Effectiveness of mobile-delivered, therapist-assisted cognitive behavioral therapy for insomnia in nationwide routine clinical care in Finland

Abstract

Study Objectives:

Internet-delivered cognitive behavioral therapies (iCBTs) are efficacious for insomnia. Few studies have as yet reported their effectiveness in routine care. The objective of this study was to examine the effectiveness of the new Finnish 7-session HUS Helsinki University Hospital–iCBT for Insomnia (HUS-iCBTI) program in nationwide routine care.

Methods:

We studied 2,464 physician-referred patients in therapist-supported iCBTI. Treatment was free of charge for patients. The primary measure was the Insomnia Severity Index (ISI); the Patient Health Questionnaire, 4-item version (PHQ-4), and the Alcohol Use Disorders Identification Test (AUDIT-C) were also administered.

Results:

As many as 75.4% completed the treatment, with the ISI scores declining, on average, by –7.04 points during treatment. Altogether, 91.1% of the completers improved in terms of having a lower ISI score in phase 7 than in phase 1 and 34.0% of the completers achieved remission by phase 7 (ISI score < 8); 46.0% responded to treatment (ISI-score reduction of ≥ 8 points from phase 1). The Cohen’s effect size (ISI-score decline) was d = –1.449, showing a large effect.

Conclusions:

We observed large benefits for this publicly funded therapist-supported treatment in its naturalistic setting as a part of routine care. Effect sizes were comparable to randomized controlled trials reported earlier. HUS-iCBTI appears to be effective as a treatment for insomnia and may also reduce the risk of other adverse health consequences associated with poor sleep.

Citation:

Stenberg J-H, Ritola V, Joffe G, Saarni S, Rosenström T. Effectiveness of mobile-delivered, therapist-assisted cognitive behavioral therapy for insomnia in nationwide routine clinical care in Finland. J Clin Sleep Med. 2022;18(11):2643–2651.

BRIEF SUMMARY

Current Knowledge/Study Rationale: Internet-delivered cognitive behavioral therapies are efficacious for insomnia but only a few studies have reported their effectiveness in routine care. We observed large benefits for our free-of-charge, therapist-supported therapy for insomnia in a naturalistic setting as a part of national routine care.

Study Impact: As many as 75.4% completed the treatment, with the Insomnia Severity Index (ISI) scores declining, on average, by –7.04 points. Altogether, 91.1% of the completers improved in terms of a reduced ISI score after treatment. As many as 34.0% of the completers achieved remission after treatment (ISI score < 8); 46.0% responded to treatment (ISI-score reduction of ≥ 8 points from the beginning of treatment). An effect size d = –1.449 is comparable to the randomized controlled trials reported.

INTRODUCTION

Sleep problems and symptoms of insomnia are common: approximately 25% of the population are dissatisfied with their sleep and 6–10% meet the criteria for insomnia disorder.¹ Furthermore, recent evidence suggests that symptoms of insomnia may also be linked to major depressive disorder, bipolar disorder, autism-spectrum disorders, and even coronary artery disease.^2–6 Thus, early and efficient treatment of sleep problems could pay large dividends in terms of public health. Incentives are considerable to achieve safe, rapidly delivered, and cost-efficient treatment.

In most cases of insomnia, the best treatment may be cognitive behavioral therapy (CBT) for insomnia (CBTI),^7–9 but training psychotherapists is costly in both time and money. Furthermore, an investment in such expertise does not scale very well, because 1 therapist can treat only a very limited number of patients. So, while insomnia can be treated safely and effectively with cognitive behavioral psychotherapy, cost-effectiveness remains limited.

Internet-delivered CBTI (iCBTI), if compared with classic CBTI, can drastically reduce both the training requirements of the therapists and the therapist time needed to treat 1 patient. iCBTI is a multidimensional approach consisting of information typically for cognitive behavioral therapies but also functional exercises like sleep restriction, stimulus control, sleep hygiene, and relaxation techniques.⁹ iCBTI can be delivered, for example, via internet or telephone or mobile app. Sometimes these treatments are called digital CBTI or, if delivered via smartphone, mobile-app–based CBTI. Different therapies can also have differences due to their content or level of interaction with the therapist. For example, in iCBTI, the patient may follow an online program designed by experts according to the principles of CBTI and can communicate with a therapist usually via written chat. There exist both purely self-guided and therapist-guided formats for iCBTI. One of the main aims of implementing scalable low-intensity treatments like iCBTI is to reduce the need of more intensive treatments. According to our clinical experience within our guided iCBTI, most patients need only occasional aid in following the treatment. Such aid can be provided with a fraction of the resources required for CBTI. Yet, iCBTI treatments, on average, achieve treatment-effect sizes comparable to those in face-to-face CBTI treatment, resulting in a 1.09 standard deviation reduction in insomnia severity at post-treatment.^9,10 However, different approaches to deliver iCBTI are associated with distinct treatment effects.¹¹

Although cost-effective, iCBTI treatments are not similar to self-help programs: treatment durations are longer and the amount of personal clinical support is higher, which is associated with larger iCBTI-treatment effects compared with those of self-help programs.⁹ All in all, iCBTI treatments appear to be well suited for mass deployment and to be a suitable solution to the CBTI availability problem.

Preliminary evidence suggests that iCBTI may reduce or prevent other mental health problems, leading to cumulative public health benefits. In a recent study, iCBTI reduced both insomnia and concurrent depression in a wide range of individuals from various demographic groups.¹² However, these authors stressed the importance of studying actual help-seeking populations in a real-world treatment setting. In their meta-analysis, Zachariae et al¹⁰ similarly noted the need for “studies examining the effectiveness of iCBTI in clinical settings, eg, in primary health care,” while most evidence for iCBTI currently comes from randomized controlled trials specifically designed to demonstrate their benefits.

This study evaluates the real-world effectiveness of the iCBTI performed by the HUS Helsinki University Hospital (HUS iCBTI) in a large naturalistic patient population with a wide range of somatic and psychiatric comorbidities, referred to iCBTI from various treatment settings. Because our study is a nationwide registry study, we compared the results with the outcome variables estimated from other studies. The primary outcome variable is the effect of treatment on the Insomnia Severity Index (ISI). However, insomnia is commonly accompanied with, and may even cause, depressive symptoms and alcohol use.^2–5 We therefore also report the effect of treatment on depression and anxiety symptoms and alcohol use as well as data on adherence and homogeneity of treatment.

METHODS

The HUS Helsinki University Hospital, Department of Psychiatry, has been producing internet- delivered, therapist-supported CBT programs since 2013. Internet-delivered treatment programs now total 11; the most popular therapies are those for depression, for generalized anxiety, and for insomnia.¹⁴ The HUS iCBTI program was developed to meet the increasing need for treatment modalities for insomnia in Finland. The HUS Psychiatry Department and HealthFOX Ltd have collaborated to create the iCBTI based on the technical platform of the HealthFOX Easier Tomorrow service concept for health care professionals and patients. As a university hospital, HUS clinical experts were responsible for all of the treatment content and treatment processes, and HealthFOX Ltd was responsible for the technical execution of the therapy platform. Patients can be referred to the HUS iCBTI program by any licensed medical doctor from any level of health care in any corner of the country, with minimal exclusion criteria. The treatment is paid by municipalities, meaning that all patients who are experiencing insomnia in Finland could easily be referred to iCBTI at no cost to the patient.

Patients

Our registry sample consisted of all patients accepted to the HUS iCBTI program between April 2019 and June 2021. The patients were referred to HUS iCBTI from primary health care, student health care, and occupational health care as well as from various private health care settings from different localities in Finland as part of their normal treatment procedure for insomnia. Patients had been diagnosed with Insomnia F51.0 by their referring physician.¹⁵ There was no repeated confirmation of the diagnosis in the treatment program. Exclusion criteria were as follows (1) acute psychosis or mania/hypomania in an active phase of symptoms, (2) current untreated addiction problem, (3) acute suicidality, (4) neurological or neuropsychiatric disease with cognitive decline, or (5) bipolar disorder type I.

Intervention

The iCBTI consisted of 7 consecutive phases and a follow-up phase 3 months after treatment completion. The phases are planned to be executed on a weekly basis. The treatment had a restricted completion time of 30 days for each phase. The program was based on the theory of CBTI. The core issues are to help each patient to identify and eliminate negative thoughts and worries that keep one awake and to replace behavior that worsens sleep problems with habits that promote sound sleep. More specifically, treatment includes psychoeducation for better sleep, education for stimulus-control therapy, sleep restriction, sleep hygiene, sleep environment improvement, mindfulness-based relaxation training, and the skill of remaining passively awake—these all are standard components of cognitive behavioral therapy for primary insomnia.⁹ The therapy phases included information via text as well as educational illustrations, pedagogic case stories, a generic (not titrated uniquely) sleep diary, and therapeutic assignments. The main contents of the sessions are described in Table 1.

Table 1.

Content by phases.

	Description
Phase 1	Introduction to HUS Helsinki University Hospital internet-delivered cognitive behavioral therapy for the insomnia program, cognitive behavioral therapy model for insomnia; included “pretreatment” ISI reporting.
Phase 2	Information about sleep, good sleep conditions, sleep pressure, and conscious relaxation
Phase 3	Sleep maintenance, the impact of the day’s events on sleep, and dealing with worries during the day
Phase 4	Stimulus control, sleep restriction; ISI report
Phase 5	Core beliefs and distortions of thought related to sleep
Phase 6	Fear of insomnia, more relaxed relationship with sleep, accepting presence
Phase 7	Compassion, achievement of goals, and plans for the future; ISI report
Phase 8	Three-month follow-up session; ISI report

ISI = Insomnia Severity Index.

Delivery of the HUS iCBTI was via a mobile application on a smartphone or tablet. After approval, the patients received an e-mail and an SMS prompting them to sign in and begin the treatment. The patients downloaded the application from an application store and logged into the program via strong electronic authentication.

The therapists in the treatment were either trained clinical psychologists or psychiatric nurses with additional studies in CBT, and all were working at HUS Helsinki University Hospital Department of Psychiatry. All therapists did receive training on internet delivery of CBT, intervention protocols, and good clinical practices. Therapists also had regular group supervision and could seek consultation with a senior psychologist or psychiatrist at any time. The actions of the therapists in the program were manualized.

The therapists’ role was to remotely monitor and asynchronously support the patients’ progress in the otherwise automated program. Therapeutic commentary was offered weekly via in-program chat and additionally when the patient contacted the therapist for some reason. In cases in which the patient was out of contact for 2 weeks, the therapist attempted contact via phone, SMS, or chat. If the patient still did not resume contact, and no reply came in, further contact was attempted via phone or SMS or both, and a deadline was provided for proceeding in the program. Therapist time per patient was not monitored, but a quota of patients per dedicated working hours was the rule. On average, this would mean 9 to 11 minutes per patient per week. If a therapist ended a work contract or went on leave, that therapist’s patients were changed to another therapist temporarily or permanently.

Measures

Here, we used the ISI as our primary outcome measure. The ISI is a reliable and valid instrument to quantify perceived insomnia severity and can serve as a screening device or as an outcome measure in insomnia-treatment research.^16,17 The ISI was measured after sessions 1, 4, 5, and 7.

Symptoms of depression and anxiety were screened with the Patient Health Questionnaire (PHQ)—4-item version (PHQ-4).¹⁸ The PHQ-4 is a composite measure containing both the PHQ-2 and two-item measure for anxiety (GAD-2, Generalized Anxiety Assessment-2).^18,19 Despite its brevity, the PHQ-4 has excellent criterion validity, its clinical classification accuracy is on par with the longer measures, and it can detect meaningful changes in treatment.^20–22 PHQ-4 was measured after sessions 1 and 7.

Alcohol use was measured with the 3-question version of the Alcohol Use Disorders Identification Test (AUDIT-C).^23,24 AUDIT-C asks about current alcohol use: frequency of use, typical quantity, and frequency of heavy consumption. AUDIT-C has had good criterion validity for current alcohol abuse in the Finnish general population.²⁵ AUDIT-C was measured after sessions 1 and 7.

Statistical methods

First, we tested homogeneity of the HUS iCBTI service. Specifically, we tested whether the treatment outcome depended on the patient’s iCBTI therapist, coded as either a fixed or a random effect on the outcome.²⁶ We also tested whether the patient characteristics and treatment outcomes stayed constant over the treatment program’s history. For the latter purpose, we estimated (local polynomial regression) scatterplot smoothers of ISI score and change in the score vs therapy entry date, using the default parameters of the R software version 3.6.3 (R Foundation for Statistical Computing, Vienna, Austria).²⁷ We further used simple logistic regression models and correlation coefficients to find and understand covariates of completion rates and treatment response. Second, we computed both unstandardized and Cohen’s d effect sizes and remission and response rates among those who completed the HUS iCBTI program by finishing phase 7 of the program.^28,29

In line with the literature, “remission” was defined as an ISI score < 8 and “response” as an ISI-score reduction of ≥ 8 points.²⁹ Some patients whom we intend to treat may fail to complete the program either because they notice it is not working for them or because of general motivational issues for any treatment. However, in iCBT treatments there seems to exist a subgroup of patients who benefit from the therapy rapidly and discontinue treatment after the benefits.³⁰ These patients may regard further treatment as unnecessary. Hence, discontinuation should not be considered, by default, a treatment failure. Treatment discontinuation associated with observed preceding symptom reductions would lead to attrition bias, which sometimes can be reduced via multiple imputation with chained equations.³¹ We implemented multiple imputation using the “mice” R package.³² We ran 100 predictive mean matching imputation chains for 100 iterations, monitoring their satisfactory mixing and sensible imputation distributions. This methodology cannot correct for effects of discontinuation due to unreported benefits, but reporting required little effort in the mobile app.

Ethical considerations

The study was registry based, allowing patients to receive treatment regardless of participation in the study. The research was reviewed by an institutional review board and approved by the ethics committee of the Helsinki University Hospital.

RESULTS

Between the dates April 16, 2019, and June 8, 2021, the registry held 3,059 unique treatments for 3,027 different patients and 18 unique online psychotherapists. Of these treatments, 2,645 had ended by the date of data shipment, with 1,994 fully completed treatments (reached phase 7 and therapist tagged as “completed”) and with 651 stamped as terminated before completion (“interrupted”). A crude completion rate based on these numbers is 75.39%. On average, it took 53.52 days for a patient to proceed from the first phase 1 entry to the last phase 7 entry (standard deviation = 30.28; range 32.70–484.45 days), with 61.45% completing the 7 phases in 7 weeks.

Altogether, 87.38% of patients in the system had ended their therapy, either through completion or through interruption. The remaining 12.62%, which we did not further analyze, were still ongoing or uncancelled treatments. We considered the 2,645 patients with ended therapies as our intention-to-treat sample (26.8% men, mean age 41.5 years, minimum age 16 years, maximum age 84 years, and standard deviation 13.5). We discarded from analyses the patients whose therapy process was still ongoing to ensure that our imputation models correct for attrition rather than for being a recently registered patient. The patients in the intention-to-treat sample were treated by 1 of the 18 online psychotherapists (Figure 1A; all of the therapists also had other responsibilities than delivering online treatments). Many of the interrupting patients nevertheless followed through a large part of the program (Figure 1B) and benefited from it in a manner quite comparable to the completers (Figure 1B, Figure 1C; note that the very few patients interrupting at the final phase of therapy may be a special group).

Figure 1. Therapy characteristics.

(A) Numbers of patients per therapist. (B) Count of initiated phases among patients who terminated before completing entire program (7 phases). (C) ISI scores per therapy phase for patients who completed the treatment program. The violin plot shows densities of the ISI score values, horizontal lines show the median values, and dots and whiskers show the mean values and their 95% confidence intervals. (D) Same as panel c but for patients who interrupted therapy before completing the program. ISI = Insomnia Severity Index.

Treatment homogeneity and predictors of therapy completion

The present study pertains to a growing naturalistic treatment register of online therapies rather than a pre-designed trial, which allows one to follow possible changes in the referred patient material and in the effect of the psychotherapy as time passes. Assuming a treatment bottleneck has existed prior to the introduction of the online therapy, the patients referred to the treatment when the program was new could represent more severe cases than the later incoming patients. However, we did not find a significant overall correlation between therapy starting date and initial ISI score (r = –.025, P = .197), or between starting date and treatment effect (change in ISI score from phase 1 to 7, r = –.022, P = .317). A visual examination suggested that the earliest data may have contained slightly more difficult cases on average, but after the first year of HUS iCBTI service, the effect of starting date on initial insomnia severity flattened out (Figure 2). Up until this point, the therapists in the registry had made a rather uneven contribution to total patient effect (Figure 1A), and we found no evidence for treatment-effect differences between therapists (therapist explained negligible variance as a random effect and was also a statistically nonsignificant fixed effect, $F_{\mathrm{17,1962}}$ = 1.463, P = .099).

Figure 2. Therapy starting date, initial ISI score, and ISI change.

(A) Scatterplot of ISI score at therapy entrance against service date, with a local regression line. (B) Zoom-in on the local regression curve with 95% confidence boundaries (dashed lines). (C) Scatterplot of ISI change from phase 1 to phase 7 against service date at entrance, with a local regression line. (D) Zoom-in on the local regression line with 95% confidence boundaries (dashed lines). ISI = Insomnia Severity Index.

None of these factors (starting date, starting ISI, phase 4 ISI, or phase 5 ISI) influenced the probability of therapy completion, but age, sex, mood symptoms, and alcohol use did (Table 2). Young patients, males, and those with present depression or anxiety were less likely to complete the treatment. While the odds ratio (OR) for age may appear small when expressed in units of single years in Table 2, each additional 30 years of age increased the odds of completing the therapy to (up to) 2.1-fold (note: if y = c · x is a scaled version of predictor x, then the OR of y scales as exp(c · log(OR_x)]). Age was also slightly positively correlated with treatment response assessed by ISI change from session 1 to 7 (r = .039, P = .081). Alcohol use reduced the odds of therapy completion (Table 2) but did not correlate with treatment outcome (r = –.001, P = .955). Depression and anxiety symptoms both reduced the odds of treatment completion (Table 2) and correlated with the ISI change (r = –.051, P = .022). These variables could potentially help in multiple imputation of missing data when predicting treatment response.

Table 2.

Predictors of treatment completion in simple logistic regressions.

Variable	OR	95% CI	P
Age (years)	1.025	1.018, 1.032	<.001
Sex (male)	0.720	0.593, 0.873	<.001
Entry date (SD)	1.024	0.937, 1.118	.602
Phase 1 ISI	1.002	0.981, 1.024	.849
Phase 4 ISI	0.991	0.969, 1.014	.433
Phase 5 ISI	0.980	0.957, 1.004	.099
Phase 1 PHQ-4	0.945	0.918, 0.972	<.001
Phase 1 AUDIT-C	0.886	0.849, 0.924	<.001

Phases 4 and 5 were the only intermediate treatment phases where patients filled in the ISI. The PHQ-4 and the short AUDIT-C were available only from the first and last phases. AUDIT-C, Alcohol Use Disorders Identification Test, CI = confidence interval, ISI = Insomnia Severity Index, OR = odds ratio, PHQ-4, 4-item Primary Health Questionnaire, SD = standard deviation.

Crude effects among treatment completers

Among therapy completers, their ISI scores declined, on average, by –7.037 points during iCBTI (from phase 1 to 7; Table 3). A total of 91.06% of the completers improved in the sense of having a lower ISI score in phase 7 than in phase 1. A total of 34.0% of the completers achieved “remission” by phase 7 (if defined as ISI score < 8) and 46.0% “responded” to treatment (defined as an ISI-score reduction of ≥ 8 points from phase 1; definitions from Vedaa et al²⁹). In terms of Cohen’s effect size, the ISI-score decline was $d$ = –1.449—a large effect. Although not equally large, a substantial decline was evident in mood symptoms measured with the PHQ-4 questionnaire over the course of the therapy (Cohen’s d = –0.610). Only a very small (d = –0.110), albeit still statistically significant (P < .001), decline occurred for alcohol use measure AUDIT-C. One interesting question is what would be the effect of the therapy on insomnia if the patients who terminated the program also had gone through the therapy and fulfilled the given therapy assignments? In what follows, we statistically evaluate this counterfactual question.

Table 3.

Average symptom scores across the HUS iCBTI treatment.

Phases	ISI	SD(ISI)	n	nmiss(ISI)	PHQ-4	SD(PHQ-4)	AUDIT-C	SD(AUDIT-C)
Phase 1	17.43	4.16	1,994	18	5.17	3.12	2.72	2.02
Phase 4	14.58	4.73	1,994	13	‒	‒	‒	‒
Phase 5	12.99	5.20	1,994	16	‒	‒	‒	‒
Phase 7	10.41	5.46	1,994	18	3.33	2.91	2.49	1.95

The table shows the change in outcome variables of treatment completers at the measurement stages of treatment. AUDIT-C = Alcohol Use Disorders Identification Test, HUS iCBTI, internet-delivered cognitive behavioral therapies for insomnia manufactured by HUS Helsinki University Hospital, ISI = Insomnia Severity Index, nmiss = treatment completers with missing data, PHQ-4 = 4-item Patient Health Questionnaire, SD = standard deviation of measured data.

Imputed intention-to-treat effects for all ended therapies (completed or not)

Some patients start a therapy with an intention to treat themselves but for some reason fail to complete the therapy. For these patients, the symptom outcome of the full therapy can be characterized as “missing data.” Yet, their other covariate data may provide statistical clues toward the eventual total effect of the therapy had we been able to record the outcome data for all of the patients entering the program. The use of such statistical clues is called imputation. Based, for example, on associations in Table 2, we used the following imputation variables in our multiple imputation algorithm: age, sex, phase 5 ISI score, phase 1 PHQ-4 score, phase 1 AUDIT-C score, and completion status (Table 4).

Table 4.

Effect sizes for change in symptom scores from phase 1 to phase 7.

	Nonimputed (Crude) Results				Multiply Imputed Results
Symptoms	ΔScore	95% CI	d	n	ΔScore	95% CI	d	n
Insomnia (ISI)	–7.04	–7.25, –6.82	–1.45	1,966	–6.69	–7.31, –6.06	–1.32	2,645
Mood (PHQ-4)	–1.84	–1.97, –1.71	–0.61	1,968	–1.71	–2.06, –1.36	–0.61	2,645
Alcohol (AUDIT-C)	–0.22	–0.31, –0.13	–0.11	1,964	–0.03	–0.22, 0.16	–0.02	2,645

AUDIT-C = Alcohol Use Disorders Identification Test, d = Cohen’s d, ISI = Insomnia Severity Index, n = sample size per comparison, PHQ-4 = 4-item Patient Health Questionnaire, Δscore = change in sum score, 95% CI = 95% Wald’s confidence interval.

From the imputed Cohen’s d, we can conclude that ISI scores declined for approximately $\mathit{\Phi }(-d)\times 100\%\approx$ 90.63% of the patients from phase 1 to phase 7—an estimate very close to the value observed for completers (91.06%; here, Φ refers to cumulative distribution function of the standard normal distribution). This is a type of upper-limit intention-to-treat estimate for those who benefit from the therapy. Overall, the imputation had very little effect on the treatment-response estimates in terms of ISI and PHQ-4 scores, but it did nullify the treatment effect on alcohol use (Table 4). This suggests that the apparent benefits of the iCBTI regarding alcohol use among completers may have derived from lower completion rates among those patients not reducing their alcohol use.

Some patients (n = 367) who lacked phase 7 ISI score and other correlates nevertheless had phase 1 plus phase 4 or 5 ISI score, allowing us to compute estimates with the last available (phase 4 or 5) ISI observation carried forward to the final score (total n = 2,347). This observed treatment response across all data was –6.52 ISI points (d = –1.32), being very close to the imputed-data estimates (see Table 4). Of course, it contained mostly the same observations and the treatment response was markedly lower in the specific group of 367 patients having phase 4 or 5 ISI score as their last reported ISI value (–3.78 ISI points and d = –0.78).

DISCUSSION

In our routine-care patients, ISI scores declined by 7.04 (95% confidence interval –7.25, –6.82) points, on average, among patients completing the program. Also noteworthy was the commitment of patients to treatment: 75.39% completed the program. The quality of our treatment was homogeneous in the sense that the treating therapist or date of service had no significant effect on treatment outcome. In standardized units, the average uncontrolled within-individual treatment effect was large (Cohen’s $d$ –1.45 for completers; imputed d = –1.32; last-observation-carried-forward d = –1.32). Altogether, 91% of the completing patients improved rather than deteriorated. Over the course of the therapy, comorbid mood problems and alcohol problems showed smaller but still significant improvements. The treatment effect on insomnia symptoms and mood problems withstood imputation analysis for noncompleters, whereas that on alcohol problems did not.

Based on data from randomized controlled trials, we expected to find within-subject decreases of 3.4 to 8.8 points on the ISI score.^13,29 Thus, our naturalistic data revealed a treatment effect comparable to trial-reported treatment effects, which is a surprisingly good naturalistic outcome given that some publication bias has been reported in randomized trials,^7,10 and given that they tend to select for inclusion patients with fewer comorbidities compared with naturalistic studies. We next compared our results with those of other studies shown in the paper of Seyffert et al.¹³ They showed average (standard deviation) within-subject pre- to post-therapy differences of –5.6 (3.8), –3.4 (4.6), –8.2 (3.6), and –8.8 (3.7) ISI scores in 4 different trials, implying standardized effects of –1.47, –0.74, –2.28, and –2.38, respectively. We observed an ISI change of –7.0 and standardized effect of –1.45. Their waiting-list improvement was, on average, 2.23 ISI points. While we did not have our own waiting-list data, we can use their numbers as a rough reference point because these are just patients not yet receiving treatment. If we remove the average waiting-list effect observable in those studies, the controlled treatment effect of the HUS iCBTI treatment would be 4.46 ISI points upon therapy completion (“counterfactual” imputations) and 4.81 for those who actually completed the therapy. The absolute effect of that size falls on the upper half of the interventions presented in Seyffert et al¹³ (they reported controlled effects of –4.49, –2.40, –3.70, and 6.60). Furthermore, our ISI-score reduction of 7.0 points is slightly lower than the 8.8 points recently reported by Vedaa et al²⁹ for the Norwegian Sleep Healthy data and the Internet program, but their sample had slightly higher baseline symptoms (19.2 points vs our 17.4 points). Initially higher symptom severity might have allowed for more dramatic treatment effects or might have hindered treatment—or both—making accurate comparisons of treatment programs difficult when based on these data. To whichever the 2 programs pertain, Finnish or Norwegian, we agree with the conclusion of Vedaa et al²⁹ that empirical support exists for the “widespread dissemination” of iCBTI.

HUS iCBTI is a guided online treatment, but no differences between therapists were observed in this study, which suggests that the role of therapist might not be crucial in our treatment. We have compared our results with effective treatment data from fully automated iCBTI from Vedaa et al²⁹ and data from meta-analysis from Seyffert et al,¹³ including data from both guided and automated nonguided therapies. As therapist-supported treatments have been more effective compared with automated ones in some cases^33,34 a question worth exploring is whether the treatment of insomnia in general or treatment with a mobile-application format can be particularly effective with little or no guidance from a therapist.

Interestingly, Vedaa et al²⁹ also concluded that “research is needed to identify the moderators of response and to improve targeting.” Our analysis of the naturalistic register database may provide some important information to researchers pursuing those aims. First, iCBTI seemed to work better for those who were elderly and for female participants who seemed to complete therapy more often than did those who were younger or male (see Table 2: predictors of treatment completion in simple logistic regressions), as frequently observed for iCBT treatments.³⁴ Second, and importantly, comorbid alcohol use and depression and anxiety symptoms can reduce treatment compliance and thereby its benefits. Before using these correlates for targeting, one should, however, keep in mind that those with depression and anxiety symptoms benefited from the iCBTI in terms of both sleep and depression and anxiety symptoms. Further, those who did not complete the treatment nevertheless showed benefits (Figure 1D). Therefore, the health care utility of treating patients with iCBTI is not a simple function of sleep problems or of completion probabilities.

Some of our findings should be interpreted in light of their limitations. First, the estimated therapist effect on treatment outcome may not be entirely accurate. Some patients may have been reallocated to another therapist if their original therapist’s work contract ended; similarly, if a therapist took a leave of absence, although often the therapists kept the online patients despite such absence. Such minor inaccuracies due to patient reallocations should not have diluted strong therapist effects but might have done so for small effects. Second, the intention-to-treat (ie, imputation) analysis required a technical missing-at-random assumption that cannot be directly tested and may not strictly hold for online psychotherapies.^31,32 In contrast to randomized trials, however, intention-to-treat effects are of secondary interest for this naturalistic study.

Other limitations of our register study are its sparse measurement strategy optimized for treatment rather than research. Additional limitations due to the study design were lack of data concerning the sleep diary and the therapists’ contact with patients as well as lack of a comparator group. However, the nationwide registry-based design also brings with it the significant strengths of the study: a large sample of nationwide routine care when only a few studies have reported their effectiveness in routine care.

In summary, we observed large benefits for HUS iCBTI treatment in its naturalistic setting. The effect sizes were comparable to previously reported randomized controlled trials. Mobile-application–based, guided iCBTI appears to be a promising treatment for insomnia symptoms and may also reduce the risk for other adverse health consequences associated with poor sleep.

DISCLOSURE STATEMENT

All authors have seen and approved the manuscript. Work for this study was performed at Helsinki University Hospital. This study was funded by the Government of Finland and the Hospital Region of Helsinki and Uusimaa (HUS). Tom Rosenström is supported by the Academy of Finland (grant numbers 334057 and 335901). In Finland, this treatment is provided by the hospital district with public funds. The treatment is not aimed at financial gain. The authors report no conflicts of interest.

ABBREVIATIONS

AUDIT-C

Alcohol Use Disorders Identification Test

CBT

cognitive behavioral therapy

HUS iCBTI

internet-delivered cognitive behavioral therapies for insomnia manufactured by HUS Helsinki University Hospital

iCBT

internet-delivered cognitive behavioral therapy

iCBTI

internet-delivered cognitive behavioral therapy for insomnia

ISI

Insomnia Severity Index

PHQ

Patient Health Questionnaire