Sleep-AID: a cross-sectional analysis of subjective sleep quality, psychosocial measures and real-world glycemic outcomes in people with diabetes using automated insulin delivery systems Authors: Drew Cooper, Tebbe Ubben, Christine Knoll, Christine Blume, Wolf-Julian Neumann, Katarina Braune Published: 2026 ABSTRACT AbstractStudy ObjectivesSleep disturbances are common in people with diabetes, yet their relationship to automated insulin delivery (AID) use has not been well described in large, real-world cohorts. We assessed subjective sleep quality, fear of hypoglycemia, and their associations with glycemic outcomes among adults with diabetes using open-source and commercial AID compared with non-users.Materials and MethodsWe conducted a cross-sectional analysis of 529 adults from the OPEN Project (2020–2023). Participants completed the Pittsburgh Sleep Quality Index (PSQI) and Hypoglycemia Fear Survey-II (HFS-II) short form. All participants self-reported glycated hemoglobin (HbA1c) levels; a subsample of 60 AID users contributed ≥25 days of continuous glucose monitoring (CGM) data. Free-text responses to PSQI item 5j were thematically coded.ResultsAID users reported better sleep (median 5 [IQR 3–7] vs. 7 [IQR 5–9], p < .001) and lower fear of hypoglycemia (median 7 [IQR 3–11] vs. 13 [IQR 8–19], p < .001) than non-users. Nevertheless, 43% of AID users and 69% of non-users exceeded poor sleep thresholds. Women reported worse PSQI and HFS-II scores than men (both p < .001). Diabetes-related causes of sleep disruption were more common among non-users (p < .05). PSQI and HFS-II scores correlated with HbA1c in the subsample, but associations with CGM metrics were not significant.ConclusionsAdults using AID systems reported better sleep quality and lower fear of hypoglycemia than those managing diabetes without automation, yet a large proportion of all participants continued to experience poor sleep. Diabetes-related nocturnal disruptions remained common among non-users, suggesting that AID may alleviate—but not eliminate—the nighttime diabetes burden. Statement of SignificanceThis is a first-of-its-kind study assessing subjective sleep quality, fear of hypoglycemia, diabetes outcome measures, continuous glucose monitoring (CGM)–derived glycemic metrics and free-text responses simultaneously, integrating quantitative and qualitative methods across a broad spectrum of people with diabetes including multiple diabetes subtypes and treatment paradigms. Our results show that while automated insulin delivery (AID) use is associated with better sleep quality and lessened fears of hypoglycemia, poor sleep still persists in all participant subgroups, indicating the need for more targeted approaches to improving sleep. Open-source AID users demonstrated better sleep quality than commercial users, which suggests that more granular control over AID system settings leads to better outcomes. Future studies should aim to recruit a more diverse cohort, and collect glucose data from all participant subgroups. Introduction Sleep is a fundamental determinant of health, and is intrinsically linked to cognitive, psychological, and cardiometabolic well-being. Specific to metabolic health, insufficient sleep quality has been shown to reduce insulin sensitivity, increase glucose variability, and impair hormonal regulation [1]. Regular sleep disruption may create a vicious cycle in which the accumulation of sleep debt further disturbs individuals’ sleep–wake cycles, leading to compensatory sleep behaviour, negatively impacting quality of life and compounding adverse health outcomes [2]. This kind of altered sleep has been reported in people with diabetes, and may be a consequence of both the disease and its management [3]. Diabetes is characterized by impaired insulin response and/or production, which causes high blood glucose concentration (hyperglycemia). While mild hyperglycemia may not pose immediate danger, it can affect sleep, emotional well-being, and contribute to long-term cardiovascular and neurological complications [4–10]. In extreme cases it can precipitate diabetic ketoacidosis, a life-threatening state marked by dehydration, metabolic acidosis, and potential coma. For people with diabetes requiring insulin therapy, insulin is either delivered via multiple daily injections (MDI), or administered continuously via an insulin pump. Excess insulin can lead to low blood glucose concentration (hypoglycemia), which in severe cases can lead to loss of consciousness and seizures. Diabetes management is thus a balancing act between these two extremes, requiring constant vigilance day and night. Sleep is therefore a particularly challenging aspect of life for people with diabetes to navigate, as hypo- and hyperglycemic episodes need to be remedied with relative immediacy. Individuals need to be awake to consume carbohydrates or manually dose insulin, while for those using diabetes technology such as glucose monitors and insulin pumps, device alarms will directly interrupt sleep [11]. The fear of a nocturnal hypo- or hyperglycemic events may itself impact sleep quality; diabetes symptoms, their perceived impact, and treatment can all induce significant mental strain [12–15]. People with diabetes are increasingly using automated insulin delivery (AID) systems, which integrate continuous glucose monitors (CGMs) and insulin pumps with an algorithm into a “closed-loop” system, automating insulin dosing and maintaining glucose levels within target range [16]. Users of open-source AID reported that these systems reduced diabetes distress and fear of hypoglycemia, improved subjective sleep quality, and allowed them to sleep through the night with fewer diabetes-related disruptions or device alarms, all while maintaining an average time in glycemic range exceeding international treatment targets [17–23]. Commercial AID systems have demonstrated similar benefits across a range of glycemic and psychosocial outcomes [24–26]. Despite evidence that AID use is associated with improved sleep-related outcomes, important gaps remain. Few studies have examined subjective sleep quality, specific reasons for sleep disturbance, and fear of hypoglycemia using validated questionnaires, in a mixed population of people with diabetes using different diabetes management strategies. It also remains unclear if such quality of life factors are linked to glycemic outcomes such as HbA1c or CGM-derived glycemic variability. In this secondary analysis of data from the “Outcomes of Patients’ Evidence With Novel, Do-It-Yourself Artificial Pancreas Technology” (OPEN) project, we examined differences in sleep quality and fear of hypoglycemia between AID users and non-users, assessed associations with glycemic outcomes, and analyzed open-ended questionnaire responses to better characterize reasons for sleep disturbance. By integrating subjective sleep assessments, psychosocial measures, glycemic outcomes, and qualitative data, this study provides a more comprehensive understanding of the lived experience of sleep in people with diabetes. Materials and methods Study design and recruitment We conducted a cross-sectional analysis of data from the OPEN project, an international research collaboration investigating outcomes of AID systems with a specific focus on open-source AID systems [27]. Participants were recruited online between September 2020 and April 2023 through social media and diabetes community channels, including Facebook groups, Twitter, Diabetes Daily, and through the OPEN project’s newsletter. The study was approved by the Life Sciences Human Research Ethics Committee, University College Dublin (LS-20-37), and the Charité – Universitätsmedizin Berlin ethics committee (EA2/206/21), with all participants electronically providing informed consent. Survey data were collected and managed using REDCap, while medical device data were donated to the Open Humans data repository [28–30]. The survey was conducted in English. Eligibility criteria Eligible participants were adults (≥18 years) living with diabetes who could read English and provide online consent. Specific to diabetes type, those with either Type 1 (T1D), Type 2 (T2D), latent autoimmune diabetes in adults (LADA), or maturity-onset diabetes of the young (MODY) were eligible. Participants were either users of an AID system (open-source or commercial) or non-users using other treatment modalities (continuous glucose monitors with multiple daily insulin injections, or non-automated insulin pump therapy including predictive low glucose suspend). There was no minimum diabetes duration. For analyses linking sleep outcomes to glucose metrics, we included only participants who donated ≥25 days of continuous device data within the corresponding 30-day period covered by the survey and who uploaded device data via Nightscout or AndroidAPS. Individuals were grouped based on type of diabetes, specifically in relation to shared metabolic characteristics between types; those with T1D or LADA in one group, and those with T2D or MODY in another. Individuals were also categorized into further subgroups based on AID usage (incl. Additional diabetes management peripherals for those “non-users” of AID systems), type of AID (commercial vs. open source), gender, ethnicity, and education. Measures Subjective sleep quality was assessed with the original 19-item Pittsburgh Sleep Quality Index (PSQI) questionnaire, which evaluates sleep over the preceding month and yields a global score from 0–21; scores >5 indicate poor sleep quality [31,32]. Responses to the PSQI open-ended item 5j (“other reasons for sleep disturbance”) were thematically coded by three independent reviewers (DC, TU, KB) using a predefined deductive codebook (Table 1). Discrepancies were resolved by consensus. Table 1Deductive coding framework for responses to PSQI item 5j (“other reasons for sleep disturbance”) Code Label Example 1A) Diabetes-relatedDisturbances related to symptoms, treatment, and day-to-day management of diabetes (e.g., hypoglycemia, hyperglycemia, itching/painful injection site, needing to take insulin or carbohydrates during the night).“Frequent and hard to treat hypos”1B) Diabetes TechnologyDisturbances specifically related to diabetes technology (e.g., CGM and pump alerts, connectivity issues, worry about device functioning).“Bluetooth failure between [AndroidAPS] and Combo - alarm”2) ChildcareDisturbances related to parenting or caring for children (e.g., infant care, nighttime waking of children) or managing a child’s diabetes.“Small child, sometimes crying at night”3) Stress/Mental HealthDisturbances related to perceived stress, worry, mental load, and psychological concerns.“I can’t stop thinking about what I didn’t do for work that day that urgently needs to be done.”4) Physical Health/Well-beingDisturbances related to physical health symptoms, conditions, self-care, or well-being not specific to diabetes (e.g., getting up for micturition, hot flashes, sleep apnea, restless legs).“Pain in frozen shoulder”5) EnvironmentDisturbances related to external stimuli (e.g., noise, light, temperature, pets/partners/roommates/co-habitators/neighbor activity).“My [husband] snore[s] loudly”6) OtherDisturbances not covered by the categories above.“Night shift”Codes were applied to classify free-text responses into predefined categories, with examples provided for clarity. The table includes three columns: “Code”, which lists the codes by name; “Label”, which describes the potential reasons for sleep disturbance that would fall under that particular code; and “Example”, which describes a text example of the code in question Deductive coding framework for responses to PSQI item 5j (“other reasons for sleep disturbance”) Codes were applied to classify free-text responses into predefined categories, with examples provided for clarity. The table includes three columns: “Code”, which lists the codes by name; “Label”, which describes the potential reasons for sleep disturbance that would fall under that particular code; and “Example”, which describes a text example of the code in question Fear of hypoglycemia was measured using the Hypoglycemia Fear Survey–II (HFS-II) short form, which provides total, worry, and behavior subscale scores. Participants self-reported their most recent HbA1c. A subset of AID users also contributed device data through Nightscout or AndroidAPS, with which we assessed the following CGM-derived glycemic metrics: mean glucose, standard deviation glucose (SD), coefficient of variation (CV), percentage of time in range (TIR: 70–180 mg/dL; 3.9–10.0 mmol/L), time in tight range (TITR: 70–140 mg/dL; 3.9–7.8 mmol/L), time above range (TAR: >180 mg/dL; 10.0 mmol/L)—including levels 1 (TAR1: 181–250 mg/dL; 10.1–13.9 mmol/L) and 2 (TAR2: >250 mg/dL; >13.9 mmol/L)—and time below range (TBR: <70 mg/dL; 3.9 mmol/L)—also including levels 1 (TBR1: 54–69 mg/dL; 3.0–3.9 mmol/L) and 2 (TBR2: <54 mg/dL; <3.0 mmol/L). Statistical analysis Analyses were performed in R (v4.5.1). Device data were cleaned and aligned to the PSQI reporting window to create complete glucose histories. Participants with incomplete survey data or insufficient device data were excluded listwise from the corresponding analyses. For group comparisons along questionnaire scores we used Mann–Whitney U tests; open-format responses were first scored and sorted into deductive coding categories, then analyzed using χ2 or Fisher’s exact tests, reported with odds ratios and 95% confidence intervals where appropriate. Associations between continuous variables were assessed using Spearman’s rank correlation. Two-tailed p-values <0.05 were considered statistically significant after applying a Holm-Bonferroni correction for multiple comparisons within the same category and cohort, where relevant (unadjusted p-values are reported as such). Effect size was assessed using the impact effect size, which is a more robust measure for non-normally distributed data [33]. Reporting guidelines This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for observational studies, specifically cross-sectional studies [34]. Results Participant characteristics A total of 694 adults participated in the survey (AID users: n = 521, non-users: n = 173). Of these, 165 participants (AID users: n = 100, non-users: n = 65) were excluded due to incomplete PSQI or HFS-II responses. The final analytic sample comprised 529 adults (AID users: n = 421; non-users: n = 108). Most participants identified as White (91.5%), with small proportions reporting Asian, Arab, Black/African American, Hispanic/Latino, mixed, or other ethnic backgrounds. The majority lived with T1D (90.4%), with smaller groups reporting T2D (0.8%), LADA (3.8%), or MODY (0.9%). The mean age was 43.8 ± 12.6 years for AID users and 41.0 ± 13.7 years for non-users. Women were more frequent in the non-user group (61.1%) than in the AID group (41.1%). Baseline demographic characteristics are summarized in Table 2. Table 2Baseline demographics and outcome measures of study participants Overall (n = 529) Users (n = 421) Non-users (n = 108) Age (years), mean ± SD 43.2 ± 12.843.8 ± 12.641.0 ± 13.7 Duration of diabetes (years), mean ± SD 25.3 ± 13.326.6 ± 13.020.2 ± 13.3 Body mass index (kg/m 2 ), mean ± SD 25.9 ± 4.726.1 ± 4.725.1 ± 4.6 Sex, n (%) Women239 (45.2)173 (41.1)66 (61.1)Men289 (54.6)248 (58.9)41 (38.0)Not reported1 (0.2)0 (0)1 (0.9) Type of diabetes, n (%) Type 1498 (94.1)403 (95.7)95 (88.0)LADAa20 (3.8)12 (2.9)8 (7.4)Type 24 (0.8)2 (0.5)2 (1.9)MODYb5 (0.9)3 (0.7)2 (1.9)Not reported2 (0.4)1 (0.2)1 (0.9) Diabetes management, n (%) Open-source AIDc401 (75.8%)401 (95.2%)0 (0%)Commercial AIDd20 (3.8%)20 (4.8%)0 (0%)SAPTe21 (4%)1 (0.2%)20 (18.5%)rtCGM/isCGMf73 (13.8%)2 (0.5%)71 (65.7%)SMBGg38 (7.2%)4 (1%)34 (31.5%)Insulin pump37 (7%)1 (0.2%)36 (33.3%)MDIh45 (8.5%)0 (0%)45 (41.7%)Otheri10 (1.9%)1 (0.2%)9 (8.3%) Ethnicity, n (%) White484 (91.5)389 (92.4)95 (88.0)Mixed/Multiple ethnic groups/Other19 (3.6%)15 (3.6%)4 (3.7%)I’d rather not say8 (1.5)6 (1.4)2 (1.9)Arab6 (1.1)4 (1.0)2 (1.9)Asian6 (1.1)2 (0.5)4 (3.7)Hispanic/Latino5 (0.9)4 (1.0)1 (0.9)Black/African American1 (0.2)1 (0.2)0 (0) Education, n (%) Undergraduate degree or equivalent195 (36.9%)147 (34.9%)48 (44.4%)Post-graduate degree or equivalent224 (42.3%)185 (43.9%)39 (36.1%)High school degree or equivalent52 (9.8)41 (9.7)11 (10.2)Some college, no degree38 (7.2)30 (7.1)8 (7.4)Less than high school diploma8 (1.5)7 (1.7)1 (0.9)I’d rather not say6 (1.1)5 (1.2)1 (0.9)None of the above6 (1.1)6 (1.4)0 (0) Person-reported outcome measures, median [IQR] Pittsburgh Sleep Quality Index (PSQI)†5.0 [4.0–8.0]5.0 [3.0–7.0]7.0 [5.0–9.0]Hypoglycemia Fear Survey II (HFS-II)‡8.0 [4.0–13.0]7.0 [3.0–11.0]13.0 [8.0–19.0]HbA1c, % [IQR]6.2 [5.8–6.7]6.1 [5.8–6.5]6.8 [6.2–7.2]The table summarizes demographic characteristics [such as age, sex, type of diabetes, diabetes management, ethnicity, and education] for the overall cohort, users of AID systems (both commercial and open-source) and non-users. Person-reported outcome measures are reported following demographicsaLADA: Latent Autoimmune Diabetes in AdultsbMODY: Maturity-Onset Diabetes in the YoungcOS-AID: Open-source automated insulin delivery (AndroidAPS, Loop, OpenAPS)dC-AID: Commercial Automated Insulin Delivery (Medtronic 640G, Medtronic 670G, Tandem Control-IQ)eSAPT: Sensor-Augmented Pump Therapy (Tandem, Insulet)frtCGM/isCGM: real-time CGM / intermittent-scanning CGMgSMBG: Self-Monitored Blood GlucosehMDI: Multiple Daily [Insulin] InjectionsiOther: Inhaled insulin, SGLT2-inhibitors, and other unlisted methods †Impact effect size PSQI: 0.17 ‡Impact effect size HFS-II: 0.06 Baseline demographics and outcome measures of study participants The table summarizes demographic characteristics [such as age, sex, type of diabetes, diabetes management, ethnicity, and education] for the overall cohort, users of AID systems (both commercial and open-source) and non-users. Person-reported outcome measures are reported following demographics LADA: Latent Autoimmune Diabetes in Adults MODY: Maturity-Onset Diabetes in the Young OS-AID: Open-source automated insulin delivery (AndroidAPS, Loop, OpenAPS) C-AID: Commercial Automated Insulin Delivery (Medtronic 640G, Medtronic 670G, Tandem Control-IQ) SAPT: Sensor-Augmented Pump Therapy (Tandem, Insulet) rtCGM/isCGM: real-time CGM / intermittent-scanning CGM SMBG: Self-Monitored Blood Glucose MDI: Multiple Daily [Insulin] Injections Other: Inhaled insulin, SGLT2-inhibitors, and other unlisted methods †Impact effect size PSQI: 0.17 ‡Impact effect size HFS-II: 0.06 Sleep quality The median global PSQI score for all respondents was 5 [4–8], with 72% scoring >5, indicating poor sleep quality. AID users reported significantly better subjective sleep quality than non-users, with median PSQI scores of 5 [IQR 3–7] and 7 [IQR 5–9], respectively (p < .001). Despite this difference, 43% of AID users and 69% of non-users exceeded the poor sleep threshold (p < .001). Women reported significantly poorer sleep quality than men across the entire cohort (median PSQI 6 [IQR 4–8] vs. 5 [IQR 3–7]; p < .01). This effect persisted within the AID subgroup. Participants with T2D or MODY demonstrated higher PSQI scores than those with T1D or LADA (median 7 [IQR 6–12] vs. 5 [IQR 4–8]; p = .045, unadjusted). Within AID users alone, however, sleep quality did not differ significantly by diabetes type. Users of open-source AID systems achieved significantly lower PSQI scores than commercial users (median 5 [IQR 3–7] vs. 8 [IQR 6.75–9.25], p = .001, unadjusted). Neither age nor diabetes duration were significantly correlated to PSQI scores. PSQI score distributions are visualized as violin plots in Figure 1. Violin plots of Pittsburgh Sleep Quality Index (PSQI) scores. These violin plots show PSQI score distributions for the entire study cohort, as well as gender, diabetes type, AID type, and AID user/non-user subgroups. A horizontal line at 5 marks the threshold to separate “good sleep” PSQI scores (<5) from “poor sleep” PSQI scores. The overall cohort had a median score of 5 out of a maximum possible score of 21; women scored higher than men, people with Type 2 or MODY scored higher than those with Type 1 or LADA, commercial AID users scored markedly higher than open-source users, and non-users of AID scored higher than users. LADA: Latent Autoimmune Diabetes in Adults. MODY: Maturity-Onset Diabetes of the Young. OS-AID: Open-source automated insulin delivery (AndroidAPS, Loop, OpenAPS). C-AID: Commercial Automated Insulin Delivery (Medtronic 640G, Medtronic 670G, Tandem Control-IQ). User: User of OS-AID or C-AID. Non-user: incl. SAPT, rtCGM/isCGM, SMBG, Insulin pump, MDI, Inhaled insulin, SGLT2-inhibitors, and other unlisted methods. Analysis of free-text responses to PSQI item 5j yielded 142 coded entries from 123 participants (103 entries from AID users, 39 entries from non-users). Inter-rater reliability was high (Krippendorff’s alpha = 0.95) [35,36]. The most frequently reported themes were physical-health issues (n = 44), childcare-related disturbances (n = 23), and issues with stress or mental health (n = 19) as well as diabetes- (n = 18) or diabetes-technology-related (n = 23) factors. Diabetes-related reasons for sleep disturbance were significantly more common among non-users than AID users (OR [95% CI]: 0.18 [0.06, 0.52], p < .05). The frequencies of diabetes-technology related disturbances did not differ significantly between both groups. Results from other examined categories also showed equal frequencies in users and non-users and were not significant. Fear of Hypoglycemia AID users demonstrated markedly lower total HFS-II scores than non-users (median 7 [IQR 3–11] vs. 13 [IQR 8–19]; p < .001). Subscale analyses showed lower scores in AID users for both behaviour (3 [IQR 1–5] vs. 4.5 [IQR 2–7], p < .001) and worry subscales (4 [IQR 2–7] vs. 7.5 [IQR 5–11], p < .001). Women reported higher fear-of-hypoglycemia scores than men (9 [IQR 6–15] vs 7 [IQR 3–11]; p < .001). Item-level analyses revealed that non-users were particularly more likely to report worries about experiencing hypoglycemia during sleep (2 [1–3] vs. 1 [0–1], p < .001) and not recognizing a low blood glucose (2 [1, 2] vs. 1 [0–1], p < .001). PSQI and HFS-II scores were positively correlated (ρ = 0.32, p < .001). Associations were consistent within AID users (ρ = 0.26, p < .001) and non-users (ρ = 0.34, p < .001). Binary (≥5 and < 5) PSQI scores and HFS-II scores remained significantly positively correlated (ρ = 0.26, p < .001); both for users (ρ = 0.20, p < .001) and non-users (ρ = 0.24, p < .05). HFS-II scores were not significantly different between users of open-source and commercial AID systems. Age (ρ = -0.20, p < .001) and diabetes duration (ρ = -0.20, p < .001) were significantly negatively correlated to HFS-II scores; however, this relationship was not observed in the AID subgroup, nor did controlling for age and diabetes duration impact other HFS-II analyses. HFS-II score distributions are visualized as violin plots in Figure 2. Violin plots of Hypoglycemia Fear II short form (HFS-II) scores. These violin plots show HFS-II score distributions for the entire study cohort, as well as gender, diabetes type, AID type, and AID user/non-user subgroups. The overall cohort had a median score of 8 out of a maximum possible score of 44; women scored higher than men, and non-users of AID scored higher than users. There were insignificant differences between diabetes types and AID types. Score distributions tended to skew lower for the entire cohort and across subgroups, with similar interquartile ranges throughout. LADA: Latent Autoimmune Diabetes in Adults. MODY: Maturity-Onset Diabetes of the Young. OS-AID: Open-source automated insulin delivery (AndroidAPS, Loop, OpenAPS). C-AID: Commercial Automated Insulin Delivery (Medtronic 640G, Medtronic 670G, Tandem Control-IQ). User: User of OS-AID or C-AID. Non-user: incl. SAPT, rtCGM/isCGM, SMBG, Insulin pump, MDI, Inhaled insulin, SGLT2-inhibitors, and other unlisted methods. Glycemic outcomes HbA1c values were self-reported with a median gap of 79 [45–138] days relative to the date of questionnaire responses, with a range of 5–2318 days. Median HbA1c was lower in AID users (6.1% [5.8–6.5]) than in non-users (6.8% [6.2–7.2]; p < .001). HbA1c showed a modest positive correlation with both PSQI (ρ = 0.20, p < .001) and HFS-II (ρ = 0.28, p < .001); these associations were evident among AID users (PSQI: ρ = 0.13, p < .05; HFS: ρ = 0.21, p < .001) but not significant in non-users. Users of open-source AID reported a significantly lower HbA1c than commercial users (median 6.1 [IQR 5.8–6.5] vs. 6.4 [IQR 6.1–7.1], p < .05). Neither age, nor diabetes duration correlated significantly to HbA1c. Gendered differences were—however—significant, with men reporting lower HbA1c than women (median 6.2 [IQR 5.8–6.5] vs. 6.3 [IQR 5.9–7.0], p < .05). A subsample of 60 participants in the AID subgroup provided ≥25 days of matched CGM data aligned to the PSQI reporting window (supplemental fig. S4). The results of their continuous glucose metrics (supplemental table S2) align with targets commonly recommended in international consensus statements. Neither PSQI nor HFS-II scores were associated with CGM-derived glycemic metrics in this subset (supplemental table S3). Demographics of the subsample are detailed in supplementary table S1. Discussion Results summary and interpretation In this large, multinational real-world cohort, adults using AID—both open-source and commercial—reported substantially better subjective sleep quality than non-users. The prevalence of poor sleep among AID users (43%) approached levels typically observed in general population studies (30–40%) [37–39], whereas non-users experienced markedly poorer sleep (69%), which is comparable to similar reported patient populations [40]. Fear of hypoglycemia was also substantially lower among AID users. These findings align with prior work demonstrating that AID use is associated with improved sleep-related outcomes and reduced nocturnal vigilance demands [19,41,42]. Although the between-group difference in PSQI scores was smaller than the typically cited 3-point threshold for a minimal clinically important difference [43], the markedly lower prevalence of poor sleep among AID users (43% vs. 69%) indicates a substantial clinical impact [39]. In contrast, differences in HFS-II scores exceeded thresholds considered clinically relevant, suggesting that AID use may have a more pronounced effect on alleviating fear of hypoglycemia than on global sleep quality [44]. From a clinical perspective, these results suggest that improvements in fear of hypoglycemia may represent an earlier or more sensitive benefit marker from AID use than global sleep quality scores, which may change more gradually or be less responsive to technology-related interventions. While this study’s cross-sectional design precludes any causative links, these findings suggest that AID delivery may reduce the cognitive and emotional burden of nocturnal diabetes management, thereby improving perceived sleep quality and reducing fear of hypoglycemia. PSQI and HFS-II scores were modestly correlated with HbA1c—particularly among AID users—suggesting that psychosocial and physiological factors may jointly influence sleep experiences in this population [45]. For both the PSQI and HFS-II, effect sizes were quite small at 0.17 and 0.06 respectively—while both questionnaires represent important quality of life elements to consider for people with diabetes, they are unlikely to be the primary drivers of the observed group differences between users and non-users of AID systems. Among subgroups, the difference in PSQI scores between users of open-source AID and commercial AID is notable. Whereas previous studies reported no significant sleep-related differences between users of AID systems [46], we observed a statistically significant improvement in scores from users of open-source AID compared with commercial systems. Additionally—where other researchers found open-source users to experience greater fear of hypoglycemia than commercial users—we found no significant difference [46]. We can contextualize these favourable results for open-source AID users by understanding the pivotal role these systems played in the broader adoption of AID. When and where commercial systems were unavailable or inaccessible, OpenAPS, Loop, and related systems already offered users a high degree of personalization and early adoption of new features. Those using open-source AID may therefore have more granular control of their systems and greater familiarity with system quirks compared to commercial users, which could contribute to better quality of life outcomes. They may also have greater trust in their medical devices as a consequence of their familiarity, which in turn may limit fears of nocturnal hypoglycemia. The gendered differences observed in both PSQI and HFS-II outcomes echo similar outcomes from other studies. These differences are both expected and meaningful; women with type 1 diabetes are more likely to report persistent fatigue following hypoglycemic episodes, and female sex has been independently associated with higher odds of poor sleep quality [47,48]. However, given the evidence for gendered sleep-specific differences in the broader population, future studies should not only recapitulate these results but identify specific causal factors leading to worse outcomes in those identifying as women [49]. Responses to the open-ended question of the PSQI revealed physical-health issues, childcare-related disturbances, stress or mental health concerns, and diabetes- or diabetes-technology-related issues to be the most commonly reported reasons for sleep disturbance. Non-users reported diabetes-related reasons for sleep disturbance significantly more often than users of AID—namely waking to treat high and/or low blood glucose—suggesting that AID use may curtail these disruptions for most users. This pattern supports the interpretation that diabetes-specific nocturnal disruptions remain a prominent burden for non-users but appear less common among individuals using AID. However, diabetes-technology-related disturbances did not differ significantly between groups, despite the assumption that such issues might be more common in AID users due to the additional components involved, especially in open-source systems. These observations should be interpreted cautiously. Free-text reporting is vulnerable to recall bias, language differences, and selective responding, and many participants provided no response to item 5j. Accordingly, these findings should be considered exploratory, highlighting the need for more sensitive, diabetes-specific sleep assessments capable of capturing nocturnal disruptions such as CGM and pump alerts, glucose fluctuations, and diabetes-related vigilance more completely. Finally, this study extends prior work by including people with T2D, LADA and MODY in the assessment of PSQI and HFS-II outcomes among people using AID systems, although subgroup sizes were small and analyses were primarily descriptive. Interpretation of the CGM subsample A subset of 60 participants using AID systems contributed matched CGM data, enabling exploratory analysis of glucose patterns in relation to subjective sleep quality and fear of hypoglycemia. This group demonstrated glucose metrics well above internationally recommended targets, as well as relatively low PSQI and HFS-II scores with limited variability. Notably, the subsample differed demographically from the broader cohort: individuals who donated device data were more often men, identified as White, and were predominantly based in Germany. Because data donation required navigating several additional technical steps (e.g., exporting and uploading CGM files via Nightscout or AndroidAPS), this process may have introduced a participation barrier that selectively reduced data contribution from other demographic groups. Given these structural differences and the restricted variability in both glycemic measures and PROMs, the absence of significant associations between CGM metrics and subjective sleep outcomes should be interpreted with caution. These findings likely reflect limited variance and the non-representativeness of this subsample rather than a true lack of relationship. They highlight the need for more inclusive data-donation approaches and future studies that integrate objective sleep measures in larger and more diverse populations. Strengths This study has several notable strengths. First, it represents one of the largest real-world studies to date examining subjective sleep quality and fear of hypoglycemia in adults with diabetes; spanning both open-source and commercial AID users, as well as non-users. By leveraging data from the OPEN project—an international, patient-led research initiative—the study captures a broad range of experiences across people with diabetes using different treatment modalities and geographical regions. Second, the simultaneous assessment of sleep quality, fear of hypoglycemia, clinical outcome measures, device data, and qualitative free-text responses provides a comprehensive perspective on the multidimensional nature of sleep in diabetes. The integration of quantitative and qualitative methods strengthens the interpretation of findings and offers insight into the specific nocturnal disruptions that people with diabetes experience. Third, free-text responses were systematically analyzed using a predefined, transparent coding framework with excellent inter-rater reliability, enhancing the robustness of the qualitative analysis. Few studies in diabetes or sleep medicine have examined open-ended reasons for sleep disturbance at this scale. Fourth, distinguishing between open-source and commercial AID systems allowed us to examine whether different technological approaches relate differently to sleep and psychosocial outcomes–an area of growing relevance as AID technologies continue to evolve and diversify. Finally, the study was preregistered, followed established ethical procedures, and made analysis code openly available, supporting transparency and reproducibility. Together, these strengths position the findings as an important contribution to understanding the lived experience of sleep among adults with diabetes in the context of modern diabetes technologies. Limitations Several limitations should also be acknowledged. The OPEN project was designed to examine the experiences and outcomes of people using open-source AID systems, and recruitment occurred primarily through online communities. As a result, the sample is highly purposive and may not be representative of the broader population of people living worldwide with diabetes. Participants were predominantly White, highly educated, and from higher income brackets, reflecting known sociodemographic patterns in diabetes technology communities and introducing potential selection bias [50,51]. Further, all outcomes were derived from self-reported questionnaires, which may be susceptible to recall bias or underreporting of nocturnal disturbances, particularly diabetes-specific disruptions that may not be fully captured by existing instruments. The cross-sectional design of our study also precludes potential assessment of causal inference, limiting us to establishing associations rather than any specific causative link between demographic characteristics, diabetes outcomes and quality of life. And although this work includes a broader range of diabetes types than prior OPEN analyses, the number of adults with T2D or MODY diabetes was small—especially relative to those with T1D—limiting comparative analyses across diabetes types. It is also worth mentioning that given the absence of systematically collected biological sex data, we were limited to exploratory questionnaire sub-analyses using self-reported gender identity only. Our choice of questionnaires also presented some late-discovered limitations. While the PSQI includes a threshold cutoff between “good” and “poor” sleep, there is no universally accepted clinical cutoff for fear of hypoglycemia. Previous studies have suggested a cutoff applicable only to the worry subscale of the full-length questionnaire [whereas we used the HFS-II short form] and only within a study population of those with type 2 diabetes [52]. This limited us to dichotomisation of scores only in the PSQI, leaving the HFS-II short form as a continuous measure. Furthermore, the HFS-II short form was unsuitable for exploratory subscale analyses given that several items are missing relative to the full-length questionnaire. For example, the “Sought Safety” measure is represented by only a single remaining item in the HFS-II short form, limiting the interpretability and psychometric validity. We therefore decided not to perform potentially underpowered or structurally incomplete subscale analyses. The markedly higher HbA1c values and PSQI scores observed among commercial AID users should be viewed with some caution, as this subgroup comprised only 20 individuals. Comparisons of clinical outcomes between open-source and commercial AID users should therefore be considered as exploratory rather than confirmatory. Additionally, HbA1c reflects average glycemia over ~3 months and does not adequately capture hypoglycemia frequency, variability, or nocturnal events, which may be particularly relevant to sleep quality and fear of hypoglycemia. Self-reporting of HbA1c also limited interpretability of our results, given the time lag between test validity and study commencement. However, concurrent HbA1c testing was not feasible given our international cohort, and remote work structure. Despite these limitations, lower HbA1c values among open-source users align with existing literature, and observed differences in sleep outcomes suggest that greater customisability and interoperability of open-source AID systems may contribute to improved outcomes. Between users and non-users of AID, the frequency of diabetes-technology-related disturbances did not significantly differ. Despite the suggested benefits of AID systems, individuals must still contend with alarms and device troubleshooting when problems do inevitably arise; factors known to interfere with sleep quality in this population [53–55]. Future work should examine which system-level and user-experience factors, such as alarm configuration, automation aggressiveness, and personalization, most strongly influence sleep and quality-of-life outcomes. Implications for sleep assessment in diabetes An important observation was that relatively few participants explicitly cited diabetes or diabetes technology as reasons for nocturnal disturbance in the PSQI open-ended item, despite diabetes being widely recognized as a contributor to disrupted sleep. This may be a consequence of how people living with chronic conditions recalibrate their perception of what “good” quality of life is; in this case, people with diabetes having a skewed perception of what constitutes “normal” sleep, after so many years of regular disruptions, alarms, and fatigue. This could potentially lead to underreporting of disease-specific disturbances, i.e., what we saw in the PSQI open-ended item. While the PSQI is commonly used in psychosocial outcome research in diabetes, our results suggest there is a need for a validated, chronic disease-sensitive sleep questionnaire, which has the granularity needed to detect differences in diabetes-related sleep experiences, particularly in technologically supported populations. The PSQI was not designed to capture condition-specific nocturnal disruptions such as device alerts, glucose fluctuations, or diabetes-related vigilance, and chronic diseases can only be mentioned in item 5j, limiting their influence on the overall score. Specific reporting on obstructive sleep apnea—which is missing in both the PSQI and other study questionnaires—is especially important to include in the context of sleep health for people with diabetes, as are other chronic sleep disorders. A chronic disease-sensitive questionnaire would allow for understanding sleep quality in the context of diabetes, and could also be effective for future work integrating subjective assessments with objective metrics such as actigraphy or device-logged nocturnal alarms. Together, these limitations highlight the need for future research using more diverse recruitment strategies, validated diabetes-specific sleep instruments, and longitudinal or mixed-methods designs to better understand the interplay between psychosocial experiences, sleep, and diabetes technology use. Conclusion Adults using AID systems reported better subjective sleep quality and markedly lower fear of hypoglycemia than non-users. While associations between sleep, psychosocial outcomes, and glycemia were modest, they underscore the complex interplay between physiological and psychosocial factors shaping sleep-related experiences of people living with diabetes. Digital biomarkers (like those derived from actigraphy-based wearables), greater assessment of user experience vis à vis system alerts and usability, and longitudinal analyses across more diverse populations would all go towards developing actionable insights for people with diabetes using AID systems. Ultimately, these findings highlight the importance of incorporating sleep-related outcomes into diabetes care, research, and technology evaluation, and they point to a need for more sensitive diabetes-specific sleep measures and inclusive multimodal studies that integrate subjective, objective, and device-derived data. Supplementary Material