Evaluating the impact of unmeasured confounding with internal validation data: an example cost evaluation in type 2 diabetes

unmconf : an R package for Bayesian regression with unmeasured confounders

The inability to correctly account for unmeasured confounding can lead to bias in parameter estimates, invalid uncertainty assessments, and erroneous conclusions. Sensitivity analysis is an approach to investigate the impact of unmeasured confounding in observational studies. However, the adoption of this approach has been slow given the lack of accessible software. An extensive review of available R packages to account for unmeasured confounding list deterministic sensitivity analysis methods, but no R packages were listed for probabilistic sensitivity analysis. The R package unmconf implements the first available package for probabilistic sensitivity analysis through a Bayesian unmeasured confounding model. The package allows for normal, binary, Poisson, or gamma responses, accounting for one or two unmeasured confounders from the normal or binomial distribution. The goal of unmconf is to implement a user friendly package that performs Bayesian modeling in the presence of unmeasured confounders, with simple commands on the front end while performing more intensive computation on the back end. We investigate the applicability of this package through novel simulation studies. The results indicate that credible intervals will have near nominal coverage probability and smaller bias when modeling the unmeasured confounder(s) for varying levels of internal/external validation data across various combinations of response-unmeasured confounder distributional families.

Real-world evidence: a practical toolbox for collecting health state utilities

Health state utilities (HSU) data collected in real-world evidence studies are at risk of bias. Although numerous guidance documents are available, practical advice to avoid bias in HSU studies is limited. Thus, the objective of this article was to develop a concise toolbox intended for investigators seeking to collect HSU in a real-world setting. The proposed toolbox builds on existing guidance and provides practical steps to help investigators perform good quality research. The toolbox aims at increasing the credibility of HSU data for future reimbursement decision making.

Cancer-Related Pain Is an Independent Predictor of In-Hospital Opioid Overdose: A Propensity-Matched Analysis

BACKGROUND

About 50% of patients with cancer who have undergone surgery suffer from cancer-related pain (CP). The use of opioids for postoperative pain management presents the potential for overdose, especially among these patients.

OBJECTIVE

The primary objective of this study was to determine the association between CP and postoperative opioid overdose among inpatients who had undergone major elective procedures. The secondary objective was to assess the relationship between CP and inpatient mortality, total hospital charge, and length of stay in this population.

METHODS

Data of adults 18 years and older from the National Inpatient Sample (NIS) were analyzed. Variables were identified using ICD-9 codes. Propensity-matched regression models were employed in evaluating the association between CP and outcomes of interest.

RESULTS

Among 4,085,355 selected patients, 0.8% (N = 2,665) had CP, whereas 99.92% (N = 4,082,690) had no diagnosis of CP. We matched patients with CP (N = 2,665) and no CP (N = 13,325) in a 1:5 ratio. We found higher odds of opioid overdose (adjusted odds ratio [aOR] = 4.82, 95% confidence interval [CI] = 2.68-8.67, P < 0.0001) and inpatient mortality (aOR = 1.39, 95% CI = 1.11-1.74, P = 0.0043) in patients with CP vs no CP. Also, patients with CP were more likely to stay longer in the hospital (12.76 days vs 7.88 days) with higher total hospital charges ($140,220 vs $88,316).

CONCLUSIONS

CP is an independent risk factor for opioid overdose, increased length of stay, and increased total hospital charges.

Causal inference for evaluating prescription opioid abuse using trend-in-trend design

PURPOSE

One response to the opioid crisis in the United States has been the development of opioid analgesics with properties intended to reduce non-oral use. Previous evaluations of abuse in the community have relied on population averaged interrupted time series Poisson models with utilization offsets. However, competing interventions and secular trends complicate interpretation of time-series analyses. An alternative research design, trend-in-trend, accounts for heterogeneity in per capita opioid dispensing and unmeasured time-varying confounding, which provides a causal evaluation, provided that underlying assumptions are met.

METHODS

Trend-in-trend can be modeled using a logistic regression framework. In logistic regression, exposure was any product-specific outpatient dispensing by three-digit ZIP code and calendar quarter, for 22 opioids. The outcome was any product-specific abuse case ascertained from poison centers and drug treatment programs, covering 94% of the US population, between July 2009 and December 2016. Product-specific odds ratios compared places without dispensing with places with any dispensing; the causal contrast represents the odds of product-specific abuse in the community given exposure.

RESULTS

Dispensing of new and low-volume opioids varied considerably across the country, with no region showing high of all products. Of 22 opioids analyzed, the three with approved labeling as intended to deter abuse ranked near the lowest in both absolute (population-adjusted rates: 1.7, 0.9, and 8.2 per million people per quarter, respectively) and relative measures (trend-in-trend ORs: 1.96, 1.79, 1.69, respectively).

CONCLUSIONS

Postmarketing studies of prescription opioid abuse may benefit by evolving from unadjusted surveillance rates to a causal inference approach.

Exploring pathways linking greenspace to health: Theoretical and methodological guidance

BACKGROUND

In a rapidly urbanizing world, many people have little contact with natural environments, which may affect health and well-being. Existing reviews generally conclude that residential greenspace is beneficial to health. However, the processes generating these benefits and how they can be best promoted remain unclear.

OBJECTIVES

During an Expert Workshop held in September 2016, the evidence linking greenspace and health was reviewed from a transdisciplinary standpoint, with a particular focus on potential underlying biopsychosocial pathways and how these can be explored and organized to support policy-relevant population health research.

DISCUSSIONS

Potential pathways linking greenspace to health are here presented in three domains, which emphasize three general functions of greenspace: reducing harm (e.g. reducing exposure to air pollution, noise and heat), restoring capacities (e.g. attention restoration and physiological stress recovery) and building capacities (e.g. encouraging physical activity and facilitating social cohesion). Interrelations between among the three domains are also noted. Among several recommendations, future studies should: use greenspace and behavioural measures that are relevant to hypothesized pathways; include assessment of presence, access and use of greenspace; use longitudinal, interventional and (quasi)experimental study designs to assess causation; and include low and middle income countries given their absence in the existing literature. Cultural, climatic, geographic and other contextual factors also need further consideration.

CONCLUSIONS

While the existing evidence affirms beneficial impacts of greenspace on health, much remains to be learned about the specific pathways and functional form of such relationships, and how these may vary by context, population groups and health outcomes. This Report provides guidance for further epidemiological research with the goal of creating new evidence upon which to develop policy recommendations.

Illustration of the Impact of Unmeasured Confounding Within an Economic Evaluation Based on Nonrandomized Data

Background: Propensity score (PS) methods are frequently used within economic evaluations based on nonrandomized data to adjust for measured confounders, but many researchers omit the fact that they cannot adjust for unmeasured confounders. Objective: To illustrate how confounding due to unmeasured confounders can bias an economic evaluation despite PS matching. Methods: We used data from a previously published nonrandomized study to select a prematched population consisting of 121 patients (46.5%) who received endovascular aneurysm repair (EVAR) and 139 patients (53.5%) who received open surgical repair (OSR), in which sufficient data regarding eight measured confounders were available. One-to-one PS matching was used within this population to select two PS-matched subpopulations. The Matched PS-Smoking Excluded Subpopulation was selected by matching patients using a PS model that omitted patients’ smoking status (one of the measured confounders), whereas the Matched PS-Smoking Included Subpopulation was selected by matching patients using a PS model that included all eight measured confounders. Incremental cost-effectiveness ratios (ICERs) were assessed within both subpopulations. Results: Both subpopulations were composed of two different sets of 164 patients. Balance within the Matched PS-Smoking Excluded Subpopulation was achieved on all confounders except for patients’ smoking status, whereas balance within the Matched PS-Smoking Included Subpopulation was achieved on all confounders. Results indicated that the ICER of EVAR over OSR differed between both subpopulations; the ICER was estimated at $157,909 per life-year gained (LYG) within the Matched PS-Smoking Excluded Subpopulation, while it was estimated at $235,074 per LYG within the Matched PS-Smoking Included Subpopulation. Discussion: Although effective in controlling for measured confounding, PS matching may not adjust for unmeasured confounders that may bias the results of an economic evaluation based on nonrandomized data.

Bias Analysis for Uncontrolled Confounding in the Health Sciences

Uncontrolled confounding due to unmeasured confounders biases causal inference in health science studies using observational and imperfect experimental designs. The adoption of methods for analysis of bias due to uncontrolled confounding has been slow, despite the increasing availability of such methods. Bias analysis for such uncontrolled confounding is most useful in big data studies and systematic reviews to gauge the extent to which extraneous preexposure variables that affect the exposure and the outcome can explain some or all of the reported exposure-outcome associations. We review methods that can be applied during or after data analysis to adjust for uncontrolled confounding for different outcomes, confounders, and study settings. We discuss relevant bias formulas and how to obtain the required information for applying them. Finally, we develop a new intuitive generalized bias analysis framework for simulating and adjusting for the amount of uncontrolled confounding due to not measuring and adjusting for one or more confounders.

Methods to control for unmeasured confounding in pharmacoepidemiology: an overview

Background Unmeasured confounding is one of the principal problems in pharmacoepidemiologic studies. Several methods have been proposed to detect or control for unmeasured confounding either at the study design phase or the data analysis phase. Aim of the Review To provide an overview of commonly used methods to detect or control for unmeasured confounding and to provide recommendations for proper application in pharmacoepidemiology. Methods/Results Methods to control for unmeasured confounding in the design phase of a study are case only designs (e.g., case-crossover, case-time control, self-controlled case series) and the prior event rate ratio adjustment method. Methods that can be applied in the data analysis phase include, negative control method, perturbation variable method, instrumental variable methods, sensitivity analysis, and ecological analysis. A separate group of methods are those in which additional information on confounders is collected from a substudy. The latter group includes external adjustment, propensity score calibration, two-stage sampling, and multiple imputation. Conclusion As the performance and application of the methods to handle unmeasured confounding may differ across studies and across databases, we stress the importance of using both statistical evidence and substantial clinical knowledge for interpretation of the study results.

Adjustment for missing confounders in studies based on observational databases: 2-stage calibration combining propensity scores from primary and validation data

Bias caused by missing or incomplete information on confounding factors constitutes an important challenge in observational studies. The incorporation of external data on more detailed confounding information into the main study data may help remove confounding bias. This work was motivated by a study of the association between chronic obstructive pulmonary disease and herpes zoster. Analyses were based on administrative databases in which information on important confounders-cigarette smoking and alcohol consumption-was lacking. We consider adjusting for the confounding bias arising from missing confounders by incorporating a validation sample with data on smoking and alcohol consumption obtained from a small-scale National Health Interview Survey study. We propose a 2-stage calibration (TSC) method, which summarizes the confounding information through propensity scores and combines the analysis results from the main and the validation study samples, where the propensity score adjustment from the main sample is crude and that from the validation sample is more precise. Unlike the existing methods, the validity of the TSC approach does not rely on any specific measurement error model. When applying the TSC method to the motivating study above, the odds ratio of herpes zoster associated with chronic obstructive pulmonary disease is 1.91 (95% confidence interval: 1.62, 2.26) after adjustment for cumulative smoking and alcohol consumption.

Pre-study feasibility and identifying sensitivity analyses for protocol pre-specification in comparative effectiveness research

The use of healthcare databases for comparative effectiveness research (CER) is increasing exponentially despite its challenges. Researchers must understand their data source and whether outcomes, exposures and confounding factors are captured sufficiently to address the research question. They must also assess whether bias and confounding can be adequately minimized. Many study design characteristics may impact on the results; however, minimal if any sensitivity analyses are typically conducted, and those performed are post hoc. We propose pre-study steps for CER feasibility assessment and to identify sensitivity analyses that might be most important to pre-specify to help ensure that CER produces valid interpretable results.