Directed acyclic graphs (DAGs) are an increasingly popular approach for identifying confounding variables that require conditioning when estimating causal effects. This review examined the use of ...DAGs in applied health research to inform recommendations for improving their transparency and utility in future research.
Original health research articles published during 1999-2017 mentioning 'directed acyclic graphs' (or similar) or citing DAGitty were identified from Scopus, Web of Science, Medline and Embase. Data were extracted on the reporting of: estimands, DAGs and adjustment sets, alongside the characteristics of each article's largest DAG.
A total of 234 articles were identified that reported using DAGs. A fifth (n = 48, 21%) reported their target estimand(s) and half (n = 115, 48%) reported the adjustment set(s) implied by their DAG(s). Two-thirds of the articles (n = 144, 62%) made at least one DAG available. DAGs varied in size but averaged 12 nodes interquartile range (IQR): 9-16, range: 3-28 and 29 arcs (IQR: 19-42, range: 3-99). The median saturation (i.e. percentage of total possible arcs) was 46% (IQR: 31-67, range: 12-100). 37% (n = 53) of the DAGs included unobserved variables, 17% (n = 25) included 'super-nodes' (i.e. nodes containing more than one variable) and 34% (n = 49) were visually arranged so that the constituent arcs flowed in the same direction (e.g. top-to-bottom).
There is substantial variation in the use and reporting of DAGs in applied health research. Although this partly reflects their flexibility, it also highlights some potential areas for improvement. This review hence offers several recommendations to improve the reporting and use of DAGs in future research.
The past few decades have seen substantial improvements in cancer survival, but concerns exist about long-term cardiovascular disease risk in survivors. Evidence is scarce on the risks of specific ...cardiovascular diseases in survivors of a wide range of cancers to inform prevention and management. In this study, we used large-scale electronic health records data from multiple linked UK databases to address these evidence gaps.
For this population-based cohort study, we used linked primary care, hospital, and cancer registry data from the UK Clinical Practice Research Datalink to identify cohorts of survivors of the 20 most common cancers who were 18 years or older and alive 12 months after diagnosis and controls without history of cancer, matched for age, sex, and general practice. We compared risks for a range of cardiovascular disease outcomes using crude and adjusted Cox models. We fitted interactions to investigate effect modification, and flexible parametric survival models to estimate absolute excess risks over time.
Between Jan 1, 1990, and Dec 31, 2015, 126 120 individuals with a diagnosis of a cancer of interest still being followed up at least 1 year later were identified and matched to 630 144 controls. After exclusions, 108 215 cancer survivors and 523 541 controls were included in the main analyses. Venous thromboembolism risk was elevated in survivors of 18 of 20 site-specific cancers compared with that of controls; adjusted hazard ratios (HRs) ranged from 1·72 (95% CI 1·57–1·89) in patients after prostate cancer to 9·72 (5·50–17·18) after pancreatic cancer. HRs decreased over time, but remained elevated more than 5 years after diagnosis. We observed increased risks of heart failure or cardiomyopathy in patients after ten of 20 cancers, including haematological (adjusted HR 1·94, 1·66–2·25, with non-Hodgkin lymphoma; 1·77, 1·50–2·09, with leukaemia; and 3·29, 2·59–4·18, with multiple myeloma), oesophageal (1·96, 1·46–2·64), lung (1·82, 1·52–2·17) kidney (1·73, 1·38–2·17) and ovarian (1·59, 1·19–2·12). Elevated risks of arrhythmia, pericarditis, coronary artery disease, stroke, and valvular heart disease were also observed for multiple cancers, including haematological malignancies. HRs for heart failure or cardiomyopathy and venous thromboembolism were greater in patients without previous cardiovascular disease and in younger patients. However, absolute excess risks were generally greater with increasing age. Increased risks of these outcomes seemed most pronounced in patients who had received chemotherapy.
Survivors of most site-specific cancers had increased medium-term to long-term risk for one or more cardiovascular diseases compared with that for the general population, with substantial variations between cancer sites.
Wellcome Trust and Royal Society.
Longitudinal data is commonly analysed to inform prevention policies for diseases that may develop throughout life. Commonly methods interpret the longitudinal data as a series of discrete ...measurements or as continuous patterns. Some of the latter methods condition on the outcome, aiming to capture 'average' patterns within outcome groups, while others capture individual-level pattern features before relating these to the outcome. Conditioning on the outcome may prevent meaningful interpretation. Repeated measurements of a longitudinal exposure (weight) and later outcome (glycated haemoglobin levels) were simulated to match three scenarios: one with no causal relationship between growth rate and glycated haemoglobin; two with a positive causal effect of growth rate on glycated haemoglobin. Two methods that condition on the outcome and one that did not were applied to the data in 1000 simulations. The interpretation of the two-step method matched the simulation in all causal scenarios, but that of the methods conditioning on the outcome did not. Methods that condition on the outcome do not accurately represent a causal relationship between a longitudinal pattern and outcome. Researchers considering longitudinal data should carefully determine if they wish to analyse longitudinal data as a series of discrete time points or by extracting pattern features.
Celotno besedilo
Dostopno za:
DOBA, IZUM, KILJ, NUK, PILJ, PNG, SAZU, SIK, UILJ, UKNU, UL, UM, UPUK
Many models for the analysis of spatio-temporal networks specify time as a series of discrete steps. This either requires evenly spaced measurement times or the aggregation of data into measurement ...windows. This can lead to the introduction of bias. An alternative is to use continuous-time models, for example, multilevel models. Models capturing complex spatio-temporal variation are often difficult to visualise and interpret. This can be addressed by simplifying the results, for example by extracting ‘features’ of interest (such as maxima or minima) of temporal patterns associated with different network connections.
This paper uses simulation to evaluate the accuracy and precision with which b-spline-based multilevel models (a flexible form of continuous-time model that can easily capture complex variation associated with a spatio-temporal network structure) capture the timing and extent of maximum delays to journeys made between pairs of stations in a small railway network.
On average models captured the timing and extent of maximum delay with small bias, but there was evidence of overestimation and underestimation of low and high values of these features, respectively. This systematic bias may have partially caused the undercoverage of credible intervals for the pattern features. Alternative model specifications – specifically to capture x-axis random variation, for example – should be considered in future work.
Autoregressive and moving average models for temporally dynamic networks treat time as a series of discrete steps which assumes even intervals between data measurements and can introduce bias if this ...assumption is not met. Using real and simulated data from the London Underground network, this paper illustrates the use of continuous time multilevel models to capture temporal trajectories of edge properties without the need for simultaneous measurements, along with two methods for producing interpretable summaries of model results. These including extracting ‘features’ of temporal patterns (e.g. maxima, time of maxima) which have utility in understanding the network properties of each connection and summarising whole-network properties as a continuous function of time which allows estimation of network properties at any time without temporal aggregation of non-simultaneous measurements. Results for temporal pattern features in the response variable were captured with reasonable accuracy. Variation in the temporal pattern features for the exposure variable was underestimated by the models. The models showed some lack of precision. Both model summaries provided clear ‘real-world’ interpretations and could be applied to data from a range of spatio-temporal network structures (e.g. rivers, social networks). These models should be tested more extensively in a range of scenarios, with potential improvements such as random effects in the exposure variable dimension.
For five decades, blood pressure lowering treatment has been recommended for patients with hypertension (currently defined as blood pressure of ≥140/90 mm Hg). In the past 20 years, guidelines for ...treatment began incorporating predicted absolute cardiovascular disease risk (predicted risk) and reducing blood pressure thresholds. The blood pressure threshold at which to start treatment has become a secondary consideration in some countries. We aimed to provide descriptive data to assess the relative importance of blood pressure thresholds versus predicted risk on the subsequent rate of cardiovascular disease to inform treatment decisions.
In this English population-based cohort study, we used linked data from the Clinical Practice Research Datalink (CPRD) GOLD, Hospital Episode Statistics Admitted Patient Care, and the Office for National Statistics mortality data, and area-based deprivation indices (Townsend scores). Eligible patients were aged 30–79 years on Jan 1, 2011 (cohort entry date) and could be linked to hospital, mortality, and deprivation data. Patients were followed up until death, end of CPRD follow-up, or Nov 31, 2018. We examined three outcomes: cardiovascular disease, markers of potential target organ damage, and incident dementia without a known cause. The rate of each outcome was estimated and stratified by systolic blood pressure and predicted 10-year risk of cardiovascular disease (QRISK2 algorithm).
Between Jan 1, 2011, and Nov 31, 2018, 1 098 991 patients were included in the cohort and followed up for a median of 4·3 years (IQR 2·6–6·0; total follow-up of 4·6 million person-years). Median age at entry was 52 years (IQR 42–62) and 629 711 (57·3%) patients were female. There were 51 996 cardiovascular disease events and the overall rate of cardiovascular disease was 11·2 per 1000 person-years (95% CI 11·1–11·3). Median QRISK2 10-year predicted risk was 4·6% (IQR 1·4–12·0) and mean systolic blood pressure before cohort entry was 129·1 mm Hg (SD 15·7). Within strata of predicted risk, the effect of increasing systolic blood pressure on outcomes was small. For example, in the group with 10·0–19·9% predicted risk, rates of all cardiovascular disease rose from 20·1 to 23·6 per 1000 person-years between systolic blood pressures less than 110 mm Hg and 180 and higher mm Hg. But among patients with systolic blood pressure 140·0–149·9 mm Hg, rates rose from 6·9 to 52·3 per 1000 person-years between those with less than 10·0% risk and those with 30·0% or higher predicted risk.
For a wide range of blood pressures, the rate of cardiovascular disease and effectiveness of blood pressure drug treatment was mainly determined by predicted risk, with blood pressure thresholds 140/90 mm Hg or 160/100 mm Hg—ubiquitous in most countries—adding little useful information. When medium-term predicted risk is low, there is no urgency to initiate drug treatment, allowing time to attempt non-pharmacological blood pressure reduction.
National Institute for Health Research.
Worldwide treatment recommendations for lowering blood pressure continue to be guided predominantly by blood pressure thresholds, despite strong evidence that the benefits of blood pressure reduction ...are observed in patients across the blood pressure spectrum. In this study, we aimed to investigate the implications of alternative strategies for offering blood pressure treatment, using the UK as an illustrative example.
We did a retrospective cohort study in primary care patients aged 30–79 years without cardiovascular disease, using data from the UK's Clinical Practice Research Datalink linked to Hospital Episode Statistics and Office for National Statistics mortality. We assessed and compared four different strategies to determine eligibility for treatment: using 2011 UK National Institute for Health and Care Excellence (NICE) guideline, or proposed 2019 NICE guideline, or blood pressure alone (threshold ≥140/90 mm Hg), or predicted 10-year cardiovascular risk alone (QRISK2 score ≥10%). Patients were followed up until the earliest occurrence of a cardiovascular disease diagnosis, death, or end of follow-up period (March 31, 2016). For each strategy, we estimated the proportion of patients eligible for treatment and number of cardiovascular events that could be prevented with treatment. We then estimated eligibility and number of events that would occur during 10 years in the UK general population.
Between Jan 1, 2011, and March 31, 2016, 1 222 670 patients in the cohort were followed up for a median of 4·3 years (IQR 2·5–5·2). 271 963 (22·2%) patients were eligible for treatment under the 2011 NICE guideline, 327 429 (26·8%) under the proposed 2019 NICE guideline, 481 859 (39·4%) on the basis of a blood pressure threshold of 140/90 mm Hg or higher, and 357 840 (29·3%) on the basis of a QRISK2 threshold of 10% or higher. During follow-up, 32 183 patients were diagnosed with cardiovascular disease (overall rate 7·1 per 1000 person-years, 95% CI 7·0–7·2). Cardiovascular event rates in patients eligible for each strategy were 15·2 per 1000 person-years (95% CI 15·0–15·5) under the 2011 NICE guideline, 14·9 (14·7–15·1) under the proposed 2019 NICE guideline, 11·4 (11·3–11·6) with blood pressure threshold alone, and 16·9 (16·7–17·1) with QRISK2 threshold alone. Scaled to the UK population, we estimated that 233 152 events would be avoided under the 2011 NICE guideline (28 patients needed to treat for 10 years to avoid one event), 270 233 under the 2019 NICE guideline (29 patients), 301 523 using a blood pressure threshold (38 patients), and 322 921 using QRISK2 threshold (27 patients).
A cardiovascular risk-based strategy (QRISK2 ≥10%) could prevent over a third more cardiovascular disease events than the 2011 NICE guideline and a fifth more than the 2019 NICE guideline, with similar efficiency regarding number treated per event avoided.
National Institute for Health Research.
Cancer survivors have a higher risk for developing cardiovascular diseases than the general population.
The aim of this study was to investigate whether cardiovascular mortality overtakes ...cancer-specific mortality during cancer survivorship and, if so, at what point cardiovascular disease becomes the dominant cause of death.
This cohort study used linked English electronic health records, including death registration data. The study population included 104,028 adults ≥40 years of age whose first cancer diagnosis was for 1 of 9 common cancers and who were alive and followed up at least 1 year after diagnosis. Age-stratified mortality rates were estimated from cardiovascular disease or cancer by predicting from Poisson models incorporating categorical age at diagnosis and time since diagnosis. Where cardiovascular disease mortality overtook cancer mortality, the crossover point was estimated using interpolation.
Mortality from cardiovascular causes overtook mortality due to the primary cancer at 2 to 11 years after cancer diagnosis in survivors of all 9 cancer types ≥80 years of age at diagnosis and after 5 to 17 years in survivors of 7 cancer types 60 to 79 years of age at diagnosis. Cardiovascular mortality overtook all cancer mortality for 6 and 2 cancer sites in the ≥80-year and 60- to 79-year age groups, respectively, over a longer time period. Cardiovascular mortality did not overtake cancer mortality during the observation period in patients aged 40 to 59 years, except among survivors of uterine cancer.
In older survivors of 9 common cancers, cardiovascular mortality becomes dominant over mortality from the primary cancer, though not always over total cancer mortality, as time passes since cancer diagnosis.
Abstract only
11571
Background: There are concerns about long-term cardiovascular disease (CVD) risk in cancer survivors, but few studies have quantified the risks for a wide range of cancers and ...specific CVD outcomes. Methods: Using UK electronic health records, we identified cohorts of adults alive one year after a cancer diagnosis at 20 different sites. Risks of a range of CVD outcomes were compared to age, sex and general practice matched cancer free controls using Cox regression; crude and adjusted models were compared to investigate the role of shared cancer/CVD risk factors (e.g. smoking and diabetes). Results: 126 120 cancer survivors and 603 144 controls were followed over a median (IQR) 4.6 (2.5-8.1) and 5.6 (3.2-9.2) years. Crude and adjusted hazard ratios (HRs) were similar. In adjusted models, there was strong evidence (p<0.01) of increased risk of CVDs among cancer survivors compared with controls: venous thromboembolism (VTE, 18 cancers), heart failure/cardiomyopathy (7 cancers), arrhythmia (4 cancers), and stroke (3 cancers). In stratified analyses HRs were higher in younger people and continued beyond 5 years post diagnosis. Conclusions: We found increased long term CVD risk among survivors of several cancers compared to the general population, which varied by cancer site and specific CVD outcome.Table: see text
PDF
