Mid-Treatment Sleep Duration Predicts Clinically Significant Knee Osteoarthritis Pain reduction at 6 months: Effects From a Behavioral Sleep Medicine Clinical Trial

Abstract

Study Objectives:

To determine the relative influence of sleep continuity (sleep efficiency, sleep onset latency, total sleep time [TST], and wake after sleep onset) on clinical pain outcomes within a trial of cognitive behavioral therapy for insomnia (CBT-I) for patients with comorbid knee osteoarthritis and insomnia.

Methods:

Secondary analyses were performed on data from 74 patients with comorbid insomnia and knee osteoarthritis who completed a randomized clinical trial of 8-session multicomponent CBT-I versus an active behavioral desensitization control condition (BD), including a 6-month follow-up assessment. Data used herein include daily diaries of sleep parameters, actigraphy data, and self-report questionnaires administered at specific time points.

Results:

Patients who reported at least 30% improvement in self-reported pain from baseline to 6-month follow-up were considered responders (N = 31). Pain responders and nonresponders did not differ significantly at baseline across any sleep continuity measures. At mid-treatment, only TST predicted pain response via t tests and logistic regression, whereas other measures of sleep continuity were nonsignificant. Recursive partitioning analyses identified a minimum cut-point of 382 min of TST achieved at mid-treatment in order to best predict pain improvements 6-month posttreatment. Actigraphy results followed the same pattern as daily diary-based results.

Conclusions:

Clinically significant pain reductions in response to both CBT-I and BD were optimally predicted by achieving approximately 6.5 hr sleep duration by mid-treatment. Thus, tailoring interventions to increase TST early in treatment may be an effective strategy to promote long-term pain reductions. More comprehensive research on components of behavioral sleep medicine treatments that contribute to pain response is warranted.

Statement of Significance

Insomnia and osteoarthritis are highly comorbid, and for some patients, behavioral sleep treatment results in clinically significant reductions in pain. However, research on short-term, within treatment predictors of posttreatment pain reduction, is limited. We evaluated mid-treatment (4 sessions) sleep continuity predictors of clinical pain improvement 6 months after a CBT-I or BD. Total sleep time (TST) was the best predictor of 6-month posttreatment pain response. Recursive partitioning analyses indicated that individuals who had achieved at least 382 min of TST by mid-treatment were much more likely to be pain responders at 6-month follow-up. Thus, achieving sufficient TST is important in reducing pain, and future research on sleep restriction when compared to other behavioral sleep interventions is warranted.

INTRODUCTION

Knee osteoarthritis (KOA) is a common degenerative arthritic disorder that affects approximately 10%–25% of individuals older than 50 years worldwide.¹ Most patients with KOA have chronic pain, and approximately 30% suffer at least moderate functional disability.¹ In the absence of efficacious disease-modifying medications, a substantial amount of research has investigated putative mechanisms contributing to pain and disability in KOA. Variation in sleep continuity is one potential mechanism, given the high rates of insomnia comorbidity observed among KOA patients^2,3 and the large, and growing, literature base demonstrating that poor sleep predicts increased pain in both experimental and clinical studies.⁴

If poor sleep directly contributes to pain in KOA, it follows that treating sleep should reduce pain. Clinical trials have begun to test this possibility among patients with comorbid KOA and insomnia (KOA-I). Cognitive-behavioral therapy for insomnia (CBT-I) is the gold standard treatment for patients with insomnia, with efficacy rates and durability that meet or exceed those of leading pharmaceutical treatments but with far fewer side effects.^5–7 Although clinical trials have demonstrated that that CBT-I is efficacious for treating insomnia in patients with KOA and other chronic pain disorders, the effects of CBT-I on clinical pain have been mixed (see: Finan and colleagues for a review).⁸ Vitiello and colleagues⁹ compared CBT for pain with a hybrid CBT for both pain and insomnia and an education control. While the hybrid intervention outperformed the other arms on sleep measures, all three arms failed to show substantive reductions in pain severity at postintervention or follow-up. In a subsequent report of subgroup analyses, Vitiello and colleagues¹⁰ collapsed across treatment assignment and demonstrated that patients who evidenced posttreatment improvements in insomnia severity of 30% or greater had modest but statistically significant reductions in pain at 9- and 18-month follow-up compared to those who did not have a clinically significant insomnia response to treatment.

Recently, we showed that CBT-I, relative to a placebo behavioral desensitization active control (BD), was more efficacious for improving wake after sleep onset (WASO), a cardinal sign of sleep maintenance insomnia among patients with KOA-I.³ Both conditions evidenced similar statistically significant reductions in pain severity by postintervention that were maintained at 3- and 6-month follow-up. Notably, posttreatment (8 weeks) changes in WASO were predictive of later reductions in pain. However, despite those gains, 67% of patients did not experience an improvement in pain of 30% or more, a commonly accepted threshold for clinically significant change in pain.^11,12

A possible reason for these results could have been that TST, which is negatively correlated with pain severity,¹³ was slow to change in many patients, especially in the CBT-I condition. Therapeutic techniques could have contributed to the variation in TST change throughout the treatment. For example, CBT-I utilized sleep restriction therapy (SRT), which actively curtails sleep opportunity in the early phases of treatment to achieve sleep consolidation. SRT has been shown to substantially reduce TST in the short term¹⁴ and is associated with a number of negative side effects, such as fatigue, sleepiness, and headaches,¹⁵ which could potentially influence the experience of chronic musculoskeletal pain.

Current Study

Against the backdrop of variable evidence in support of the efficacy of behavioral sleep medicine interventions for pain, it is critical to understand which mechanisms contribute to clinically meaningful reductions in pain. Therefore, we conducted secondary analyses on our recently completed clinical trial³ to determine whether levels of sleep continuity (WASO, TST, sleep onset latency [SOL], and sleep efficiency [SE], both self-report and actigraphy) at mid-treatment (4 weeks) predicted clinically meaningful, longer term change in pain severity (>30%) at 6-month follow-up. Self-reported sleep continuity parameters were our primary variables of interest due to the critical role of subjective symptom report in the diagnosis of insomnia and clinical decision-making in CBT-I and sleep restriction therapy. However, we also secondarily evaluated actigraphy data to corroborate and complement our findings, and as emerging data suggest that patients with insomnia who demonstrate objectively curtailed sleep may be at the highest risk of severe medical comorbidities.¹⁶ Additionally, given that CBT-I and BD had similar effects on pain, revealed through primary analyses,³ we collapsed across treatment conditions and hypothesized that patients who achieved clinically significant reductions in pain by 6-month follow-up would have greater average TST and less WASO at mid-treatment than nonresponders.

We were particularly interested in evaluating the impact of mid-treatment sleep continuity factors (as opposed to posttreatment or follow-up) on pain outcomes, as the earlier that a course of treatment can be altered to address individual differences in response or lack thereof, the better outcomes patients can have and the more cost-effectively treatments can be delivered. This sentiment is reflected across areas of mental/behavioral health intervention, in which researchers have begun to investigate predictors or outcomes of changes in early treatment.^17–19 Additionally, primary results from the clinical trial described herein³ indicated that the majority of gains were achieved by mid-treatment.

METHOD

Participants

100 patients were randomized to one of the two treatment arms and 74 patients returned for a 6-month follow-up visit; see the Consort diagram in the main trial paper³ for details on recruitment.

Procedure

100 patients were randomized to receive either 8 weeks of CBT-I or 8 weeks of BD. At baseline, all patients received in-home polysomnography to rule out the presence of occult sleep disorders and completed approximately 2 weeks of diary assessments of pain and sleep variables. Daily diary assessments were continued throughout the course of the study, and weekly data were aggregated before each visit. Questionnaires were repeated at mid-treatment, posttreatment, and 3- and 6-month follow-up visits. Detailed information about study procedures, inclusion/exclusion criteria, and the treatment conditions can be found in Smith and colleagues.³

Measures

Daily Diary

Patients used electronic personal digital assistants (PDAs) to complete the sleep diaries; all entries were stamped with the time and date. At baseline, entries were made upon waking each morning for approximately 2 weeks. Following randomization, each PDA sleep diary collection period spanned approximately 1 week before each visit. Diary-recorded measures of sleep continuity (SE, SOL, TST, and WASO) were quantified by averaging across days.²⁰ For each assessment period, the number of days available for recording varied slightly among participants due to scheduling/availability. During the baseline assessment, 62% of participants met the goal of 14 days of diary recordings, 81% of participants recorded on at least 10 days, and 96% of participants recorded at least 1 week of data. Comprehensive sleep data were available for 84.2% of 1693 available baseline data points. For the mid-treatment assessment period, 76.3% of participants met the goal of 7 days of diary recordings, and all participants had at least 5 days of data. Comprehensive sleep data were available for 81.1% of 944 available mid-treatment data points. Specific calculations of sleep continuity measures based on PDA recordings are as follows:

• SOL. Time difference (in minutes) between time participant got in bed to go to sleep and time participant fell asleep.

• TST. Time spent asleep (in minutes), calculated based on time participant initially fell asleep and time participant woke up the next morning, subtracting out any time awake during the night.

• WASO. Total time (in minutes) spent awake between time the participant initially fell asleep and time participant woke up for the final time the next morning.

• SE. TST divided by total time spent in bed.

Actigraphy

Patients wore a MiniMitter Actiwatch 2 triaxial accelerometer continuously on the nondominant wrist for 2 weeks²¹ at baseline, mid-treatment, posttreatment, and at 3-month and 6-month follow-ups. For each of these 2-week time periods, data were averaged across days to calculate sleep continuity parameters (SE, SOL, TST, and WASO) according to standardized methods.²²

Western Ontario and McMaster Universities Osteoarthritis Index

The Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) is a validated measure for the assessment of pain, stiffness, and physical function among patients with KOA.²³ Items were rated on a 10-cm visual analog scale, and a total score is made by summing the item scores for each subscale. In this study, only the 5-item pain subscale was used. Pain outcomes were calculated based on percent change in pain score from baseline to 6 months posttreatment.

Analytic Strategy

Data Reduction

Treatment groups were initially compared on WOMAC pain scores at each assessment point to evaluate the possibility of combining groups. Analyses indicated that CBT-I and BD groups did not differ significantly at any time point (all ps > .35); thus, we collapsed across groups for the present analyses. Pain responders were defined as participants who reported at least 30% improvement in clinical pain from baseline to the 6-month follow-up visit, as measured by the WOMAC; 30% or greater is generally considered to be indicative of clinically meaningful improvement.^11,12 We only used participants with 6-month follow-up data in our analyses to enhance the fidelity of the recursive partitioning algorithm. Given that (a) this method of data analysis involves evaluation of patterns present in the data, (b) 26% of participants were lost to 6-month follow-up, and (c) our sample size is small, we chose to use only those participants with 6-month follow-up data for these analyses in order to not artificially bias the results.

Missing Data

We evaluated the potential relationships between dropout and demographic, pain, and sleep continuity variables against a Bonferroni-corrected p value for each set of analyses (corrected values: demographic: p < .008; sleep continuity: p < .013; pain: p < .017). Participants with and without 6-month data did not differ significantly on demographic variables (age, race, sex, employment status, marital status, or BMI), baseline sleep continuity, mid-treatment sleep continuity, or pain (baseline, change from baseline to mid-treatment, or mid-treatment WOMAC-A scores). Based on these analyses, we elected to remove participants without 6-month follow-up data from further analyses, as (a) they did not differ significantly from participants with 6-month data and thus likely do not represent a distinct group and (b) given that the goals were to evaluate mechanisms contributing to clinically meaningful change in pain, we were exclusively interested in observed, rather than extrapolated, data. Of the remaining 74 participants, 31 (42%) were classified as pain responders and 43 (58%) were classified as pain nonresponders.

Data Analysis

To identify measures of sleep continuity that uniquely predicted pain response, we used forward stepwise logistic regressions. Logistic regression tests the extent to which a one-unit increase in the independent variable (TST) is associated with increased/decreased odds of the dependent variable occurring (pain response). Thus, given that a one-unit increase in minutes (i.e., 1 min) is not clinically useful, we converted TST to hours for this analysis. To address the need for empirical determination of mid-treatment CBT-I sleep parameters that predict outcomes, we used binary recursive partitioning, a nonparametric method of computing classification trees using the “Party” package and “ctree” function in R²⁴. This approach is particularly meaningful and applicable to CBT-I, as it produces a numerical cut-point in the independent variable, and CBT-I providers use specific numerical cut-points to make clinical decisions. Further, recursive partitioning is appropriate for this type of research²⁵ and has been used effectively in samples of similar size or smaller.^26–29

More specifically, recursive partitioning first identifies a binary split in the predictor variable (sleep continuity measures) and then repeats this process recursively until there are no more cut-offs, and each group (pain responders vs. nonresponders) is maximally homogeneous. This particular method of classification tree is unique from traditional methods because it requires splits to be significant at Bonferroni-corrected p values (set at p < .05), thereby eliminating the possibility of overfitting and need for post hoc pruning or cross-validation.^24,25 Further, due to the method of selecting the split based on p values, recursive partitioning in “Party” also does not make splits based on missing data, whereas other classification tree programs may be more susceptible to this error.

RESULTS

Demographics

Demographic and clinical characteristics of the full sample are presented in the primary trial outcomes paper,³ and basic demographics for the 74 included participants are presented herein. Participants had a mean age of 59.55 years (SD = 9.88), and 57 (77%) were women. With regard to education, 41.9% had some college, 27% were college graduates, and 31.1% had a graduate degree. Finally, 40.5% were African American and 59.5% were Caucasian. To evaluate potential demographic covariates, pain responders and nonresponders were compared based on age, sex, employment status, race, and marital status. Independent samples t tests did not yield any significant differences, although age trended toward significance, t(72) = 1.90, p = .06, with pain responders being approximately 4 years younger (SDs = 9.7). However, with a Bonferroni correction applied (.05/5), this result is nonsignificant.

Of the 74 patients who provided 6-month follow-up data, 78.4% did not miss any treatment visits, and 16.2% missed one visit (mean visits missed = .31, SD = .72). Pain responders and nonresponders were not significantly different on number of treatment visits missed, t(72) = .86, p = .39. Further, number of treatment visits missed was also not significantly correlated with any of the baseline or mid-treatment sleep continuity measures (all ps > .27).

Finally, 36 (48.6%) participants received CBT-I, while 38 (51.4%) received behavioral desensitization. Of the 31 pain responders, 11 (35.5%) had received CBT-I, while 20 (64.5%) had received behavioral desensitization, a difference that trended toward significance, χ² (1, N = 74) = 3.70, p = .05.

Self-Reported Sleep Continuity Differences in Pain Responders vs. Nonresponders

Analyses were based on a Bonferroni-corrected p value for significance of .013 (.05/4) for baseline and mid-treatment measures. While the Bonferroni correction may be overly conservative, it is the most applicable correction to the t tests used herein. Further, a more moderate correction to the p value for significance (e.g., .02 or .03) would yield the same results. Means, SDs, and results of t tests for pain responder and nonresponder groups at baseline and mid-treatment are presented in Table 1. At baseline, the two response groups did not differ significantly on all four measures of self-reported sleep continuity, including SOL, SE, TST, and WASO (all ps > .10). At mid-treatment, data yielded no significant differences between pain responders and nonresponders in SOL or WASO (ps > .18). However, there was a trend toward significance for pain responders to have greater sleep efficiency than nonresponders, t(70) = −1.84, p = .07. Additionally, pain responders had significantly greater TST at mid-treatment than nonresponders, t(70) = −2.66, p = .01.

Table 1.

Sleep Continuity Characteristics in Pain Responders and Nonresponders at Baseline and Mid-Treatment.

	Baseline						Mid-treatment
	Pain responders		NorResponders		Comparison		Pain responders		Nonresponders		Comparison
	M	SD	M	SD	t	p	M	SD	M	SD	t	p
Sleep efficiency	0.70	0.11	0.67	0.15	−0.99	.33	0.84	0.09	0.79	0.12	−1.84	.07
Sleep onset latency	41.43	27.91	50.37	37.11	1.13	.26	21.96	13.33	26.66	26.77	0.89	.38
Total sleep time	343.34	76.96	313.60	73.59	−1.68	.10	378.86	77.96	334.45	63.49	−2.66	.01
Wake after sleep onset	67.31	38.76	70.77	46.69	0.34	.74	29.60	33.86	41.95	41.54	1.34	.18

Prediction of Pain Response by Self-Reported Sleep Continuity

The relationships between pain response and sleep continuity at baseline and mid-treatment were also evaluated via forward likelihood ratio logistic regression as to most fully evaluate the possible contributions of each of these predictors. Criteria for inclusion was set at p < .05, and criteria for exclusion was set at p > .10. At baseline, none of the four sleep continuity variables met criteria for inclusion. At mid-treatment, TST was the most significant predictor of pain response, B = .56, SE = .23, odds ratio (OR) = 1.74 (95% confidence interval [CI] from 1.12–2.71), p = .01. SE, SOL, and WASO did not contribute significantly to the variability in pain response above and beyond TST. Correlations among sleep continuity variables at mid-treatment are presented in Table 2. While SE is highly correlated with the other three measures of sleep continuity, suggesting colinearity, this is likely because SE is calculated directly from these other variables and is not necessarily a distinct measure. SOL, WASO, and TST all have correlations with one another within the moderate range with the exception of TST and SOL in both the pain responder and the nonresponder groups, and TST and WASO in the nonresponder group. However, TST and WASO were significantly correlated in the pain responder group.

Table 2.

Correlations Among Mid-Treatment Sleep Continuity Characteristics.

	SE	SOL	TST	WASO
SE	—	−0.59**	0.60**	−0.82***
SOL	−0.66***	—	−0.16	0.38*
TST	0.55**	−0.22	—	−0.41*
WASO	−0.79***	0.32*	−0.22	—

Abbreviations: SOL, sleep onset latency; SE, sleep efficiency; TST, total sleep time; WASO, wake after sleep onset.

^a p < .01.

^b p < .001.

^c p < .05.

Correlations among mid-treatment sleep continuity characteristics in participants with 6-month follow-up data. Correlations in pain responders are presented above the diagonal and correlations in non-responders below the diagonal.

Binary Cut-Points of Self-Reported Sleep Continuity Variables

Following regression analyses, we used recursive partitioning (with “ctree” function in the R “Party” package²⁴) to identify cut-points in sleep continuity variables that maximally differentiated pain responders and nonresponders. A cut-point in TST that maximally differentiated pain responders and nonresponders was identified, with a significance of p = .01. As is demonstrated in Figure 1, participants who achieved at least 381.67 min of sleep per night at mid-treatment (n = 25) were significantly more likely to be pain responders at 6 months. Using this cutoff to detect pain response has a sensitivity of 54.8%, a specificity of 81.4%, a positive predictive value of 68%, and a negative predictive value of 71.4%. A chi-square test of independence also indicated that these two categories of TST are significantly different from one another with regard to pain response, χ² (N = 74) = 10.93, p = .001. Recursive partitioning analyses did not identify a significant split for SOL, SE, or WASO in the prediction of pain response.

Figure 1.

Recursive partitioning analyses of total sleep time cut-point for differentiating pain responders and nonresponders.

The cut-point of 381.67 min of TST at mid-treatment also predicted long-term achievement of greater TST at both posttreatment (356.58 min vs. 448.71 min, t(69) = −5.56, p < .001) and 6-month follow-up (357.33 min vs. 439.45 min, t(64) = −6.07, p < .001). Similarly, those who achieved at least 382 min TST by mid-treatment also had significantly greater change in TST from baseline to mid-treatment; 11.05 min vs. 51.79 min, t(70) = −2.55, p = .01. Participants who were randomized to the BD group were significantly more likely than those assigned to CBT-I to have achieved at least 382 min of TST by mid-treatment, χ² (1, N = 72) = 4.23, p = .035; 23% of those in the CBT-I group achieved 382 min of TST by mid-treatment, whereas 46% of the BD group had achieved 382 min TST.

Actigraphy Analyses

Analyses were repeated with actigraphy-derived sleep continuity measures. Generally, actigraphy results corroborated self-report results, with mid-treatment TST being the strongest predictor of 6-month pain response across each of the different analyses. Similar to the diary findings, independent samples t tests yielded no significant differences between pain responders and nonresponders at baseline. At mid-treatment, pain responders evidenced greater SE (p = .01; 72% vs. 62%) and TST (p = .005; 317.3 min vs. 268.1 min). Logistic regression analyses also followed diary data results; TST again emerged as the only significant predictor of pain response (p = .009, OR = 1.87). In the recursive partitioning analyses, participants who achieved less than 241.4 min of actigraphy-derived TST per night at mid treatment were significantly less likely to be pain responders (p = .006). This cutoff had a sensitivity of 96.8%, a specificity of 37.2%, a positive predictive value of 52.6%, and a negative predictive value of 94.1%. Full details of these analyses can be found in the Supplementary Section.

DISCUSSION

In a subsample of 74 patients who provided complete 6-month follow-up data, we observed that self-reported and objective total sleep time (TST) was significantly higher at mid-treatment in individuals classified as pain responders than in nonresponders. Further, other measures of sleep continuity (sleep efficiency, SOL, and WASO) did not explain pain response above and beyond the variance explained by TST. Finally, TST and WASO were significantly correlated in pain responders but not in nonresponders, suggesting that WASO and TST covary in the context of pain reduction.

We evaluated the relationship between pain response and TST using recursive partitioning, an analytic method that identifies a cut-point in the independent variable (TST) which maximally differentiates two groups (i.e., pain responders and nonresponders). Analyses indicated that patients who had achieved a self-reported TST of at least ~382 min by mid-treatment were significantly more likely to be pain responders 6-month posttreatment. While this cut-point was not particularly sensitive (54.8%), it was specific (81.4%), indicating that on average, if ~382 min or more of diary-derived TST is achieved by mid-treatment, the likelihood of experiencing clinically meaningful improvements in pain is greatly increased. The actigraphy data provided highly complementary findings. While the diary TST cut-point (382 min) was quite specific in identifying pain responders, it was not sensitive in detecting those unlikely to improve. The actigraphy-derived cut-point of 241.4 min, however, was highly sensitive (96.8%) in identifying subjects unlikely to report clinically meaning long-term pain improvement but not specific (37.2%).

While we presented actigraphy data alongside daily diary data in the report of primary study outcomes,³ we recently reported that actigraphy may not adequately distinguish good sleeping individuals with KOA from those with KOA-I³⁰. Moreover, standard CBT-I uses sleep diary data rather than objective data to make changes to the prescribed sleep schedule. Future research identifying objective measures of sleep that are valid in this population and may be utilized in clinical decision-making is warranted and may contribute to greater understanding of these relationships and more valid identification of different sleep continuity disturbances.

Despite potential limitations in the validity of actigraphy within this population, the pattern of results did not vary based on differences in measurement of sleep constructs (self-report vs. actigraphy). They suggest that of the sleep continuity measures assessed at mid-treatment, TST was the strongest predictor of clinically meaningful pain response to behavioral sleep medicine interventions, regardless of measurement strategy. They are consistent with past clinical and experimental research, which suggest that insufficient TST is related to worse clinical pain.^31,32 However, they are not consistent with a recent observational study showing that WASO was a stronger predictor of daily pain than TST in a longitudinal cohort study of patients with sickle cell disease.³³ This discrepancy may be explained by differences in study design and assessment. While WASO may have a greater day-to-day influence on pain flares than TST,³³ it appears that in the context of behavioral sleep medicine interventions, patients who are able to quickly increase TST may stand the best chance of experiencing long-term clinical pain benefits. Past research has demonstrated that posttreatment sleep and pain outcomes are predictive of long-term pain and functioning improvements in a subgroup of subjects demonstrating high baseline levels of insomnia and pain.⁹ However, the clinical relevance of behavioral sleep medicine interventions for patients in chronic pain has been unclear because the observed changes in pain were of small magnitude, and the addition of pain-related CBT modules to the standard CBT-I package have not consistently produced additive effects on pain.⁸

By comparison, the present study suggests that mid-treatment (after 4 sessions) TST levels are predictive of clinically significant long-term pain outcomes in patients undergoing either CBT-I or an active control. These data suggest that early identification of sleep continuity response patterns may be warranted to assist practitioners in making individually based treatment adaptations in an effort to maximize the efficacy of behavioral sleep medicine interventions for pain.

Implications for Clinical Practice and Research

Clinically, this research highlights the importance of achieving at least 382 min of sleep per night within 4–6 weeks of initiating treatment to optimize long-term pain outcomes. These findings coincide with research indicating that consistently sleeping <6.5 hr per night (390 min) is detrimental to one’s health^34–36 and that 6.5–8 hr of sleep per night is optimal.^37,38 Strikingly, our analysis empirically determined that 382 min was the optimal cut-point for TST predicting long-term pain response without a priori input or standardization based on baseline demographic or sleep continuity variables. Further, patients who achieved at least 382 min TST at mid-treatment went on to achieve ~449 min TST at posttreatment and ~439 min at 6-month follow-up. Thus, the mid-treatment TST cut-point of 382 min not only predicts long-term pain outcomes, but it also predicts greater TST long term, further supporting the validity of this mid-treatment target. Patients with comorbid chronic pain and insomnia may see greater reductions in clinical pain if treatment emphasizes sleep duration from the start. Such an approach would represent a shift away from the prevailing model of CBT-I, which introduces stimulus control and sleep restriction protocols early in the treatment³⁹ and emphasizes early sleep consolidation as a means to bolster TST in the long term. Sleep restriction therapy procedures, which have been deployed as components in all the clinical trials of CBT-I for insomnia in chronic pain to date, may need to be modified to optimize long-term pain outcomes, particularly in light of our finding that CBT-I produced approximately half as many pain responders than the behavioral desensitization control, which did not include a sleep restriction therapy (SRT) component.

Traditional SRT instructions initially prescribe curtailing sleep opportunity to coincide with diary-derived average total sleep time, with the caveat of prescribing sleep opportunity ≥5.5 hr as a clinical floor to avoid the adverse effects of severe daytime sleepiness. In the current data set, mean baseline TST was 5.43 hr (5.38 for CBT and 5.49 for BD). Thus, nearly half of subjects randomized to CBT-I (47.2%) were prescribed an initial sleep opportunity of 5.5 hr. The goals of sleep restriction therapy are to achieve normal sleep consolidation at the short-term expense of TST. Once subjects demonstrate SE average scores of 90%, then sleep opportunity is increased by 15 min per week. The current data suggest that this prescription may be too restrictive in the context of chronic pain, since under this algorithm, many patients randomized to CBT would need to demonstrate normal sleep efficiency within the first week in order to achieve the possibility of 382 min after 4 sessions.

Several modifications could be made to standard multicomponent CBT-I that would require empirical testing. For example, one strategy could be to lower the sleep efficiency threshold to trigger upward titration of sleep opportunity from 90%–95% to 80% or 85%. This modification might make particular sense in older adults with chronic pain for whom the 90%–95% benchmark may not be realistic. Alternatively, less restrictive forms of sleep curtailment, such as sleep compression⁴⁰ or upward titration of sleep opportunity in 30-min increments,⁴¹ could be tested in patients with chronic pain. Another approach would be to determine whether dropping SRT altogether and focusing more heavily on stimulus control and other CBT components may be more beneficial.

The current analyses did not find other sleep continuity parameters to be predictive of pain improvement above and beyond TST and therefore suggest that approaches that maximize TST may be most beneficial in chronic pain. However, the data do not suggest that sleep consolidation is unimportant. In fact, as described earlier, the significant, negative association between WASO and TST in pain responders suggests that responders tended to have greater TST and less WASO and that these different sleep continuity parameters are more closely related in pain responders than nonresponders. These findings suggest that future studies should consider intervention approaches sensitive to balancing sleep consolidation and total sleep time.

Meta-analytic reviews of CBT-I^42–44 consistently demonstrate that the effects CBT-I on TST are smaller than the effect sizes for other major sleep continuity parameters, especially in the short term. Combined with our data, this body of literature suggests that innovative approaches aimed at improving TST may be warranted, especially when applying CBT-I to the treatment of chronic pain.

Greater understanding of the mechanisms involved in the relationship between sleep and pain may enable clinicians and researchers to tailor interventions to specific groups. Our clinical trial demonstrated that a placebo behavioral desensitization group produced similar reductions in clinical pain to CBT-I³⁰, thereby highlighting the need for further evaluation of the active components of treatment. While emerging literature suggests that stimulus control, sleep restriction, and the combination of stimulus control and sleep restriction are broadly equally efficacious for improving sleep continuity,⁴⁵ researchers are yet to perform dismantling studies evaluating the relative contributions of stimulus control when compared to sleep restriction on pain outcomes. The present results raise the possibility that a stimulus control monotherapy that aims to preserve and even bolster TST may be a better first-line treatment than the combination of stimulus control and sleep restriction for patients with KOA-I. Of course, future studies with varying and possibly adaptive treatment arms will be needed to explicitly evaluate this possibility.

Limitations

Despite its strength for clinical applicability, our analytic strategy did not allow us to control for individual variance due to relevant person-level factors such as demographics or medication usage. We did, however, exclude subjects taking narcotic pain medications, sedative hypnotics, antidepressants, and other psychotropic medications. Further, although analyses indicated that pain response did not differ significantly based on demographic variables, future research that more thoroughly evaluates the impact of individual differences on treatment outcomes is needed. This study included only treatment completers and participants who provided 6-month treatment outcome data. While analyses indicated that these 74 participants were not significantly different from the excluded 26 participants on demographics or baseline variables, this selection of participants may limit generalizability of results to those patients actively committed to treatment.

CONCLUSIONS

Overall, this research supports a body of literature linking sleep continuity to variability in pain and suggests that achieving sufficient TST is particularly important for long-term reductions in clinical pain following behavioral sleep medicine interventions. However, it is important to note that while these findings indicate that retaining TST is key in pain response, given the literature support for other sleep continuity parameters, we cannot conclude at present that TST should not be prioritized at the expense of other aspects of sleep continuity. Future research evaluating the relative contributions of different intervention components to improvements in pain is necessary for greater understanding of the complex relationships between sleep and pain.

SUPPLEMENTARY MATERIAL

Supplementary material is available at SLEEP online.

FUNDING

This work was supported by grant T32 NS070201-11 (MTS); R01 AR05487 (MTS) and grant K23 DA035915 (PHF)

AUTHORS’ NOTE

All work was performed at Johns Hopkins School of Medicine. ClinicalTrials.gov; Sleep in Osteoarthritis Project (SOAP); Identifier: NCT00592449

DISCLOSURE STATEMENT

JS has no conflict of interest. MS is a consultant for Fitbit, Inc. and Pain Care, LLC. He owns an equity stake in BMED interactive. PF is a consultant for PainCare, LLC. He declares no other potential conflicts of interest