Moderate alcohol consumption on the risk of stroke in the Million Veteran Program

BACKGROUND

There is inconsistent evidence on the association of moderate alcohol consumption and stroke risk in the general population and is not well studied among U.S. Veterans. Furthermore, it is unclear whether primarily drinking beer, wine, or liquor is associated with a difference in stroke risk.

METHODS

The study included 185,323 Million Veteran Program participants who self-reported alcohol consumption on the Lifestyle Survey. Moderate consumption was defined as 1-2 drinks/day and beverage preference of beer, wine or liquor was defined if ≥ 50% of total drinks consumed were from a single type of beverage. Strokes were defined using ICD-9 and ICD-10 codes from the participants' electronic health record.

RESULTS

The mean (sd) age of the sample was 64 (13) years and 11% were women. We observed 4,339 (94% ischemic; 6% hemorrhagic) strokes over a median follow-up of 5.2 years. In Cox models adjusted for age, sex, race, education, income, body mass index, smoking, exercise, diet, cholesterol, prevalent diabetes, prevalent hypertension, lipid-lowering medication, antihypertensive medication, and diabetes medication, moderate alcohol consumption (1-2 drinks/day) was associated with a 22% lower risk of total stroke compared with never drinking [Hazards ratio (HR) 95% confidence interval (CI): 0.78 (0.67, 0.92)]. When stratifying by stroke type, we observed a similar protective association with moderate consumption and ischemic stroke [HR (95% CI): 0.76 (0.65, 0.90)], but a non-statistically significant higher risk of hemorrhagic stroke [HR (95% CI): 1.29 (0.64, 2.61)]. We did not observe a difference in ischemic or hemorrhagic stroke risk among those who preferred beer, liquor or wine vs. no beverage preference. When stratifying by prior number of hospital visits (≤ 15, 16-33, 34-64, ≥ 65) as a proxy for health status, we observed attenuation of the protective association with greater number of visits [HR (95% CI): 0.87 (0.63, 1.19) for ≥ 65 visits vs. 0.80 (0.59, 1.08) for ≤ 15 visits].

CONCLUSIONS

We observed a lower risk of ischemic stroke, but not hemorrhagic stroke with moderate alcohol consumption and did not observe substantial differences in risk by beverage preference among a sample of U.S. Veterans. Healthy user bias of moderate alcohol consumption may be driving some of the observed protective association.

SAS and R code for probabilistic quantitative bias analysis for misclassified binary variables and binary unmeasured confounders

Systematic error from selection bias, uncontrolled confounding, and misclassification is ubiquitous in epidemiologic research but is rarely quantified using quantitative bias analysis (QBA). This gap may in part be due to the lack of readily modifiable software to implement these methods. Our objective is to provide computing code that can be tailored to an analyst's dataset. We briefly describe the methods for implementing QBA for misclassification and uncontrolled confounding and present the reader with example code for how such bias analyses, using both summary-level data and individual record-level data, can be implemented in both SAS and R. Our examples show how adjustment for uncontrolled confounding and misclassification can be implemented. Resulting bias-adjusted point estimates can then be compared to conventional results to see the impact of this bias in terms of its direction and magnitude. Further, we show how 95% simulation intervals can be generated that can be compared to conventional 95% confidence intervals to see the impact of the bias on uncertainty. Having easy to implement code that users can apply to their own datasets will hopefully help spur more frequent use of these methods and prevent poor inferences drawn from studies that do not quantify the impact of systematic error on their results.

Bias Analysis in Healthcare Time Series (BAHT) Decision Support Systems from Meta Data

One of the hindrances in the widespread acceptance of deep learning-based decision support systems in healthcare is bias. Bias in its many forms occurs in the datasets used to train and test deep learning models and is amplified when deployed in the real world, leading to challenges such as model drift. Recent advancements in the field of deep learning have led to the deployment of deployable automated healthcare diagnosis decision support systems at hospitals as well as tele-medicine through IoT devices. Research has been focused primarily on the development and improvement of these systems leaving a gap in the analysis of the fairness. The domain of FAccT ML (fairness, accountability, and transparency) accounts for the analysis of these deployable machine learning systems. In this work, we present a framework for bias analysis in healthcare time series (BAHT) signals such as electrocardiogram (ECG) and electroencephalogram (EEG). BAHT provides a graphical interpretive analysis of bias in the training, testing datasets in terms of protected variables, and analysis of bias amplification by the trained supervised learning model for time series healthcare decision support systems. We thoroughly investigate three prominent time series ECG and EEG healthcare datasets used for model training and research. We show the extensive presence of bias in the datasets leads to potentially biased or unfair machine-learning models. Our experiments also demonstrate the amplification of identified bias with an observed maximum of 66.66%. We investigate the effect of model drift due to unanalyzed bias in datasets and algorithms. Bias mitigation though prudent is a nascent area of research. We present experiments and analyze the most prevalently accepted bias mitigation strategies of under-sampling, oversampling, and the use of synthetic data for balancing the dataset through augmentation. It is important that healthcare models, datasets, and bias mitigation strategies should be properly analyzed for a fair unbiased delivery of service.

Spinal pain and major depression in a military cohort: bias analysis of dependent misclassification in electronic medical records

BACKGROUND

Spinal pain and major depression are prevalent conditions in adult populations and are particularly impactful in the military. However, the temporal relationship between these two conditions remains poorly understood.

METHODS

Using data extracted from electronic medical records, we assessed the association between incident diagnoses of spinal pain and major depression in a cohort of 48,007 Canadian Armed Forces personnel followed from January 2017 to August 2018. We used multivariate Poisson regression to measure the association between the period prevalence of these two conditions. We used probabilistic bias modelling to correct our estimates for misclassification of spinal pain and major depression.

RESULTS

After correcting for misclassification with probabilistic bias modelling, subjects newly diagnosed with spinal pain during the study period were 1.41 times (95% interval 1.25, 1.59) more likely also to be diagnosed with incident major depression, and personnel newly diagnosed with major depression were 1.28 times (95% interval 1.17, 1.39) more likely also to be diagnosed with spinal pain, compared to undiagnosed counterparts of the same age and sex. Without bias corrections, we would have overestimated the magnitude of the association between major depression and spinal pain by a factor of approximately 2.0.

CONCLUSION

Our results highlight a moderate and bi-directional association between two of the most prevalent disorders in military populations. Our results also highlight the importance of correcting for misclassification in electronic medical record data research.

The validity of the claims-based definition of rheumatoid arthritis evaluated in 64 hospitals in Japan

BACKGROUND

An administrative database covering a whole population such as the national database in Japan may be used to estimate the nationwide prevalence of diseases including rheumatoid arthritis (RA) when a well-validated definition of the disease is available. In Japan, the record linkage between the administrative database and medical charts in hospitals is strictly prohibited. A "hospital-based" validation study is one of few possible validation studies where claims kept inside the study hospital are rearranged into the database structure.

METHODS

We selected random samples of 19,734 patients from approximately 1.6 million patients who received medical care between February 2018 and January 2019 in one of the 64 hospitals of the Tokushukai Medical Group. We excluded patients whose observation period was less than 365 days and identified 334 patients who met the definition of "possible cases of RA" whose medical charts were then independently evaluated by two rheumatologists. In a sensitivity analysis, we assessed bias due to misclassifying some patients with RA who did not meet the definition of "possible cases of RA" as a patient with no RA.

RESULTS

The kappa coefficient between the two rheumatologists was 0.80. The prevalence of RA in the study population was estimated to be 0.56%. We found that [condition code of RA] and ([any disease-modifying antirheumatic drug] or [oral corticosteroid with no systemic autoimmune diseases (other than RA) and no polymyalgia rheumatica]) had a relatively high sensitivity (approximately 73%) and a high positive predictive value (approximately 80%). In a sensitivity analysis, we found that when some patients with RA who did not meet the definition of "possible cases of RA" were misclassified as a patient with no RA, then this would lead to underestimation of the prevalence of the definition-positive patients and the adjusted prevalence.

CONCLUSIONS

We recommend using the claims-based definition of RA (found in the current validation study) to estimate the prevalence of RA in Japan. We also suggest estimating the adjusted prevalence using the quantitative bias analysis method, since the prevalence of the disease in the "hospital-based" validation study is different from that in the administrative database.

TRIAL REGISTRATION

The current study is not a clinical trial and hence not subject to trial registration.

Depression and cardiovascular disease events among patients with type 2 diabetes: A systematic review and meta-analysis with bias analysis

AIMS

To provide updated systematic and quantitative summary of the association between depression and the risk of CVD events among individuals with type 2 diabetes. We also aimed to examine the sensitivity of the association to uncontrolled confounding.

METHODS

Data sources included Medline, Embase, and PsycInfo through September 2019. Two independent reviewers selected cohort studies that evaluated the association between depression and fatal or non-fatal CVD events among individuals with type 2 diabetes. Bias analysis was performed using the bias formula approach.

RESULTS

Of 2527 citations screened, 17 eligible studies with a total of 1,033,131 participants were identified. Based on random-effects meta-analysis, depression was associated with higher risks of non-fatal CVD events (relative risk 1.35, 95% confidence interval [CI] 1.20 to 1.53) and fatal CVD event (relative risk 1.47, 95% CI 1.21 to 1.77). Bias analysis indicated that unmeasured confounders alone may not explain the observed association between depression and CVD events among individuals with type 2 diabetes.

CONCLUSIONS

Depression was associated with a higher risk of non-fatal and fatal CVD events among individuals with type 2 diabetes. Our findings provide updated and robust evidence about the association between depression and CVD events among individuals with type 2 diabetes.

Towards reduction in bias in epidemic curves due to outcome misclassification through Bayesian analysis of time-series of laboratory test results: case study of COVID-19 in Alberta, Canada and Philadelphia, USA

BACKGROUND

Despite widespread use, the accuracy of the diagnostic test for SARS-CoV-2 infection is poorly understood. The aim of our work was to better quantify misclassification errors in identification of true cases of COVID-19 and to study the impact of these errors in epidemic curves using publicly available surveillance data from Alberta, Canada and Philadelphia, USA.

METHODS

We examined time-series data of laboratory tests for SARS-CoV-2 viral infection, the causal agent for COVID-19, to try to explore, using a Bayesian approach, the sensitivity and specificity of the diagnostic test.

RESULTS

Our analysis revealed that the data were compatible with near-perfect specificity, but it was challenging to gain information about sensitivity. We applied these insights to uncertainty/bias analysis of epidemic curves under the assumptions of both improving and degrading sensitivity. If the sensitivity improved from 60 to 95%, the adjusted epidemic curves likely falls within the 95% confidence intervals of the observed counts. However, bias in the shape and peak of the epidemic curves can be pronounced, if sensitivity either degrades or remains poor in the 60-70% range. In the extreme scenario, hundreds of undiagnosed cases, even among the tested, are possible, potentially leading to further unchecked contagion should these cases not self-isolate.

CONCLUSION

The best way to better understand bias in the epidemic curves of COVID-19 due to errors in testing is to empirically evaluate misclassification of diagnosis in clinical settings and apply this knowledge to adjustment of epidemic curves.

Cancers Preceding Parkinson's Disease after Adjustment for Bias in a Danish Population-Based Case-Control Study

Cancer patients are at lower risk of developing Parkinson's disease (PD) compared with the general population. One explanation is the negative association between smoking and PD, but PD risk is also lower for cancers not related to smoking. Another explanation is survival bias where death from cancer may act as a competing risk. We conducted a large population-based case-control study in Denmark and investigated whether cancer diagnosis reduced the risk of developing PD even after adjusting for important risk factors including smoking, physical activity, and lifetime oestrogen status. Using probabilistic bias analysis we quantified the influence of survival bias. We estimated negative point estimates (ORs) between cancers and PD for all cancers except skin, female breast, and ill-defined and unspecified 0.85 (95% CI 0.59-1.21); smoking-related cancers 0.75 (95% CI 0.45-1.23); and cancers not related to smoking 0.82 (95% CI 0.49-1.38) that are very similar to those previously reported for a much larger Danish register only based study, even though our confidence intervals include the null. These effect estimates shifted towards the null after accounting for survival bias but most bias-adjusted ORs remained below 1 within the range of priors considered in simulations. Overall, cancer patients have a lower risk of developing PD even after controlling for cancer-related lifestyles factors and correcting for survival bias.

How to deal with the Poisson-gamma model to forecast patients' recruitment in clinical trials when there are pauses in recruitment dynamic?

Recruiting patients is a crucial step of a clinical trial. Estimation of the trial duration is a question of paramount interest. Most techniques are based on deterministic models and various ad hoc methods neglecting the variability in the recruitment process. To overpass this difficulty the so-called Poisson-gamma model has been introduced involving, for each centre, a recruitment process modelled by a Poisson process whose rate is assumed constant in time and gamma-distributed. The relevancy of this model has been widely investigated. In practice, rates are rarely constant in time, there are breaks in recruitment (for instance week-ends or holidays). Such information can be collected and included in a model considering piecewise constant rate functions yielding to an inhomogeneous Cox model. The estimation of the trial duration is much more difficult. Three strategies of computation of the expected trial duration are proposed considering all the breaks, considering only large breaks and without considering breaks. The bias of these estimations procedure are assessed by means of simulation studies considering three scenarios of breaks simulation. These strategies yield to estimations with a very small bias. Moreover, the strategy with the best performances in terms of prediction and with the smallest bias is the one which does not take into account of breaks. This result is important as, in practice, collecting breaks data is pretty hard to manage.

Quantitative bias analysis in an asthma study of rescue-recovery workers and volunteers from the 9/11 World Trade Center attacks

PURPOSE

When learning bias analysis, epidemiologists are taught to quantitatively adjust for multiple biases by correcting study results in the reverse order of the error sequence. To understand the error sequence for a particular study, one must carefully examine the health study's epidemiologic data-generating process. In this article, we describe the unique data-generating process of a man-made disaster epidemiologic study.

METHODS

We described the data-generating process and conducted a bias analysis for a study associating September 11, 2001 dust cloud exposure and self-reported newly physician-diagnosed asthma among rescue-recovery workers and volunteers. We adjusted an odds ratio (OR) estimate for the combined effect of missing data, outcome misclassification, and nonparticipation.

RESULTS

Under our assumptions about systematic error, the ORs adjusted for all three biases ranged from 1.33 to 3.84. Most of the adjusted estimates were greater than the observed OR of 1.77 and were outside the 95% confidence limits (1.55, 2.01).

CONCLUSIONS

Man-made disasters present some situations that are not observed in other areas of epidemiology. Future epidemiologic studies of disasters could benefit from a proactive approach that focuses on the technical aspect of data collection and gathers information on bias parameters to provide more meaningful interpretations of results.

Analysis of error-prone survival data under additive hazards models: measurement error effects and adjustments

Covariate measurement error occurs commonly in survival analysis. Under the proportional hazards model, measurement error effects have been well studied, and various inference methods have been developed to correct for error effects under such a model. In contrast, error-contaminated survival data under the additive hazards model have received relatively less attention. In this paper, we investigate this problem by exploring measurement error effects on parameter estimation and the change of the hazard function. New insights of measurement error effects are revealed, as opposed to well-documented results for the Cox proportional hazards model. We propose a class of bias correction estimators that embraces certain existing estimators as special cases. In addition, we exploit the regression calibration method to reduce measurement error effects. Theoretical results for the developed methods are established, and numerical assessments are conducted to illustrate the finite sample performance of our methods.

Investigation of the association between the test day milk fat-protein ratio and clinical mastitis using a Poisson regression approach for analysis of time-to-event data

The objective of the present study was to investigate the association between the milk fat-protein ratio and the incidence rate of clinical mastitis including repeated cases of clinical mastitis to determine the usefulness of this association to monitor metabolic disorders as risk factors for udder health. Herd records from 10 dairy herds of Holstein cows in Saxony, Germany, from September 2005-2011 (36,827 lactations of 17,657 cows) were used for statistical analysis. A mixed Poisson regression model with the weekly incidence rate of clinical mastitis as outcome variable was fitted. The model included repeated events of the outcome, time-varying covariates and multilevel clustering. Because the recording of clinical mastitis might have been imperfect, a probabilistic bias analysis was conducted to assess the impact of the misclassification of clinical mastitis on the conventional results. The lactational incidence of clinical mastitis was 38.2%. In 36.2% and 34.9% of the lactations, there was at least one dairy herd test day with a fat-protein ratio of <1.0 or >1.5, respectively. Misclassification of clinical mastitis was assumed to have resulted in bias towards the null. A clinical mastitis case increased the incidence rate of following cases of the same cow. Fat-protein ratios of <1.0 and >1.5 were associated with higher incidence rates of clinical mastitis depending on week in milk. The effect of a fat-protein ratio >1.5 on the incidence rate of clinical mastitis increased considerably over the course of lactation, whereas the effect of a fat-protein ratio <1.0 decreased. Fat-protein ratios <1.0 or >1.5 on the precedent test days of all cows irrespective of their time in milk seemed to be better predictors for clinical mastitis than the first test day results per lactation.

Test for rare variants by environment interactions in sequencing association studies

We consider in this article testing rare variants by environment interactions in sequencing association studies. Current methods for studying the association of rare variants with traits cannot be readily applied for testing for rare variants by environment interactions, as these methods do not effectively control for the main effects of rare variants, leading to unstable results and/or inflated Type 1 error rates. We will first analytically study the bias of the use of conventional burden-based tests for rare variants by environment interactions, and show the tests can often be invalid and result in inflated Type 1 error rates. To overcome these difficulties, we develop the interaction sequence kernel association test (iSKAT) for assessing rare variants by environment interactions. The proposed test iSKAT is optimal in a class of variance component tests and is powerful and robust to the proportion of variants in a gene that interact with environment and the signs of the effects. This test properly controls for the main effects of the rare variants using weighted ridge regression while adjusting for covariates. We demonstrate the performance of iSKAT using simulation studies and illustrate its application by analysis of a candidate gene sequencing study of plasma adiponectin levels.

Bias from conditioning on live birth in pregnancy cohorts: an illustration based on neurodevelopment in children after prenatal exposure to organic pollutants

Only 60-70% of fertilized eggs may result in a live birth, and very early fetal loss mainly goes unnoticed. Outcomes that can only be ascertained in live-born children will be missing for those who do not survive till birth. In this article, we illustrate a common bias structure (leading to 'live-birth bias') that arises from studying the effects of prenatal exposure to environmental factors on long-term health outcomes among live births only in pregnancy cohorts. To illustrate this we used prenatal exposure to perfluoroalkyl substances (PFAS) and attention-deficit/hyperactivity disorder (ADHD) in school-aged children as an example. PFAS are persistent organic pollutants that may impact human fecundity and be toxic for neurodevelopment. We simulated several hypothetical scenarios based on characteristics from the Danish National Birth Cohort and found that a weak inverse association may appear even if PFAS do not cause ADHD but have a considerable effect on fetal survival. The magnitude of the negative bias was generally small, and adjusting for common causes of the outcome and fetal loss can reduce the bias. Our example highlights the need to identify the determinants of pregnancy loss and the importance of quantifying bias arising from conditioning on live birth in observational studies.

Bias and sensitivity analysis when estimating treatment effects from the cox model with omitted covariates

Omission of relevant covariates can lead to bias when estimating treatment or exposure effects from survival data in both randomized controlled trials and observational studies. This paper presents a general approach to assessing bias when covariates are omitted from the Cox model. The proposed method is applicable to both randomized and non-randomized studies. We distinguish between the effects of three possible sources of bias: omission of a balanced covariate, data censoring and unmeasured confounding. Asymptotic formulae for determining the bias are derived from the large sample properties of the maximum likelihood estimator. A simulation study is used to demonstrate the validity of the bias formulae and to characterize the influence of the different sources of bias. It is shown that the bias converges to fixed limits as the effect of the omitted covariate increases, irrespective of the degree of confounding. The bias formulae are used as the basis for developing a new method of sensitivity analysis to assess the impact of omitted covariates on estimates of treatment or exposure effects. In simulation studies, the proposed method gave unbiased treatment estimates and confidence intervals with good coverage when the true sensitivity parameters were known. We describe application of the method to a randomized controlled trial and a non-randomized study.