Document resource
WHAT IS ALREADY KNOWN ON THIS TOPIC Hyperosmolar hyperglycemic state (HHS) is a life-threatening diabetic emergency, yet no national surveillance system exists and guideline adherence is unknown.WHAT THIS STUDY ADDS We developed a surveillance model to monitor real-time trends in patient outcomes and adherence to clinical guidelines. Age, serum sodium, urea, and osmolality emerged as key predictors of mortality, supporting their role in risk stratification, and inter-hospital variation in guideline adherence identified current lapses in performance standards.HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE, OR POLICY This study demonstrates a scalable and reproducible model for standardizing HHS surveillance, with data that can inform future guidelines and strengthen a sparse evidence base on HHS management.Introduction Hyperosmolar hyperglycemic state (HHS) is a life-threatening acute metabolic complication most commonly observed in individuals with type 2 diabetes mellitus, though it can occur across the diabetes spectrum. 1–3 It is characterized by profound hyperglycemia, marked hyperosmolarity, and severe dehydration resulting from sustained osmotic diuresis secondary to relative insulin deficiency.1 2 In contrast to diabetic ketoacidosis (DKA), HHS typically presents with minimal or absent ketonemia and acidosis. Despite a lower prevalence than DKA, HHS carries a significantly higher mortality rate—up to 10 times greater in some reports—due to its insidious onset, delayed recognition, and more severe biochemical derangements.1 4While population-level data from the USA suggest a declining trend in inpatient HHS mortality, with recent estimates as low as 0.77%, the generalizability of these findings to other healthcare systems is unclear.5 In the UK and Europe, contemporary evidence on the epidemiology, clinical management, and outcomes of HHS remains scarce. Existing studies are frequently limited to single-center or regional datasets, rely on retrospective designs, and are largely descriptive in nature.6–9 Consequently, a critical gap remains in the literature to inform high-quality, evidence-based care pathways for HHS within the UK context.Current clinical guidance in the UK, issued by the Joint British Diabetes Society (JBDS), provides structured recommendations for managing HHS.10 However, these guidelines are predominantly consensus-driven and lack a strong empirical foundation due to the paucity of large-scale, multicenter data. Moreover, there is no dedicated national surveillance mechanism to monitor real-world adherence to these guidelines or to assess variation in care delivery across institutions.10 11 The National Diabetes Inpatient Safety Audit, while valuable, does not capture the granularity required to evaluate HHS-specific processes and outcomes.12 This limits the ability of healthcare systems to benchmark performance, identify areas of suboptimal care, or implement evidence-informed improvements.Variability in the implementation of guideline-based care for HHS is well documented, often reflecting systemic and behavioral barriers at both individual and organizational levels.13 Such variation in practice can lead to suboptimal outcomes, including prolonged hospital stay, preventable complications, and increased mortality.5 14 Our prior work has demonstrated that routine feedback on key process and outcome measures can significantly enhance adherence to best-practice guidelines and reduce unwarranted variation in inpatient diabetes care.15 In this context, there is an urgent need to develop and implement a structured, multicenter surveillance system to systematically evaluate HHS care and drive continuous quality improvement. Therefore, we conducted this study:To develop and implement a standardized multicenter surveillance system for the systematic collection of data on clinical presentation, management practices, and outcomes in patients with HHS.To assess interhospital variation in adherence to national HHS management guidelines and its association with patient outcomes.To explore the barriers and facilitators influencing adoption and sustained implementation of the HHS surveillance model across diverse clinical settings.Methods This was a multicenter, mixed-methods observational study conducted across 12 acute National Health Service (NHS) hospitals in the UK between January 2021 and November 2024. The study aimed to establish and evaluate a standardized surveillance system for the management of HHS. Each participating site secured local institutional approval and registered the project in line with clinical governance frameworks (registration numbers: Hereford Hospital (HHS-QIP 011), Ipswich Hospital (SE-MEDIPS23-1553), Norfolk and Norwich University Hospital (DIAB-22–23 A08), Russells Hall Hospital (Diab/QI/2023-24/04), Royal Free Hospital London (RFHBU_79623/24), Sandwell and West Birmingham Hospitals (SG1913), University Hospitals Birmingham (CARMS 20986), Walsall Manor Hospital (QI20-21/LTC/01), Warwick Hospital (2593)). Data collection adhered to Caldicott principles and institutional data governance protocols to ensure patient confidentiality and compliance with national data protection regulations. 16Development of the HHS surveillance system The HHS surveillance model was developed using the Digital Evaluation of Ketosis and Other Diabetes Emergencies (DEKODE)-DKA framework established by the DEVI collaboration at the University of Birmingham, which facilitates structured data collection to monitor adherence to national guidelines and identify variation in care delivery. 17 We codesigned a standardized data collection tool—a secure, structured Google Form—through an expert consultation process involving contributors to the JBDS HHS guidelines. The form was piloted at a lead site to assess usability, data completeness, and interpretability. Based on iterative feedback from clinicians, revisions were made to enhance clarity, particularly in recording temporally sequenced clinical interventions. Final domains captured included: demographics (age, sex, ethnicity, diabetes type, Body Mass Index, Charlson Comorbidity Index (CCI), HbA1c, preadmission insulin requirements in 24 hours, diabetes pharmacotherapy and residential status), precipitating factors (intercurrent illness, dehydration, poor adherence, steroid-induced hyperglycemia, alcohol-related causes, or new-onset diabetes), biochemical parameters (pH, bicarbonate, osmolality, ketones, glucose, lactate, urea, electrolytes), management (fluid volumes, administration of 0.450% saline, insulin dosing, frequency of capillary blood glucose and ketone monitoring), and outcomes (time to resolution of HHS, time to medically fit for discharge, delays in discharge, complications (eg, hypoglycemia, intensive care unit admission), and in-hospital mortality. Data were pseudonymized at the point of entry using site-specific identifiers, with no patient-identifiable information shared externally.Case identification and inclusion criteria Adult inpatients (≥18 years) admitted with suspected HHS were identified using combined informatics-based methods: (1) diagnostic coding via ICD-10-CM (E11.00), and (2) pharmacy records indicating initiation of fixed-rate intravenous insulin infusions. All potential cases underwent screening against diagnostic criteria defined by the JBDS guidelines, which included serum osmolality ≥320 mOsm/kg, serum glucose ≥30 mmol/L, absence of significant ketonemia (≤3.0 mmol/L), and absence of acidosis (pH ≥7.3 and bicarbonate ≥15 mmol/L). 10 Patients with mixed DKA-HHS presentations or those not meeting the above criteria were excluded to maintain diagnostic fidelity.Implementation and roll-out across sites Following pilot refinement, the surveillance tool was disseminated to all centers engaged in the DEKODE initiative. Structured training sessions were delivered to individuals expressing interest in participating in the DEKODE HHS surveillance model by experienced DEKODE team members, ensuring a uniform understanding of JBDS-aligned diagnostic criteria, data definitions, and form usage. 10 Training materials and walkthroughs were shared digitally and reinforced through follow-up consultations to maintain fidelity to protocol.Comparative analysis of clinical practice To assess intersite variation, individual hospital-level data were benchmarked against median values aggregated across all participating centers. Comparative analyses were also stratified by hospital size and population demographics to control for contextual variability. Each site received anonymized, tailored feedback highlighting areas of congruence or divergence from national guideline recommendations.Evaluation of implementation: barriers and facilitators To gain insight into the implementation experience, we administered structured surveys to the clinical teams involved at each site. Survey domains included perceived utility of the surveillance system, ease of integration into existing workflows, organizational readiness, and barriers to data collection. Respondents were invited to participate in semistructured interviews to explore these domains further in depth.Implementation success was evaluated using the Reach, Effectiveness, Adoption, Implementation, Maintenance (RE-AIM) framework.18 This structured model enabled multidimensional assessment of feasibility, sustainability, and potential for scale-up.Data analysis Quantitative analysis Descriptive statistics were computed for all clinical and outcome variables. Continuous variables were reported as medians with IQRs and compared using Wilcoxon rank-sum tests following assessment for normality via the Shapiro-Wilk test. Categorical variables were reported as counts (N) with frequencies and compared using χ 2 or Fisher’s exact tests, as appropriate. Multivariable logistic regression was performed to identify independent predictors of in-hospital mortality. Variables were selected based on clinical relevance and univariate significance. Calculated osmolality was excluded from the final model due to conceptual collinearity with included variables (sodium and urea). Multicollinearity was assessed using variance inflation factors (VIFs), with all predictors demonstrating acceptable levels (VIFs <1.2). Model performance was assessed using the Hosmer-Lemeshow goodness-of-fit test and Nagelkerke’s R². Results are presented as adjusted ORs (aORs) with 95% CIs and p values. All statistical analyses were conducted using IBM SPSS Statistics for Mac, V.29.0.1.1 (IBM, Armonk, New York, USA). A two-tailed p value of <0.05 was considered statistically significant.Qualitative analysis Interview data were transcribed verbatim and analyzed inductively using thematic analysis. Coding was conducted independently by two researchers using NVivo V.14.24.3 (QSR International, 2025). Discrepancies were resolved through discussion to achieve consensus. Emergent themes were mapped to the RE-AIM framework and synthesized to generate actionable insights for system refinement and wider implementation. 18Results Implementation of the DEKODE-HHS surveillance model The DEKODE-HHS surveillance system was successfully implemented across 12 acute NHS trusts in England, aligning with audit principles from the National Institute for Health and Care Excellence (NICE) and the Healthcare Quality Improvement Partnership (HQIP) ( online supplemental file 1).19 20 The surveillance model demonstrated feasibility for identifying, monitoring, and evaluating the management and outcomes of HHS across a range of hospital settings.SP110.1136/bmjdrc-2025-005489.supp1Supplementary dataFrom 245 patients initially identified through coding and pharmacy records, 218 episodes met JBDS-IP diagnostic criteria for HHS and were included in the final analysis.10 Excluded cases included miscoded diagnoses (n=17), mixed DKA/HHS presentations (n=6), and incomplete records (n=4).Improvement initiatives driven by DEKODE-HHS feedback At Hospital A, early engagement of diabetes specialist nurses, use of real-time feedback loops, and adaptation of JBDS guidance into a streamlined, color-coded A4 protocol were facilitated by the feedback from DEKODE HHS surveillance. 10 Barriers to improving HHS care included workforce shortages, intermittent access to point-of-care testing equipment, and variable interdepartmental communication. Clinicians reported challenges with the practical application of fluid resuscitation and insulin titration protocols, particularly in out-of-hours settings. To address these challenges, the local guideline was revised to include simplified biochemical monitoring intervals and highlighted early referral to diabetes specialist teams. Updates to local guidelines were informed by the 2024 American Diabetes Association (ADA) consensus on hyperglycemic emergencies to align with current international standards and were piloted successfully at Hospital A, with plans for scaling up across other sites.21Characteristics of the HHS cohort The median age of included patients was 77.0 years (IQR 64.0–85.0), and 84.4% (n=184/218) had type 2 diabetes. New diagnoses of diabetes accounted for 8.7% (n=19/218) of cases, while 2.4% (n=5/218) had secondary forms of diabetes (eg, post-transplant, steroid-induced). Preadmission glycemic control was suboptimal, with a median HbA1c of 81 mmol/mol (IQR 61.5–116). 41.7% (n=91/218) of patients required insulin therapy, with a median daily dose of 30 units (IQR 16–57.5) prior to admission. Metformin was the most commonly used oral antiglycemic agent (37.2%, n=81/218), and 41.3% (n=90/218) of patients were not receiving any glucose-lowering therapy prior to admission for HHS. The burden of multimorbidity was high, with a median CCI score of 6 (IQR 4–7). The ethnic profile broadly reflected regional diversity: White (57.8%, n=126/218), Asian (19.3%, n=42/218), and Black (10.5%, n=23/218). Most patients (37.2%, n=81/218) were living at home prior to admission, while the remainder resided in long-term care facilities, such as nursing or care homes ( table 1).Table 1Baseline characteristics, precipitating factors, biochemical parameters at diagnosis and management of the overall cohort and comparison with Hospital ACharacteristicsOverall(n=218)Hospital A(n=44)P valueAge, median (IQR)77.0 (64.0–85.0)73.0 (60.25–83.75)0.006Gender, n (%)0.07 Male104 (47.7)27 (61.4) Female114 (52.3)17 (38.6)Ethnicity, n (%) 0.326 White126 (57.8)32 (72.7) Asian42 (19.3)6 (13.6) Black23 (10.5)5 (11.4) Other ethnic group3 (1.4)0 (0.0) Not stated24 (11.0)1 (2.3)BMI, median (IQR)25.7 (22.0–29.2)26.3 (22–30.9)0.496Type of diabetes, n (%)0.193 Type 110 (4.6)7 (15.9) Type 2184 (84.4)34 (77.3) Type 3c3 (1.4)0 (0.0) First presentation19 (8.7)2 (4.5) Steroid-induced1 (0.5)0 (0.0) Post-renal transplant1 (0.5)1 (2.3)Preadmission HbA1c in mmol/mol, median (IQR)81 (61.5–116)70 (59–93)0.343Medications, n (%) None90 (41.3)21 (47.7)0.238 Insulin91 (41.7)19 (43.2)0.481 Metformin81 (37.2)16 (36.4)0.5 Dipeptidyl peptidase-4 inhbitors (DPP4i)47 (21.6)10 (22.7)0.5 Sulfonylureas30 (13.8)3 (6.8)0.130Sodium-glucose co-transporter-2 inhibitors (SGLT2i)11 (5.0)2 (4.5)0.418Preadmission insulin requirements in 24 hours, median (IQR)30 (16–57.5)28 (14.5–55.5)Residential status before admission, n (%)0.004 Own home81 (37.2)18 (40.9) Care/residential home19 (8.7)7 (15.9) Nursing home13 (6.0)7 (15.9) Assisted living1 (0.5)0 (0.0) Sheltered accommodation1 (0.5)0 (0.0) Unknown103 (47.2)12 (27.3)CCI, median (IQR) Overall6 (4-7)5.5 (4-7)0.194 Discharged alive6 (4-7)5 (4-7)0.194 In-hospital death5 (4-6)––Precipitating factors, n (%)0.131 Intercurrent illness108 (49.5)26 (59.1) Infectious diseases35 (16.0)3 (6.8) Suboptimal compliance to treatment26 (11.9)5 (11.4) Dehydration/reduced oral intake12 (5.5)5 (11.4) First presentation11 (5.0)0 (0) Steroid-induced5 (2.3)0 (0) Alcohol-related 1 (0.5)1 (2.3) Unknown20 (9.2)4 (9.1)Biochemical parameter at diagnosis, median (IQR) pH 7.35 (7.31–7.40)7.37 (7.30–7.41)0.542 Bicarbonate (mmol/L)23.05 (20.7–26.4)23.0 (19.1–25.8)0.772 Glucose (mmol/L)33.0 (30.1–38.9)33.0 (31.3–33.3)0.184 Serum osmolality (mOsm/L)354.5 (338.3–375.4)354 (328.8–364.8)0.357 Sodium (mmol/L)147.0 (139.2–154.0)145 (137–154)0.894 Potassium (mmol/L)4.6 (4.02–5.3)4.6 (4.14–4.95)0.990 Urea (mmol/L)19.35 (13.6–25.25)19.5 (14.4–25.3)0.717 Lactate (mmol/L)2.8 (2.17–3.80)3.2 (2.2–4.8)0.073 Ketones (mmol/L)0.8 (0.3–2.5)0.7 (0.1–2.4)0.104 Management, median (IQR) Total fluids (L)6.5 (4.0–9.7)6.5 (4.5–10.1)0.122 Total insulin (units)69.0 (30.8–116.0)107 (50–161.7)0.001 Adherence to hourly glucose monitoring (%)65.9 (47.5–88.0)86.33 (68.61–105.01)<0.001 Adherence to hourly ketone monitoring (%)28.9 (14.9–49.7)21.5 (13.0–31.9)0.079Administration of 0.45% saline, n (%)70 (32.1)29 (67.4)<0.001Significant p values (<.05) are highlighted in bold. BMI, Body Mass Index; CCI, Charlson Comorbidity Index; HHS, hyperosmolar hyperglycemic state.Intercurrent illness was the most commonly documented precipitant (49.5%, n=108/218), followed by infections (16.0%, n=35/218). Biochemical parameters at presentation were consistent with severe HHS: median glucose 33.0 mmol/L (IQR 30.1–38.9), sodium 147.0 mmol/L (IQR 139.2–154.0), and serum osmolality 354.5 mOsm/kg (IQR 338.3–375.4) (table 1). The median time to formal HHS diagnosis was 1.95 hours (IQR, 0.77–6.00), although 7.8% (n=17/218) were diagnosed more than 24 hours post admission. Median intravenous fluid volume administered during the acute phase was 6.5 L (IQR 4.0–9.7), and median time to resolution was 48.2 hours (IQR 24.9–74.2) (tables 1 and 2). Length of stay was 10.3 days (IQR 6.0–17.0), with common delays in discharge attributed to ongoing investigations or care coordination (table 2).Table 2Outcomes, complications, and discharge types of the overall cohort and comparison with Hospital ACharacteristicsOverall(n=218)Hospital A(n=44)P valueOutcomes, median (IQR) Time from admission to diagnosis (hours)1.95 (0.77–6.0)1.78 (0.85–4.13)0.165 HHS >24 hours after admission, n (%)17 (7.8)1 (2.3)0.146 HHS duration (hours)48.2 (24.9–74.15)53.4 (27.6–68.0)0.458 Time from HHS resolution to MFFD (days)7.13 (2.98–12.89)11.4 (4.6–16.5)0.012 Time from MFFD to discharge (days)1.0 (0.15–1.86)0.99 (0.34–1.88)0.989 Length of stay (days)10.3 (6.03–16.98)13.0 (5.77–19.05)0.145Complications, n (%) Hypoglycemic episodes32 (14.7)10 (23.3)0.085 ITU admission12 (5.5)2 (4.7)0.5 Death35 (16.1)1 (2.3)0.011Factors affecting discharge Until MFFD, n (%)0.513 Ongoing investigations and treatment 61 (28.0)22 (66.7) Glycemic control34 (15.6)6 (18.2) DSN review15 (6.9)3 (9.1) Allied health professional review6 (2.8)1 (3.0) Medical specialist review2 (0.9)0 (0.0) Awaiting IP Rehab2 (0.9)1 (3.0)After MFFD, n (%)0.538 Ongoing investigations and treatment 19 (8.7)5 (15.6) Glycemic control3 (1.4)0 (0.0) DSN review11 (5.0)3 (9.4) Allied health professional review4 (1.8)0 (0.0) Discharge letter or TTO awaited for >3 hours18 (8.3)10 (31.3) Awaiting placement/POC23 (10.6)7 (21.9) Awaiting transport8 (3.7)4 (12.5) No delay8 (3.7)3 (9.4)Significant p values (<.05) are highlighted in bold. DSN, diabetes specialist nurse; HHS, hyperosmolar hyperglycaemic state; IP, inpatient; ITU, intensive therapy unit; MFFD, medically fit for discharge; POC, package of care; TTO, to take out medications.Interhospital variation in clinical practice and outcomes Across participating sites, notable variation in management practices and outcomes was observed. In comparative analyses, Hospital A exhibited distinct patterns in several domains. Compared with the overall cohort, patients in Hospital A were younger (median 73.0 years vs 77.0 years; p=0.006), received higher total insulin doses (median 107 units vs 69 units; p=0.001), and demonstrated significantly better adherence to hourly capillary glucose monitoring (86.3% vs 65.9%; p<0.001) ( table 1).Despite a longer time from HHS resolution to medically fit-for-discharge status at Hospital A (11.4 days vs 7.13 days; p=0.012), in-hospital mortality was significantly lower compared with the overall cohort (2.3% vs 16.1%; p=0.011) (table 2). These differences persisted in a direct comparison with Hospital B, which had similar baseline characteristics but lower insulin dosing and adherence to monitoring (tables 3 and 4).Table 3Comparison of outcomes, complications, and discharge types between Hospital A and Hospital BCharacteristicsHospital A(n=44)Hospital B(n=43)P valueOutcomes, median (IQR) / n (%) Time from admission to diagnosis (hours)1.78 (0.85–4.13)2.1 (0.76–8.87)0.562 HHS >24 hours after admission1 (2.3)7 (16.7)0.03 HHS duration (hours)53.4 (27.6–68.0)55.7 (30.9–73.2)0.804 Time from HHS resolution to MFFD (days)11.4 (4.6–16.5)6.08 (2.3–12.06)0.103 Time from MFFD to discharge (days)0.99 (0.34–1.88)0.83 (0.16–2.87)0.482 Length of stay (days)13.0 (5.77–19.05)10.4 (7.03–20.86)0.815Complications, n (%) ITU admission2 (4.7)0 (0.0)0.196 Death1 (2.3)7 (16.3)0.024 Hypoglycemic episodes10 (23.3)10 (27.8)0.645Factors affecting discharge Until MFFD, n (%)0.287 Ongoing investigations and treatment 22 (66.7)18 (60) Glycemic control6 (18.2)9 (30) DSN review3 (9.1)0 (0.0) Allied health professional review1 (3.0)2 (6.7) Medical specialist review0 (0.0)1 (3.3) Awaiting IP Rehab1 (3.0)0 (0.0)After MFFD, n (%)0.521 Ongoing investigations and treatment 5 (15.6)5 (19.2) Glycemic control0 (0.0)1 (3.8) DSN review3 (9.4)3 (11.5) Allied health professional review0 (0.0)1 (3.8) Discharge letter or TTO awaited for>3 hours10 (31.3)5 (19.2) Awaiting placement/POC7 (21.9)9 (34.6) Awaiting transport4 (12.5)1 (3.8) No delay3 (9.4)1 (3.8)Significant p values (<.05) are highlighted in bold. DSN, diabetes specialist nurse; HHS, hyperosmolar hyperglycaemic state; IP, inpatient; ITU, intensive therapy unit; MFFD, medically fit for discharge; POC, package of care; TTO, to take out medications.Table 4Comparison of baseline characteristics, precipitating factors, biochemical parameters at diagnosis and management between Hospital A and Hospital BCharacteristicsHospital A(n=44)Hospital B(n=43)P valueAge, median (IQR)73.0 (60.2–83.7)77.0 (69.0–84.0)0.139Gender, n (%) 0.391 Male27 (61.4)22 (51.2) Female17 (38.6)21 (48.8)Ethnicity, n (%) 0.003 White32 (72.7)21 (48.8) Asian6 (13.6)10 (23.3) Black5 (11.4)1 (2.3) Other ethnic group0 (0.0)0 (0.0) Not stated1 (2.3)11 (25.6)BMI, median (IQR)26.3 (22–30.9)25.8 (22.1–28.7)0.674Type of diabetes, n (%)0.091 Type 17 (15.9)1 (2.3) Type 234 (77.3)41 (95.3) Type 3c0 (0.0)0 (0.0) First presentation2 (4.5)1 (2.3) Steroid-induced0 (0.0)0 (0.0) Post-renal transplant1 (2.3)0 (0.0)Preadmission HbA1c in mmol/mol, median (IQR)70 (59–93)76 (60–100)0.598Medications, n (%) None21 (47.7)12 (27.9)0.077 Insulin19 (43.2)22 (51.2)0.522 Metformin16 (36.4)17 (39.5)0.827 DPP4i10 (22.7)14 (32.6)0.345 Sulfonylureas3 (6.8)9 (20.9)0.068 SGLT2i2 (4.5)4 (9.3)0.434Preadmission insulin requirements in 24 hours, median (IQR) 28 (14.5–55.5)26 (14 - 54)0.149Residential status before admission, n (%)0.320 Own home18 (40.9)19 (44.2) Care/residential home7 (15.9)5 (11.6) Nursing home7 (15.9)2 (4.7) Assisted living0 (0.0)1 (2.3) Sheltered accommodation0 (0.0)0 (0.0) Unknown12 (27.3)16 (37.2)CCI, median (IQR) Overall5.5 (4-7)6 (4–7.75)0.674 Discharged Alive5 (4-7)6 (4-8)0.403 In-hospital Death–5 (4-7)–Precipitating factors, n (%)0.627 Intercurrent illness26 (59.1)26 (60.5) Infectious diseases3 (6.8)7 (16.3) Suboptimal compliance to treatment5 (11.4)3 (7) Dehydration/reduced oral intake5 (11.4)4 (9.3) Steroid-induced0 (0)1 (2.3) Alcohol-related 1 (2.3)0 (0) Unknown4 (9.1)2 (4.7)Biochemical parameter at diagnosis, median (IQR) pH 7.37 (7.30–7.41)7.37 (7.32–7.40)0.607 Bicarbonate (mmol/L)23.0 (19.1–25.8)22.1 (20.9–25.2)0.989 Glucose (mmol/L)33.0 (31.3–33.3)32.5 (29.5–35.8)0.852 Serum osmolality (mOsm/L)354 (328.8–364.8)357 (345–380.9)0.048 Sodium (mmol/L)145 (137–154)149 (143–159)0.273 Potassium (mmol/L)4.6 (4.14–4.95)4.5 (3.86–5.26)0.805 Urea (mmol/L)19.5 (14.4–25.3)19.7 (13.6–30.65)0.566 Lactate (mmol/L)3.2 (2.27–4.8)2.55 (2.18–3.68)0.452 Ketones (mmol/L)0.7 (0.1–2.4)0.6 (0.2–1.8)0.647 Management, median (IQR) Total fluids (L)6.5 (4.5–10.1)7 (5.3–8.0)0.653 Total insulin (units)107 (50–161.7)65 (27–101)0.016 Adherence to hourly glucose monitoring (%)86.33 (68.61–105.01)64.88 (52.5–74.6)<0.001 Adherence to hourly ketone monitoring (%)21.5 (13.0–31.9)27.9 (9.53–41.0)0.406Administration of 0.45% saline, n (%)29 (67.4)17 (47.2)0.108Significant p values (<.05) are highlighted in bold. BMI, Body Mass Index; CCI, Charlson Comorbidity Index; HHS, hyperosmolar hyperglycaemic state.Univariate logistic regression identified older age, higher sodium, urea, and serum osmolality at presentation as significant predictors of in-hospital mortality. In the final multivariable logistic regression model, older age (aOR 1.049 per year; 95% CI 1.012 to 1.087; p=0.015), higher serum sodium at presentation (aOR 1.043 per mmol/L; 95% CI 1.009 to 1.081; p=0.016), and higher serum urea (aOR 1.051 per mmol/L; 95% CI 1.006 to 1.097; p=0.024) remained independent predictors of in-hospital mortality. The mortality difference between Hospitals A and B did not retain significance after adjustment (OR 5.255; 95% CI 0.589 to 46.897; p=0.137). The model showed good calibration (Hosmer-Lemeshow χ² = 6.431, p=0.599) and moderate explanatory power (Nagelkerke R² = 0.236), with no evidence of multicollinearity (all VIFs <1.2) (table 5).Table 5Multivariable logistic regression analysis for HHS mortalityVariableaOR95% CIP valueAge (years)1.0491.012 to 1.0870.015Sodium (mmol/L)1.0431.009 to 1.0810.016Urea (mmol/L)1.0511.006 to 1.0970.024HospitalHospital A (ref)10.589 to 46.8970.137Hospital B5.255aOR, adjusted OR; HHS, hyperosmolar hyperglycemic state.Stakeholder feedback: training, utility, and motivators 17 stakeholders from participating sites completed the DEKODE-HHS Participant Involvement Questionnaire. The majority had contributed to data collection (76.5%, n=13/17) and audit activities (58.8%, n=10/17), with motivations centered on academic development (76.5%, n=13/17) and quality improvement (64.7%, n=11/17). Participants rated the data collection tool positively for clarity (mean score 8.2/10) and usability (7.1/10). Challenges included incomplete electronic health records, variability in guideline interpretation, and frequent misclassification of HHS as DKA. Engagement was high, with 88.2% (n=15/17) indicating interest in contributing to future protocol development and dissemination.Eight participants from five NHS trusts participated in follow-up interviews. Thematic analysis (κ=0.68) revealed four core domains:Implementation logistics: familiarity with DEKODE initiatives facilitated rapid approval processes. As participant 3 explained: “So there was absolutely no challenge in getting approval. Everybody was happy. (DEKODE) is quite known, I must say throughout England, so no objection from anyone in setting up the project”. However, logistical barriers, including rotating staff, non-intuitive Electronic Health Records, and Virtual Private Network limitations for remote data entry, impeded data collection.Educational impact: the project enhanced clinicians’ understanding of HHS, with feedback facilitating structured teaching, improving confidence in biochemistry interpretation, and clarifying fluid and insulin management. Participants also became more involved in updating trust-specific guidelines. As participant 7 shared: “…to recognize which patients are just hyperglycemia or whether they are DKA and HSS are just differentiating between the kind of biochemistry on presentation and then also just appreciating. The different types of fluids, depending on how the patient responds to the initial management, because I think sort of an A&E, the initial few steps…”.Engagement and motivation: sustained engagement was driven by opportunities for academic gain and leadership opportunities, with suggestions to embed surveillance activities into foundation training and medical school placements. Participant 4 commented, “So I’d already been collecting data for the DKA part of it for quite a while. Most, which was mostly because I wanted to learn more about diabetes, … that I didn’t have as much knowledge in that as I would have liked really.”Evaluation and future directions: participants cited the Google form’s clarity and user-friendly design as a key strength of data collection, assuming baseline clinical knowledge. Regular monthly meetings helped maintain momentum and foster collaboration across centers. Communication with consultant leads was efficient and responsive. However, participants expressed a need for more regular feedback on project progress and findings at the local level. Nevertheless, the project helped identify institutional gaps in HHS management and areas requiring focused education. As one participant observed: “It really brought out where we were lacking in terms of management and where we had to focus more in educating the resident doctors.” (Participant 1)Evaluation using the RE-AIM framework The DEKODE-HHS model demonstrated success across all RE-AIM domains ( online supplemental file 3)18:Reach: involved 12 NHS sites with a nationally representative HHS cohort.Effectiveness: identified modifiable variations in care and predictors of adverse outcomes.Adoption: secured rapid uptake across DEKODE-affiliated trusts via structured training and protocol standardization.Implementation: achieved high fidelity through the use of a uniform audit tool, embedded training, and support structures.Maintenance: iterative refinements and site-specific feedback loops were integrated, with several sites initiating plans to embed HHS surveillance within routine audit and governance cycles.This structured, scalable approach demonstrated both feasibility and impact, offering a replicable model for improving care quality in other rare and complex inpatient diabetes emergencies.Discussion This study represents the first successful implementation of a standardized, multicenter surveillance system for HHS in the UK. It led to the most extensive UK dataset to date on the epidemiology, clinical presentation, management, and outcomes of HHS, and demonstrates how structured surveillance can reveal modifiable variation in care and inform improvements in practice. The model aligns with national audit standards from NICE and HQIP and establishes a foundation for embedding continuous quality improvement into routine clinical pathways for rare but high-risk diabetes emergencies. 19 20Our findings demonstrate that the DEKODE-HHS model facilitates both clinical insight and institutional benchmarking. By enabling hospitals to evaluate their local data against aggregate medians and similar peers, the model supports early identification of adverse outcomes, areas of low compliance, and good practice. These data can inform future iterations of national guidelines and contribute to a stronger evidence base for optimal HHS care.To our knowledge, this is the first UK study to report national multicenter data on HHS duration and length of hospitalization. The median length of stay in our cohort was 10.3 days, significantly longer than the 3.8 days reported in some US studies, suggesting differences in care pathways, discharge planning, or underlying health system structures.6 Delays following HHS resolution were frequently attributed to non-clinical factors, including care coordination and ongoing investigation needs, indicating the importance of system-level interventions to support timely discharge.The demographic profile of our cohort is consistent with previously reported population-level estimates.3 6 Notably, a significant minority, approximately one-third of HHS cases, occurred in individuals under 70 years of age, including 8.3% aged 30–49 years. This likely reflects the rising prevalence of type 2 diabetes across younger age groups and underscores the importance of recognizing HHS risk in a wider age spectrum.22 23 Consistent with earlier reports, intercurrent illness and infection were the leading precipitants, accounting for over 70% of cases.4 9 Furthermore, 10.8% of admissions were linked to poor treatment adherence, indicated by a median preadmission HbA1c of 81 mmol/mol, highlighting opportunities for primary and secondary prevention through education and proactive follow-up, particularly in long-term care facilities, since the majority of our cohort resided in nursing or care homes prior to admission.24 25While 84.4% of our cohort had pre-existing type 2 diabetes, 8.7% presented with new diagnoses. This aligns with findings by Rosager et al, who reported that nearly one-third of HHS cases lacked a prior diabetes diagnosis.3 26 This reinforces the importance of diagnosing and managing HHS based on clinical and biochemical criteria rather than relying solely on known diabetes status.Importantly, the study revealed considerable variation in HHS management, particularly in insulin dosing and capillary glucose monitoring. At Hospital A, higher insulin dosing and superior adherence to hourly glucose monitoring were observed alongside lower inpatient mortality. Notably, baseline glycemic control indicated by HbA1c and preadmission insulin use did not differ between hospitals, suggesting that differences in insulin management during HHS likely reflect institutional variations in clinical practice rather than prior diabetes care. These findings are notable given the absence of trial-based guidance on insulin and fluid regimens in HHS, with current protocols often extrapolated from DKA pathways, and reinforces the importance of adherence to a standardized protocol to reduce variations in HHS care.10 While causality cannot be established, the observed differences in mortality may warrant further evaluation in prospective or interventional studies.Inpatient mortality in this cohort was 16.1%, comparable to 10–17% observed in international cohorts.3 5 7 Multivariate analysis revealed patient-level factors rather than management variations predominantly determined patient outcomes. Higher serum sodium and urea at presentation emerged as significant independent predictors of mortality, reflecting greater severity of HHS. These findings corroborate prior evidence and suggest their potential utility in risk stratification models to guide enhanced monitoring or early therapeutic escalation.9 27 28 Advanced age also remained a significant predictor of mortality in our cohort, likely a reflection of decreased physiological reserve and increased vulnerability to metabolic stress. The lack of association between comorbidity burden (CCI) and mortality in our cohort differs from some previous studies, possibly due to the uniformly high comorbidity burden across the cohort or sample size limitations.8 29 30 However, this could also suggest that acute physiological derangement may be superior to chronic disease burden in determining patient outcomes in HHS care.While some intersite variation in adherence to guideline-recommended practices was observed, differences in outcomes were not statistically significant after adjustment, suggesting that institutional characteristics alone may not explain mortality differences. Instead, the DEKODE-HHS model highlights how structured feedback and local ownership can drive targeted quality improvement. For example, Hospital A adapted the ADA international consensus for HHS care into a simplified, color-coded tool, enhanced educational activities, and demonstrated higher monitoring compliance.Qualitative feedback further emphasized the educational value of the project, with many participants reporting increased confidence in recognizing and managing HHS. Thematic analysis revealed key implementation facilitators including alignment with existing quality improvement (QI) structures and engagement from clinical leadership, as well as persistent barriers such as workforce limitations, misclassification of HHS, and challenges with data entry and case identification.This study illustrates the value of a mixed-methods evaluation in implementation science. The RE-AIM framework was instrumental in assessing real-world impact, highlighting the model’s wide reach, feasibility, and adaptability across varied institutional contexts.18 Importantly, stakeholder engagement was high, and feedback has already informed refinements to both the surveillance tool and educational resources.Strengths and limitations This study presents the largest UK multicenter dataset on HHS, integrating clinical outcomes with implementation insights to offer a comprehensive evaluation of care delivery. Standardized diagnostic criteria and an embedded feedback model enhanced comparability, adaptability, and scalability. However, key limitations include incomplete documentation of critical variables such as hydration and neurological status, limiting prognostic analysis. Additionally, our analysis did not include the granular specifics of insulin management including loading doses and infusion parameters, imposing constraints on the evaluation of treatment effects given the pivotal role of insulin in HHS management. The study was not powered to detect adjusted mortality differences between sites, and voluntary participation may introduce selection bias. Nonetheless, the findings provide valuable data to inform future guidelines and highlight the feasibility of a surveillance-based quality improvement model for rare acute metabolic emergencies.Conclusion The DEKODE-HHS surveillance system demonstrates a feasible, scalable model for monitoring and improving care for patients with HHS. It supports early identification of unwarranted variation, informs clinical decision-making, and provides a platform for national benchmarking and future research. Age and serum sodium at presentation emerged as key predictors of mortality and should be further explored in risk-stratification efforts. Institutional adaptation, education, and continuous feedback are critical to enabling improvement. This model may serve as a blueprint for similar efforts in other complex inpatient conditions where national data are sparse and variation in care remains a barrier to safety and equity.