Introduction
Type 2 diabetes (T2D) has increased globally, especially in younger adults.1 Still, studies reporting prevalence of young-onset T2D (YOD), defined as diabetes diagnosis before age 40, are scarce.2 YOD is associated with higher complication rates and reduced life expectancy.3 4 Recent reductions in adverse outcomes in T2D have been less evident in YOD.5 6
Individuals with YOD have higher body mass index (BMI) and hemoglobin A1c (HbA1c) than in later-onset T2D and retinopathy is the diabetes complication with highest excess risk.7–9 In T2D, incidence of retinopathy and symptomatic eye disease has been reduced due to less smoking, better glycemic and blood pressure control, and earlier detection and treatment.10 However, it is less known to what extent individuals with YOD have benefited. In YOD, longer diabetes duration and poorer glycemic control seem to be the most important factors explaining higher prevalence of retinopathy, but it is unresolved whether YOD is an inherently more aggressive disease phenotype.9 11
In men, the absolute risk of acquiring T2D and YOD and of developing diabetes complications is higher compared with women.1 12 13 Men also develop T2D at lower BMI.14 Sex differences in macrovascular complications are widely studied, while population-based or primary care studies comparing microvascular complications by sex are scarce. Therefore, we aimed to identify the prevalence of diabetes onset before age 40 among adults with T2D in Norwegian general practice and investigate associations between YOD and retinopathy overall and stratified by sex.
Research design and methods
Design, data collection and participants
Data source was the ROSA 4 study, described in detail elsewhere.15 16 In short, ROSA 4 is a cross-sectional dataset on all individuals with a diabetes mellitus diagnosis identified in the electronic medical records (EMRs) of 282 Norwegian general practitioner (GPs). Data were collected in 2015 from three of four health regions. We invited smaller and larger practices from both urban and rural areas of low and high socioeconomic status and mixed ethnic backgrounds. For the included GPs, age, sex distribution and number of patients on their lists were comparable with the Norwegian average.15 Data include patients’ sex, year of birth and diabetes diagnosis, diabetes type, relevant clinical examinations and laboratory results, diabetic complications, prescribed diabetes medications, referrals for diabetes to and summaries from secondary healthcare.
Data on adults (≥18 years), with a diabetes diagnosis recorded in each GP’s EMR between 2012 and 2014, were extracted using a customized software program (Medrave). Research nurses manually validated all diagnoses and year of diagnosis according to the study protocol. The dataset comprised 11 428 diabetes cases; 10 248 with T2D, 1180 with T1D and 49 with other or unknown type.15 For individuals with age of onset <50 years, three clinicians used medication history and clinical data to quality check diabetes type. Some cases (<10) were reclassified, resulting in 10 241 cases with T2D. Education and country background variables from Statistics Norway were linked to ROSA 4.17
Variables
The exposure variable, age at diabetes diagnosis, was stratified into three groups: diagnosis (1) before age 40 years (YOD), (2) age 40–49 years and (3) 50 years or older.
The primary outcome was retinopathy diagnosed by an ophthalmologist, registered in the GP’s EMR as a diagnosis code, free text or in notes from secondary care. Treatment of retinopathy with injections, laser or other methods was recorded, including year of first treatment. Impaired vision was recorded as visual acuity lower than 0.33.
The last recorded clinical measurements before January 1, 2015 were included. For hemoglobin A1c (HbA1c %, converted to mmol/mol), 95% of individuals had repeated measurements from 2012 to 2014, with an average of seven recordings per person. Of these, 1836 (19%) were diagnosed with T2D in this time period and thus had a recorded HbA1c at diagnosis. Reduced foot sensibility/neuropathy, coronary heart disease and stroke were recorded from GPs’ EMR including secondary care notes. Chronic kidney disease was defined as estimated glomerular filtration rate <60 mL/min/1.73 m2. Glucose-lowering treatment was categorized as either: (1) no antidiabetic medication or “diet only,” (2) “insulin,” alone or in combination with other agents or (3) “other glucose-lowering agents” only. Current smokers were defined as smoking recorded during the last 5 years. Highest education was categorized as compulsory education (primary and lower secondary), upper secondary or higher education. Country of origin was divided into three groups, Europe and North America, South Asia and other.
Statistical analysis
Descriptive statistics are presented with means, proportions and 95% CIs as appropriate. Individuals with and without a recorded year of diagnosis and retinopathy status were compared descriptively. Statistical analyses were undertaken in StataSE16 (StataCorp).
Associations between age at diabetes diagnosis, either age or diabetes duration and diabetes complications were estimated using logistic regression. Further, average adjusted predictions (AAPs) of complications in each age-at-onset group were calculated using the margins command. AAPs give the expected average complication prevalence over the observed distribution of current age or diabetes duration if all individuals were diagnosed before age 40, at age 40–49 or at age 50 years.
A directed acyclic graph was drawn in Dagitty V.3.0 prior to analysis to identify possible clinically relevant relationships between exposure (YOD), primary outcome (retinopathy) and potential confounding and mediating variables (online supplemental figure 1).
HbA1c progression was analyzed using longitudinal data. Average HbA1c was estimated by age at diagnosis and diabetes duration, using the first observed measurement in each year. Average HbA1c progression in the first year was analyzed by estimating local linear regressions by sex and age at diagnosis, using the lpoly command. Associations between age at diagnosis, sex and HbA1c up to 10 years diabetes after diagnosis were analyzed by linear regression.
To reduce potential bias from missing values for retinopathy, multilevel multiple imputation was undertaken in R V.3.5.2 under the assumption of missing at random. The R package miceadds was used to generate 25 imputed datasets, exported to Stata for analysis.
Multivariate logistic regression analyses were undertaken in imputed data and in complete cases, to assess associations between YOD and retinopathy, stratified by sex. In Model 1, all identified potential confounders (age, education, country of origin, BMI) were adjusted for. In Model 2, potential mediators (HbA1c, diabetes duration, systolic blood pressure, low-density lipoprotein (LDL) cholesterol) were added. All covariates were tested for interactions by cross-product terms, but no significant interactions were found. Potential mediators were included in the final models if the absolute difference in OR for retinopathy was more than 0.1.
Regression analyses were repeated for complete cases and for missing values for retinopathy assigned to “no retinopathy.” Results are presented as adjusted OR (aOR) with 95% CIs.
Predicted retinopathy prevalences based on the imputed logistic regression analyses were plotted using mimargins and marginsplot.