Data resource profile: cardiovascular disease research using linked bespoke studies and electronic health records (CALIBER)

Excess health care use is significantly and persistently reduced following diagnosis of late-onset epilepsy

OBJECTIVE

The incidence of late-onset epilepsy (LOE) is rising, and these patients may use an excess of health care resources. This study aimed to measure pre-/post-diagnostic health care use (HCU) for patients with LOE compared to controls.

METHODS

This was an observational open cohort study covering years 1998-2019 using UK population-based linked primary care (Clinical Practice Research Datalink [CPRD]) and hospital (HES) electronic health records. The participants included patients with incident LOE enrolled in CPRD and 1:10 age-, sex-, and general practice-matched controls. The exposure was incident LOE (diagnosed at age ≥65) using a 5-year washout. The main outcome was all HCU (primary care [PC], accident and emergency [A&E], admitted patient and outpatient care) using inverse proportional weighting to PC use and HCU by setting. An interrupted time-series analysis was used to examine pre-/post-diagnostic HCU between patients with LOE and controls over 4 years either side of diagnosis/matching date. An adjusted mixed-effects negative binomial regression was used for post-diagnosis HCU interactions.

RESULTS

Of 2 569 874 people ≥65 years of age, 1048 (4%) developed incident LOE. Mean weighted total HCU increased by 32 visits per patient-year (95% confidence interval [95% CI]: 13-50, p = .003) until LOE diagnosis, and then dropped by a mean of 60 visits per patient-year (95% CI: -81 to -40). There was an acute rise and fall over the 1-2 years immediately pre-/post-diagnosis. Incident HCU remained higher for LOE compared to controls post-diagnosis (adjusted incidence rate ratio: 1.72; 95% CI: 1.65-1.70; p < .001), including A&E, outpatient, and admitted care.

SIGNIFICANCE

Health care use demonstrates an acute on chronic rise over the 4 years before diagnosis of LOE. To what extent the partial reversal of the acute pre-diagnosis rise, and the mediators of the accelerated increase compared to controls are attributed to epilepsy, comorbid and bidirectional disease states, or a combination of both warrants further exploration.

Health map for HealthGap: Defining a geographical catchment to examine cardiovascular risk in Victoria, Australia

The HealthGap study aimed to understand cardiovascular risk among Indigenous Australians in Victoria using linked administrative data. A key challenge was differing spatial coverages of sources: state-level data for risk factors but cardiovascular outcomes for three hospitals. Catchments were defined based on hospital postcodes to estimate denominator populations for risk modelling: first- and second-order neighbours, and spatial distribution of outcomes ('spatial event distribution'). Catchment coverage was assessed through proportions of patients presenting to study hospitals from catchment postcodes. The spatial event distribution performed best, capturing 82% events overall (first-order:40%; second-order:64%) and 65% Indigenous (27% and 45%). No approach excluded proximal non-study hospitals. Spatial event distributions could help define denominator populations when geographic information on outcome data is available but may not avoid potential misclassification.

Use of heart failure medical therapy before and after a cancer diagnosis: A longitudinal study

AIMS

We aim to evaluate change in the use of prognostic guideline-directed medical therapies (GDMTs) for heart failure (HF) before and after a cancer diagnosis as well as the matched non-cancer controls, including renin-angiotensin-system inhibitors (RASIs), beta-blockers, and mineralocorticoid receptor antagonists (MRAs).

METHODS AND RESULTS

We conducted a longitudinal study in patients with HF in the UK Clinical Practice Research Datalink between 2005 and 2021. We selected patients with probable HF with reduced ejection fraction (HFrEF) based on diagnostic and prescription records. We described the longitudinal trends in the use and dosing of GDMTs before and after receiving an incident cancer diagnosis. HF patients with cancer were matched with a 1:1 ratio to HF patients without cancer to investigate the association between cancer diagnosis and treatment adherence, persistence, initiation, and dose titration as odds ratios (ORs) with 95% confidence intervals (CIs) using multivariable logistic regression models. Of 8504 eligible HFrEF patients with incident cancer, 4890 were matched to controls without cancer. The mean age was 75.7 (±8.4) years and 73.9% were male. In the 12 months following a cancer diagnosis, patients experienced reductions in the use and dosing of GDMT. Compared with the non-cancer controls, patients with cancer had higher risks for poor adherence for all three medication classes (RASIs: OR = 1.51, 95% CI = 1.35-1.68; beta-blockers: OR = 1.22, 95% CI = 1.08-1.37; MRAs: OR = 1.31, 95% CI = 1.08-1.59) and poor persistence (RASIs: OR = 2.04, 95% CI = 1.75-2.37; beta-blockers: OR = 1.35, 95% CI = 1.12-1.63; MRAs: OR = 1.49, 95% CI = 1.16-1.93), and higher risks for dose down-titration for RASIs (OR = 1.69, 95% CI = 1.40-2.04) and beta-blockers (OR = 1.31, 95% CI = 1.05-1.62). Cancer diagnosis was not associated with treatment initiation or dose up-titration. Event rates for HF hospitalization and mortality were higher in patients with poor adherence or persistence to GDMTs.

CONCLUSIONS

Following a cancer diagnosis, patients with HFrEF were more likely to have reduced use of GDMTs for HF.

Impact of linkage level on inferences from big data analyses in health and medical research: an empirical study

BACKGROUND

Linkage errors that occur according to linkage levels can adversely affect the accuracy and reliability of analysis results. This study aimed to identify the differences in results according to personally identifiable information linkage level, sample size, and analysis methods through empirical analysis.

METHODS

The difference between the results of linkage in directly identifiable information (DII) and indirectly identifiable information (III) linkage levels was set as III linkage based on name, date of birth, and sex and DII linkage based on resident registration number. The datasets linked at each level were named as databaseIII (DBIII) and databaseDII (DBDII), respectively. Considering the analysis results of the DII-linked dataset as the gold standard, descriptive statistics, group comparison, incidence estimation, treatment effect, and moderation effect analysis results were assessed.

RESULTS

The linkage rates for DBDII and DBIII were 71.1% and 99.7%, respectively. Regarding descriptive statistics and group comparison analysis, the difference in effect in most cases was "none" to "very little." With respect to cervical cancer that had a relatively small sample size, analysis of DBIII resulted in an underestimation of the incidence in the control group and an overestimation of the incidence in the treatment group (hazard ratio [HR] = 2.62 [95% confidence interval (CI): 1.63-4.23] in DBIII vs. 1.80 [95% CI: 1.18-2.73] in DBDII). Regarding prostate cancer, there was a conflicting tendency with the treatment effect being over or underestimated according to the surveillance, epidemiology, and end results summary staging (HR = 2.27 [95% CI: 1.91-2.70] in DBIII vs. 1.92 [95% CI: 1.70-2.17] in DBDII for the localized stage; HR = 1.80 [95% CI: 1.37-2.36] in DBIII vs. 2.05 [95% CI: 1.67-2.52] in DBDII for the regional stage).

CONCLUSIONS

To prevent distortion of the analyses results in health and medical research, it is important to check that the patient population and sample size by each factor of interest (FOI) are sufficient when different data are linked using DBDII. In cases involving a rare disease or with a small sample size for FOI, there is a high likelihood that a DII linkage is unavoidable.

Estimating disease burden using national linked electronic health records: a study using an English population-based cohort

Background

Electronic health records (EHRs) have the potential to be used to produce detailed disease burden estimates. In this study we created disease estimates using national EHR for three high burden conditions, compared estimates between linked and unlinked datasets and produced stratified estimates by age, sex, ethnicity, socio-economic deprivation and geographical region.

Methods

EHRs containing primary care (Clinical Practice Research Datalink), secondary care (Hospital Episode Statistics) and mortality records (Office for National Statistics) were used. We used existing disease phenotyping algorithms to identify cases of cancer (breast, lung, colorectal and prostate), type 1 and 2 diabetes, and lower back pain. We calculated age-standardised incidence of first cancer, point prevalence for diabetes, and primary care consultation prevalence for low back pain.

Results

7.2 million people contributing 45.3 million person-years of active follow-up between 2000-2014 were included. CPRD-HES combined and CPRD-HES-ONS combined lung and bowel cancer incidence estimates by sex were similar to cancer registry estimates. Linked CPRD-HES estimates for combined Type 1 and Type 2 diabetes were consistently higher than those of CPRD alone, with the difference steadily increasing over time from 0.26% (2.99% for CPRD-HES vs. 2.73 for CPRD) in 2002 to 0.58% (6.17% vs. 5.59) in 2013. Low back pain prevalence was highest in the most deprived quintile and when compared to the least deprived quintile the difference in prevalence increased over time between 2000 and 2013, with the largest difference of 27% (558.70 per 10,000 people vs 438.20) in 2013.

Conclusions

We use national EHRs to produce estimates of burden of disease to produce detailed estimates by deprivation, ethnicity and geographical region. National EHRs have the potential to improve disease burden estimates at a local and global level and may serve as more automated, timely and precise inputs for policy making and global burden of disease estimation.

The Polypill: A New Alternative in the Prevention and Treatment of Cardiovascular Disease

Cardiovascular disease (CVD) is the primary cause of death and disability worldwide. Although age-standardized CVD mortality rates decreased globally by 14.5% between 2006 and 2016, the burden of CVD remains disproportionately higher in low- and middle-income countries compared to high-income countries. Even though proven, effective approaches based on multiple-drug intake aimed at the prevention and treatment of CVD are currently available, poor adherence, early discontinuation of treatment, and suboptimal daily execution of the prescribed therapeutic regimes give rise to shortfalls in drug exposure, leading to high variability in the responses to the prescribed medications. Wald and Law, in their landmark paper published in BMJ 2003, hypothesized that the use of a fixed-dose combination of statins, β-blockers, angiotensin receptor blockers, angiotensin-converting enzyme inhibitors, and aspirin (classic Polypill composition) may increase adherence and decrease CVD by up to 80% when prescribed as primary prevention or in substitution of traditional protocols. Since then, many clinical trials have tested this hypothesis, with comparable results. This review aims to describe the available clinical trials performed to assess the impact of fixed-dose combinations on adherence, cost-effectiveness, and the risk factors critical to the onset of CVD.

Sexual and reproductive health and rights of migrant women attending primary care in England: A population-based cohort study of 1.2 million individuals of reproductive age (2009-2018)

Background

Evidence on the sexual and reproductive health and rights (SRHR) of migrants is lacking globally. We describe SRHR healthcare resource use and long-acting reversible contraceptives (LARCs) prescriptions for migrant versus non-migrant women attending primary care in England (2009-2018).

Methods

This population-based observational cohort study, using Clinical Practice Research Datalink (CPRD) GOLD, included females living in England aged 15 to 49. Migration was defined using a validated codelist. Rates per 100 person years at risk (pyar) and adjusted rate ratios (RRs) were measured in migrants versus non-migrants for consultations related to all-causes, six exemplar SRHR outcomes, and LARC prescriptions. Proportions of migrants and non-migrants ever prescribed LARC were calculated.

Findings

There were 25,112,116 consultations across 1,246,353 eligible individuals. 98,214 (7.9 %) individuals were migrants. All-cause consultation rates were lower in migrants versus non-migrants (509 vs 583/100pyar;RR 0.9;95 %CI 0.9-0.9), as were consultations rates for emergency contraception (RR 0.7;95 %CI 0.7-0.7) and cervical screening (RR 0.96;95 %CI 0.95-0.97). Higher rates of consultations were found in migrants for abortion (RR 1.2;95 %CI 1.1-1.2) and management of fertility problems (RR 1.39;95 %CI 1.08-1.79). No significant difference was observed for chlamydia testing and domestic violence. Of 1,205,258 individuals eligible for contraception, the proportion of non-migrants ever prescribed LARC (12.2 %;135,047/1,107,894) was almost double that of migrants (6.91 %;6,728/97,364). Higher copper intrauterine devices prescription rates were found in migrants (RR 1.53;95 %CI 1.45-1.61), whilst hormonal LARC rates were lower for migrants: levonorgestrel intrauterine device (RR 0.63;95 %CI 0.60-0.66), subdermal implant (RR 0.72;95 %CI 0.69-0.75), and progesterone-only injection (RR 0.35;95 %CI 0.34-0.36).

Interpretation

Healthcare resource use differs between migrant and non-migrant women of reproductive age. Opportunities identified for tailored interventions include access to primary care, LARCs, emergency contraception and cervical screening. An inclusive approach to examining health needs is essential to actualise sexual and reproductive health as a human right.

Thiazide Diuretics and Risk of Colorectal Cancer: A Population-Based Cohort Study

Evidence from clinical trials and observational studies on the association between thiazide diuretics and colorectal cancer risk is conflicting. We aimed to determine whether thiazide diuretics are associated with an increased colorectal cancer risk compared with dihydropyridine calcium channel blockers (dCCBs). A population-based, new-user cohort was assembled using the UK Clinical Practice Research Datalink. Between 1990-2018, we compared thiazide diuretic initiators with dCCB initiators and estimated hazard ratios (HR) with 95% confidence intervals (CIs) of colorectal cancer using Cox proportional hazard models. Models were weighted using standardized morbidity ratio weights generated from calendar time-specific propensity scores. The cohort included 377,760 thiazide diuretic initiators and 364,300 dCCB initiators, generating 3,619,883 person-years of follow-up. Compared with dCCBs, thiazide diuretics were not associated with colorectal cancer (weighted HR = 0.97, 95% CI: 0.90, 1.04). Secondary analyses yielded similar results, although an increased risk was observed among patients with inflammatory bowel disease (weighted HR = 2.45, 95% CI: 1.13, 5.35) and potentially polyps (weighted HR = 1.46, 95% CI: 0.93, 2.30). Compared with dCCBs, thiazide diuretics were not associated with an overall increased colorectal cancer risk. While these findings provide some reassurance, research is needed to corroborate the elevated risks observed among patients with inflammatory bowel disease and history of polyps.

Cardiovascular disease incidence rates: a study using routinely collected health data

BACKGROUND

There is substantial evidence that systemic anticancer therapies and radiotherapy can increase the long-term risk of cardiovascular disease (CVD). Optimal management decisions for cancer patients therefore need to take into account the likely risks from a proposed treatment option, as well as its likely benefits. For CVD, the magnitude of the risk depends on the incidence of the disease in the general population to which the patient belongs, including variation with age and sex, as well as on the treatment option under consideration. The aim of this paper is to provide estimates of CVD incidence rates in the general population of England for use in cardio-oncology and in other relevant clinical, research and health policy contexts.

METHODS

We studied a population-based representative cohort, consisting of 2,633,472 individuals, derived by electronic linkage of records from primary care with those of admitted-patient care in England during April 1, 2010, to April 1, 2015. From 38 individual CVDs available via the linked dataset we identified five relevant categories of CVD whose risk may be increased by cancer treatments: four of heart disease and one of stroke.

RESULTS

We calculated incidence rates by age-group and sex for all relevant CVD categories combined, for the four relevant categories of heart disease combined, and for the five relevant CVD categories separately. We present separate incidence rates for all 38 individual CVDs available via the linked dataset. We also illustrate how our data can be used to estimate absolute CVD risks in a range of people with Hodgkin lymphoma treated with chemotherapy and radiotherapy.

CONCLUSIONS

Our results provide population-based CVD incidence rates for a variety of uses, including the estimation of absolute risks of CVD from cancer treatments, thus helping patients and clinicians to make appropriate individualized cancer treatment decisions. Graphical Abstract: Cardiovascular incidence rates for use in cardio-oncology and elsewhere: A presentation of age- and sex-specific cardiovascular disease (CVD) incidence rates for use in calculation of absolute cardiovascular risks of cancer treatments, and in other clinical, research and health policy contexts. Abbreviations - CVD: cardiovascular disease; y: years.

Incidence of 12 common cardiovascular diseases and subsequent mortality risk in the general population

BACKGROUND

Incident events of cardiovascular diseases (CVDs) are heterogenous and may result in different mortality risks. Such evidence may help inform patient and physician decisions in CVD prevention and risk factor management.

AIMS

This study aimed to determine the extent to which incident events of common CVD show heterogeneous associations with subsequent mortality risk in the general population.

METHODS AND RESULTS

Based on England-wide linked electronic health records, we established a cohort of 1 310 518 people ≥30 years of age initially free of CVD and followed up for non-fatal events of 12 common CVD and cause-specific mortality. The 12 CVDs were considered as time-varying exposures in Cox's proportional hazards models to estimate hazard rate ratios (HRRs) with 95% confidence intervals (CIs). Over the median follow-up of 4.2 years (2010-16), 81 516 non-fatal CVD, 10 906 cardiovascular deaths, and 40 843 non-cardiovascular deaths occurred. All 12 CVDs were associated with increased risk of cardiovascular mortality, with HRR (95% CI) ranging from 1.67 (1.47-1.89) for stable angina to 7.85 (6.62-9.31) for haemorrhagic stroke. All 12 CVDs were also associated with increased non-cardiovascular and all-cause mortality risk but to a lesser extent: HRR (95% CI) ranged from 1.10 (1.00-1.22) to 4.55 (4.03-5.13) and from 1.24 (1.13-1.35) to 4.92 (4.44-5.46) for transient ischaemic attack and sudden cardiac arrest, respectively.

CONCLUSION

Incident events of 12 common CVD show significant adverse and markedly differential associations with subsequent cardiovascular, non-cardiovascular, and all-cause mortality risk in the general population.

South Asian, Black and White ethnicity and the effect of potentially modifiable risk factors for dementia: A study in English electronic health records

INTRODUCTION

We aimed to investigate ethnic differences in the associations of potentially modifiable risk factors with dementia.

METHODS

We used anonymised data from English electronic primary care records for adults aged 65 and older between 1997 and 2018. We used Cox regression to investigate main effects for each risk factor and interaction effects between each risk factor and ethnicity.

RESULTS

We included 865,674 people with 8,479,973 person years of follow up. Hypertension, dyslipidaemia, obesity and diabetes were more common in people from minority ethnic groups than White people. The impact of hypertension, obesity, diabetes, low HDL and sleep disorders on dementia risk was increased in South Asian people compared to White people. The impact of hypertension was greater in Black compared to White people.

DISCUSSION

Dementia prevention efforts should be targeted towards people from minority ethnic groups and tailored to risk factors of particular importance.

Association of comorbid-socioeconomic clusters with mortality in late onset epilepsy derived through unsupervised machine learning

BACKGROUND AND OBJECTIVES

Late-onset epilepsy is a heterogenous entity associated with specific aetiologies and an elevated risk of premature mortality. Specific multimorbid-socioeconomic profiles and their unique prognostic trajectories have not been described. We sought to determine if specific clusters of late onset epilepsy exist, and whether they have unique hazards of premature mortality.

METHODS

We performed a retrospective observational cohort study linking primary and hospital-based UK electronic health records with vital statistics data (covering years 1998-2019) to identify all cases of incident late onset epilepsy (from people aged ≥65) and 1:10 age, sex, and GP practice-matched controls. We applied hierarchical agglomerative clustering using common aetiologies identified at baseline to define multimorbid-socioeconomic profiles, compare hazards of early mortality, and tabulating causes of death stratified by cluster.

RESULTS

From 1,032,129 people aged ≥65, we identified 1048 cases of late onset epilepsy who were matched to 10,259 controls. Median age at epilepsy diagnosis was 68 (interquartile range: 66-72) and 474 (45%) were female. The hazard of premature mortality related to late-onset epilepsy was higher than matched controls (hazard ratio [HR] 1.73; 95% confidence interval [95%CI] 1.51-1.99). Ten unique phenotypic clusters were identified, defined by 'healthy' males and females, ischaemic stroke, intracerebral haemorrhage (ICH), ICH and alcohol misuse, dementia and anxiety, anxiety, depression in males and females, and brain tumours. Cluster-specific hazards were often similar to that derived for late-onset epilepsy as a whole. Clusters that differed significantly from the base late-onset epilepsy hazard were 'dementia and anxiety' (HR 5.36; 95%CI 3.31-8.68), 'brain tumour' (HR 4.97; 95%CI 2.89-8.56), 'ICH and alcohol misuse' (HR 2.91; 95%CI 1.76-4.81), and 'ischaemic stroke' (HR 2.83; 95%CI 1.83-4.04). These cluster-specific risks were also elevated compared to those derived for tumours, dementia, ischaemic stroke, and ICH in the whole population. Seizure-related cause of death was uncommon and restricted to the ICH, ICH and alcohol misuse, and healthy female clusters.

SIGNIFICANCE

Late-onset epilepsy is an amalgam of unique phenotypic clusters that can be quantitatively defined. Late-onset epilepsy and cluster-specific comorbid profiles have complex effects on premature mortality above and beyond the base rates attributed to epilepsy and cluster-defining comorbidities alone.

Periodontal disease in people with a history of psychosis: Results from the UK biobank population-based study

OBJECTIVES

To test the hypotheses that: (1) Prevalence of periodontal disease would be higher in people with a history of psychosis when compared to the general population and (2) Demographic, life-style related factors and co-morbid medical conditions would predict periodontal disease in people experiencing psychosis.

METHODS

The authors performed cross-sectional analysis of baseline data from the UK Biobank study (2007-2010), identifying cases with psychosis using clinical diagnosis, antipsychotic medication, and self-report. Demographic (age, gender, ethnicity, socioeconomic status), lifestyle-related(BMI, blood pressure, smoking and alcohol intake, physical activity) and physical co-morbidities (cancer, cardiovascular, respiratory, inflammatory disease and metabolic conditions) were included as potential risk factors for periodontal disease among people with a history of psychosis using logistic regression analyses. The analysis sample included 502,505 participants.

RESULTS

Risk of periodontal disease was higher in people with psychosis, regardless of how cases were identified. Patients with a clinical diagnosis had the highest proportion of periodontal disease compared to the general population (21.3% vs. 14.8%, prevalence ratio 1.40, 95% CI: 1.26-1.56). Older and female cases were more likely to experience periodontal disease. Lifestyle factors (smoking) and comorbidities (cardiovascular, cancer or respiratory disease) were associated with periodontal disease among people with a history of psychosis.

CONCLUSIONS

The findings suggest that periodontal disease is more common in people with a history of psychosis, compared to the general population. Prevention and early diagnosis of periodontal disease should be a priority for oral health promotion programmes, which should also address modifiable risk factors like smoking which also contribute to co-morbid systemic disease.

Evaluating Metformin Strategies for Cancer Prevention: A Target Trial Emulation Using Electronic Health Records

BACKGROUND

Metformin users appear to have a substantially lower risk of cancer than nonusers in many observational studies. These inverse associations may be explained by common flaws in observational analyses that can be avoided by explicitly emulating a target trial.

METHODS

We emulated target trials of metformin therapy and cancer risk using population-based linked electronic health records from the UK (2009-2016). We included individuals with diabetes, no history of cancer, no recent prescription for metformin or other glucose-lowering medication, and hemoglobin A1c (HbA1c) <64 mmol/mol (<8.0%). Outcomes included total cancer and 4 site-specific cancers (breast, colorectal, lung, and prostate). We estimated risks using pooled logistic regression with adjustment for risk factors via inverse-probability weighting. We emulated a second target trial among individuals regardless of diabetes status. We compared our estimates with those obtained using previously applied analytic approaches.

RESULTS

Among individuals with diabetes, the estimated 6-year risk differences (metformin - no metformin) were -0.2% (95% CI = -1.6%, 1.3%) in the intention-to-treat analysis and 0.0% (95% CI = -2.1%, 2.3%) in the per-protocol analysis. The corresponding estimates for all site-specific cancers were close to zero. Among individuals regardless of diabetes status, these estimates were also close to zero and more precise. By contrast, previous analytic approaches yielded estimates that appeared strongly protective.

CONCLUSIONS

Our findings are consistent with the hypothesis that metformin therapy does not meaningfully influence cancer incidence. The findings highlight the importance of explicitly emulating a target trial to reduce bias in the effect estimates derived from observational analyses.

Determining cardiovascular risk in patients with unattributed chest pain in UK primary care: an electronic health record study

AIMS

Most adults presenting in primary care with chest pain symptoms will not receive a diagnosis ('unattributed' chest pain) but are at increased risk of cardiovascular events. To assess within patients with unattributed chest pain, risk factors for cardiovascular events and whether those at greatest risk of cardiovascular disease can be ascertained by an existing general population risk prediction model or by development of a new model.

METHODS AND RESULTS

The study used UK primary care electronic health records from the Clinical Practice Research Datalink linked to admitted hospitalizations. Study population was patients aged 18 plus with recorded unattributed chest pain 2002-2018. Cardiovascular risk prediction models were developed with external validation and comparison of performance to QRISK3, a general population risk prediction model. There were 374 917 patients with unattributed chest pain in the development data set. The strongest risk factors for cardiovascular disease included diabetes, atrial fibrillation, and hypertension. Risk was increased in males, patients of Asian ethnicity, those in more deprived areas, obese patients, and smokers. The final developed model had good predictive performance (external validation c-statistic 0.81, calibration slope 1.02). A model using a subset of key risk factors for cardiovascular disease gave nearly identical performance. QRISK3 underestimated cardiovascular risk.

CONCLUSION

Patients presenting with unattributed chest pain are at increased risk of cardiovascular events. It is feasible to accurately estimate individual risk using routinely recorded information in the primary care record, focusing on a small number of risk factors. Patients at highest risk could be targeted for preventative measures.

Independent Associations of Incident Epilepsy and Enzyme-Inducing and Non-Enzyme-Inducing Antiseizure Medications With the Development of Osteoporosis

Importance

Both epilepsy and enzyme-inducing antiseizure medications (eiASMs) having varying reports of an association with increased risks for osteoporosis.

Objective

To quantify and model the independent hazards for osteoporosis associated with incident epilepsy and eiASMS and non-eiASMs.

Design, Setting, and Participants

This open cohort study covered the years 1998 to 2019, with a median (IQR) follow-up of 5 (1.7-11.1) years. Data were collected for 6275 patients enrolled in the Clinical Practice Research Datalink and from hospital electronic health records. No patients who met inclusion criteria (Clinical Practice Research Datalink-acceptable data, aged 18 years or older, follow-up after the Hospital Episode Statistics patient care linkage date of 1998, and free of osteoporosis at baseline) were excluded or declined.

Exposure

Incident adult-onset epilepsy using a 5-year washout and receipt of 4 consecutive ASMs.

Main Outcomes and Measures

The outcome was incident osteoporosis as determined through Cox proportional hazards or accelerated failure time models where appropriate. Incident epilepsy was treated as a time-varying covariate. Analyses controlled for age, sex, socioeconomic status, cancer, 1 or more years of corticosteroid use, body mass index, bariatric surgery, eating disorders, hyperthyroidism, inflammatory bowel disease, rheumatoid arthritis, smoking status, falls, fragility fractures, and osteoporosis screening tests. Subsequent analyses (1) excluded body mass index, which was missing in 30% of patients; (2) applied propensity score matching for receipt of an eiASM; (3) restricted analyses to only those with incident onset epilepsy; and (4) restricted analyses to patients who developed epilepsy at age 65 years or older. Analyses were performed between July 1 and October 31, 2022, and in February 2023 for revisions.

Results

Of 8 095 441 adults identified, 6275 had incident adult-onset epilepsy (3220 female [51%] and 3055 male [49%]; incidence rate, 62 per 100 000 person-years) with a median (IQR) age of 56 (38-73) years. When controlling for osteoporosis risk factors, incident epilepsy was independently associated with a 41% faster time to incident osteoporosis (time ratio [TR], 0.59; 95% CI, 0.52-0.67; P < .001). Both eiASMs (TR, 0.91; 95% CI, 0.87-0.95; P < .001) and non-eiASMs (TR, 0.77; 95% CI, 0.76-0.78; P < .001) were also associated with significant increased risks independent of epilepsy, accounting for 9% and 23% faster times to development of osteoporosis, respectively. The independent associations among epilepsy, eiASMs, and non-eiASMs remained consistent in propensity score-matched analyses, cohorts restricted to adult-onset epilepsy, and cohorts restricted to late-onset epilepsy.

Conclusions and Relevance

These findings suggest that epilepsy is independently associated with a clinically meaningful increase in the risk for osteoporosis, as are both eiASMs and non-eiASMs. Routine screening and prophylaxis should be considered in all people with epilepsy.

What quantifies good primary care in the United States? A review of algorithms and metrics using real-world data

Primary care physicians (PCPs) play an indispensable role in providing comprehensive care and referring patients for specialty care and other medical services. As the COVID-19 outbreak disrupts patient access to care, understanding the quality of primary care is critical at this unprecedented moment to support patients with complex medical needs in the primary care setting and inform policymakers to redesign our primary care system. The traditional way of collecting information from patient surveys is time-consuming and costly, and novel data collection and analysis methods are needed. In this review paper, we describe the existing algorithms and metrics that use the real-world data to qualify and quantify primary care, including the identification of an individual's likely PCP (identification of plurality provider and major provider), assessment of process quality (for example, appropriate-care-model composite measures), and continuity and regularity of care index (including the interval index, variance index and relative variance index), and highlight the strength and limitation of real world data from electronic health records (EHRs) and claims data in determining the quality of PCP care. The EHR audits facilitate assessing the quality of the workflow process and clinical appropriateness of primary care practices. With extensive and diverse records, administrative claims data can provide reliable information as it assesses primary care quality through coded information from different providers or networks. The use of EHRs and administrative claims data may be a cost-effective analytic strategy for evaluating the quality of primary care.

Evaluation of data processing pipelines on real-world electronic health records data for the purpose of measuring patient similarity

BACKGROUND

The ever-growing size, breadth, and availability of patient data allows for a wide variety of clinical features to serve as inputs for phenotype discovery using cluster analysis. Data of mixed types in particular are not straightforward to combine into a single feature vector, and techniques used to address this can be biased towards certain data types in ways that are not immediately obvious or intended. In this context, the process of constructing clinically meaningful patient representations from complex datasets has not been systematically evaluated.

AIMS

Our aim was to a) outline and b) implement an analytical framework to evaluate distinct methods of constructing patient representations from routine electronic health record data for the purpose of measuring patient similarity. We applied the analysis on a patient cohort diagnosed with chronic obstructive pulmonary disease.

METHODS

Using data from the CALIBER data resource, we extracted clinically relevant features for a cohort of patients diagnosed with chronic obstructive pulmonary disease. We used four different data processing pipelines to construct lower dimensional patient representations from which we calculated patient similarity scores. We described the resulting representations, ranked the influence of each individual feature on patient similarity and evaluated the effect of different pipelines on clustering outcomes. Experts evaluated the resulting representations by rating the clinical relevance of similar patient suggestions with regard to a reference patient.

RESULTS

Each of the four pipelines resulted in similarity scores primarily driven by a unique set of features. It was demonstrated that data transformations according to each pipeline prior to clustering can result in a variation of clustering results of over 40%. The most appropriate pipeline was selected on the basis of feature ranking and clinical expertise. There was moderate agreement between clinicians as measured by Cohen's kappa coefficient.

CONCLUSIONS

Data transformation has downstream and unforeseen consequences in cluster analysis. Rather than viewing this process as a black box, we have shown ways to quantitatively and qualitatively evaluate and select the appropriate preprocessing pipeline.

Impact of a Prior Cancer Diagnosis on Quality of Care and Survival Following Acute Myocardial Infarction: Retrospective Population-Based Cohort Study in England

BACKGROUND

An increasing proportion of patients with cancer experience acute myocardial infarction (AMI). We investigated differences in quality of AMI care and survival between patients with and without previous cancer diagnoses.

METHODS

A retrospective cohort study using Virtual Cardio-Oncology Research Initiative data. Patients aged 40+ years hospitalized in England with AMI between January 2010 and March 2018 were assessed, ascertaining previous cancers diagnosed within 15 years. Multivariable regression was used to assess effects of cancer diagnosis, time, stage, and site on international quality indicators and mortality.

RESULTS

Of 512 388 patients with AMI (mean age, 69.3 years; 33.5% women), 42 187 (8.2%) had previous cancers. Patients with cancer had significantly lower use of ACE (angiotensin-converting enzyme) inhibitors/angiotensin receptor blockers (mean percentage point decrease [mppd], 2.6% [95% CI, 1.8-3.4]) and lower overall composite care (mppd, 1.2% [95% CI, 0.9-1.6]). Poorer quality indicator attainment was observed in patients with cancer diagnosed in the last year (mppd, 1.4% [95% CI, 1.8-1.0]), with later stage disease (mppd, 2.5% [95% CI, 3.3-1.4]), and with lung cancer (mppd, 2.2% [95% CI, 3.0-1.3]). Twelve-month all-cause survival was 90.5% in noncancer controls and 86.3% in adjusted counterfactual controls. Differences in post-AMI survival were driven by cancer-related deaths. Modeling improving quality indicator attainment to noncancer patient levels showed modest 12-month survival benefits (lung cancer, 0.6%; other cancers, 0.3%).

CONCLUSIONS

Measures of quality of AMI care are poorer in patients with cancer, with lower use of secondary prevention medications. Findings are primarily driven by differences in age and comorbidities between cancer and noncancer populations and attenuated after adjustment. The largest impact was observed in recent cancer diagnoses (<1 year) and lung cancer. Further investigation will determine whether differences reflect appropriate management according to cancer prognosis or whether opportunities to improve AMI outcomes in patients with cancer exist.

Hospital-wide natural language processing summarising the health data of 1 million patients

Electronic health records (EHRs) represent a major repository of real world clinical trajectories, interventions and outcomes. While modern enterprise EHR's try to capture data in structured standardised formats, a significant bulk of the available information captured in the EHR is still recorded only in unstructured text format and can only be transformed into structured codes by manual processes. Recently, Natural Language Processing (NLP) algorithms have reached a level of performance suitable for large scale and accurate information extraction from clinical text. Here we describe the application of open-source named-entity-recognition and linkage (NER+L) methods (CogStack, MedCAT) to the entire text content of a large UK hospital trust (King's College Hospital, London). The resulting dataset contains 157M SNOMED concepts generated from 9.5M documents for 1.07M patients over a period of 9 years. We present a summary of prevalence and disease onset as well as a patient embedding that captures major comorbidity patterns at scale. NLP has the potential to transform the health data lifecycle, through large-scale automation of a traditionally manual task.

Sex Differences in Oral Anticoagulation Therapy in Patients Hospitalized With Atrial Fibrillation: A Nationwide Cohort Study

Background Important disparities in the treatment and outcomes of women and men with atrial fibrillation (AF) are well recognized. Whether introduction of direct oral anticoagulants has reduced disparities in treatment is uncertain. Methods and Results All patients who had an incident hospitalization from 2010 to 2019 with nonvalvular AF in Scotland were included in the present cohort study. Community drug dispensing data were used to determine prescribed oral anticoagulation therapy and comorbidity status. Logistic regression modeling was used to evaluate patient factors associated with treatment with vitamin K antagonists and direct oral anticoagulants. A total of 172 989 patients (48% women [82 833 of 172 989]) had an incident hospitalization with nonvalvular AF in Scotland between 2010 and 2019. By 2019, factor Xa inhibitors accounted for 83.6% of all oral anticoagulants prescribed, while treatment with vitamin K antagonists and direct thrombin inhibitors declined to 15.9% and 0.6%, respectively. Women were less likely to be prescribed any oral anticoagulation therapy compared with men (adjusted odds ratio [aOR], 0.68 [95% CI, 0.67-0.70]). This disparity was mainly attributed to vitamin K antagonists (aOR, 0.68 [95% CI, 0.66-0.70]), while there was less disparity in the use of factor Xa inhibitors between women and men (aOR, 0.92 [95% CI, 0.90-0.95]). Conclusions Women with nonvalvular AF were significantly less likely to be prescribed vitamin K antagonists compared with men. Most patients admitted to the hospital in Scotland with incident nonvalvular AF are now treated with factor Xa inhibitors and this is associated with fewer treatment disparities between women and men.

Identifying and visualising multimorbidity and comorbidity patterns in patients in the English National Health Service: a population-based study

BACKGROUND

Globally, there is a paucity of multimorbidity and comorbidity data, especially for minority ethnic groups and younger people. We estimated the frequency of common disease combinations and identified non-random disease associations for all ages in a multiethnic population.

METHODS

In this population-based study, we examined multimorbidity and comorbidity patterns stratified by ethnicity or race, sex, and age for 308 health conditions using electronic health records from individuals included on the Clinical Practice Research Datalink linked with the Hospital Episode Statistics admitted patient care dataset in England. We included individuals who were older than 1 year and who had been registered for at least 1 year in a participating general practice during the study period (between April 1, 2010, and March 31, 2015). We identified the most common combinations of conditions and comorbidities for index conditions. We defined comorbidity as the accumulation of additional conditions to an index condition over an individual's lifetime. We used network analysis to identify conditions that co-occurred more often than expected by chance. We developed online interactive tools to explore multimorbidity and comorbidity patterns overall and by subgroup based on ethnicity, sex, and age.

FINDINGS

We collected data for 3 872 451 eligible patients, of whom 1 955 700 (50·5%) were women and girls, 1 916 751 (49·5%) were men and boys, 2 666 234 (68·9%) were White, 155 435 (4·0%) were south Asian, and 98 815 (2·6%) were Black. We found that a higher proportion of boys aged 1-9 years (132 506 [47·8%] of 277 158) had two or more diagnosed conditions than did girls in the same age group (106 982 [40·3%] of 265 179), but more women and girls were diagnosed with multimorbidity than were boys aged 10 years and older and men (1 361 232 [80·5%] of 1 690 521 vs 1 161 308 [70·8%] of 1 639 593). White individuals (2 097 536 [78·7%] of 2 666 234) were more likely to be diagnosed with two or more conditions than were Black (59 339 [60·1%] of 98 815) or south Asian individuals (93 617 [60·2%] of 155 435). Depression commonly co-occurred with anxiety, migraine, obesity, atopic conditions, deafness, soft-tissue disorders, and gastrointestinal disorders across all subgroups. Heart failure often co-occurred with hypertension, atrial fibrillation, osteoarthritis, stable angina, myocardial infarction, chronic kidney disease, type 2 diabetes, and chronic obstructive pulmonary disease. Spinal fractures were most strongly non-randomly associated with malignancy in Black individuals, but with osteoporosis in White individuals. Hypertension was most strongly associated with kidney disorders in those aged 20-29 years, but with dyslipidaemia, obesity, and type 2 diabetes in individuals aged 40 years and older. Breast cancer was associated with different comorbidities in individuals from different ethnic groups. Asthma was associated with different comorbidities between males and females. Bipolar disorder was associated with different comorbidities in younger age groups compared with older age groups.

INTERPRETATION

Our findings and interactive online tools are a resource for: patients and their clinicians, to prevent and detect comorbid conditions; research funders and policy makers, to redesign service provision, training priorities, and guideline development; and biomedical researchers and manufacturers of medicines, to provide leads for research into common or sequential pathways of disease and inform the design of clinical trials.

FUNDING

UK Research and Innovation, Medical Research Council, National Institute for Health and Care Research, Department of Health and Social Care, Wellcome Trust, British Heart Foundation, and The Alan Turing Institute.

Incidence, morbidity, mortality and disparities in dementia: A population linked electronic health records study of 4.3 million individuals

INTRODUCTION

We report dementia incidence, comorbidities, reasons for health-care visits, mortality, causes of death, and examined dementia patterns by relative deprivation in the UK.

METHOD

A longitudinal cohort analysis of linked electronic health records from 4.3 million people in the UK was conducted to investigate dementia incidence and mortality. Reasons for hospitalization and causes of death were compared in individuals with and without dementia.

RESULTS

From 1998 to 2016 we observed 145,319 (3.1%) individuals with incident dementia. Repeated hospitalizations among senior adults for infection, unknown morbidity, and multiple primary care visits for chronic pain were observed prior to dementia diagnosis. Multiple long-term conditions are present in half of the individuals at the time of diagnosis. Individuals living in high deprivation areas had higher dementia incidence and high fatality.

DISCUSSION

There is a considerable disparity of dementia that informs priorities of prevention and provision of patient care.

Dihydropyridine Calcium Channel Blockers and Risk of Pancreatic Cancer: A Population-Based Cohort Study

Background Recent studies have reported that dihydropyridine calcium channel blockers (dCCBs) may increase the risk of pancreatic cancer, but these studies had methodological limitations. We thus aimed to determine whether dCCBs are associated with an increased risk of pancreatic cancer compared with thiazide diuretics, a clinically relevant comparator. Methods and Results We conducted a new user, active comparator, population-based cohort study using the UK Clinical Practice Research Datalink. We identified new users of dCCBs and new users of thiazide diuretics between 1990 and 2018, with follow-up until 2019. Cox proportional hazards models were used to estimate hazard ratios (HRs) with 95% CIs for pancreatic cancer, comparing dCCBs with thiazide diuretics. Models were weighted using standardized morbidity ratio weights based on calendar time-specific propensity scores. We also conducted secondary analyses by cumulative duration of use, time since initiation, and individual drugs and assessed for the presence of effect modification by age, sex, smoking status, body mass index, history of chronic pancreatitis, and diabetes. The cohort included 344 480 initiators of dCCBs and 357 968 initiators of thiazide diuretics, generating 3 360 745 person-years of follow-up. After a median follow-up of 4.5 years, the weighted incidence rate per 100 000 person-years was 37.2 (95% CI, 34.1-40.4) for dCCBs and 39.4 (95% CI, 36.1-42.9) for thiazide diuretics. Overall, dCCBs were not associated with an increased risk of pancreatic cancer (weighted HR, 0.93; 95% CI, 0.80-1.09). Similar results were observed in secondary analyses. Conclusions In this large, population-based cohort study, dCCBs were not associated with an increased risk of pancreatic cancer compared with thiazide diuretics. These findings provide reassurance regarding the long-term pancreatic cancer safety of these drugs.

Neural-signature methods for structured EHR prediction

Models that can effectively represent structured Electronic Healthcare Records (EHR) are central to an increasing range of applications in healthcare. Due to the sequential nature of health data, Recurrent Neural Networks have emerged as the dominant component within state-of-the-art architectures. The signature transform represents an alternative modelling paradigm for sequential data. This transform provides a non-learnt approach to creating a fixed vector representation of temporal features and has shown strong performances across an increasing number of domains, including medical data. However, the signature method has not yet been applied to structured EHR data. To this end, we follow recent work that enables the signature to be used as a differentiable layer within a neural architecture enabling application in high dimensional domains where calculation would have previously been intractable. Using a heart failure prediction task as an exemplar, we provide an empirical evaluation of different variations of the signature method and compare against state-of-the-art baselines. This first application of neural-signature methods in real-world healthcare data shows a competitive performance when compared to strong baselines and thus warrants further investigation within the health domain.

Role of Antihypertensive Treatment and Blood Pressure Control in the Occurrence of Adverse Pregnancy Outcomes: a Population-Based Study of Linked Electronic Health Records

BACKGROUND

Chronic hypertension (CH) adversely impacts pregnancy. It remains unclear whether antihypertensive treatment alters these risks. We examined the role of antihypertensive treatment in the association between CH and adverse pregnancy outcomes.

METHODS

Electronic health records from the UK Caliber and Clinical Practice Research Datalink were used to define a cohort of women delivering between 1997 and 2016. Primary outcomes were preeclampsia, preterm birth (PTB), and fetal growth restriction (FGR). We used multivariable logistic regression to compare outcomes in women with CH to women without CH and propensity score matching to compare antihypertensive agents.

RESULTS

The study cohort consisted of 1 304 679 women and 1 894 184 births. 14 595 (0.77%) had CH, and 6786 (0.36%) were prescribed antihypertensive medications in pregnancy. Overall, women with CH (versus no CH), had higher odds of preeclampsia (adjusted odds ratio [aOR], 5.74 [95% CI, 5.44-6.07]); PTB (aOR, 2.53 [2.39-2.67]); and FGR (aOR, 2.51 [2.31-2.72]). Women with CH prescribed treatment (versus untreated women) had higher odds of preeclampsia (aOR, 1.17 [1.05-1.30]), PTB (1.25 [1.12-1.39]), and FGR (1.80 [1.51-2.14]). Women prescribed methyldopa (versus β-blockers) had higher odds of preeclampsia (aOR, 1.43 [1.19-1.73]); PTB (1.59 [1.30-1.93]), but lower odds of FGR (aOR, 0.66 [0.48-0.90]). Odds of adverse outcomes were similar in relation to calcium channel blockers (versus β-blockers) except for PTB (aOR, 1.94 [1.15-3.27]). Among women prescribed treatment, lower average blood pressure (<135/85 mm Hg) was associated with better pregnancy outcomes.

CONCLUSIONS

Treatment with antihypertensive agents and control of hypertension ameliorates some effects but higher risks of adverse outcomes persist. β-Blockers versus methyldopa may be associated with better pregnancy outcomes except for FGR. Powered trials are needed to inform optimal treatment of CH during pregnancy.

The additive effect of periodontitis with hypertension on risk of systemic disease and mortality

BACKGROUND

Recent evidence suggests that periodontitis (PD) causes hypertension, which is a precursor to development of other systemic diseases. The aim of this study was to examine the effect of hypertension and PD on the risk of subsequent systemic disease.

METHODS

This longitudinal cohort study included 244,393 UK Biobank participants who were free of systemic disease other than hypertension at baseline. Self-reported responses of painful gums or loose teeth were surrogates for PD. Hypertensives were identified by clinical diagnosis, or elevated blood pressure (≥140/90 mmHg). Systemic diseases including cancer, cardiovascular disease (CVD), and diabetes were identified from linked diagnostic codes. Multivariable Cox proportional hazard models were used to quantify the risk of systemic diseases and all-cause mortality, stratified by hypertensive and PD status.

RESULTS

The average age of the study population was 55.4 years (standard deviation [SD:] 8.1 years), and 130,220 (53.3%) participants were female. At baseline, 131,566 (53.8%) participants were hypertensive and 4.5% reported PD. The incidence rates of all systemic diseases were higher in hypertensive than non-hypertensive participants of the same PD status. In hypertensives, an additive effect was observed for PD on the risks of CVD (adjusted hazard ratio [HR]: 1.35, 95% confidence interval [CI]: 1.21-1.53) and respiratory disease (HR: 1.11, 95% CI: 0.95-1.30) compared to hypertensive healthy controls.

CONCLUSIONS

Hypertensives with PD have exacerbated risks of several systemic diseases. Future interventional studies should consider the effect of periodontal treatment on systemic outcomes in targeted hypertensive populations.

Treatment and prescribing trends of antihypertensive drugs in 2.7 million UK primary care patients over 31 years: a population-based cohort study

OBJECTIVES

To describe the prescribing trends of antihypertensive drugs in primary care patients and assess the trajectory of antihypertensive drug prescriptions, from first-line to third-line, in patients with hypertension according to changes to the United Kingdom (UK) hypertension management guidelines.

DESIGN

Population-based cohort study.

SETTING AND PARTICIPANTS

We used the UK Clinical Practice Research Datalink, an electronic primary care database representative of the UK population. Between 1988 and 2018, we identified all adult patients with at least one prescription for a thiazide diuretic, angiotensin-converting enzyme (ACE) inhibitor, angiotensin receptor blocker, beta-blocker or calcium channel blocker (CCB).

PRIMARY AND SECONDARY OUTCOME MEASURES

We estimated the period prevalence of patients with antihypertensive drug prescriptions for each calendar year over a 31-year period. Treatment trajectory was assessed by identifying patients with hypertension newly initiating an antihypertensive drug, and treatment changes were defined by a switch or add-on of a new class. This cohort was stratified before and after 2007, the year following important changes to UK hypertension management guidelines.

RESULTS

The cohort included 2 709 241 patients. The prevalence of primary care patients with antihypertensive drug prescriptions increased from 7.8% (1988) to 21.9% (2018) and was observed for all major classes except thiazide diuretics. Patients with hypertension initiated thiazide diuretics (36.8%) and beta-blockers (23.6%) as first-line drugs before 2007, and ACE inhibitors (39.9%) and CCBs (31.8%) after 2007. After 2007, 17.3% were not prescribed guideline-recommended first-line agents. Overall, patients were prescribed a median of 2 classes (IQR 1-2) after first-line treatment.

CONCLUSION

Nearly one-quarter of primary care patients were prescribed antihypertensive drugs by the end of the study period. Most patients with hypertension initiated guideline-recommended first-line agents. Not all patients, particularly females, were prescribed recommended agents however, potentially leading to suboptimal cardiovascular outcomes. Future research should aim to better understand the implication of this finding.

Target Trial Emulation for Transparent and Robust Estimation of Treatment Effects for Health Technology Assessment Using Real-World Data: Opportunities and Challenges

Evidence about the relative effects of new treatments is typically collected in randomised controlled trials (RCTs). In many instances, evidence from RCTs falls short of the needs of health technology assessment (HTA). For example, RCTs may not be able to capture longer-term treatment effects, or include all relevant comparators and outcomes required for HTA purposes. Information routinely collected about patients and the care they receive have been increasingly used to complement RCT evidence on treatment effects. However, such routine (or real-world) data are not collected for research purposes, so investigators have little control over the way patients are selected into the study or allocated to the different treatment groups, introducing biases for example due to selection or confounding. A promising approach to minimise common biases in non-randomised studies that use real-world data (RWD) is to apply design principles from RCTs. This approach, known as 'target trial emulation' (TTE), involves (1) developing the protocol with respect to core study design and analysis components of the hypothetical RCT that would answer the question of interest, and (2) applying this protocol to the RWD so that it mimics the data that would have been gathered for the RCT. By making the 'target trial' explicit, TTE helps avoid common design flaws and methodological pitfalls in the analysis of non-randomised studies, keeping each step transparent and accessible. It provides a coherent framework that embeds existing analytical methods to minimise confounding and helps identify potential limitations of RWD and the extent to which these affect the HTA decision. This paper provides a broad overview of TTE and discusses the opportunities and challenges of using this approach in HTA. We describe the basic principles of trial emulation, outline some areas where TTE using RWD can help complement RCT evidence in HTA, identify potential barriers to its adoption in the HTA setting and highlight some priorities for future work.

What makes administrative data "research-ready"? A systematic review and thematic analysis of published literature

Introduction

Administrative data are a valuable research resource, but are under-utilised in the UK due to governance, technical and other barriers (e.g., the time and effort taken to gain secure data access). In recent years, there has been considerable government investment in making administrative data "research-ready", but there is no definition of what this term means. A common understanding of what constitutes research-ready administrative data is needed to establish clear principles and frameworks for their development and the realisation of their full research potential.

Objective

To define the characteristics of research-ready administrative data based on a systematic review and synthesis of existing literature.

Methods

On 29th June 2021, we systematically searched seven electronic databases for (1) peer-reviewed literature (2) related to research-ready administrative data (3) written in the English language. Following supplementary searches and snowball screening, we conducted a thematic analysis of the identified relevant literature.

Results

Overall, we screened 2,375 records and identified 38 relevant studies published between 2012 and 2021. Most related to administrative data from the UK and US and particularly to health data. The term research-ready was used inconsistently in the literature and there was some conflation with the concept of data being ready for statistical analysis. From the thematic analysis, we identified five defining characteristics of research-ready administrative data: (a) accessible, (b) broad, (c) curated, (d) documented and (e) enhanced for research purposes.

Conclusions

Our proposed characteristics of research-ready administrative data could act as a starting point to help data owners and researchers develop common principles and standards. In the more immediate term, the proposed characteristics are a useful framework for cataloguing existing research-ready administrative databases and relevant resources that can support their development.

Long-Term Cardiovascular Risk and Management of Patients Recorded in Primary Care With Unattributed Chest Pain: An Electronic Health Record Study

Background Most adults presenting with chest pain will not receive a diagnosis and be recorded with unattributed chest pain. The objective was to assess if they have increased risk of cardiovascular disease compared with those with noncoronary chest pain and determine whether investigations and interventions are targeted at those at highest risk. Methods and Results We used records from general practices in England linked to hospitalization and mortality information. The study population included patients aged 18 years or over with a new record of chest pain with a noncoronary cause or unattributed between 2002 and 2018, and no cardiovascular disease recorded up to 6 months (diagnostic window) afterward. We compared risk of a future cardiovascular event by type of chest pain, adjusting for cardiovascular risk factors and alternative explanations for chest pain. We determined prevalence of cardiac diagnostic investigations and preventative medication during the diagnostic window in patients with estimated cardiovascular risk ≥10%. There were 375 240 patients with unattributed chest pain (245 329 noncoronary chest pain). There was an increased risk of cardiovascular events for patients with unattributed chest pain, highest in the first year (hazard ratio, 1.25 [95% CI, 1.21-1.29]), persistent up to 10 years. Patients with unattributed chest pain had consistently increased risk of myocardial infarction over time but no increased risk of stroke. Thirty percent of patients at higher risk were prescribed lipid-lowering medication. Conclusions Patients presenting to primary care with unattributed chest pain are at increased risk of cardiovascular events. Primary prevention to reduce cardiovascular events appears suboptimal in those at higher risk.

Cardiovascular risk prediction in type 2 diabetes: a comparison of 22 risk scores in primary care settings

AIMS/HYPOTHESIS

We aimed to compare the performance of risk prediction scores for CVD (i.e., coronary heart disease and stroke), and a broader definition of CVD including atrial fibrillation and heart failure (CVD+), in individuals with type 2 diabetes.

METHODS

Scores were identified through a literature review and were included irrespective of the type of predicted cardiovascular outcome or the inclusion of individuals with type 2 diabetes. Performance was assessed in a contemporary, representative sample of 168,871 UK-based individuals with type 2 diabetes (age ≥18 years without pre-existing CVD+). Missing observations were addressed using multiple imputation.

RESULTS

We evaluated 22 scores: 13 derived in the general population and nine in individuals with type 2 diabetes. The Systemic Coronary Risk Evaluation (SCORE) CVD rule derived in the general population performed best for both CVD (C statistic 0.67 [95% CI 0.67, 0.67]) and CVD+ (C statistic 0.69 [95% CI 0.69, 0.70]). The C statistic of the remaining scores ranged from 0.62 to 0.67 for CVD, and from 0.64 to 0.69 for CVD+. Calibration slopes (1 indicates perfect calibration) ranged from 0.38 (95% CI 0.37, 0.39) to 0.74 (95% CI 0.72, 0.76) for CVD, and from 0.41 (95% CI 0.40, 0.42) to 0.88 (95% CI 0.86, 0.90) for CVD+. A simple recalibration process considerably improved the performance of the scores, with calibration slopes now ranging between 0.96 and 1.04 for CVD. Scores with more predictors did not outperform scores with fewer predictors: for CVD+, QRISK3 (19 variables) had a C statistic of 0.68 (95% CI 0.68, 0.69), compared with SCORE CVD (six variables) which had a C statistic of 0.69 (95% CI 0.69, 0.70). Scores specific to individuals with diabetes did not discriminate better than scores derived in the general population: the UK Prospective Diabetes Study (UKPDS) scores performed significantly worse than SCORE CVD (p value <0.001).

CONCLUSIONS/INTERPRETATION

CVD risk prediction scores could not accurately identify individuals with type 2 diabetes who experienced a CVD event in the 10 years of follow-up. All 22 evaluated models had a comparable and modest discriminative ability.

Development and Validation of a Primary Care Electronic Health Record Phenotype to Study Migration and Health in the UK

International migrants comprised 14% of the UK's population in 2020; however, their health is rarely studied at a population level using primary care electronic health records due to difficulties in their identification. We developed a migration phenotype using country of birth, visa status, non-English main/first language and non-UK-origin codes and applied it to the Clinical Practice Research Datalink (CPRD) GOLD database of 16,071,111 primary care patients between 1997 and 2018. We compared the completeness and representativeness of the identified migrant population to Office for National Statistics (ONS) country-of-birth and 2011 census data by year, age, sex, geographic region of birth and ethnicity. Between 1997 to 2018, 403,768 migrants (2.51% of the CPRD GOLD population) were identified: 178,749 (1.11%) had foreign-country-of-birth or visa -status codes, 216,731 (1.35%) non-English-main/first-language codes, and 8288 (0.05%) non-UK-origin codes. The cohort was similarly distributed versus ONS data by sex and region of birth. Migration recording improved over time and younger migrants were better represented than those aged ≥50. The validated phenotype identified a large migrant cohort for use in migration health research in CPRD GOLD to inform healthcare policy and practice. The under-recording of migration status in earlier years and older ages necessitates cautious interpretation of future studies in these groups.

Assortative mating and within-spouse pair comparisons

Spousal comparisons have been proposed as a design that can both reduce confounding and estimate effects of the shared adulthood environment. However, assortative mating, the process by which individuals select phenotypically (dis)similar mates, could distort associations when comparing spouses. We evaluated the use of spousal comparisons, as in the within-spouse pair (WSP) model, for aetiological research such as genetic association studies. We demonstrated that the WSP model can reduce confounding but may be susceptible to collider bias arising from conditioning on assorted spouse pairs. Analyses using UK Biobank spouse pairs found that WSP genetic association estimates were smaller than estimates from random pairs for height, educational attainment, and BMI variants. Within-sibling pair estimates, robust to demographic and parental effects, were also smaller than random pair estimates for height and educational attainment, but not for BMI. WSP models, like other within-family models, may reduce confounding from demographic factors in genetic association estimates, and so could be useful for triangulating evidence across study designs to assess the robustness of findings. However, WSP estimates should be interpreted with caution due to potential collider bias.

Association of Enzyme-Inducing Antiseizure Drug Use With Long-term Cardiovascular Disease

Importance

Enzyme-inducing antiseizure medications (eiASMs) have been hypothesized to be associated with long-term risks of cardiovascular disease.

Objective

To quantify and model the putative hazard of cardiovascular disease secondary to eiASM use.

Design, Setting, and Participants

This cohort study covered January 1990 to March 2019 (median [IQR] follow-up, 9 [4-15], years). The study linked primary care and hospital electronic health records at National Health Service hospitals in England. People aged 18 years or older diagnosed as having epilepsy after January 1, 1990, were included. All eligible patients were included with a waiver of consent. No patients were approached who withdrew consent. Analysis began January 2021 and ended August 2021.

Exposures

Receipt of 4 consecutive eiASMs (carbamazepine, eslicarbazepine, oxcarbazepine, phenobarbital, phenytoin, primidone, rufinamide, or topiramate) following an adult-onset (age ≥18 years) epilepsy diagnosis or repeated exposure in a weighted cumulative exposure model.

Main Outcomes and Measures

Three cohorts were isolated, 1 of which comprised all adults meeting a case definition for epilepsy diagnosed after 1990, 1 comprised incident cases diagnosed after 1998 (hospital linkage date), and 1 was limited to adults diagnosed with epilepsy at 65 years or older. Outcome was incident cardiovascular disease (ischemic heart disease or ischemic or hemorrhagic stroke). Hazard of incident cardiovascular disease was evaluated using adjusted propensity-matched survival analyses and weighted cumulative exposure models.

Results

Of 10 916 166 adults, 50 888 (0.6%) were identified as having period-prevalent cases (median [IQR] age, 32 [19-50] years; 16 584 [53%] female), of whom 31 479 (62%) were diagnosed on or after 1990 and were free of cardiovascular disease at baseline. In a propensity-matched Cox proportional hazards model adjusted for age, sex, baseline socioeconomic status, and cardiovascular risk factors, the hazard ratio for incident cardiovascular disease was 1.21 (95% CI, 1.06-1.39) for those receiving eiASMs. The absolute difference in cumulative hazard diverges by more than 1% and greater after 10 years. For those with persistent exposure beyond 4 prescriptions, the median hazard ratio increased from amedian (IQR) of 1.54 (1.28-1.79) when taking a relative defined daily dose of an eiASM of 1 to 2.38 (1.52-3.56) with a relative defined daily dose of 2 throughout a maximum of 25 years' follow-up compared with those not receiving an eiASM. The hazard was elevated but attenuated when restricting analyses to incident cases or those diagnosed when older than 65 years.

Conclusions and Relevance

The hazard of incident cardiovascular disease is higher in those receiving eiASMs. The association is dose dependent and the absolute difference in hazard seems to reach clinical significance by approximately 10 years from first exposure.

Internal-external cross-validation helped to evaluate the generalizability of prediction models in large clustered datasets

OBJECTIVE

To illustrate how to evaluate the need of complex strategies for developing generalizable prediction models in large clustered datasets.

STUDY DESIGN AND SETTING

We developed eight Cox regression models to estimate the risk of heart failure using a large population-level dataset. These models differed in the number of predictors, the functional form of the predictor effects (non-linear effects and interaction) and the estimation method (maximum likelihood and penalization). Internal-external cross-validation was used to evaluate the models' generalizability across the included general practices.

RESULTS

Among 871,687 individuals from 225 general practices, 43,987 (5.5%) developed heart failure during a median follow-up time of 5.8 years. For discrimination, the simplest prediction model yielded a good concordance statistic, which was not much improved by adopting complex strategies. Between-practice heterogeneity in discrimination was similar in all models. For calibration, the simplest model performed satisfactorily. Although accounting for non-linear effects and interaction slightly improved the calibration slope, it also led to more heterogeneity in the observed/expected ratio. Similar results were found in a second case study involving patients with stroke.

CONCLUSION

In large clustered datasets, prediction model studies may adopt internal-external cross-validation to evaluate the generalizability of competing models, and to identify promising modelling strategies.

Multimorbid disease trajectories for people with periodontitis

AIM

Periodontitis is a multifactorial condition linked to increased risk of systemic diseases. This study aimed to identify disease trajectories of people with periodontitis using the process mining technique as a heuristic approach.

MATERIALS AND METHODS

A total of 188,863 participants from the UK Biobank cohort were included. Self-reported oral health indicators (bleeding gums, painful gums, loose teeth) were surrogates for periodontitis at baseline. Systemic disease diagnoses and dates formed the process mining event log. Relative risk (RR) of systemic diseases, disease trajectories, and Cox proportional hazard ratio models for mortality were compared to age- and sex-matched controls who did not report a history of periodontitis.

RESULTS

Participants with loose teeth had shorter median time to most systemic diseases, and crude RR was increased for several diseases including cardiovascular disease (crude RR: 1.15, 95% CI: 1.03-1.28), hypertension (crude RR: 1.14, 95% CI: 1.05-1.24), and depression (crude RR: 1.33, 95% CI: 1.09-1.61). Participants with loose teeth had increased RR for 20 disease trajectories, though these were not significant after adjustments. Participants with bleeding/painful gums had similar disease trajectories as those of matched controls.

CONCLUSIONS

Self-reported periodontitis may be associated with early and frequent multimorbidity development, though further evidence is required to confirm this hypothesis. People with periodontitis should be informed of the risks of disease progression and be targeted in prevention initiatives.

Additive Effect of Periodontal Disease and Obesity on COVID-19 Outcomes

This study aims to examine the impact of periodontal disease in obesity on COVID-19 infection and associated outcomes. This retrospective longitudinal study included 58,897 UK Biobank participants tested for COVID-19 between March 2020 and February 2021. Self-reported oral health indicators (bleeding gums, painful gums, and loose teeth) were used as surrogates for periodontal disease. Body fat levels were quantified by body mass index (BMI) and categorized as normal weight (18.5 to 24.9 kg/m²), overweight (25 to 29.9 kg/m²), and obese (≥30 kg/m²). Multivariable logistic regression and Cox proportional hazard models were used to quantify risk of COVID-19 infection, hospital admission, and mortality, adjusted for participants’ demographics and covariates. Of 58,897 participants, 14,466 (24.6%) tested positive for COVID-19 infection. COVID-19 infection was higher for participants who were overweight (odds ratio, 1.18; 95% CI, 1.12 to 1.24) and obese (odds ratio, 1.33; 95% CI, 1.26 to 1.41) as compared with those of normal weight, but infection was not affected by periodontal disease. The hospital admission rate was 57% higher (hazard ratio, 1.57; 95% CI, 1.25 to 1.97) in the obese group with periodontal disease than without periodontal disease, and admission rates increased with BMI category (normal weight, 4.4%; overweight, 6.8%; obese, 10.1%). Mortality rates also increased with BMI category (normal weight, 1.9%; overweight, 3.17%; obese, 4.5%). In addition, for participants with obesity, the mortality rate was much higher (hazard ratio, 3.11; 95% CI, 1.91 to 5.06) in participants with periodontal disease than those without. Obesity is associated with higher hospitalization and mortality rates, and periodontal disease may exacerbate this impact. The results could inform health providers, policy makers, and the general public of the importance to maintain good oral health through seamless provision of dental services and public oral health prevention initiatives.

Risk factors for dementia development, frailty, and mortality in older adults with epilepsy - A population-based analysis

OBJECTIVE

Although the prevalence of comorbid epilepsy and dementia is expected to increase, the impact is not well understood. Our objectives were to examine risk factors associated with incident dementia and the impact of frailty and dementia on mortality in older adults with epilepsy.

METHODS

The CALIBER scientific platform was used. People with incident epilepsy at or after age 65 were identified using Read codes and matched by age, sex, and general practitioner to a cohort without epilepsy (10:1). Baseline cohort characteristics were compared using conditional logistic regression models. Multivariate Cox proportional hazard regression models were used to examine the impact of frailty and dementia on mortality, and to assess risk factors for dementia development.

RESULTS

One thousand forty eight older adults with incident epilepsy were identified. The odds of having dementia at baseline were 7.39 [95% CI 5.21-10.50] times higher in older adults with epilepsy (n = 62, 5.92%) compared to older adults without epilepsy (n = 88, 0.86%). In the final multivariate Cox model (n = 326), age [HR: 1.20, 95% CI 1.09-1.32], Charlson comorbidity index score [HR: 1.26, 95% CI 1.10-1.44], and sleep disturbances [HR: 2.41, 95% CI 1.07-5.43] at baseline epilepsy diagnosis were significantly associated with an increased hazard of dementia development over the follow-up period. In a multivariate Cox model (n = 1047), age [HR: 1.07, 95% CI 1.03-1.11], baseline dementia [HR: 2.66, 95% CI 1.65-4.27] and baseline e-frailty index score [HR: 11.55, 95% CI 2.09-63.84] were significantly associated with a higher hazard of death among those with epilepsy. Female sex [HR: 0.77, 95% CI 0.59-0.99] was associated with a lower hazard of death.

SIGNIFICANCE

The odds of having dementia were higher in older adults with incident epilepsy. A higher comorbidity burden acts as a risk factor for dementia, while prevalent dementia and increasing frailty were associated with mortality.

Ethnicity-specific BMI cutoffs for obesity based on type 2 diabetes risk in England: a population-based cohort study

BACKGROUND

National and global recommendations for BMI cutoffs to trigger action to prevent obesity-related complications like type 2 diabetes among non-White populations are questionable. We aimed to prospectively identify ethnicity-specific BMI cutoffs for obesity based on the risk of type 2 diabetes that are risk-equivalent to the BMI cutoff for obesity among White populations (≥30 kg/m2).

METHODS

In this population-based cohort study, we used electronic health records across primary care (Clinical Practice Research Datalink) linked to secondary care records (Hospital Episodes Statistics) from a network of general practitioner practices in England. Eligible participants were aged 18 years or older, without any past or current diagnosis of type 2 diabetes, had a BMI of 15·0-50·0 kg/m2 and complete ethnicity data, were registered with a general practitioner practice in England at any point between Sept 1, 1990, and Dec 1, 2018, and had at least 1 year of follow-up data. Patients with type 2 diabetes were identified by use of a CALIBER phenotyping algorithm. Self-reported ethnicity was collapsed into five main categories. Age-adjusted and sex-adjusted negative binomial regression models, with fractional polynomials for BMI, were fitted with incident type 2 diabetes and ethnicity data.

FINDINGS

1 472 819 people were included in our study, of whom 1 333 816 (90·6%) were White, 75 956 (5·2%) were south Asian, 49 349 (3·4%) were Black, 10 934 (0·7%) were Chinese, and 2764 (0·2%) were Arab. After a median follow-up of 6·5 years (IQR 3·2-11·2), 97 823 (6·6%) of 1 472 819 individuals were diagnosed with type 2 diabetes. For the equivalent age-adjusted and sex-adjusted incidence of type 2 diabetes at a BMI of 30·0 kg/m2 in White populations, the BMI cutoffs were 23·9 kg/m2 (95% CI 23·6-24·0) in south Asian populations, 28·1 kg/m2 (28·0-28·4) in Black populations, 26·9 kg/m2 (26·7-27·2) in Chinese populations, and 26·6 kg/m2 (26·5-27·0) in Arab populations.

INTERPRETATION

Revisions of ethnicity-specific BMI cutoffs are needed to ensure that minority ethnic populations are provided with appropriate clinical surveillance to optimise the prevention, early diagnosis, and timely management of type 2 diabetes.

FUNDING

National Institute for Health Research.

Transforming and evaluating electronic health record disease phenotyping algorithms using the OMOP common data model: a case study in heart failure

OBJECTIVE

The aim of the study was to transform a resource of linked electronic health records (EHR) to the OMOP common data model (CDM) and evaluate the process in terms of syntactic and semantic consistency and quality when implementing disease and risk factor phenotyping algorithms.

MATERIALS AND METHODS

Using heart failure (HF) as an exemplar, we represented three national EHR sources (Clinical Practice Research Datalink, Hospital Episode Statistics Admitted Patient Care, Office for National Statistics) into the OMOP CDM 5.2. We compared the original and CDM HF patient population by calculating and presenting descriptive statistics of demographics, related comorbidities, and relevant clinical biomarkers.

RESULTS

We identified a cohort of 502 536 patients with the incident and prevalent HF and converted 1 099 195 384 rows of data from 216 581 914 encounters across three EHR sources to the OMOP CDM. The largest percentage (65%) of unmapped events was related to medication prescriptions in primary care. The average coverage of source vocabularies was >98% with the exception of laboratory tests recorded in primary care. The raw and transformed data were similar in terms of demographics and comorbidities with the largest difference observed being 3.78% in the prevalence of chronic obstructive pulmonary disease (COPD).

CONCLUSION

Our study demonstrated that the OMOP CDM can successfully be applied to convert EHR linked across multiple healthcare settings and represent phenotyping algorithms spanning multiple sources. Similar to previous research, challenges mapping primary care prescriptions and laboratory measurements still persist and require further work. The use of OMOP CDM in national UK EHR is a valuable research tool that can enable large-scale reproducible observational research.

Concept libraries for automatic electronic health record based phenotyping: A review

Introduction

Electronic health records (EHR) are linked together to examine disease history and to undertake research into the causes and outcomes of disease. However, the process of constructing algorithms for phenotyping (e.g., identifying disease characteristics) or health characteristics (e.g., smoker) is very time consuming and resource costly. In addition, results can vary greatly between researchers. Reusing or building on algorithms that others have created is a compelling solution to these problems. However, sharing algorithms is not a common practice and many published studies do not detail the clinical code lists used by the researchers in the disease/characteristic definition. To address these challenges, a number of centres across the world have developed health data portals which contain concept libraries (e.g., algorithms for defining concepts such as disease and characteristics) in order to facilitate disease phenotyping and health studies.

Objectives

This study aims to review the literature of existing concept libraries, examine their utilities, identify the current gaps, and suggest future developments.

Methods

The five-stage framework of Arksey and O'Malley was used for the literature search. This approach included defining the research questions, identifying relevant studies through literature review, selecting eligible studies, charting and extracting data, and summarising and reporting the findings.

Results

This review identified seven publicly accessible Electronic Health data concept libraries which were developed in different countries including UK, USA, and Canada. The concept libraries (n = 7) investigated were either general libraries that hold phenotypes of multiple specialties (n = 4) or specialized libraries that manage only certain specialities such as rare diseases (n = 3). There were some clear differences between the general libraries such as archiving data from different electronic sources, and using a range of different types of coding systems. However, they share some clear similarities such as enabling users to upload their own code lists, and allowing users to use/download the publicly accessible code. In addition, there were some differences between the specialized libraries such as difference in ability to search, and if it was possible to use different searching queries such as simple or complex searches. Conversely, there were some similarities between the specialized libraries such as enabling users to upload their own concepts into the libraries and to show where they were published, which facilitates assessing the validity of the concepts. All the specialized libraries aimed to encourage the reuse of research methods such as lists of clinical code and/or metadata.

Conclusion

The seven libraries identified have been developed independently and appear to replicate similar concepts but in different ways. Collaboration between similar libraries would greatly facilitate the use of these libraries for the user. The process of building code lists takes time and effort. Access to existing code lists increases consistency and accuracy of definitions across studies. Concept library developers should collaborate with each other to raise awareness of their existence and of their various functions, which could increase users’ contributions to those libraries and promote their wide-ranging adoption.

Healthcare use by people who use illicit opioids (HUPIO): development of a cohort based on electronic primary care records in England

Background: People who use illicit opioids such as heroin have substantial health needs, but there are few longitudinal studies of general health and healthcare in this population. Most research to date has focused on a narrow set of outcomes, including overdoses and HIV or hepatitis infections. We developed and validated a cohort using UK primary care electronic health records (Clinical Practice Research Datalink GOLD and AURUM databases) to facilitate research into healthcare use by people who use illicit opioid use (HUPIO). Methods: Participants are patients in England with primary care records indicating a history of illicit opioid use. We identified codes including prescriptions of opioid agonist therapies (methadone and buprenorphine) and clinical observations such as 'heroin dependence'. We constructed a cohort of patients with at least one of these codes and aged 18-64 at cohort entry, with follow-up between January 1997 and March 2020. We validated the cohort by comparing patient characteristics and mortality rates to other cohorts of people who use illicit opioids, with different recruitment methods. Results: Up to March 2020, the HUPIO cohort included 138,761 patients with a history of illicit opioid use. Demographic characteristics and all-cause mortality were similar to existing cohorts: 69% were male; the median age at index for patients in CPRD AURUM (the database with more included participants) was 35.3 (interquartile range 29.1-42.6); the average age of new cohort entrants increased over time; 76% had records indicating current tobacco smoking; patients disproportionately lived in deprived neighbourhoods; and all-cause mortality risk was 6.6 (95% CI 6.5-6.7) times the general population of England. Conclusions: Primary care data offer new opportunities to study holistic health outcomes and healthcare of this population. The large sample enables investigation of rare outcomes, whilst the availability of linkage to external datasets allows investigation of hospital use, cancer treatment, and mortality.

Development and validation of phenotype classifiers across multiple sites in the observational health data sciences and informatics network

Objective

Accurate electronic phenotyping is essential to support collaborative observational research. Supervised machine learning methods can be used to train phenotype classifiers in a high-throughput manner using imperfectly labeled data. We developed 10 phenotype classifiers using this approach and evaluated performance across multiple sites within the Observational Health Data Sciences and Informatics (OHDSI) network.

Materials and Methods

We constructed classifiers using the Automated PHenotype Routine for Observational Definition, Identification, Training and Evaluation (APHRODITE) R-package, an open-source framework for learning phenotype classifiers using datasets in the Observational Medical Outcomes Partnership Common Data Model. We labeled training data based on the presence of multiple mentions of disease-specific codes. Performance was evaluated on cohorts derived using rule-based definitions and real-world disease prevalence. Classifiers were developed and evaluated across 3 medical centers, including 1 international site.

Results

Compared to the multiple mentions labeling heuristic, classifiers showed a mean recall boost of 0.43 with a mean precision loss of 0.17. Performance decreased slightly when classifiers were shared across medical centers, with mean recall and precision decreasing by 0.08 and 0.01, respectively, at a site within the USA, and by 0.18 and 0.10, respectively, at an international site.

Discussion and Conclusion

We demonstrate a high-throughput pipeline for constructing and sharing phenotype classifiers across sites within the OHDSI network using APHRODITE. Classifiers exhibit good portability between sites within the USA, however limited portability internationally, indicating that classifier generalizability may have geographic limitations, and, consequently, sharing the classifier-building recipe, rather than the pretrained classifiers, may be more useful for facilitating collaborative observational research.

Natural Language Processing and Machine Learning for Identifying Incident Stroke From Electronic Health Records: Algorithm Development and Validation

Background

Stroke is an important clinical outcome in cardiovascular research. However, the ascertainment of incident stroke is typically accomplished via time-consuming manual chart abstraction. Current phenotyping efforts using electronic health records for stroke focus on case ascertainment rather than incident disease, which requires knowledge of the temporal sequence of events.

Objective

The aim of this study was to develop a machine learning–based phenotyping algorithm for incident stroke ascertainment based on diagnosis codes, procedure codes, and clinical concepts extracted from clinical notes using natural language processing.

Methods

The algorithm was trained and validated using an existing epidemiology cohort consisting of 4914 patients with atrial fibrillation (AF) with manually curated incident stroke events. Various combinations of feature sets and machine learning classifiers were compared. Using a heuristic rule based on the composition of concepts and codes, we further detected the stroke subtype (ischemic stroke/transient ischemic attack or hemorrhagic stroke) of each identified stroke. The algorithm was further validated using a cohort (n=150) stratified sampled from a population in Olmsted County, Minnesota (N=74,314).

Results

Among the 4914 patients with AF, 740 had validated incident stroke events. The best-performing stroke phenotyping algorithm used clinical concepts, diagnosis codes, and procedure codes as features in a random forest classifier. Among patients with stroke codes in the general population sample, the best-performing model achieved a positive predictive value of 86% (43/50; 95% CI 0.74-0.93) and a negative predictive value of 96% (96/100). For subtype identification, we achieved an accuracy of 83% in the AF cohort and 80% in the general population sample.

Conclusions

We developed and validated a machine learning–based algorithm that performed well for identifying incident stroke and for determining type of stroke. The algorithm also performed well on a sample from a general population, further demonstrating its generalizability and potential for adoption by other institutions.

Temporal trends in heart failure medication prescription in a population-based cohort study

OBJECTIVE

We examined temporal heart failure (HF) prescription patterns in a large representative sample of real-world patients in the UK, using electronic health records (EHR).

METHODS

From primary and secondary care EHR, we identified 85 732 patients with a HF diagnosis between 2002 and 2015. Almost 50% of patients with HF were women and the median age was 79.1 (IQR 70.2-85.7) years, with age at diagnosis increasing over time.

RESULTS

We found several trends in pharmacological HF management, including increased beta blocker prescriptions over time (29% in 2002-2005 and 54% in 2013-2015), which was not observed for mineralocorticoid receptor-antagonists (MR-antagonists) (18% in 2002-2005 and 18% in 2013-2015); higher prescription rates of loop diuretics in women and elderly patients together with lower prescription rates of angiotensin-converting enzyme inhibitors and/or angiotensin II receptor blockers, beta blockers or MR-antagonists in these patients; little change in medication prescription rates occurred after 6 months of HF diagnosis and, finally, patients hospitalised for HF who had no recorded follow-up in primary care had considerably lower prescription rates compared with patients with a HF diagnosis in primary care with or without HF hospitalisation.

CONCLUSION

In the general population, the use of MR-antagonists for HF remained low and did not change throughout 13 years of follow-up. For most patients, few changes were seen in pharmacological management of HF in the 6 months following diagnosis.

Data-driven identification of ageing-related diseases from electronic health records

Reducing the burden of late-life morbidity requires an understanding of the mechanisms of ageing-related diseases (ARDs), defined as diseases that accumulate with increasing age. This has been hampered by the lack of formal criteria to identify ARDs. Here, we present a framework to identify ARDs using two complementary methods consisting of unsupervised machine learning and actuarial techniques, which we applied to electronic health records (EHRs) from 3,009,048 individuals in England using primary care data from the Clinical Practice Research Datalink (CPRD) linked to the Hospital Episode Statistics admitted patient care dataset between 1 April 2010 and 31 March 2015 (mean age 49.7 years (s.d. 18.6), 51% female, 70% white ethnicity). We grouped 278 high-burden diseases into nine main clusters according to their patterns of disease onset, using a hierarchical agglomerative clustering algorithm. Four of these clusters, encompassing 207 diseases spanning diverse organ systems and clinical specialties, had rates of disease onset that clearly increased with chronological age. However, the ages of onset for these four clusters were strikingly different, with median age of onset 82 years (IQR 82–83) for Cluster 1, 77 years (IQR 75–77) for Cluster 2, 69 years (IQR 66–71) for Cluster 3 and 57 years (IQR 54–59) for Cluster 4. Fitting to ageing-related actuarial models confirmed that the vast majority of these 207 diseases had a high probability of being ageing-related. Cardiovascular diseases and cancers were highly represented, while benign neoplastic, skin and psychiatric conditions were largely absent from the four ageing-related clusters. Our framework identifies and clusters ARDs and can form the basis for fundamental and translational research into ageing pathways.