Selection Bias In Observational Studies Research Articles

Abstract P1-07-04: Claims-derived comorbidities associations with real-world overall survival (rwOS) among patients with metastatic breast cancer in an electronic health record-claims linked dataset

Abstract Introduction: Comorbid conditions can significantly impact various aspects of cancer care and therefore are often utilized as covariates to control for selection bias in observational studies. Oncology based electronic health record (EHR) data is often used as a source for comorbidity information but may be subject to missingness due to differences in clinician documentation. Utilizing additional data sources, such as health care claims data, may allow for greater capture of comorbidities. This study used a claims data linked to oncology EHR to identify comorbidities and related patient characteristics, as well as the association between a claims-derived comorbidity index and real-world overall survival (rwOS) among outpatient metastatic breast cancer (mBC) patients, to characterize the comorbidities data from this data source. Methods: We selected patients diagnosed with mBC between 2013 and 2021 from a linkage of mBC de-identified EHR-derived Flatiron Health Research Database (FHRD) and Komodo Health claims data. Charlson Comorbidity Index (CCI) were identified based on the presence of a pre-specified International Classification of Diseases (ICD) codes within 12 months preceding cancer diagnosis. To understand comorbidity data within this linkage, patient characteristics were assessed after the study cohort was stratified by the presence of any comorbidity assessed as part of CCI. We calculated the CCI for the study cohort as well as the prevalence of comorbid conditions that are used as part of the CCI. Kaplan-Meier estimates and Cox proportional hazards model were used to estimate median survival time and hazard ratios (HRs) of rwOS among patients with a CCI score of 1 and 2+. Results: The study cohort had 3,213 patients with a median age of 63 [IQR(52, 76)] and 98.7% were female. White patients accounted for 60.1% of the study population while Black and Hispanic/Latino patients accounted for 12.5% and 7.4% of the study cohort, respectively. Patients with comorbidities were older (mean age of 65 vs 61), had a higher (2+) ECOG performance status (PS) score (14.0% vs 9.7%), and were more likely to be Black (16.0% vs 11.0%) or Hispanic/Latino 10.8% vs 6.3%). Twenty percent of the patients had diabetes, 8.2% had peripheral vascular disease, 7.5% had mild liver disease, and 6.3% had congestive heart failure. According to index values, 21.0% of the study cohort had a CCI score of 1, and 18.0% had a score of 2 and above. The median survival times in years for patients with a CCI score of 1 and 2+ were 2.51 [95% CI 2.29,2.81] and 2.05 [95% CI 1.83,2.42], respectively. Based on univariate analysis, patients with a CCI score greater than zero had a higher risk of mortality (CCI 1: HR = 1.25 [95%CI 1.10, 1.41, p< 0.01], CCI 2+: HR = 1.54 [95%CI 1.35, 1.76, p< 0.01]). In the multivariate analysis, adjusted for age, ethnicity/race, and baseline ECOG PS score, the HR for the CCI 1 group was not statistically significant (HR=1.12 [95%CI 0.99, 1.27, p=0.08]). However, the risk of mortality was statistically significant for the CCI 2+ group (HR=1.31 [95%CI 1.14, 1.50], p < 0.01) in the adjusted analysis. Conclusion: Claims data linked to EHR can be used to identify comorbidities and describe patient characteristics. As anticipated, generally, the presence of more comorbid conditions was associated with worse rwOS for patients with mBC. This finding supports use of this linked dataset for similar future studies. Citation Format: Mustafa S. Ascha, Alemseged A. Asfaw, Prakirthi Yerram, Samantha N. Reiss, Sarah D. Brake, Niquelle B. Wadé. Claims-derived comorbidities associations with real-world overall survival (rwOS) among patients with metastatic breast cancer in an electronic health record-claims linked dataset [abstract]. In: Proceedings of the 2022 San Antonio Breast Cancer Symposium; 2022 Dec 6-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2023;83(5 Suppl):Abstract nr P1-07-04.

Author Response: Prevalence and Significance of Impaired Microvascular Tissue Reperfusion Despite Macrovascular Angiographic Reperfusion (No-Reflow).

We thank the readers for their interest in our research.1 Methodological differences may account for the discrepancy in the reported rates of no-reflow. Using prospective trial data with prespecified follow-up perfusion imaging, we assessed 146 of 147 (99%, 1 patient was excluded because of hemicraniectomy) patients with TICI 2c/3 angiographic reperfusion,1 while only 25% (n = 51/203) were assessed in the study by Ter Schiphorst et al.2 This is important because the association between no-reflow and poor neurologic recovery (24-hour NIHSS 7 [no-reflow] vs 2 [without no-reflow]; p < 0.0001) may confer inherent selection bias in observational studies because patients with more severe symptoms, such as those with no-reflow, may be less likely to complete prolonged research MRIs and therefore be excluded from analysis. Despite intermodality differences, a recent study by Rosso et al. found a similar rate of hypoperfusion on arterial spin labeling among patients with complete recanalization (n = 34/140; 24.3%),3 compared with our analysis (25.3%) using perfusion-weighted imaging and CT perfusion.1 Rosso et al. added to the growing body of literature that demonstrates a remarkably consistent prevalence of no-reflow across imaging modalities (transcranial Doppler 27.9%,4 single-photon emission computed tomography 25%5), which is in support of the external validity of our results.

Exploring the impact of selection bias in observational studies of COVID-19: a simulation study.

Non-random selection of analytic subsamples could introduce selection bias in observational studies. We explored the potential presence and impact of selection in studies of SARS-CoV-2 infection and COVID-19 prognosis. We tested the association of a broad range of characteristics with selection intoCOVID-19 analytic subsamples in the Avon Longitudinal Study of Parents and Children (ALSPAC) and UK Biobank (UKB). We then conducted empirical analyses and simulations to explore the potential presence, direction and magnitude of bias due to this selection (relative to our defined UK-based adult target populations) when estimating the association of body mass index (BMI) with SARS-CoV-2 infection and death-with-COVID-19. In both cohorts, a broad range of characteristics was related to selection, sometimes in opposite directions (e.g. more-educated people were more likely to have data on SARS-CoV-2 infection in ALSPAC, but less likely in UKB). Higher BMI was associated with higher odds of SARS-CoV-2 infection and death-with-COVID-19. We found non-negligible bias in many simulated scenarios. Analyses using COVID-19 self-reported or national registry data may be biased due to selection. The magnitude and direction of this bias depend on the outcome definition, the true effect of the risk factor and the assumed selection mechanism; these are likely to differ between studies with different target populations. Bias due to sample selection is a key concern in COVID-19 research based on national registry data, especially as countries end free mass testing. The framework we have used can be applied by other researchers assessing the extent to which their results may be biased for their research question of interest.

Contralateral advanced radiographic knee osteoarthritis predicts radiographic progression and future arthroplasty in ipsilateral knee with early-stage osteoarthritis.

To explore whether the severity of contralateral knee osteoarthritis (OA) is associated with OA progression in ipsilateral knee with early OA. Knees in early OA (Kellgren-Lawrence grade (KLG):1-2) with intact baseline demographic and clinical data were retrieved from OAI database and defined as target knees. The target knees were divided into the exposure group (contralateral knees KLG 3 to 4) and the control group (contralateral knees KLG 0 to 2). Both groups underwent propensity score matching (PSM) concerning demographic data, as well as radiographic and clinical outcomes at the baseline. The primary outcome was the upgrade of KLG in the target knee in the first 12 and 24months. The secondary outcome was the incidence of knee arthroplasty in ipsilateral knee during the first 108months. One thousand seven hundred fifty-two knees were included, with 449 in the exposure cohort and 1276 in the control cohort. Four hundred thirty-four knees in each group were matched after PSM. Target knees in the exposure cohort showed a significantly higher rate of radiographic progression in the first 12months (12.9% vs. 5.1%, P < 0.001) and 24months (19.6% vs. 8.1%, P < 0.001). As for the risk of future arthroplasty, a significant difference was also found between the two groups (7.8% vs. 4.0%, P = 0.02). Kaplan-Meier analysis showed that the 108-month accumulated knee survival rate was significantly lower in the exposure group (P = 0.01). The ipsilateral knee with early-stage OA is prone to have worse early to mid-, and long-term prognosis in the circumstance of contralateral radiographic advanced knee OA. Key Points •Identifying early knee osteoarthritis (OA) with a high risk of radiographic progression and future arthroplasty enables early personalized intervention. •This is a novel study to investigate the relationship between the risk of future arthroplasty and contralateral knee status. •Propensity score matching holds promise to minimize selection bias in observational studies. •Knees with early OA are prone to have a high risk of radiographic progression and future arthroplasty in the circumstance of contralateral advanced knee OA.

Efficacy and safety of aprotinin in paediatric cardiac surgery: A systematic review and meta-analysis.

The relicensing of aprotinin in Europe and Canada has stimulated discussions on its usefulness in paediatric cardiac surgery. To systematically evaluate the available evidence on the efficacy and safety of aprotinin in paediatric cardiac surgery. Systematic review of all randomised and observational studies comparing aprotinin with tranexamic acid, epsilon aminocaproic acid, placebo or no drug in paediatric cardiac surgery. Meta-analyses were performed on efficacy and safety outcomes. PubMed, Cochrane Central Register of Controlled Trials, Web of Science and Embase were searched from January 2000 to March 2021. Studies that enrolled children under 18 years undergoing cardiac surgery with cardiopulmonary bypass. Thirty-two studies enrolling a total of 63 894 paediatric cardiac procedures were included. Aprotinin significantly reduced total blood loss [mean difference -4.70 ml kg-1, 95% confidence interval (CI), -7.88 to -1.53; P = 0.004], postoperative transfusion requirements and the incidence of surgical re-exploration for bleeding [odds ratio (OR) 0.74, 95% CI, 0.56 to 0.97; P = 0.03]. Aprotinin had no effects on 30-day mortality (OR 1.02, 95% CI, 0.93 to 1.11; P = 0.73) and on other safety outcomes, except for the incidence of renal replacement therapy (RRT), which was significantly increased in patients given aprotinin (OR 1.29, 95% CI, 1.08 to 1.54; P = 0.006). Findings from observational and randomised controlled trials did not largely differ. A sub-group analysis in neonates showed that aprotinin significantly reduced packed red blood cell transfusions and the incidence of postoperative surgical re-exploration for bleeding and/or tamponade. When compared with lysine analogues, aprotinin was more effective at reducing bleeding and transfusion without increasing the risk of side effects. This meta-analysis suggests that aprotinin is effective and well tolerated in paediatric cardiac surgery. Given the large heterogeneity of the results and the risk of selection bias in observational studies, large randomised controlled trials are warranted.

Using dagR to identify minimal sufficient adjustment sets and to simulate data based on directed acyclic graphs

Abstract Motivation Directed acyclic graphs (DAGs) are causal diagrams that can be used to identify confounding or selection bias in observational studies, and are increasingly used in many areas of medical research. In the add-on package dagR for the statistical software R, a set of simple graphical rules was implemented to identify minimal sufficient adjustment sets or harmful adjustment. General features The dagR package allows an automated approach for realistic causal structures featuring numerous variables and dependencies, where a manual approach may be too tedious and error-prone. The algorithmic adherence to the graphical step-by-step approach often used in DAG theory introductory courses, together with functionalities for plotting and simulating data conforming to the causal structure of an arbitrary DAG, renders the dagR package particularly useful for both teaching purposes and methodological research. Availability dagR is available under the GNU general public licence (GPL-2) from within R or by download at [https://CRAN-R-project.org/package=dagR].

Selection Bias in Observational Studies of Palliative Care: Lessons Learned

ContextPalliative care (PC) programs are typically evaluated using observational data, raising concerns about selection bias. ObjectivesTo quantify selection bias because of observed and unobserved characteristics in a PC demonstration program. MethodsProgram administrative data and 100% Medicare claims data in two states and a 20% sample in eight states (2013–2017). The sample included 2983 Medicare fee-for-service beneficiaries aged 65+ participating in the PC program and three matched cohorts: regional; two states; and eight states. Confounding because of observed factors was measured by comparing patient baseline characteristics. Confounding because of unobserved factors was measured by comparing days of follow-up and six-month and one-year mortality rates. ResultsAfter matching, evidence for observed confounding included differences in observable baseline characteristics, including race, morbidity, and utilization. Evidence for unobserved confounding included significantly longer mean follow-up in the regional, two-state, and eight-state comparison cohorts, with 207 (P < 0.001), 192 (P < 0.001), and 187 (P < 0.001) days, respectively, compared with the 162 days for the PC cohort. The PC cohort had higher six-month and one-year mortality rates of 53.5% and 64.5% compared with 43.5% and 48.0% in the regional comparison, 53.4% and 57.4% in the two-state comparison, and 55.0% and 59.0% in the eight-state comparison. ConclusionThis case study demonstrates that selection of comparison groups impacts the magnitude of measured and unmeasured confounding, which may change effect estimates. The substantial impact of confounding on effect estimates in this study raises concerns about the evaluation of novel serious illness care models in the absence of randomization. We present key lessons learned for improving future evaluations of PC using observational study designs.

Potential Selection Bias in Observational Studies Comparing Cervical Disc Arthroplasty to Anterior Cervical Discectomy and Fusion.

Retrospective study using a national administrative database. To define the cohort differences in patient characteristics between patients undergoing cervical disc arthroplasty (CDA) and anterior cervical discectomy and fusion (ACDF) in a large national sample, and to describe the impact of those baseline patient characteristics on analyses of costs and complications. CDA was initially studied in high quality, randomized trials with strict inclusion criteria. Recently a number of non-randomized, observational studies have been published an attempt to expand CDA indications. These trials are predisposed to falsely attributing differences in outcomes to an intervention due to selection bias. Adults undergoing ACDF or CDA between 2004 and 2014 were identified using International Classification of Diseases, 9, Clinical Modification (ICD-9-CM) diagnosis and procedure codes. Perioperative demographics, comorbidities, complications, and costs were queried. Patient characteristics were compared via chi-square and t tests. Cost, mortality, and complications were compared between ACDF and CDA cohorts using models that adjusted for demographics and comorbidities, as well as "naïve" models that did not. A total of 290,419 procedures, 98.2% ACDF and 1.8% CDA, were included in the sample. Compared with ACDF patients, CDA patients were younger, healthier as evidenced by number of comorbidities, and had an improved socioeconomic status as measured by income and insurance. The naïve logistic regression model showed that hospital costs for CDA were $549 lower than ACDF. In the fully specified model, CDA was $574 more expensive. The naïve model for medical complications suggests a protective advantage for CDA over ACDF, odds ratio of 0.627, P < 0.01. No statistically significant difference was found in the fully specified model in terms of complications. Patients undergoing CDA were younger and healthier with higher socioeconomic statuses compared with ACDF patients. Accounting for these baseline differences significantly attenuated the apparent benefit for CDR on costs and medical complications. 3.

Effective Promotional Strategies Selection in Social Media: A Data-Driven Approach

Nowdays, many companies, organizations and individuals are using the function of sharing or retweeting information to promote their products, policies, and ideas on social media. While a growing body of research has focused on identifying the promoters from millions of users, the promoters themselves are seeking to know which strategy can improve promotional effectiveness, which is rarely studied in the literature. In this work, we investigate an open problem of effective promotional strategy selection via causal analysis which is challenging in identifying and quantifying promotional strategies as well as the selection bias when estimating the causal effect of promotional strategies from observational data. We study the promotional strategies not only on the content level (what to promote) but also on the context level (when and how to promote). To alleviate the issue of selection bias in observational studies, we propose a data-driven approach that is a Propensity Score Matching (PSM) based method, which helps to evaluate the causal effect of each promotional strategy and discover the set of effective strategies to predict the promotional effectiveness (i.e., the number of users infected by the promotion). We evaluate our proposed method on a real social dataset including 194 million users and 5 million promoted messages. Experimental results show that (1) the top-ranked strategies by our PSM based method significantly and consistently outperform the correlation based feature selection methods in predicting promotional effectiveness; (2) we conclude our observations from the real data with three interpretable and practical ideas for steering social media promotion.

A cohort study of mammography screening finds that comorbidity measures are insufficient for controlling selection bias

ObjectiveTo examine the potential of claims-based comorbidity measures for controlling selection bias in observational studies of mammography screening. Study Design and SettingBased on claims data of a large German Statutory Health Insurance fund, the single comorbidities considered by the Charlson, Elixhauser, Multipurpose Australian Comorbidity Scoring System, and M3 comorbidity measures were identified for mammography screening participants and nonparticipants. Total death rates within 4 years after screening invitation were compared. Cox proportional hazards regressions were performed unadjusted and adjusted for age, federal state of residence, and comorbidity. ResultsAmong 1,247,919 insured women aged 50–68 years (56.2% participants), 10,311 participants (death rate 375.8/100,000 person-years) and 18,113 nonparticipants (death rate 854.8/100,000 person-years) died from any cause during the follow-up. The unadjusted hazard ratio (HR) for death from any cause for participants vs. nonparticipants was 0.44 (99.9% confidence interval 0.42–0.46). Adjustments attenuated the HR to a maximum of 0.52 (0.50–0.54). ConclusionThe lower short-term all-cause mortality among participants cannot be explained by mammography screening effects and should be interpreted as selection bias. Adjusting for comorbidities only slightly attenuated this bias. Future studies should examine whether claims data include further information that is beneficial to adequately control selection bias in observational studies of mammography screening.

Reflection on modern methods: selection bias-a review of recent developments.

Selection bias remains a more difficult bias to understand than confounding or measurement error. Past definitions have not always been illuminating and a simple method (such as the change-in-estimate method for confounding) has not been available to determine its presence and magnitude in the study sample. A better understanding of the nature of the bias has led to the definition of endogenous selection bias. It is the result of conditioning on a collider variable, itself caused by two other variables; the latter variables become spuriously associated. Conditioning on a variable in the analysis that is a collider or on an indicator of sample selection has the same effect. Note that selection bias is possible even in the absence of a collider, but in the presence of endogenous selection bias, the concern is whether it is possible to identify a causal effect in the sample. Conditions have been outlined to determine that. However, even if conditions are met to identify a causal effect in the study sample, its generalization to a defined target population is not a given.We discuss the concept of endogeneity and the sources of endogenous selection bias in observational studies. We then briefly address the terms generalizability, target population (or alternative formulations) and transportability. We outline the explicit conditions to identify causal effects in studies affected by selection bias: they involve exchangeability between exposed and unexposed and exchangeability between sampled and unsampled. We briefly describe methods to generalize estimated causal effects to the target population. The latter usually require data from the target population. Finally we discuss sensitivity analyses; some are limited to providing an indication of the presence and direction of the bias and others can provide corrected estimates with user-supplied selection bias parameters.

Score de propension : intérêts, utilisation et limites. Un guide pratique pour le clinicien

Propensity scores have been proposed in the early 1980s, and are increasingly used in epidemiology since the 2000s. They are is used to minimize the selection bias in observational studies, leading to a comparability between the exposure groups close to that observed in randomized trials. However, they have important limitations. Besides, new statistical techniques to improve the propensity score performances are more and more complex, while the build and the use of propensity score require a strict methodology to avoid bias, imprecision and non-reproducibility. This overview, designed for clinicians, is aimed at describing the advantages, techniques of use and limitations of propensity scores. A reading grid is provided in order to help interpreting studies using propensity scores.

Adoption of renewable energy technologies in rural Tigray, Ethiopia: An analysis of the impact of cooperatives

With the increasing pollution of the biosphere, the development of renewable energy (RE) has become a major societal challenge. Though increasing the share of RE is high on the policy agenda, the general public has been slow to adopt it. In Ethiopia, the dissemination of RE has been limited due to lack of institutional capacity, among others. The purpose of this article is to empirically test the impact of cooperatives (institutions taking an active role in the Ethiopian economy) on RE adoption in rural Tigray, Ethiopia. To this end, we use the treatment effects model in a bid to address the potential problem of selection bias in observational studies. Findings show that co-op members invest about 1198 more birr (birr = Ethiopian currency: 1 birr ≈ 0.05 USD) in RE than non-members.

Selection bias and relationships between alcohol consumption and mortality.

Commentary to: Selection biases in observational studies affect associations between ‘moderate’ alcohol consumption and mortality

Pretest Measures of the Study Outcome and the Elimination of Selection Bias: Evidence from Three Within Study Comparisons.

This paper examines how pretest measures of a study outcome reduce selection bias in observational studies in education. The theoretical rationale for privileging pretests in bias control is that they are often highly correlated with the outcome, and in many contexts, they are also highly correlated with the selection process. To examine the pretest's role in bias reduction, we use the data from two within study comparisons and an especially strong quasi-experiment, each with an educational intervention that seeks to improve achievement. In each study, the pretest measures are consistently highly correlated with post-intervention measures of themselves, but the studies vary the correlation between the pretest and the process of selection into treatment. Across the three datasets with two outcomes each, there are three cases where this correlation is low and three where it is high. A single wave of pretest always reduces bias across the six instances examined, and it eliminates bias in three of them. Adding a second pretest wave eliminates bias in two more instances. However, the pattern of bias elimination does not follow the predicted pattern-that more bias reduction ensues as a function of how highly the pretest is correlated with selection. The findings show that bias is more complexly related to the pretest's correlation with selection than we hypothesized, and we seek to explain why.

The protective effects of moderate drinking: lies, damned lies, and… selection biases?

Commentary to: Selection biases in observational studies affect associations between ‘moderate’ alcohol consumption and mortality

Beneficial effects of moderate alcohol use-a case for Occam's razor?

Commentary to: Selection biases in observational studies affect associations between ‘moderate’ alcohol consumption and mortality

On bias in alcohol epidemiology and the search for the perfect study.

Commentary to: Selection biases in observational studies affect associations between ‘moderate’ alcohol consumption and mortality

Selection biases in observational studies affect associations between 'moderate' alcohol consumption and mortality.

Selection biases may lead to systematic overestimate of protective effects from 'moderate' alcohol consumption. Overall, most sources of selection bias favor low-volume drinkers in relation to non-drinkers. Studies that attempt to address these types of bias generally find attenuated or non-significant relationships between low-volume alcohol consumption and cardiovascular disease, which is the major source of possible protective effects on mortality from low-volume consumption. Furthermore, observed mortality effects among established low-volume consumers are of limited relevance to health-related decisions about whether to initiate consumption or to continue drinking purposefully into old age. Short of randomized trials with mortality end-points, there are a number of approaches that can minimize selection bias involving low-volume alcohol consumption.

Cigarette Smoking and Pterygium: A Propensity Score Matching Analysis.

To assess the association between smoking and pterygium among elderly adults by applying a propensity score matching (PSM) approach and to examine a potential dose-response relationship. The study was designed as a community-based cross-sectional study based on a Chinese cohort aged 60 years or older in China. Anterior segment examination was performed without pupil dilation using a slitlamp. Pterygium was defined as a raised fleshy triangular fibrovascular tissue growth of the conjunctiva encroaching onto the clear cornea. Information regarding smoking history was collected via a structured interviewer-administered questionnaire. The propensity scores for myopia were formulated using seven potential confounders. We matched the propensity scores for smokers and nonsmokers within a caliper of 0.01 of logit function of propensity scores. Slitlamp examination was successfully performed in 4567 adults. The odds ratio of pterygium for cigarette smoking before matching was 0.65 (95% confidence interval, 0.56 to 0.75; p < 0.001). There were significant covariate imbalances between comparison groups, and after PSM, covariate imbalances were significantly reduced. After PSM, the magnitude of association was slightly reduced (odds ratio, 0.70; 95% confidence interval, 0.55 to 0.89; p = 0.004). Adults with more daily cigarette consumption were less likely to be affected by pterygium in multivariate analysis (p for trend = 0.02). Smokers were less likely to be affected by pterygium. The PSM approach may be a useful method to address selection bias in observational studies when randomized trials cannot ethically be conducted.