Automated identification of miscoded and misclassified cases of diabetes from computer records

The pattern of incident fractures according to fracture site in people with T1D

Higher incidences of fractures are seen in people with type 1 diabetes (T1D), but knowledge on different fracture sites is sparse. We found a higher incidence mainly for distal fracture sites in people with T1D compared to controls. It must be further studied which fractures attributed to the higher incidence rates (IRs) at specific sites.

INTRODUCTION

People with T1D have a higher incidence of fractures compared to the general population. However, sparse knowledge exists on the incidence rates of individual fracture sites. Therefore, we examined the incidence of various fracture sites in people with newly treated T1D compared to matched controls.

METHODS

All people from the UK Clinical Practice Research Datalink GOLD (1987-2017), of all ages with a T1D diagnosis code (n = 6381), were included. People with T1D were matched by year of birth, sex, and practice to controls (n = 6381). Fracture IRs and incidence rate ratios (IRRs) were calculated. Analyses were stratified by fracture site and sex.

RESULTS

The IR of all fractures was significantly higher in people with T1D compared to controls (IRR: 1.39 (CI95%: 1.24-1.55)). Compared to controls, the IRR for people with T1D was higher for several fracture sites including carpal (IRR: 1.41 (CI95%: 1.14-1.75)), clavicle (IRR: 2.10 (CI95%: 1.18-3.74)), foot (IRR: 1.70 (CI95%: 1.23-2.36)), humerus (IRR: 1.46 (CI95%: 1.04-2.05)), and tibia/fibula (IRR: 1.67 CI95%: 1.08-2.59)). In women with T1D, higher IRs were seen at the ankle (IRR: 2.25 (CI95%: 1.10-4.56)) and foot (IRR: 2.11 (CI95%: 1.27-3.50)), whereas in men with T1D, higher IRs were seen for carpal (IRR: 1.45 (CI95%: 1.14-1.86)), clavicle (IRR: 2.13 (CI95%: 1.13-4.02)), and humerus (IRR: 1.77 (CI95%: 1.10-2.83)) fractures.

CONCLUSION

The incidence of carpal, clavicle, foot, humerus, and tibia/fibula fractures was higher in newly treated T1D, but there was no difference at other fracture sites compared to controls. Therefore, the higher incidence of fractures in newly treated people with T1D has been found mainly for distal fracture sites.

Detecting Miscoded Diabetes Diagnosis Codes in Electronic Health Records for Quality Improvement: Temporal Deep Learning Approach

BACKGROUND

Diabetes affects more than 30 million patients across the United States. With such a large disease burden, even a small error in classification can be significant. Currently billing codes, assigned at the time of a medical encounter, are the "gold standard" reflecting the actual diseases present in an individual, and thus in aggregate reflect disease prevalence in the population. These codes are generated by highly trained coders and by health care providers but are not always accurate.

OBJECTIVE

This work provides a scalable deep learning methodology to more accurately classify individuals with diabetes across multiple health care systems.

METHODS

We leveraged a long short-term memory-dense neural network (LSTM-DNN) model to identify patients with or without diabetes using data from 5 acute care facilities with 187,187 patients and 275,407 encounters, incorporating data elements including laboratory test results, diagnostic/procedure codes, medications, demographic data, and admission information. Furthermore, a blinded physician panel reviewed discordant cases, providing an estimate of the total impact on the population.

RESULTS

When predicting the documented diagnosis of diabetes, our model achieved an 84% F1 score, 96% area under the curve-receiver operating characteristic curve, and 91% average precision on a heterogeneous data set from 5 distinct health facilities. However, in 81% of cases where the model disagreed with the documented phenotype, a blinded physician panel agreed with the model. Taken together, this suggests that 4.3% of our studied population have either missing or improper diabetes diagnosis.

CONCLUSIONS

This study demonstrates that deep learning methods can improve clinical phenotyping even when patient data are noisy, sparse, and heterogeneous.

Does Renal Function or Heart Failure Diagnosis Affect Primary Care Prescribing for Sodium-Glucose Co-Transporter 2 Inhibitors in Type 2 Diabetes?

INTRODUCTION

Sodium-glucose co-transporter 2 inhibitors (SGLT2is) are a unique class of drugs currently used in the management of type 2 diabetes (T2D). There are emerging data from cardiovascular outcome trials confirming renal and heart failure benefits of these drugs independent of glucose lowering. By contrast, the current licencing indications of these drugs are mainly limited to their glucose-lowering effects, and not to renal or heart failure benefits. It is therefore timely to ascertain whether the presence of these clinical conditions may influence prescribing choices for patients with T2D. Our aims are to report prescribing of SGLT2is in people with T2D according to their renal function and presence of heart failure. Co-prescribing with diuretics will also be explored.

METHODS

We will perform a cross-sectional analysis of people with T2D in the Royal College of General Practitioners (RCGP) Research and Surveillance Centre (RSC) network. The RCGP RSC includes more than 1500 volunteer practices throughout England and parts of Wales, and a representative sample of over 10 million patients. The proportion of adults with T2D ever prescribed an SGLT2i will be determined. Within this cohort, we will calculate the percentage of SGLT2is prescribed according to renal function, and the proportion of prescriptions in people with co-morbid heart failure, stratified by body mass index categories. The percentage of SGLT2is prescribed as an add-on to a diuretic or following discontinuation of prescribing for a diuretic will also be reported. Multilevel logistic regression will be performed to explore the association between heart failure and renal function, and propensity to prescribe SGLT2is.

PLANNED OUTPUTS

The study findings will be submitted to a primary care/diabetes-focused conference, and for publication in a peer reviewed journal.

Weight loss and mortality risk in patients with different adiposity at diagnosis of type 2 diabetes: a longitudinal cohort study

BACKGROUND

Undiagnosed comorbid diseases that independently lead to weight loss before type 2 diabetes mellitus (T2DM) diagnosis could explain the observed increased mortality risk in T2DM patients with normal weight.

OBJECTIVES

To evaluate the impact of weight change patterns before the diagnosis of T2DM on the association between body mass index (BMI) at diagnosis and mortality risk.

METHODS

This was a longitudinal cohort study using 145,058 patients from UK primary care, with newly diagnosed T2DM from January 2000. Patients aged 18-70, without established disease history at diagnosis (defined as the presence of cardiovascular diseases, cancer, and renal diseases on or before diagnosis) were followed up to 2014. Longitudinal 6-monthly measures of bodyweight three years before (used to define groups of patients who lost bodyweight or not before diagnosis) and 2 years after diagnosis were obtained. The main outcome was all-cause mortality.

RESULTS

At diagnosis, mean (SD) age was 52 (12) years, 56% were male, 52% were current or ex-smokers, mean BMI was 33 kg/m², and 66% were obese. Normal weight and overweight patients experienced a small but significant reduction in body weight 6 months before diagnosis. Among all categories of obese patients, consistently increasing body weight was observed within the same time window. Among patients who did not lose body weight pre-diagnosis (n = 117,469), compared with the grade 1 obese, normal weight patients had 35% (95% CI of HR: 1.17, 1.55) significantly higher adjusted mortality risk. However, among patients experiencing weight loss before diagnosis (n = 27,589), BMI at diagnosis was not associated with mortality risk (all p > 0.05).

CONCLUSIONS

Weight loss before the diagnosis of T2DM was not associated with the observed increased mortality risk in normal weight patients with T2DM. This emphasises the importance of addressing risk factors post diagnosis for excess mortality in this group.

Data Mining Approach to Identify Disease Cohorts from Primary Care Electronic Medical Records: A Case of Diabetes Mellitus

Background:

Identification of diseased patients from primary care based electronic medical records (EMRs) has methodological challenges that may impact epidemiologic inferences.

Objective:

To compare deterministic clinically guided selection algorithms with probabilistic machine learning (ML) methodologies for their ability to identify patients with type 2 diabetes mellitus (T2DM) from large population based EMRs from nationally representative primary care database.

Methods:

Four cohorts of patients with T2DM were defined by deterministic approach based on disease codes. The database was mined for a set of best predictors of T2DM and the performance of six ML algorithms were compared based on cross-validated true positive rate, true negative rate, and area under receiver operating characteristic curve.

Results:

In the database of 11,018,025 research suitable individuals, 379 657 (3.4%) were coded to have T2DM. Logistic Regression classifier was selected as best ML algorithm and resulted in a cohort of 383,330 patients with potential T2DM. Eighty-three percent (83%) of this cohort had a T2DM code, and 16% of the patients with T2DM code were not included in this ML cohort. Of those in the ML cohort without disease code, 52% had at least one measure of elevated glucose level and 22% had received at least one prescription for antidiabetic medication.

Conclusion:

Deterministic cohort selection based on disease coding potentially introduces significant mis-classification problem. ML techniques allow testing for potential disease predictors, and under meaningful data input, are able to identify diseased cohorts in a holistic way.

Life Expectancy and Cause-Specific Mortality in Type 2 Diabetes: A Population-Based Cohort Study Quantifying Relationships in Ethnic Subgroups

OBJECTIVES

This study 1) investigated life expectancy and cause-specific mortality rates associated with type 2 diabetes and 2) quantified these relationships in ethnic subgroups.

RESEARCH DESIGN AND METHODS

This was a cohort study using Clinical Practice Research Datalink data from 383 general practices in England with linked hospitalization and mortality records. A total of 187,968 patients with incident type 2 diabetes from 1998 to 2015 were matched to 908,016 control subjects. Abridged life tables estimated years of life lost, and a competing risk survival model quantified cause-specific hazard ratios (HRs).

RESULTS

A total of 40,286 deaths occurred in patients with type 2 diabetes. At age 40, white men with diabetes lost 5 years of life and white women lost 6 years compared with those without diabetes. A loss of between 1 and 2 years was observed for South Asians and blacks with diabetes. At age older than 65 years, South Asians with diabetes had up to 1.1 years' longer life expectancy than South Asians without diabetes. Compared with whites with diabetes, South Asians with diabetes had lower adjusted risks for mortality from cardiovascular (HR 0.82; 95% CI 0.75, 0.89), cancer (HR 0.43; 95% CI 0.36, 0.51), and respiratory diseases (HR 0.60; 95% CI 0.48, 0.76). A similar pattern was observed in blacks with diabetes compared with whites with diabetes.

CONCLUSIONS

Type 2 diabetes was associated with more years of life lost among whites than among South Asians or blacks, with older South Asians experiencing longer life expectancy compared with South Asians without diabetes. The findings support optimized cardiovascular disease risk factor management, especially in whites with type 2 diabetes.

Quality of recording of diabetes in the UK: how does the GP's method of coding clinical data affect incidence estimates? Cross-sectional study using the CPRD database

OBJECTIVE

To assess the effect of coding quality on estimates of the incidence of diabetes in the UK between 1995 and 2014.

DESIGN

A cross-sectional analysis examining diabetes coding from 1995 to 2014 and how the choice of codes (diagnosis codes vs codes which suggest diagnosis) and quality of coding affect estimated incidence.

SETTING

Routine primary care data from 684 practices contributing to the UK Clinical Practice Research Datalink (data contributed from Vision (INPS) practices).

MAIN OUTCOME MEASURE

Incidence rates of diabetes and how they are affected by (1) GP coding and (2) excluding 'poor' quality practices with at least 10% incident patients inaccurately coded between 2004 and 2014.

RESULTS

Incidence rates and accuracy of coding varied widely between practices and the trends differed according to selected category of code. If diagnosis codes were used, the incidence of type 2 increased sharply until 2004 (when the UK Quality Outcomes Framework was introduced), and then flattened off, until 2009, after which they decreased. If non-diagnosis codes were included, the numbers continued to increase until 2012. Although coding quality improved over time, 15% of the 666 practices that contributed data between 2004 and 2014 were labelled 'poor' quality. When these practices were dropped from the analyses, the downward trend in the incidence of type 2 after 2009 became less marked and incidence rates were higher.

CONCLUSIONS

In contrast to some previous reports, diabetes incidence (based on diagnostic codes) appears not to have increased since 2004 in the UK. Choice of codes can make a significant difference to incidence estimates, as can quality of recording. Codes and data quality should be checked when assessing incidence rates using GP data.

Glucose test provenance recording in UK primary care: was that fasted or random?

AIMS

To describe the proportion of glucose tests with unrecorded provenance in routine primary care data and identify the impact on clinical practice.

METHODS

A cross-sectional analysis was conducted of blood glucose measurements from the Royal College of General Practitioner Research and Surveillance Centre database, which includes primary care records from >100 practices across England and Wales. All blood glucose results recorded during 2013 were identified. Tests were grouped by provenance (fasting, oral glucose tolerance test, random, none specified and other). A clinical audit in a single primary care practice was also performed to identify the impact of failing to record glucose provenance on diabetes diagnosis.

RESULTS

A total of 2 137 098 people were included in the cross-sectional analysis. Of 203 350 recorded glucose measurements the majority (117 893; 58%) did not have any provenance information. The most commonly reported provenance was fasting glucose (75 044; 37%). The distribution of glucose values where provenance was not recorded was most similar to that of fasting samples. The glucose measurements of 256 people with diabetes in the audit practice (size 11 514 people) were analysed. The initial glucose measurement had no provenance information in 164 cases (64.1%). A clinician questioned the provenance of a result in 41 cases (16.0%); of these, 14 (34.1%) required repeating. Lack of provenance led to delays in the diagnosis of diabetes [median (range) 30 (3-614) days].

CONCLUSIONS

The recording of glucose provenance in UK primary care could be improved. Failure to record provenance causes unnecessary repeated testing, delayed diagnosis and wasted clinician time.

Is diabetes mellitus correctly registered and classified in primary care? A population-based study in Catalonia, Spain

OBJECTIVE

To assess the prevalence of miscoding, misclassification, misdiagnosis and under-registration of diabetes mellitus (DM) in primary health care in Catalonia (Spain), and to explore use of automated algorithms to identify them.

METHODS

In this cross-sectional, retrospective study using an anonymized electronic general practice database, data were collected from patients or users with a diabetes-related code or from patients with no DM or prediabetes code but treated with antidiabetic drugs (unregistered DM). Decision algorithms were designed to classify the true diagnosis of type 1 DM (T1DM), type 2 DM (T2DM), and undetermined DM (UDM), and to classify unregistered DM patients treated with antidiabetic drugs.

RESULTS

Data were collected from a total of 376,278 subjects with a DM ICD-10 code, and from 8707 patients with no DM or prediabetes code but treated with antidiabetic drugs. After application of the algorithms, 13.9% of patients with T1DM were identified as misclassified, and were probably T2DM; 80.9% of patients with UDM were reclassified as T2DM, and 19.1% of them were misdiagnosed as DM when they probably had prediabetes. The overall prevalence of miscoding (multiple codes or UDM) was 2.2%. Finally, 55.2% of subjects with unregistered DM were classified as prediabetes, 35.7% as T2DM, 8.5% as UDM treated with insulin, and 0.6% as T1DM.

CONCLUSIONS

The prevalence of inappropriate codification or classification and under-registration of DM is relevant in primary care. Implementation of algorithms could automatically flag cases that need review and would substantially decrease the risk of inappropriate registration or coding.

Association of smoking and concomitant metformin use with cardiovascular events and mortality in people newly diagnosed with type 2 diabetes

BACKGROUND

The cardiovascular and mortality risk in patients with incident type 2 diabetes (T2D) in relation to smoking status and concurrent use of metformin is not well known.

METHODS

A cohort study was performed in 82,205 incident T2D patients from the U.K. Clinical Practice Research Datalink. In the present study, the risks of myocardial infarction (MI), stroke, and mortality in incident T2D patients were evaluated in relation to their smoking status with and without concurrent use of metformin.

RESULTS

Over a median 5.4 years of follow-up, of patients without a history of cardiovascular disease (CVD) before a diagnosis of diabetes (n = 63,166), current smokers with and without metformin had an 8% (hazard ratio [HR] 1.08; 95% confidence interval [CI] 0.81, 1.45) and 32% (HR 1.32; 95% CI 1.07, 1.65) increased risk of MI or stroke, respectively, compared with non-smokers without metformin treatment. The respective HRs (95% CI) for mortality in these patients were 0.96 (0.83, 1.11) and 1.86 (1.68, 2.07). The HR for mortality among ex-smokers with and without concurrent metformin treatment was 0.92 (95% CI 0.83, 1.11) and 1.19 (95% CI 1.10, 1.30), respectively. Similar beneficial modifiable effects of metformin among ex- and current smokers were observed in patients with CVD before diagnosis of diabetes (n = 19,039).

CONCLUSIONS

In T2D patients, concurrent treatment with metformin attenuates the observed higher cardiovascular and mortality risk in ex- and current smokers. In addition to smoking cessation support, treatment with metformin, particularly in ex- and current smokers, should be encouraged.

Delay in treatment intensification increases the risks of cardiovascular events in patients with type 2 diabetes

BACKGROUND

The aim of the study was to evaluate the effect of delay in treatment intensification (IT; clinical inertia) in conjunction with glycaemic burden on the risk of macrovascular events (CVE) in type 2 diabetes (T2DM) patients.

METHODS

A retrospective cohort study was carried out using United Kingdom Clinical Practice Research Datalink, including T2DM patients diagnosed from 1990 with follow-up data available until 2012.

RESULTS

In the cohort of 105,477 patients mean HbA1c was 8.1% (65 mmol/mol) at diagnosis, 11% had a history of cardiovascular disease, and 7.1% experienced at least one CVE during 5.3 years of median follow-up. In patients with HbA1c consistently above 7/7.5% (53/58 mmol/mol, n = 23,101/11,281) during 2 years post diagnosis, 26/22% never received any IT. Compared to patients with HbA1c <7% (<53 mmol/mol), in patients with HbA1c ≥7% (≥53 mmol/mol), a 1 year delay in receiving IT was associated with significantly increased risk of MI, stroke, HF and composite CVE by 67% (HR CI: 1.39, 2.01), 51% (HR CI: 1.25, 1.83), 64% (HR CI: 1.40, 1.91) and 62% (HR CI: 1.46, 1.80) respectively. One year delay in IT in interaction with HbA1c above 7.5% (58 mmol/mol) was also associated with similar increased risk of CVE.

CONCLUSIONS

Among patients with newly diagnosed T2DM, 22% remained under poor glycaemic control over 2 years, and 26% never received IT. Delay in IT by 1 year in conjunction with poor glycaemic control significantly increased the risk of MI, HF, stroke and composite CVE.

Obesity paradox in people newly diagnosed with type 2 diabetes with and without prior cardiovascular disease

AIM

To address the debate on 'obesity paradox' in patients with type 2 diabetes mellitus (T2DM) by evaluating the cardiovascular and mortality risks associated with normal and overweight patients compared to obese at diagnosis of diabetes, separately for patients with and without cardiovascular disease (CVD) before diagnosis.

METHODS

A retrospective study with two study cohorts with/without prior CVD (n = 10237/37272) with complete measures of body mass index (BMI) at diagnosis of T2DM from UK General Practice Research Database. Primary outcomes were long-term risks of cardiovascular events (CVEs) and all-cause mortality in patients with normal weight, overweight and obesity at diagnosis.

RESULTS

The mortality rates per 1000 person-years in normal weight, overweight and obese patients among patients without prior CVD were 13.1, 8.6 and 6.0, respectively, during 5 years of median follow-up. For patients with prior CVD, these estimates were 30.1, 21.1 and 15.5, respectively. Among patients without and with prior CVD, normal weight patients had 47% (hazard ratio, HR CI: 1.29, 1.69) and 30% (HR CI: 1.11, 1.53) increased mortality risk respectively compared to obese patients. In the cohort without prior CVD, compared to obese patients, those with normal body weight did not have increased CVE risk. Interactions between age, HbA1c and BMI at diagnosis were observed in both cohorts.

CONCLUSIONS

Adults with normal weight at the diagnosis of T2DM have significantly higher mortality risk compared to those who are obese, with significant interactions between age, BMI and HbA1c. Elevated cardiovascular risk was not observed in normal weight patients without prior CVD.

Prevalence and characteristics in coding, classification and diagnosis of diabetes in primary care

INTRODUCTION

Approximately 366 million people worldwide live with diabetes and this figure is expected to rise. Among the correct diagnosis, there will be errors in the diagnosis, classification and coding, resulting in adverse health and financial implications.

AIM

To determine the prevalence and characteristics of diagnostic errors in people with diabetes managed in primary care settings.

METHODS

We conducted a cross-sectional study in nine general practices in Leicester, UK, from May to August 2011, using a validated electronic toolkit. Searches identified cases with potential errors which were manually checked for accuracy.

RESULTS

There were 54 088 patients and 2434 (4.5%) diagnosed with diabetes. Out of 316 people identified with potential errors with the toolkit, 180 (57%) had confirmed errors after manually reviewing the records, resulting in an error prevalence of 7.4%. Correctly coded people on registers had significantly greater glycated haemoglobin (HbA1c) reductions. There were no significant differences between patients with and without errors in their HbA1C, body mass index, age and size of practice. There was also no significant association of the errors with pay-for-performance initiatives; however, those patients not on disease register had worse glycaemic control.

CONCLUSIONS

A high prevalence of diabetic diagnostic errors was confirmed using medication, biochemical and demographic data. Larger studies are needed to more accurately assess the scale of this problem. Automation of these processes might be possible, which would allow searches to be even more user friendly.

Evaluating tools to support a new practical classification of diabetes: excellent control may represent misdiagnosis and omission from disease registers is associated with worse control

AIMS

To conduct a service evaluation of usability and utility on-line clinical audit tools developed as part of a UK Classification of Diabetes project to improve the categorisation and ultimately management of diabetes.

METHOD

We conducted the evaluation in eight volunteer computerised practices all achieving maximum pay-for-performance (P4P) indicators for diabetes; two allowed direct observation and videotaping of the process of running the on-line audit. We also reported the utility of the searches and the national levels of uptake.

RESULTS

Once launched 4235 unique visitors accessed the download pages in the first 3 months. We had feedback about problems from 10 practices, 7 were human error. Clinical audit naive staff ran the audits satisfactorily. However, they would prefer more explanation and more user-familiar tools built into their practice computerised medical record system. They wanted the people misdiagnosed and misclassified flagged and to be convinced miscoding mattered. People with T2DM misclassified as T1DM tended to be older (mean 62 vs. 47 years old). People misdiagnosed as having T2DM have apparently 'excellent' glycaemic control mean HbA1c 5.3% (34 mmol/mol) vs. 7.2% (55 mmol/mol) (p<0.001). People with vague codes not included in the P4P register (miscoded) have worse glycaemic control [HbA1c 8.1% (65 mmol/mol) SEM=0.42 vs.7.0% (53mmol/mol) SEM=0.11, p=0.006].

CONCLUSIONS

There was scope to improve diabetes management in practice achieving quality targets. Apparently 'excellent' glycaemic control may imply misdiagnosis, while miscoding is associated with worse control. On-line clinical audit toolkits provide a rapid method of dissemination and should be added to the armamentarium of quality improvement interventions.