Abstract
Background:
Osteoarthritis-related insomnia is the most common form of comorbid insomnia among older Americans. A randomized clinical trial found that six sessions of telephone-delivered cognitive behavioral therapy for insomnia (CBT-I) improved sleep outcomes in this population. Using these data, we evaluated the incremental cost-effectiveness of CBT-I from a healthcare sector perspective.
Methods:
The study was based on 325 community-dwelling older adults with insomnia and osteoarthritis pain enrolled with Kaiser Permanente of Washington State.
We measured quality-adjusted life years (QALYs) using the EuroQol 5-dimension scale. Arthritis-specific quality of life was measured using the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC). Insomnia-specific quality of life was measured using the Insomnia Severity Index (ISI) and nights without clinical insomnia (i.e., “insomnia-free nights”). Total healthcare costs included intervention and healthcare utilization costs.
Results:
Over the 12 months after randomization, CBT-I improved ISI and WOMAC by −2.6 points (95% CI: −2.9 to −2.4) and −2.6 points (95% CI: −3.4 to −1.8), respectively. The ISI improvement translated into 89 additional insomnia-free nights (95% CI: 79 to 98) over the 12 months. CBT-I did not significantly reduce total healthcare costs (−$1072 [95% CI: −$1968 to $92]). Improvements in condition-specific measures were not reflected in QALYs gained (−0.01 [95% CI: −0.01 to 0.01]); at a willingness-to-pay of $150,000 per QALY, CBT-I resulted in a positive net monetary benefit of $369 with substantial uncertainty (95% CI: −$1737 to $2270).
Conclusion:
CBT-I improved sleep and arthritis function without increasing costs. These findings support the consideration of telephone CBT-I for treating insomnia among older adults with comorbid OA. Our findings also suggest potential limitations of the general quality of life measures in assessing interventions designed to improve sleep and arthritis outcomes.
Keywords: cognitive behavioral therapy, cost-effectiveness analysis, insomnia, osteoarthritis, sleep initiation and maintenance disorders
INTRODUCTION
Osteoarthritis (OA)-related insomnia is the most common form of comorbid insomnia, affecting approximately 14 million Americans aged 65 and older.1,2 In addition to exacerbating OA pain and diminishing physical function and overall quality of life,3 insomnia is independently associated with 50% higher healthcare costs.4 With the number of Americans aged 65 and older expected to increase by 17 million (to 73 million) in the next decade,2 these substantial clinical and economic burdens are likely to increase, raising the need to identify cost-effective interventions for OA-related insomnia.
Strong evidence supports the effectiveness of cognitive behavioral therapy for insomnia (CBT-I) in the general insomnia population,5 with CBT-I recognized as the first-line treatment with long-term efficacy and safety. There is also growing evidence demonstrating the effectiveness of CBT-I specifically for OA-related insomnia.6,7 We recently showed that telephone delivery of a brief CBT-I intervention was effective for sustained improvement in OA-related insomnia and fatigue over 12 months.8 The results of our trial were particularly promising for reducing the substantial clinical and economic burdens of OA-related insomnia because most prior trials had relied on longer in-person sessions.6,9
Given high U.S. healthcare spending and competing demands for new services, health plan coverage decisions are guided in part by assessing the cost-effectiveness of interventions.10 For instance, information on the cost-effectiveness of telephone-based interventions for smoking cessation and depression treatment has accelerated their adoption by payers.11–13 However, there are few cost-effectiveness studies of CBT-I, prompting experts to call this knowledge gap a “dramatic need for inclusion of economic endpoints in insomnia treatment studies.”14 Specifically in the OA population, we found no cost-effectiveness evaluations of CBT-I. The objective of this study was to use data from our recently published trial to evaluate the incremental cost-effectiveness of telephone-delivered CBT-I for insomnia-specific, arthritis-specific, and general quality of life outcomes.15 Evidence on the costs and effectiveness of CBT-I can inform coverage and clinical practice.
METHODS
Trial description
Details of the trial protocol and main trial outcomes have been published.8,15 Briefly, our study included adults aged 60 and over who had moderate to severe insomnia and OA pain symptoms on 2 screens, 3 weeks apart, to ensure persistent symptoms. We randomized eligible participants to 6 telephone sessions delivered over 8 weeks. Participants randomized to CBT-I received coaching on sleep restriction, stimulus control, sleep hygiene, and cognitive restructuring, with homework.16,17 Participants randomized to an education-only control (EOC) received information about sleep and OA but did not receive sleep restriction, stimulus control, or cognitive restructuring.18 Both groups kept sleep diaries. The CBT-I versus EOC comparison provides conservative effect estimates of CBT-I, as the nonspecific benefits of CBT-I (i.e., social support, education, behavioral monitoring) are also included in EOC. This study was approved by the University of Washington Institutional Review Board.
Overview of economic evaluation
We compared the costs and effects of CBT-I and EOC, applying a healthcare sector perspective and a 12-month time horizon from the date of randomization. Many CBT-I studies demonstrate positive effects on sleep over 12 months, supporting our time horizon.19 In contrast to a societal perspective, the healthcare sector perspective does not include several cost categories such as transportation or time costs, or changes in work productivity.20 As the intervention was delivered telephonically over a short period with the minimal patient time required, and because most individuals receiving CBT-I were older and less likely to be employed, a healthcare sector perspective is reasonable for this study. However, insomnia is associated with presenteeism (i.e., reduced work productivity).21,22 If CBT-I substantially reduced presenteeism, then our cost estimates would be conservative by not including these productivity gains.19 Our study reporting conforms to Consolidated Health Economic Evaluation Reporting Standards (CHEERS).23
Data collection
We collected investigator-blinded health outcomes at baseline (i.e., immediately before randomization), and 2 and 12 months after randomization. To calculate average values for each health outcome in the 12 months after randomization, we calculated areas under the curve using baseline, 2-month, and 12-month values for each health outcome of interest.24 We obtained healthcare utilization measures from Kaiser Permanente Washington (KPWA) electronic health records (EHR) for individuals in the 12 months before and 12 months after randomization.
We excluded 2 individuals who died for reasons unrelated to the intervention as deemed by an independent safety officer, because potentially high end-of-life costs could skew mean cost estimates. However, as a model check, we describe costs including these individuals.
Health outcomes
We collected information on patient-reported general health-related quality of life using the EuroQol 5-dimension, 5-level health state description (EQ-5D).25 We calculated utility values using preference weights of the EQ-5D health states that were derived from a study that used a U.S. community-based sample.26
As QALY measures may be less sensitive to OA-specific and insomnia-specific treatment benefits,14,27 we also included well-accepted condition-specific measures: the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) and the Insomnia Severity Index (ISI).28 The WOMAC is a 24-item questionnaire assessing levels of difficulty with pain, stiffness, and physical function.29 The WOMAC score is expressed on a percentage scale with 100% being extreme difficulty in all areas. The ISI is a 7-item questionnaire assessing insomnia severity with scores from 0–28, with higher scores indicating worse insomnia.28,30 We constructed an additional insomnia measure—insomnia-free nights—to quantify the number of nights in which a person is insomnia-free, applying methods used to estimate depression-free days.31,32 We transformed the ISI score into a binary value; ISI scores of 7 or less were considered to be insomnia-free30 and were assigned an insomnia-free score of 1, otherwise the insomnia-free score was 0. With this baseline, 2- and 12-month insomnia-free scores, we computed the average insomnia-free score over the 12 months.24 We then multiplied this average insomnia-free score by 365 to estimate the number of insomnia-free nights.
Intervention costs
We calculated the costs of delivering telephonic coaching for CBT-I by multiplying the time spent per session by the maximum number of sessions (6) and by total hourly compensation. The time spent per session to conduct the CBT-I intervention including telephone sessions and pre- and post-call duties was estimated at 45 min. The total hourly compensation was the sum of hourly wages, benefits, and employer-paid payroll taxes. Reflecting the training of coaches participating in the trial, we obtained hourly wage estimates for “Registered Nurses” and “Healthcare Social Workers” in 2019 from the national median hourly wages reported by the U.S. Bureau of Labor Statistics, $35.24 and $27.29 per hour, respectively.33 Benefits were estimated using the percent of total compensation taken as benefits from the 2019 Bureau of Labor Statistics National Compensation Survey for national private industry healthcare and social assistance workers (30.025%).34 We added 6.2% and 1.45% to account for payroll taxes paid by the employer for Social Security and Medicare, respectively, which were not accounted for in the wage estimates.35,36 We did not include overhead costs (e.g., leased space) because those are expected to be negligible for the maximum intervention time of 4.5 hours per patient.
Healthcare utilization costs
A strength of KPWA is that its population tends to be stable, with nearly all care obtained within the KPWA system. Care provided outside the system and billed to KPWA is recorded in claims data. We included the costs of all healthcare utilization from 12 months before and 12 months after randomization, as recorded in the KPWA EHR. We assigned Medicare costs to each unit of healthcare use by applying a previously developed Standardized Relative Resource Cost Algorithm.37 The algorithm comprehensively extracts healthcare utilization data from the KPWA EHR (e.g., procedure codes, Diagnosis-Related Group, National Drug Codes) and assigns costs using the Centers for Medicare and Medicaid Services fee schedules. This algorithm has been applied in multiple studies to produce estimates of healthcare costs using EHR data from Kaiser Permanente.38,39 All costs were inflation-adjusted to 2019 U. S. dollars using the personal consumption expenditure-health index.40
Total healthcare costs
Total healthcare costs used in the cost-effectiveness analyses were the sum of intervention costs and healthcare utilization costs. Costs of providing the EOC sessions were not considered for participants receiving EOC.
Covariates
To produce adjusted estimates of cost and health outcomes, we included healthcare utilization costs in the 12 months before randomization, and baseline age, race, education, comorbidity (using the Charlson index), condition-specific quality of life (ISI, WOMAC), and general health-related quality of life (EQ-5D utility score and EQ-5D visual analog scale score). Additionally, models were adjusted for baseline depression (Patient Health Questionnaire-8 item [PHQ]41), pain (Brief Pain Inventory-short form42), and fatigue (Flinders Fatigue Scale43) as in the main trial analyses.8 We also adjusted for whether individuals used antidepressants, opioids, and/or sedative/hypnotics in the 3 months before randomization.
Analyses—Imputation methods
The trial was powered to detect differences in changes in ISI, a much less variable outcome than healthcare costs.15,44 We used multiple imputation to reduce potential bias related to missing data and to use the full sample of randomized participants for our cost-effectiveness analyses.45 The imputation method used here is similar to the method used as a sensitivity analysis of our published main trial outcomes8 (Supplementary Methods 1 has imputation details). Briefly, we imputed missing data on outcomes, costs, and utilization under the assumption that data are missing at random (MAR) conditional on observed data. Ten imputed datasets were generated separately for each of the two randomization groups using the method of imputation by chained equations.45
Cost imputations were truncated at the maximum observed cost for that category and time period. As a model check to assess the effect of truncation, we also calculated the observed mean difference in total healthcare costs using an imputation model that did not truncate costs. As another model check, we estimated observed mean differences in health outcomes and healthcare utilization costs using only the complete case sample (i.e., excluding individuals who were missing any health outcomes or cost measures due to nonresponse or health plan disenrollment).
Analyses—Modeling cost and effectiveness
We modeled cost and health outcomes separately within each imputed dataset and calculated the adjusted mean differences using the method of recycled predictions.24 A generalized linear model with errors from the gamma distribution with a log link was used to model costs based on goodness of fit.24,46 Multiple linear regression was performed to model the health outcomes: QALYs, ISI, WOMAC, and insomnia-free nights. The bootstrap percentile method with 1000 samples was used to construct 95% confidence intervals for all estimates. To combine bootstrap and imputation samples, bootstrap samples were taken within each imputed dataset and were stratified by the intervention arm. Estimates were combined over the 10 imputed datasets using Rubin’s rules47 to calculate incremental cost-effectiveness for each health outcome of interest.
Our primary outcome was total healthcare costs (i.e., intervention plus healthcare utilization costs) per QALY, reflecting standard practice in cost-effectiveness analysis to assess effectiveness in a common metric. (Supplementary Methods S2 has details) Our secondary outcomes were total healthcare costs per ISI score, per insomnia-free night, and per WOMAC score, reflecting condition-specific outcomes more likely to be influenced by an insomnia intervention. Differences between this cost-effectiveness analysis and the main trial analysis8 include use of the 12-month average outcomes, multiple imputation for the base case analysis, and additional baseline covariates.
Imputations were performed in SAS v9.4 (SAS Institute, Inc) for Microsoft Windows using IVEWare: Imputation and Variance Estimation Software, version 0.3 (University of Michigan). All other analyses were performed using R version 4.0.3 and SAS 9.4.
RESULTS
The study sample included 325 people randomized to CBT-I (n = 162) and EOC (n = 163), after excluding two individuals who died (1 in each arm) post randomization. Response rates for prospectively collected patient-reported health outcomes were 100% at baseline, 86% at 2 months, and 76% at 12 months, and were balanced between intervention and control groups (Supplementary Table S1). In the 12 months following randomization, there was little (5%, n = 15) disenrollment from the health plan (the data source for the healthcare utilization and cost data). The CBT-I group had higher percentages of individuals with non-Hispanic white race and college education but were otherwise well balanced in other demographics, and baseline healthcare utilization and outcome values (Table 1).
TABLE 1.
Baseline characteristics of study participants
| Characteristic | CBT-I n = 162 | EOC n = 163 |
|---|---|---|
| Age, mean (SD) | 70.0 (7.1) | 70.3 (6.5) |
| Female sex, n (%) | 123 (75.9) | 120 (73.6) |
| Race, n (%) | ||
| Non-Hispanic White | 142 (87.7) | 122 (74.9) |
| Non-White | 16 (9.9) | 31 (19.0) |
| Hispanic | 4(2.5) | 10 (6.1) |
| College graduate or higher, n (%) | 86 (53.1) | 72 (44.2) |
| Married/living with partner, n (%) | 107 (66.1) | 108 (66.3) |
| Charlson Comorbidity Index score, n (%) | ||
| 0 | 106 (65.4) | 96 (58.9) |
| 1 | 28 (17.3) | 29 (17.8) |
| ≥2 | 28 (17.3) | 38 (23.3) |
| Any antidepressant use in the 3 months before randomization, n (%) | 23 (14.2) | 28 (17.2) |
| Any opioid use in the 3 months before randomization, n (%) | 34 (21.0) | 29 (17.8) |
| Any sedative/hypnotic use in the 3 months before randomization, n (%) | 8 (4.9) | 10 (6.1) |
| Insomnia Severity Index score, mean (SD) | 15.5 (3.3) | 15.5 (3.3) |
| Brief Pain Inventory-short form | ||
| Interference subscale, mean (SD) | 5.0 (1.8) | 4.8 (1.8) |
| Severity subscale, mean (SD) | 4.7 (1.6) | 4.6 (1.5) |
| Patient Health Questionnaire, depression scale score, mean (SD) | 6.5 (3.7) | 7.2 (3.7) |
| Flinders Fatigue Scale | 14.2 (5.8) | 14.1 (5.0) |
| EQ-5D utility score | 0.64 (0.2) | 0.64 (0.2) |
| EQ VAS score | 66.8 (16.2) | 66.6 (16.0) |
| WOMAC | 42.2 (16.2) | 41.3 (16.2) |
| Total healthcare utilization costs in the 12 months before randomization, mean (SD) | 7729 (10675) | 7898 (9868) |
Abbreviations: CBT-I, cognitive behavioral therapy for insomnia; EOC, education-only control; EQ-5D, EuroQol five dimension, five level; EQ VAS, EuroQol Group visual analogue scale; ISI, Insomnia Severity Index score; PHQ-8, Patient Health Questionnaire, depression scale; WOMAC, Western Ontario and McMaster Universities Osteoarthritis Index.
Health outcomes
Observed health outcomes in our analytic sample improved in both groups post randomization (Table 2). The improvement in adjusted ISI and adjusted WOMAC were greater in the CBT-I group than the EOC group with a mean difference in ISI and WOMAC improvement of −2.6 (95% CI: −2.9 to −2.4) and −2.6 (95% CI: −3.4 to −1.8), respectively. CBT-I provided 89 additional insomnia-free nights (95% CI: 79 to 98) in the 12 months after randomization relative to the EOC group. However, these improvements in condition-specific measures were not reflected in the conventional cost-effectiveness outcome measure (EQ-5D), which showed no statistically significant difference in QALYs.
TABLE 2.
Observed and model-adjusted effectiveness outcomes by treatment arm in the 12 months after randomization
| Effectiveness outcome | CBT-I Mean (95% CI) | EOC Mean (95% CI) | Difference Mean (95% CI) |
|---|---|---|---|
| EQ-5D utility score | |||
| Observed | 0.69 (0.65 to 0.72) | 0.68 (0.65 to 0.71) | 0.01 (−0.04 to 0.05) |
| Adjusted | 0.68 (0.67 to 0.69) | 0.69 (0.68 to 0.69) | −0.01 (−0.02 to 0.01) |
| WOMAC | |||
| Observed | 34.1 (31.7 to 36.5) | 36.6 (33.8 to 39.5) | −2.6 (−6.4 to 1.2) |
| Adjusted | 34.1 (33.5 to 34.7) | 36.7 (35.9 to 37.4) | −2.6 ( 3.4 to −1.8) |
| ISI score | |||
| Observed | 8.6 (8 to 9.3) | 11.3 (10.7 to 12) | −2.7 (−3.6 to −1.8) |
| Adjusted | 8.7 (8.5 to 8.8) | 11.3 (11.1 to 11.5) | −2.6 (−2.9 to −2.4) |
| Total insomnia-free nightsa | |||
| Observed | 176 (150 to 201) | 82 (63 to 101) | 94 (62 to 125) |
| Adjusted | 173 (166 to 180) | 85 (79 to 90) | 89 (79 to 98) |
Abbreviations: CBT-I, cognitive behavioral therapy for insomnia; EOC, education-only control; EQ-5D, EuroQol five dimension, five level; ISI, Insomnia Severity Index score; WOMAC, Western Ontario and McMaster Universities Osteoarthritis Index.
“Insomnia-free” is defined as having ISI score of 7 or less.
Intervention and healthcare utilization costs
Most participants (80%) completed all 6 CBT-I sessions, averaging 2.1 h of actual telephone time per individual in the CBT-I group.8 As a conservative estimate, we used the maximum intervention time of 4.5 hours per individual (45 minutes per session and assuming individuals received 6 sessions), for a total intervention cost of $193.73 (Table 3). Mean observed healthcare utilization costs increased in both groups (Table 3), with nonsignificant trends toward higher costs in the EOC group. These somewhat higher EOC-group costs were partially explained by greater increases in observed costs for joint replacement surgeries, medications, hospitalizations, and other healthcare (Table 4). The adjusted difference in total costs (including costs of delivering CBT-I) between the two groups post randomization was not statistically significant (−$1072 [95% CI: -$1968 to $92]) (Table 3).
TABLE 3.
Cost of delivering CBT-I, observed healthcare utilization costs and adjusted total costs (cost of delivering CBT-I plus healthcare utilization costs) by treatment arm, in the 12 months after randomization
| Observed costs | CBT-I (n = 162) Mean (95% CI) | EOC (n = 163) Mean (95% CI) | Difference Mean (95% CI) |
|---|---|---|---|
| Cost of delivering CBT-I | |||
| Intervention time, hours | 2.1 | – | 2.1 |
| Intervention costs, $ | 193.70 | – | 193.70 |
| Observed healthcare utilization costs, $a | 9395 (7243 to 11,548) | 11,312 (8350 to 14,274) | −1916 (−5567 to 1734) |
| Adjusted total cost, $a | 10,035 (9431 to 10,892) | 11,107 (10,182 to 11,942) | −1072 (−1968 to 92) |
Abbreviations: CBT-I, cognitive behavioral therapy for insomnia; EOC, education-only control.
Based on imputed dataset.
TABLE 4.
Observed healthcare utilization costs by category, treatment arm, and period in the 12 months before and after randomization
| Observed costs | CBT-I (n = 162) Mean (95% CI) | EOC (n = 163) Mean (95% CI) | Difference Mean (95% CI) |
|---|---|---|---|
| Medication costs, $ | |||
| Before randomization | 2457 (1828 to 3086) | 2687 (2210 to 3164) | −230 (−1016 to 556) |
| After randomization | 2285 (1628 to 2941) | 2920 (2034 to 3806) | −636 (−1685 to 413) |
| Ambulatory visits, $ | |||
| Before randomization | 2420 (2114 to 2725) | 2353 (2063 to 2643) | 66 (−353 to 486) |
| After randomization | 3194 (2629 to 3759) | 2839 (2424 to 3255) | 355 (−350 to 1060) |
| Hospitalizations, $ | |||
| Before randomization | 1051 (36 to 2066) | 1190 (8 to 2373) | −139 (−1693 to 1414) |
| After randomization | 748 (−340 to 1837) | 1149 (412 to 1886) | −401 (−1708 to 907) |
| Other healthcare costs, $a | |||
| Before randomization | 1801 (1025 to 2578) | 1668 (1256 to 2080) | 134 (−741 to 1008) |
| After randomization | 2277 (1543 to 3011) | 2762 (897 to 4627) | −485 (−2516 to 1545) |
| Joint replacement surgeries, $ | |||
| Before randomization | 0 (0 to 0) | 0 (0 to 0) | 0 (0 to 0) |
| After randomization | 892 (222 to 1561) | 1641 (438 to 2843) | −749 (−2123 to 624) |
Abbreviations: CBT-I, cognitive behavioral therapy for insomnia; EOC, education-only control.
Sum of costs for: skilled nursing facility, hospital ambulatory care, home health, hospice, dialysis, ambulatory surgery, durable medical equipment, other institutional care, and chemotherapy.
Cost-effectiveness
The point estimates for costs and QALYs are both negative. (Figure 1A) CBT-I had 95% or greater probability of being cost-effective compared with EOC at a willingness-to-pay per QALY threshold of $8137/QALY or less; but this probability decreases as a healthcare payer’s willingness to pay for a QALY increases (Supplementary Figure S1; Supplementary - Table S2). Presented alternatively, at a willingness-to-pay of $150,000 per QALY, CBT-I provides a positive net monetary benefit of $369 (i.e., is cost-effective) with substantial uncertainty (95% CI: -$1737 to $2270). Our choice of the $150,000/QALY threshold makes CBT-I appear less cost-effective than if we had chosen a lower threshold such as $100,000/QALY (Supplementary Methods S2 has details).
FIGURE 1.
Incremental cost-effectiveness planes of cognitive behavioral therapy for insomnia (CBT-I) compared with education-only control (EOC) with 95% confidence ellipse. The point estimate is red and each of the 1000 bootstrap replicates are black. Panel A. Effectiveness measured by quality-adjusted life years (QALYs). The point estimate is in the South West quadrant, meaning CBT-I produces fewer QALYs (−0.005) and reduces costs (−$1072) Panel B. Effectiveness measured by Insomnia Severity Index (ISI). The point estimate is South East quadrant, meaning CBT-I reduces insomnia (−2.6 ISI) and costs (−$1072) Panel C. Effectiveness measured by Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC). The point estimate is in the South East quadrant, meaning CBT-I reduces arthritis-related functional limitations (−2.6 WOMAC) and reduces costs (−$1072) Panel D. Effectiveness measured by insomnia-free nights (i.e., nights without clinical insomnia). The point estimate is South East quadrant, meaning CBT-I increases insomnia-free nights (88.6 nights) and reduces costs (−$1072)
In the analyses of condition-specific outcomes, CBT-I improved condition-specific quality of life whereas not increasing mean costs (Figures 1B–D). Consequently, at any positive willingness-to-pay level (i.e., if decision makers are willing to pay anything at or above $0) for an improvement in ISI, insomnia-free nights, or WOMAC, there is a greater than 95% probability of CBT-I being cost-effective than EOC. Viewed another way, setting aside any possible reductions in healthcare costs that might be attributable to CBT-I and considering only the added costs of delivering CBT-I, the average cost per additional insomnia-free night was $2.18 (estimated intervention cost of $193.73 divided by 89 nights with insomnia averted).
Model checks
The observed health outcomes in the complete case sample were similar to the analytic sample used for our primary and secondary analyses (Supplementary Table S3). The observed mean difference in healthcare utilization costs (1) in the complete case sample, (2) in the sample including the 2 individuals who died, and (3) with the imputation that did not cap costs at observed maximums for each healthcare utilization category were all higher than the observed mean difference in total healthcare costs in our analytic sample (−$2208, −$2876, and −$2503 respectively, compared with −$1916) (Supplementary Table S4 compared with Table 3). These results suggest that our cost savings estimates are conservative (i.e., we may be underestimating the actual cost savings of CBT-I).
DISCUSSION
In this study, we found that CBT-I for older adults with concomitant insomnia and OA pain not only improved sleep as previously reported,15 but it also improved arthritis function. Thus, CBT-I significantly improved condition-specific outcomes and did not increase total healthcare costs (including costs of delivering CBT-I). However, there were no statistically significant changes in QALYs.
Following conventional practice, we prespecified cost per QALY as the primary outcome. Advantages of QALYs include allowing for direct comparisons of health outcome changes across broad therapeutic areas and accepted ranges of willingness-to-pay per QALY, which supports statements about whether the treatment represents good value. In contrast, there is no commonly accepted willingness to pay thresholds for condition-specific measures. However, researchers have noted the limitations of generic quality of life instruments for assessing the health benefits of improved sleep, functional status related to comorbid conditions, and enhanced psychological well-being.14,27 Consistent with our findings, a randomized trial of CBT-I conducted in the UK found a large (3.9 point) improvement in ISI without improving QALYs.48 Yet, the Institute of Medicine has noted the “enormous public health burden” of sleep disorders, and others have called for sleep assessment to be considered an “additional vital sign” that providers should routinely perform.48,49
This is the first economic evaluation of the cost-effectiveness of CBT-I among patients with OA and one of only a few well-controlled studies on the cost-effectiveness of CBT-I overall. Prior cost-effectiveness studies based on CBT-I clinical trials that were not restricted to the OA population found small, nonsignificant improvements in QALYs. Some found that CBT-I reduced healthcare costs whereas others found CBT-I increased healthcare costs. We found essentially no change in QALYs. Although we found a potentially meaningful 11% reduction in total healthcare costs (and larger reductions based on our model checks), we caution against overinterpretation because the finding was nonsignificant. Future studies with larger sample sizes and sufficient power to detect changes in costs44 are needed to adequately assess effects of CBT-I on costs, because that is an important outcome for healthcare payers.
Limitations
Our study has limitations. First, we measured costs and health outcome changes over 12-months, whereas the health benefits of CBT-I may continue for at least 3 years after intervention.50 It would be ideal to measure longer-term changes in costs and health outcomes.51 Nevertheless, our study represents an advance in understanding the longer-term effects of CBT-I on costs and health outcomes because prior economic studies used even shorter time horizons (8 weeks to 6 months).52
Second, the use of an EOC group resulted in an underestimate of CBT-I effects. To calculate the incremental effectiveness of CBT-I, we subtracted the health outcomes experienced by the EOC group, which also improved. A potential reason for the improvement in the EOC group is nonspecific effects such as therapist attention and patient expectations that might have led to improved quality of life. Such effects are arguably part of the effect of any healthcare intervention but in our study were controlled by the use of an EOC group.18 Therefore, to the degree that nonspecific effects improved quality of life in study patients, our incremental effectiveness estimates of CBT-I are conservative (i.e., shifted toward the null).
Third, when we calculated the incremental cost of CBT-I, we did not include the costs for delivering EOC. We also used the maximum number of CBT-I sessions rather than the observed number of sessions delivered. Further, we calculated the costs of CBT-I based upon the expertise of coaches who delivered the intervention in the clinical trial. The costs of delivering CBT-I by organizations routinely offering such services could be lower due to economies of scale and the use of lower-cost personnel. These considerations would make our cost estimates more conservative, although intervention costs were a small proportion of total healthcare costs in our study.
Fourth, a limitation of our novel secondary measure, “insomnia free night” is that (just like “depression-free days,” the PHQ, and the ISI) it is based on a 2-week recall period collected at 3 time points.31,32 However, we previously reported that there was substantial stability in insomnia remission throughout this study,8 providing more confidence in our use of the insomnia-free night as an additional secondary measure.
Finally, there was some imbalance in race and education between CBT-I and EOC groups. Although race and education may be associated with some of the outcomes, all our regression models adjusted for baseline outcome values as well as race and education as covariates.
Conclusion
CBT-I is a recognized first-line treatment for chronic insomnia in the general insomnia population. CBT-I delivered by phone likely increases access and scalability of providing CBT-I. To our knowledge, this is the first study to report on the cost-effectiveness of CBT-I among older adults with OA, a clinically important and costly patient population with comorbid insomnia. We found that at typical willingness-to-pay thresholds per QALY, CBT-I did not provide a statistically significant net monetary benefit. However, we have reason to believe that the conventional QALY measure was insensitive to CBT-I benefits that matter to patients, such as improved sleep and improved arthritis function. Indeed, CBT-I improved sleep and arthritis function (using condition-specific measures) whereas total healthcare costs were not significantly affected.
Ignoring potential cost savings of CBT-I and considering only intervention costs, we conservatively estimate that the added cost of gaining an additional insomnia-free night was about $2. Our findings on sleep and arthritis function support consideration of telephone CBT-I as an option for treating insomnia among older adults with co-morbid OA.
Supplementary Material
Appendix S1. Supporting Information.
Supplementary Methods S1. Multiple imputation details.
Supplementary Methods S2. Definition, calculation, and interpretation of three terms: incremental cost-effectiveness ratios, willingness to pay, and net monetary benefit.
Supplementary Table S1. Number and percentage of individuals who had responded to patient-reported health outcome surveys at a given time point or disenrolled from health plan.
Supplementary Table S2. Percent of estimates from the 1000 bootstrap resamples that fall into each quadrant of the incremental cost-effectiveness plane shown in Figure 1 of the manuscript.
Supplemental Table S3. Observed effectiveness outcomes by treatment arm and period (baseline and over the 12 months after randomization) in the complete case sample (n = 236).
Supplemental Table S4. Observed mean healthcare utilization costs over the 12 months after randomization.
Supplemental Figure S1. Cost-effectiveness acceptability curve of CBT-I for quality-adjusted life years (derived from the as measured by the EuroQol five dimension) in the 12 months after randomization.
Key Points.
Phone-based cognitive behavioral therapy for insomnia (CBT-I) improved sleep and arthritis function.
The intervention costs $194 to deliver but did not increase total healthcare costs (intervention costs plus healthcare utilization costs).
Why Does this Paper Matter?
Phone CBT-I is efficacious for sleep- and arthritis-specific outcomes and is low cost.
ACKNOWLEDGMENTS
We thank Kenneth Pike, University of Washington, and Malia Oliver, Kaiser Permanente Washington Health Research Institute, for data management and programming support in producing the analytic dataset.
SPONSOR’S ROLE
This study was supported by a grant from the National Institute on Aging (R01AG053221). The funder had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Funding information
National Institute on Aging, Grant/Award Number: R01AG053221
Charles Morin received research support from Eisai, Idorsia, Canopy Health, and Lallemand Health Solutions; and served on advisory boards for Eisai, Merck, Pear Therapeutics, and Sunovion.
Footnotes
SUPPORTING INFORMATION
Additional supporting information may be found in the online version of the article at the publisher’s website.
CONFLICT OF INTEREST
The authors have no conflicts.
REFERENCES
- 1.Foley D, Ancoli-Israel S, Britz P, Walsh J. Sleep disturbances and chronic disease in older adults: results of the 2003 National Sleep Foundation Sleep in America Survey. J Psychosom Res. 2004;56(5):497–502. [DOI] [PubMed] [Google Scholar]
- 2.Vespa J, Medina L, Armstrong DM. Demographic Turning Points for the United States: Population Projections for 2020 to 2060. United States Census Bureau; 2020. [Google Scholar]
- 3.Abad VC, Sarinas PS, Guilleminault C. Sleep and rheumatologic disorders. Sleep Med Rev. 2008;12(3):211–228. [DOI] [PubMed] [Google Scholar]
- 4.Anderson LH, Whitebird RR, Schultz J, McEvoy CE, Kreitzer MJ, Gross CR. Healthcare utilization and costs in persons with insomnia in a managed care population. Am J Manag Care. 2014;20(5):e157–e165. [PubMed] [Google Scholar]
- 5.Mitchell MD, Gehrman P, Perlis M, Umscheid CA. Comparative effectiveness of cognitive behavioral therapy for insomnia: a systematic review. BMC Fam Pract. 2012;13:40. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Vitiello MV, McCurry SM, Shortreed SM, et al. Cognitive-behavioral treatment for comorbid insomnia and osteoarthritis pain in primary care: the lifestyles randomized controlled trial. J Am Geriatr Soc. 2013;61(6):947–956. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Smith MT, Finan PH, Buenaver LF, et al. Cognitive-behavioral therapy for insomnia in knee osteoarthritis: a randomized, double-blind, active placebo-controlled clinical trial. Arthritis Rheumatol. 2015;67(5):1221–1233. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.McCurry SM, Zhu W, Von Korff M, et al. Effect of telephone cognitive behavioral therapy for insomnia in older adults with osteoarthritis pain: a randomized clinical trial. JAMA Intern Med. 2021;181(4):530–538. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.McCurry SM, Shortreed SM, Von Korff M, et al. Who benefits from CBT for insomnia in primary care? Important patient selection and trial design lessons from longitudinal results of the Lifestyles trial. Sleep. 2014;37(2):299–308. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Wickwire EM, Shaya FT, Scharf SM. Health economics of insomnia treatments: the return on investment for a good night’s sleep. Sleep Med Rev. 2016;30:72–82. [DOI] [PubMed] [Google Scholar]
- 11.Curry SJ, Grothaus LC, McAfee T, Pabiniak C. Use and cost effectiveness of smoking-cessation services under four insurance plans in a health maintenance organization. N Engl J Med. 1998;339(10):673–679. [DOI] [PubMed] [Google Scholar]
- 12.Simon GE, Ludman EJ, Rutter CM. Incremental benefit and cost of telephone care management and telephone psychotherapy for depression in primary care. Arch Gen Psychiatry. 2009; 66(10):1081–1089. [DOI] [PubMed] [Google Scholar]
- 13.Quit For Life Program. Accessed February 10, 2021. https://www.optum.com/business/solutions/population-health/prevention/smoking-cessation.html
- 14.Wickwire EM, Morin CM. Advancing a value framework for sleep: update on economic aspects of cognitive-behavioral treatments for insomnia. Sleep Med Rev. 2020;54:101387. [DOI] [PubMed] [Google Scholar]
- 15.McCurry SM, Von Korff M, Morin CM, et al. Telephone interventions for co-morbid insomnia and osteoarthritis pain: The OsteoArthritis and Therapy for Sleep (OATS) randomized trial design. Contemp Clin Trials. 2019;87:105851. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.McCurry SM, Guthrie KA, Morin CM, et al. Telephone-based cognitive behavioral therapy for insomnia in perimenopausal and postmenopausal women with vasomotor symptoms: a MsFLASH randomized clinical trial. JAMA Intern Med. 2016; 176(7):913–920. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Bastien CH, Morin CM, Ouellet MC, Blais FC, Bouchard S. Cognitive-behavioral therapy for insomnia: comparison of individual therapy, group therapy, and telephone consultations. J Consult Clin Psychol. 2004;72(4):653–659. [DOI] [PubMed] [Google Scholar]
- 18.Balderson BH, McCurry SM, Vitiello MV, et al. Information without implementation: a practical example for developing a best practice education control group. Behav Sleep Med. 2016; 14(5):514–527. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Geiger-Brown JM, Rogers VE, Liu W, Ludeman EM, Downton KD, Diaz-Abad M. Cognitive behavioral therapy in persons with comorbid insomnia: a meta-analysis. Sleep Med Rev. 2015;23:54–67. [DOI] [PubMed] [Google Scholar]
- 20.Neumann PJ, Sanders GD, Russell LB, Siegel JE, Ganiats TG. Cost-Effectiveness in Health and Medicine. Oxford University Press; 2016. [Google Scholar]
- 21.Bolge SC, Doan JF, Kannan H, Baran RW. Association of insomnia with quality of life, work productivity, and activity impairment. Qual Life Res. 2009;18(4):415–422. [DOI] [PubMed] [Google Scholar]
- 22.Bostock S, Luik AI, Espie CA. Sleep and productivity benefits of digital cognitive behavioral therapy for insomnia: a randomized controlled trial conducted in the workplace environment. J Occup Environ Med. 2016;58(7):683–689. [DOI] [PubMed] [Google Scholar]
- 23.Husereau D, Drummond M, Petrou S, et al. Consolidated Health Economic Evaluation Reporting Standards (CHEERS) statement. BMJ. 2013;346:f1049. [DOI] [PubMed] [Google Scholar]
- 24.Glick HA, Doshi JA, Sonnad SS, Polsky D. Economic Evaluation in Clinical Trials. 1st ed. Oxford University Press; 2007. [Google Scholar]
- 25.EuroQol G. EuroQol–a new facility for the measurement of health-related quality of life. Health Policy. 1990;16(3): 199–208. [DOI] [PubMed] [Google Scholar]
- 26.Pickard AS, Law EH, Jiang R, et al. United States Valuation of EQ-5D-5L Health States Using an International Protocol. Value Health. 2019;22(8):931–941. [DOI] [PubMed] [Google Scholar]
- 27.Knapp M, Mangalore R. The trouble with QALYs…. Epidemiol Psichiatr Soc. 2007;16(4):289–293. [DOI] [PubMed] [Google Scholar]
- 28.Bastien CH, Vallieres A, Morin CM. Validation of the Insomnia Severity Index as an outcome measure for insomnia research. Sleep Med. 2001;2(4):297–307. [DOI] [PubMed] [Google Scholar]
- 29.WOMAC Osteoarthritis Index. WOMAC 3.1 Index. 2012. Accessed April 21, 2021. https://www.womac.com/services
- 30.Morin CM, Belleville G, Belanger L, Ivers H. The Insomnia Severity Index: psychometric indicators to detect insomnia cases and evaluate treatment response. Sleep. 2011;34(5): 601–608. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Vannoy SD, Arean P, Unutzer J. Advantages of using estimated depression-free days for evaluating treatment efficacy. Psychiatr Serv. 2010;61(2):160–163. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Lave JR, Frank RG, Schulberg HC, Kamlet MS. Cost-effectiveness of treatments for major depression in primary care practice. Arch Gen Psychiatry. 1998;55(7):645–651. [DOI] [PubMed] [Google Scholar]
- 33.U.S. Bureau of Labor Statistics. May 2019 National Occupational Employment and Wage Estimates. Occupational Employment Statistics; 2019. Accessed February 10, 2021. https://www.bls.gov/oes/current/oes_nat.htm [Google Scholar]
- 34.U.S. Bureau of Labor Statistics. NCS Databases. National Compensation Survey. Accessed February 10, 2021. https://www.bls.gov/ncs/data.htm [Google Scholar]
- 35.26 U.S. Code § 3121 - Definitions. Accessed January 3, 2019. https://www.law.cornell.edu/uscode/text/26/3121
- 36.Tax on Self-Employment Income, 1402. Definitions. 26 U.S.C.; 2011. [Google Scholar]
- 37.O’Keeffe-Rosetti MC, Hornbrook MC, Fishman PA, et al. A standardized relative resource cost model for medical care: application to cancer control programs. J Natl Cancer Inst Monogr. 2013;46:106–116. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Tabano DC, Anderson ML, Ritzwoller DP, et al. Estimating the impact of diabetes mellitus on worker productivity using self-report, electronic health record and human resource data. J Occup Environ Med. 2018;60(11):e569–e574. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Ritzwoller DP, Fishman PA, Banegas MP, et al. Medical care costs for recurrent versus De Novo stage IV cancer by age at diagnosis. Health Serv Res. 2018;53(6):5106–5128. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Dunn A, Grosse SD, Zuvekas SH. Adjusting health expenditures for inflation: a review of measures for health services research in the United States. Health Serv Res. 2018;53(1):175–196. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41.Kroenke K, Strine TW, Spitzer RL, Williams JB, Berry JT, Mokdad AH. The PHQ-8 as a measure of current depression in the general population. J Affect Disord. 2009;114(1–3):163–173. [DOI] [PubMed] [Google Scholar]
- 42.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]
- 43.Gradisar M, Lack L, Richards H, et al. The Flinders Fatigue Scale: preliminary psychometric properties and clinical sensitivity of a new scale for measuring daytime fatigue associated with insomnia. J Clin Sleep Med. 2007;3(7):722–728. [PMC free article] [PubMed] [Google Scholar]
- 44.Briggs A. Economic evaluation and clinical trials: size matters. BMJ. 2000;321(7273):1362–1363. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 45.Raghunathan TE, Lepkowski JM, Van Hoewyk J, Solenberger P. A multivariate technique for multiply imputing missing values using a sequence of regression models. Surv Methodol. 2001;27(1):85–96. [Google Scholar]
- 46.Pregibon D. Goodness of link tests for generalized linear models. J Appl Stat. 1980;29(1):15. [Google Scholar]
- 47.Rubin DB. Multiple imputation for nonresponse in surveys. John Wiley & Sons; 2004. [Google Scholar]
- 48.Grandner MA, Malhotra A. Sleep as a vital sign: why medical practitioners need to routinely ask their patients about sleep. Sleep Health. 2015;1(1):11–12. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 49.Ramar K, Malhotra RK, Carden KA, et al. Sleep is essential to health: an American Academy of Sleep Medicine position statement. J Clin Sleep Med. 2021. 10.5664/jcsm.9476 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 50.Backhaus J, Hohagen F, Voderholzer U, Riemann D. Long-term effectiveness of a short-term cognitive-behavioral group treatment for primary insomnia. Eur Arch Psychiatry Clin Neurosci. 2001;251(1):35–41. [DOI] [PubMed] [Google Scholar]
- 51.Natsky AN, Vakulin A, Chai-Coetzer CL, et al. Economic evaluation of cognitive behavioural therapy for insomnia (CBT-I) for improving health outcomes in adult populations: a systematic review. Sleep Med Rev. 2020;54:101351. [DOI] [PubMed] [Google Scholar]
- 52.Watanabe N, Furukawa TA, Shimodera S, et al. Cost-effectiveness of cognitive behavioral therapy for insomnia comorbid with depression: analysis of a randomized controlled trial. Psychiatry Clin Neurosci. 2015;69(6):335–343. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Appendix S1. Supporting Information.
Supplementary Methods S1. Multiple imputation details.
Supplementary Methods S2. Definition, calculation, and interpretation of three terms: incremental cost-effectiveness ratios, willingness to pay, and net monetary benefit.
Supplementary Table S1. Number and percentage of individuals who had responded to patient-reported health outcome surveys at a given time point or disenrolled from health plan.
Supplementary Table S2. Percent of estimates from the 1000 bootstrap resamples that fall into each quadrant of the incremental cost-effectiveness plane shown in Figure 1 of the manuscript.
Supplemental Table S3. Observed effectiveness outcomes by treatment arm and period (baseline and over the 12 months after randomization) in the complete case sample (n = 236).
Supplemental Table S4. Observed mean healthcare utilization costs over the 12 months after randomization.
Supplemental Figure S1. Cost-effectiveness acceptability curve of CBT-I for quality-adjusted life years (derived from the as measured by the EuroQol five dimension) in the 12 months after randomization.