Bias and efficiency of multiple imputation compared with complete-case analysis for missing covariate values

Impact of missing data on bias and precision when estimating change in patient-reported outcomes from a clinical registry

BACKGROUND

Clinical registries, which capture information about the health and healthcare use of patients with a health condition or treatment, often contain patient-reported outcomes (PROs) that provide insights about the patient's perspectives on their health. Missing data can affect the value of PRO data for healthcare decision-making. We compared the precision and bias of several missing data methods when estimating longitudinal change in PRO scores.

METHODS

This research conducted analyses of clinical registry data and simulated data. Registry data were from a population-based regional joint replacement registry for Manitoba, Canada; the study cohort consisted of 5631 patients having total knee arthroplasty between 2009 and 2015. PROs were measured using the 12-item Short Form Survey, version 2 (SF-12v2) at pre- and post-operative occasions. The simulation cohort was a subset of 3000 patients from the study cohort with complete PRO information at both pre- and post-operative occasions. Linear mixed-effects models based on complete case analysis (CCA), maximum likelihood (ML) and multiple imputation (MI) without and with an auxiliary variable (MI-Aux) were used to estimate longitudinal change in PRO scores. In the simulated data, bias, root mean squared error (RMSE), and 95% confidence interval (CI) coverage and width were estimated under varying amounts and types of missing data.

RESULTS

Three thousand two hundred thirty (57.4%) patients in the study cohort had complete data on the SF-12v2 at both occasions. In this cohort, mixed-effects models based on CCA resulted in substantially wider 95% CIs than models based on ML and MI methods. The latter two methods produced similar estimates and 95% CI widths. In the simulation cohort, when 50% of the data were missing, the MI-Aux method, in which a single hypothetical auxiliary variable was strongly correlated (i.e., 0.8) with the outcome, reduced the 95% CI width by up to 14% and bias and RMSE by up to 50 and 45%, respectively, when compared with the MI method.

CONCLUSIONS

Missing data can substantially affect the precision of estimated change in PRO scores from clinical registry data. Inclusion of auxiliary information in MI models can increase precision and reduce bias, but identifying the optimal auxiliary variable(s) may be challenging.

Threat sensitivity is associated with the healthcare source used most often: doctor's office, emergency room, or none at all

A significant proportion of American adults do not have a regular source of healthcare and the reasons for this shortfall are not fully understood. The objective of this study was to examine the relationship between individual differences in threat sensitivity and healthcare utilization in a survey of 483 African American men. Demographics, psychological characteristics, and health behaviors were assessed. The primary outcomes were: 1) most frequent source of healthcare utilization (doctor's office or clinic vs. emergency room vs. no place), and 2) frequency of healthcare utilization (one or more vs. no healthcare visits in the previous year). Data were analyzed with multivariable logistic regression. Results showed that threat sensitivity, insurance status, and age were associated with the most frequent source of healthcare utilization. Compared to men who most commonly used a doctor's office or clinic, men who tended to use an emergency room had higher levels of threat sensitivity and those with no usual healthcare source had lower levels of threat sensitivity. These findings fit with leading neurobiological theories of personality regarding threat sensitivity. From a pragmatic standpoint, these findings may also lend insight to the tailoring of health marketing messages designed to optimize utilization of healthcare resources.

The proportion of missing data should not be used to guide decisions on multiple imputation

OBJECTIVES

Researchers are concerned whether multiple imputation (MI) or complete case analysis should be used when a large proportion of data are missing. We aimed to provide guidance for drawing conclusions from data with a large proportion of missingness.

STUDY DESIGN AND SETTING

Via simulations, we investigated how the proportion of missing data, the fraction of missing information (FMI), and availability of auxiliary variables affected MI performance. Outcome data were missing completely at random or missing at random (MAR).

RESULTS

Provided sufficient auxiliary information was available; MI was beneficial in terms of bias and never detrimental in terms of efficiency. Models with similar FMI values, but differing proportions of missing data, also had similar precision for effect estimates. In the absence of bias, the FMI was a better guide to the efficiency gains using MI than the proportion of missing data.

CONCLUSION

We provide evidence that for MAR data, valid MI reduces bias even when the proportion of missingness is large. We advise researchers to use FMI to guide choice of auxiliary variables for efficiency gain in imputation analyses, and that sensitivity analyses including different imputation models may be needed if the number of complete cases is small.

Emergence times and airway reactions during general anaesthesia with remifentanil and a laryngeal mask airway: A multicentre randomised controlled trial

BACKGROUND

Avoidance of airway complications and rapid emergence from anaesthesia are indispensable for the use of a laryngeal mask airway (LMA). Evidence from adequately powered randomised studies with a low risk of bias for the optimal anaesthetic in this context is limited.

OBJECTIVE

We tested the hypothesis that when using remifentanil-based intra-operative analgesia, desflurane would be the most suitable anaesthetic: with noninferiority in the occurrence of upper airway complications and superiority in emergence times compared with sevoflurane or propofol.

DESIGN

A randomised, multicentre, partially double-blinded, three-arm, parallel-group study.

SETTING

Two university and two regional German hospitals, from February to October 2015.

PATIENTS

A total of 352 patients (age 18 to 75 years, ASA physical status I to III, BMI less than 35 kg m⁻² and fluent in German) were enrolled in this study. All surgery was elective with a duration of 0.5 to 2 h, and general anaesthesia with a LMA was feasible.

INTERVENTION

The patients were randomised to receive desflurane, sevoflurane or propofol anaesthesia.

MAIN OUTCOME MEASURES

This study was powered for the primary outcome ‘time to state date of birth’ and the secondary outcome ‘intra-operative cough’. Time to emergence from anaesthesia and the incidence of upper airway complications were assessed on the day of surgery.

RESULTS

The primary outcome was analysed for 343 patients: desflurane (n=114), sevoflurane (n=111) and propofol (n=118). The desflurane group had the fastest emergence. The mean (± SD) times to state the date of birth following desflurane, sevoflurane and propofol were 8.1 ± 3.6, 10.1 ± 4.0 and 9.8 ± 5.1 min, respectively (P < 0.01). There was no difference in upper airway complications (cough and laryngospasm) across the groups, but these complications were less frequent than in previous studies.

CONCLUSION

When using a remifentanil infusion for intra-operative analgesia in association with a LMA, desflurane was associated with a significantly faster emergence and noninferiority in the incidence of intra-operative cough than either sevoflurane or Propofol.

TRIAL REGISTRATION

ClinicalTrials.gov identifier: NCT02322502; EudraCT identifier: 2014-003810-96.

Cancer-related symptoms, mental well-being, and psychological distress in men diagnosed with prostate cancer treated with androgen deprivation therapy

Purpose

There are known associations between treatment of prostate cancer (PCa) involving Androgen Deprivation Therapy (ADT) and psychological and physical side effects. We investigate the associations between cancer-related symptoms, health-related quality of life (HRQL), and poor psychological outcomes in men whose treatment for PCa involved ADT.

Methods

A cross-sectional postal questionnaire was administered to UK men 18–42 months post diagnosis of PCa. Men completed items on functional outcomes using the Expanded Prostate Cancer Index Composite (EPIC-26), EuroQol-5D (EQ-5D), and the European Organisation for Research and Treatment of Cancer (EORTC) Fatigue subscale. Psychological outcomes (mental well-being and psychological distress) were assessed using the Short Warwick–Edinburgh Mental Well-being Scale (SWEMWBS) and the Kessler 6-item scale (K6), respectively. Associations between explanatory variables and psychological outcomes were assessed using stepped logistic regression.

Results

13,097 men treated with ADT completed a questionnaire. A minority of men reported poor mental well-being (15.5%) or severe psychological distress (6.6%). After controlling for sociodemographic and clinical variables, reporting clinically significant fatigue was strongly associated with severe psychological distress (OR 9.92; 95% CI 7.63 to 12.89) and poor well-being (OR 3.86; 95% CI 3.38 to 4.42). All cancer-related symptoms and HRQL variables were associated with both psychological outcomes.

Conclusions

While the majority of men treated with ADT did not report poor psychological outcomes, a small proportion reported severe problems. Clinically significant fatigue was demonstrated as a possible indicator of poor outcomes. Healthcare systems need to have clear protocols in place which specifically and routinely target psychological distress and fatigue.

Electronic supplementary material

The online version of this article (10.1007/s11136-019-02212-x) contains supplementary material, which is available to authorized users.

Using simulation studies to evaluate statistical methods

Simulation studies are computer experiments that involve creating data by pseudo-random sampling. A key strength of simulation studies is the ability to understand the behavior of statistical methods because some "truth" (usually some parameter/s of interest) is known from the process of generating the data. This allows us to consider properties of methods, such as bias. While widely used, simulation studies are often poorly designed, analyzed, and reported. This tutorial outlines the rationale for using simulation studies and offers guidance for design, execution, analysis, reporting, and presentation. In particular, this tutorial provides a structured approach for planning and reporting simulation studies, which involves defining aims, data-generating mechanisms, estimands, methods, and performance measures ("ADEMP"); coherent terminology for simulation studies; guidance on coding simulation studies; a critical discussion of key performance measures and their estimation; guidance on structuring tabular and graphical presentation of results; and new graphical presentations. With a view to describing recent practice, we review 100 articles taken from Volume 34 of Statistics in Medicine, which included at least one simulation study and identify areas for improvement.

Regional Variations in Quality of Survival Among Men with Prostate Cancer Across the United Kingdom

BACKGROUND

Prostate cancer incidence, treatment, and survival rates vary throughout the UK, but little is known about regional differences in quality of survival.

OBJECTIVE

To investigate variations in patient-reported outcomes between UK countries and English Cancer Alliances.

DESIGN, SETTING, AND PARTICIPANTS

A cross-sectional postal survey of prostate cancer survivors diagnosed 18-42mo previously.

OUTCOME MEASUREMENTS AND STATISTICAL ANALYSIS

Urinary, bowel, and sexual problems and vitality were patient reported using the Expanded Prostate Cancer Index Composite (EPIC-26) questionnaire. General health was also self-assessed. Regional variations were identified using multivariable log-linear regression.

RESULTS AND LIMITATIONS

A total of 35823 men responded, 60.8% of those invited. Self-assessed health was significantly lower than the UK average in Wales and Scotland. Respondents reported more urinary incontinence in Scotland, more urinary irritation/obstruction in Scotland and Northern Ireland (NI), poorer bowel function in Scotland and NI, worse sexual function in Scotland, and reduced vitality/hormonal function in Scotland, Wales, and NI. Self-assessed health was poorer than the English average in South Yorkshire and North-East and Cumbria, with more urinary incontinence in North-East and Cumbria and Peninsula, greater sexual problems in West Midlands, and poorer vitality in North-East and Cumbria and West Midlands. Limitations include difficulty identifying clinically significant differences and limited information on pretreatment conditions.

CONCLUSIONS

Despite adjustment for treatment, and clinical and sociodemographic factors, quality of survival among prostate cancer survivors varied by area of residence. Adoption of best practice from areas performing well could support enhanced survival quality in poorer performing areas, particularly with regard to bowel problems and vitality, where clinically relevant differences were reported.

PATIENT SUMMARY

We conducted a UK-wide survey of patient's quality of life after treatment for prostate cancer. Outcomes were found to vary depending upon where patients live. Different service providers need to ensure that all prostate cancer patients receive the same follow-up care.

Caregiver burden and quality of life 2 years after attendance at a memory clinic

OBJECTIVES

We aimed to describe (1) the burden and health-related quality of life (HRQL) of informal caregivers of new patients attending a memory assessment service (MAS), (2) changes in these outcomes over 2 years, and (3) satisfaction with services.

METHODS

Informal caregivers of patients attending one of 73 MASs throughout England completed questionnaires at the patient's first appointment, and 6 and 12 months later. Participants from 30 of these MASs were also followed up at 24 months. Questionnaires covered caregivers' sociodemographic characteristics, Zarit Burden Interview, EQ-5D-3L, and satisfaction with services. We used multivariable linear regression to assess relationships between burden, HRQL, and caregiver and patient characteristics.

RESULTS

Of 1020 caregivers at baseline, 569 were followed up at 6 months, 452 at 12 months, and 187 at 24 months. There was a small increase in caregiver burden over 2 years (effect size 0.30 SD). These changes were not associated with most caregiver or patient characteristics, except socio-economic deprivation, which was associated with larger increases in burden at 2 years. Caregivers' HRQL was weakly associated with burden and showed a small reduction over time (0.2 SD). Most caregivers were satisfied with services, but caregivers who were not satisfied with the services they received reported greater increases in burden.

CONCLUSIONS

Increases in caregiver burden and reductions in HRQL appear to be small over the first 2 years after attending a MAS. However, the longer term impact on caregivers and those they care for needs investigating, as do strategies to reduce their burden.

Stage-matched survival differences by ethnicity among gastric cancer patients of Asian ancestry treated in the United States

BACKGROUND

Differences have been noted in overall survival (OS) in gastric cancer (GC) between trials conducted in Western vs Asian countries. The National Cancer Database (NCDB) reports outcomes and patient/disease variables relevant to OS.

METHODS

Using NCDB, we identified 89 558 adult patients with GC diagnosed (2004-2012), where self-reported race/ethnicity was available. Cox proportional hazard model was used to calculate multivariable hazard ratio (HR) of death, adjusting for race/ethnicity, age, gender, insurance, histology, grade, location, stage, and treatment type.

RESULTS

After adjustment, Asian patients had improved OS (HR = 0.74, 95% confidence intervals [CI] = 0.71-0.77). There were differences in OS between Asian ethnicities compared with white patients (n = 69 945), notably with Korean (n = 1249, HR = 0.70, 95% CI = 0.64-0.75), Chinese (n = 1271, HR = 0.69, 95% CI = 0.64-0.75), and Indian/Pakistani patients (n = 492, HR = 0.68, 95% CI = 0.61-0.76). Japanese (n = 829, HR = 0.84, 95% CI = 0.77-0.91) and Vietnamese (n = 560, HR = 0.79, 95% CI = 0.71-0.88) OS was also improved (P < 0.0001), while Filipino patients showed no difference (n = 415, HR = 1.00). Black patients had slightly improved OS (n = 13 500, HR = 0.98, 95% CI = 0.95-1.00, P = 0.035).

CONCLUSIONS

This analysis supports improved OS in Asian patients independent of stage, treatment, and known patient or disease characteristics in this large US cohort, and is the largest to define OS differences between Asian ethnicities.

Vulnerable narcissism and addiction: The mediating role of shame

Problem drinking and gambling are addictive behaviours experienced by young adults and commonly occur with narcissism. Research has acknowledged two distinct forms of narcissism: grandiose and vulnerable. There has been work that has examined the relationship between grandiose and vulnerable narcissism and addictive behaviours, but it has been limited, particularly with vulnerable narcissism. Evidence suggests vulnerable narcissism, but not grandiose narcissism, is associated with greater negative affect. Accordingly, shame, a potent social emotion could be a mediator in the narcissism-addiction pathway. Shame has been implicated in both vulnerable narcissism and problem drinking and gambling. Thus, we hypothesized that shame would mediate the relationship between vulnerable narcissism and addictive behaviours. Young adults (N = 210) completed self-report baseline (Time 1) and 1-month follow-up (Time 2) assessments of shame, narcissism, problem drinking, and gambling. As predicted, those with elevated vulnerable narcissism at Time 1 had increased shame at Time 2, which predicted problem drinking and gambling at Time 2. This relationship was not observed for grandiose narcissism. Overall, our results suggest that feelings of shame are essential to understanding the vulnerable narcissism-addiction pathway, and an important consideration when designing clinical interventions for at-risk young adults.

Realist evaluation of cancer rehabilitation services in South Wales (REEACaRS): a mixed methods study protocol

INTRODUCTION

Long-term and late effects of cancer treatments can cause functional limitations and reduce quality of life. Cancer rehabilitation services, which can comprise physical exercise, psychological support and educational interventions depending on the individual's needs, have been found to have a positive effect on health-related quality of life worldwide. However, accessibility or the lack of awareness on available help can act as barriers and influence the uptake of services, resulting in people having unmet rehabilitation needs. In Wales, UK, 41% of people, who have had health and social care needs resulting from cancer and its treatments, reported that they did not receive care when needed. The reason for this lack of support has not yet been fully investigated. The aim of this study is to investigate the conditions in which cancer rehabilitation services work and their underpinning mechanisms in South Wales, UK, specifically addressing barriers, facilitators and costs.

METHODS AND ANALYSIS

Realist evaluation, which explains for whom a service works in what circumstances and how through context-mechanism-outcome pattern conjunctions, will be used in three phases to investigate the conditions in which cancer rehabilitation services work and their underpinning mechanisms. Phase 1 will be secondary analysis of a cancer rehabilitation database from a local Health Board to give context to who are accessing rehabilitation. Phase 2 will be thematic analysis of face-to-face, semistructured rehabilitation participant (n=20) and healthcare professional (n=20) interviews to explore the mechanisms of how cancer rehabilitation works. Phase 3 will be two case studies and cost-consequences analysis of cancer rehabilitation services.

ETHICS AND DISSEMINATION

This study received favourable ethical opinion from London South-East Research Ethics Committee (17/LO/2123) in December 2017. This project is part of the author's PhD thesis and it is expected that the findings will be disseminated in academic journals and at local and international conferences.

Population-calibrated multiple imputation for a binary/categorical covariate in categorical regression models

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.

Estimating trends in the incidence rate with interval censored data and time-dependent covariates

We propose a multiple imputation method for estimating the incidence rate with interval censored data and time-dependent (and/or time-independent) covariates. The method has two stages. First, we use a semi-parametric G-transformation model to estimate the cumulative baseline hazard function and the effects of the time-dependent (and/or time-independent covariates) on the interval censored infection times. Second, we derive the participant's unique cumulative distribution function and impute infection times conditional on the covariate values. To assess performance, we simulated infection times from a Cox proportional hazards model and induced interval censoring by varying the testing rate, e.g., participants test 100%, 75%, 50% of the time, etc. We then compared the incidence rate estimates from our G-imputation approach with single random-point and mid-point imputation. By comparison, our G-imputation approach gave more accurate incidence rate estimates and appropriate standard errors for models with time-independent covariates only, time-dependent covariates only, and a mixture of time-dependent and time-independent covariates across various testing rates. We demonstrate, for the first time, a multiple imputation approach for incidence rate estimation with interval censored data and time-dependent (and/or time-independent) covariates.

Socioeconomic status and improvement in functional ability among older adults in Japan: a longitudinal study

BACKGROUND

Recovery from functionally disabled status is an important target of public health measures for older adults. This study aimed to examine socioeconomic inequalities in the improvement of functional ability among older adults stratified by the level of disability at baseline.

METHODS

In the Japan Gerontological Evaluation Study, we conducted a mail survey of community-dwelling older adults (1937 men and 2212 women) who developed functional impairment during 2010-2014. The survey data were individually linked to the longitudinal records of changes in the levels of functional disability based on the Public Long-Term Care Insurance System.

RESULTS

The mean (standard deviation) follow-up period was 316 (269) days. During follow-up, 811 participants (19.5%) showed improved functional ability. Among those with severe disabilities at baseline, men with 13 or more years of education were more likely to improve functional ability than men with 9 or fewer years of education (hazard ratio: 1.97, 95% confidence interval: 1.12-3.45). A similar association was observed among women (hazard ratio: 2.16, 95% confidence interval: 1.03-4.53). Neither income nor occupation was statistically associated with improved functional ability.

CONCLUSIONS

There are education-related inequalities in the improvement of functional ability, especially among older adults with severe disabilities. Health policy makers and practitioners should consider the educational background of individuals with reduced functionality in formulating strategies to improve their functional ability.

Health indicator recording in UK primary care electronic health records: key implications for handling missing data

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.

Constrained empirical-likelihood confidence regions in nonignorable covariate-missing data problems

Missing covariates in regression analysis are a pervasive problem in medical, social, and economic researches. We study empirical-likelihood confidence regions for unconstrained and constrained regression parameters in a nonignorable covariate-missing data problem. For an assumed conditional mean regression model, we assume that some covariates are fully observed but other covariates are missing for some subjects. By exploitation of a probability model of missingness and a working conditional score model from a semiparametric perspective, we build a system of unbiased estimating equations, where the number of equations exceeds the number of unknown parameters. Based on the proposed estimating equations, we introduce unconstrained and constrained empirical-likelihood ratio statistics to construct empirical-likelihood confidence regions for the underlying regression parameters without and with constraints. We establish the asymptotic distributions of the proposed empirical-likelihood ratio statistics. Simulation results show that the proposed empirical-likelihood methods have a better finite-sample performance than other competitors in terms of coverage probability and interval length. Finally, we apply the proposed empirical-likelihood methods to the analysis of a data set from the US National Health and Nutrition Examination Survey.

Long-term Impact of Crystalloid versus Colloid Solutions on Renal Function and Disability-free Survival after Major Abdominal Surgery

Editor’s Perspective

What We Already Know about This Topic

What This Article Tells Us That Is New

Background

The authors recently demonstrated that administration of balanced hydroxyethyl starch solution as part of intraoperative goal-directed fluid therapy was associated with better short-term outcomes than administration of a balanced crystalloid solution in patients having major open abdominal surgery. In the present study, a 1-yr follow-up of renal and disability outcomes in these patients was performed.

Methods

All patients enrolled in the earlier study were followed up 1 yr after surgery for renal function and disability using the World Health Organization Disability Assessment Schedule 2.0 (WHODAS). The main outcome measure was the estimated glomerular filtration rate. Other outcomes were serum creatinine, urea, pruritus, and WHODAS score. Groups were compared on a complete-case analysis basis, and modern imputation methods were then used in mixed-model regressions to assess the stability of the findings taking into account the missing data.

Results

Of the 160 patients enrolled in the original study, follow-up data were obtained for renal function in 129 and for WHODAS score in 114. There were no statistically significant differences in estimated glomerular filtration rate at 1 yr (ml min−1 1.73 m−2): 80 [65 to 92] for crystalloids versus 74 [64 to 94] for colloids; 95% CI [−10 to 7], P = 0.624. However, the WHODAS score (%) was statistically significantly lower in the colloid than in the crystalloid group (2.7 [0 to 12] vs. 7.6 [1.3 to 18]; P = 0.015), and disability-free survival was higher (79% vs. 60%; 95% CI [2 to 39]; P = 0.024).

Conclusions

In patients undergoing major open abdominal surgery, there was no evidence of a statistically significant difference in long-term renal function between a balanced hydroxyethyl starch and a balanced crystalloid solution used as part of intraoperative goal-directed fluid therapy, although there was only limited power to rule out a clinically significant difference. However, disability-free survival was significantly higher in the colloid than in the crystalloid group.

A guide to missing data for the pediatric nephrologist

Missing data is an important and common source of bias in clinical research. Readers should be alert to and consider the impact of missing data when reading studies. Beyond preventing missing data in the first place, through good study design and conduct, there are different strategies available to handle data containing missing observations. Complete case analysis is often biased unless data are missing completely at random. Better methods of handling missing data include multiple imputation and models using likelihood-based estimation. With advancing computing power and modern statistical software, these methods are within the reach of clinician-researchers under guidance of a biostatistician. As clinicians reading papers, we need to continue to update our understanding of statistical methods, so that we understand the limitations of these techniques and can critically interpret literature.

Health administrative data enrichment using cohort information: Comparative evaluation of methods by simulation and application to real data

Background

Studies using health administrative databases (HAD) may lead to biased results since information on potential confounders is often missing. Methods that integrate confounder data from cohort studies, such as multivariate imputation by chained equations (MICE) and two-stage calibration (TSC), aim to reduce confounding bias. We provide new insights into their behavior under different deviations from representativeness of the cohort.

Methods

We conducted an extensive simulation study to assess the performance of these two methods under different deviations from representativeness of the cohort. We illustrate these approaches by studying the association between benzodiazepine use and fractures in the elderly using the general sample of French health insurance beneficiaries (EGB) as main database and two French cohorts (Paquid and 3C) as validation samples.

Results

When the cohort was representative from the same population as the HAD, the two methods are unbiased. TSC was more efficient and faster but its variance could be slightly underestimated when confounders were non-Gaussian. If the cohort was a subsample of the HAD (internal validation) with the probability of the subject being included in the cohort depending on both exposure and outcome, MICE was unbiased while TSC was biased. The two methods appeared biased when the inclusion probability in the cohort depended on unobserved confounders.

Conclusion

When choosing the most appropriate method, epidemiologists should consider the origin of the cohort (internal or external validation) as well as the (anticipated or observed) selection biases of the validation sample.

Overall Survival Trends and Clinical Characteristics of Plasmacytoma in the United States: A National Cancer Database Analysis

BACKGROUND

Given the rarity of plasmacytoma, large-scale database analysis can provide useful information regarding the clinical presentation and patient-related factors impacting overall survival (OS).

MATERIALS AND METHODS

The National Cancer Data Base was queried for patients with plasmacytoma between 2004 and 2013, excluding patients with systemic disease. Plasmacytomas were classified as originating in bone (P-bone), in extramedullary tissue (P-EM), or unspecified. Survival was estimated using the Kaplan-Meier and log-rank test method. We used Cox regression to determine specific outcomes adjusting for demographic, socioeconomic, geographic, facility type, year of diagnosis, and comorbid factors.

RESULTS

In total, 6225 patients were identified, of which 61.5% were men. The median age at diagnosis was 64 years (range, 18-90 years), and the median follow-up was 58 months. The primary site of disease was P-bone in 4056 (65.1%) patients and P-EM in 1468 (23.6%), and the remaining 701 patients were P-unspecified. The unadjusted median survival for solitary P-bone was 89 months (95% confidence interval, 82.9-95.0 months), and for solitary P-EM was 117.3 months (95% confidence interval, 108.8 months to not reached). Factors associated with improved OS include younger age, private insurance, higher income, solitary lesion, and lower comorbidity score. Patients with P-bone disease treated at academic facilities had improved OS. Only 65% of patients with solitary plasmacytoma lesions received radiation treatment. Age greater than 75 years and increased distance to treatment facility was associated with a decreased likelihood of receiving radiation.

CONCLUSIONS

This is the largest study examining outcomes of patients with plasmacytoma using a large database analysis, revealing unique aspects of P-EM versus P-bone and underutilization of radiation treatment.

Rebutting Existing Misconceptions About Multiple Imputation as a Method for Handling Missing Data

Missing data is a problem that occurs frequently in many scientific areas. The most sophisticated method for dealing with this problem is multiple imputation. Contrary to other methods, like listwise deletion, this method does not throw away information, and partly repairs the problem of systematic dropout. Although from a theoretical point of view multiple imputation is considered to be the optimal method, many applied researchers are reluctant to use it because of persistent misconceptions about this method. Instead of providing an(other) overview of missing data methods, or extensively explaining how multiple imputation works, this article aims specifically at rebutting these misconceptions, and provides applied researchers with practical arguments supporting them in the use of multiple imputation.

How handling missing data may impact conclusions: A comparison of six different imputation methods for categorical questionnaire data

Objectives:

Missing data is a recurrent issue in many fields of medical research, particularly in questionnaires. The aim of this article is to describe and compare six conceptually different multiple imputation methods, alongside the commonly used complete case analysis, and to explore whether the choice of methodology for handling missing data might impact clinical conclusions drawn from a regression model when data are categorical.

Methods:

In addition to the commonly used complete case analysis, we tested the following six imputation methods: multiple imputation using expectation–maximization with bootstrapping, multiple imputation using multiple correspondence analysis, multiple imputation using latent class analysis, multiple hot deck imputation and multivariate imputation by chained equations with two different model specifications: logistic regression and random forests. The methods are tested on real data from a questionnaire-based study in the Norwegian opioid maintenance treatment programme.

Results:

All methods performed relatively well when the sample size was large (n = 1000). For a smaller sample size (n = 200), the regression estimates depend heavily on the level of missing. When the amount of missing was ⩾20%, in particular, complete case analysis, hot deck and random forests had biased estimates with too low coverage. Multiple imputation using multiple correspondence analysis had the best performance all over.

Conclusion:

The choice of missing handling methodology has a significant impact on the clinical interpretation of the accompanying statistical analyses. With missing data, the choice of whether to impute or not, and choice of imputation method, can influence clinical conclusion drawn from a regression model and should therefore be given sufficient consideration.

Polybrominated diphenyl ethers and incident pregnancy loss: The LIFE Study

BACKGROUND

Polybrominated diphenyl ethers (PBDEs) have not been studied in relation to incident pregnancy loss in human populations, despite their ubiquitous exposure and purported reproductive toxicity.

OBJECTIVES

To investigate the association between preconception serum PBDE concentrations and incident pregnancy loss.

METHODS

A preconception cohort of 501 couples was followed while trying to become pregnant, and for whom serum concentrations of 10 PBDE congeners were measured using gas chromatography-high resolution mass spectrometry. Pregnancy was prospectively identified as a positive home pregnancy test on the day of expected menstruation. Incident pregnancy loss was defined for 344 singleton pregnancies as a conversion to a negative home pregnancy test, menses, or clinical diagnosis depending upon gestational age. Cox proportional hazards models were used to estimate the hazard ratios (HRs) and 95% confidence intervals (CIs) for individual and summed PBDEs and incident pregnancy loss, adjusting for relevant covariates and male partners' information. In sensitivity analyses, inverse probability weighting was used to account for couples not becoming pregnant and, thereby, not at risk for loss.

RESULTS

The incidence of prospectively observed pregnancy loss was 28%, and the serum concentrations of PBDE congeners in females were consistently associated with a higher hazard of incident pregnancy loss. Specifically, statistically significant hazard ratios (HRs) for incident pregnancy loss were observed for lower brominated PBDE congeners: 17 (HR 1.23; CI: 1.07-1.42), 28 (HR 1.25; CI: 1.03-1.52), 66 (HR 1.23; CI: 1.07-1.42), and homolog triBDE (HR: 1.25; CI: 1.05-1.49). Findings were robust to various model specifications explored in sensitivity analyses.

CONCLUSIONS

Maternal preconception serum concentrations of specific PBDE congeners may increase the hazard of incident pregnancy.

Partner Bereavement and Detection of Dementia: A UK-Based Cohort Study Using Routine Health Data

BACKGROUND

In the UK, an estimated one third of people with dementia have not received a diagnosis. Good evidence suggests that dementia risk is increased among widowed individuals; however, it is not clear if they are being diagnosed in routine primary care.

OBJECTIVE

This study aimed to investigate if bereavement influenced the probability of having received a dementia diagnosis.

METHODS

A population-based cohort study using UK electronic health records, between 1997 and 2017, among 247,586 opposite-sex partners. Those experiencing partner bereavement were matched (age, sex, and date of bereavement) to a non-bereaved person living in a partnership. Multivariate cox regression was performed.

RESULTS

Partner bereavement was associated with an increased risk of receiving a diagnosis of dementia in the first three months (hazard ratio (HR) 1.43, 95% CI 1.20-1.71) and first six months (HR 1.24, 95% CI 1.09-1.41), while there was a small reduced risk of getting a dementia diagnosis over all follow-up (HR 0.94, 95% CI 0.89-0.98).

CONCLUSIONS

Partner bereavement appears to lead to a short-term increased risk of the surviving partner receiving a diagnosis of dementia, suggesting that bereavement unmasks existing undiagnosed dementia. Over the longer term, however, bereaved individuals are less likely to have a diagnosis of dementia in their health records than non-bereaved individuals.

Cost-effectiveness, cost-utility and the budget impact of antidepressants versus preventive cognitive therapy with or without tapering of antidepressants

BACKGROUND

As depression has a recurrent course, relapse and recurrence prevention is essential.AimsIn our randomised controlled trial (registered with the Nederlands trial register, identifier: NTR1907), we found that adding preventive cognitive therapy (PCT) to maintenance antidepressants (PCT+AD) yielded substantial protective effects versus antidepressants only in individuals with recurrent depression. Antidepressants were not superior to PCT while tapering antidepressants (PCT/-AD). To inform decision-makers on treatment allocation, we present the corresponding cost-effectiveness, cost-utility and budget impact.

METHOD

Data were analysed (n = 289) using a societal perspective with 24-months of follow-up, with depression-free days and quality-adjusted life years (QALYs) as health outcomes. Incremental cost-effectiveness ratios were calculated and cost-effectiveness planes and cost-effectiveness acceptability curves were derived to provide information about cost-effectiveness. The budget impact was examined with a health economic simulation model.

RESULTS

Mean total costs over 24 months were €6814, €10 264 and €13 282 for AD+PCT, antidepressants only and PCT/-AD, respectively. Compared with antidepressants only, PCT+AD resulted in significant improvements in depression-free days but not QALYs. Health gains did not significantly favour antidepressants only versus PCT/-AD. High probabilities were found that PCT+AD versus antidepressants only and antidepressants only versus PCT/-AD were dominant with low willingness-to-pay thresholds. The budget impact analysis showed decreased societal costs for PCT+AD versus antidepressants only and for antidepressants only versus PCT/-AD.

CONCLUSIONS

Adding PCT to antidepressants is cost-effective over 24 months and PCT with guided tapering of antidepressants in long-term users might result in extra costs. Future studies examining costs and effects of antidepressants versus psychological interventions over a longer period may identify a break-even point where PCT/-AD will become cost-effective.Declaration of interestC.L.H.B. is co-editor of PLOS One and receives no honorarium for this role. She is also co-developer of the Dutch multidisciplinary clinical guideline for anxiety and depression, for which she receives no remuneration. She is a member of the scientific advisory board of the National Insure Institute, for which she receives an honorarium, although this role has no direct relation to this study. C.L.H.B. has presented keynote addresses at conferences, such as the European Psychiatry Association and the European Conference Association, for which she sometimes receives an honorarium. She has presented clinical training workshops, some including a fee. She receives royalties from her books and co-edited books and she developed preventive cognitive therapy on the basis of the cognitive model of A. T. Beck. W.A.N. has received grants from the Netherlands Organisation for Health Research and Development and the European Union and honoraria and speakers' fees from Lundbeck and Aristo Pharma, and has served as a consultant for Daleco Pharma.

Adverse pregnancy and perinatal outcome in patients with recurrent pregnancy loss: Multiple imputation analyses with propensity score adjustment applied to a large-scale birth cohort of the Japan Environment and Children's Study

PROBLEM

Several studies have reported the increased risk of preterm birth, premature rupture of membranes, and low birth weight in patients with recurrent pregnancy loss (RPL). There have been a limited number of large population-based studies examining adverse pregnancy and perinatal outcome after RPL. Multiple-imputed analyses (MIA) adjusting for biases due to missing data is also lacking.

METHOD OF STUDY

A nationwide birth cohort study known as the "Japan Environment and Children's Study (JECS)" was conducted by the Ministry of the Environment. The subjects consisted of 104 102 registered children (including fetuses or embryos).

RESULTS

No increased risk of a congenital anomaly, aneuploidy, neonatal asphyxia, or a small for date infant was observed among the children from women with a history of RPL. A novel increased risk of placental adhesion and uterine infection was found. The adjusted ORs using MIA in women with three or more PL were 1.76 (95% CI, 1.04-2.96) for a stillbirth, 1.68 (1.12-2.52) for a pregnancy loss, 2.53 (1.17-5.47) for placental adhesion, 1.87 (1.37-2.55) and 1.60 (.99-2.57) for mild and severe hypertensive disorders of pregnancy, respectively, 1.94 (1.06-3.55) for uterine infection, 1.28 (1.11-1.47) for caesarean section and .86 (.76-.98) for a male infant.

CONCLUSION

MIA better quantified the risk, which could encourage women who might hesitate to attempt a subsequent pregnancy.

Canonical Causal Diagrams to Guide the Treatment of Missing Data in Epidemiologic Studies

With incomplete data, the "missing at random" (MAR) assumption is widely understood to enable unbiased estimation with appropriate methods. While the need to assess the plausibility of MAR and to perform sensitivity analyses considering "missing not at random" (MNAR) scenarios has been emphasized, the practical difficulty of these tasks is rarely acknowledged. With multivariable missingness, what MAR means is difficult to grasp, and in many MNAR scenarios unbiased estimation is possible using methods commonly associated with MAR. Directed acyclic graphs (DAGs) have been proposed as an alternative framework for specifying practically accessible assumptions beyond the MAR-MNAR dichotomy. However, there is currently no general algorithm for deciding how to handle the missing data given a specific DAG. Here we construct "canonical" DAGs capturing typical missingness mechanisms in epidemiologic studies with incomplete data on exposure, outcome, and confounding factors. For each DAG, we determine whether common target parameters are "recoverable," meaning that they can be expressed as functions of the available data distribution and thus estimated consistently, or whether sensitivity analyses are necessary. We investigate the performance of available-case and multiple-imputation procedures. Using data from waves 1-3 of the Longitudinal Study of Australian Children (2004-2008), we illustrate how our findings can guide the treatment of missing data in point-exposure studies.

Frequent solitary drinking mediates the associations between negative affect and harmful drinking in emerging adults

INTRODUCTION

Depression and anxiety are highly comorbid conditions that are associated with harmful drinking. Social learning theory proposes that depressed or anxious individuals learn that drinking can reduce negative affect, which makes them susceptible to harmful drinking. Consistent with theory, evidence suggests that negative affect increases risk for harmful drinking. But, less is known about mediators of negative affect-pathways to harmful drinking. Research has demonstrated that solitary drinking is an underlying mechanism of harmful drinking among emerging adults with high levels of negative affect. However, much of this work is cross-sectional. We conducted a longitudinal study to examine solitary drinking as a key explanatory variable in the negative-affect pathway to harmful drinking.

METHODS

Emerging adults (N = 308) completed online self-reports of depression and anxiety at Time 1 (the beginning of their first semester in university), and drinking context and harmful drinking at Time 2 (the end of their first semester of university).

RESULTS

Structural equation modeling supported indirect effects from negative affect to harmful drinking via solitary drinking. Specifically, emerging adults with high levels of negative affect at the beginning of their first semester engaged in frequent solitary drinking over the rest of the semester, and subsequently experienced increased harmful drinking (controlling for Time 1 solitary/social drinking and harmful use). Social drinking was not a mediator of this effect.

CONCLUSIONS

The findings reveal that solitary drinking uniquely mediated the relation between negative affect and harmful drinking. Reducing solitary drinking might mitigate negative affect-related risk for harmful drinking in emerging adults.

Association of BMI with overall and cause-specific mortality: a population-based cohort study of 3·6 million adults in the UK

BACKGROUND

BMI is known to be strongly associated with all-cause mortality, but few studies have been large enough to reliably examine associations between BMI and a comprehensive range of cause-specific mortality outcomes.

METHODS

In this population-based cohort study, we used UK primary care data from the Clinical Practice Research Datalink (CPRD) linked to national mortality registration data and fitted adjusted Cox regression models to examine associations between BMI and all-cause mortality, and between BMI and a comprehensive range of cause-specific mortality outcomes (recorded by International Classification of Diseases, 10th revision [ICD-10] codes). We included all individuals with BMI data collected at age 16 years and older and with subsequent follow-up time available. Follow-up began at whichever was the latest of: start of CPRD research-standard follow up, the 5-year anniversary of the first BMI record, or on Jan 1, 1998 (start date for death registration data); follow-up ended at death or on March 8, 2016. Fully adjusted models were stratified by sex and adjusted for baseline age, smoking, alcohol use, diabetes, index of multiple deprivation, and calendar period. Models were fitted in both never-smokers only and the full study population. We also did an extensive range of sensitivity analyses. The expected age of death for men and women aged 40 years at baseline, by BMI category, was estimated from a Poisson model including BMI, age, and sex.

FINDINGS

3 632 674 people were included in the full study population; the following results are from the analysis of never-smokers, which comprised 1 969 648 people and 188 057 deaths. BMI had a J-shaped association with overall mortality; the estimated hazard ratio per 5 kg/m2 increase in BMI was 0·81 (95% CI 0·80-0·82) below 25 kg/m2 and 1·21 (1·20-1·22) above this point. BMI was associated with all cause of death categories except for transport-related accidents, but the shape of the association varied. Most causes, including cancer, cardiovascular diseases, and respiratory diseases, had a J-shaped association with BMI, with lowest risk occurring in the range 21-25 kg/m2. For mental and behavioural, neurological, and accidental (non-transport-related) causes, BMI was inversely associated with mortality up to 24-27 kg/m2, with little association at higher BMIs; for deaths from self-harm or interpersonal violence, an inverse linear association was observed. Associations between BMI and mortality were stronger at younger ages than at older ages, and the BMI associated with lowest mortality risk was higher in older individuals than in younger individuals. Compared with individuals of healthy weight (BMI 18·5-24·9 kg/m2), life expectancy from age 40 years was 4·2 years shorter in obese (BMI ≥30·0 kg/m2) men and 3·5 years shorter in obese women, and 4·3 years shorter in underweight (BMI <18·5 kg/m2) men and 4·5 years shorter in underweight women. When smokers were included in analyses, results for most causes of death were broadly similar, although marginally stronger associations were seen among people with lower BMI, suggesting slight residual confounding by smoking.

INTERPRETATION

BMI had J-shaped associations with overall mortality and most specific causes of death; for mental and behavioural, neurological, and external causes, lower BMI was associated with increased mortality risk.

FUNDING

Wellcome Trust.

Economic Evaluation of an Internet-Based Preventive Cognitive Therapy With Minimal Therapist Support for Recurrent Depression: Randomized Controlled Trial

Background

Major depressive disorder (MDD) is highly recurrent and has a significant disease burden. Although the effectiveness of internet-based interventions has been established for the treatment of acute MDD, little is known about their cost effectiveness, especially in recurrent MDD.

Objectives

Our aim was to evaluate the cost effectiveness and cost utility of an internet-based relapse prevention program (mobile cognitive therapy, M-CT).

Methods

The economic evaluation was performed alongside a single-blind parallel group randomized controlled trial. Participants were recruited via media, general practitioners, and mental health care institutions. In total, 288 remitted individuals with a history of recurrent depression were eligible, of whom 264 were randomly allocated to M-CT with minimal therapist support added to treatment as usual (TAU) or TAU alone. M-CT comprised 8 online lessons, and participants were advised to complete 1 lesson per week. The economic evaluation was performed from a societal perspective with a 24-month time horizon. The health outcomes were number of depression-free days according to Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, (DSM-IV) criteria assessed with the Structured Clinical Interview for DSM-IV axis I disorders by blinded interviewers after 3, 12, and 24 months. Quality-adjusted life years (QALYs) were self-assessed with the three level version of the EuroQol Five Dimensional Questionnaire (EQ-5D-3L). Costs were assessed with the Trimbos and Institute for Medical Technology Assessment Questionnaire on Costs Associated with Psychiatric Illness (TiC-P). Incremental cost-effectiveness ratios were calculated and cost-effectiveness planes and cost-effectiveness acceptability curves were displayed to assess the probability that M-CT is cost effective compared to TAU.

Results

Mean total costs over 24 months were €8298 (US $9415) for M-CT and €7296 (US $8278) for TAU. No statistically significant differences were found between M-CT and TAU regarding depression-free days and QALYs (P=.37 and P=.92, respectively). The incremental costs were €179 (US $203) per depression-free day and €230,816 (US $261,875) per QALY. The cost-effectiveness acceptability curves suggested that for depression-free days, high investments have to be made to reach an acceptable probability that M-CT is cost effective compared to TAU. Regarding QALYs, considerable investments have to be made but the probability that M-CT is cost effective compared to TAU does not rise above 40%.

Conclusions

The results suggest that adding M-CT to TAU is not effective and cost effective compared to TAU alone. Adherence rates were similar to other studies and therefore do not explain this finding. The participants scarcely booked additional therapist support, resulting in 17.3 minutes of mean total therapist support. More studies are needed to examine the cost effectiveness of internet-based interventions with respect to long-term outcomes and the role and optimal dosage of therapist support. Overall, more research is needed on scalable and cost-effective interventions that can reduce the burden of recurrent MDD.

Trial registration

Netherlands Trial Register NTR2503; http://www.trialregister.nl/trialreg/admin/rctview.asp?TC=2503 (Archived by WebCite at http://www.webcitation.org/73aBn41r3)

Treating loss-to-follow-up as a missing data problem: a case study using a longitudinal cohort of HIV-infected patients in Haiti

Background

HIV programs are often assessed by the proportion of patients who are alive and retained in care; however some patients are categorized as lost to follow-up (LTF) and have unknown vital status. LTF is not an outcome but a mixed category of patients who have undocumented death, transfer and disengagement from care. Estimating vital status (dead versus alive) among this category is critical for survival analyses and program evaluation.

Methods

We used three methods to estimate survival in the cohort and to ascertain factors associated with death among the first cohort of HIV positive patients to receive antiretroviral therapy in Haiti: complete case (CC) (drops missing), Inverse Probability Weights (IPW) (uses tracking data) and Multiple Imputation with Chained Equations (MICE) (imputes missing data). Logistic regression was used to calculate odds ratios and 95% confidence intervals for adjusted models for death at 10 years. The logistic regression models controlled for sex, age, severe poverty (living on <$1 USD per day), Port-au-Prince residence and baseline clinical characteristics of weight, CD4, WHO stage and tuberculosis diagnosis.

Results

Age, severe poverty, baseline weight and WHO stage were statistically significant predictors of AIDS related mortality across all models. Gender was only statistically significant in the MICE model but had at least a 10% difference in odds ratios across all models.

Conclusion

Each of these methods had different assumptions and differed in the number of observations included due to how missing values were addressed. We found MICE to be most robust in predicting survival status as it allowed us to impute missing data so that we had the maximum number of observations to perform regression analyses. MICE also provides a complementary alternative for estimating survival among patients with unassigned vital status. Additionally, the results were easier to interpret, less likely to be biased and provided an alternative to a problem that is often commented upon in the extant literature.

Electronic supplementary material

The online version of this article (10.1186/s12889-018-6115-0) contains supplementary material, which is available to authorized users.

Missing data treatments matter: an analysis of multiple imputation for anterior cervical discectomy and fusion procedures

BACKGROUND CONTEXT

The presence of missing data is a limitation of large datasets, including the National Surgical Quality Improvement Program (NSQIP). In addressing this issue, most studies use complete case analysis, which excludes cases with missing data, thus potentially introducing selection bias. Multiple imputation, a statistically rigorous approach that approximates missing data and preserves sample size, may be an improvement over complete case analysis.

PURPOSE

The present study aims to evaluate the impact of using multiple imputation in comparison with complete case analysis for assessing the associations between preoperative laboratory values and adverse outcomes following anterior cervical discectomy and fusion (ACDF) procedures.

STUDY DESIGN/SETTING

This is a retrospective review of prospectively collected data.

PATIENT SAMPLE

Patients undergoing one-level ACDF were identified in NSQIP 2012-2015.

OUTCOME MEASURES

Perioperative adverse outcome variables assessed included the occurrence of any adverse event, severe adverse events, and hospital readmission.

METHODS

Missing preoperative albumin and hematocrit values were handled using complete case analysis and multiple imputation. These preoperative laboratory levels were then tested for associations with 30-day postoperative outcomes using logistic regression.

RESULTS

A total of 11,999 patients were included. Of this cohort, 63.5% of patients had missing preoperative albumin and 9.9% had missing preoperative hematocrit. When using complete case analysis, only 4,311 patients were studied. The removed patients were significantly younger, healthier, of a common body mass index, and male. Logistic regression analysis failed to identify either preoperative hypoalbuminemia or preoperative anemia as significantly associated with adverse outcomes. When employing multiple imputation, all 11,999 patients were included. Preoperative hypoalbuminemia was significantly associated with the occurrence of any adverse event and severe adverse events. Preoperative anemia was significantly associated with the occurrence of any adverse event, severe adverse events, and hospital readmission.

CONCLUSIONS

Multiple imputation is a rigorous statistical procedure that is being increasingly used to address missing values in large datasets. Using this technique for ACDF avoided the loss of cases that may have affected the representativeness and power of the study and led to different results than complete case analysis. Multiple imputation should be considered for future spine studies.

A comparison of different methods to handle missing data in the context of propensity score analysis

Propensity score analysis is a popular method to control for confounding in observational studies. A challenge in propensity methods is missing values in confounders. Several strategies for handling missing values exist, but guidance in choosing the best method is needed. In this simulation study, we compared four strategies of handling missing covariate values in propensity matching and propensity weighting. These methods include: complete case analysis, missing indicator method, multiple imputation and combining multiple imputation and missing indicator method. Concurrently, we aimed to provide guidance in choosing the optimal strategy. Simulated scenarios varied regarding missing mechanism, presence of effect modification or unmeasured confounding. Additionally, we demonstrated how missingness graphs help clarifying the missing structure. When no effect modification existed, complete case analysis yielded valid causal treatment effects even when data were missing not at random. In some situations, complete case analysis was also able to partially correct for unmeasured confounding. Multiple imputation worked well if the data were missing (completely) at random, and if the imputation model was correctly specified. In the presence of effect modification, more complex imputation models than default options of commonly used statistical software were required. Multiple imputation may fail when data are missing not at random. Here, combining multiple imputation and the missing indicator method reduced the bias as the missing indicator variable can be a proxy for unobserved confounding. The optimal way to handle missing values in covariates of propensity score models depends on the missing data structure and the presence of effect modification. When effect modification is present, default settings of imputation methods may yield biased results even if data are missing at random.

Behavioral predictors of improved speech output in minimally verbal children with autism

We investigated the relationship between eight theoretically motivated behavioral variables and a spoken-language-related outcome measure, after 25 sessions of treatment for speech production in 38 minimally verbal children with autism. After removing potential predictors that were uncorrelated with the outcome variable, two remained. We used both complete-case and multiple-imputation analyses to address missing predictor data and performed linear regressions to identify significant predictors of change in percent syllables approximately correct after treatment. Baseline phonetic inventory (the number of English phonemes repeated correctly) was the most robust predictor of improvement. In the group of 17 participants with complete data, ADOS score also significantly predicted the outcome. In contrast to some earlier studies, nonverbal IQ, baseline levels of expressive language, and younger age did not significantly predict improvement. The present results are not only consistent with previous studies showing that verbal imitation and autism severity significantly predict spoken language outcomes in preschool-aged minimally verbal children with autism, but also extend these findings to older minimally verbal children with autism. Autism Res 2018, 11: 1356-1365. © 2018 International Society for Autism Research, Wiley Periodicals, Inc. LAY SUMMARY: We wished to understand what baseline factors predicted whether minimally verbal children with autism would improve after treatment for spoken language. The outcome measure was change in percentage (%) syllables approximately correct on a set of 30 two-syllable words or phrases. Fifteen were both practiced in treatment and tested; the remainder were not practiced in treatment, but only tested, to assess how well children were able to generalize their new skills to an untrained set of words. Potential predictors tested were sex, age, expressive language, phonetic inventory (the number of English speech sounds repeated correctly), autism severity, and nonverbal IQ. Phonetic inventory and (for some children) autism severity predicted children's posttreatment improvement. Nonverbal IQ and expressive language ability did not predict improvement, nor did younger age, suggesting that some older children with autism may be candidates for speech therapy.

Association of the kidney allocation system with dialysis exposure before deceased donor kidney transplantation by preemptive wait-listing status

BACKGROUND

It is unknown whether the new kidney transplant allocation system (KAS) has attenuated the advantages of preemptive wait-listing as a strategy to minimize pretransplant dialysis exposure.

METHODS

We performed a retrospective study of adult US deceased donor kidney transplant (DDKT) recipients between December 4, 2011-December 3, 2014 (pre-KAS) and December 4, 2014-December 3, 2017 (post-KAS). We estimated pretransplant dialysis durations by preemptive listing status in the pre- and post-KAS periods using multivariable gamma regression models.

RESULTS

Among 65 385 DDKT recipients, preemptively listed recipients (21%, n = 13 696) were more likely to be white (59% vs 34%, P < 0.001) and have private insurance (64% vs 30%, P < 0.001). In the pre- and post-KAS periods, average adjusted pretransplant dialysis durations for preemptively listed recipients were <2 years in all racial groups. Compared to recipients who were listed after starting dialysis, preemptively listed recipients experienced 3.85 (95% Confidence Interval [CI] 3.71-3.99) and 4.53 (95% CI 4.32-4.74) fewer average years of pretransplant dialysis in the pre- and post-KAS periods, respectively (P < 0.001 for all comparisons).

CONCLUSIONS

Preemptively wait-listed DDKT recipients continue to experience substantially fewer years of pretransplant dialysis than recipients listed after dialysis onset. Efforts are needed to improve both socioeconomic and racial disparities in preemptive wait-listing.

Food Services Using Energy- and Protein-Fortified Meals to Assist Vulnerable Community-Residing Older Adults Meet Their Dietary Requirements and Maintain Good Health and Quality of Life: Findings from a Pilot Study

The effects of “standard (STD)” vs. “protein- and energy-enriched (HEHP)” food-service meals on the nutrient intake, nutritional status, functional capacity, and wellbeing of older adults was investigated using a 12 week, double-blinded, parallel group design. All participants received dietetics counseling and either an STD (2.3 MJ and 30 g protein per meal) or a HEHP (4.6 MJ and 60 g protein) hot lunchtime meal for at least 3 days/week; those who did not want food-service meals were included in the control group (CON). Twenty-nine participants completed the study (STD = 7; HEHP = 12; CON = 10). From baseline to week 12, the HEHP subjects increased their mean daily energy intake from 6151 ± 376 kJ to 8228 ± 642 kJ (p = 0.002 for effect of time) and protein intake from 67 ± 4 g to 86 ± 8 g (p = 0.014 for effect of time). The MNA (Mini Nutritional Assessment) score was increased significantly in HEHP by 4.0 ± 1.1 points (p = 0.001), but not in the STD and CON groups (2.8 ± 2.1 points and 1.8 ± 1.1 points, p > 0.05). No difference was found for other clinical outcomes between the groups. The findings indicate that provision of HEHP-fortified food-service meals can increase energy and protein intake and improve the nutritional status of nutritionally at-risk older people.

Avoidance, pacing, or persistence in multidisciplinary functional rehabilitation for chronic musculoskeletal pain: An observational study with cross-sectional and longitudinal analyses

Background

Three main activity patterns have been distinguished in describing chronic pain (avoidance, pacing and persistence). However, their influence on patient outcomes remains a question of debate. This observational study aimed to measure the associations between the avoidance, pacing, and persistence (labelled overdoing) scales of the Patterns of Activity Measure–Pain (POAM-P), self-reported outcomes (pain-interference, depression, functional ability), and observational outcomes (walking, lifting test, physical fitness).

Methods

We conducted an observational study with cross-sectional and longitudinal analyses. The data were collected prospectively before and after treatment, which was a 5-week functional rehabilitation including vocational aspects. In addition to self-reported and observational outcomes, patients were asked if they thought they would be able to return to work at 6 months. Analyses were conducted with treatment effect sizes, correlations, and multiple regression models.

Results

In this sample (891 patients), we found on average small to moderate improvements for pain-interference and observational outcomes (Cohen’s d: 0.37 to 0.64). According to the multivariable models, overdoing was associated with most of the beneficial psychosocial and observational outcomes (β -0.13 to 0.17; all p<0.01). Avoidance was related to negative psychosocial outcomes before treatment (β -0.09 to 0.17; all p<0.015). Pacing, which had moderate correlation with avoidance (r = 0.46), was not associated with most of the outcomes. The feeling that the goal of returning to work was attainable was associated with lower avoidance scores (adjusted OR 0.97; p = 0.024).

Conclusions

The overdoing POAM-P scale probably measures a task-contingent persistence, which appears appropriate in the setting of this study. Persistent behavior was indeed related to small or moderate positive biopsychosocial outcomes, before and after treatment. Moreover feeling able to return to work was related to lower avoidance. Further studies should test the efficacy of motivational strategies that may promote functional task-contingent persistence and reduce avoidance of painful tasks.

Comparison of intent-to-treat analysis strategies for pre-post studies with loss to follow-up

In pre-post studies when all outcomes are completely observed, previous studies have shown that analysis of covariance (ANCOVA) is more powerful than a change-score analysis in testing the treatment effect. However, there have been few studies comparing power under missing post-test values. This paper was motivated by the Behavior and Exercise for Physical Health Intervention (BePHIT) Study, a pre-post study designed to compare two interventions on postmenopausal women's walk time. The goal of this study was to compare the power of two methods which adhere to the intent-to-treat (ITT) principle when post-test data are missing: ANCOVA after multiple imputation (MI) and the mixed model applied to all-available data (AA). We also compared the two ITT analysis strategies to two methods which do not adhere to ITT principles: complete-case (CC) ANCOVA and the CC mixed model. Comparisons were made through analyses of the BePHIT data and simulation studies conducted under various sample sizes, missingness rates, and missingness scenarios. In the analysis of the BePHIT data, ANCOVA after MI had the smallest p-value for the test of the treatment effect of the four methods. Simulation results demonstrated that the AA mixed model was usually more powerful than ANCOVA after MI. The power of ANCOVA after MI dropped the fastest as the missingness rate increased; in most simulated scenarios, ANCOVA after MI had the smallest power when 50% of the post-test outcomes were missing.

Is frailty a stable predictor of mortality across time? Evidence from the Cognitive Function and Ageing Studies

Background

age-specific mortality reduction has been accompanied by a decrease in the prevalence of some diseases and an increase in others. Whether populations are becoming ‘healthier’ depends on which aspect of health is being considered. Frailty has been proposed as an integrative measure to quantify health status.

Objective

to investigate changes in the near-term lethality of frailty before and after a 20-year interval using the frailty index (FI), a summary of age-related health deficit accumulation.

Design

baseline data from the Cognitive Function and Ageing Studies (CFAS) in 1991 (n = 7,635) and 2011 (n = 7,762).

Setting

three geographically distinct UK centres (Newcastle, Cambridgeshire and Nottingham).

Subjects

individuals aged 65 and over (both institutionalised and community-living).

Methods

a 30-item frailty score was used, which includes morbidities, risk factors and subjective measures of disability. Missing items were imputed using multiple imputations by chained equations. Binomial regression was used to investigate the relationship between frailty, age, sex and cohort. Two-year mortality was modelled using logistic regression.

Results

mean frailty was slightly higher in CFAS II (0.19, 95% confidence interval (CI): 0.19–0.20) than CFAS I (0.18, 95% CI: 0.17–0.18). Two-year mortality in CFAS I was higher than in CFAS II (odds ratio (OR) = 1.16, 95% CI: 1.03–1.30). The association between frailty and 2-year mortality was non-linear with an OR of ~1.6 for each 0.10 increment in the FI.

Conclusions

the relationship between frailty and mortality did not significantly differ across the studies. Severe frailty as an indicator of mortality is shown to be a stable construct.

Predictive Value of the Fear-Avoidance Model on Functional Capacity Evaluation

Purpose Measuring the predictive value of the Fear-Avoidance Model (FAM) on lifting tasks in Functional Capacity Evaluation (FCE), and on reasons for stopping the evaluation (safe maximal effort, versus self-limited). Methods A monocentric prospective study was conducted on 298 consecutive inpatients. Components of the FAM were analyzed using the Cumulative Psychosocial Factor Index (CPFI: kinesiophobia, catastrophizing, depressive mood) and perceived disability (Hand/Spinal Function Sort: HFS/SFS). Floor-to-waist, waist-to-overhead and dominant-hand lifting tests were measured according to the FCE guidelines. Maximal safe performance was judged by certified FCE assessors. Analyses were conducted with linear multiple regression models. Results The CPFI was significantly associated with the 3 FCE lifting tests: floor-to-waist (ß = - 1.12; p = 0.039), waist-to-overhead (ß = - 0.88; p = 0.011), and dominant-handed lifting (ß = - 1.21; p = 0.027). Higher perceived disability was also related to lower performances: floor-to-waist (ß = 0.09; p < 0.001), waist-to-overhead (ß = 0.04; p < 0.001), and dominant-handed lifting (ß = 0.06; p < 0.001). The CPFI was not related to performances of patients with self-limited effort despite higher psychological scores, while a relationship was found for patients who achieved a safe maximal performance. Higher perceived disability was related to performances in both situations. Conclusions FAM components should be taken into account when interpreting maximal physical performance in FCE. This study also suggests that factors other than pain-related fears may influence patients with self-limited effort.

Missing Data Techniques and the Statistical Conclusion Validity of Survey-Based Alcohol and Drug Use Research Studies: A Review and Comment on Reproducibility

The goal of the present review was to examine whether or not the use of modern missing data techniques impacts the statistical conclusion validity of research on alcohol and drug use outcomes in survey-based research studies. We identified 28 papers and received complete case data from the authors of 12 studies. Seven studies (25%) reported the missing data pattern (missing not at random [MNAR], missing at random [MAR], missing completely at random [MCAR]), 15 studies (53.6%) indicated the amount of missing observations in the data set, and a significant proportion of studies ( n = 13, 46.4%) did not report any of the conditions or assumptions under which the missing data analysis was performed or implemented. Six of the 12 (50%) studies analyzed reported a different number of statistically significant associations between the complete case and full sample analyses. Efforts should be made to make missing data analysis more accessible, easy to implement and report to improve transparency and reproducibility of findings.

Does pattern mixture modelling reduce bias due to informative attrition compared to fitting a mixed effects model to the available cases or data imputed using multiple imputation?: a simulation study

BACKGROUND

Informative attrition occurs when the reason participants drop out from a study is associated with the study outcome. Analysing data with informative attrition can bias longitudinal study inferences. Approaches exist to reduce bias when analysing longitudinal data with monotone missingness (once participants drop out they do not return). However, findings may differ when using these approaches to analyse longitudinal data with non-monotone missingness.

METHODS

Different approaches to reduce bias due to informative attrition in non-monotone longitudinal data were compared. To achieve this aim, we simulated data from a Whitehall II cohort epidemiological study, which used the slope coefficients from a linear mixed effects model to investigate the association between smoking status at baseline and subsequent decline in cognition scores. Participants with lower cognitive scores were thought to be more likely to drop out. By using a simulation study, a range of scenarios using distributions of variables which exist in real data were compared. Informative attrition that would introduce a known bias to the simulated data was specified and the estimates from a mixed effects model with random intercept and slopes when fitted to: available cases; data imputed using multiple imputation (MI); imputed data adjusted using pattern mixture modelling (PMM) were compared. The two-fold fully conditional specification MI approach, previously validated for non-monotone longitudinal data under ignorable missing data assumption, was used. However, MI may not reduce bias because informative attrition is non-ignorable missing. Therefore, PMM was applied to reduce the bias, usually unknown, by adjusting the values imputed with MI by a fixed value equal to the introduced bias.

RESULTS

With highly correlated repeated outcome measures, the slope coefficients from a mixed effects model were found to have least bias when fitted to available cases. However, for moderately correlated outcome measurements, the slope coefficients from fitting a mixed effects model to data adjusted using PMM were least biased but still underestimated the true coefficients.

CONCLUSIONS

PMM may potentially reduce bias in studies analysing longitudinal data with suspected informative attrition and moderately correlated repeated outcome measurements. Including additional auxiliary variables in the imputation model may also reduce any remaining bias.

Association of Broad-Based Genomic Sequencing With Survival Among Patients With Advanced Non-Small Cell Lung Cancer in the Community Oncology Setting

IMPORTANCE

Broad-based genomic sequencing is being used more frequently for patients with advanced non-small cell lung cancer (NSCLC). However, little is known about the association between broad-based genomic sequencing and treatment selection or survival among patients with advanced NSCLC in a community oncology setting.

OBJECTIVE

To compare clinical outcomes between patients with advanced NSCLC who received broad-based genomic sequencing vs a control group of patients who received routine testing for EGFR mutations and/or ALK rearrangements alone.

DESIGN, SETTING, AND PARTICIPANTS

Retrospective cohort study of patients with chart-confirmed advanced NSCLC between January 1, 2011, and July 31, 2016, and who received care at 1 of 191 oncology practices across the United States using the Flatiron Health Database. Patients were diagnosed with stage IIIB/IV or unresectable nonsquamous NSCLC who received at least 1 line of antineoplastic treatment.

EXPOSURES

Receipt of either broad-based genomic sequencing or routine testing (EGFR and/or ALK only). Broad-based genomic sequencing included any multigene panel sequencing assay examining more than 30 genes prior to third-line treatment.

MAIN OUTCOMES AND MEASURES

Primary outcomes were 12-month mortality and overall survival from the start of first-line treatment. Secondary outcomes included frequency of genetic alterations and treatments received.

RESULTS

Among 5688 individuals with advanced NSCLC (median age, 67 years [interquartile range, 41-85], 63.6% white, 80% with a history of smoking); 875 (15.4%) received broad-based genomic sequencing and 4813 (84.6%) received routine testing. Among patients who received broad-based genomic sequencing, 4.5% received targeted treatment based on testing results, 9.8% received routine EGFR/ALK targeted treatment, and 85.1% received no targeted treatment. Unadjusted mortality rates at 12 months were 49.2% for patients undergoing broad-based genomic sequencing and 35.9% for patients undergoing routine testing. Using an instrumental variable analysis, there was no significant association between broad-based genomic sequencing and 12-month mortality (predicted probability of death at 12 months, 41.1% for broad-based genomic sequencing vs 44.4% for routine testing; difference -3.6% [95% CI, -18.4% to 11.1%]; P = .63). The results were consistent in the propensity score-matched survival analysis (42.0% vs 45.1%; hazard ratio, 0.92 [95% CI, 0.73 to 1.11]; P = .40) vs unmatched cohort (hazard ratio, 0.69 [95% CI, 0.62 to 0.77]; log-rank P < .001).

CONCLUSIONS AND RELEVANCE

Among patients with advanced non-small cell lung cancer receiving care in the community oncology setting, broad-based genomic sequencing directly informed treatment in a minority of patients and was not independently associated with better survival.

Attrition Bias Related to Missing Outcome Data: A Longitudinal Simulation Study

BACKGROUND

Most longitudinal studies do not address potential selection biases due to selective attrition. Using empirical data and simulating additional attrition, we investigated the effectiveness of common approaches to handle missing outcome data from attrition in the association between individual education level and change in body mass index (BMI).

METHODS

Using data from the two waves of the French RECORD Cohort Study (N = 7,172), we first examined how inverse probability weighting (IPW) and multiple imputation handled missing outcome data from attrition in the observed data (stage 1). Second, simulating additional missing data in BMI at follow-up under various missing-at-random scenarios, we quantified the impact of attrition and assessed how multiple imputation performed compared to complete case analysis and to a perfectly specified IPW model as a gold standard (stage 2).

RESULTS

With the observed data in stage 1, we found an inverse association between individual education and change in BMI, with complete case analysis, as well as with IPW and multiple imputation. When we simulated additional attrition under a missing-at-random pattern (stage 2), the bias increased with the magnitude of selective attrition, and multiple imputation was useless to address it.

CONCLUSIONS

Our simulations revealed that selective attrition in the outcome heavily biased the association of interest. The present article contributes to raising awareness that for missing outcome data, multiple imputation does not do better than complete case analysis. More effort is thus needed during the design phase to understand attrition mechanisms by collecting information on the reasons for dropout.

Inferior long-term allograft and patient outcomes among recipients of offspring living donor kidneys

While offspring-to-parent living donor kidney transplantations may represent an ideal donor-recipient combination to optimize long-term transplantation outcomes, the sex-specific long-term success of these transplantations remains unclear. We hypothesize that allograft and recipient survivals in offspring-to-parent living donor kidney transplantation differ between men and women due to donor-specific alloimmunization during pregnancy. We retrospectively analyzed long-term allograft and patient survival among men and women who received an offspring living donor kidney compared with those who received other haplotype-matched living donor kidneys. Based on multivariable Cox proportional hazards modeling of Organ Procurement and Transplantation Network data from 2001 to 2015, we found that both men and women who received offspring living donor kidneys had significantly increased mortality compared with recipients who received nonoffspring living donor kidneys. While male recipients of any living donor kidney had greater risk of mortality and allograft failure than female recipients, there was no significant difference in all-cause allograft failure or mortality in male versus female recipients of offspring living donor kidney transplantations. Our analysis demonstrated no significant interaction between recipient sex and donor offspring status. We conclude that nonoffspring living donors should be considered whenever feasible for both men and women with multiple donor options.

Shame mediates the relationship between depression and addictive behaviours

BACKGROUND

Alcohol and gambling problems are common in young adults. Self-medication theory states that young adults with depression drink and/or gamble to escape negative emotions. Research shows that depression is a risk factor for drinking/gambling problems, but more work is needed to examine mediators underlying these associations. One potential mediator is shame. Shame is a self-directed emotion that follows a negative life event and is characterized by intense feelings of inferiority, worthlessness, and embarrassment. Depressed individuals are especially susceptible to shame (and associated emotions). Shame has also been implicated in risk for addiction. Accordingly, we predicted that elevated shame would explain why depression is associated with both alcohol and gambling problems.

METHODS

A longitudinal design was used to examine this hypothesis. Undergraduates (N = 210) completed self-reports of depression at baseline (Time 1) and then completed self-reports of shame, alcohol misuse, gambling problems one month later (Time 2).

RESULTS

Results showed that individuals with elevated depression at Time 1 endorsed high levels of shame at Time 2, which in turn predicted more gambling (β = .038, 95% CI [.010, .087]) and alcohol problems (β = .249, 95% CI [.123, .435]) at Time 2. We found that increased levels of shame explained the effects of depression on problem drinking and gambling.

CONCLUSIONS

Study findings improve our understanding of the depressive pathway to addiction by providing evidence for shame as a potential mechanism of this pathway.

IMPACT

Reducing shame can be a target of clinical interventions for young adults with depression and alcohol/gambling problems.

Strategies for imputing missing covariates in accelerated failure time models

Missing covariates often occur in biomedical studies with survival outcomes. Multiple imputation via chained equations (MICE) is a semi-parametric and flexible approach that imputes multivariate data by a series of conditional models, one for each incomplete variable. When applying MICE, practitioners tend to specify the conditional models in a simple fashion largely dictated by the software, which could lead to suboptimal results. Practical guidelines for specifying appropriate conditional models in MICE are lacking. Motivated by a study of time to hip fractures in the Women's Health Initiative Observational Study using accelerated failure time models, we propose and experiment with some rationales leading to appropriate MICE specifications. This strategy starts with specifying a joint model for the variables involved. We first derive the conditional distribution of each variable under the joint model, then approximate these conditional distributions to the extent which can be characterized by commonly used regression models. We propose to fit separate models to impute incomplete variables by the failure status, which is key to generating appropriate MICE specifications for survival outcomes. The proposed strategy can be conveniently implemented with all available imputation software that uses fully conditional specifications. Our simulation results show that some commonly used simple MICE specifications can produce suboptimal results, while those based on the proposed strategy appear to perform well and be robust toward model misspecifications. Hence, we warn against a mechanical use of MICE and suggest careful modeling of the conditional distributions of variables to ensure proper performance.

Promises and pitfalls of electronic health record analysis

Routinely collected electronic health records (EHRs) are increasingly used for research. With their use comes the opportunity for large-scale, high-quality studies that can address questions not easily answered by randomised clinical trials or classical cohort studies involving bespoke data collection. However, the use of EHRs generates challenges in terms of ensuring methodological rigour, a potential problem when studying complex chronic diseases such as diabetes. This review describes the promises and potential of EHRs in the context of diabetes research and outlines key areas for caution with examples. We consider the difficulties in identifying and classifying diabetes patients, in distinguishing between prevalent and incident cases and in dealing with the complexities of diabetes progression and treatment. We also discuss the dangers of introducing time-related biases and describe the problems of inconsistent data recording, missing data and confounding. Throughout, we provide practical recommendations for good practice in conducting EHR studies and interpreting their results.