UK Primary Care Electronic Health Research Articles

IDENTIFYING PRODROMAL FEATURES OF DEMENTIA IN UK PRIMARY CARE: A NESTED CASE-CONTROL STUDY USING ELECTRONIC HEALTH RECORDS FROM 92,623 DEMENTIA PATIENTS

The process of diagnosing dementia is often lengthy and based on a narrow range of cognitive presentations. We aimed to identify prodromal features of dementia in UK primary care electronic health records, in order to facilitate more timely diagnosis. A systematic review was conducted to identify prodromal features of dementia. We generated a study population from the Clinical Practice Research Datalink (CPRD). Based on a dementia phenotyping algorithm, patients were matched with disease-free individuals (1:5). We investigated prodromal features’ six-monthly prevalence and applied Least Absolute Shrinkage and Selection Operator (LASSO) regression to fit models and calculate odds ratios (OR) for dementia diagnosis at thirteen time intervals 20 years prior to diagnosis. We identified 92,623 dementia patients (median age: 83, female: 65.1%, Alzheimer's: 31.6%) and 329 prodromal features. From the period prevalence, medications were the earliest discernable features between cases and controls, including Antiplatelet Drugs and Analgesics, 19 years before diagnosis. Feature odds ratios showed that medications had the greatest number of negatively associated features, physiological features were most consistently positively associated with dementia and features of health system utilization had the greatest extremes of association. Model recall (sensitivity) ranged from 0.69 at 20–18 years to diagnosis, to 0.76 at 20–1 years to diagnosis. Overall precision of the models ranged from 0.13 to 0.30 due to the high proportion of false positives.

IDENTIFYING PRODROMAL FEATURES OF DEMENTIA IN U.K. PRIMARY CARE: A NESTED CASE-CONTROL STUDY USING ELECTRONIC HEALTH RECORDS FROM 92,623 DEMENTIA PATIENTS

The process of diagnosing dementia is often lengthy and based on a narrow range of cognitive presentations. We aimed to identify prodromal features of dementia in UK primary care electronic health records, in order to facilitate more timely diagnosis. A systematic review was conducted to identify prodromal features of dementia. We generated a study population from the Clinical Practice Research Datalink (CPRD). Based on a dementia phenotyping algorithm, patients were matched with disease-free individuals (1:5). We investigated prodromal features’ six-monthly prevalence and applied Least Absolute Shrinkage and Selection Operator (LASSO) regression to fit models and calculate odds ratios (OR) for dementia diagnosis at thirteen time intervals 20 years prior to diagnosis. We identified 92,623 dementia patients (median age: 83, female: 65.1%, Alzheimer's: 31.6%) and 329 prodromal features. From the period prevalence, medications were the earliest discernable features between cases and controls, including Antiplatelet Drugs and Analgesics, 19 years before diagnosis. Feature odds ratios showed that medications had the greatest number of negatively associated features, physiological features were most consistently positively associated with dementia and features of health system utilization had the greatest extremes of association. Model recall (sensitivity) ranged from 0.69 at 20–18 years to diagnosis, to 0.76 at 20–1 years to diagnosis. Overall precision of the models ranged from 0.13 to 0.30 due to the high proportion of false positives.

Antibiotic prescribing for common infections in UK general practice: variability and drivers.

ObjectivesTo examine variations across general practices and factors associated with antibiotic prescribing for common infections in UK primary care to identify potential targets for improvement and optimization of prescribing.MethodsOral antibiotic prescribing for common infections was analysed using anonymized UK primary care electronic health records between 2000 and 2015 using the Clinical Practice Research Datalink (CPRD). The rate of prescribing for each condition was observed over time and mean change points were compared with national guideline updates. Any correlation between the rate of prescribing for each infectious condition was estimated within a practice. Predictors of prescribing were estimated using logistic regression in a matched patient cohort (1:1 by age, sex and calendar time). ResultsOver 8 million patient records were examined in 587 UK general practices. Practices varied considerably in their propensity to prescribe antibiotics and this variance increased over time. Change points in prescribing did not reflect updates to national guidelines. Prescribing levels within practices were not consistent for different infectious conditions. A history of antibiotic use significantly increased the risk of receiving a subsequent antibiotic (by 22%–48% for patients with three or more antibiotic prescriptions in the past 12 months), as did higher BMI, history of smoking and flu vaccinations. Other drivers for receiving an antibiotic varied considerably for each condition.ConclusionsLarge variability in antibiotic prescribing between practices and within practices was observed. Prescribing guidelines alone do not positively influence a change in prescribing, suggesting more targeted interventions are required to optimize antibiotic prescribing in the UK.

Health indicator recording in UK primary care electronic health records: key implications for handling missing data.

BackgroundClinical databases are increasingly used for health research; many of them capture information on common health indicators including height, weight, blood pressure, cholesterol level, smoking status, and alcohol consumption. However, these are often not recorded on a regular basis; missing data are ubiquitous. We described the recording of health indicators in UK primary care and evaluated key implications for handling missing data.MethodsWe examined the recording of health indicators in The Health Improvement Network (THIN) UK primary care database over time, by demographic variables (age and sex) and chronic diseases (diabetes, myocardial infarction, and stroke). Using weight as an example, we fitted linear and logistic regression models to examine the associations of weight measurements and the probability of having weight recorded with individuals’ demographic characteristics and chronic diseases.ResultsIn total, 6,345,851 individuals aged 18–99 years contributed data to THIN between 2000 and 2015. Women aged 18–65 years were more likely than men of the same age to have health indicators recorded; this gap narrowed after age 65. About 60–80% of individuals had their height, weight, blood pressure, smoking status, and alcohol consumption recorded during the first year of registration. In the years following registration, these proportions fell to 10%–40%. Individuals with chronic diseases were more likely to have health indicators recorded, particularly after the introduction of a General Practitioner incentive scheme. Individuals’ demographic characteristics and chronic diseases were associated with both observed weight measurements and missingness in weight.ConclusionMissing data in common health indicators will affect statistical analysis in health research studies. A single analysis of primary care data using the available information alone may be misleading. Multiple imputation of missing values accounting for demographic characteristics and disease status is recommended but should be considered and implemented carefully. Sensitivity analysis exploring alternative assumptions for missing data should also be evaluated.

The internal validation of weight and weight change coding using weight measurement data within the UK primary care Electronic Health Record.

PurposeTo use recorded weight values to internally validate weight status and weight change coding in the primary care Electronic Health Record (EHR).Patients and methodsWe included adult patients with weight-related Read codes recorded in the UK’s Clinical Practice Research Datalink EHR between 2000 and 2017. Weight status codes were compared to weight values recorded on the same day and positive predictive values (PPVs) were calculated for commonly used codes. Weight change codes were validated using three methods: the percentage (%) difference in kilograms at the time of the code and 1) the previous weight measurement, 2) the weight predicted using linear regression, and 3) the historic mean weight. Weight change codes were validated if estimates were consistent across two out of three methods.ResultsA total of 8,108,481 weight codes were recorded in 1,000,002 patients’ EHR. Twice as many were recorded in females (n=5,208,593, 64%). The mean body mass index for “overweight” codes ranged from 31.9 kg/m2 to 46.9 kg/m2 and from 17.4 kg/m2 to 19.2 kg/m2 for “underweight” codes. PPVs for the most commonly used weight status codes ranged from 81.3% (80%–82.5%) to 99.3% (99.2%–99.4%). Across the estimation methods, and using only validated weight change codes, mean weight loss ranged from – 5.2% (SD 5.8%) to −7.9% (SD 7.3%) and mean weight gain from 4.2 % (SD 5.5%) to 7.9 % (SD 8.2%). The previous and predicted weight methods were most consistent.ConclusionWe have developed an internationally applicable methodology to internally validate weight-related EHR coding by using available weight measurement data. We demonstrate the UK Read codes that can be confidently used to classify weight status and weight change in the absence of weight values. We provide the first evidence from primary care that a Read code for unexpected weight loss represents a mean loss of ≥ 5 % in a 6-month period, which was broadly consistent across age groups and gender.

Population-calibrated multiple imputation for a binary/categorical covariate in categorical regression models.

Multiple imputation (MI) has become popular for analyses with missing data in medical research. The standard implementation of MI is based on the assumption of data being missing at random (MAR). However, for missing data generated by missing not at random mechanisms, MI performed assuming MAR might not be satisfactory. For an incomplete variable in a given data set, its corresponding population marginal distribution might also be available in an external data source. We show how this information can be readily utilised in the imputation model to calibrate inference to the population by incorporating an appropriately calculated offset termed the “calibrated‐δ adjustment.” We describe the derivation of this offset from the population distribution of the incomplete variable and show how, in applications, it can be used to closely (and often exactly) match the post‐imputation distribution to the population level. Through analytic and simulation studies, we show that our proposed calibrated‐δ adjustment MI method can give the same inference as standard MI when data are MAR, and can produce more accurate inference under two general missing not at random missingness mechanisms. The method is used to impute missing ethnicity data in a type 2 diabetes prevalence case study using UK primary care electronic health records, where it results in scientifically relevant changes in inference for non‐White ethnic groups compared with standard MI. Calibrated‐δ adjustment MI represents a pragmatic approach for utilising available population‐level information in a sensitivity analysis to explore potential departures from the MAR assumption.

Depression, depressive symptoms and treatments in women who have recently given birth: UK cohort study

ObjectivesTo investigate how depression is recognised in the year after child birth and treatment given in clinical practice.DesignCohort study based on UK primary care electronic health records.SettingPrimary care.ParticipantsWomen who have...

Low salivary cortisol levels in patients with rheumatoid arthritis exposed to oral glucocorticoids: a cross-sectional study set within UK electronic health records

BackgroundGlucocorticoids (GCs) suppress endogenous cortisol levels which can lead to adrenal insufficiency (AI). The frequency of GC-induced AI remains unclear. In this cross-sectional study, low morning salivary cortisol (MSC) levels...

Underrecording of osteoarthritis in United Kingdom primary care electronic health record data.

PurposePrimary care electronic health records are increasingly used to estimate the occurrence of osteoarthritis (OA). We aimed to estimate the extent and trend over time of underrecording of severe OA patients in UK primary care electronic health records using first primary total hip and knee replacements (THR/TKR) – >90% of which are performed for OA – as the reference population.Patients and methodsWe identified patients with a first primary THR or TKR recorded in the UK Clinical Practice Research Datalink between 2000 and 2015. We then searched for a diagnostic/problem code for OA up to 10 years prior to THR/TKR using 3 definitions: “diagnosed OA (joint-specific),” “diagnosed OA (any joint),” “clinical OA” (diagnosed OA or relevant peripheral joint pain symptom code).ResultsAmong 34,299 THR patients identified, 28.1%, 53.4%, and 74.4% had a prior record of diagnosed OA (hip), diagnosed OA (any), and clinical OA, respectively. Among 47,588 TKR patients, the corresponding figures were, 25.5% (diagnosed OA [knee]), 43.7%, and 74.8%. In the UK Clinical Practice Research Datalink, the proportion of patients with prior recorded OA decreased between 2000 and 2015.ConclusionAn increasing trend of underrecording of OA or joint pain among patients with THR or TKR (severe OA patients) between 2000 and 2015 was identified. An underestimate health care demand could be derived based on consultation incidence and prevalence of OA from electronic health record data that relies on osteoarthritis diagnostic codes. Further studies are warranted to investigate the validity of OA or joint pain recorded in primary care settings, which might be used to correct the consultation incidence and prevalence of OA.

Risks of smoking and benefits of smoking cessation on hospitalisations for cardiovascular events and respiratory infection in patients with rheumatoid arthritis: a retrospective cohort study using the Clinical Practice Research Datalink

ObjectivesTo investigate the associations between smoking status, smoking cessation and hospitalisations for cardiovascular events (CVE) and respiratory tract infections (RTI) in an inception cohort of patients with rheumatoid arthritis (RA).MethodsThe...

An algorithm for identification and classification of individuals with type 1 and type 2 diabetes mellitus in a large primary care database.

BackgroundResearch into diabetes mellitus (DM) often requires a reproducible method for identifying and distinguishing individuals with type 1 DM (T1DM) and type 2 DM (T2DM).ObjectivesTo develop a method to identify individuals with T1DM and T2DM using UK primary care electronic health records.MethodsUsing data from The Health Improvement Network primary care database, we developed a two-step algorithm. The first algorithm step identified individuals with potential T1DM or T2DM based on diagnostic records, treatment, and clinical test results. We excluded individuals with records for rarer DM subtypes only. For individuals to be considered diabetic, they needed to have at least two records indicative of DM; one of which was required to be a diagnostic record. We then classified individuals with T1DM and T2DM using the second algorithm step. A combination of diagnostic codes, medication prescribed, age at diagnosis, and whether the case was incident or prevalent were used in this process. We internally validated this classification algorithm through comparison against an independent clinical examination of The Health Improvement Network electronic health records for a random sample of 500 DM individuals.ResultsOut of 9,161,866 individuals aged 0–99 years from 2000 to 2014, we classified 37,693 individuals with T1DM and 418,433 with T2DM, while 1,792 individuals remained unclassified. A small proportion were classified with some uncertainty (1,155 [3.1%] of all individuals with T1DM and 6,139 [1.5%] with T2DM) due to unclear health records. During validation, manual assignment of DM type based on clinical assessment of the entire electronic record and algorithmic assignment led to equivalent classification in all instances.ConclusionThe majority of individuals with T1DM and T2DM can be readily identified from UK primary care electronic health records. Our approach can be adapted for use in other health care settings.

Comment: Does l -thyroxine prevent or cause stroke in hypothyroidism?

The association between overt hypothyroidism and atherosclerotic risk factors, especially hypertension and dyslipidemia, is clear. To date, only a few small and methodologically flawed epidemiologic studies have investigated the relationship between hypothyroidism and stroke, precluding definitive conclusions. In the current issue of Neurology®, Karch and Thomas(1) present a large well-designed case-control study investigating this relationship. They compared 34,907 patients with autoimmune thyroiditis on treatment with l-thyroxine and 149,632 matched individuals without autoimmune thyroiditis from a UK primary care electronic health record system. They found a slight increase in the risk of stroke in patients with autoimmune thyroiditis (adjusted relative risk = 1.10, 95% confidence interval: 1.01-1.20) compared with controls. Some of this increased risk of stroke was independent of cardiovascular risk factors.

Bisphosphonates and risk of upper gastrointestinal cancer--a case control study using the General Practice Research Database (GPRD).

BackgroundConcerns have been raised as to the safety of bisphosphonates; in particular a possible link between bisphosphonate use and upper gastrointestinal (GI) cancer. Two published studies using different study populations but drawn from earlier versions of the same national UK database, reached differing conclusions: one finding no evidence for an increase in the risk of gastric or oesophageal cancer in bisphosphonate users and one finding a small but significantly increased risk of oesophageal cancer linked to duration of bisphosphonate use.Methodology/ Principal FindingsDesign-A case control study comparing bisphosphonate prescribing in cases of upper GI cancer from 1995 to 2007 using UK primary care electronic health records (GPRD).Main Outcome Measure-Relative Risk (approximated to Odds Ratio for rare events) for oesophageal and gastric cancer development in bisphosphonate users compared to non–users. The odds of being a case of oesophageal cancer, adjusted for smoking status, were significantly increased in women who had had one or more bisphosphonate prescriptions, odds ratio 1·54 (95% CI 1·27–1·88) compared to non-users. There was no significant effect on gastric cancer in women, odds ratio adjusted for smoking status, 1.06 (95% CI 0.83–1.37) and also no apparent risk in men for either oesophageal or gastric cancer, odds ratio adjusted for smoking status 0.78 (95%CI 0.56–1.09) and 0.87 (95% CI 0.55–1.36) respectively.Conclusions/ SignificanceOur results support a small but significant increased risk of oesophageal cancer in women prescribed bisphosphonates and is based on the largest number of exposed cases to date in the UK.