Computerized extraction of information on the quality of diabetes care from free text in electronic patient records of general practitioners

Tailoring Type II Diabetes Treatment: Investigating the Effect of 5-HTT Polymorphisms on HbA1c Levels after Metformin Initiation

Aims

To investigate the effect of serotonin transporter (5-HTT) polymorphisms on change in HbA1c levels six months after metformin initiation in type 2 diabetes patients.

Materials and Methods

Participants of PROVALID (PROspective cohort study in patients with type 2 diabetes mellitus for VALidation of biomarkers) within the GIANTT (Groningen Initiative to ANalyse Type 2 Diabetes Treatment) cohort who initiated metformin were genotyped for combined 5-HTTLPR/rs25531 (L∗L∗, L∗S∗, and S∗S∗) and 5-HTT VNTR (STin 2.12, 12/-, and 10/-) polymorphisms, respectively. Multiple linear regression was applied to determine the change in HbA1c level from baseline date to six months across 5-HTTLPR/VNTR genotype groups, adjusted for baseline HbA1c, age, gender, triglyceride level, low-density lipoprotein level, and serum creatinine.

Results

157 participants were included, of which 56.2% were male. The average age was 59.3 ± 9.23 years, and the mean baseline HbA1c was 7.49% ± 1.21%. 5-HTTLPR was characterized in 46 patients as L∗L∗, 70 patients as L∗S∗, and 41 patients as S∗S∗ genotypes. No significant association was found between 5-HTTLPR and 5-HTT VNTR genotypes and change in HbA1c after adjustments.

Conclusions

5-HTT polymorphisms did not affect HbA1c levels six months after the start of metformin. Further long-term studies in large samples would be relevant to determine which polymorphisms can explain the variation in response to metformin treatment.

RegEMR: a natural language processing system to automatically identify premature ovarian decline from Chinese electronic medical records

BACKGROUND

The ovarian reserve is a reservoir for reproductive potential. In clinical practice, early detection and treatment of premature ovarian decline characterized by abnormal ovarian reserve tests is regarded as a critical measure to prevent infertility. However, the relevant data are typically stored in an unstructured format in a hospital's electronic medical record (EMR) system, and their retrieval requires tedious manual abstraction by domain experts. Computational tools are therefore needed to reduce the workload.

METHODS

We presented RegEMR, an artificial intelligence tool composed of a rule-based natural language processing (NLP) extractor and a knowledge-based disease scoring model, to automatize the screening procedure of premature ovarian decline using Chinese reproductive EMRs. We used regular expressions (REs) as a text mining method and explored whether REs automatically synthesized by the genetic programming-based online platform RegexGenerator +  + could be as effective as manually formulated REs. We also investigated how the representativeness of the learning corpus affected the performance of machine-generated REs. Additionally, we translated the clinical diagnostic criteria into a programmable disease diagnostic model for disease scoring and risk stratification. Four hundred outpatient medical records were collected from a Chinese fertility center. Manual review served as the gold standard, and fivefold cross-validation was used for evaluation.

RESULTS

The overall F-score of manually built REs was 0.9444 (95% CI 0.9373 to 0.9515), with no significant difference (paired t test p > 0.05) compared with machine-generated REs that could be affected by training set sizes and annotation portions. The extractor performed effectively in automatically tracing the dynamic changes in hormone levels (F-score 0.9518-0.9884) and ultrasonographic measures (F-score 0.9472-0.9822). Applying the extracted information to the proposed diagnostic model, the program obtained an accuracy of 0.98 and a sensitivity of 0.93 in risk screening. For each specific disease, the automatic diagnosis in 76% of patients was consistent with that of the clinical diagnosis, and the kappa coefficient was 0.63.

CONCLUSION

A Chinese NLP system named RegEMR was developed to automatically identify high risk of early ovarian aging and diagnose related diseases from Chinese reproductive EMRs. We hope that this system can aid EMR-based data collection and clinical decision support in fertility centers.

Role of Serum Lipids, Blood Glucose and Blood Pressure in Breast Cancer Risk for Women with Type 2 Diabetes Mellitus

PURPOSE

Women with type 2 diabetes mellitus (T2DM) have an increased risk of breast cancer. We aimed to determine the contribution of lipids, glucose and blood pressure to this risk based on the multifactorial nature of T2DM.

PATIENTS AND METHODS

This population-based cohort study used data from a Dutch database (the Groningen Initiative to Analyse Type 2 Diabetes Treatment) for the period 2004-2013. The cohort included women diagnosed with T2DM, aged 30-80 years, with no history of breast cancer and with follow-up data for at least 1 year. We used Cox proportional hazards models to estimate the associations of exposures with breast cancer occurrence, reporting adjusted hazard ratios (aHR) with 95% confidence intervals (CI). Exposures of interest included total cholesterol, high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), triglycerides, glycated hemoglobin A (HbA1c) and systolic blood pressure (SBP).

RESULTS

During a median of 4.45 years' follow-up, 183 of 10,183 included women received a breast cancer diagnosis. We observed U-shaped associations with breast cancer incidence for total cholesterol and HDL-C at baseline. Compared with moderate elevations, women had significantly higher breast cancer risks associated with high total cholesterol (aHR, 95% CI: 1.72, 1.15-2.55) and HDL-C (aHR, 95% CI: 1.74, 1.18-2.58) levels, while low total cholesterol (aHR, 95% CI: 1.43, 0.94-2.19) and HDL-C (aHR, 95% CI: 1.44, 0.95-2.17) levels produced marginal effects without significance. Women with high LDL-C levels more often received a breast cancer diagnosis than those with medium levels (aHR, 95% CI: 1.56, 1.03-2.35).

CONCLUSION

This real-world dataset highlights the importance of balancing lipid profiles, particularly total cholesterol and HDL-C. Dysregulation of the lipid profile, not the glucose or blood pressure profiles, may increase the risk of breast cancer in women with T2DM.

Identifying patients at increased risk of hypoglycaemia in primary care: Development of a machine learning-based screening tool

INTRODUCTION

In primary care, identifying patients with type 2 diabetes (T2D) who are at increased risk of hypoglycaemia is important for the prevention of hypoglycaemic events. We aimed to develop a screening tool based on machine learning to identify such patients using routinely available demographic and medication data.

METHODS

We used a cohort study design and the Groningen Initiative to ANalyse Type 2 diabetes Treatment (GIANTT) medical record database to develop models for hypoglycaemia risk. The first hypoglycaemic event in the observation period (2007-2013) was the outcome. Demographic and medication data were used as predictor variables to train machine learning models. The performance of the models was compared with a model using additional clinical data using fivefold cross validation with the area under the receiver operator characteristic curve (AUC) as a metric.

RESULTS

We included 13,876 T2D patients. The best performing model including only demographic and medication data was logistic regression with least absolute shrinkage and selection operator, with an AUC of 0.71. Ten variables were included (odds ratio): male gender (0.997), age (0.990), total drug count (1.012), glucose-lowering drug count (1.039), sulfonylurea use (1.62), insulin use (1.769), pre-mixed insulin use (1.109), insulin count (1.827), insulin duration (1.193), and antidepressant use (1.05). The proposed model obtained a similar performance to the model using additional clinical data.

CONCLUSION

Using demographic and medication data, a model for identifying patients at increased risk of hypoglycaemia was developed using machine learning. This model can be used as a tool in primary care to screen for patients with T2D who may need additional attention to prevent or reduce hypoglycaemic events.

The Identification of Diabetes Mellitus Subtypes Applying Cluster Analysis Techniques: A Systematic Review

Diabetes Mellitus is a chronic and lifelong disease that incurs a huge burden to healthcare systems. Its prevalence is on the rise worldwide. Diabetes is more complex than the classification of Type 1 and 2 may suggest. The purpose of this systematic review was to identify the research studies that tried to find new sub-groups of diabetes patients by using unsupervised learning methods. The search was conducted on Pubmed and Medline databases by two independent researchers. All time publications on cluster analysis of diabetes patients were selected and analysed. Among fourteen studies that were included in the final review, five studies found five identical clusters: Severe Autoimmune Diabetes; Severe Insulin-Deficient Diabetes; Severe Insulin-Resistant Diabetes; Mild Obesity-Related Diabetes; and Mild Age-Related Diabetes. In addition, two studies found the same clusters, except Severe Autoimmune Diabetes cluster. Results of other studies differed from one to another and were less consistent. Cluster analysis enabled finding non-classic heterogeneity in diabetes, but there is still a necessity to explore and validate the capabilities of cluster analysis in more diverse and wider populations.

Predicting short- and long-term glycated haemoglobin response after insulin initiation in patients with type 2 diabetes mellitus using machine-learning algorithms

AIM

To assess the potential of supervised machine-learning techniques to identify clinical variables for predicting short-term and long-term glycated haemoglobin (HbA1c) response after insulin treatment initiation in patients with type 2 diabetes mellitus (T2DM).

MATERIALS AND METHODS

We included patients with T2DM from the Groningen Initiative to Analyse Type 2 diabetes Treatment (GIANTT) database who started insulin treatment between 2007 and 2013 and had a minimum follow-up of 2 years. Short- and long-term responses at 6 (±2) and 24 (±2) months after insulin initiation, respectively, were assessed. Patients were defined as good responders if they had a decrease in HbA1c ≥ 5 mmol/mol or reached the recommended level of HbA1c ≤ 53 mmol/mol. Twenty-four baseline clinical variables were used for the analysis and an elastic net regularization technique was used for variable selection. The performance of three traditional machine-learning algorithms was compared for the prediction of short- and long-term responses and the area under the receiver-operating characteristic curve (AUC) was used to assess the performance of the prediction models.

RESULTS

The elastic net regularization-based generalized linear model, which included baseline HbA1c and estimated glomerular filtration rate, correctly classified short- and long-term HbA1c response after treatment initiation, with AUCs of 0.80 (95% CI 0.78-0.83) and 0.81 (95% CI 0.79-0.84), respectively, and outperformed the other machine-learning algorithms. Using baseline HbA1c alone, an AUC = 0.71 (95% CI 0.65-0.73) and 0.72 (95% CI 0.66-0.75) was obtained for predicting short-term and long-term response, respectively.

CONCLUSIONS

Machine-learning algorithm performed well in the prediction of an individual's short-term and long-term HbA1c response using baseline clinical variables.

No significant association of type 2 diabetes-related genetic risk scores with glycated haemoglobin levels after initiating metformin or sulphonylurea derivatives

AIM

To explore the added value of diabetes-related genetic risk scores (GRSs) to readily available clinical variables in the prediction of glycated haemoglobin (HbA1c) levels after initiation of glucose-regulating drugs.

MATERIALS AND METHODS

We conducted a cohort study in people with type 2 diabetes (T2DM) from the Groningen Initiative to Analyse Type 2 Diabetes Treatment (GIANTT) database who initiated metformin (MET) or sulphonylurea derivatives (SUs) and for whom blood samples were genotyped. The primary outcome was HbA1c level at 6 months, adjusted for baseline HbA1c. GRSs were based on single nucleotide polymorphisms linked to insulin sensitivity, β-cell activity, and T2DM risk in general. Associations were analysed using multiple linear regression to assess whether adding the GRSs increased the explained variance in a prediction model that included age, gender, diabetes duration and cardio-metabolic biomarkers.

RESULTS

We included 282 patients initiating MET and 89 patients initiating SUs. In the MET prediction model, diabetes duration of >3 months when starting MET was associated with 2.7-mmol/mol higher HbA1c level. For SUs, no significant clinical predictors were identified. Addition of the GRS linked to insulin sensitivity (for MET), β-cell activity (for SUs) and T2DM risk (for both) to the models did not improve the explained variance significantly (22% without vs. 22% with GRS) for the MET and (14% without vs. 14% with GRS) for the SUs model, respectively.

CONCLUSION

This study did not indicate a significant effect of GRS related to T2DM in general or to the drugs' mechanism of action for prediction of inter-individual HbA1c variability in the short term after initiation of MET or SU therapy.

When drug treatments bias genetic studies: Mediation and interaction

Background

Increasingly, genetic analyses are conducted using information from subjects with established disease, who often receive concomitant treatment. We determined when treatment may bias genetic associations with a quantitative trait.

Methods

Graph theory and simulated data were used to explore the impact of drug prescriptions on (longitudinal) genetic effect estimates. Analytic derivations of longitudinal genetic effects are presented, accounting for the following scenarios: 1) treatment allocated independently of a genetic variant, 2) treatment that mediates the genetic effect, 3) treatment that modifies the genetic effect. We additionally evaluate treatment modelling strategies on bias, the root mean squared error (RMSE), coverage, and rejection rate.

Results

We show that in the absence of treatment by gene effect modification or mediation, genetic effect estimates will be unbiased. In simulated data we found that conditional models accounting for treatment, confounding, and effect modification were generally unbiased with appropriate levels of confidence interval coverage. Ignoring the longitudinal nature of treatment prescription, however (e.g. because of incomplete records in longitudinal data), biased these conditional models to a similar degree (or worse) as simply ignoring treatment.

Conclusion

The mere presence of (drug) treatment affecting a GWAS phenotype is insufficient to bias genetic associations with quantitative traits. While treatment may bias associations through effect modification and mediation, this might not occur frequently enough to warrant general concern at the presence of treated subjects in GWAS. Should treatment by gene effect modification or mediation be present however, current GWAS approaches attempting to adjust for treatment insufficiently account for the multivariable and longitudinal nature of treatment trajectories and hence genetic estimates may still be biased.

HbA1c response after insulin initiation in patients with type 2 diabetes mellitus in real life practice: Identifying distinct subgroups

AIMS

To identify subgroups of patients with type 2 diabetes mellitus (T2DM) following distinct trajectories of HbA1c after insulin initiation and explore underlying differences in clinical characteristics.

MATERIALS AND METHODS

A cohort study was conducted in patients with T2DM initiating insulin in 2007-2013 with a follow-up of 2 to 4 years. Data were collected from the Groningen Initiative to Analyze Type 2 Diabetes Treatment (GIANTT) database. The primary outcome was subgroups with different trajectories of HbA1c patterns after insulin initiation, as identified by latent class growth modeling. Differences between subgroups were tested using one-way ANOVA, Kruskal-Wallis or chi-square tests, where appropriate.

RESULTS

From 1459 patients, three subgroups with distinct HbA1c patterns were identified. Group 1 (8%) initially showed a moderate decrease followed by an increase in HbA1c 2 years later, despite receiving more comedication. Group 2 (84%) showed a stable decrease. Group 3 (8%) had a high initial level of HbA1c and a rapid decline within the first year, followed by a slow increase thereafter. Group 1 patients were on average 6-7 years younger than patients in groups 2 and 3 and were more likely to receive sulfonylureas than Group 3 patients. Group 3 patients had a shorter diabetes duration and were less well-controlled for HbA1c, systolic blood pressure and LDL-cholesterol at insulin initiation.

CONCLUSIONS

Most patients showed a stable HbA1c response, but one out of six patients showed either a poor response, or a rapid initial response only after insulin initiation. Response patterns were associated with age, diabetes duration and risk-factor controls at the time of insulin initiation.

Research Electronic Data Capture (REDCap®) used as an audit tool with a built-in database

The aim of this study was to develop an audit tool with a built-in database using Research Electronic Data Capture (REDCap®) as part of an antimicrobial stewardship program at a regional hospital in the Central Denmark Region, and to analyse the need, if any, to involve more than one expert in the evaluation of cases of antimicrobial treatment, and the level of agreement among the experts. Patients treated with systemic antimicrobials in the period from 1 September 2015 to 31 August 2016 were included, in total 722 cases. Data were collected retrospectively and entered manually. The audit was based on seven flow charts regarding: (1) initiation of antimicrobial treatment (2) infection (3) prescription and administration of antimicrobials (4) discontinuation of antimicrobials (5) reassessment within 48 h after the first prescription of antimicrobials (6) microbiological sampling in the period between suspicion of infection and the first administration of antimicrobials (7) microbiological results. The audit was based on automatic calculations drawing on the entered data and on expert assessments. Initially, two experts completed the audit, and in the cases in which they disagreed, a third expert was consulted. In 31.9% of the cases, the two experts agreed on all elements of the audit. In 66.2%, the two experts reached agreement by discussing the cases. Finally, 1.9% of the cases were completed in cooperation with a third expert. The experts assessed 3406 flow charts of which they agreed on 75.8%. We succeeded in creating an audit tool with a built-in database that facilitates independent expert evaluation using REDCap. We found a large inter-observer difference that needs to be considered when constructing a project based on expert judgements. Our two experts agreed on most of the flow charts after discussion, whereas the third expert's intervention did not have any influence on the overall assessment.

Identification of Four Novel Loci in Asthma in European American and African American Populations

RATIONALE

Despite significant advances in knowledge of the genetic architecture of asthma, specific contributors to the variability in the burden between populations remain uncovered.

OBJECTIVES

To identify additional genetic susceptibility factors of asthma in European American and African American populations.

METHODS

A phenotyping algorithm mining electronic medical records was developed and validated to recruit cases with asthma and control subjects from the Electronic Medical Records and Genomics network. Genome-wide association analyses were performed in pediatric and adult asthma cases and control subjects with European American and African American ancestry followed by metaanalysis. Nominally significant results were reanalyzed conditioning on allergy status.

MEASUREMENTS AND MAIN RESULTS

The validation of the algorithm yielded an average of 95.8% positive predictive values for both cases and control subjects. The algorithm accrued 21,644 subjects (65.83% European American and 34.17% African American). We identified four novel population-specific associations with asthma after metaanalyses: loci 6p21.31, 9p21.2, and 10q21.3 in the European American population, and the PTGES gene in African Americans. TEK at 9p21.2, which encodes TIE2, has been shown to be involved in remodeling the airway wall in asthma, and the association remained significant after conditioning by allergy. PTGES, which encodes the prostaglandin E synthase, has also been linked to asthma, where deficient prostaglandin E₂ synthesis has been associated with airway remodeling.

CONCLUSIONS

This study adds to understanding of the genetic architecture of asthma in European Americans and African Americans and reinforces the need to study populations of diverse ethnic backgrounds to identify shared and unique genetic predictors of asthma.

Natural language processing systems for capturing and standardizing unstructured clinical information: A systematic review

We followed a systematic approach based on the Preferred Reporting Items for Systematic Reviews and Meta-Analyses to identify existing clinical natural language processing (NLP) systems that generate structured information from unstructured free text. Seven literature databases were searched with a query combining the concepts of natural language processing and structured data capture. Two reviewers screened all records for relevance during two screening phases, and information about clinical NLP systems was collected from the final set of papers. A total of 7149 records (after removing duplicates) were retrieved and screened, and 86 were determined to fit the review criteria. These papers contained information about 71 different clinical NLP systems, which were then analyzed. The NLP systems address a wide variety of important clinical and research tasks. Certain tasks are well addressed by the existing systems, while others remain as open challenges that only a small number of systems attempt, such as extraction of temporal information or normalization of concepts to standard terminologies. This review has identified many NLP systems capable of processing clinical free text and generating structured output, and the information collected and evaluated here will be important for prioritizing development of new approaches for clinical NLP.

Development and validation of prescribing quality indicators for patients with type 2 diabetes

AIM

Quality indicators are used to measure whether healthcare professionals act according to guidelines, but few indicators focus on the quality of pharmacotherapy for diabetes. The aim of this study was to develop and validate a set of prescribing quality indicators (PQIs) for type 2 diabetes in primary care, and to apply this set in practice. To take into account the stepwise treatment of chronic disease, clinical action indicators were specifically considered.

METHODS

Potential PQIs were derived from clinical practice guidelines and evaluated using the RAND/UCLA Appropriateness Method, a modified Delphi panel. Thereafter, the feasibility of calculating the PQIs was tested in two large Dutch primary care databases including >80 000 diabetes patients in 2012.

RESULTS

32 PQIs focusing on treatment with glucose, lipid, blood pressure and albuminuria lowering drugs, and on vaccination, medication safety and adherence were assessed by ten experts. After the Delphi panel, the final list of twenty PQIs was tested for feasibility. All PQIs definitions were feasible for measuring the quality of medication treatment using these databases. Indicator scores ranged from 18.8% to 90.8% for PQIs focusing on current medication use, clinical action and medication choice, and from 2.1% to 37.2% for PQIs focusing on medication safety.

DISCUSSION AND CONCLUSIONS

Twenty PQIs focusing on treatment with glucose, lipid, blood pressure and albuminuria lowering drugs, and on medication safety in type 2 diabetes were developed, considered valid and operationally feasible. Results showed room for improvement, especially in initiation and intensification of treatment as measured with clinical action indicators.

Predictors of HbA1c levels in patients initiating metformin

OBJECTIVE

The aim was to assess demographic and clinical factors as predictors of short (6 months) and long term (18 months) HbA1c levels in diabetes patients initiating metformin treatment.

RESEARCH DESIGN AND METHODS

We conducted a cohort study including type 2 diabetes patients who received their first metformin prescription between 2007 and 2013 in the Groningen Initiative to Analyze Type 2 Diabetes Treatment (GIANTT) database. The primary outcome was HbA1c level at follow-up adjusted for baseline HbA1c; the secondary outcome was failing to achieve the target HbA1c level of 53 mmol/mol. Associations were analyzed by linear and logistic regression. Multiple imputation was used for missing data. Additional analyses stratified by dose and adherence level were conducted.

RESULTS

The cohort included 6050 patients initiating metformin. Baseline HbA1c at target consistently predicted better HbA1c outcomes. Longer diabetes duration and lower total cholesterol level at baseline were predictors for higher HbA1c levels at 6 months. At 18 months, cholesterol level was not a predictor. Longer diabetes duration was also associated with not achieving the target HbA1c at follow-up. The association for longer diabetes duration was especially seen in patients starting on low dose treatment. No consistent associations were found for comorbidity and comedication.

CONCLUSIONS

Diabetes duration was a relevant predictor of HbA1c levels after 6 and 18 months of follow-up in patients initiating metformin treatment. Given the study design, no causal inference can be made. Our study suggests that prompt treatment intensification may be needed in patients who have a longer diabetes duration at treatment initiation.

Role of Patient and Practice Characteristics in Variance of Treatment Quality in Type 2 Diabetes between General Practices

Background

Accounting for justifiable variance is important for fair comparisons of treatment quality. The variance between general practices in treatment quality of type 2 diabetes (T2DM) patients may be attributed to the underlying patient population and practice characteristics. The objective of this study is to describe the between practice differences in treatment, and identify patient and practice level characteristics that may explain these differences.

Methods

The data of 24,607 T2DM patients from 183 general practices in the Netherlands were used. Treatment variance was assessed in a cross-sectional manner for: glucose-lowering drugs/metformin, lipid-lowering drugs/statins, blood pressure-lowering drugs/ACE-inhibitor or ARB. Patient characteristics tested were age, gender, diabetes duration, comorbidity, comedication. Practice characteristics were number of T2DM patients, practice type, diabetes assistant available. Multilevel logistic regression was used to examine the between practice variance in treatment and the effect of characteristics on this variance.

Results

Treatment rates varied considerably between practices (IQR 9.5–13.9). The variance at practice level was 7.5% for glucose-lowering drugs, 3.6% for metformin, 3.1% for lipid-lowering drugs, 10.3% for statins, 8.6% for blood pressure-lowering drugs, and 3.9% for ACE-inhibitor/ARB. Patient and practice characteristics explained 19.0%, 7.5%, 20%, 6%, 9.9%, and 13.4% of the variance respectively. Age, multiple chronic drugs, and ≥3 glucose-lowering drugs were the most relevant patient characteristics. Number of T2DM patients per practice was the most relevant practice characteristic.

Discussion

Considerable differences exist between practices in treatment rates. Patients’ age was identified as characteristic that may account for justifiable differences in especially lipid-lowering treatment. Other patient or practice characteristics either do not explain or do not justify the differences.

The completeness of electronic medical record data for patients with Type 2 Diabetes in primary care and its implications for computer modelling of predicted clinical outcomes

AIM

To describe the completeness of routinely collected primary care data that could be used by computer models to predict clinical outcomes among patients with Type 2 Diabetes (T2D).

METHODS

Data on blood pressure, weight, total cholesterol, HDL-cholesterol and glycated haemoglobin levels for regular patients were electronically extracted from the medical record software of 12 primary care practices in Australia for the period 2000-2012. The data was analysed for temporal trends and for associations between patient characteristics and completeness. General practitioners were surveyed to identify barriers to recording data and strategies to improve its completeness.

RESULTS

Over the study period data completeness improved up to around 80% complete although the recording of weight remained poorer at 55%. T2D patients with Ischaemic Heart Disease were more likely to have their blood pressure recorded (OR 1.6, p=0.02). Practitioners reported not experiencing any major barriers to using their computer medical record system but did agree with some suggested strategies to improve record completeness.

CONCLUSION

The completeness of routinely collected data suitable for input into computerised predictive models is improving although other dimensions of data quality need to be addressed.

Overcoming the Challenges of Unstructured Data in Multisite, Electronic Medical Record-based Abstraction

BACKGROUND

Unstructured data encountered during retrospective electronic medical record (EMR) abstraction has routinely been identified as challenging to reliably abstract, as these data are often recorded as free text, without limitations to format or structure. There is increased interest in reliably abstracting this type of data given its prominent role in care coordination and communication, yet limited methodological guidance exists.

OBJECTIVES

As standard abstraction approaches resulted in substandard data reliability for unstructured data elements collected as part of a multisite, retrospective EMR study of hospital discharge communication quality, our goal was to develop, apply and examine the utility of a phase-based approach to reliably abstract unstructured data. This approach is examined using the specific example of discharge communication for warfarin management.

RESEARCH DESIGN

We adopted a "fit-for-use" framework to guide the development and evaluation of abstraction methods using a 4-step, phase-based approach including (1) team building; (2) identification of challenges; (3) adaptation of abstraction methods; and (4) systematic data quality monitoring.

MEASURES

Unstructured data elements were the focus of this study, including elements communicating steps in warfarin management (eg, warfarin initiation) and medical follow-up (eg, timeframe for follow-up).

RESULTS

After implementation of the phase-based approach, interrater reliability for all unstructured data elements demonstrated κ's of ≥0.89-an average increase of +0.25 for each unstructured data element.

CONCLUSIONS

As compared with standard abstraction methodologies, this phase-based approach was more time intensive, but did markedly increase abstraction reliability for unstructured data elements within multisite EMR documentation.

Using primary care electronic health record data for comparative effectiveness research: experience of data quality assessment and preprocessing in The Netherlands

AIM

Details of data quality and how quality issues were solved have not been reported in published comparative effectiveness studies using electronic health record data.

METHODS

We developed a conceptual framework of data quality assessment and preprocessing and apply it to a study comparing angiotensin-converting enzyme inhibitors with angiotensin receptor blockerss on renal function decline in diabetes patients.

RESULTS

The framework establishes a line of thought to identify and act on data issues. The core concept is to evaluate whether data are fit-for-use for research tasks. Possible quality problems are listed through specific signal detections, and verified whether they are true problems. Optimal solutions are selected for the identified problems.

CONCLUSION

This framework can be used in observational studies to improve validity of results.

Prescribing patterns, adherence and LDL-cholesterol response of type 2 diabetes patients initiating statin on low-dose versus standard-dose treatment: a descriptive study

AIMS

The aim of this study was to describe and compare treatment modifications and discontinuation, adherence levels and response to treatment in patients with type 2 diabetes initiating on low-dose vs. standard-dose statin treatment.

METHODS

A 2-year follow-up cohort study was performed using data from the Groningen Initiative to Analyse Type 2 Diabetes Treatment (GIANTT) database in patients with type 2 diabetes initiating statin treatment between January 2007 and December 2012. First, we determined whether there were differences in treatment modifications and discontinuation after statin initiation between patients starting on a low-dose vs. standard-dose. Second, we looked at differences in adherence and LDL-cholesterol response after 2 years follow-up between these groups.

RESULTS

Around 22% of patients initiated statin treatment on a dose lower than recommended. More than half of them remained on a low dose during a 2-year follow-up period, whereas less than 15% received a dose increase. Of the patients initiating on standard-dose, also more than half remained on the same treatment during this period, whereas 8% received a dose decrease without subsequent increase. Over 25% of patients starting on low-dose or standard-dose treatment discontinued treatment, often within the first 180 days after initiation or after a first treatment change. Patients on low-dose treatment had lower adherence levels and were less likely to have adequate LDL-cholesterol response compared with patients on standard-dose after 2 years follow-up.

CONCLUSIONS

Current patterns of statin treatment in patients with type 2 diabetes are suboptimal, with discontinuation, inadequate adherence levels and lack of treatment intensification seen in those who had inadequate LDL-cholesterol response after 2 years of follow-up. Patients starting on low-dose had more treatment modifications, discontinuation and adherence problems as compared with those starting on standard-dose treatment, which calls for a closer look at the rationale of starting patients on low-dose statin treatment.

Comparing the effect of angiotensin-converting enzyme inhibitors and angiotensin receptor blockers on renal function decline in diabetes

AIM

To compare effectiveness of angiotensin-converting enzyme inhibitors (ACEis)/angiotensin receptor blockers (ARBs) for protecting Type 2 diabetes mellitus (DM2) patients from renal function decline in a real-world setting.

METHODS

Retrospective cohort study of new ACEi/ARB users in 2007-2012 in an unselected primary care DM2 population. Outcome is decline in renal function stage (combining estimated glomerular filtration rate and albuminuria). Patients were matched on a propensity score. Extended Cox models with time-varying covariates were used to estimate hazard ratios of outcome.

RESULTS

The time to renal function decline for ARB users was slightly, but not significantly longer than for ACEi users (hazard ratio: 0.80; 95% CI: 0.58-1.10; p = 0.166).

CONCLUSION

This study did not show significant differences between the classes in preventing renal function decline in DM2 patients in primary care.

Identification of major cardiovascular events in patients with diabetes using primary care data

Background

Routine primary care data are increasingly being used for evaluation and research purposes but there are concerns about the completeness and accuracy of diagnoses and events captured in such databases. We evaluated how well patients with major cardiovascular disease (CVD) can be identified using primary care morbidity data and drug prescriptions.

Methods

The study was conducted using data from 17,230 diabetes patients of the GIANTT database and Dutch Hospital Data register. To estimate the accuracy of the different measures, we analyzed the sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) relative to hospitalizations and/or records with a diagnosis indicating major CVD, including ischaemic heart diseases and cerebrovascular events.

Results

Using primary care morbidity data, 43 % of major CVD hospitalizations could be identified. Adding drug prescriptions to the search increased the sensitivity up to 94 %. A proxy of at least one prescription of either a platelet aggregation inhibitor, vitamin k antagonist or nitrate could identify 85 % of patients with a history of major CVD recorded in primary care, with an NPV of 97 %. Using the same proxy, 57 % of incident major CVD recorded in primary or hospital care could be identified, with an NPV of 99 %.

Conclusions

A substantial proportion of major CVD hospitalizations was not recorded in primary care morbidity data. Drug prescriptions can be used in addition to diagnosis codes to identify more patients with major CVD, and also to identify patients without a history of major CVD.

Electronic supplementary material

The online version of this article (doi:10.1186/s12913-016-1361-2) contains supplementary material, which is available to authorized users.

Development and initial validation of prescribing quality indicators for patients with chronic kidney disease

BACKGROUND

Quality assessment is a key element for improving the quality of care. Currently, a comprehensive indicator set for measuring the quality of medication treatment in patients with chronic kidney disease (CKD) is lacking. Our aim was to develop and validate a set of prescribing quality indicators (PQIs) for CKD care, and to test the feasibility of applying this set in practice.

METHODS

Potential indicators were based on clinical practice guidelines and evaluated using the RAND/UCLA Appropriateness Method. This is a structured process in which an expert panel assesses the validity of the indicators. Feasibility was tested in a Dutch primary care database including >4500 diabetes patients with CKD.

RESULTS

An initial list of 22 PQIs was assessed by 12 experts. After changing 10 PQIs, adding 2 and rejecting 8, a final list of 16 indicators was accepted by the expert panel as valid. These PQIs focused on the treatment of hypertension, albuminuria, mineral and bone disorder, statin prescribing and possible unsafe medication. The indicators were successfully applied to measure treatment quality in the primary care database, but for some indicators the number of eligible patients was too small for reliable calculation. Results showed that there was room for improvement in the treatment quality of this population.

CONCLUSIONS

We developed a set of 16 PQIs for measuring the quality of treatment in CKD patients, which had sufficient content and face validity as well as operational feasibility. These PQIs can be used to point out priority areas for improvement.

Adherence to standard-dose or low-dose statin treatment and low-density lipoprotein cholesterol response in type 2 diabetes patients

OBJECTIVE

To determine the association between adherence, dose and low-density lipoprotein (LDL) cholesterol response in patients with type 2 diabetes initiating statin treatment.

RESEARCH DESIGN AND METHODS

This cohort study was performed using data for 2007-2012 from the Groningen Initiative to Analyse Type 2 Diabetes Treatment (GIANTT) database. The association between adherence to a standard-dose statin and LDL cholesterol response was assessed using linear regression, adjusting for covariates. The effect of low-dose versus standard-dose was assessed in a propensity-score matched cohort. Adherence rates, defined as the proportion of days covered (PDC), were estimated between statin initiation and LDL outcome measurement.

MAIN OUTCOME MEASURE

LDL cholesterol level at follow-up.

RESULTS

The effect of adherence on LDL cholesterol response, measured in 2160 patients, was dependent on the baseline LDL cholesterol level. For patients with a baseline LDL cholesterol of 3.7 mmol/l and an adherence rate of 80%, a 40% reduction in LDL cholesterol was predicted. In the matched sample of 1144 patients, the treatment dose showed a difference in impact on the outcome for adherence rates higher than 50%. It was estimated that a patient with a baseline LDL cholesterol of 3.7 mmol/l will need an adherence rate of at least 76% on low-dose and 63% on standard-dose treatment to reach the LDL cholesterol target of 2.5 mmol/l.

LIMITATIONS

Adherence was measured as the PDC, which is known to overestimate actual adherence. Also, we were not able to adjust for lifestyle factors.

CONCLUSIONS

We determined the concurrent effect of treatment adherence and dose on LDL cholesterol outcomes. Given the adherence levels seen in clinical practice, diabetes patients initiating statin treatment are at high risk of not reaching the recommended cholesterol target, especially when they start on a low-dose statin.

Data-Driven Information Extraction from Chinese Electronic Medical Records

OBJECTIVE

This study aims to propose a data-driven framework that takes unstructured free text narratives in Chinese Electronic Medical Records (EMRs) as input and converts them into structured time-event-description triples, where the description is either an elaboration or an outcome of the medical event.

MATERIALS AND METHODS

Our framework uses a hybrid approach. It consists of constructing cross-domain core medical lexica, an unsupervised, iterative algorithm to accrue more accurate terms into the lexica, rules to address Chinese writing conventions and temporal descriptors, and a Support Vector Machine (SVM) algorithm that innovatively utilizes Normalized Google Distance (NGD) to estimate the correlation between medical events and their descriptions.

RESULTS

The effectiveness of the framework was demonstrated with a dataset of 24,817 de-identified Chinese EMRs. The cross-domain medical lexica were capable of recognizing terms with an F1-score of 0.896. 98.5% of recorded medical events were linked to temporal descriptors. The NGD SVM description-event matching achieved an F1-score of 0.874. The end-to-end time-event-description extraction of our framework achieved an F1-score of 0.846.

DISCUSSION

In terms of named entity recognition, the proposed framework outperforms state-of-the-art supervised learning algorithms (F1-score: 0.896 vs. 0.886). In event-description association, the NGD SVM is superior to SVM using only local context and semantic features (F1-score: 0.874 vs. 0.838).

CONCLUSIONS

The framework is data-driven, weakly supervised, and robust against the variations and noises that tend to occur in a large corpus. It addresses Chinese medical writing conventions and variations in writing styles through patterns used for discovering new terms and rules for updating the lexica.

Presentation of respiratory symptoms prior to diagnosis in general practice: a case-control study examining free text and morbidity codes

OBJECTIVE

General practitioners can record patients' presenting symptoms by using a code or free text. We compared breathlessness and wheeze symptom codes and free text recorded prior to diagnosis of ischaemic heart disease (IHD), chronic obstructive pulmonary disease (COPD) and asthma.

DESIGN

A case-control study.

SETTING

11 general practices in North Staffordshire, UK, contributing to the Consultations in Primary Care Archive consultation database.

PARTICIPANTS

Cases with an incident diagnosis of IHD, COPD or asthma in 2010 were matched to controls (four per case) with no such diagnosis. All prior consultations with codes for breathlessness or wheeze symptoms between 2004 and 2010 were identified. Free text of cases and controls were also searched for mention of these symptoms.

RESULTS

592 cases were identified, 194 (33%) with IHD, 182 (31%) with COPD and 216 (37%) with asthma. 148 (25%) cases and 125 (5%) controls had a prior coded consultation for breathlessness. Prevalence of a prior coded symptom of breathlessness or wheeze was 30% in cases, 6% in controls. Median time from first coded symptom to diagnosis among cases was 57 weeks. After adding symptoms recorded in text, prevalence rose to 62% in cases and 25% in controls. Median time from first recorded symptom increased to 144 weeks. The associations between diagnosis of cases and prior symptom codes was strong IHD relative risk ratio (RRR) 3.21 (2.15 to 4.79); COPD RRR 9.56 (6.74 to 13.60); asthma RRR 10.30 (7.17 to 14.90).

CONCLUSIONS

There is an association between IHD, COPD and asthma diagnosis and earlier consultation for respiratory symptoms. Symptoms are often noted in free text by GPs long before they are coded. Free text searching may aid investigation of early presentation of long-term conditions using GP databases, and may be an important direction for future research.

Cost-effectiveness of statins for primary prevention in patients newly diagnosed with type 2 diabetes in the Netherlands

BACKGROUND

Statins are lipid-lowering drugs that reduce the risk of cardiovascular events in patients with diabetes.

OBJECTIVES

The objective of this study was to determine whether statin treatment for primary prevention in newly diagnosed type 2 diabetes is cost-effective, taking nonadherence, baseline risk, and age into account.

METHODS

A cost-effectiveness analysis was performed by using a Markov model with a time horizon of 10 years. The baseline 10-year cardiovascular risk was estimated in a Dutch population of primary prevention patients with newly diagnosed diabetes from the Groningen Initiative to Analyse Type 2 Diabetes Treatment (GIANTT) database, using the United Kingdom Prospective Diabetes Study risk engine. Statin adherence was measured as pill days covered in the IADB.nl pharmacy research database. Cost-effectiveness was measured in costs per quality-adjusted life-year (QALY) from the health care payers' perspective.

RESULTS

For an average patient aged 60 years, the base case, statin treatment was highly cost-effective at €2245 per QALY. Favorable cost-effectiveness was robust in sensitivity analysis. Differences in age and 10-year cardiovascular risk showed large differences in cost-effectiveness from almost €100,000 per QALY to almost being cost saving. Treating all patients younger than 45 years at diabetes diagnosis was not cost-effective (weighted cost-effectiveness of almost €60,000 per QALY).

CONCLUSIONS

Despite the nonadherence levels observed in actual practice, statin treatment is cost-effective for primary prevention in patients newly diagnosed with type 2 diabetes. Because of large differences in cost-effectiveness according to different risk and age groups, the efficiency of the treatment could be increased by targeting patients with relatively higher cardiovascular risk and higher ages.

Medication beliefs, treatment complexity, and non-adherence to different drug classes in patients with type 2 diabetes

OBJECTIVE

To assess the relationship of patients' medication beliefs and treatment complexity with unintentional and intentional non-adherence for three therapeutic groups commonly used by patients with type 2 diabetes.

METHODS

Survey data about adherence (Medication Adherence Report Scale) and beliefs about medicines (Beliefs about Medicines Questionnaire) were combined with prescription data from the Groningen Initiative to ANalyse Type 2 diabetes Treatment (GIANTT) database. Patients were classified as being adherent, mainly unintentional non-adherent, or partly intentional non-adherent per therapeutic group (glucose-, blood pressure-, and lipid-lowering drugs). Treatment complexity was measured using the Medication Regimen Complexity Index, which includes the dosage form, dosing frequency and additional directions of taking the drug. Analyses were performed using Kruskal-Wallis and Mann-Whitney U-tests.

RESULTS

Of 257 contacted patients, 133 (52%) returned the questionnaire. The patients had a mean age of 66years and 50% were females. Necessity beliefs were not significantly different between the adherers, mainly unintentional non-adherers, and partly intentional non-adherers (differences smaller than 5 points on a scale from 5 to 25). For blood pressure-lowering drugs, patients reporting intentional non-adherence had higher concern beliefs than adherers (8 point difference, P=0.01). Treatment complexity scores were lower for adherers but similar for mainly unintentional and partly intentional non-adherers to glucose- and blood pressure-lowering drugs.

CONCLUSION

Treatment complexity was related to non-adherence in general. Beliefs about necessity were not strongly associated with non-adherence, while patients' concern beliefs may be associated with intentional non-adherence. However, the role of these determinants differs per therapeutic group.

Potential overtreatment and undertreatment of diabetes in different patient age groups in primary care after the introduction of performance measures

OBJECTIVE

To assess whether after the introduction of diabetes performance measures decreases in undertreatment correspond with increases in overtreatment for blood pressure (BP) and glycemic control in different patient age groups.

RESEARCH DESIGN AND METHODS

We conducted a cohort study using data from the Groningen Initiative to Analyse Type 2 Diabetes Treatment (GIANTT) database. General practices were included when data were available from 1 year before to at least 1 year after the introduction of diabetes performance measures. Included patients had a confirmed diagnosis of type 2 diabetes. Potential overtreatment was defined as prescribing maximum treatment or a treatment intensification to patients with a sustained low-risk factor level. Potential undertreatment was defined as a lack of treatment intensification in patients with a sustained high-risk factor level. Percentages of over- and undertreated patients at baseline were compared with those in subsequent years, and stratified analyses were performed for different patient age groups.

RESULTS

For BP, undertreatment significantly decreased from 61 to 57% in the first year after the introduction of performance measures. In patients >75 years of age, undertreatment decreased from 65 to ∼61%. Overtreatment was relatively stable (∼16%). For glycemic control, undertreatment significantly increased from 49 to 53%, and overtreatment remained relatively stable (∼7%).

CONCLUSIONS

The improvement of BP undertreatment after introduction of the performance measures did not correspond with an increase in overtreatment. The performance measures appeared to have little impact on improving glucose-regulating treatment. The trends did not differ among patient age groups.

Practical challenges in integrating genomic data into the electronic health record

Genetic testing has had limited impact on routine clinical care. Widespread adoption of electronic health records presents a promising means of disseminating genetic testing into diverse care settings. Practical challenges to integration of genomic data into electronic health records include size and complexity of genetic test results, inadequate use of standards for clinical and genetic data, and limitations in electronic health record capacity to store and analyze genetic data. Related challenges include uncertainty in the interpretation of regulatory requirements for return of results, and privacy concerns specific to genetic testing. Successful integration of genomic data may require significant redesign of existing electronic health record systems.

Optimising the use of electronic health records to estimate the incidence of rheumatoid arthritis in primary care: what information is hidden in free text?

Background

Primary care databases are a major source of data for epidemiological and health services research. However, most studies are based on coded information, ignoring information stored in free text. Using the early presentation of rheumatoid arthritis (RA) as an exemplar, our objective was to estimate the extent of data hidden within free text, using a keyword search.

Methods

We examined the electronic health records (EHRs) of 6,387 patients from the UK, aged 30 years and older, with a first coded diagnosis of RA between 2005 and 2008. We listed indicators for RA which were present in coded format and ran keyword searches for similar information held in free text. The frequency of indicator code groups and keywords from one year before to 14 days after RA diagnosis were compared, and temporal relationships examined.

Results

One or more keyword for RA was found in the free text in 29% of patients prior to the RA diagnostic code. Keywords for inflammatory arthritis diagnoses were present for 14% of patients whereas only 11% had a diagnostic code. Codes for synovitis were found in 3% of patients, but keywords were identified in an additional 17%. In 13% of patients there was evidence of a positive rheumatoid factor test in text only, uncoded. No gender differences were found. Keywords generally occurred close in time to the coded diagnosis of rheumatoid arthritis. They were often found under codes indicating letters and communications.

Conclusions

Potential cases may be missed or wrongly dated when coded data alone are used to identify patients with RA, as diagnostic suspicions are frequently confined to text. The use of EHRs to create disease registers or assess quality of care will be misleading if free text information is not taken into account. Methods to facilitate the automated processing of text need to be developed and implemented.

Advancing current approaches to disease management evaluation: capitalizing on heterogeneity to understand what works and for whom

BACKGROUND

Evaluating large-scale disease management interventions implemented in actual health care settings is a complex undertaking for which universally accepted methods do not exist. Fundamental issues, such as a lack of control patients and limited generalizability, hamper the use of the 'gold-standard' randomized controlled trial, while methodological shortcomings restrict the value of observational designs. Advancing methods for disease management evaluation in practice is pivotal to learn more about the impact of population-wide approaches. Methods must account for the presence of heterogeneity in effects, which necessitates a more granular assessment of outcomes.

METHODS

This paper introduces multilevel regression methods as valuable techniques to evaluate 'real-world' disease management approaches in a manner that produces meaningful findings for everyday practice. In a worked example, these methods are applied to retrospectively gathered routine health care data covering a cohort of 105,056 diabetes patients who receive disease management for type 2 diabetes mellitus in the Netherlands. Multivariable, multilevel regression models are fitted to identify trends in clinical outcomes and correct for differences in characteristics of patients (age, disease duration, health status, diabetes complications, smoking status) and the intervention (measurement frequency and range, length of follow-up).

RESULTS

After a median one year follow-up, the Dutch disease management approach was associated with small average improvements in systolic blood pressure and low-density lipoprotein, while a slight deterioration occurred in glycated hemoglobin. Differential findings suggest that patients with poorly controlled diabetes tend to benefit most from disease management in terms of improved clinical measures. Additionally, a greater measurement frequency was associated with better outcomes, while longer length of follow-up was accompanied by less positive results.

CONCLUSIONS

Despite concerted efforts to adjust for potential sources of confounding and bias, there ultimately are limits to the validity and reliability of findings from uncontrolled research based on routine intervention data. While our findings are supported by previous randomized research in other settings, the trends in outcome measures presented here may have alternative explanations. Further practice-based research, perhaps using historical data to retrospectively construct a control group, is necessary to confirm results and learn more about the impact of population-wide disease management.

Association between performance measures and glycemic control among patients with diabetes in a community-wide primary care cohort

BACKGROUND

Performance measures are used for assessing quality of care. Higher performance shown by these measures is expected to reflect better care, but little is known whether they predict better patient outcomes.

OBJECTIVE

To assess the predictive value of performance measures of glucose management on glycemic control, and evaluate the impact of patient characteristics on this association.

RESEARCH DESIGN

Cohort study (2007-2009).

SUBJECTS

A total of 15,454 type 2 diabetes patients (mean age, 66.5 y; 48% male) from the GIANTT cohort.

MEASURES

We included performance measures assessing frequency of HbA1c monitoring, glucose-lowering treatment status, and treatment intensification. Associations between performance and glycemic control were tested using multivariate linear regression adjusted for confounding, reporting estimated differences in HbA1c with 95% confidence intervals (CI). Impact of patient characteristics was examined through interactions.

RESULTS

Annual HbA1c monitoring was associated with better glycemic control when compared with no such monitoring (HbA1c -0.29%; 95% CI -0.37, -0.22). This association lost significance in patients with lower baseline HbA1c, older age, and without macrovascular comorbidity. Treatment status was associated with better glycemic control only in patients with elevated baseline HbA1c. Treatment intensification after elevated HbA1c levels was associated with better glycemic control compared with no intensification (HbA1c -0.21; 95% CI -0.26, -0.16).

CONCLUSIONS

Performance measures of annual HbA1c monitoring and of treatment intensification did predict better patient outcomes, whereas the measure of treatment status did not. Predictive value of annual monitoring and of treatment status varied across patient characteristics, and it should be used with caution when patient characteristics cannot be taken into account.

Is albuminuria screening and treatment optimal in patients with type 2 diabetes in primary care? Observational data of the GIANTT cohort

BACKGROUND

Failure of diagnosing and treatment of albuminuria play a role in morbidity and mortality in type 2 diabetes (T2DM). We evaluated guideline adherence and factors associated with albuminuria screening and treatment in T2DM patients in primary care.

METHODS

Guidelines recommend annual measurement of albuminuria and, if increased, treatment with renin-angiotensin-aldosterone system (RAAS) blockers. We performed a cohort study of T2DM patients managed by 182 Dutch general practitioners (GPs; Groningen Initiative to Analyse Type 2 diabetes Treatment database), and evaluated guideline adherence in the years 2007-2009. We assessed whether demographic, clinical, organizational or provider factors determined guideline adherence with multilevel analyses.

RESULTS

Data were available for 14 120 T2DM patients [47.6% male, mean age 67.3 ± 11.7 years, median diabetes duration 6 (IQR: 3-10) years]. The albumin-creatinine ratio (ACR) was measured in 45.2% in 2007, 57.4% in 2008 and 56.8% in 2009. Only 23.7% of all patients were measured every year and 21.4% were never measured. The ACR was more often measured in patients <75 years, with a previous ACR measurement, using anti-diabetic medication, and receiving additional care by a diabetes support facility. RAAS treatment was prescribed to 78.4% of patients with prevalent micro/macroalbuminuria, 66.5% with incident micro/macroalbuminuria, 59.3% with normoalbuminuria and 52.1% of those without ACR measurements. In those not treated with RAAS blockers, it was initiated in 14.3, 12.3, 3.0 and 2.3%, respectively. The presence of micro/macroalbuminuria, higher blood pressure, incidence of cardiovascular events and treatment with antihypertensive medication were the determinants of RAAS-treatment initiation.

CONCLUSIONS

Guideline implementation regarding the management of albuminuria in T2DM patients in primary care should be further improved.

The effect of a patient-oriented treatment decision aid for risk factor management in patients with diabetes (PORTDA-diab): study protocol for a randomised controlled trial

Background

To improve risk factor management in diabetes, we need to support effective interactions between patients and healthcare providers. Our aim is to develop and evaluate a treatment decision aid that offers personalised information on treatment options and outcomes, and is intended to empower patients in taking a proactive role in their disease management. Important features are: (1) involving patients in setting goals together with their provider; (2) encourage them to prioritise on treatments that maximise relevant outcomes; and (3) integration of the decision aid in the practice setting and workflow. As secondary aim, we want to evaluate the impact of different presentation formats, and learn more from the experiences of the healthcare providers and patients with the decision aid.

Methods and design

We will conduct a randomised trial comparing four formats of the decision aid in a 2×2 factorial design with a control group. Patients with type 2 diabetes managed in 18 to 20 primary care practices in The Netherlands will be recruited. Excluded are patients with a recent myocardial infarction, stroke, heart failure, angina pectoris, terminal illness, cognitive deficits, >65 years at diagnosis, or not able to read Dutch. The decision aid is offered to the patients immediately before their quarterly practice consultation. The same decision information will be available to the healthcare provider for use during consultation. In addition, the providers receive a set of treatment cards, which they can use to discuss the benefits and risks of different options. Patients in the control group will receive care as usual. We will measure the effect of the intervention on patient empowerment, satisfaction with care, beliefs about medication, negative emotions, health status, prescribed medication, and predicted cardiovascular risk. Data will be collected with questionnaires and automated extraction from medical records in 6 months before and after the intervention.

Discussion

This decision aid is innovative in supporting patients and their healthcare providers to make shared decisions about multiple treatments, using the patient’s data from electronic medical records. The results can contribute to the further development and implementation of electronic decision support tools for the management of chronic diseases.

Trial registration

Dutch Trial register NTR1942.

Representativeness of diabetes patients participating in a web-based adverse drug reaction monitoring system

PURPOSE

Lareb Intensive Monitoring, LIM, is a non-interventional observational cohort method which follows first-time users of certain drugs during a certain period of time and collects information about adverse drug reactions, ADRs. In order for LIM to be a useful pharmacovigilance tool, it is important to know whether the LIM population is comparable to the whole population using the drug. The aim of this study is to compare the LIM diabetes population with an external diabetes reference population on characteristics that may influence the patient's susceptibility for ADRs.

METHODS

In this study, a LIM diabetes population was compared to a reference diabetes population derived from The Groningen Initiative to ANalyse Type 2 diabetes Treatment project. Comparisons were made regarding age, gender, body mass index and polypharmacy, as well as diabetes medication used and disease/treatment duration.

RESULTS

LIM patients were more often men (58.5% vs 50.8%) and in general younger (59.1 vs 64.7 years) and healthier, by that meaning they had a higher percentage of de novo treated patients (55.5% vs 53.2%), a shorter diabetes treatment duration (3.7 vs 5.5 years) and used less co-medication than patients in the reference population.

CONCLUSIONS

This study shows that diabetes patients participating in a web-based intensive monitoring system differ from a reference population. The observed differences might lead to an underestimation of ADRs, but it is not clear whether this would also influence the type or time-course of the reported ADRs. When interpreting results from LIM studies, one should take these differences into account.

Towards automatic diabetes case detection and ABCS protocol compliance assessment

OBJECTIVE

According to the American Diabetes Association, the implementation of the standards of care for diabetes has been suboptimal in most clinical settings. Diabetes is a disease that had a total estimated cost of $174 billion in 2007 for an estimated diabetes-affected population of 17.5 million in the United States. With the advent of electronic medical records (EMR), tools to analyze data residing in the EMR for healthcare surveillance can help reduce the burdens experienced today. This study was primarily designed to evaluate the efficacy of employing clinical natural language processing to analyze discharge summaries for evidence indicating a presence of diabetes, as well as to assess diabetes protocol compliance and high risk factors.

METHODS

Three sets of algorithms were developed to analyze discharge summaries for: (1) identification of diabetes, (2) protocol compliance, and (3) identification of high risk factors. The algorithms utilize a common natural language processing framework that extracts relevant discourse evidence from the medical text. Evidence utilized in one or more of the algorithms include assertion of the disease and associated findings in medical text, as well as numerical clinical measurements and prescribed medications.

RESULTS

The diabetes classifier was successful at classifying reports for the presence and absence of diabetes. Evaluated against 444 discharge summaries, the classifier's performance included macro and micro F-scores of 0.9698 and 0.9865, respectively. Furthermore, the protocol compliance and high risk factor classifiers showed promising results, with most F-measures exceeding 0.9.

CONCLUSIONS

The presented approach accurately identified diabetes in medical discharge summaries and showed promise with regards to assessment of protocol compliance and high risk factors. Utilizing free-text analytic techniques on medical text can complement clinical-public health decision support by identifying cases and high risk factors.

Differential effects of comorbidity on antihypertensive and glucose-regulating treatment in diabetes mellitus--a cohort study

Background

Comorbidity is often mentioned as interfering with “optimal” treatment decisions in diabetes care. It is suggested that diabetes-related comorbidity will increase adequate treatment, whereas diabetes-unrelated comorbidity may decrease this process of care. We hypothesized that these effects differ according to expected priority of the conditions.

Methods

We evaluated the relationship between comorbidity and treatment intensification in a study of 11,248 type 2 diabetes patients using the GIANTT (Groningen Initiative to Analyse type 2 diabetes Treatment) database. We formed a cohort of patients with a systolic blood pressure ≥140 mmHg (6,820 hypertensive diabetics), and a cohort of patients with an HbA1c ≥7% (3,589 hyperglycemic diabetics) in 2007. We differentiated comorbidity by diabetes-related or unrelated conditions and by priority. High priority conditions include conditions that are life-interfering, incident or requiring new medication treatment. We performed Cox regression analyses to assess association with treatment intensification, defined as dose increase, start, or addition of drugs.

Results

In both the hypertensive and hyperglycemic cohort, only patients with incident diabetes-related comorbidity had a higher chance of treatment intensification (HR 4.48, 2.33–8.62 (p<0.001) for hypertensives; HR 2.37, 1.09–5.17 (p = 0.030) for hyperglycemics). Intensification of hypertension treatment was less likely when a new glucose-regulating drug was prescribed (HR 0.24, 0.06–0.97 (p = 0.046)). None of the prevalent or unrelated comorbidity was significantly associated with treatment intensification.

Conclusions

Diabetes-related comorbidity induced better risk factor treatment only for incident cases, implying that appropriate care is provided more often when complications occur. Diabetes-unrelated comorbidity did not affect hypertension or hyperglycemia management, even when it was incident or life-interfering. Thus, the observed “undertreatment” in diabetes care cannot be explained by constraints caused by such comorbidity.

The age-phenome database

Data linking specific ages or age ranges with disease are abundant in biomedical literature. However, these data are organized such that searching for age-phenotype relationships is difficult. Recently, we described the Age-Phenome Knowledge-base (APK), a computational platform for storage and retrieval of information concerning age-related phenotypic patterns. Here, we report that data derived from over 1.5 million human-related PubMed abstracts have been added to APK. Using a text-mining pipeline, 35,683 entries which describe relationships between age and phenotype (such as disease) have been introduced into the database. Comparing the results to those obtained by a human reader reveals that the overall accuracy of these entries is estimated to exceed 80%. The usefulness of these data for obtaining new insight regarding age-disease relationships is demonstrated using clustering analysis, which is shown to capture obvious, as well as potentially interesting relationships between diseases. In addition, a new tool for browsing and searching the APK database is presented. We thus present a unique resource and a new framework for studying age-disease relationships and other phenotypic processes.

Part-of-speech tagging for clinical text: wall or bridge between institutions?

Part-of-speech (POS) tagging is a fundamental step required by various NLP systems. The training of a POS tagger relies on sufficient quality annotations. However, the annotation process is both knowledge-intensive and time-consuming in the clinical domain. A promising solution appears to be for institutions to share their annotation efforts, and yet there is little research on associated issues. We performed experiments to understand how POS tagging performance would be affected by using a pre-trained tagger versus raw training data across different institutions. We manually annotated a set of clinical notes at Kaiser Permanente Southern California (KPSC) and a set from the University of Pittsburg Medical Center (UPMC), and trained/tested POS taggers with intra- and inter-institution settings. The cTAKES POS tagger was also included in the comparison to represent a tagger partially trained from the notes of a third institution, Mayo Clinic at Rochester. Intra-institution 5-fold cross-validation estimated an accuracy of 0.953 and 0.945 on the KPSC and UPMC notes respectively. Trained purely on KPSC notes, the accuracy was 0.897 when tested on UPMC notes. Trained purely on UPMC notes, the accuracy was 0.904 when tested on KPSC notes. Applying the cTAKES tagger pre-trained with Mayo Clinic's notes, the accuracy was 0.881 on KPSC notes and 0.883 on UPMC notes. After adding UPMC annotations to KPSC training data, the average accuracy on tested KPSC notes increased to 0.965. After adding KPSC annotations to UPMC training data, the average accuracy on tested UPMC notes increased to 0.953. The results indicated: first, the performance of pre-trained POS taggers dropped about 5% when applied directly across the institutions; second, mixing annotations from another institution following the same guideline increased tagging accuracy for about 1%. Our findings suggest that institutions can benefit more from sharing raw annotations but less from sharing pre-trained models for the POS tagging task. We believe the study could also provide general insights on cross-institution data sharing for other types of NLP tasks.

Development of a minimal set of prescribing quality indicators for diabetes management on a general practice level

OBJECTIVE

To identify the relevant prescribing quality domains of type 2 diabetes mellitus care as a basis for the selection of a minimal set of prescribing quality indicators from a set of previously validated indicators.

METHODS

We used the principal factor analysis to identify the underlying dimensions or domains of prescribing quality for 76 general practitioners participating to the Groningen Initiative to Analyse Type 2 Diabetes Treatment project in the Netherlands. From a set of 10 prescribing quality indicators covering various aspects of cardiovascular and metabolic management, we selected a subset of indicators with the highest loading within each identified domain. Next, we evaluated the effect of using this subset on the quintile ranking of practices on their prescribing quality scores.

RESULTS

We identified five prescribing quality domains in our data set: two assessing initiation of pharmacotherapy for different risk factors in diabetic patients, two on stepwise intensification of treatment, and one on treatment of patients with cardiovascular disease. A composite score comprising the indicators selected from each of the domains showed good agreement with the composite score comprising all indicators with 82% of general practitioners either not changing their position or shifting their ranking by only one quintile.

CONCLUSIONS

We showed that a minimal set of prescribing quality indicators for type 2 diabetes mellitus care should not just focus on the management of different clinical risk factors but also reflect different steps of treatment intensification. The results of our study are relevant for stakeholders when selecting quality indicators to assess the quality of prescribing in diabetic patients.

A longitudinal study examining adherence to guidelines in diabetes care according to different definitions of adequacy and timeliness

BACKGROUND

Performance indicators assessing quality of diabetes care often look at single processes, e.g. whether an HbA1c test was conducted. Adequate care, however, consists of consecutive processes which should be taken in time (clinical pathways). We assessed quality of diabetes care by looking at single processes versus clinical pathways. In addition, we evaluated the impact of time period definitions on this quality assessment.

METHODOLOGY

We conducted a cohort study in 2007-2008 using the GIANTT (Groningen Initiative to Analyse type 2 diabetes Treatment) database. Proportions of patients adequately managed for HbA1c, systolic blood pressure (SBP), LDL-cholesterol (LDL-C), and albumin/creatinine ratio (ACR) were calculated for the pathway of (1) risk factor level testing, (2) treatment intensification when indicated, (3) response to treatment evaluation. Strict and wide time periods for each step were defined. Proportions of patients adequately managed regarding the overall pathway and single steps, using strict or wide time periods were compared using odds ratios (OR) with 95% confidence intervals.

FINDINGS

Of 11,176 patients diagnosed with type 2 diabetes, 9439 with complete follow-up were included. The majority received annual examination of HbA1c (86%) and SBP (86%), whereas this was 67% for LDL-C and 49% for ACR. Adequate management regarding the three-step pathway was observed in 73%, 53%, 46%, 41% of patients for HbA1c, SBP, LDL-C, and ACR respectively. Quality scores reduced significantly due to the second step (OR 0.43, 0.18, 0.44, 0.74), but were not much further reduced by the third step. Timely treatment evaluation occurred in 88% for HbA1c, 87% for SBP, 83% for LDL-C, and 76% for ACR. The overall score was not significantly changed by using strict time windows.

CONCLUSION

Quality estimates of glycemic, blood pressure and cholesterol management are substantially reduced when looking at clinical pathways as compared to estimates based on commonly used simple process measures.

Identifying targets to improve treatment in type 2 diabetes; the Groningen Initiative to aNalyse Type 2 diabetes Treatment (GIANTT) observational study

PURPOSE

Assessment of quality of cardiometabolic risk management in diabetes in primary care.

METHODS

In a descriptive cohort study including 95 Dutch general practices, we assessed medication treatment in relation to the level of control for HbA1c, systolic blood pressure (SBP) and LDL-cholesterol (LDL-c) in 2007. We also applied a prospective measure of treatment quality by assessing treatment modifications in not well-controlled patients. In a subpopulation of 23 practices, we studied trends in these quality indicators from 2004 (2059 patients) to 2007 (2929 patients).

RESULTS

In 2007, averages for HbA1c, SBP and LDL-c were 6.9%, 142 mmHg and 2.3 mmol/l, respectively. Of the patients with an HbA1c >  8.5%, 16% were treated with one oral drug class and 50% used insulin. In 27% of these patients, therapy modification occurred subsequently. During the 4-year period, a slight decrease in average HbA1c was observed, but no changes in treatment level. In 2007, 56% of the patients had an SBP ≥ 140 mmHg, 19% of whom were not using antihypertensives. In the 13% with an SBP > 160 mmHg, 23% received a therapy modification. During the 4-year period, the average SBP decreased with 6 mmHg but the treatment level showed no substantial increase. In 2007, 39% had an LDL-c level ≥ 2.5 mmol/l, 49% of whom were not using statins. Of the patients with an LDL-c > 3.5 mmol/l, only 9% received a therapy modification.

CONCLUSIONS

The decreasing population averages of HbA1c, SBP and LDL-c values suggest improvement in quality of care. However, the relatively few therapy modifications observed in insufficiently controlled patients show room for improvement.

Under-documentation of chronic kidney disease in the electronic health record in outpatients

OBJECTIVE

To ascertain if outpatients with moderate chronic kidney disease (CKD) had their condition documented in their notes in the electronic health record (EHR).

DESIGN

Outpatients with CKD were selected based on a reduced estimated glomerular filtration rate and their notes extracted from the Columbia University data warehouse. Two lexical-based classification tools (classifier and word-counter) were developed to identify documentation of CKD in electronic notes.

MEASUREMENTS

The tools categorized patients' individual notes on the basis of the presence of CKD-related terms. Patients were categorized as appropriately documented if their notes contained reference to CKD when CKD was present.

RESULTS

The sensitivities of the classifier and word-count methods were 95.4% and 99.8%, respectively. The specificity of both was 99.8%. Categorization of individual patients as appropriately documented was 96.9% accurate. Of 107 patients with manually verified moderate CKD, 32 (22%) lacked appropriate documentation. Patients whose CKD had not been appropriately documented were significantly less likely to be on renin-angiotensin system inhibitors or have urine protein quantified, and had the illness for half as long (15.1 vs 30.7 months; p<0.01) compared to patients with documentation.

CONCLUSION

Our studies show that lexical-based classification tools can accurately ascertain if appropriate documentation of CKD is present in a EHR. Using this method, we demonstrated under-documentation of patients with moderate CKD. Under-documented patients were less likely to receive CKD guideline recommended care. A tool that prompts providers to document CKD might shorten the time to implementing guideline-based recommendations.

Leveraging informatics for genetic studies: use of the electronic medical record to enable a genome-wide association study of peripheral arterial disease

BACKGROUND

There is significant interest in leveraging the electronic medical record (EMR) to conduct genome-wide association studies (GWAS).

METHODS

A biorepository of DNA and plasma was created by recruiting patients referred for non-invasive lower extremity arterial evaluation or stress ECG. Peripheral arterial disease (PAD) was defined as a resting/post-exercise ankle-brachial index (ABI) less than or equal to 0.9, a history of lower extremity revascularization, or having poorly compressible leg arteries. Controls were patients without evidence of PAD. Demographic data and laboratory values were extracted from the EMR. Medication use and smoking status were established by natural language processing of clinical notes. Other risk factors and comorbidities were ascertained based on ICD-9-CM codes, medication use and laboratory data.

RESULTS

Of 1802 patients with an abnormal ABI, 115 had non-atherosclerotic vascular disease such as vasculitis, Buerger's disease, trauma and embolism (phenocopies) based on ICD-9-CM diagnosis codes and were excluded. The PAD cases (66+/-11 years, 64% men) were older than controls (61+/-8 years, 60% men) but had similar geographical distribution and ethnic composition. Among PAD cases, 1444 (85.6%) had an abnormal ABI, 233 (13.8%) had poorly compressible arteries and 10 (0.6%) had a history of lower extremity revascularization. In a random sample of 95 cases and 100 controls, risk factors and comorbidities ascertained from EMR-based algorithms had good concordance compared with manual record review; the precision ranged from 67% to 100% and recall from 84% to 100%.

CONCLUSION

This study demonstrates use of the EMR to ascertain phenocopies, phenotype heterogeneity and relevant covariates to enable a GWAS of PAD. Biorepositories linked to EMR may provide a relatively efficient means of conducting GWAS.

A systematic literature review: prescribing indicators related to type 2 diabetes mellitus and cardiovascular risk management

PURPOSE

Valid prescribing indicators (PI) are needed for reliable assessment of prescribing quality. The purpose of this study is to describe the validity of existing PI for type 2 diabetes mellitus and cardiovascular risk management.

METHODS

We conducted a systematic literature search for studies describing the development and assessment of relevant PIs between January 1990 and January 2009. We grouped identified PI as drug- or disease-oriented, and according to the aspects of prescribing addressed and the additional clinical information included. We reviewed the clinimetric characteristics of the different types of PI.

RESULTS

We identified 59 documents describing the clinimetrics of 16 types of PI covering relevant prescribing aspects, including first-choice treatment, safety issues, dosing, costs, sufficient and timely treatment. We identified three types of drug-oriented, and five types of disease-oriented PI with proven face and content validity as well as operational feasibility in different settings. PI focusing on treatment modifications were the only indicators that showed concurrent validity. Several solutions were proposed for dealing with case-mix and sample size problems, but their actual effect on PI scores was insufficiently assessed. Predictive validity of individual PI is not yet known.

CONCLUSION

We identified a range of existing PI that are valid for internal quality assessment as they are evidence-based, accepted by professionals, and reliable. For external use, problems of patient case-mix and sample size per PI should be better addressed. Further research is needed for selecting indicators that predict clinical outcomes.

Methods to identify the target population: implications for prescribing quality indicators

Background

Information on prescribing quality is increasingly used by policy makers, insurance companies and health care providers. For reliable assessment of prescribing quality it is important to correctly identify the patients eligible for recommended treatment. Often either diagnostic codes or clinical measurements are used to identify such patients. We compared these two approaches regarding the outcome of the prescribing quality assessment and their ability to identify treated and undertreated patients.

Methods

The approaches were compared using electronic health records for 3214 diabetes patients from 70 general practitioners. We selected three existing prescribing quality indicators (PQI) assessing different aspects of treatment in patients with hypertension or who were overweight. We compared population level prescribing quality scores and proportions of identified patients using definitions of hypertension or being overweight based on diagnostic codes, clinical measurements or both.

Results

The prescribing quality score for prescribing any antihypertensive treatment was 93% (95% confidence interval 90-95%) using the diagnostic code-based approach, and 81% (78-83%) using the measurement-based approach. Patients receiving antihypertensive treatment had a better registration of their diagnosis compared to hypertensive patients in whom such treatment was not initiated. Scores on the other two PQI were similar for the different approaches, ranging from 64 to 66%. For all PQI, the clinical measurement -based approach identified higher proportions of both well treated and undertreated patients compared to the diagnostic code -based approach.

Conclusions

The use of clinical measurements is recommended when PQI are used to identify undertreated patients. Using diagnostic codes or clinical measurement values has little impact on the outcomes of proportion-based PQI when both numerator and denominator are equally affected. In situations when a diagnosis is better registered for treated than untreated patients, as we observed for hypertension, the diagnostic code-based approach results in overestimation of provided treatment.