A total of 75 patients, 39 women and 36 men, were included in the sample with a mean age of 62 years (see Table 1). Self-reported mean duration of type 2 diabetes in the patients was 13.4 (SD = 9.2) years. In the patient sample 45% were working full-time, 28% were on sick leave, and 27% were retired. Use of antihypertensive medication and statins, including ezetimibe, was also noted.

Regarding diabetes medication, some adjustments were made during the intervention (see Table 2). The table describes number of patients using different medication before and at the end of the concentrated intervention. In particular a large proportion of patients initiated GLP-1 agonists or SGLT-2 inhibitor treatment.

The treatment was performed by a multidisciplinary team. The diabetes team consisted of endocrinologists, diabetes nurses, a psychologist, a pharmacist, physical therapists and a clinical nutritionist. All were trained by participating in 4-day treatment groups in a master-novice manner where they first observed a minimum of one 4-day treatment group (led by a master) before they led one. For the present study, all groups were led by the same group leaders, a psychologist (first author AWL) as main leader and a diabetes nurse as co-therapist. The group leader had observed and been co-therapist in seven 4-day treatment groups for other disorders before leading those for type 2 diabetes. Several from the interdisciplinary team joined parts of the 4-day intervention and we used manuals and step-by-step group leader instructions to ensure fidelity.

The patients were referred by their general practitioner or other physician responsible for the treatment of their type 2 diabetes. Eligible patients were invited to participate in the programme. Upon referral, the patients received a telephone call by a clinician, who informed them about the intervention and the PUSH-project. A week prior to the intervention, the patients were contacted by a therapist to ensure all necessary information about the program had been given.

For a more detailed explanation of the intervention, we refer to the protocol paper (Kvale et al., 2021). In short, the intervention consisted of three phases: preparing for change (phase 1), The concentrated intervention (phase 2), Integrating change into everyday life (phase 3). In general, the treatment focused on what the patient can do something about in terms of behavioral change (i.e., movement that increases insulin sensitivity) and not only on the result (i.e., weight, blood glucose, Hba1c) which is not subject to immediate change.

The concept of micro-choices entails identifying when symptoms, habits, or automated and unhelpful behavior occur (Bendixen et al., 2025b). By intentionally breaking unhelpful patterns, replace them with health-promoting micro-choices, the aim is to increase flexibility and a sense of being in control. A shift in focus was introduced from targeting glucose levels to monitoring the everyday micro-choices that may lead to increased insulin sensitivity. A shift away from trying to regulate the blood-sugar level to focus on something controllable; the actions (eating, activity, regulate stress, sleep) that impact insulin sensitivity, has the potential to increase a sense of being more in control of ones‘ illness. This shift implies that change is within reach, as opposed to physiological parameters that are not under immediate control such as blood-sugar levels or weight. The intervention also used the concept of individual goal setting. The individual goals were Specific, Measurable, Achievable, Relevant, and Time-specific according to SMART-goals (Bjerke and Renger, 2017). The sleep education lasted about 1 h and was delivered as a lecture by a somnologist (certified sleep expert) with wide clinical experience in treating insomnia and other sleep disorders. This knowledge was incorporated in other lectures as well so that all disciplines gave lectures that built on each other.

A group introduction meeting was held during this time, in which essential topics, such as challenging behavioral patterns, therapist-assisted behavior activation and basic bodily rhythms were covered. At the same meeting, the necessity of actively choose to participate in all aspects of the program was emphasized. If the patients were reluctant, they were recommended to wait until ready.

A multidisciplinary approach was used during the concentrated treatment. Interactive lectures were given in groups of 6–10 patients, and were held by a diabetes nurse, a diabetes psychologist, a pharmacist, and clinical nutritionist. Lecture topics were on how to achieve behavioral change, diabetes type 2, nutrition, medication, sleep-wake rhythms, brief mindfulness-sessions, and exercise sessions. Patients were concomitantly given practical information and hands-on coaching regarding their SMART-goals. In short, patients must first specify and find a measurable and relevant objective, then identify current baseline to be able to concretise an achievable goal in an appropriate time frame. During the treatment, patients underwent individual consultations with the endocrinologist and diabetes nurses. The main goal of the consultation was to address struggles regarding their SMART-goals, and also suggest changes in medication if necessary.

The overall goal of the third phase was to help the patient integrate changes into everyday life. Seven to 10 days post-intervention a diabetes nurse conducted phone calls to each patient focusing on progress and how to maintain the change. At 3-months follow-up, the diabetes nurses were available through the project application for contact regarding progress and potential struggles.

Sleep disorders were assessed using the questionnaires Berlin Questionnaire (BQ), Bergen Insomnia Scale (BIS), Insomnia Severity Index (ISI), and Epworth Sleepiness Scale (ESS). The questionnaires were answered at a secure online application/website at baseline, 1 week after the treatment and at 3-, 6- and 12-month follow up.

The BQ (Netzer et al., 1999) is a questionnaire for classifying patients in low or high risk of obstructive sleep apnoea. The questionnaire consists of 10 items, further divided into three categories. Category 1 is related to snoring, asking for frequency and severity; category 2 explores daytime sleepiness, including sleepiness while driving a motor vehicle; category 3 asks for presence of high blood pressure. The questionnaire also asks for age, sex, height, and weight. Categories 1 and 2 are positive if there are two points or more, category 3 is positive if the answer to item 10 (“Do you have high blood pressure?”) is yes or if the patient has a BMI >30 kg/m². The patient is classified as high risk if two or more categories are positive, or low risk if one or less categories are positive.

The BIS is a self-report form that consist of six questions corresponding to the DSMV-IV criteria for insomnia (Pallesen et al., 2008). Each question ranges from 0 to 7 days per week, adding to a possible composite score ranging from 0 to 42. The first three questions concern sleep onset, maintaining sleep and preterm awakening, whilst the latter three pertain to the feeling of being rested, functioning during daytime, and being dissatisfied with current sleep. The BIS also has a dichotomous score relating to clinical likelihood of having insomnia disorder (yes or no) that corresponds to scoring 3 or higher on one of questions 1–4 and 3 or higher on one of questions 5–6.

The ISI is a self-report tool to measure subjective insomnia symptoms (Morin et al., 2011). The form has seven items, pertaining to severity of problems with sleep latency, early awakening, satisfaction with sleep pattern, and impact on daily functioning. Each item is scored between 0 and 4, with the total score ranging between 0 and 28. A score between 0 and 7 signifies no clinically significant insomnia, 8–14 signifies subthreshold insomnia, 15–21 and 22–28 suggest moderate and severe insomnia, respectively.

The ESS was initially made to offer an affordable and less time-consuming way to screen for daytime sleepiness than for instance Multiple Sleep Latency Test (MSLT; Johns, 1991). It gives 8 scenarios for which the patient is to rate the likelihood of dozing off or falling asleep. The 8 scenarios are scored from 0 (“No chance of dozing“) to 3 (“High chance of dozing”), with total scores ranging between 0 and 24. A cut-off of ≥11 is considered sleepiness above normal (Johns, 1991).

COMISA can be calculated by high risk of OSA on the BQ in addition to “moderate” or “severe” on ISI and by high risk of OSA on the BQ in addition to fulfilling the dichotomous score relating to clinical likelihood of having insomnia disorder (yes or no) on the BIS.

Since this is a non-randomized pre-post pilot study, no sample size calculation was made in advance. Mixed-effects regression models were used to test for pre- to post changes in continuous BIS, ISI, and ESS scores at different timepoints [1 week (BIS only), 3 months, 6 months and 12 months].

Changes in proportions scoring below or above the predefined cut-offs on ISI, BIS, BQ, and ESS were examined first with Cochran's Q tests. Used longitudinally as here, the Cochran's Q test can indicate whether the proportions in the different cut-off categories at the different time points are the same in the population. Statistically significant Cochran's Q tests were followed by post hoc McNemar's tests comparing baseline proportions to all subsequent time points. The four ISI categories were combined into two categories for the sake of these analyses: a non-insomnia/subthreshold category consisting of the first two categories and a moderate/severe insomnia category consisting of the last two categories.

There were some missing values in ISI and BQ. For ISI the missing values were n = 8 at 3 months follow-up, n = 12 at 6 months follow-up, and n = 30 at 12 months follow-up. For BQ missing values were n = 8 at 3 months follow-up, n = 11 at 6 months follow-up, and n = 23 at 12 months follow-up. Mixed-effect models are flexible and can be fitted even if some outcome data are missing and can produce unbiased estimates provided that missingness is not Missing Not at Random (MNAR). For the analyses on the continuous outcomes all participants were thus included, irrespective of missing data at any of the assessment points. We know that many participants needed help to log on to the website where the questionnaires were filled out and this fact suggests that their ability or willingness to access the site is related to their age or technical skills, which are likely associated with underlying factors such as digital literacy. This indicates that the missing data are systematically related to unobserved characteristics (e.g., comfort with technology), meaning that the data are MNAR but are instead missing due to these underlying factors.

To address missing data on the categorical outcomes, multiple imputation was performed using chained equations (MICE). A combination of logistic and ordinal regression models was used to generate 30 datasets. Following imputation, analyses were conducted separately within each dataset. As Cochran's Q and McNemar's χ² are not parameter estimates, it does not make sense to use Rubin's rule to pool them across imputations. Instead, we used the range of Q and χ² statistics and associated p-values across imputations to assess the consistency and robustness of the findings. Although the original sample size was 75, the number of complete cases available for analysis after imputation varied slightly across time points, due to residual missingness in variables not fully resolved by the imputation models. For descriptive purposes, the average proportion of participants classified as high risk on ISI and BQ were calculated across the 30 datasets and are presented in the Results section.

Written consent was signed digitally by all participants before data collection. The project was approved by the ethics committee in Western Norway (REK West 2020/101.648) and was conducted in accordance with the Helsinki declaration. ClinicalTrials.gov Identifier: NCT05234281.

The mixed-effects regression showed that total ESS score decreased from baseline to 1-week post-treatment. At 3-month follow-up reduction in score from baseline was −1.56 [b = −1.56, (95% CI: −2.78, −0.34), p = 0.013], at 6-months −2.39 [b = −2.39, (95% CI: −3,61, −1,16), p < 0.001], and at 12-months −4.31 [b = −4.31, (95% CI: −5.53, −3.08), p < 0.001]. Table 3 presents additional marginal means computed from the mixed regression.

At baseline, 28% of patients (n = 21) scored above the cut-off on ESS (see Table 4). The proportion decreased to 21.3% at 3-month follow-up, 18.7% at 6-months, and 12% at 12 months. The Cochran's Q was statistically significant, Q(3) = 10.571, p = 0.014, and the post hoc McNemar's tests showed that the proportion scoring 11 or above was statistically significantly lower at 12-month follow-up than at baseline, χ²(1) = 8, p < 0.01. There were no statistically significant differences between baseline and either 3-month or 6-month follow-up.

The mixed-effects regression showed a reduction in the total BIS score from baseline to 1-week post-treatment [b = −1.85, (95% CI: −3.67, −0.035), p = 0.046, Table 3]. As can be seen in Table 3, the mean level of BIS reverted to baseline level at 3-months follow-up (p = 0.914), and then decreased again at 6-months [b = −3.52, (95% CI: −5.71, −1.34), p = 0.002] and 12-months follow-up [b = −3.01, (95% CI: −5.15, −0.87), p = 0.006].

At baseline, 44 (58.7%) patients out of a total 75 were categorized as likely to have insomnia (see Table 4). The Cochrane's Q test showed that the proportion changed over time, Q(4) = 45.565, p < 0.001. Follow-up McNemar's tests showed a statistically significant reduction in proportions at 6-months [25.3%, χ²(1) = 21.55, p < 0.001] and 12-months [29.3%, χ²(1) = 15.12, p < 0.001]. The differences in proportions were not found to be statistically significant at 1 week and 3 months follow-up.

The mixed-effects regression showed that total ISI scores improved from baseline to post-treatment (see Table 3). The reduction from baseline to 3-month follow-up was −1.59 [b = −1.59, (95% CI: −2.63, −0.55), p = 0.003], from baseline to 6-month follow-up −1.80 [b = −1.80, (95% CI: −2.86, −0.74), p = 0.001], and to 12-months −2.48 [b = 2.48, (95% CI: −3.67, −1.29), p < 0.001].

At baseline, 25.3% of patients fell into the moderate/severe insomnia category (n = 15 in moderate and n = 4 in severe, see Table 4). The Cochran's Q statistic was statistically significant in all 30 imputations, ranging between 21.091 and 27.462 (all ps < 0.001). Follow-up McNemar's tests showed a statistically significant reduction in proportions at 3-months (χ² range: 9.31–13, all ps < 0.01), 6-months (χ² range: 8.33–12, all ps < 0.01), and 12-months (χ² range: 10.89–14.22, all ps < 0.01). The average proportion of participants classified as moderate/severe insomnia across the 30 imputations was 8.4% at 3-months (based on n = 73), 9.9% at 6-months (based on n = 72), and 4.2% at 12-months (based on n = 72).

At baseline, 56% (n = 42) of patients were found to have high risk of OSA (see Table 4). The Cochran's Q statistic was statistically significant in all 30 imputations, ranging between 21.344 and 33.308 (all ps < 0.001). Post hoc McNemar's tests showed that the proportions having high risk of OSA were statistically significantly lower at 3-months (χ² range: 5–9.78, p range: 0.002–0.025), 6-months (χ² range: 13.37–19.59, all ps < 0.001), and 12-months follow-up (χ² range: 6.26–18.62, p range: 0.000–0.012) compared with baseline. The average proportion of participants classified as high risk across the 30 imputations was 39.2% at 3-months (based on n = 73), 27.2% at 6-months (based on n = 73), and 31.9% at 12-months (based on n = 73).

At baseline, 21.3% and 36% fulfilled the definition of COMISA based on BQ + ISI and BQ + BIS, respectively. The percentages were reduced to 5.6% and 28.9% at 3 months, 4.3% and 9.6% at 6 months and 2.8% and 11% at 12-months follow up.

Our study design poses as a limiting factor, as the non-randomized approach might give rise to selection bias. Missing data, particularly regarding the Berlin Questionnaire, might contribute to selection bias as we cannot evaluate the condition of those not able to adhere to the follow-up forms (Lachin, 2016). We have adjusted for missing data by mixed-effect models that are flexible and can be fitted even if some outcome data are missing and can produce unbiased estimates provided that missingness is not MNAR, which we argue that is the case. We used MICE to address missing data on the categorical outcomes, where multiple imputation was performed. This procedure produces less biased parameter estimates compared to simpler methods.

As part of the intervention, changes were made in the patients' diabetes medication (see Table 2). The contributing effect of such changes on sleep problems in patients with type-2 diabetes may also be interesting to explore in future studies, as SGLT-2 inhibitors have yielded promising results for the treatment of OSA (Tanriover et al., 2023), and we cannot rule out that effect. In our sample 23% of the participants were on SGLT-2 inhibitors pre intervention while 53% were on them at the end of the intervention. Consequently, such changes might have reduced OSA symptoms in our study population. Clearly, the study design does not allow for teasing out the individual effect components and changes in medication may have had a large impact. Also, we cannot rule out drug side-effects in relation to the changes in the sleep parameters. However, we did not make changes in medication other than related to their diabetes during the intervention.

The same two group leaders led all intervention groups. Hence, we can not rule out that unique characteristics or biases of the group leaders could influence outcomes, making it difficult to determine if observed effects are due to the intervention or the leaders' individual traits.

A strength of our study is that we have investigated common sleep-related disturbances in type 2 diabetes and measured them at multiple and long-term time points. Such study designs and data are rare in type 2 diabetes, and hence provide novel and potentially important information. Also, including lectures on how to improve sleep in diabetes treatment programs is not common. The sleep-related education session lasted only 45–60 min and we do not know how much of the improvements has to with what they learned about sleep-wake rhythms (we did not test their sleep knowledge pre – post) and how much of the improvement is residual to the changes they made in activity, food intake and so on. The dismantling of these effects should be investigated in future studies. However, in previous studies from concentrated treatment format for OCD, the authors conclude that 4 days may be too short to reduce insomnia symptoms short-term (Hagen et al., 2021), while we found an immediate reduction (after 1 week) in insomnia-symptoms when including more sleep-specific education in concentrated treatment for depression and for anxiety (Wilhelmsen-Langeland et al., 2024).

If this concentrated treatment reduces the risk of OSA in patients with type 2 diabetes, it may be highly health-economically positive as other treatment modalities for OSA are costly, and so is the assessment of OSA. The combination of OSA and type 2 diabetes also increase the risk of cardiovascular disease, hence the results we report on here may be of great importance, pointing at novel ways to reduce diabetes type 2 complications in a short amount of time. A physiological assessment of OSA is however still the gold standard.

The PUSH-project's goal, to explore treatment alternatives for the next generation of public health care solutions, constitutes the core value of our study. Our results indicate an effect of the 4-day concentrated treatment format on different modalities of sleep in our sample of diabetes patients. The primary goal of the treatment given was to change focus from end-results (e.g., glucose levels) to behavioral daily health-promoting micro-choices, thus emphasizing a holistic and hands-on approach. As such, the exact cause of improvement is not enlightened through our study. As sleep is a cornerstone in maintaining good health and wellbeing, it may be beneficial to look for treatment options which include sleep as an important outcome target.