Document resource
WHAT IS ALREADY KNOWN ON THIS TOPIC In the UK, mortality rates are substantially higher in adults with learning disabilities, and they are also at higher risk of developing type 2 diabetes mellitus (T2DM). However, there has been no large-scale investigation into the impacts of learning disabilities on T2DM management and outcomes.WHAT THIS STUDY ADDS Though individuals with learning disabilities had better glycemic control, they also had faster progression to severe diabetes (proxied by faster treatment escalation) and higher risk of mortality, despite having similar risks of vascular complications.HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY Further work into the causes of these effects could be utilized in primary care to help reduce health disparities for people with T2DM and learning disabilities.Background There are approximately 1.5 million people with learning disabilities (including intellectual disabilities) in the UK, with around 950 000 of these being adults. 1 2 Adults with learning disabilities are a vulnerable population in the UK at substantially higher risk of a range of chronic conditions, including type 2 diabetes (T2DM), related to a variety of interconnecting genetic, lifestyle, and socioeconomic factors.3 4 Mortality rates are also substantially higher in people with learning disabilities, with life expectancy estimated to be around 20 years lower than people without learning disabilities.5Management of T2DM in those with learning disabilities presents a sizeable problem for both patients and healthcare professionals. Self-management of T2DM requires a substantial amount of monitoring and management for patients, such as diet planning, blood glucose recording, exercising, and taking medication.6 People with learning disabilities may have particular challenges coping with some or all of these aspects of care, potentially leading to poorer glycemic control. Cognitive impairment may also lead to difficulties communicating with carers and healthcare professionals, perhaps compromising management of their condition relative to the general population.7 8 This study sought to better understand how learning disabilities may affect T2DM care and outcomes.Previous literature indicates that people with learning disabilities have a greater risk of vascular complications and mortality than people without learning disabilities.9 However, there has been no large-scale investigation into the impacts of learning disabilities on T2DM management and outcomes. The aim of this study was to determine whether people with T2DM who have a learning disability differ from those without a learning disability with respect to glycemic control, progression to microvascular and macrovascular complications, initiation of insulin therapy, and risk of all-cause and diabetes-related mortality. We hypothesized that people with learning disabilities would have poorer glycemic control after diagnosis and higher risks of vascular complications and mortality.Methods Study design We conducted an observational cohort study using data from the UK Clinical Practice Research Datalink (CPRD) GOLD linked to the Index of Multiple Deprivation (IMD) and the Office for National Statistics (ONS) mortality data. CPRD GOLD collects anonymized patient data from electronic health records from 886 UK general practices using Vision software, covering approximately 16 million currently registered patients. CPRD GOLD contains coded data that capture demographic characteristics, diagnoses, symptoms, drug histories, laboratory tests, and referrals. 10 11 ONS data were used to obtain data on mortality in a subset of CPRD participants from English general practices that have consented to CPRD linkage.Pseudonymized data for adults aged 18 years and over newly diagnosed with T2DM between January 2004 and January 2021 were extracted from the January 2021 version of CPRD GOLD. Individuals were excluded from the study if they had less than 6 months follow-up after T2DM diagnosis to improve the data quality. To ensure that the population included only people with incident T2DM, individuals with diagnoses within 6 months of joining the CPRD were excluded, as these could have represented recording of historic diagnoses captured during initial consultation or registration with a new general practice.12 Participants were also excluded if they were prescribed insulin either in the 12 months prior or 2 years after their T2DM diagnosis, to ensure the exclusion of individuals who may have type 1 diabetes mellitus. This is because it would be unlikely for a newly diagnosed patient with type 1 diabetes mellitus to go 2 years without insulin, as it is the first-line therapy.13The study follow-up period began at the latest of (1) the date the patient first registered with a general practice contributing to CPRD GOLD, (2) the general practitioner (GP) practice up-to-standard date, or (3) T2DM diagnosis date. This follow-up period ended at the earliest of (1) the date the patient transferred out of the practice, (2) the date of last data collection by the practice, or (3) date of death. This study was informed using a simplified conceptual model showing potential associations to be examined (figure 1). This shows our hypothesized confounders that we adjusted for and potential effect modifiers that may lie on the causal pathway between learning disability and T2DM-related outcomes.Figure 1Conceptual model diagram identifying potential confounders and effect modifiers of the interaction between learning disability and type 2 diabetes mellitus (T2DM) outcomes. BMI, body mass index; GP, general practitioner; HbA1c, glycated hemoglobin.Exposure variable The primary exposure was the presence of a learning disability diagnosis prior to T2DM diagnosis. Though there are not generally standardized definitions for diagnoses for CPRD coding, learning disability can be defined by the UK Department of Health and Social Care definition as:… significantly reduced ability to understand new or complex information, to learn new skills (impaired intelligence), with a reduced ability to cope independently (impaired social functioning); which started before adulthood, with a lasting effect on development.14Outcome variables Glycemic control Glycated hemoglobin (HbA1c), which measures average blood glucose levels for the past 3 months, was analyzed at two time points: 2 years and 5 years post-T2DM diagnosis. These measurements were taken as close to 2 and 5 years as possible ±6 months. Two years was chosen as the first time point as it is long enough from diagnosis for the effects of clinical care of T2DM diagnosis to be observed, while 5 years was chosen to assess the impacts of learning disabilities on the health of T2DM patients in the longer term. HbA1c was analyzed both as a continuous and as a binary variable of good versus poor glycemic control, with poor control considered to be above 7%/53 mmol/L.Initiation of insulin Newly initiated insulin therapy more than 2 years after T2DM diagnosis was used as a proxy indicator for severe T2DM. This is because insulin is only indicated in T2DM once there has been failure to achieve target HbA1c using lifestyle changes and oral antidiabetic medications. 13Diabetes complications T2DM-related complications were analyzed in two separate categories: macrovascular and microvascular. Macrovascular complications include stroke, coronary heart disease, heart failure, peripheral vascular disease, or amputation more than 6 months following T2DM diagnosis. Microvascular complications include the new diagnosis of either diabetic nephropathy, retinopathy, or neuropathy more than 6 months following T2DM diagnosis. Existing complications prior to T2DM diagnosis and complication diagnoses within 6 months of diagnosis were excluded to limit the probability that complications arose independent of T2DM.Mortality Deaths occurring any time after T2DM diagnosis were considered for the mortality outcome. Mortality data were obtained from both primary care records and from linked ONS data where the death date was derived from the ONS death record. Risk of all-cause mortality was assessed in the entire study cohort using the earliest death date from either the primary care record or the ONS death records, where available. Risk of diabetes-related mortality was assessed in the ONS linked cohort and was defined as underlying cause of death due to microvascular and macrovascular complications. ONS linkage was available for 43% of the study cohort.Covariates Covariates were all defined at the date of T2DM diagnosis and included: age (categorised as <50, 50–59, 60–69, and 70+ years); sex; self-reported ethnicity (categorized as White, South Asian, Black, Other, and Mixed); deprivation (quintiles of IMD); number of consultations in the year prior to the index date; systolic blood pressure (categorised as <120 mmHg, 120–139 mmHg, 140–159 mmHg, and >160 mmHg); HbA1c (categorised as ≤7% good glycemic control and >7% poor glycemic control), body mass index (BMI) (categorised as <25 kg/m 2 normal weight, 25–29.99 kg/m2 overweight, 30–39.99 kg/m2 obese, >40 kg/m2 severely obese); smoking status at diagnosis (categorized as non-smoker, current smoker, or ex-smoker); comorbidities (hypertension, microvascular, macrovascular), and use of lipid- and blood pressure-lowering medication.Code lists for all study variables can be found at https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7228040/.15Statistical analysis Descriptive analysis The baseline characteristics of the sample were described using simple cross-tabulations to show the distribution of variables by diagnosis of learning disability. Analysis of these characteristics was carried out by performing chi-squared tests on binary characteristics and t-tests for continuous variables to obtain p-values for their associations. Crude rates of microvascular and macrovascular complications, insulin initiation, and death were calculated and reported per 1000 person-years.Multivariable analysis A complete case analysis was conducted and included individuals with complete data for gender, age, and ethnicity. Data for learning disability, medications, and comorbidities contained no missing values as these were considered present if coded and absent if not coded. An analysis of the distribution of missing data by exposure status was also conducted.To analyze glycemic control as an outcome, both logistic and linear regression were used to explore HbA1c as a binary variable (poor/good glycemic control) and continuous variable (absolute percentage value) respectively. To examine the rates of complications, insulin initiation, and death, Cox proportional hazards regression models were used. Wald and likelihood ratio tests were used to test the strength of association, and the proportional hazards assumptions were tested in all Cox regression models. A priori confounders included in the regression models were age at diagnosis, gender, and ethnicity. Variables identified as potentially lying on the causal pathway between learning disability and T2DM-related outcomes were examined as potential effect modifiers (figure 1). Clustering of patients by GP practice was also accounted for using robust standard errors across all types of regression model.For multivariable analyses, ethnicity was grouped into White, Other, and Unknown. This was due to the high proportion of unknown ethnicity (48.99%) leading to low numbers of individuals with learning disabilities of South Asian, Black, Other, and Mixed ethnicity.Stata 17 was used to conduct all analyses.Ethical considerations This project was approved by the London School of Hygiene & Tropical Medicine (LSHTM) Research Ethics Committee (Approval Ref: 25836) and the Clinical Practice Research Datalink (CPRD) Independent Scientific Advisory Committee (Ref: 21_000432).Results Of the 352 215 individuals diagnosed with T2DM between 2004 and 2021 extracted from the data, 280 300 met the criteria to be included in the study, with 2074 (0.74%) having a learning disability at T2DM diagnosis ( figure 2). Individuals with learning disabilities were younger at baseline (mean age 51 vs 64 years), had shorter duration of follow-up (5.3 vs 6.0 years), higher proportions of men, people of white ethnicity, people with severe obesity, and people residing in highly deprived geographic areas. Compared with those without learning disabilities, people with learning disabilities had higher HbA1c, lower systolic blood pressure, and more prescriptions of antihypertensive and lipid-lowering medications at baseline. Those with learning disabilities also had considerably more diabetes-related complications at the time of T2DM diagnosis. In the year preceding T2DM diagnosis, people with learning disabilities had a higher average number of consultations than people without learning disabilities (15 vs 12) (table 1).Figure 2Flowchart showing the application of exclusion criteria to the Clinical Practice Research Datalink (CPRD) cohort dataset. *Type 2 diabetes mellitus (T2DM) diagnoses over the age of 18 years between January 2004 and January 2021 from the January 2021 version of the CPRD.Table 1Baseline characteristics of study cohort stratified by learning disability diagnosis at baselineBaseline characteristicLearning disability (n (%) or mean average (95% CI))P-value*No learning disability (N=278 226)Learning disability (N=2074)Sex Male155 891 (56.03)1188 (57.28)0.253 Female122 355 (43.97)886 (42.72)Age at T2DM diagnosis (years) <5048 772 (17.53)936 (45.13)<0.001 50–5964 426 (23.16)561 (27.05) 60–6976 692 (27.56)361 (17.41) 70+88 336 (31.75)216 (10.41) Mean (95% CI)62.43 (62.38 to 62.48)51.08 (50.47 to 51.69)Length of follow-up (years) Mean (95% CI)6.03 (6.01 to 6.04)5.29 (5.13 to 5.45)<0.0001Ethnicity White128 141 (46.06)1190 (57.38)<0.001 South Asian7780 (2.80)31 (1.49) Black3209 (1.15)15 (0.72) Other1979 (0.71)<5 (0.14) Mixed621 (0.22)8 (0.39) Missing136 496 (49.06)827 (39.87)IMD quintile 1 (least deprived)47 636 (17.12)249 (12.01)<0.001 248 583 (17.46)314 (15.14) 358 113 (20.89)402 (19.38) 460 303 (21.67)517 (24.93) 5 (most deprived)63 591 (22.86)592 (28.54)Glycemic control (HbA1c%) Good glycemic control (HbA1c ≤7%)107 801 (38.06)665 (32.06)<0.001 Poor glycemic control (HbA1c >7%)118 219 (42.49)998 (48.12) Missing52 206 (18.76)411 (19.82)BMI (kg/m2) <2525 221 (9.06)171 (8.24)<0.001 25–29.9972 968 (26.23)393 (18.95) 30–39.99115 073 (41.36)841 (40.55) 40+26 551 (9.54)321 (15.48) Missing38 413 (13.81)348 (16.78)Smoking status Non-smoker109 290 (39.28)1119 (53.95)<0.001 Current smoker45 642 (16.40)356 (17.16) Ex-smoker89 677 (32.23)282 (13.60) Missing33 617 (12.08)317 (15.28)Statin prescription in 12 months prior to T2DM diagnosis160 875 (57.82)850 (40.98)<0.001Presence of diabetes-related complications61 733 (22.22)268 (10.67)<0.001Systolic blood pressure (mmHg) <12027 437 (9.86)340 (16.39)<0.001 120–139115 964 (41.68)932 (44.94) 140–15990 893 (32.67)503 (24.25) 160+32 258 (11.59)141 (6.80) Missing11 674 (4.20)158 (7.62)Antihypertensive prescription in 12 months prior to T2DM diagnosis185 459 (66.66)954 (46.00)<0.001Mean number of consultations in 12 months prior to T2DM diagnosis Mean (95% CI)12.18 (12.14 to 12.22)14.99 (14.42 to 15.56)<0.0001 Missing120 (0.04)<5 (0.14)*P-values obtained by chi-squared test for categorical variables and two-tailed t-tests for continuous variables, comparing those with learning disabilities to those without.HbA1c, glycated hemoglobin; IMD, Index of Multiple Deprivation; T2DM, type 2 diabetes mellitus.Missing data Eight variables contained missing data values, which are shown in online supplemental table S1. These were ethnicity (48.99%), smoking status (12.11%), systolic blood pressure (4.22%), BMI (13.83%), consultations in prior year to diagnosis (0.04%), as well as HbA1c at baseline (18.77%), 2 years (29.54%), and 5 years (53.73%). Analysis of the distribution of missing HbA1c data showed no evidence of an association between missingness of baseline HbA1c and learning disability. However, there was strong evidence to suggest that those with learning disabilities had a greater number of missing HbA1c values at 2 and 5 years than those without learning disabilities.SP110.1136/bmjdrc-2024-004879.supp1Supplementary dataGlycemic control After adjustment for a priori confounders, there was strong evidence to suggest that those with learning disabilities had lower average HbA1c% than those without learning disabilities at 2 years (−0.05, 95% CI −0.07 to −0.02, p<0.001) and 5 years (−0.05, 95% CI −0.08 to −0.02, p=0.004) post-diagnosis (table 2). Considering HbA1c as a binary variable, people with learning disabilities had reduced odds of poor glycemic control compared with people without learning disabilities at 2 years (OR 0.82, 95% CI 0.73 to 0.92, p=0.001) and 5 years (OR 0.81, 95% CI 0.70 to 0.94, p=0.004) post-T2DM diagnosis, after adjustment for confounders (table 3).Table 2Crude and adjusted association between learning disability and glycated hemoglobin percentage at 2 and 5 years after type 2 diabetes mellitus diagnosisTime pointLearning disabilityNMean HbA1c% (SD)Crude coefficient (95% CI)P-value*Adjusted coefficient (95% CI)†P-value*2 yearsNo learning disability196 1467.04 (1.23)0.02 (−0.01 to 0.05)0.150−0.05 (−0.07 to –0.02)<0.001Learning disability13517.15 (1.47)5 yearsNo learning disability128 8987.28 (1.36)0.04 (0.00 to 0.07)0.031−0.05 (−0.08 to –0.02)0.004Learning disability7977.52 (1.74)*Calculated using Wald tests in logistic regression models with robust standard errors to adjust for clustering by general practitioner practice.†Adjusted for sex, ethnicity, and age at type 2 diabetes mellitus diagnosis.CI, confidence interval; HbA1c, glycated hemoglobin; SD, standard deviation.Table 3Crude and adjusted odds ratios for the association between learning disability and glycemic control at 2 and 5 years after type 2 diabetes mellitus diagnosisTime pointLearning disabilityNn (%) with poor glycemic control >7% HbA1cCrude OR (95% CI)P-value*Adjusted OR (95% CI)†P-value*2 yearsNo learning disability196 14676 167 (38.83)11Learning disability1351551 (40.78)1.08 (0.97 to 1.21)0.1470.82 (0.73 to 0.92)0.0015 yearsNo learning disability128 89861 135 (47.43)11Learning disability797408 (51.19)1.16 (1.01 to 1.33)0.0310.81 (0.70 to 0.94)0.004*Calculated using Wald tests in logistic regression models with robust standard errors to adjust for clustering by general practitioner practice.†Adjusted for sex, ethnicity, and age at type 2 diabetes mellitus diagnosis.CI, confidence interval; HbA1c, glycated hemoglobin; OR, odds ratio.Potential effect modifiers were added to the logistic regression model to assess their effect on the adjusted odds ratio for the relationship between glycemic control and learning disability status (online supplemental table S2). None of these variables showed any sign of effect modification.Insulin initiation and diabetes-related complications After adjustment for confounders, there was no evidence of an association between learning disability and the risk of microvascular (hazard ratio (HR) 1.09, 95% CI 0.96 to 1.23, p=0.190) and macrovascular (HR 1.13, 95% CI 0.89 to 1.44, p=0.308) complications. However, there was weak evidence that people with learning disabilities initiated insulin faster than those without learning disabilities (HR 1.20, 95% CI 1.00 to 1.45, p=0.052) ( table 4).Table 4Crude and adjusted hazard ratios for the association between learning disability and type 2 diabetes mellitus-related outcomesOutcomeLearning disabilityRate per 1000 person-years (95% CI)Crude HR (95% CI)P-value for crude HR*Adjusted HR (95% CI)†P-value for adjusted HR*Insulin initiationNo learning disability8.66 (8.52 to 8.81)11Learning disability10.66 (8.90 to 12.78)1.49 (1.24 to 1.80)<0.0011.20 (1.00 to 1.45)0.052Microvascular complicationsNo learning disability32.60 (32.33 to 32.88)11Learning disability27.98 (25.02 to 31.29)1.00 (0.89 to 1.13)1.0001.09 (0.96 to 1.23)0.190Macrovascular complicationsNo learning disability9.76 (9.61 to 9.91)11Learning disability7.11 (5.69 to 8.87)0.84 (0.67 to 1.06)0.1451.13 (0.89 to 1.44)0.308*Calculated using Wald tests in Cox regression models with robust standard errors to adjust for clustering by general practitioner practice.†Adjusted for sex, ethnicity, and age at type 2 diabetes mellitus diagnosis.CI, confidence interval; HR, hazard ratio.Potential effect modifiers of the relationship between insulin initiation and having a learning disability were added to the Cox regression model in online supplemental table S3. Adjusting for baseline glycemic control produced a reduction in the effect size (HR 1.04, 95% CI 0.81 to 1.34, p=0.767), suggesting that it acts as an effect modifier for the association between learning disability and insulin initiation. Models including the other potential effect modifiers showed no considerable signs of effect modification.Mortality Of the total cohort, 43% were eligible for ONS linkage, limiting the analysis of diabetes-related mortality to 121 004 individuals. The ONS-linked cohort had a smaller proportion of people with learning disabilities compared with the whole study population (0.63% vs 0.74%). After adjustment for confounders, we found strong evidence that people with learning disabilities had approximately twice the risk of both all-cause mortality (HR 1.81, 95% CI 1.60 to 2.04, p<0.001) and diabetes-related mortality (HR 2.01, 95% CI 1.43 to 2.84, p<0.001) ( table 5).Table 5Crude and adjusted hazard ratios for the association between learning disability and mortalityOutcomeLearning disabilitynRate per 1000 person-years (95% CI)Crude HR (95% CI)P-value for crude HR*Adjusted HR (95% CI)†P-value for adjusted HR*All-cause mortality (N=280 300)No learning disability278 22636.26 (35.99 to 36.57)11Learning disability207430.44 (27.35 to 33.89)0.87 (0.77 to 0.97)0.0171.81 (1.60 to 2.04)<0.001ONS-linked diabetes-related mortality (N=121 004)No learning disability120 24211.57 (11.31 to 11.84)11Learning disability76210.74 (7.82 to 14.76)0.98 (0.69 to 1.40)0.9152.01 (1.43 to 2.84)<0.001*Calculated using Wald tests in Cox regression models with robust standard errors to adjust for clustering by general practitioner practice.†Adjusted for sex, ethnicity, and age at type 2 diabetes mellitus diagnosis.CI, confidence interval; HR, hazard ratio; ONS, Office for National Statistics.Discussion Using large-scale UK electronic health records, we conducted an observational cohort study of adults with T2DM to determine the association between learning disability and glycemic control, vascular outcomes, and mortality. We found strong evidence to suggest that those with learning disabilities had better glycemic control than those without learning disabilities. However, people with learning disabilities had higher rates of insulin initiation, our proxy for severe diabetes, suggesting that a greater proportion had severe diabetes as they required treatment escalation beyond lifestyle changes and oral antidiabetic medications. While we found no difference in the risk of microvascular and macrovascular complications between people with and without learning disabilities, we found strong evidence across both the CPRD and ONS-linked cohorts that people with learning disabilities were at substantially higher risk of all-cause and diabetes-related mortality than those without learning disabilities.Contrary to our initial hypothesis, we found that glycemic control was better in individuals with a learning disability. This could be for several reasons. First, HbA1c recording was poorer in those with learning disabilities. GPs who are more diligent at recording HbA1c may also provide superior T2DM care overall. Therefore, the population with complete HbA1c data may have had better glycemic control than the population with missing data, who were disproportionately represented among those with learning disabilities. Another possible explanation is that those with learning disabilities may have had better glycemic control because they are in more frequent contact with primary care.3 This echoes the findings that multimorbid T2DM patients are more likely to be treated with insulin, so often achieve lower HbA1c levels than those without comorbidities.16 This is also combined with findings that show that risk factor management is often better in those with multimorbidity than in those with single conditions.17 Another possible contributor could be that those with learning disabilities are more likely to have caregivers or live in assisted living. There is some evidence that among adults with poorly controlled diabetes, those who receive assistance from a caregiver may have higher medication adherence than those without.18 Family support in diabetes care has also been associated with improved blood sugar control.19 20 In this way, the presence of caregivers could potentially lead to better adherence to diabetes monitoring and treatment regimens for those with learning disabilities.This aligns with our second finding, namely that people with learning disabilities had higher rates of insulin initiation than those without learning disabilities. Our finding that poor glycemic control at baseline acted as an effect modifier for this relationship suggests that increased insulin use in those with learning disabilities may have been related to poorer glycemic control at diagnosis. This increased insulin use may have then led to people with learning disabilities having better controlled HbA1c throughout the observation period.Our finding that people with learning disabilities were at far higher risk of all-cause and diabetes-related mortality echoes those of a similar study which reported that people with learning disabilities are more likely to be hospitalized and have higher mortality due to diabetes-related causes, especially cardiovascular-related complications.21Strengths and limitations A key strength of this study was the use of large-scale, routinely collected, electronic patient data, which allowed us to study a rare exposure with sufficient statistical power over a substantial follow-up time. We had access to linked deprivation and mortality data, the latter of which enabled us to examine cause-specific mortality in this population.The CPRD has been found to be representative of the UK population, ensuring good generalizability, despite the fact that contributing GP practices are not evenly distributed throughout the UK.10 Data quality is also generally good for T2DM due to the UK’s Quality and Outcome Framework (QoF), which provides general practice remuneration for recorded quality of care for chronic conditions including T2DM and learning disabilities.22Selection bias may have been introduced when using the ONS-linked cohort to examine all-cause and cause-specific mortality. Only 43.17% of the cohort were eligible for ONS-linkage, and the distribution of people with and without learning disabilities differed from that of the whole cohort. However, effect sizes from the analyses of whole cohort and ONS-linked cohorts were comparable, suggesting that selection bias had a limited effect on our findings.Those with learning disabilities had a shorter follow-up time than those without a learning disability. This higher loss to follow-up may be due to higher general rates of adverse health events, death, and censoring in this population, introducing selection bias in the form of survivor bias and making the exposure groups less comparable. Differences in follow-up time could also be due to social or structural factors, such as difficulties in communication, care, and transport for people with learning disabilities.Of those diagnosed with a learning disability at any point, 438 (17.44%) were diagnosed after their T2DM diagnosis. These patients were therefore classified as unexposed for the duration of the study, resulting in potential exposure misclassification.Recording of treatment and complication data is highly dependent on the individual data collection processes in the general practice electronic health record systems. Any variation in quality and completeness of routinely collected data could be related to a multitude of both patient- and GP practice-related factors.10 As CPRD data are GP-recorded health data captured during routine consultations, lifestyle factors such as diet, exercise, living situation, and insurance status are infrequently recorded, so they cannot be reliably examined with the available data.23 It is a recognized limitation of the CPRD dataset that there are not generally standardized definitions for diagnoses, which can lead to some subjectivity between data recorders.10There were large numbers of missing values for the outcome variables for glycemic control. Those with learning disabilities were more likely to have missing glycemic control values both at diagnosis and at 2 and 5 years post-T2DM diagnosis. Following this pattern, it is possible that there may have been either underdiagnosing or under-recording of complications rates in those with learning disabilities compared with those without.Missing data, notably HbA1c, were more common in those with learning disabilities than those without, with the difference increasing over time. This increasing missingness may indicate greater loss to follow-up in those with learning disabilities, which may be due to higher rates of adverse health events. As those with learning disabilities have a higher burden of comorbid chronic disease, it is possible that less attention would be given to long-term diabetes care alone while other comorbid illnesses are addressed.2 3 Data completeness may also act as a proxy for healthcare-seeking behavior, while missing data may indicate poorer access to healthcare, poorer quality of care, and poor attendance. This pattern of missing data could lead to an overestimation of the protective effect of learning disability on glycemic control, as those with worse glycemic control may be more likely to have been lost to follow-up at 2 and 5 years. In this way, the high missingness of HbA1c may have partially contributed to the unexpected findings that those with learning disabilities have better glycemic control, biasing our results away from the null hypothesis.Another variable with a very large number of missing values is ethnicity, with nearly half of all patients having no ethnicity data. Due to this, there were very low numbers of recorded South Asian, Black, Other, and Mixed ethnicity T2DM patients with learning disabilities. This greatly limited our ability to examine the role of confounding by ethnicity or for the exploration of the differences in the effect of learning disability on T2DM outcomes by ethnicity. Our data suggested that those with learning disabilities had a greater proportion of missing data in these variables. This variability of missingness is a recognized limitation of the CPRD dataset and raises concerns about worse data collection in those with learning disabilities.10Future directions Our finding of higher rates of insulin initiation in those with learning disabilities warrants further investigation into whether this is due to poorer glycemic control at presentation (and therefore faster advancing T2DM) or due to having a greater degree of clinical surveillance. The potential damaging effect of overuse of insulin and hypoglycemia, as seen in those with multimorbidity, could also be explored.People with learning disabilities often present with multiple comorbidities and have substantially higher rates of health service usage.24 Mortality rates are therefore also substantially higher in the learning disability population. However, people with learning disabilities can have difficulties accessing healthcare and can often be faced with social and organizational limitations of care.25 It will also be important to explore the effect of health and social care on T2DM outcomes in people with learning disabilities to understand the extent to which quality and frequency of care have a protective effect. Any shortcomings in healthcare for people with learning disabilities may be related to general barriers to healthcare access for those with learning disabilities, such as overlooked pain, lower screening rates, and insufficient training of staff.26–28People with learning disabilities have substantially higher rates of health service usage.24 However, cognitive impairment can lead to difficulties in communicating and being understood by health services. The Confidential Inquiry into the premature deaths of those with learning disabilities found a far greater burden of avoidable deaths from causes that could be amenable to higher-quality care than in the general population.29 This presents particular difficulties to healthcare providers as GPs often feel they lack experience and training in caring for patients with learning disabilities.30 Qualitative research has indicated that rigid health and social care systems may not fit the needs of individuals with learning disabilities.31 In this way, reasonable adjustments by healthcare staff to understand the needs of those with learning disabilities could be beneficial, such as allowing sufficient time and space for information exchanges and consultations.Conclusions This study provides novel evidence of the effect of having a learning disability on various outcomes of T2DM. Further research into the roles of insulin initiation and social care could be helpful in aiding understanding of any disparities.