Heterogeneity in glycaemic control in children and adolescents with type 1 diabetes: A latent class trajectory analysis of Danish nationwide data.

AIMS: Suboptimal glycaemic control in children and adolescents with type 1 diabetes is prevalent and associated with increased risk of diabetes-related complications and mortality later in life. First, we aimed to identify distinct glycated haemoglobin (HbA1c) trajectories in children and adolescents (2-19 years) with type 1 diabetes. Second, we examined their associations with clinical and socio-demographic factors. METHODS: Data were obtained from the Danish Registry of Childhood and Adolescent Diabetes (DanDiabKids) comprising all Danish children and adolescents diagnosed with type 1 diabetes from 1996 to 2019. Subgroups of distinct mean trajectories of HbA1c were identified using data-driven latent class trajectory modelling. RESULTS: A total of 5889 children (47% female) had HbA1c measured a median of 6 times (interquartile range 3-8) and contributing to 36,504 measurements. We identified four mean HbA1c trajectories, referred to as 'Stable but elevated HbA1c' (83%), 'Increasing HbA1c' (5%), 'Late HbA1c peak' (7%), and 'Early HbA1c peak' (5%). Compared to the 'Stable but elevated HbA1c' group, the three other groups presented rapidly deteriorating glycaemic control during late childhood or adolescence, had higher HbA1c at study entry, and included fewer pump users, higher frequency of inadequate blood glucose monitoring, more severe hypoglycaemic events, lower proportions with Danish origin, and worse educational status of parents. The groups also represented significant differences by healthcare region. CONCLUSIONS: Children and adolescents with type 1 diabetes experience heterogenous trajectories with different timings and magnitudes of the deterioration of HbA1c levels, although the majority follow on average a stable, yet elevated HbA1c trajectory. The causes and long-term health implications of these heterogenous trajectories need to be addressed.

Large Language Models for Epidemiological Research via Automated Machine Learning: Case Study Using Data From the British National Child Development Study.

Background: Large language models have had a huge impact on natural language processing (NLP) in recent years. However, their application in epidemiological research is still limited to the analysis of electronic health records and social media data. objectives: To demonstrate the potential of NLP beyond these domains, we aimed to develop prediction models based on texts collected from an epidemiological cohort and compare their performance to classical regression methods. Methods: We used data from the British National Child Development Study, where 10,567 children aged 11 years wrote essays about how they imagined themselves as 25-year-olds. Overall, 15% of the data set was set aside as a test set for performance evaluation. Pretrained language models were fine-tuned using AutoTrain (Hugging Face) to predict current reading comprehension score (range: 0-35) and future BMI and physical activity (active vs inactive) at the age of 33 years. We then compared their predictive performance (accuracy or discrimination) with linear and logistic regression models, including demographic and lifestyle factors of the parents and children from birth to the age of 11 years as predictors. Results: NLP clearly outperformed linear regression when predicting reading comprehension scores (root mean square error: 3.89, 95% CI 3.74-4.05 for NLP vs 4.14, 95% CI 3.98-4.30 and 5.41, 95% CI 5.23-5.58 for regression models with and without general ability score as a predictor, respectively). Predictive performance for physical activity was similarly poor for the 2 methods (area under the receiver operating characteristic curve: 0.55, 95% CI 0.52-0.60 for both) but was slightly better than random assignment, whereas linear regression clearly outperformed the NLP approach when predicting BMI (root mean square error: 4.38, 95% CI 4.02-4.74 for NLP vs 3.85, 95% CI 3.54-4.16 for regression). The NLP approach did not perform better than simply assigning the mean BMI from the training set as a predictor. Conclusions: Our study demonstrated the potential of using large language models on text collected from epidemiological studies. The performance of the approach appeared to depend on how directly the topic of the text was related to the outcome. Open-ended questions specifically designed to capture certain health concepts and lived experiences in combination with NLP methods should receive more attention in future epidemiological studies.

ChatGPT- versus human-generated answers to frequently asked questions about diabetes: A Turing test-inspired survey among employees of a Danish diabetes center.

Large language models have received enormous attention recently with some studies demonstrating their potential clinical value, despite not being trained specifically for this domain. We aimed to investigate whether ChatGPT, a language model optimized for dialogue, can answer frequently asked questions about diabetes. We conducted a closed e-survey among employees of a large Danish diabetes center. The study design was inspired by the Turing test and non-inferiority trials. Our survey included ten questions with two answers each. One of these was written by a human expert, while the other was generated by ChatGPT. Participants had the task to identify the ChatGPT-generated answer. Data was analyzed at the question-level using logistic regression with robust variance estimation with clustering at participant level. In secondary analyses, we investigated the effect of participant characteristics on the outcome. A 55% non-inferiority margin was pre-defined based on precision simulations and had been published as part of the study protocol before data collection began. Among 311 invited individuals, 183 participated in the survey (59% response rate). 64% had heard of ChatGPT before, and 19% had tried it. Overall, participants could identify ChatGPT-generated answers 59.5% (95% CI: 57.0, 62.0) of the time, which was outside of the non-inferiority zone. Among participant characteristics, previous ChatGPT use had the strongest association with the outcome (odds ratio: 1.52 (1.16, 2.00), p = 0.003). Previous users answered 67.4% (61.7, 72.7) of the questions correctly, versus non-users' 57.6% (54.9, 60.3). Participants could distinguish between ChatGPT-generated and human-written answers somewhat better than flipping a fair coin, which was against our initial hypothesis. Rigorously planned studies are needed to elucidate the risks and benefits of integrating such technologies in routine clinical practice.

Autonomic dysfunction is associated with the development of arterial stiffness: the Whitehall II cohort.

This study aims to examine the association between baseline level and change of autonomic nervous function with subsequent development of arterial stiffness. Autonomic nervous function was assessed in 4901 participants of the Whitehall II occupational cohort by heart rate variability (HRV) indices and resting heart rate (rHR) three times between 1997 and 2009, while arterial stiffness was assessed by carotid-femoral pulse wave velocity (PWV) measured twice between 2007 and 2013. First, individual HRV/rHR levels and annual changes were estimated. Then, we modelled the development of PWV by HRV/rHR using linear mixed effect models. First, we adjusted for sex and ethnicity (model 1), and then for socioeconomic and lifestyle factors, various clinical measurements, and medications (model 2). A decrease in HRV and unchanged rHR was associated with subsequent higher levels of PWV, but the effect of a change in HRV was less pronounced at higher ages. A typical individual aged 65 years with a SDNN level of 30 ms and a 2% annual decrease in SDNN had 1.32 (0.95; 1.69) higher PWV compared to one with the same age and SDNN level but with a 1% annual decrease in SDNN. Further adjustment had no major effect on the results. People who experience a steeper decline in autonomic nervous function have higher levels of arterial stiffness. The association was stronger in younger people.

Body Mass Index Trajectories From Childhood to Adulthood and Age at Onset of Overweight and Obesity: The Influence of Parents' Weight Status.

We investigated the influence of parents' weight status on their children's growth trajectories and its association with age at onset of overweight and obesity. We used 16,396 height and weight records from 3,284 youths from the Panel Study of Income Dynamics, followed across childhood into adulthood (United States, 1997-2017). Across age groups, we modeled body mass index trajectories (ages 5-32 years) according to parents' weight status, using mixed-effect models to estimate age at onset of overweight and obesity and proportion with obesity from childhood to adulthood. There were large differences in growth patterns according to parents' weight status: Children of parents with obesity had, on average, overweight at age 6 (95% confidence interval (CI): 5, 7) and steep growth trajectories until age 12; children of normal-weight parents had slower increases in body mass index, reaching overweight on average at age 25 (95% CI: 24, 27). By age 30, 30% (95% CI: 28, 31) of youths had obesity. Differences in early-life growth persisted into adulthood: 48% (95% CI: 45, 52) of adult children of parents with obesity had obesity versus 16% (95% CI: 14, 19) of those of normal-weight parents. Trajectories to unhealthy weight were heavily influenced by parents' weight status, especially before age 12, children of parents with obesity having overweight 19 earlier in life than children of normal-weight parents.

Trajectory and predictors of HbA1c in children and adolescents with type 1 diabetes-A Danish nationwide cohort study.

OBJECTIVES: Poor glycemic control in type 1 diabetes increases the risk of chronic complications and it is essential to identify life periods and predictors associated with deteriorating HbA1c . The aim was to describe specific HbA1c trajectories in Danish children and adolescents with type 1 diabetes and study associations with clinical and sociodemographic factors. RESEARCH DESIGN AND METHODS: 5889 children with type 1 diabetes were included from the nationwide Danish Registry of Childhood and Adolescent Diabetes with annual visits during 1996-2019. Trajectories of HbA1c were modeled with linear mixed-effects models (using age as time scale, included as cubic spline) and with an individual-specific random intercept and slope. The following cofactors were included stepwise into the model: sex, age at diagnosis, calendar year, parental education, immigrant status, health care region, blood glucose monitoring (BGM) frequency, treatment modalities: continuous subcutaneous insulin infusion (pump) versus multiple daily insulin injection therapy (pen) and continuous glucose monitoring. RESULTS: HbA1c overall increased during age while there was a significant decreasing secular trend. Older age at diagnosis was associated with a steeper trajectory, and non-Danish origin and shorter parental education were each associated with higher levels of HbA1c across age. A lower BGM frequency was associated with a markedly poorer HbA1c trajectory, while no significant differences were shown for different treatment modalities. CONCLUSIONS: Glycemic outcome worsened with age during childhood and adolescence, which is of clinical concern. Important predictors for a poorer glycemic trajectory were later age at diabetes diagnosis, shorter parental education, non-Danish origin and, in particular low BGM frequency.

Changes in type 2 diabetes incidence and mortality associated with introduction of HbA1c as diagnostic option: A Danish 24-year population-based study.

BACKGROUND: In 2011, the World Health Organization began recommending glycated haemoglobin (HbA1c) as a measure for diagnosing type 2 diabetes (T2D). This initiative may have changed basic T2D epidemiology. Consequently, we examined time changes in T2D incidence and mortality during 1995-2018. METHODS: In this population-based cohort study, we included 415,553 individuals with incident T2D. We calculated annual age-standardized incidence rates of T2D. We examined HbA1c testing and used Poisson-regression to investigate all-cause mortality among the T2D patients and a matched comparison cohort from the general population over successive 3-year periods. FINDINGS: From 1995 to the 2012 introduction of HbA1c testing as a diagnostic option in Denmark, the annual standardized incidence rate (SIR) of T2D doubled, from 193 to 396 per 100,000 persons (4.1% increase annually). From 2012 onwards, the T2D incidence declined by 36%, reaching 253 per 100,000 persons in 2018 (5.7% decrease annually). This was driven by fewer patients starting treatment with an HbA1c measurement of <6·5% or without prior HbA1c testing. Mortality per 1,000 person-years following a T2D diagnosis decreased by 44% between 1995-1997 and 2010-2012, from 69 deaths to 38 deaths (adjusted mortality rate ratio: 0·55 (95% CI: 0·54-0·56)). After the low level during 2010-2012, mortality increased again by 27% to 48 per 1,000 person-years (95% CI: 46-50) by 2016-2018. INTERPRETATION: Our findings suggest that introducing HbA1c as a diagnostic option may have changed basic T2D epidemiology by leaving patients undiagnosed, that previously would have been diagnosed and treated. FUNDING: Aarhus University funded the study and had no further involvement.

Transfer learning for non-image data in clinical research: A scoping review.

BACKGROUND: Transfer learning is a form of machine learning where a pre-trained model trained on a specific task is reused as a starting point and tailored to another task in a different dataset. While transfer learning has garnered considerable attention in medical image analysis, its use for clinical non-image data is not well studied. Therefore, the objective of this scoping review was to explore the use of transfer learning for non-image data in the clinical literature. METHODS AND FINDINGS: We systematically searched medical databases (PubMed, EMBASE, CINAHL) for peer-reviewed clinical studies that used transfer learning on human non-image data. We included 83 studies in the review. More than half of the studies (63%) were published within 12 months of the search. Transfer learning was most often applied to time series data (61%), followed by tabular data (18%), audio (12%) and text (8%). Thirty-three (40%) studies applied an image-based model to non-image data after transforming data into images (e.g. spectrograms). Twenty-nine (35%) studies did not have any authors with a health-related affiliation. Many studies used publicly available datasets (66%) and models (49%), but fewer shared their code (27%). CONCLUSIONS: In this scoping review, we have described current trends in the use of transfer learning for non-image data in the clinical literature. We found that the use of transfer learning has grown rapidly within the last few years. We have identified studies and demonstrated the potential of transfer learning in clinical research in a wide range of medical specialties. More interdisciplinary collaborations and the wider adaption of reproducible research principles are needed to increase the impact of transfer learning in clinical research.

Discordance Between Glucose Levels Measured in Interstitial Fluid vs in Venous Plasma After Oral Glucose Administration: A Post-Hoc Analysis From the Randomised Controlled PRE-D Trial.

Aims: The oral glucose tolerance test (OGTT) is together with haemoglobin A1c (HbA1c) gold standard for diagnosing prediabetes and diabetes. The objective of this study was to assess the concordance between glucose values obtained from venous plasma versus interstitial fluid after oral glucose administration in 120 individuals with prediabetes and overweight/obesity. Methods: 120 adults with prediabetes defined by HbA1c 39-47 mmol/mol and overweight or obesity who participated in the randomised controlled PRE-D trial were included in the study. Venous plasma glucose concentrations were measured at 0, 30, 60 and 120 minutes during a 75 g oral glucose tolerance test (OGTT) performed on three different occasions within a 26 weeks period. During the OGTT, the participants wore a CGM device (IPro2, Medtronic), which assessed glucose concentrations every five minutes. Results: A total of 306 OGTTs with simultaneous CGM measurements were obtained. Except in fasting, the CGM glucose values were below the OGTT values throughout the OGTT period with mean (SD) differences of 0.2 (0.7) mmol/L at time 0 min, -1.1 (1.3) at 30 min, -1.4 (1.8) at 60 min, and -0.5 (1.1) at 120 min). For measurements at 0 and 120 min, there was a proportional bias with an increasing mean difference between CGM and OGTT values with increasing mean of the two measurements. Conclusions: Due to poor agreement between the OGTT and CGM with wide 95% limits of agreement and proportional bias at 0 and 120 min, the potential for assessing glucose tolerance in prediabetes using CGM is questionable.

Towards precision medicine in diabetes? A critical review of glucotypes.

In response to a study previously published in PLOS Biology, this Formal Comment thoroughly examines the concept of 'glucotypes' with regard to its generalisability, interpretability and relationship to more traditional measures used to describe data from continuous glucose monitoring.

Spousal concordance in pathophysiological markers and risk factors for type 2 diabetes: a cross-sectional analysis of The Maastricht Study.

INTRODUCTION: We compared the degree of spousal concordance in a set of detailed pathophysiological markers and risk factors for type 2 diabetes to understand where in the causal cascade spousal similarities are most relevant. RESEARCH DESIGN AND METHODS: This is a cross-sectional analysis of couples who participated in The Maastricht Study (n=172). We used quantile regression models to assess spousal concordance in risk factors for type 2 diabetes, including four adiposity measures, two dimensions of physical activity, sedentary time and two diet indicators. We additionally assessed beta cell function and insulin sensitivity and glucose metabolism status with fasting and 2-hour plasma glucose and hemoglobin A1c. RESULTS: The strongest spousal concordance (beta estimates) was observed for the Dutch Healthy Diet Index (DHDI) in men. A one-unit increase in wives' DHDI was associated with a 0.53 (95% CI 0.22 to 0.67) unit difference in men's DHDI. In women, the strongest concordance was for the time spent in high-intensity physical activity (HPA); thus, a one-unit increase in husbands' time spent in HPA was associated with a 0.36 (95% CI 0.17 to 0.64) unit difference in women's time spent in HPA. The weakest spousal concordance was observed in beta cell function indices. CONCLUSIONS: Spousal concordance was strongest in behavioral risk factors. Concordance weakened when moving downstream in the causal cascade leading to type 2 diabetes. Public health prevention strategies to mitigate diabetes risk may benefit from targeting spousal similarities in health-related behaviors and diabetes risk factors to design innovative and potentially more effective couple-based interventions.

A Validated Prediction Model for End-Stage Kidney Disease in Type 1 Diabetes.

OBJECTIVE: End-stage kidney disease (ESKD) is a life-threatening complication of diabetes that can be prevented or delayed by intervention. Hence, early detection of people at increased risk is essential. RESEARCH DESIGN AND METHODS: From a population-based cohort of 5,460 clinically diagnosed Danish adults with type 1 diabetes followed from 2001 to 2016, we developed a prediction model for ESKD accounting for the competing risk of death. Poisson regression analysis was used to estimate the model on the basis of information routinely collected from clinical examinations. The effect of including an extended set of predictors (lipids, alcohol intake, etc.) was further evaluated, and potential interactions identified in a survival tree analysis were tested. The final model was externally validated in 9,175 adults from Denmark and Scotland. RESULTS: During a median follow-up of 10.4 years (interquartile limits 5.1; 14.7), 303 (5.5%) of the participants (mean [SD] age 42.3 [16.5] years) developed ESKD, and 764 (14.0%) died without having developed ESKD. The final ESKD prediction model included age, male sex, diabetes duration, estimated glomerular filtration rate, micro- and macroalbuminuria, systolic blood pressure, hemoglobin A1c, smoking, and previous cardiovascular disease. Discrimination was excellent for 5-year risk of an ESKD event, with a C-statistic of 0.888 (95% CI 0.849; 0.927) in the derivation cohort and confirmed at 0.865 (0.811; 0.919) and 0.961 (0.940; 0.981) in the external validation cohorts from Denmark and Scotland, respectively. CONCLUSIONS: We have derived and validated a novel, high-performing ESKD prediction model for risk stratification in the adult type 1 diabetes population. This model may improve clinical decision making and potentially guide early intervention.

ß-Lactoglobulin Elevates Insulin and Glucagon Concentrations Compared with Whey Protein-A Randomized Double-Blinded Crossover Trial in Patients with Type Two Diabetes Mellitus.

Whey protein is an insulinotropic fraction of dairy that reduces postprandial glucose levels in patients with type 2 diabetes mellitus (T2DM). We have recently shown that ß-lactoglobulin (BLG), the largest protein fraction of whey, elevates insulin concentrations compared with iso-nitrogenous whey protein isolate (WPI) in healthy individuals. We therefore hypothesized that BLG pre-meals would lower glucose levels compared with WPI in patients with T2DM. We investigated 16 participants with T2DM using a randomized double-blinded cross-over design with two pre-meal interventions, (i) 25 g BLG and (ii) 25 g WPI prior to an oral glucose tolerance test (OGTT), followed by four days of continuous glucose monitoring (CGM) at home. BLG increased concentrations of insulin with 10%, glucagon with 20%, and glucose with 10% compared with WPI after the OGTT (all p < 0.05). Both BLG and WPI reduced the interstitial fluid (ISF) glucose concentrations (using CGM) with 2 mM and lowered glycemic variability with 10-15%, compared with tap-water (p < 0.05), and WPI lowered the ISF glucose with 0.5 mM compared with BLG from 120 min and onwards (p < 0.05). In conclusion, BLG pre-meals resulted in higher insulin, glucagon, and glucose concentrations compared with WPI in participants with T2DM. Pre-meal servings of WPI remains the most potent protein in terms of lowering postprandial glucose excursions.

Effect of familial diabetes status and age at diagnosis on type 2 diabetes risk: a nation-wide register-based study from Denmark.

AIMS/HYPOTHESIS: We assessed whether the risk of developing type 2 diabetes and the age of onset varied with the age at diabetes diagnosis of affected family members. METHODS: We performed a national register-based open cohort study of individuals living in Denmark between 1995 and 2012. The population under study consisted of all individuals aged 30 years or older without diagnosed diabetes at the start date of the cohort (1 January 1995) and who had information about their parents' identity. Individuals who turned 30 years of age during the observation period and had available parental identity information were also added to the cohort from that date (open cohort design). These criteria restricted the study population mostly to people born between 1960 and 1982. Multivariable Poisson regression models adjusted for current age and highest educational attainment were used to estimate incidence rate ratios (IRRs) of type 2 diabetes. RESULTS: We followed 2,000,552 individuals for a median of 14 years (24,034,059 person-years) and observed 76,633 new cases of type 2 diabetes. Compared with individuals of the same age and sex who did not have a parent or full sibling with diabetes, the highest risk of developing type 2 diabetes was observed in individuals with family members diagnosed at an early age. The IRR was progressively lower with a higher age at diabetes diagnosis in family members: 3.9 vs 1.4 for those with a parental age at diagnosis of 50 or 80 years, respectively; and 3.3 vs 2.0 for those with a full sibling's age at diagnosis of 30 or 60 years, respectively. CONCLUSIONS/INTERPRETATION: People with a family member diagnosed with diabetes at an earlier age are more likely to develop diabetes and also to develop it at an earlier age than those with a family member diagnosed in later life. This finding highlights the importance of expanding our understanding of the interplay between genetic diabetes determinants and the social, behavioural and environmental diabetes determinants that track in families across generations. Accurate registration of age at diagnosis should form an integral part of recording a diabetes family history, as it provides easily obtainable and highly relevant detail that may improve identification of individuals at increased risk of younger onset of type 2 diabetes. In particular, these individuals may benefit from closer risk factor assessment and follow-up, as well as prevention strategies that may involve the family.

Prospective Association Among Diabetes Diagnosis, HbA1c, Glycemia, and Frailty Trajectories in an Elderly Population.

OBJECTIVE: Frailty is a dynamic state of vulnerability in the elderly. We examined whether individuals with overt diabetes or higher levels of HbA1c or fasting plasma glucose (FG) experience different frailty trajectories with aging. RESEARCH DESIGN AND METHODS: Diabetes, HbA1c, and FG were assessed at baseline, and frailty status was evaluated with a 36-item frailty index every 2 years during a 10-year follow-up among participants from the English Longitudinal Study of Ageing (ELSA). Mixed-effects models with age as time scale were used to assess whether age trajectories of frailty differed as a function of diabetes, HbA1c, and FG. RESULTS: Among 5,377 participants (median age [interquartile range] 70 [65, 77] years, 45% men), 35% were frail at baseline. In a model adjusted for sex, participants with baseline diabetes had an increased frailty index over aging compared with those without diabetes. Similar findings were observed with higher levels of HbA1c, while FG was not associated with frailty. In a model additionally adjusted for income, social class, smoking, alcohol, and hemoglobin, only diabetes was associated with an increased frailty index. Among nonfrail participants at baseline, both diabetes and HbA1c level were associated with a higher increased frailty index over time. CONCLUSIONS: People with diabetes or higher HbA1c levels at baseline had a higher frailty level throughout later life. Nonfrail participants with diabetes or higher HbA1c also experienced more rapid deterioration of frailty level with aging. This observation could reflect a role of diabetes complications in frailty trajectories or earlier shared determinants that contribute to diabetes and frailty risk in later life.

Progressive Decline in Estimated Glomerular Filtration Rate in Patients With Diabetes After Moderate Loss in Kidney Function-Even Without Albuminuria.

OBJECTIVE: Persons with diabetes but no chronic kidney disease (CKD) and without albuminuria have the same age-related decline in kidney function as the background population. Whether this also applies following moderate loss in kidney function is unknown. We quantified the impact of albuminuria status on the development of estimated glomerular filtration rate (eGFR) trajectories following CKD stage 3 (CKD3) and assessed potential heterogeneous development patterns among the subgroup with normoalbuminuria. RESEARCH DESIGN AND METHODS: We used repeated clinical measures during up to 16 years of follow-up in 935 persons with type 1 diabetes and 1,984 with type 2 diabetes. Trajectories of eGFR by diabetes type and albuminuria status following CKD3 were estimated with spline mixed-effects models with adjustment for relevant confounders. Latent class trajectory modeling was used to find distinct patterns of eGFR development in the subgroups with normoalbuminuria. RESULTS: Mean annual declines in eGFR for normo-, micro- and macroalbuminuria the first 10 years following CKD3 were 1.9, 2.3, and 3.3 mL/min/1.73 m2 in type 1 diabetes and 1.9, 2.1, and 3.0 in type 2 diabetes, respectively. For normoalbuminuria, two distinct eGFR patterns were found, one with accelerated declining eGFR levels. This specific progression pattern was associated with less use of lipid-lowering treatment, renin-angiotensin system blockers, and other antihypertensive treatment. CONCLUSIONS: Our results support a diabetes-dependent decline in kidney function without albuminuria following CKD3, with a subgroup showing a progressive decline. Furthermore, this group seems to be undertreated in terms of cardioprotective and renal protective treatment and suggests that increased attention should be drawn to normoalbuminuric diabetic kidney disease.