Skip to main content
eScholarship
Open Access Publications from the University of California

Scholarly Works (31 results)

  • Article
  • Peer Reviewed
UCLA Previously Published Works (2014)

Background/Aims

Comparative effectiveness research (CER) investigates the effects of treatments and practices, thus requires causal inference. Routine data such as billing, pharmacy or EHR, while often incomplete on important confounding variables, are the usual sources of information for nonexperimental CER. The lack of randomization introduces important considerations regarding uncontrolled confounding, especially in large datasets, which potentially magnify systematic error. Yet, quantitative bias analysis in CER is not common practice. In this paper we formalize and demonstrate easy-to-implement record-level simulation techniques for analysis of uncontrolled confounding in cancer treatment CER.

Methods

We use recent advancements from the causal theory and risk analysis literature, specifically directed acyclic graphs (DAGs), and Monte-Carlo simulation techniques to introduce a novel form of record-level missing variable imputation that can be implemented during the core data analysis stage, making bias analysis more accessible using standard statistical packages. Further, our methods take into account varying levels of uncontrolled confounding by research center, or other clustering variable that may predict the level of unknown information in the dataset, and are specifically designed for implementation in large datasets, or data from multiple sources. We demonstrate these methods with two example sensitivity analyses of uncontrolled confounding in cancer treatment CER.

Results

Our methodology highlights the underlying causal model assumed for the main analysis in CER. Our technique uses the observed data lacking important confounding variables and informed estimates of the unmeasured variables to impute missing variables. The new variables now have a joint distribution with the observed data that would have been the case had they been observed fully under the assumed interrelationships. This technique is intuitively in line with the missing data framework and inference using partially observed distributions.

Conclusions

Sensitivity analysis for uncontrolled confounding is feasible and indispensable for CER. Unlike existing formula-intensive external adjustment techniques, the new technique can be implemented during core data analysis, is not outcome model specific, is at most semi-parametric and requires no esoteric software. Quantitative uncertainty analysis should be routine practice for CER in large observational data sources. Flexible methods accessible to all researchers should be a priority in this growing area of research.
    Cover page: B4-4: A Novel Technique for Analysis of Uncontrolled Confounding in Non-experimental Comparative Effectiveness Research
    • Article
    • Peer Reviewed
    UCLA Previously Published Works (2014)

    Background/Aims

    Comparative effectiveness research (CER) investigates the effects of treatments and practices, thus requires causal inference. Routine data such as billing, pharmacy or EHR, while often incomplete on important confounding variables, are the usual sources of information for nonexperimental CER. The lack of randomization introduces important considerations regarding uncontrolled confounding, especially in large datasets, which potentially magnify systematic error. Yet, quantitative bias analysis in CER is not common practice. In this paper we formalize and demonstrate easy-to-implement record-level simulation techniques for analysis of uncontrolled confounding in cancer treatment CER.

    Methods

    We use recent advancements from the causal theory and risk analysis literature, specifically directed acyclic graphs (DAGs), and Monte-Carlo simulation techniques to introduce a novel form of record-level missing variable imputation that can be implemented during the core data analysis stage, making bias analysis more accessible using standard statistical packages. Further, our methods take into account varying levels of uncontrolled confounding by research center, or other clustering variable that may predict the level of unknown information in the dataset, and are specifically designed for implementation in large datasets, or data from multiple sources. We demonstrate these methods with two example sensitivity analyses of uncontrolled confounding in cancer treatment CER.

    Results

    Our methodology highlights the underlying causal model assumed for the main analysis in CER. Our technique uses the observed data lacking important confounding variables and informed estimates of the unmeasured variables to impute missing variables. The new variables now have a joint distribution with the observed data that would have been the case had they been observed fully under the assumed interrelationships. This technique is intuitively in line with the missing data framework and inference using partially observed distributions.

    Conclusions

    Sensitivity analysis for uncontrolled confounding is feasible and indispensable for CER. Unlike existing formula-intensive external adjustment techniques, the new technique can be implemented during core data analysis, is not outcome model specific, is at most semi-parametric and requires no esoteric software. Quantitative uncertainty analysis should be routine practice for CER in large observational data sources. Flexible methods accessible to all researchers should be a priority in this growing area of research.
      Cover page: B4-4: A Novel Technique for Analysis of Uncontrolled Confounding in Non-experimental Comparative Effectiveness Research
      • Article
      • Peer Reviewed
      UCLA Previously Published Works (2014)

      Purpose

      Selection bias is a form of systematic error that can be severe in compromised study designs such as case-control studies with inappropriate selection mechanisms or follow-up studies that suffer from extensive attrition. External adjustment for selection bias is commonly undertaken when such bias is suspected, but the methods used can be overly simplistic, if not unrealistic, and fail to allow for simultaneous adjustment of associations of the exposure and covariates with the outcome, when of interest. Internal adjustment for selection bias via inverse probability weighting allows bias parameters to vary with the levels of covariates but has only been formalized for longitudinal studies with covariate data on patients up until loss to follow-up.

      Methods

      We demonstrate the use of inverse probability weighting and externally obtained bias parameters to perform internal adjustment of selection bias in studies lacking covariate data on unobserved participants.

      Results

      The "true" or selection-adjusted odds ratio for the association between exposure and outcome was successfully obtained by analyzing only data on those in the selected stratum (i.e., responders) weighted by the inverse probability of their being selected as function of their observed covariate data.

      Conclusions

      This internal adjustment technique using user-supplied bias parameters and inverse probability weighting for selection bias can be applied to any type of observational study.
        Cover page: Selection bias modeling using observed data augmented with imputed record-level probabilities.
        • Article
        • Peer Reviewed
        UCLA Previously Published Works (2014)

        Purpose

        Selection bias is a form of systematic error that can be severe in compromised study designs such as case-control studies with inappropriate selection mechanisms or follow-up studies that suffer from extensive attrition. External adjustment for selection bias is commonly undertaken when such bias is suspected, but the methods used can be overly simplistic, if not unrealistic, and fail to allow for simultaneous adjustment of associations of the exposure and covariates with the outcome, when of interest. Internal adjustment for selection bias via inverse probability weighting allows bias parameters to vary with the levels of covariates but has only been formalized for longitudinal studies with covariate data on patients up until loss to follow-up.

        Methods

        We demonstrate the use of inverse probability weighting and externally obtained bias parameters to perform internal adjustment of selection bias in studies lacking covariate data on unobserved participants.

        Results

        The "true" or selection-adjusted odds ratio for the association between exposure and outcome was successfully obtained by analyzing only data on those in the selected stratum (i.e., responders) weighted by the inverse probability of their being selected as function of their observed covariate data.

        Conclusions

        This internal adjustment technique using user-supplied bias parameters and inverse probability weighting for selection bias can be applied to any type of observational study.
          Cover page: Selection bias modeling using observed data augmented with imputed record-level probabilities
          • Thesis
          • Peer Reviewed
          UC San Diego Electronic Theses and Dissertations (2021)

          Background: Lung cancer is the leading cause of cancer-related death in the United States, but early diagnosis and evidence-based guideline-concordant treatment (GCT) can improve prognosis. However, disparities exist in who receives GCT for lung cancer which may be attributable to a patient’s geography, social environment, or provider relationship. We studied the relative contribution of travel time, neighborhood diversity, and healthcare provider engagement on disparities in GCT among non-small cell lung cancer (NSCLC) patients in California.Methods: For Aims 1 and 2, we analyzed geocoded California Cancer Registry linked American Community Survey data for ~23,000 stage I-II NSCLC patients (2006-2015). In Aim 3, we additionally linked these data to electronic health records (EHRs) for ~1,000 patients from a large healthcare delivery system. GCT was defined based on National Comprehensive Cancer Network guidelines. Driving and public transit travel times were estimated from a patient’s residence to their treatment facility, neighborhood diversity was based on the racial/ethnic composition of the patient’s neighborhood, and EHR variables reflecting healthcare engagement included provider sex and enrollment in an online patient portal. We used adjusted regression models to quantify the relative risks for undertreatment and delay (treatment initialization >45 days from diagnosis) associated with our target variables, stratified by detailed patient race/ethnicity. Results: In Aim 1, we observed that longer travel times reduced risk of undertreatment and delay. This counterintuitive result, which we call a ‘Travel Time Paradox’, did not benefit all patients, with longer travel times leading to reduced quality care for some racial/ethnic groups. In Aim 2, we observed that patients living in neighborhoods that are mixed or discordant from their race/ethnicity increased risk of undertreatment and delay, but these findings also varied across race/ethnicities with some non-White patient groups living in racial/ethnic concordant neighborhoods at increased risk for undertreatment and delay. In Aim 3 we observed that patients enrolled in the online patient portal were at substantially decreased risk for undertreatment and delay, but most patients were not enrolled. Conclusion: These results support the role of contextual drivers of inequitable treatment for cancer and highlight the importance of evaluating risk heterogeneity among multiethnic populations.

            Cover page: Identifying geospatial-, neighborhood- and healthcare system-related drivers of racial and ethnic disparities in lung cancer treatment in California
            • Thesis
            • Peer Reviewed
            UCLA Electronic Theses and Dissertations (2013)

            Quantitative treatment of uncontrolled bias in observational research is a neglected matter. In the dawn of the era of "big data", this is of particular concern because systematic error, as a portion of total error, can be greatly magnified when sample sizes increase. Unfortunately, considerable statistical road blocks exist between performing a basic multivariable analysis of an exposure-disease relationship and the thorough consideration of the direction and magnitude of uncontrolled bias. Most published literature points to the use of external formula adjustment for a thorough treatment of bias, but the formulas are often too simple (and thus unrealistic) or too complex (and thus unwieldy). A practical solution might be to perform the bias adjustment in the data, before analysis is performed. This solution would be especially useful in pooled data consortium projects, which are becoming increasingly popular as a way to investigate rare exposures and disease subtypes in cancer epidemiology, and often employ one shared data source used by multiple investigators simultaneously. Record-level data augmentation for bias analysis is central to a pooling project because it allows for multiple bias parameters to be placed directly in this data source. In this work we utilize causal theory, Monte-Carlo methods, and the missing data framework to contribute the literature of quantitative bias modeling, via flexible algorithms that may be used to translate bias adjustment for unmeasured confounding and non-response directly into the data source, before the analysis stage. We provide proof of concept for these methods via a series of simulation studies, and demonstrate their utility in a large multi-center pooled study of the epidemiology of endometrial cancer, employing both fixed hypothetical and probabilistic empirical priors for our bias parameters. Moving bias adjustment to the pre-analytic stage opens the door for an augmented data set to be analyzed in any conventional way, with no need for a working knowledge of the complex methodology behind existing external formula adjustments. A thorough, accessible, quantitative bias analysis can then serve as a tool to guide qualitative discussions about the impact of systematic error in multi-study data projects.

              Cover page: Multiple Bias Modeling in a Multi-Center Epidemiologic Study of Endometrial Cancer
              • Article
              • Peer Reviewed
              UC San Francisco Previously Published Works (2018)

              Purpose

              To examine the characteristics of voluntary online commitment contracts that may be associated with greater weight loss.

              Design

              Retrospective analysis of weight loss commitment contracts derived from a company that provides web-based support for personal commitment contracts. Using regression, we analyzed whether percentage weight loss differed between participants who incentivized their contract using monetary deposits and those who did not.

              Setting

              Online.

              Participants

              Users (N = 3857) who voluntarily signed up online in 2013 for a weight loss contract.

              Intervention

              Participants specified their own weight loss goal, time period, and self-reported weekly weight. Deposits were available in the following 3 categories: charity, anticharity (a nonprofit one does not like), or donations made to a friend.

              Measures

              Percentage weight loss per week.

              Analysis

              Multivariable linear regressions.

              Results

              Controlling for several participant and contract characteristics, contracts with anticharity, charity, and friend deposits had greater reported weight loss than nonincentivized contracts. Weight change per week relative to those without deposits was -0.33%, -0.28%, and -0.25% for anti-charity, charity, and friend, respectively ( P < 0.001). Contracts without a weight verification method claimed more weight loss than those with verification.

              Conclusion

              Voluntary use of commitment contracts may be an effective tool to assist weight loss. Those who choose to use monetary incentives report more weight loss. It is not clear whether this is due to the incentives or higher motivation.
                Cover page: Association Between Monetary Deposits and Weight Loss in Online Commitment Contracts
                • Article
                • Peer Reviewed
                UC San Diego Previously Published Works (2023)
                Immortal time bias is a well-recognized bias in clinical epidemiology but is rarely discussed in environmental epidemiology. Under the target trial framework, this bias is formally conceptualized as a misalignment between the start of study follow-up (time 0) and treatment assignment. This misalignment can occur when attained duration of follow-up is encoded into treatment assignment using minimums, maximums, or averages. The bias can be exacerbated in the presence of time trends commonly found in environmental exposures. Using lung cancer cases from the California Cancer Registry (2000-2010) linked with estimated concentrations of particulate matter less than or equal to 2.5 μm in aerodynamic diameter (PM2.5), we replicated previous studies that averaged PM2.5 exposure over follow-up in a time-to-event model. We compared this approach with one that ensures alignment between time 0 and treatment assignment, a discrete-time approach. In the former approach, the estimated overall hazard ratio for a 5-μg/m3 increase in PM2.5 was 1.38 (95% confidence interval: 1.36, 1.40). Under the discrete-time approach, the estimated pooled odds ratio was 0.99 (95% confidence interval: 0.98, 1.00). We conclude that the strong estimated effect in the former approach was likely driven by immortal time bias, due to misalignment at time 0. Our findings highlight the importance of appropriately conceptualizing a time-varying environmental exposure under the target trial framework to avoid introducing preventable systematic errors.
                  Cover page: Immortal Time Bias With Time-Varying Exposures in Environmental Epidemiology: A Case Study in Lung Cancer Survival.
                  • Article
                  • Peer Reviewed
                  UC San Diego Previously Published Works (2018)

                  Background

                  A "Table Fallacy," as coined by Westreich and Greenland, reports multiple adjusted effect estimates from a single model. This practice, which remains common in published literature, can be problematic when different types of effect estimates are presented together in a single table. The purpose of this paper is to quantitatively illustrate this potential for misinterpretation with an example estimating the effects of preeclampsia on preterm birth.

                  Methods

                  We analysed a retrospective population-based cohort of 2 963 888 singleton births in California between 2007 and 2012. We performed a modified Poisson regression to calculate the total effect of preeclampsia on the risk of PTB, adjusting for previous preterm birth. pregnancy alcohol abuse, maternal education, and maternal socio-demographic factors (Model 1). In subsequent models, we report the total effects of previous preterm birth, alcohol abuse, and education on the risk of PTB, comparing and contrasting the controlled direct effects, total effects, and confounded effect estimates, resulting from Model 1.

                  Results

                  The effect estimate for previous preterm birth (a controlled direct effect in Model 1) increased 10% when estimated as a total effect. The risk ratio for alcohol abuse, biased due to an uncontrolled confounder in Model 1, was reduced by 23% when adjusted for drug abuse. The risk ratio for maternal education, solely a predictor of the outcome, was essentially unchanged.

                  Conclusions

                  Reporting multiple effect estimates from a single model may lead to misinterpretation and lack of reproducibility. This example highlights the need for careful consideration of the types of effects estimated in statistical models.
                    Cover page: Revisiting the Table 2 fallacy: A motivating example examining preeclampsia and preterm birth
                    • Article
                    • Peer Reviewed
                    UCLA Previously Published Works (2014)

                    Background

                    Leveraging professionalism has been put forward as a strategy to drive improvement of patient care. We investigate professionalism as a factor influencing the uptake of quality improvement activities by physicians and nurses working in European hospitals.

                    Objective

                    To (i) investigate the reliability and validity of data yielded by using the self-developed professionalism measurement tool for physicians and nurses, (ii) describe their levels of professionalism displayed, and (iii) quantify the extent to which professional attitudes would predict professional behaviors.

                    Methods and materials

                    We designed and deployed survey instruments amongst 5920 physicians and nurses working in European hospitals. This was conducted under the cross-sectional multilevel study "Deepening Our Understanding of Quality Improvement in Europe" (DUQuE). We used psychometric and generalized linear mixed modelling techniques to address the aforementioned objectives.

                    Results

                    In all, 2067 (response rate 69.8%) physicians and 2805 nurses (94.8%) representing 74 hospitals in 7 European countries participated. The professionalism instrument revealed five subscales of professional attitude and one scale for professional behaviour with moderate to high internal consistency and reliability. Physicians and nurses display equally high professional attitude sum scores (11.8 and 11.9 respectively out of 16) but seem to have different perceptions towards separate professionalism aspects. Lastly, professionals displaying higher levels of professional attitudes were more involved in quality improvement actions (physicians: b = 0.019, P<0.0001; nurses: b = 0.016, P<0.0001) and more inclined to report colleagues' underperformance (physicians--odds ratio (OR) 1.12, 95% CI 1.01-1.24; nurses - OR 1.11, 95% CI 1.01-1.23) or medical errors (physicians--OR 1.14, 95% CI 1.01-1.23; nurses - OR 1.43, 95% CI 1.22-1.67). Involvement in QI actions was found to increase the odds of reporting incompetence or medical errors.

                    Conclusion

                    A tool that reliably and validly measures European physicians' and nurses' commitment to professionalism is now available. Collectively leveraging professionalism as a quality improvement strategy may be beneficial to patient care quality.
                      Cover page: Measuring Professionalism in Medicine and Nursing: Results of a European SurveyCreative Commons 'BY' version 4.0 license
                      Top