), those who do not self-report as White race (51.9% Black Non-Hispanic vs 48.9% Hispanic vs 61.3% White Non-Hispanic), those with any SDoH (50.2% vs 63.7% none), those with a positive screen for depression (50.2% vs 65.3%) and those with higher pain severity at baseline (52.0% vs 61.6% for those ≥7 vs <7 on BPI-SF) were less likely to complete either CBT-CP-based program. All findings described above reflect ~10% or greater differences between groups. Supplementary Table 3 presents baseline characteristics for participants in each CBT-CP-based group by session completion: no sessions completed, 1–5 sessions completed, and 6–8 sessions completed. Amongst non-completers (< 6 sessions) most completed no sessions (painTRAINER 61% (249/405) and health coach 51% (118/230)).
Table 1.
Baseline characteristics overall and by the proportion with therapeutic dose completion (6–8 sessions) by CBT-CP program
| Intervention Completion (6–8 Sessions) | ||||
|---|---|---|---|---|
| Overall | CBT-CP programs | painTRAINER | Health Coach | |
| BASELINE CHARACTERISTICS | N (Column %) | N complete/N (Completion %) | N complete/N (Completion %) | N complete/N (Completion %) |
| TOTAL | 2331 (100.0) | 919/1554 (59.1) | 371/776 (47.8) | 548/778 (70.4) |
| DEMOGRAPHICS | ||||
| Sexb | ||||
| Female or Non-Binary | 1712 (73.4) | 703/1143 (61.5) | 285/571 (49.9) | 418/572 (73.1) |
| Male | 619 (26.6) | 216/411 (52.6) | 86/205 (42.0) | 130/206 (63.1) |
| Age | ||||
| <65 years old | 1431 (61.4) | 529/958 (55.2) | 222/485 (45.8) | 307/473 (64.9) |
| ≥ 65 years old | 900 (38.6) | 390/596 (65.4) | 149/291 (51.2) | 241/305 (79.0) |
| Race and Ethnicity | ||||
| White and Non-Hispanic | 1699 (75.0) | 702/1145 (61.3) | 282/573 (49.2) | 420/572 (73.4) |
| Black/African American and Non-Hispanic | 350 (15.4) | 123/237 (51.9) | 54/119 (45.4) | 69/118 (58.5) |
| Hispanic | 77 (3.4) | 23/47 (48.9) | 6/21 (28.6) | 17/26 (65.4) |
| Other | 140 (6.2) | 47/82 (57.3) | 17/38 (44.7) | 30/44 (68.2) |
| Rural/medically underserved residencya,d | ||||
| Urban | 1301 (55.8) | 521/869 (60.0) | 219/436 (50.2) | 302/433 (69.7) |
| Rural/medically underserved residency | 1030 (44.2) | 398/685 (58.1) | 152/340 (44.7) | 246/345 (71.3) |
| Existing Social Determinate of Health | ||||
| None | 1540 (67.0) | 661/1037 (63.7) | 260/502 (51.8) | 401/535 (75.0) |
| Any | 758 (33.0) | 247/492 (50.2) | 105/261 (40.2) | 142/231 (61.5) |
| EHR DIAGNOSES AND HEALTH VARIABLES | ||||
| Musculoskeletal pain conditionsa,e | ||||
| 1 pain condition | 621 (26.6) | 253/417 (60.7) | 116/216 (53.7) | 137/201 (68.2) |
| > 1 pain condition | 1710 (73.4) | 666/1137 (58.6) | 255/560 (45.5) | 411/577 (71.2) |
| Depression and/or Anxiety diagnosisa | ||||
| None | 1366 (58.6) | 576/932 (61.8) | 246/476 (51.7) | 330/456 (72.4) |
| Depression and/or Anxiety diagnosis | 965 (41.4) | 343/622 (55.1) | 125/300 (41.7) | 218/322 (67.7) |
| Depression diagnosis | 685 (29.4) | 240/452 (53.1) | 85/216 (39.4) | 155/236 (65.7) |
| Anxiety diagnosis | 682 (29.3) | 231/441 (52.4) | 91/221 (41.2) | 140/220 (63.6) |
| Depression Symptoms | ||||
| Screen Negative | 1395 (60.5) | 607/930 (65.3) | 254/466 (54.5) | 353/464 (76.1) |
| Screen Positive | 910 (39.5) | 303/604 (50.2) | 114/301 (37.9) | 189/303 (62.4) |
| Anxiety Symptoms | ||||
| < Moderate | 1681 (72.2) | 692/1128 (61.3) | 279/568 (49.1) | 413/560 (73.8) |
| Moderate+ | 648 (27.8) | 227/424 (53.5) | 92/206 (44.7) | 135/218 (61.9) |
| BASELINE OUTCOME | ||||
| Pain Severity (0–10) | ||||
| <7 score | 1737 (74.5) | 714/1160 (61.6) | 290/581 (49.9) | 424/579 (73.2) |
| ≥7 score | 594 (25.5) | 205/394 (52.0) | 81/195 (41.5) | 124/199 (62.3) |
Data are from the electronic health record (EHR); diagnoses are based on ICD-10 codes.
Data source was self-reported sex from study survey unless missing and then used sex from EHR.
Missing is reported as n but excluded from the denominator in the % reported
Rural defined as subject’s resident Census Tract corresponds to US Census 2010 Rural-Urban Commuting Area (RUCA) Codes 4, 5, 6, 7, 8, 9 or 10. Medically underserved is defined as subject’s resident Census Tract corresponds to HRSA-designated primary care or mental health geographic or geographic high needs health professional shortage area.
Based on ICD-10 diagnoses in past year and includes following non-malignant musculoskeletal chronic pain conditions: back pain; neck pain; limb/extremity pain, joint pain, and arthritic disorders; fibromyalgia; headache; orofacial, ear, and temporomandibular disorder pain; musculoskeletal chest pain; general pain.
The only statistically significant moderator differentiating intervention completion (≥ 6 sessions) between the two CBT-CP-based groups was the presence of multiple, rather than a single, musculoskeletal chronic pain types. Those with multiple chronic pain types were 8.2% less likely to complete painTRAINER (>1 pain types 45.5% vs. 1 pain type 53.7%) relative to the health coach program (>1 pain type 71.2% vs. 1 pain type 68.2%) (Supplementary Table 4; Adjusted RR (95% CI) comparing health coach vs painTRAINER completers among >1 chronic pain type: 1.57 (1.42, 1.74) vs a single pain type: 1.27 (1.10, 1.47); Interaction P-value 0.017). Those with depression and/or anxiety diagnoses at baseline had a 10.0% lower completion percent of painTRAINER (41.7% with vs. 51.7% without) and a 4.7% lower completion percent of the health coach program (67.7% with vs. 72.4 without), but the effect of the moderator of completion status was not statistically significant (Interaction P-value=0.088). A similar finding was observed for those screening positive for depression symptoms at baseline with a 16.6% lower completion percent for painTRAINER (37.9% vs 54.5% without) and a 13.7% lower completion percent for health coach (62.4% vs 76.1% without; interaction P-value=0.066).
Regarding any intervention session completion (≥1 session), those with any SDoH need were less likely to complete any sessions of painTRAINER (59.5% with vs 71.0% without), but this was not observed for the health coach group (83.9% with vs. 85.2% without; Adjusted RR (95% CI) comparing health coach vs. painTRAINER any session completers among those with any SDOH: 1.41 (1.25, 1.60) vs. those with none: 1.20 (1.13, 1.27); Interaction P-value 0.022). Similar, but not statistically significant findings, were observed for those with a depression diagnosis at baseline and those with a positive depression screen at baseline (Supplementary Table 4).
Dosing Analyses comparing Different Intervention Effects on pain severity MCID.
Participants randomized to either CBT-CP-based program were more likely to achieve an MCID in pain severity (≥30% improvement) than those in usual care at both 3 months (Adjusted RR (95% CI): 1.41 (1.20, 1.66)) and 12 months (Adjusted RR (95% CI): 1.37 (1.20, 1.55)) (Table 2). Using an ITT approach to analysis, the health coach program was more effective than the painTRAINER program at 3 months (Adjusted RR: 1.20 (1.03, 1.40)), but this difference was not sustained at 12 months (Adjusted RR (95% CI): 1.07 (0.96, 1.18)). Yet, when restricting to those who were intervention completers (≥6 sessions), those in the health coach and painTRAINER programs had similar effectiveness at 3 months (Adjusted RR (95% CI): 0.99 (0.85, 1.16)) and at 12 months (Adjusted RR (95% CI): 0.93 (0.82, 1.05)).
Table 2.
Dosing analysis to compare the effectiveness of either CBT-CP program (Health Coach or painTRAINER) relative to Usual Care and Health Coach to painTRAINER (ITT and amongst completers) in achieving a 30% reduction in pain severity from baseline at 3 and 12 months
| 3 Months | 12 Months | |||||||
|---|---|---|---|---|---|---|---|---|
| Adjusted % (95% CI)a | Adjusted RR (95% CI) | Adjusted % (95% CI)a | Adjusted RR (95% CI) | |||||
| A | B | P Val | A | B | P Val | |||
| CBT-CP programs combined (B) versus Usual Care (A) (N=2331) | 20.9 (18.1, 24.1) | 29.5 (27.2, 31.9) | 1.41 (1.20, 1.66) | <0.001 | 27.3 (24.2, 30.7) | 37.3 (35.4, 39.3) | 1.37 (1.20, 1.55) | <0.001 |
| Health Coach (B) versus painTRAINER (A) ITT | ||||||||
| ITT Population (N=1554) | 26.8 (23.6, 30.4) | 32.2 (29.5, 35.1) | 1.20 (1.03, 1.40) | 0.019 | 36.1 (32.7, 40.0) | 38.5 (37.0, 40.1) | 1.07 (0.96, 1.18) | 0.218 |
| Amongst Completers (N=919) | 37.4 (32.9, 42.6) | 37.1 (33.9, 40.7) | 0.99 (0.85, 1.16) | 0.928 | 43.2 (38.6, 48.5) | 40.3 (38.2, 42.4) | 0.93 (0.82, 1.05) | 0.254 |
Bold indicates interaction p-value is statistically significant at the 0.05-level
Adjusted mean % and adjusted RR were calculated from a modified Poisson regression model fit using generalized estimating equations (GEE) for each binary outcome. Adjusted mean % assume the mean of the covariate distribution to calculate randomized population average effects.
Moderator Analyses for 2 CBT-CP-based Intervention Groups Combined versus Usual Care.
Among those without a depression and/or anxiety diagnosis at baseline, CBT-CP-based intervention overall (2 groups combined) was more likely to achieve an MCID in pain severity than usual care at 3 months (post-treatment) (Adjusted RR (95% CI): 1.65 (1.33, 20.5) compared to those with a depression and/or anxiety diagnosis (Adjusted RR (95% CI): 1.13 (0.89, 1.44); Interaction P-value 0.021) (Table 3). This moderation effect was consistent in separate analysis for those with anxiety diagnosis and directionally consistent, but not statistically significant, for those with a depression diagnosis, those screening positive for depression symptoms, and those with moderate symptoms of anxiety. No other moderators were statistically significant at 3 months. At 12 months there were no statistically significant moderator interactions (Supplementary Table 5).
Table 3.
Moderator analysis to compare the effectiveness of either CBT-CP program (Health Coach or painTRAINER) relative to Usual Care in achieving a 30% reduction in pain severity from baseline at 3 months for each moderator
| Adjusted % (95% CI)a | Adjusted RR (95% CI) | Subgroup | Interact | ||
|---|---|---|---|---|---|
| Usual Care | CBT-CP | P Val | P Valb | ||
| Sex | |||||
| Female or Non-Binary | 21.8 (18.5, 25.7) | 31.4 (28.9, 34.0) | 1.44 (1.20, 1.72) | <0.001 | 0.690 |
| Male | 18.5 (13.9, 24.6) | 24.6 (20.6, 29.4) | 1.33 (0.95, 1.86) | 0.092 | |
| Age | |||||
| <65 years old | 19.4 (16.1, 23.5) | 27 (23.6, 30.9) | 1.39 (1.12, 1.73) | 0.003 | 0.816 |
| ≥ 65 years old | 23.3 (18.6, 29.2) | 33.7 (29.5, 38.5) | 1.45 (1.13, 1.86) | 0.004 | |
| Race and Ethnicity | |||||
| White | 21.8 (18.5, 25.7) | 29.8 (27.1, 32.7) | 1.37 (1.13, 1.65) | 0.001 | 0.854 |
| Black/African American | 17.0 (11.1, 26.1) | 27.1 (21.8, 33.8) | 1.59 (1.01, 2.51) | 0.045 | |
| Hispanic | 17.3 (8.00, 37.6) | 32.6 (18.8, 56.7) | 1.89 (0.73, 4.89) | 0.191 | |
| Other | 20.8 (12.1, 35.9) | 31.1 (23.0, 42.3) | 1.50 (0.80, 2.80) | 0.208 | |
| Rural/medically underserved residency | |||||
| Urban | 21.4 (17.8, 25.8) | 30.9 (27.6, 34.5) | 1.44 (1.17, 1.77) | 0.001 | 0.758 |
| Rural/medically underserved | 20.2 (16.1, 25.3) | 27.7 (25, 30.7) | 1.37 (1.07, 1.75) | 0.013 | |
| Multiple musculoskeletal pain conditions | |||||
| 1 pain condition | 22.8 (17.4, 29.8) | 32.8 (28.2, 38.1) | 1.44 (1.06, 1.95) | 0.020 | 0.886 |
| > 1 pain condition | 20.3 (17.1, 24.0) | 28.4 (25.8, 31.2) | 1.4 (1.16, 1.70) | 0.001 | |
| Neuropathy or Fibromyalgiac | |||||
| No | 21.7 (18.5, 25.4) | 31.0 (28.4, 33.9) | 1.43 (1.19, 1.71) | 0.000 | 0.731 |
| Yes | 18.4 (13.2, 25.6) | 24.5 (20.8, 28.8) | 1.33 (0.92, 1.92) | 0.126 | |
| Depression and/or Anxiety diagnoses | |||||
| No | 19.4 (15.9, 23.6) | 31.9 (28.9, 35.3) | 1.65 (1.33, 2.05) | <0.001 | 0.021 |
| Yes | 22.9 (18.6, 28.1) | 25.9 (22.8, 29.5) | 1.13 (0.89, 1.44) | 0.303 | |
| Depression diagnosisc | |||||
| No | 21.1 (17.9, 25) | 31.4 (28.8, 34.3) | 1.49 (1.23, 1.79) | <0.001 | 0.311 |
| Yes | 20.2 (15.4, 26.6) | 25 (21.1, 29.5) | 1.23 (0.9, 1.69) | 0.192 | |
| Anxiety diagnosisc | |||||
| No | 19.5 (16.3, 23.3) | 30.9 (28.1, 33.9) | 1.58 (1.29, 1.93) | <0.001 | 0.036 |
| Yes | 23.9 (18.8, 30.4) | 26.2 (22.6, 30.4) | 1.1 (0.83, 1.45) | 0.519 | |
| Screen Positive Depression symptomsc | |||||
| No | 21.5 (18.0, 25.7) | 33.6 (30.9, 36.6) | 1.56 (1.28, 1.89) | <0.001 | 0.180 |
| Yes | 19.5 (15.2, 25.0) | 23.8 (20.0, 28.3) | 1.22 (0.91, 1.65) | 0.184 | |
| Moderate+ Anxiety symptomsc | |||||
| No | 20.2 (17.0, 23.9) | 30.3 (27.7, 33.2) | 1.50 (1.24, 1.82) | <0.001 | 0.271 |
| Yes | 22.7 (17.5, 29.5) | 27.7 (23.0, 33.3) | 1.22 (0.89, 1.67) | 0.222 | |
| Negative social determinants of health | |||||
| None | 23.0 (19.5, 27.1) | 30.5 (27.9, 33.4) | 1.33 (1.10, 1.60) | 0.003 | 0.263 |
| Any Negative SDOH/existing need | 16.8 (12.6, 22.5) | 27.7 (23.4, 32.7) | 1.64 (1.18, 2.29) | 0.003 | |
Bold indicates interaction p-value is statistically significant at the 0.05-level
Adjusted mean % and adjusted RR were calculated from a modified Poisson regression model fit using generalized estimating equations (GEE) for each binary outcome. Adjusted mean % assume the mean of the covariate distribution to calculate randomized population average effects.
Interact Pval is the Interaction P-value referring to the test of statistical significance of the interaction effects which determine statistically significant differences in the relative risks by subgroup
Exploratory moderators that were not specified in the trials statistical analysis plan but assessed to better understand the relationship of the pre-specified moderator of having a depression and/or anxiety diagnosis that may be harder to treat in the EHR. All other moderators evaluated were pre-specified in the statistical analysis plan
Moderator Analysis for Health Coach vs. painTRAINER.
Relative to the painTRAINER program at 3 months, the health coach program was more effective among males (Adjusted RR (95% CI): 1.80 (1.28, 2.54)) compared to females (Adjusted RR (95% CI): 1.07 (0.91, 1.26); Interaction P-value 0.006) (Table 4). This 3-month pain severity outcome moderator effect did not persist at 12 months (Supplementary Table 6) and no other moderators were statistically significant at 3- or 12-month follow-up.
Table 4.
Moderator analysis to compare the effectiveness of Health Coach program to painTRAINER program in achieving a 30% reduction in pain severity from baseline at 3 months for each moderator
| Adjusted % (95% CI)a | Adjusted RR (95% CI) | Subgroup | Interaction | ||
|---|---|---|---|---|---|
| painTRAINER | Health Coach | P Value | P Valueb | ||
| Sex | |||||
| Female or Non-Binary | 30.3 (26.4, 34.7) | 32.4 (29.4, 35.7) | 1.07 (0.91, 1.26) | 0.428 | 0.006 |
| Male | 17.7 (12.7, 24.5) | 31.8 (28.3, 35.8) | 1.80 (1.28, 2.54) | 0.001 | |
| Age | |||||
| <65 years old | 25.5 (21.6, 30.2) | 29.2 (24.5, 34.7) | 1.14 (0.92, 1.43) | 0.233 | 0.490 |
| ≥ 65 years old | 28.9 (23.4, 35.6) | 37.3 (32.1, 43.2) | 1.29 (1.02, 1.63) | 0.035 | |
| Race and Ethnicity | |||||
| White | 27.6 (23.8, 32.0) | 32.6 (29.2, 36.3) | 1.18 (0.99, 1.41) | 0.071 | 0.530 |
| Black/African American | 24.5 (17.6, 34.2) | 27.6 (21.2, 35.9) | 1.12 (0.77, 1.64) | 0.546 | |
| Hispanic | 13.4 (3.3, 54.0) | 46.7 (31.0, 70.4) | 3.48 (0.82, 14.8) | 0.091 | |
| Other | 27.4 (16.1, 46.8) | 33.9 (24.2, 47.6) | 1.24 (0.66, 2.33) | 0.508 | |
| Rural/medically underserved residency | |||||
| Urban | 26.5 (22.3, 31.4) | 34.9 (31.1, 39.2) | 1.32 (1.09, 1.6) | 0.005 | 0.106 |
| Rural/medically underserved | 27.4 (22.5, 33.3) | 28.6 (26.0, 31.5) | 1.04 (0.84, 1.30) | 0.703 | |
| Multiple musculoskeletal pain conditions | |||||
| 1 pain condition | 28.5 (22.6, 36.0) | 37.6 (31.9, 44.3) | 1.32 (0.99, 1.76) | 0.060 | 0.471 |
| > 1 pain condition | 26.3 (22.6, 30.5) | 30.3 (26.9, 34.2) | 1.15 (0.95, 1.40) | 0.142 | |
| Neuropathy or Fibromyalgiac | |||||
| No | 27.8 (24.2, 32.0) | 34.1 (31.1, 37.4) | 1.23 (1.03, 1.45) | 0.019 | 0.421 |
| Yes | 23.8 (17.8, 31.8) | 25.1 (21.3, 29.7) | 1.06 (0.76, 1.47) | 0.750 | |
| Depression and/or Anxiety diagnoses | |||||
| No | 29.6 (25.5, 34.4) | 34.4 (30.4, 38.8) | 1.16 (0.96, 1.4) | 0.124 | 0.517 |
| Yes | 22.6 (17.8, 28.6) | 29.1 (25.9, 32.7) | 1.29 (1.00, 1.67) | 0.054 | |
| Depression diagnosisc | |||||
| No | 29.1 (25.3, 33.5) | 33.7 (30.5, 37.2) | 1.16 (0.98, 1.37) | 0.091 | 0.396 |
| Yes | 21.2 (15.9, 28.3) | 28.7 (24.3, 33.9) | 1.35 (0.97, 1.88) | 0.071 | |
| Anxiety diagnosisc | |||||
| No | 28.5 (24.7, 32.9) | 33.2 (29.5, 37.4) | 1.16 (0.97, 1.4) | 0.102 | 0.595 |
| Yes | 23.0 (17.5, 30.2) | 29.6 (25.6, 34.2) | 1.29 (0.95, 1.75) | 0.103 | |
| Screen Positive Depression symptomsc | |||||
| No | 32.7 (28.4, 37.6) | 34.7 (31.5, 38.3) | 1.06 (0.90, 1.26) | 0.476 | 0.014 |
| Yes | 17.8 (13.6, 23.5) | 29.3 (25.0, 34.3) | 1.64 (1.21, 2.23) | 0.001 | |
| Moderate+ Anxiety symptomsc | |||||
| No | 28.1 (24.3, 32.5) | 32.7 (29.4, 36.4) | 1.16 (0.97, 1.39) | 0.096 | 0.621 |
| Yes | 24.0 (18.3, 31.3) | 30.8 (24.6, 38.5) | 1.29 (0.92, 1.81) | 0.147 | |
| Negative social determinants of health | |||||
| None | 27.5 (23.6, 32.0) | 33.0 (29.9, 36.4) | 1.20 (1.00, 1.44) | 0.048 | 0.715 |
| Any Negative SDOH/existing need | 24.8 (19.5, 31.4) | 31.7 (25.8, 39.0) | 1.28 (0.94, 1.74) | 0.115 | |
Bold indicates interaction p-value is statistically significant at the 0.05-level
Adjusted mean % and adjusted RR were calculated from a modified Poisson regression model fit using generalized estimating equations (GEE) for each binary outcome. Adjusted mean % assume the mean of the covariate distribution to calculate randomized population average effects.
Interaction P-value refers to the test of statistical significance of the interaction effects which determine statistically significant differences in the relative risks by subgroup
Exploratory moderators that were not specified in the trials statistical analysis plan but assessed to better understand the relationship of the pre-specified moderator of having a depression and/or anxiety diagnosis and neuropathic pain that may be harder to treat in the EHR. All other moderators evaluated were pre-specified in the statistical analysis plan
In an exploratory analysis, amongst those with a positive screen for depression symptoms at 3 months, the health coach intervention was more effective than painTRAINER (Adjusted RR (95% CI): 1.64 (1.21, 2.23)) relative to those without depression symptoms (Adjusted RR (95% CI): 1.06 (0.90, 1.26); Interaction P-value 0.014).
Moderator Analyses for Health Coach vs. painTRAINER amongst Intervention “Completers”.
The only moderator difference observed at 3 months was that the health coach program resulted in more improvement than painTRAINER among males (Adjusted RR (95% CI): 1.59 (1.03, 2.44)) relative to females (Adjusted RR (95% CI): 0.90 (0.77, 1.05)) (Table 5). The adjusted percent achieving MCID among painTRAINER completers was 41.8% (95% CI: 36.5%, 47.9%) for females, but only 23.7% (95% CI: 16.2%, 34.8%) for males. There were minimal by sex differences among health coach completers (female 37.5% (95% CI: 34.9%, 40.2%) vs. male 37.7% (95% CI: 31.1%, 45.6%)). The completers analyses found no statistically significant moderators at 12 months (Supplementary Table 7).
Table 5.
Completer moderator analysis to compare the effectiveness of Health Coach to painTRAINER in achieving a 30% reduction in pain severity from baseline at 3 months for each moderator amongst those who completed at least 6 sessions
| Adjusted % (95% CI)a | Adjusted RR (95% CI) | Subgroup | Interaction | ||
|---|---|---|---|---|---|
| painTRAINER | Health Coach | P Value | P Valueb | ||
| Sex | |||||
| Female or Non-Binary | 41.8 (36.5, 47.9) | 37.5 (34.9, 40.2) | 0.90 (0.77, 1.05) | 0.164 | 0.011 |
| Male | 23.7 (16.2, 34.8) | 37.7 (31.1, 45.6) | 1.59 (1.03, 2.44) | 0.037 | |
| Age | |||||
| <65 years old | 37.9 (32.2, 44.6) | 35.1 (29.1, 42.3) | 0.93 (0.74, 1.16) | 0.501 | 0.310 |
| ≥ 65 years old | 36.7 (29.2, 46.1) | 41.1 (35.1, 48.1) | 1.12 (0.86, 1.47) | 0.406 | |
| Race and Ethnicity | |||||
| White | 37.3 (31.9, 43.7) | 37.3 (32.8, 42.3) | 1.00 (0.81, 1.23) | 0.994 | 0.548 |
| Black/African American | 37.8 (27.6, 51.8) | 34.1 (26.6, 43.9) | 0.90 (0.63, 1.30) | 0.585 | |
| Hispanic | 15.9 (2.7, 92.6) | 53.6 (31.7, 90.5) | 3.38 (0.54, 21.28) | 0.195 | |
| Other | 35.4 (18.8, 66.4) | 41.5 (31.1, 55.3) | 1.17 (0.59, 2.35) | 0.653 | |
| Rural/medically underserved residency | |||||
| Urban | 36.4 (30.6, 43.4) | 40.2 (35.0, 46.3) | 1.10 (0.90, 1.36) | 0.348 | 0.120 |
| Rural/medically underserved | 39.1 (31.7, 48.3) | 33.9 (30.6, 37.6) | 0.87 (0.69, 1.10) | 0.233 | |
| Multiple musculoskeletal pain conditions | |||||
| 1 pain condition | 39.0 (30.9, 49.3) | 44.7 (36.3, 55.1) | 1.15 (0.84, 1.57) | 0.397 | 0.314 |
| > 1 pain condition | 37.1 (31.7, 43.3) | 35.1 (31.4, 38.9) | 0.94 (0.78, 1.14) | 0.544 | |
| Neuropathy or Fibromyalgiac | |||||
| No | 39.1 (33.9, 45.1) | 39.6 (35.7, 43.8) | 1.01 (0.85, 1.21) | 0.901 | 0.528 |
| Yes | 33.0 (24.2, 45.1) | 29.5 (25.3, 34.3) | 0.89 (0.63, 1.26) | 0.521 | |
| Depression and/or Anxiety diagnoses | |||||
| No | 38.7 (33.2, 45.2) | 38.8 (33.9, 44.5) | 1.00 (0.82, 1.23) | 0.974 | 0.948 |
| Yes | 35.5 (27.9, 45.2) | 35.2 (32.0, 38.8) | 0.99 (0.77, 1.28) | 0.954 | |
| Depression diagnosisc | |||||
| No | 37.3 (32.1, 43.3) | 38.3 (34.4, 42.7) | 1.03 (0.86, 1.24) | 0.769 | 0.461 |
| Yes | 38.8 (29.7, 50.6) | 35.0 (30.8, 39.9) | 0.90 (0.68, 1.21) | 0.497 | |
| Anxiety diagnosisc | |||||
| No | 38.6 (33.3, 44.6) | 37.5 (33.2, 42.4) | 0.97 (0.80, 1.18) | 0.774 | 0.648 |
| Yes | 34.7 (26.1, 46.3) | 37.1 (31.4, 43.7) | 1.07 (0.77, 1.48) | 0.701 | |
| Screen Positive Depression symptomsc | |||||
| No | 41.4 (35.8, 47.9) | 39.2 (34.8, 44.1) | 0.95 (0.79, 1.14) | 0.566 | 0.220 |
| Yes | 29.1 (22.0, 38.5) | 35.1 (28.6, 43.1) | 1.21 (0.87, 1.68) | 0.265 | |
| Moderate+ Anxiety symptomsc | |||||
| No | 39.1 (33.8, 45.3) | 37.6 (33.5, 42.1) | 0.96 (0.80, 1.16) | 0.667 | 0.512 |
| Yes | 33.5 (25.5, 44.0) | 36.8 (29.6, 45.8) | 1.10 (0.78, 1.55) | 0.587 | |
| Negative social determinants of health | |||||
| None | 37.0 (31.6, 43.3) | 37.6 (34, 41.6) | 1.02 (0.84, 1.23) | 0.850 | 0.883 |
| Any Negative SDOH/existing need | 38.0 (29.9, 48.2) | 37.6 (30, 47.2) | 0.99 (0.72, 1.37) | 0.954 | |
Bold indicates interaction p-value is statistically significant at the 0.05-level
Adjusted mean % and adjusted RR were calculated from a modified Poisson regression model fit using generalized estimating equations (GEE) for each binary outcome. Adjusted mean % assume the mean of the covariate distribution to calculate randomized population average effects.
Interaction P-value refers to the test of statistical significance of the interaction effects which determine statistically significant differences in the relative risks by subgroup
Exploratory moderators that were not specified in the trials statistical analysis plan but assessed to better understand the relationship of the pre-specified moderator of having a depression and/or anxiety diagnosis that may be harder to treat in the EHR. All other moderators evaluated were pre-specified in the statistical analysis plan
4. Discussion
This research explored patient treatment adherence and moderators of clinical response among participants in a large-scale pragmatic trial comparing two remote CBT-CP-based approaches to one another and to usual medical care. We examined two important elements. First, given the substantially lower patient intervention adherence in the painTRAINER group (68% completing at least one session and 43% receiving a therapeutic dose [≥6 sessions]) than the health coach group (85% completing at least one session and 70% receiving a therapeutic dose), we assessed factors related to intervention uptake (completing ≥1 session) and completion (≥6 sessions) in both groups, and conducted an analysis examining the effectiveness of each of the CBT-CP-based programs among those who received a full therapeutic dose. In addition, we examined moderators of treatment response for both groups combined as well as whether there were baseline factors associated with differential benefit of one or the other remote CBT-CP-based approaches.
In examining treatment adherence, females and older individuals (≥65 years) as well as those with lower pain severity (< 7 on BPI-SF) at baseline were more likely to be completers in either program. In addition, having one versus multiple types of musculoskeletal chronic pain predicted completion for those in the painTRAINER group (those with one pain type more likely to complete) whereas completion rates were similar across one or multiple chronic pain types among those in the health coach group. Further, those with existing SDoH needs were less likely to complete any sessions of the painTRAINER program compared to those without such needs -- a difference not found for those in the health coach group. Finally, there was a trend towards baseline depression-related factors resulting in less likelihood of completing any (≥1 session) as well as a full dose of sessions (≥6 sessions) in the painTRAINER versus health coach programs.
We did not have a priori hypotheses regarding the potential impact of demographic factors (age, sex) on differential completion. However, we had posited that therapists’ ability to adapt and personalize the treatment and identify obstacles to participating may enhance adherence for those with complex social, clinical and other health-related issues. Our findings were consistent with this, as those with more than one chronic pain type, existing SDoH needs, and/or depression-related factors exhibited lower completion of painTRAINER, yet these factors did not similarly impact adherence among those in the health coach group, suggesting that working with a therapist in real time may allow better tailoring of the treatment to those with more complex needs. These findings are consistent with other studies reporting better patient adherence to online psychological interventions[4] and less early dropout from CBT-CP[27] for women, as well as better adherence to CBT-CP among older individuals[48] and those with fewer depressive symptoms[25], yet these factors haven’t been consistent predictors of adherence across all studies. To our knowledge, lower baseline pain severity has not been reported as a predictor of patient treatment adherence in other CBT-CP studies. Similarly, we could find no reports of SDoH needs predicting lower uptake of CBT-CP or other psychological treatments, but a recent review found that these needs predicted poor medication adherence.[59]
Importantly, the exploratory analyses restricted to the “completers” subset -- those in either CBT-CP-based group who received a full therapeutic dose – identified a similar proportion meeting MCID for pain severity-related improvements, suggesting therapeutic equivalency across the approaches when adherence is not differential. Thus, future efforts might focus on how to promote engagement with the programs, particularly self-directed CBT-CP-based programs like painTRAINER.
In examining moderators of treatment response, having an EHR-identified diagnosis of depression and/or anxiety emerged as the only significant predictor of outcome. Those in the 2 CBT-CP-based treatment groups combined who did NOT have such diagnoses experienced greater improvement than did those with a depression and/or anxiety diagnosis (and this was directionally consistent with findings for depressive/anxiety symptoms at baseline). A recent review of CBT-CP studies found that depression and anxiety-related baseline factors were among the most commonly reported predictors of treatment outcome with medium to large effect sizes,[22] however, not all studies have found such factors to moderate pain-related outcomes.[7;26]
For moderator analyses examining baseline factors that predicted differential response to one or the other remote CBT-CP-based program, the health coach program was more effective than the online painTRAINER program among males and for those who screened positive for depression at baseline. No other factors moderated intervention effects at 3 months and there were no significant moderators at 12 months. To our knowledge, there are no studies that have compared different modes of delivery of a similar CBT-CP program, although Kroenke and colleagues report on a recent trial that compared two intervention models relying heavily on telecare delivery, but differing in resource intensity, finding that adding human interaction to the self-management program was comparatively more effective for their target population with chronic pain and concomitant depression or anxiety.[36]
Finally, for analyses restricted to the subset of intervention completers from the CBT-CP programs, the health coach program showed more benefit than painTRAINER among males, suggesting that poorer outcomes among males in the overall study analyses for painTRAINER relative to the health coach program were not solely adherence related. No other moderators among the intervention completer sample were seen at 3 or 12 months.
Prior examinations of patient adherence predictors and moderators of CBT-CP-based treatment effectiveness have largely focused exclusively on modifiers of CBT adherence and treatment overall [7;25–27;48] or on whether there are differential conceptually based moderators of pain outcomes for different types of psychosocial interventions.[9] In contrast, our primary focus was pragmatic, namely, might one expect differential pain-related outcomes across different remote modalities of CBT-CP, based on an individual’s baseline demographic and clinical characteristics? With the accelerating use of remote treatment, expansion of online CBT programs, and increasingly tight healthcare resources, the question of when working with a therapist in real-time may be beneficial and when a largely self-directed online program may suffice is important. We found some indication that a largely self-directed online program may be a poorer fit for those with more complex pain conditions and/or depression-related conditions for which addressing deactivation via the tailoring that real-time coaching can provide may enhance patient adherence and outcomes. Yet, what is perhaps most noteworthy about overall study findings, and good news, is that there were few instances in which one of the CBT-CP programs outperformed the other, suggesting that both remote approaches may be helpful. Further, that a largely self-directed online painTRAINER program was as effective as therapist-led CBT-CP for those completing at least 6 of the 8 intervention sessions suggests that connecting patients with this online program and augmenting efforts to keep them engaged could be an important next step for providing low-cost effective CBT-CP services. In this study, those in the online CBT-CP-based group received phone-based technical onboarding support as well as phone outreach in the event that online sessions were not being completed with regular cadence (approximately weekly) from a staff person tasked with addressing barriers to participation. As such, treatment uptake, completion, and outcomes may be substantially poorer for individuals who are only provided access to similar online programs, but not given active support, as may be more routine in many healthcare settings. In fact, the poorer adherence even with such support as provided in our study for painTRAINER suggests that an important next step might be a guided self-help approach with active check-ins with a healthcare staff person, as has been successful for addressing other psychosocial conditions.[14;54]
Strengths and Limitations.
This study conducted a large number of moderator analyses, yet few of the posited moderators show a significant effect on treatment adherence or outcomes, suggesting caution in interpreting findings and the possibility of spurious effects. Thus, future research and replication are critical to determine the reliability of these findings. Finally, while we had some indication that treatment uptake was poorer among those with SDoH needs, as we seek to make these lower resource remote forms of CBT-CP more widely available, understanding this further will be important. Despite these limitations, the study had several strengths including: a very large clinically representative sample (those with high impact chronic pain), prespecified hypotheses associated with most of the proposed moderators, and testing a clinical question of high pragmatic importance to healthcare decision-makers using data readily available to many clinicians who may be interested in referring their patients for CBT-CP.
CONCLUSION
In summary, these analyses suggest that remote CBT-CP approaches, including online programs like painTRAINER or working with a virtual therapist in real-time, show beneficial pain-related outcomes across those with a range of demographic, SDoH-related needs, and clinical characteristics regardless of modality of treatment delivery. While attention to engagement is critical in optimizing outcomes for largely self-directed online programs, our study findings suggest that flexibility in the specific modality for delivering remote CBT-CP, based on patient preference and healthcare system capacity, may enhance broader scalability and patient access to such care.
Supplementary Material
Acknowledgements
Funding/Support:
This research was supported by the National Institutes of Health through the NIH HEAL Initiative (https://heal.nih.gov/) under award number(s) UG3AG067493/UH3AG067493. Research reported in this publication was also supported by the NCATS Trial Innovation Network (TIN) under Award Numbers U24TR001608 (Clinical Coordinating Center), U24TR001597 (Data Coordinating Center), U24TR001579 (Recruitment Innovation Center), and U24TR001609 (Statistical Coordinating Center) as well as the HEAL Pain Management Effectiveness Research Network (ERN) under Award Numbers U24TR004314 (Clinical Coordinating Resource Center), U24004315 (Data Coordinating Resource Center), U24TR004316 (Statistical and Safety Resource Center), and U24TR004317 (Recruitment Resource Center).
Footnotes
Conflicts of Interest: The authors have no conflicts of interest to declare.
Contributor Information
Andrea J. Cook, Kaiser Permanente Washington Health Research Institute, Seattle, WA.
Robert D. Wellman, Kaiser Permanente Washington Health Research Institute, Seattle, WA.
Meghan Mayhew, Kaiser Permanente Center for Health Research, Portland, OR.
Benjamin H. Balderson, Kaiser Permanente Washington Health Research Institute, Seattle, WA.
Morgan Justice, Kaiser Permanente Washington Health Research Institute, Seattle, WA.
Ashli A. Owen-Smith, Georgia State University and Center for Research and Evaluation Kaiser Permanente Georgia, Atlanta, GA.
Francis J. Keefe, Duke Pain Prevention and Treatment Research Program, Department of Psychiatry and Behavioral Sciences, Duke University Medical Center, Durham, NC.
Christine Rini, Department of Medical Social Sciences, Northwestern University Feinberg School of Medicine and Robert H. Lurie Comprehensive Cancer Center of Northwestern University, Chicago, IL.
Michael Von Korff, Kaiser Permanente Washington Health Research Institute, Seattle WA.
Lynn L. DeBar, Kaiser Permanente Center for Health Research, Portland, OR.
Data Sharing Statement:
Deidentified participant data can be accessed via the NIMH Data Archive (http://nda.nih.gov/niaaa) Collection identifier: C56450. Access is for researchers whose proposed use of the data has been approved and has institutional sponsorship. This manuscript reflects the views of the authors and may not reflect the opinions or views of the NIH.
References
- [1].Aaron RV, Ravyts SG, Carnahan ND, Bhattiprolu K, Harte N, McCaulley CC, Vitalicia L, Rogers AB, Wegener ST, Dudeney J. Prevalence of Depression and Anxiety Among Adults With Chronic Pain: A Systematic Review and Meta-Analysis. JAMA Netw Open 2025;8(3):e250268. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [2].Ankawi B, Kerns RD, Edmond SN. Enhancing Motivation for Change in the Management of Chronic Painful Conditions: a Review of Recent Literature. Curr Pain Headache Rep 2019;23(10):75. [DOI] [PubMed] [Google Scholar]
- [3].Asmundson GJ, Taylor S. Role of anxiety sensitivity in pain-related fear and avoidance. J Behav Med 1996;19(6):577–586. [DOI] [PubMed] [Google Scholar]
- [4].Beatty L, Binnion C. A Systematic Review of Predictors of, and Reasons for, Adherence to Online Psychological Interventions. Int J Behav Med 2016;23(6):776–794. [DOI] [PubMed] [Google Scholar]
- [5].Beckers T, Craske MG. Avoidance and decision making in anxiety: An introduction to the special issue. Behav Res Ther 2017;96:1–2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [6].Billlioux A, Verlander K, Anthony S, Alley D. Standardized screening for health-related social needs in clinical settings: the accountable health communities screening tool. NAM Perspect 2017;7(5):1–9. [Google Scholar]
- [7].Broderick JE, Keefe FJ, Schneider S, Junghaenel DU, Bruckenthal P, Schwartz JE, Kaell AT, Caldwell DS, McKee D, Gould E. Cognitive behavioral therapy for chronic pain is effective, but for whom? Pain 2016;157(9):2115–2123. [DOI] [PubMed] [Google Scholar]
- [8].Brown MC, Cushing-Haugen K, Wellman R, Dorsey CN, Norris CM, Adams JL, Steiner JF, Shah AR, Lewis CC. Intersectional Analysis of Social Risk and Need Among a Large Representative Sample of Adults From an Integrated Health System in the U.S. AJPM Focus 2025;4(1):100283. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [9].Burns JW, Jensen MP, Thorn BE, Lillis TA, Carmody J, Gerhart J, Keefe F. Cognitive Therapy, Mindfulness-Based Stress Reduction, and Behavior Therapy for the Treatment of Chronic Pain: Predictors and Moderators of Treatment Response. J Pain 2024;25(6):104460. [DOI] [PubMed] [Google Scholar]
- [10].Burstin H, Clark KJ, Duff N, Dopp AL, Bentley E, Wattenberg S, Sandbrink F, Beale RR, Ling SM, Eaton E, Freiling E, Salman A. Integrating Telehealth and Traditional Care in Chronic Pain Management and Substance Use Disorder Treatment: An Action Agenda for Building the Future State of Hybrid Care. NAM Perspect 2023;2023. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [11].Chen JA, Anderson ML, Cherkin DC, Balderson BH, Cook AJ, Sherman KJ, Turner JA. Moderators and Nonspecific Predictors of Treatment Benefits in a Randomized Trial of Mindfulness-Based Stress Reduction vs Cognitive-Behavioral Therapy vs Usual Care for Chronic Low Back Pain. J Pain 2023;24(2):282–303. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [12].Chew MT, Chan C, Kobayashi S, Cheng HY, Wong TM, Nicholson LL. Online pain management programs for chronic, widespread musculoskeletal conditions: A systematic review with meta-analysis. Pain Pract 2023;23(6):664–683. [DOI] [PubMed] [Google Scholar]
- [13].Cleeland CS, Ryan KM. Pain assessment: global use of the Brief Pain Inventory. Ann Acad Med Singapore 1994;23(2):129–138. [PubMed] [Google Scholar]
- [14].Cuijpers P, Donker T, van Straten A, Li J, Andersson G. Is guided self-help as effective as face-to-face psychotherapy for depression and anxiety disorders? A systematic review and meta-analysis of comparative outcome studies. Psychol Med 2010;40(12):1943–1957. [DOI] [PubMed] [Google Scholar]
- [15].DeBar L, Benes L, Bonifay A, Deyo RA, Elder CR, Keefe FJ, Leo MC, McMullen C, Mayhew M, Owen-Smith A, Smith DH, Trinacty CM, Vollmer WM. Interdisciplinary team-based care for patients with chronic pain on long-term opioid treatment in primary care (PPACT) - Protocol for a pragmatic cluster randomized trial. Contemp Clin Trials 2018;67:91–99. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [16].DeBar L, Mayhew M, Benes L, Bonifay A, Deyo RA, Elder CR, Keefe FJ, Leo MC, McMullen C, Owen-Smith A, Smith DH, Trinacty CM, Vollmer WM. A Primary Care-Based Cognitive Behavioral Therapy Intervention for Long-Term Opioid Users With Chronic Pain : A Randomized Pragmatic Trial. Ann Intern Med 2022;175(1):46–55. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [17].DeBar LL, Mayhew M, Wellman RD, Balderson BH, Dickerson JF, Elder CR, Justice M, Keefe FJ, McMullen CK, Owen-Smith AA, Rini C, Von Korff M, Waring S, Yarava A, Shen Z, Thompson RE, Clark AE, Casper TC, Cook AJ. Telehealth and Online Cognitive Behavioral Therapy-Based Treatments for High-Impact Chronic Pain: A Randomized Clinical Trial. JAMA 2025;334(7):592–605. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [18].Dickerson JF, Clarke G, Weersing VR, Lynch FL, Hollon SD, Brent D, Beardslee W, Gladstone TRG, Porta G, DeBar L, Brooks N, Garber J. Cognitive behavioral program for the prevention of depression in at-risk adolescents: isolating the effects of dose. Am J Epidemiol 2025;194(1):19–26. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [19].Driscoll MA, Edwards RR, Becker WC, Kaptchuk TJ, Kerns RD. Psychological Interventions for the Treatment of Chronic Pain in Adults. Psychol Sci Public Interest 2021;22(2):52–95. [DOI] [PubMed] [Google Scholar]
- [20].Dworkin RH, Turk DC, McDermott MP, Peirce-Sandner S, Burke LB, Cowan P, Farrar JT, Hertz S, Raja SN, Rappaport BA, Rauschkolb C, Sampaio C. Interpreting the clinical importance of group differences in chronic pain clinical trials: IMMPACT recommendations. Pain 2009;146(3):238–244. [DOI] [PubMed] [Google Scholar]
- [21].Feinberg S, Feinberg R, Pohl M, Bokarius V, Darnall B. American Chronic Pain Association Resource Guide to Chronic Pain Management. An Integrated Guide to Medical, Interventional, Behavioral, Pharmacologic and Rehabilitation Therapies: Rocklin, CA: American Chronic Pain Association, Inc, 2018. [Google Scholar]
- [22].Forden G, Ronaghan S, Williams P, Fish S, Ford C. Predictors of treatment outcome in cognitive behavioural therapy for chronic pain: a systematic review. Disabil Rehabil 2024;46(21):4877–4888. [DOI] [PubMed] [Google Scholar]
- [23].Fritz JM, Davis AF, Burgess DJ, Coleman B, Cook C, Farrokhi S, Goertz C, Heapy A, Lisi AJ, McGeary DD, Rhon DI, Taylor SL, Zeliadt S, Kerns RD. Pivoting to virtual delivery for managing chronic pain with nonpharmacological treatments: implications for pragmatic research. Pain 2021;162(6):1591–1596. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [24].Gandy M, Pang STY, Scott AJ, Heriseanu AI, Bisby MA, Dudeney J, Karin E, Titov N, Dear BF. Internet-delivered cognitive and behavioural based interventions for adults with chronic pain: a systematic review and meta-analysis of randomized controlled trials. Pain 2022;163(10):e1041–e1053. [DOI] [PubMed] [Google Scholar]
- [25].Gasslander N, Alfonsson S, Jackalin A, Tengberg C, Håkansson J, Huotari L, Buhrman M. Predictors of adherence to an internet-based cognitive behavioral therapy program for individuals with chronic pain and comorbid psychological distress. BMC Psychol 2021;9(1):156. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [26].Gilpin HR, Keyes A, Stahl DR, Greig R, McCracken LM. Predictors of Treatment Outcome in Contextual Cognitive and Behavioral Therapies for Chronic Pain: A Systematic Review. J Pain 2017;18(10):1153–1164. [DOI] [PubMed] [Google Scholar]
- [27].Glombiewski JA, Hartwich-Tersek J, Rief W. Attrition in cognitive-behavioral treatment of chronic back pain. Clin J Pain 2010;26(7):593–601. [DOI] [PubMed] [Google Scholar]
- [28].Hager ER, Quigg AM, Black MM, Coleman SM, Heeren T, Rose-Jacobs R, Cook JT, Ettinger de Cuba SA, Casey PH, Chilton M, Cutts DB, Meyers AF, Frank DA. Development and validity of a 2-item screen to identify families at risk for food insecurity. Pediatrics 2010;126(1):e26–32. [DOI] [PubMed] [Google Scholar]
- [29].Health Resources & Services Administration. What is a Health Professional Shortage Area (HPSA)?, 2023.
- [30].Heapy AA, Higgins DM, Cervone D, Wandner L, Fenton BT, Kerns RD. A Systematic Review of Technology-assisted Self-Management Interventions for Chronic Pain: Looking Across Treatment Modalities. Clin J Pain 2015;31(6):470–492. [DOI] [PubMed] [Google Scholar]
- [31].Hepburn J, Williams L, McCann L. Barriers to and Facilitators of Digital Health Technology Adoption Among Older Adults With Chronic Diseases: Updated Systematic Review. JMIR Aging 2025;8:e80000. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [32].Jacobson NS, Martell CR, Dimidjian S. Behavioral activation treatment for depression: Returning to contextual roots. Clinical Psychology: Science and Practice 2001;8(3):255–270. [Google Scholar]
- [33].Kanter JW, Manos RC, Bowe WM, Baruch DE, Busch AM, Rusch LC. What is behavioral activation? A review of the empirical literature. Clin Psychol Rev 2010;30(6):608–620. [DOI] [PubMed] [Google Scholar]
- [34].Keller S, Bann CM, Dodd SL, Schein J, Mendoza TR, Cleeland CS. Validity of the brief pain inventory for use in documenting the outcomes of patients with noncancer pain. Clin J Pain 2004;20(5):309–318. [DOI] [PubMed] [Google Scholar]
- [35].Krebs EE, Lorenz KA, Bair MJ, Damush TM, Wu J, Sutherland JM, Asch SM, Kroenke K. Development and initial validation of the PEG, a three-item scale assessing pain intensity and interference. J Gen Intern Med 2009;24(6):733–738. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [36].Kroenke K, Baye F, Lourens SG, Evans E, Weitlauf S, McCalley S, Porter B, Matthias MS, Bair MJ. Automated Self-management (ASM) vs. ASM-Enhanced Collaborative Care for Chronic Pain and Mood Symptoms: the CAMMPS Randomized Clinical Trial. J Gen Intern Med 2019;34(9):1806–1814. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [37].Liang KY, Zeger SL. Longitudinal Data Analysis Using Generalized Linear Models. Biometrika 1986;73(1):13–22. [Google Scholar]
- [38].Macea DD, Gajos K, Daglia Calil YA, Fregni F. The efficacy of Web-based cognitive behavioral interventions for chronic pain: a systematic review and meta-analysis. J Pain 2010;11(10):917–929. [DOI] [PubMed] [Google Scholar]
- [39].Mayhew M, Balderson BH, Cook AJ, Dickerson JF, Elder CR, Firemark AJ, Haller IV, Justice M, Keefe FJ, McMullen CK, O'Keeffe-Rosetti MC, Owen-Smith AA, Rini C, Schneider JL, Von Korff M, Wandner LD, DeBar LL. Comparing the clinical and cost-effectiveness of remote (telehealth and online) cognitive behavioral therapy-based treatments for high-impact chronic pain relative to usual care: study protocol for the RESOLVE multisite randomized control trial. Trials 2023;24(1):196. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [40].Mayhew M, DeBar LL, Deyo RA, Kerns RD, Goulet JL, Brandt CA, Von Korff M. Development and Assessment of a Crosswalk Between ICD-9-CM and ICD-10-CM to Identify Patients with Common Pain Conditions. J Pain 2019;20(12):1429–1445. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [41].Mendoza T, Mayne T, Rublee D, Cleeland C. Reliability and validity of a modified Brief Pain Inventory short form in patients with osteoarthritis. Eur J Pain 2006;10(4):353–361. [DOI] [PubMed] [Google Scholar]
- [42].PRAPARE. The PRAPARE screening tool.
- [43].Radandt NE, Corbridge T, Johnson DB, Kim AS, Scott JM, Coldwell SE. Validation of a Two-Item Food Security Screening Tool in a Dental Setting. J Dent Child (Chic) 2018;85(3):114–119. [PMC free article] [PubMed] [Google Scholar]
- [44].Rini C, Porter LS, Somers TJ, McKee DC, DeVellis RF, Smith M, Winkel G, Ahern DK, Goldman R, Stiller JL, Mariani C, Patterson C, Jordan JM, Caldwell DS, Keefe FJ. Automated Internet-based pain coping skills training to manage osteoarthritis pain: a randomized controlled trial. Pain 2015;156(5):837–848. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [45].Rini C, Porter LS, Somers TJ, McKee DC, Keefe FJ. Retaining critical therapeutic elements of behavioral interventions translated for delivery via the Internet: recommendations and an example using pain coping skills training. Journal of medical Internet research 2014;16(12):e245. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [46].Robinson L, Delgadillo J, Kellett S. The dose-response effect in routinely delivered psychological therapies: A systematic review. Psychother Res 2020;30(1):79–96. [DOI] [PubMed] [Google Scholar]
- [47].Rosser BA, Fisher E, Janjua S, Eccleston C, Keogh E, Duggan G. Psychological therapies delivered remotely for the management of chronic pain (excluding headache) in adults. Cochrane Database Syst Rev 2023;8(8):Cd013863. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [48].Sanabria-Mazo JP, Royuela-Colomer E, Navarrete J, Rodríguez-Freire C, Robles B, McCracken LM, Feliu-Soler A, Luciano JV. Predicting adherence and clinical response of cognitive behavioral therapy among individuals with chronic low back pain plus depressive symptoms: a secondary analysis of a randomized controlled trial. Pain Med 2025;26(8):459–467. [DOI] [PubMed] [Google Scholar]
- [49].Scheidegger A, Gómez Penedo JM, Blättler LT, Aybek S, Bischoff N, Grosse Holtforth M. How Treatment Motivation Predicts Favorable Outcomes in Interdisciplinary Multimodal Pain Treatment Among Patients with Chronic Primary Pain. J Clin Psychol Med Settings 2024;31(1):48–57. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [50].Skelly AC, Brodt ED, Kantner S, Diulio-Nakamura A, Mauer K, Shetty KD. Systematic Review on Noninvasive Nonpharmacological Treatment for Chronic Pain: Surveillance Report 3: Literature Update Period: January 2021 through March 2022. Noninvasive Nonpharmacological Treatment for Chronic Pain: Surveillance Reports. Rockville (MD): Agency for Healthcare Research and Quality (US), 2022. [PubMed] [Google Scholar]
- [51].Tan G, Jensen MP, Thornby JI, Shanti BF. Validation of the Brief Pain Inventory for chronic nonmalignant pain. The Journal of Pain 2004;5(2):133–137. [DOI] [PubMed] [Google Scholar]
- [52].Tauben DJ, Langford DJ, Sturgeon JA, Rundell SD, Towle C, Bockman C, Nicholas M. Optimizing telehealth pain care after COVID-19. Pain 2020;161(11):2437–2445. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [53].Terpstra JA, van der Vaart R, van Beugen S, van Eersel RA, Gkika I, Erdős D, Schmidt J, Radstake C, Kloppenburg M, van Middendorp H, Evers AWM. Guided internet-based cognitive-behavioral therapy for patients with chronic pain: A meta-analytic review. Internet Interv 2022;30:100587. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [54].Traviss-Turner GD, West RM, Hill AJ. Guided Self-help for Eating Disorders: A Systematic Review and Metaregression. Eur Eat Disord Rev 2017;25(3):148–164. [DOI] [PubMed] [Google Scholar]
- [55].U.S. Department of Health and Human Services. Pain Management Best Practices Inter-Agency Task Force Report: Updates, Gaps, Inconsistencies, and Recommendations, 2023.
- [56].Valentín-Sívico J, Canfield C, Low SA, Gollnick C. Evaluating the impact of broadband access and internet use in a small underserved rural community. Telecomm Policy 2023;47(4):102499. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [57].Von Korff M, DeBar LL, Krebs EE, Kerns RD, Deyo RA, Keefe FJ. Graded chronic pain scale revised: mild, bothersome, and high-impact chronic pain. Pain 2020;161(3):651–661. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [58].Wang M, Fitzmaurice GM. A simple imputation method for longitudinal studies with non-ignorable non-responses. Biometrical journal Biometrische Zeitschrift 2006;48(2):302–318. [DOI] [PubMed] [Google Scholar]
- [59].Wilder ME, Kulie P, Jensen C, Levett P, Blanchard J, Dominguez LW, Portela M, Srivastava A, Li Y, McCarthy ML. The Impact of Social Determinants of Health on Medication Adherence: a Systematic Review and Meta-analysis. J Gen Intern Med 2021;36(5):1359–1370. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [60].Williams ACC, Fisher E, Hearn L, Eccleston C. Psychological therapies for the management of chronic pain (excluding headache) in adults. Cochrane Database Syst Rev 2020;8(8):CD007407. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [61].Zeger SL, Liang KY, Albert PS. Models for longitudinal data: a generalized estimating equation approach. Biometrics 1988;44(4):1049–1060. [PubMed] [Google Scholar]
- [62].Zou G A modified poisson regression approach to prospective studies with binary data. Am J Epidemiol 2004;159(7):702–706. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
Deidentified participant data can be accessed via the NIMH Data Archive (http://nda.nih.gov/niaaa) Collection identifier: C56450. Access is for researchers whose proposed use of the data has been approved and has institutional sponsorship. This manuscript reflects the views of the authors and may not reflect the opinions or views of the NIH.