Search

Introduction: Policies aiming to prevent ill health and reduce health inequalities need to consider the full complexity of health systems, including environmental determinants. A learning health system that incorporates environmental factors needs healthcare, social care and non‐health data linkage at individual and small‐area levels. Our objective was to establish privacy‐preserving household record linkage for England to ensure person‐level data remain secure and private when linked with data from households or the wider environment. Methods: A stakeholder workshop with participants from our regional health board, together with the regional data processor, and the national data provider. The workshop discussed the risks and benefits of household linkages. This group then co‐designed actionable dataflows between national and local data controllers and processors. Results: A process was defined whereby the Personal Demographics Service, which includes the addresses of all patients of the National Health Service (NHS) in England, was used to match patients to a home identifier, for the time they are recorded as living at that address. Discussions with NHS England resulted in secure and quality‐assured data linkages and a plan to flow these pseudonymised data onwards into regional health boards. Methods were established, including the generation of matching algorithms, transfer processes and information governance approvals. Our collaboration accelerated the development of a new data governance application, facilitating future public health intervention evaluations. Conclusion: These activities have established a secure method for protecting the privacy of NHS patients in England, while allowing linkage of wider environmental data. This enables local health systems to learn from their data and improve health by optimizing non‐health factors. Proportionate governance of health and linked non‐health data is practical in England for incorporating key environmental factors into a learning health system. [ABSTRACT FROM AUTHOR]

Introduction: There is currently no guidance on how to assess the calibration of multistate models used for risk prediction. We introduce several techniques that can be used to produce calibration plots for the transition probabilities of a multistate model, before assessing their performance in the presence of random and independent censoring through a simulation. Methods: We studied pseudo‐values based on the Aalen‐Johansen estimator, binary logistic regression with inverse probability of censoring weights (BLR‐IPCW), and multinomial logistic regression with inverse probability of censoring weights (MLR‐IPCW). The MLR‐IPCW approach results in a calibration scatter plot, providing extra insight about the calibration. We simulated data with varying levels of censoring and evaluated the ability of each method to estimate the calibration curve for a set of predicted transition probabilities. We also developed evaluated the calibration of a model predicting the incidence of cardiovascular disease, type 2 diabetes and chronic kidney disease among a cohort of patients derived from linked primary and secondary healthcare records. Results: The pseudo‐value, BLR‐IPCW, and MLR‐IPCW approaches give unbiased estimates of the calibration curves under random censoring. These methods remained predominately unbiased in the presence of independent censoring, even if the censoring mechanism was strongly associated with the outcome, with bias concentrated in low‐density regions of predicted transition probability. Conclusions: We recommend implementing either the pseudo‐value or BLR‐IPCW approaches to produce a calibration curve, combined with the MLR‐IPCW approach to produce a calibration scatter plot. The methods have been incorporated into the "calibmsm" R package available on CRAN. [ABSTRACT FROM AUTHOR]

Background: Adverse drug reactions (ADRs) are common and a leading cause of injury. However, information on ADR risks of individual medicines is often limited. The aim of this hypothesis‐generating study was to assess the relative importance of ADR‐related and emergency hospital admission for large group of medication classes. Methods: This study was a propensity‐matched case‐control study in English primary care. Data sources were Clinical Practice Research Databank and Aurum with longitudinal, anonymized, patient level electronic health records (EHRs) from English general practices linked to hospital records. Cases aged 65–100 with ADR‐related or emergency hospital admission were matched to up to six controls by age, sex, morbidity and propensity scores for hospital admission risk. Medication groups with systemic administration as listed in the British National Formulary (used by prescribers for medication advice). Prescribing in the 84 days before the index date was assessed. Only medication groups with 50+ cases exposed were analysed. The outcomes of interest were ADR‐related and emergency hospital admissions. Conditional logistic regression estimated odds ratios (ORs) and 95% confidence intervals (CI). Results: The overall population included 121 546 cases with an ADR‐related and 849 769 cases with emergency hospital admission. The percentage of hospitalizations with an ADR‐related code for admission diagnosis was 1.83% and 6.58% with an ADR‐related code at any time during hospitalization. A total of 137 medication groups was included in the main ADR analyses. Of these, 13 (9.5%) had statistically non‐significant adjusted ORs, 58 (42.3%) statistically significant ORs between 1.0 and 1.5, 37 (27.0%) between 1.5–2.0, 18 (13.1%) between 2.0–3.0 and 11 (8.0%) 3.0 or higher. Several classes of antibiotics (including penicillins) were among medicines with largest ORs. Evaluating the 14 medications most often associated with ADRs, a strong association was found between the number of these medicines and the risk of ADR‐related hospital admission (adjusted OR of 7.53 (95% CI 7.15–7.93) for those exposed to 6+ of these medicines). Conclusions and Relevance: There is a need for a regular systematic assessment of the harm‐benefit ratio of medicines, harvesting the information in large healthcare databases and combining it with causality assessment of individual case histories. [ABSTRACT FROM AUTHOR]

Introduction: This study considers the prediction of the time until two survival outcomes have both occurred. We compared a variety of analytical methods motivated by a typical clinical problem of multimorbidity prognosis. Methods: We considered five methods: product (multiply marginal risks), dual‐outcome (directly model the time until both events occur), multistate models (msm), and a range of copula and frailty models. We assessed calibration and discrimination under a variety of simulated data scenarios, varying outcome prevalence, and the amount of residual correlation. The simulation focused on model misspecification and statistical power. Using data from the Clinical Practice Research Datalink, we compared model performance when predicting the risk of cardiovascular disease and type 2 diabetes both occurring. Results: Discrimination was similar for all methods. The product method was poorly calibrated in the presence of residual correlation. The msm and dual‐outcome models were the most robust to model misspecification but suffered a drop in performance at small sample sizes due to overfitting, which the copula and frailty model were less susceptible to. The copula and frailty model's performance were highly dependent on the underlying data structure. In the clinical example, the product method was poorly calibrated when adjusting for 8 major cardiovascular risk factors. Discussion: We recommend the dual‐outcome method for predicting the risk of two survival outcomes both occurring. It was the most robust to model misspecification, although was also the most prone to overfitting. The clinical example motivates the use of the methods considered in this study. [ABSTRACT FROM AUTHOR]

This study evaluated drug–drug interactions (DDIs) between antibiotic and nonantibiotic drugs listed with warnings of severe outcomes in the British National Formulary based on adverse drug reaction (ADR) detectable with routine International Classification of Diseases, Tenth Revision coding. Data sources were Clinical Practice Research Databank GOLD and Aurum anonymized electronic health records from English general practices linked to hospital admission records. In propensity‐matched case‐control study, outcomes were ADR or emergency admissions. Analyzed were 121,546 ADR‐related admission cases matched to 638,238 controls. For most antibiotics, adjusted odds ratios (aORs) for ADR‐related hospital admission were large (aOR for trimethoprim 4.13; 95% confidence interval (CI), 3.97–4.30). Of the 51 DDIs evaluated for ADR‐related admissions, 38 DDIs (74.5%) had statistically increased aORs of concomitant exposure compared with nonexposure (mean aOR 3.96; range 1.59–11.42); for the 89 DDIs for emergency hospital admission, the results were 75 (84.3%) and mean aOR 2.40; range 1.43–4.17. Changing reference group to single antibiotic exposure reduced aORs for concomitant exposure by 76.5% and 83.0%, respectively. Medicines listed to cause nephrotoxicity substantially increased risks that were related to number of medicines (aOR was 2.55 (95% CI, 2.46–2.64) for current use of 1 and 10.44 (95% CI, 7.36–14.81) for 3 or more medicines). In conclusion, no evidence of substantial risk was found for multiple DDIs with antibiotics despite warnings of severe outcomes in a national formulary and flagging in electronic health record software. It is proposed that the evidence base for inclusion of DDIs in national formularies be strengthened and made publicly accessible and indiscriminate flagging, which compounds alert fatigue, be reduced. [ABSTRACT FROM AUTHOR]

Aims: We evaluated the relationship between body mass index (BMI) and cancer mortality in incident type 2 diabetes. Methods: We used the Clinical Practice Research Datalink GOLD (1998‐2015), linked with the Office of National Statistics mortalities, and derived an incident type 2 diabetes cohort (N = 176 886; aged 30‐85 years). We determined BMI ±12 months diabetes diagnosis. The primary outcome was cancer mortality, categorized into deaths from obesity‐related cancers (ORCs) and non‐ORCs. Secondary outcomes were site‐specific cancer mortality and main causes of deaths [cancer, cardiovascular disease (CVD), non‐cancer non‐CVD]. We developed gender‐specific Cox models and expressed risk as hazard ratios and 95% confidence intervals, stratified by smoking status. Results: With 886 850 person‐years follow‐up, 7593 cancer deaths occurred. Among women who never smoked, there were positive associations between BMI and deaths from endometrial (hazard ratios per 5 kg/m2: 1.43; 95% confidence interval 1.26‐1.61). Among men, associations between BMI and ORC mortality were inverse but attenuated towards null among never smokers and excluding deaths in the first 2 years. In men, the proportion of CVD deaths increased from 36.8% in BMI category 22.5 to 24.9 kg/m2 to 43.6% in BMI category ≥40 kg/m2 (p <.001). Conclusions: We found some relationships between BMI and cancer mortality in patients with type 2 diabetes, but interpretations need to account for smoking status, reverse causality and deaths from CVD. [ABSTRACT FROM AUTHOR]

Clinical prediction models (CPMs) can predict clinically relevant outcomes or events. Typically, prognostic CPMs are derived to predict the risk of a single future outcome. However, there are many medical applications where two or more outcomes are of interest, meaning this should be more widely reflected in CPMs so they can accurately estimate the joint risk of multiple outcomes simultaneously. A potentially naïve approach to multi-outcome risk prediction is to derive a CPM for each outcome separately, then multiply the predicted risks. This approach is only valid if the outcomes are conditionally independent given the covariates, and it fails to exploit the potential relationships between the outcomes. This paper outlines several approaches that could be used to develop CPMs for multiple binary outcomes. We consider four methods, ranging in complexity and conditional independence assumptions: namely, probabilistic classifier chain, multinomial logistic regression, multivariate logistic regression, and a Bayesian probit model. These are compared with methods that rely on conditional independence: separate univariate CPMs and stacked regression. Employing a simulation study and real-world example, we illustrate that CPMs for joint risk prediction of multiple outcomes should only be derived using methods that model the residual correlation between outcomes. In such a situation, our results suggest that probabilistic classification chains, multinomial logistic regression or the Bayesian probit model are all appropriate choices. We call into question the development of CPMs for each outcome in isolation when multiple correlated or structurally related outcomes are of interest and recommend more multivariate approaches to risk prediction. [ABSTRACT FROM AUTHOR]

Clinical prediction models (CPMs) can inform decision making about treatment initiation, which requires predicted risks assuming no treatment is given. However, this is challenging since CPMs are usually derived using data sets where patients received treatment, often initiated postbaseline as "treatment drop-ins." This study proposes the use of marginal structural models (MSMs) to adjust for treatment drop-in. We illustrate the use of MSMs in the CPM framework through simulation studies that represent randomized controlled trials and real-world observational data and the example of statin initiation for cardiovascular disease prevention. The simulations include a binary treatment and a covariate, each recorded at two timepoints and having a prognostic effect on a binary outcome. The bias in predicted risk was examined in a model ignoring treatment, a model fitted on treatment-naïve patients (at baseline), a model including baseline treatment, and the MSM. In all simulation scenarios, all models except the MSM underestimated the risk of outcome given absence of treatment. These results were supported in the statin initiation example, which showed that ignoring statin initiation postbaseline resulted in models that significantly underestimated the risk of a cardiovascular disease event occurring within 10 years. Consequently, CPMs that do not acknowledge treatment drop-in can lead to underallocation of treatment. In conclusion, when developing CPMs to predict treatment-naïve risk, researchers should consider using MSMs to adjust for treatment drop-in, and also seek to exploit the ability of MSMs to allow estimation of individual treatment effects. [ABSTRACT FROM AUTHOR]

There is growing interest in developing clinical prediction models (CPMs) to aid local healthcare decision-making. Frequently, these CPMs are developed in isolation across different populations, with repetitive de novo derivation a common modelling strategy. However, this fails to utilise all available information and does not respond to changes in health processes through time and space. Alternatively, model updating techniques have previously been proposed that adjust an existing CPM to suit the new population, but these techniques are restricted to a single model. Therefore, we aimed to develop a generalised method for updating and aggregating multiple CPMs. The proposed "hybrid method" re-calibrates multiple CPMs using stacked regression while concurrently revising specific covariates using individual participant data (IPD) under a penalised likelihood. The performance of the hybrid method was compared with existing methods in a clinical example of mortality risk prediction after transcatheter aortic valve implantation, and in 2 simulation studies. The simulation studies explored the effect of sample size and between-population-heterogeneity on the method, with each representing a situation of having multiple distinct CPMs and 1 set of IPD. When the sample size of the IPD was small, stacked regression and the hybrid method had comparable but highest performance across modelling methods. Conversely, in large IPD samples, development of a new model and the hybrid method gave the highest performance. Hence, the proposed strategy can inform the choice between utilising existing CPMs or developing a model de novo, thereby incorporating IPD, existing research, and prior (clinical) knowledge into the modelling strategy. [ABSTRACT FROM AUTHOR]

Aims This study was designed to evaluate whether survival rates in patients with heart failure ( HF) are better than those in patients with diagnoses of the four most common cancers in men and women, respectively, in a contemporary primary care cohort in the community in Scotland. Methods and results Data were obtained from the Primary Care Clinical Informatics Unit from a database of 1.75 million people registered with 393 general practices in Scotland. Sex-specific survival modelling was undertaken using Cox proportional hazards models, adjusted for potential confounders. A total of 56 658 subjects were eligible for inclusion in the study. These represented a total of 147 938 person-years of follow-up (median follow-up: 2.04 years). In men, HF (reference group; 5-year survival: 55.8%) had worse mortality outcomes than prostate cancer [hazard ratio ( HR) 0.61, 95% confidence interval ( CI) 0.57-0.65; 5-year survival: 68.3%], and bladder cancer ( HR 0.88, 95% CI 0.81-0.96; 5-year survival: 57.3%), but better outcomes than lung cancer ( HR 3.86, 95% CI 3.65-4.07; 5-year survival: 8.4%) and colorectal cancer ( HR 1.23, 95% CI 1.16-1.31; 5-year survival: 48.9%). In women, HF (reference group; 5-year survival: 49.5%) had worse mortality outcomes than breast cancer ( HR 0.55, 95% CI 0.51-0.59; 5-year survival 77.7%), but better outcomes than colorectal cancer (HR 1.21, 95% CI 1.13-1.29; 5-year survival 51.5%), lung cancer ( HR 3.82, 95% CI 3.60-4.05; 5-year survival 10.4%), and ovarian cancer ( HR 1.98, 95% CI 1.80-2.17; 5-year survival 38.2%). Conclusions Despite advances in management, HF remains as 'malignant' as some of the common cancers in both men and women. [ABSTRACT FROM AUTHOR]

Cardiac implantable electronic devices include remote monitoring tools intended to guide heart failure management. The monitoring focus has been on averting hospitalizations by predicting worsening heart failure. However, although device measurements including intrathoracic impedance correlate with risk of decompensation, they individually predict hospitalizations with limited accuracy. Current 'crisis detection' methods involve repeatedly screening for impending decompensation, and do not adhere to the principles of diagnostic testing. Complex substrate, limited test performance, low outcome incidence, and long test to outcome times inevitably generate low positive and high negative predictive values. When combined with spectrum bias, the generalizability, incremental value, and cost-effectiveness of device algorithms are questionable. To avoid these pitfalls, remote monitoring may need to shift from crisis detection to health maintenance, keeping the patient within an ideal physiological range through continuous 'closed loop' interaction and dynamic therapy adjustment. Test performance must also improve, possibly through combination with physiological sensors in different dimensions, static baseline characteristics, and biomarkers. Complex modelling may tailor monitoring to individual phenotypes, and thus realize a personalized medicine approach. Future randomized controlled trials should carefully consider these issues, and ensure that the interventions tested are generalizable to clinical practice. [ABSTRACT FROM AUTHOR]

We thank Sabour and Ghajari for their letter1 and interest in our recent paper.2 Regrettably, we feel our paper has been misunderstood in several places, and we therefore would like to take this opportunity to respond to the points raised. REFERENCES 1 Sabour S, Ghajari H. Clinical prediction models to predict the risk of multiple binary outcomes: methodological issues. [Extracted from the article]

Background Identifying different patterns of allergens and understanding their predictive ability in relation to asthma and other allergic diseases is crucial for the design of personalized diagnostic tools. Methods Allergen-IgE screening using Immuno CAP ISAC® assay was performed at age 11 yrs in children participating a population-based birth cohort. Logistic regression ( LR) and nonlinear statistical learning models, including random forests ( RF) and Bayesian networks ( BN), coupled with feature selection approaches, were used to identify patterns of allergen responses associated with asthma, rhino-conjunctivitis, wheeze, eczema and airway hyper-reactivity ( AHR, positive methacholine challenge). Sensitivity/specificity and area under the receiver operating characteristic ( AUROC) were used to assess model performance via repeated validation. Results Serum sample for IgE measurement was obtained from 461 of 822 (56.1%) participants. Two hundred and thirty-eight of 461 (51.6%) children had at least one of 112 allergen components IgE > 0 ISU. The binary threshold >0.3 ISU performed less well than using continuous IgE values, discretizing data or using other data transformations, but not significantly (p = 0.1). With the exception of eczema ( AUROC~0.5), LR, RF and BN achieved comparable AUROC, ranging from 0.76 to 0.82. Dust mite, pollens and pet allergens were highly associated with asthma, whilst pollens and dust mite with rhino-conjunctivitis. Egg/bovine allergens were associated with eczema. Conclusions After validation, LR, RF and BN demonstrated reasonable discrimination ability for asthma, rhino-conjunctivitis, wheeze and AHR, but not for eczema. However, further improvements in threshold ascertainment and/or value transformation for different components, and better interpretation algorithms are needed to fully capitalize on the potential of the technology. [ABSTRACT FROM AUTHOR]

Existing literature on inter-rater reliability focuses on quantifying the disagreement between raters. In this paper, we introduce a method to correct for inter-rater disagreement (or observer bias), where raters are assigning scores on a continuous scale. To do this, we propose a two-stage approach. In the first stage, we standardise the distributions of rater scores to account for each rater's subjective interpretation of the continuous scale. In the second stage, we correct for case-mix differences between raters by exploiting pairwise information where two raters have read the same entity on a case. We illustrate the use of our procedure on clinicians' visual assessments of breast density (a risk factor for breast cancer). After applying our procedure, 229 out of 1398 women who were originally classified as high density were re-classified as non-high density, and 382 out of 12 348 women were re-classified from non-high to high density. A simulation study also demonstrates good performance of the proposed method over a range of scenarios. Copyright © 2013 John Wiley & Sons, Ltd. [ABSTRACT FROM AUTHOR]

In many time-to-event studies, particularly in epidemiology, the time of the first observation or study entry is arbitrary in the sense that this is not a time of risk modification. We present a formal argument that, in these situations, it is not advisable to take the first observation as the time origin, either in accelerated failure time or proportional hazards models. Instead, we advocate using birth as the time origin. We use a two-stage process to account for the fact that baseline observations may be made at different ages in different subjects. First, we marginally regress any potentially age-varying covariates against age, retaining the residuals. These residuals are then used as covariates in fitting an accelerated failure time or proportional hazards model - we call the procedures residual accelerated failure time regression and residual proportional hazards regression, respectively. We compare residual accelerated failure time regression with the standard approach, demonstrating superior predictive ability of the residual method in realistic examples and potentially higher power of the residual method. This highlights flaws in current approaches to communicating risks from epidemiological evidence to support clinical and health policy decisions. Copyright © 2012 John Wiley & Sons, Ltd. [ABSTRACT FROM AUTHOR]