Publications
scientifiques

BACKGROUND: Pancreatic cancer is the third leading cause of cancer-related death in the United States. Total serum cholesterol (TSC) may predict cancer risk, although its role independent of statins remains elusive. We examined the association between TSC and pancreatic cancer risk independent of statins. METHODS: A nested case-control analysis was conducted among statin-naïve patients within The Health Improvement Network (THIN), a United Kingdom-based general practice database. Cases were >40 years old and diagnosed with pancreatic cancer after at least 6 months of follow-up. Controls were selected by incidence density sampling and matched by age, sex, practice site, and follow-up. Primary exposure was TSC (mmol/L) prior to index date. Conditional logistic regression estimated ORs for pancreatic cancer risk associated with TSC. Sensitivity analyses were conducted among nondiabetics. RESULTS: Among 1,241 cases and 3,307 matched controls, an average 8% reduction was observed in pancreatic cancer risk per mmol/L increase in TSC [OR 0.92, 95% confidence interval (CI): 0.85-1.00; nondiabetics: OR 0.91, 95% CI: 0.83-0.99]. When TSC was measured at 12-month intervals prior to diagnosis, the OR between TSC and pancreatic cancer was 0.88 at 0 to 12 months (95% CI: 0.77-1.00; nondiabetics: OR 0.81, 95% CI: 0.68-0.96). No significant association was seen at subsequent discrete intervals before index date. CONCLUSIONS: TSC is a significant predictor of short-term risk for pancreatic cancer. This risk increase associated with lower TSC was independent of statins. IMPACT: TSC could serve as a biomarker for risk stratification, screening, and early diagnosis of pancreatic cancer in future clinical prediction models.

PURPOSE: The purpose of the study is to assess the validity of codes or algorithms used to identify dementia in UK electronic health record (EHR) primary care and hospitalisation databases. METHODS: Relevant studies were identified by searching the MEDLINE/EMBASE databases from inception to June 2018, hand-searching reference lists, and consulting experts. The search strategy included synonyms for "Dementia", "Europe", and "EHR". Studies were included if they validated dementia diagnoses in UK primary care or hospitalisation databases, irrespective of validation method used. The Quality Assessment for Diagnostic Accuracy Studies-2 (QUADAS-2) tool was used to assess risk of bias. RESULTS: From 1469 unique records, 14 relevant studies were included. Thirteen validated individual diagnoses against a reference standard, reporting high estimates of validity. Most reported only the positive predictive value (PPV), with estimates ranging between 0.09 and 1.0 and 0.62 and 0.85 in primary care and hospitalisation databases, respectively. One study performed a rate comparison, indicating good generalisability of dementia diagnoses in The Health Improvement Network (THIN) database to the UK population. Studies were of low methodological quality. As studies were not comparable, no summary validity estimates were produced. CONCLUSION: While heterogenous across studies, reported validity estimates were generally high. However, the credibility of these estimates is limited by the methodological quality of studies, primarily resulting from insufficient blinding of researchers interpreting the reference test. Inadequate reporting, particularly of the specific codes validated, hindered comparison of estimates across studies. Future validation studies should make use of more robust reference tests, follow established reporting guidelines, and calculate all measures of validity.

BACKGROUND AND OBJECTIVE: Tendon rupture can result from fluoroquinolone exposure. The objective of this study was to quantify relative and absolute risk and determine how risk is affected by timing of exposure. METHODS: The UK Health Improvement Network primary care database was used to perform a nested case-control study measuring the association between fluoroquinolone exposure and tendon rupture. Adults with tendon rupture were matched on age, sex, general practice and calendar time to four controls selected from a cohort prescribed systemic fluoroquinolone or co-amoxiclav antibiotics. The relative and absolute risk of tendon rupture with fluoroquinolone exposure was calculated. RESULTS: Current fluoroquinolone exposure was associated with an increased risk of any tendon rupture (adjusted incidence rate ratio [aIRR] 1.61, 95% CI 1.25-2.09) and Achilles tendon rupture (aIRR 3.14, 95% CI 2.11-4.65) that persisted for 60 days. Risk increased with cumulative exposure and was greatest when co-prescribed with oral corticosteroids (aIRR 19.36, 95% CI 7.78-48.19 for Achilles tendon rupture). The adjusted rate difference (aRD) with fluoroquinolone exposure was 2.9 and 2.1 per 10,000 patients for any and Achilles tendon rupture, respectively, and was greatest in people aged ≥ 60 years prescribed concomitant oral corticosteroid therapy (aDR 19.6 for any tendon and 6.6 Achilles tendon rupture per 10,000). No association was seen with co-amoxiclav or statin exposure, or with biceps or other tendon ruptures. CONCLUSIONS: Risk of tendon rupture with fluoroquinolones depends on timing, cumulative dose and concomitant exposure to oral corticosteroids. Absolute risk significantly varied by age and concomitant corticosteroid exposure, affecting elderly patients the greatest.

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

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

OBJECTIVE: To evaluate the duration of prescriptions for antibiotic treatment for common infections in English primary care and to compare this with guideline recommendations. DESIGN: Cross sectional study. SETTING: General practices contributing to The Health Improvement Network database, 2013-15. PARTICIPANTS: 931 015 consultations that resulted in an antibiotic prescription for one of several indications: acute sinusitis, acute sore throat, acute cough and bronchitis, pneumonia, acute exacerbation of chronic obstructive pulmonary disease (COPD), acute otitis media, acute cystitis, acute prostatitis, pyelonephritis, cellulitis, impetigo, scarlet fever, and gastroenteritis. MAIN OUTCOME MEASURES: The main outcomes were the percentage of antibiotic prescriptions with a duration exceeding the guideline recommendation and the total number of days beyond the recommended duration for each indication. RESULTS: The most common reasons for antibiotics being prescribed were acute cough and bronchitis (386 972, 41.6% of the included consultations), acute sore throat (239 231, 25.7%), acute otitis media (83 054, 8.9%), and acute sinusitis (76 683, 8.2%). Antibiotic treatments for upper respiratory tract indications and acute cough and bronchitis accounted for more than two thirds of the total prescriptions considered, and 80% or more of these treatment courses exceeded guideline recommendations. Notable exceptions were acute sinusitis, where only 9.6% (95% confidence interval 9.4% to 9.9%) of prescriptions exceeded seven days and acute sore throat where only 2.1% (2.0% to 2.1%) exceeded 10 days (recent guidance recommends five days). More than half of the antibiotic prescriptions were for longer than guidelines recommend for acute cystitis among females (54.6%, 54.1% to 55.0%). The percentage of antibiotic prescriptions exceeding the recommended duration was lower for most non-respiratory infections. For the 931 015 included consultations resulting in antibiotic prescriptions, about 1.3 million days were beyond the durations recommended by guidelines. CONCLUSION: For most common infections treated in primary care, a substantial proportion of antibiotic prescriptions have durations exceeding those recommended in guidelines. Substantial reductions in antibiotic exposure can be accomplished by aligning antibiotic prescription durations with guidelines.

BACKGROUND: IgA vasculitis (IgAV, Henoch-Schönlein purpura) is a small-vessel vasculitis most common in children but also occurring in adults. Case series have suggested that IgAV may be associated with cardiovascular disease and venous thromboembolism, but this has not been evaluated in population-based studies. Renal disease and hypertension are possible complications of the disease with unknown incidence. METHODS: Using a large UK primary care database, we conducted an open retrospective matched cohort study of cardiovascular, venous thrombotic and renal outcomes in adult-onset and childhood-onset IgAV. Control participants were selected at a 2:1 ratio, matched for age and sex. Adjusted HRs (aHRs) were calculated using Cox proportional hazards models. RESULTS: 2828 patients with adult-onset IgAV and 10 405 patients with childhood-onset IgAV were compared with age-matched and sex-matched controls. There was significantly increased risk of hypertension (adult-onset aHR 1.42, 95% CI 1.19 to 1.70, p < 0.001; childhood-onset aHR 1.52, 95% CI 1.22 to 1.89, p < 0.001) and stage G3-G5 chronic kidney disease (adult-onset aHR 1.54, 95% CI 1.23 to 1.93, p < 0.001; childhood-onset aHR 1.89, 95% CI 1.16 to 3.07, p=0.010). There was no evidence of association with ischaemic heart disease, cerebrovascular disease or venous thromboembolism. All-cause mortality was increased in the adult-onset IgAV cohort compared with controls (aHR 1.27, 95% CI 1.07 to 1.50, p=0.006). CONCLUSIONS: Patients with IgAV are at increased risk of hypertension and chronic kidney disease (CKD) compared with individuals without IgAV; analysis restricted to adult-onset IgAV patients showed increased mortality. Appropriate surveillance and risk factor modification could improve long-term outcomes in these patients.

BACKGROUND: The incidence of hepatobiliary cancer is steadily increasing. It is unclear if this rise is related to increasing trends in obesity, metabolic syndrome, and lifestyle changes. METHODS: A case-control study was performed using the Health Improvement Network (THIN) database. Cases with a diagnosis of liver, bile duct, and gallbladder cancers were matched in a 1:2 fashion with controls and analyzed for potential associations between hepatobiliary cancer and obesity/metabolic syndrome. RESULTS: Four thousand two hundred and eighty-seven patients (62% male, 38% female) with hepatobiliary cancers were matched with 8574 controls. On univariate analysis, body mass index (BMI), smoking, diabetes, alcohol consumption, ischemic heart disease, and hypertension were associated with hepatobiliary cancer. Statin use and non-smoking status had an inverse association. On multivariate analysis, BMI, diabetes, hypertension, ischemic heart disease, and insulin use were associated with the risk of hepatobiliary cancer. Statin use and non-smoking status were protective. On modeling BMI, each of diabetes and hypertension as a single covariate, there was a significant association with hepatobiliary cancer (1.59 [1.49-1.69], p < 0.001) which persisted despite adjusting for increasing age (1.006 [1005-1.006], p < 0.001) and background liver cirrhosis (1.037 [1.03-1.044], p < 0.001). CONCLUSIONS: Obesity and metabolic syndrome are associated with the risk of hepatobiliary cancer. Statin use seems to be protective.

PURPOSE: The purpose of the study is to evaluate the impact of validation on the identification of major bleeding events in The Health Improvement Network (THIN) database in patients receiving anticoagulant therapy. METHODS: Patients aged 2 to 89 years with a first prescription for an anticoagulant (rivaroxaban or warfarin) between 2012 and 2015 were identified in THIN. Major bleeding events, defined as bleeding events necessitating hospitalization or referral to accident and emergency services or a specialist clinic, were identified using a 2-step ascertainment process based on read codes only, and then validated using a 2-step process requiring manual review of patients' records. RESULTS: The positive predictive value for the ascertainment of major intracranial (IC) bleeds using only read codes was 96.9%, compared with 70.4% for gastrointestinal (GI) bleeds and 64.1% for urogenital (UG) bleeds. The incidence rate of major IC bleeding events was therefore similar when it was calculated before and after validation (0.32 per 100 person-years and 0.31 per 100 person-years, respectively). The incidence rate of major GI bleeds identified using read codes alone was reduced following validation from 2.05 to 0.94 per 100 person-years, and that of major UG bleeds decreased from 2.45 to 1.11 per 100 person-years. CONCLUSIONS: Major GI and UG bleeding events ascertained from THIN using read codes require validation using additional information to prevent outcome misclassification. The absence of validation may lead to overestimated incidence rates of major bleeding for GI and UG bleeds.

BACKGROUND: Overt albuminuria (urinary albumin-creatinine ratio [ACR] > 300 mg/g) is an established risk factor for progression of nephropathy and total mortality. However, whether a reduction in ACR translates into a reduction in mortality and/or cardiovascular (CV) events among insulin-treated patients with type 2 diabetes (T2D) in routine practice is currently not known. METHODS: We obtained data on a large cohort of insulin users with T2D and nephropathy (baseline ACR ≥300 mg/g) from UK general practices between 2007 and 2014. Their corresponding ACR values after 1year of follow-up were thereafter categorized into: (1) < 300 mg/g (i.e., albuminuria regression) or (2) > 300 mg/g (i.e., nonregression of albuminuria), and the cohort was followed-up for 5 years for all-cause mortality and CV events. Cox proportional hazard models were fitted to estimate the risk of all-cause death. RESULTS: A total of 11,074 patients with insulin-treated T2D met the inclusion criteria. Their mean age was 62.3 (13.6) years; mean HbA1c: 8.7 (1.8) and 53% were male. A total of 682 deaths occurred after a follow-up period of 43,393 person-years with a mortality rate of 16 per 1,000 person-years. Five-year survival was markedly reduced in the group whose proteinuria persisted or progressed (91 vs. 95%; log-rank p value < 0.001). Compared to patients whose ACR levels remained above 300 mg/g, all-cause mortality and CV events were 31 and 27% lower in those whose albuminuria regressed to < 300 mg/g (adjusted hazard ratio [aHR] 0.69; 95% CI 0.52-0.91; p = 0.008 and aHR 0. 73; 95% CI 0.54-0.98; p = 0.041), respectively. CONCLUSION: In patients with insulin-treated T2D and nephropathy in routine practice, a regression in albuminuria (e.g., via better BP or glycemic control) is associated with a significant reduction in all-cause mortality. Thus, albuminuria is not only simply a risk marker of renal and CV disease but also an independent target for therapy. Albuminuria reduction should be viewed as a goal for renal and CV protection.

OBJECTIVE: Previous studies suggest that androgens have a sexually dimorphic impact on metabolic dysfunction. However, the sex-specific link between circulating androgens and risk of type 2 diabetes mellitus (T2DM) has not been examined in a large scale, longitudinal cohort, a task we undertook in this study. DESIGN: A retrospective cohort study in a UK primary care database. PATIENTS: We included men and women with available serum testosterone and sex hormone-binding globulin (SHBG) results. MEASUREMENTS: We categorized serum concentrations according to clinically relevant cut-off points and calculated crude and adjusted T2DM Incidence Rate Ratios (IRRs and aIRRs). RESULTS: Serum testosterone concentrations were available in 70 541 men and 81 889 women; serum SHBG was available in 15 907 men and 42 034 women. In comparison to a reference cohort with serum testosterone ≥20 nmol/L, men with lower serum testosterone had a significantly increased risk of T2DM, with the highest risk in those with serum testosterone <7 nmol/L (aIRR 2.71, 95% CI 2.34-3.14, P < 0.001). In women, the risk of T2DM started to increase significantly when serum testosterone concentrations exceeded 1.5 nmol/L, with the highest risk in women with serum testosterone ≥3.5 nmol/L (aIRR 1.98, 95% CI 1.55-2.52, P < 0.001). These observations were verified in a continuous rather than categorized analysis. The risk of T2DM increased in men and women with serum SHBG <40 and <50 nmol/L, respectively. CONCLUSIONS/INTERPRETATION: In this longitudinal study, we found sexually dimorphic associations between serum testosterone and risk of incident T2DM. Androgen deficiency and excess should be considered important risk factors for diabetes in men and women, respectively.

OBJECTIVE: Atrial fibrillation (AF) is the most common cardiac arrhythmia and an important risk factor for stroke. Treatment with anticoagulants substantially reduces risk of stroke. Current prevalence and treatment rates of AF in the UK as well as changes in recent years are not known. The aim of this analysis was to determine trends in age-sex specific prevalence and treatment of AF in the UK from 2000 to 2016. METHODS: 17 sequential cross-sectional analyses were carried out between 2000 and 2016 using a large database of electronic primary care records of patients registered with UK general practitioners. These determined the prevalence of patients diagnosed with AF, the stroke risk of those with AF and the proportion of AF patients currently receiving anticoagulants. Stroke risk was assessed using CHA DS -VASc score. RESULTS: Age-sex standardised AF prevalence increased from 2.14% (95% CI 2.11% to 2.17%) in 2000 to 3.29% (95% CI 3.27% to 3.32%) in 2016. Between 2000 and 2016, the proportion of patients with AF prescribed anticoagulants increased from 35.4% (95% CI 34.7% to 36.1%) to 75.5% (95% CI 75.1% to 75.8%) in those with high stroke risk (p for change over time <0.001) and from 32.8% (95% CI 30.5% to 35.2%) to 47.1% (95% CI 45.4% to 48.7%) in those with moderate stroke risk (p<0.001). In patients with low risk of stroke, the proportion decreased from 19.9% (95% CI 17.8% to 22.2%) to 9.7% (95% CI 8.4% to 11.1%) (p<0.001). Anticoagulant prescribing performance varied between practices; in 2016, the proportion of eligible patients treated was 82.9% (95% CI 82.2% to 83.7%) and 62.0% (95% CI 61.0% to 63.0%) in the highest-performing and lowest-performing practice quintiles, respectively. There was poor agreement in individual practice performance over time from 2006 to 2016: linear-weighted κ=0.10 (95% CI 0.02 to 0.19). CONCLUSIONS: From 2000 to 2016, the prevalence of recorded AF has increased in all age groups and both sexes. Anticoagulant treatment of eligible patients with AF has more than doubled, with marked improvements since 2011, alongside a reduction in the use of anticoagulants in ineligible patients with AF.

BACKGROUND:  Many patients on warfarin therapy often present with supratherapeutic international normalized ratio (INR) levels, resulting from the influence of several patient-specific factors, which have been associated with adverse outcomes. OBJECTIVE:  This article aims to identify risk factors for over-anticoagulation (INR levels ≥4) in a cohort of patients taking warfarin. METHODS:  A cohort of warfarin users aged 18 to 85 years from January 2005 to April 2013 was identified in The Health Improvement Network U.K. primary care database (  = 12,506). A random date was assigned to all patients within their eligible person-time (index date), and a nested case-control analysis was performed with individuals presenting a first episode of INR level ≥4 after the index date used as cases (  = 699) and patients with non-supratherapeutic INR values (≤3) as controls (  = 9,798). Using unconditional logistic regression models, odds ratios with 95% confidence intervals were calculated adjusted for potential confounders. Two sensitivity analyses were performed with alternative definitions of over-anticoagulation (INR levels ≥5 or > 3). RESULTS:  Among the factors examined, the strongest predictors of over-anticoagulation were warfarin indication (in particular, valvular atrial fibrillation and valve replacement), renal failure (with the risk increasing steeply with decreasing estimated glomerular filtration rate), cancer, anaemia, respiratory infections treated with antibiotics, chronic obstructive pulmonary disease treated with β2-agonists, polypharmacy (≥10 medications), low socio-economic status and residency in rural areas. Similar results were obtained when supratherapeutic levels were defined as INR ≥5 or, alternatively, as INR > 3. CONCLUSION:  Predictors of supratherapeutic INR levels found in this study might help physicians identify patients where closer INR monitoring is warranted.

BACKGROUND: Neither the incidence of indications for childhood tonsillectomy nor the proportion of tonsillectomies that are evidence-based is known. AIM: To determine the incidence of indications for tonsillectomy in UK children, and the proportion of tonsillectomies meeting evidence-based criteria. DESIGN AND SETTING: A retrospective cohort study of electronic medical records of children aged 0-15 years registered with 739 UK general practices contributing to a research database. METHOD: Children with recorded indications for tonsillectomy were identified from electronic medical records. Evidence-based indications included documented sore throats of sufficient frequency and severity (Paradise criteria); periodic fever, aphthous stomatitis, pharyngitis, and cervical adenitis syndrome (PFAPA); or tonsillar tumour. Other indications were considered non-evidence-based. The numbers of children subsequently undergoing tonsillectomy was then identified. The numbers with evidence-based and non-evidence-based indications for surgery among children who had undergone tonsillectomy were determined. RESULTS: The authors included 1 630 807 children followed up for 7 200 159 person-years between 2005 and 2016. Incidence of evidence-based indications for tonsillectomy was 4.2 per 1000 person years; 13.6% (2144/15 760) underwent tonsillectomy. Incidence of childhood tonsillectomy was 2.5 per 1000 person years; 11.7% (2144/18 281) had evidence-based indications, almost all with Paradise criteria. The proportion of evidence-based tonsillectomies was unchanged over 12 years. Most childhood tonsillectomies followed non-evidence-based indications: five to six sore throats (12.4%) in 1 year, two to four sore throats (44.6%) in 1 year, sleep disordered breathing (12.3%), or obstructive sleep apnoea (3.9%). CONCLUSION: In the UK, few children with evidence-based indications undergo tonsillectomy and seven in eight of those who do (32 500 of 37 000 annually) are unlikely to benefit.

Objective Type 2 diabetes is associated with an increased risk of fracture. Any factor that incrementally increases this risk should be taken into account when individualising treatment. Hypoglycaemia is a common complication of antidiabetes medications and suggested as a risk factor for fractures; yet, its real-life clinical impact is unclear. Design A population-based, retrospective open cohort study using routinely collected data between 1st of January 1995 and 1st of May 2016 in The Health Improvement Network (THIN) database. Methods Patients with type 2 diabetes mellitus with documented hypoglycaemic events were compared to randomly matched patients with type 2 diabetes mellitus without documented hypoglycaemic events matched to exposed patients on age, sex, duration of diabetes and BMI. The primary outcome was any incident fracture. Secondary outcome was incident fragility (osteoporotic) fracture. Results A total of 41 163 patients with type 2 diabetes were included: 14 147 patients in the exposed cohort and 27 016 patients in the unexposed cohort. Patients with a documented hypoglycaemic event were significantly more likely to sustain any fracture compared to patients with no record of hypoglycaemic events: adjusted IRR = 1.20 (95% CI: 1.12-1.30; P < 0.0001). Patients who had a documented hypoglycaemic event were significantly more likely to suffer a fragility fracture compared to controls: adjusted IRR = 1.24 (95% CI: 1.13-1.37; P < 0.0001). Conclusions Hypoglycaemic events are a significant risk factor for fractures in patients with diabetes mellitus. This observation is clinically relevant when individualising targets for glycaemic control and selecting antidiabetic agents.

Background and objectives Glucocorticoids are used to manage adrenal insufficiency (AI). We describe treatments used in the United Kingdom and real-world clinical outcomes for each treatment. Methods We used 2010-2016 primary care data from The Health Improvement Network (THIN). Descriptive analyses were conducted, and differences in variables between patients prescribed immediate-release hydrocortisone (IR HC), prednisolone or modified-release hydrocortisone (MR HC) were assessed using Fisher's exact test. Results Overall, 2648 patients were included: 1912 on IR HC (72%), 691 on prednisolone (26%) and 45 (2%) on MR HC. A total of 1174 (44.3%) had primary and 1150 (43.4%) had secondary AI. Patients on prednisolone were older (P < 0.001) and had a greater history of smoking (292/691, P < 0.001) and CVD (275/691, P < 0.001). Patients on MR HC had more PCOS (3/45, P = 0.001) and diabetes (27/45, P = 0.004). The number of GP visits/patient/year was 6.50 in IR HC, 9.54 in prednisolone and 9.11 in MR HC cohorts. The mean number of A&E visits and inpatient and outpatient hospital admissions ranged from 0.42 to 0.93 visits/patient/year. The mean number of adrenal crises/patient/year was between 0.02 and 0.03 for all cohorts. Conclusion IR HC is most commonly used for the management of AI in the United Kingdom, followed by prednisolone. Few patients receive MR HC. The prednisolone and MR HC cohorts displayed a greater prevalence of vascular risk factors compared with IR HC. The occurrence of AC and primary and secondary resource use were similar between treatment cohorts, and they indicate significant resource utilisation. Improved treatment and management of patients with AI is needed.

BACKGROUND: Epidemiological research on small cell lung cancer (SCLC) is limited and based on cancer registry data. We evaluated the feasibility and validity of using primary care electronic health records (The Health Improvement Network [THIN]) in the UK to identify and characterise SCLC. METHODS: We searched THIN records of individuals aged 18-89 years between 2000 and 2014 for a first diagnostic code suggestive of lung cancer (group 1) or small cell cancer (SCC; group 2) and for text strings among free text comments to identify and characterise incident SCLC cases. We validated our case identification strategy by manual review of patient EHRs, including free text comments, for a random sample of 400 individuals initially detected with a diagnostic code (300 from group 1 and 100 from group 2). RESULTS: Twenty five thousand two hundred fourty one individuals had a code for lung cancer (n = 24,508 [97.1%]) or SCC (733 [2.9%]). Following free-text searches, there were 3530 incident SCLC cases (2956 from group 1; 574 from group 2) corresponding to an incidence rate of 1.01 per 10,000 person-years. In the validation exercise, SCLC confirmation rates were 99% (group 1) and 85% (group 2). Mean age at diagnosis among confirmed cases was 68.5 years; staging information was present in 63.5% of cases of whom 17.8% had limited disease and 82.2% had extensive disease. The majority (84.5%) had a recorded symptom suggestive of lung cancer; chest infection was the most common (18%) followed by cough (15.8%) and chest/abdominal/back pain (15.2%). The first year crude mortality rates was 9.9 per 100 person-months (95% confidence interval [CI] 9.5-10.4), was higher among men and those aged 80 years and above. A total of 144 (37.8%) confirmed cases had metastases recorded. Median survival among the whole study cohort was 7.37 months. CONCLUSIONS: Our SCLC case identification method appears to be valid and could potentially be adapted to identify other cancer types. However, complete characterisation of staging requires information from additional data sources including cancer registries.