Variable Selection When Estimating Effects in External Target Populations.

External validity is an important part of epidemiologic research. To validly estimate effects in specific external target populations using a chosen effect measure (i.e., "transport"), some methods require that one account for all effect measure modifiers [EMMs]. However, little is known about how including other variables that are not EMMs (i.e., non-EMMs) in adjustment sets impacts estimates. Using simulations, we evaluated how inclusion of non-EMMs affected estimation of the transported risk difference (RD) by assessing impacts of covariates that A) differ (or not) between the trial and the target, B) are associated with the outcome (or not), and C) modify the RD (or not). We assessed variation and bias when covariates with each possible combination of these factors were used to transport RDs using outcome modeling or inverse odds weighting. Including variables that differed in distribution between the populations but were non-EMMs reduced precision, regardless of whether they were associated with the outcome. However, non-EMMs associated with selection did not amplify bias resulting from omitting necessary EMMs. Including all variables associated with the outcome may result in unnecessarily imprecise estimates when estimating treatment effects in external target populations.

Comparative effectiveness of extended release naltrexone and sublingual buprenorphine for treatment of opioid use disorder among Medicaid patients.

Aims: To compare the real-world effectiveness of extended release naltrexone (XR-NTX) and sublingual buprenorphine (SL-BUP) for the treatment of opioid use disorder (OUD). Design: An observational active comparator, new user cohort study. Setting: Medicaid claims records for patients in New Jersey and California, 2016-2019. Participants/Cases: Adult Medicaid patients aged 18-64 years who initiated XR-NTX or SL-BUP for maintenance treatment of OUD and did not use medications for OUD in the 90-days before initiation. Comparators: New initiation with XR-NTX versus SL-BUP for the treatment of OUD. Measurements: We examined two outcomes up to 180 days after medication initiation, 1) composite of medication discontinuation and death, and 2) composite of overdose and death. Findings: Our cohort included 1,755 XR-NTX and 9,886 SL-BUP patients. In adjusted analyses, treatment with XR-NTX was more likely to result in discontinuation or death by the end of follow-up than treatment with SL-BUP: cumulative risk 76% (95% confidence interval [CI] 75%, 78%) versus 62% (95% CI 61%, 63%), respectively (risk difference 14 percentage points, 95% CI 13, 16). There was minimal difference in the cumulative risk of overdose or death by the end of follow-up: XR-NTX 3.8% (95% CI 2.9%, 4.7%) versus SL-BUP 3.3% (95% 2.9%, 3.7%); risk difference 0.5 percentage points, 95%CI -0.5, 1.5. Results were consistent across sensitivity analyses. Conclusions: Longer medication retention is important because risks of negative outcomes are elevated after discontinuation. Our results support selection of SL-BUP over XR-NTX. However, most patients discontinued medication by 6 months indicating that more effective tools are needed to improve medication retention, particularly after initiation with XR-NTX, and to identify which patients do best on which medication.

M-estimation for common epidemiological measures: introduction and applied examples.

M-estimation is a statistical procedure that is particularly advantageous for some comon epidemiological analyses, including approaches to estimate an adjusted marginal risk contrast (i.e. inverse probability weighting and g-computation) and data fusion. In such settings, maximum likelihood variance estimates are not consistent. Thus, epidemiologists often resort to bootstrap to estimate the variance. In contrast, M-estimation allows for consistent variance estimates in these settings without requiring the computational complexity of the bootstrap. In this paper, we introduce M-estimation and provide four illustrative examples of implementation along with software code in multiple languages. M-estimation is a flexible and computationally efficient estimation procedure that is a powerful addition to the epidemiologist's toolbox.

Leveraging External Validation Data: The Challenges of Transporting Measurement Error Parameters.

Approaches to address measurement error frequently rely on validation data to estimate measurement error parameters (e.g., sensitivity and specificity). Acquisition of validation data can be costly, thus secondary use of existing data for validation is attractive. To use these external validation data, however, we may need to address systematic differences between these data and the main study sample. Here, we derive estimators of the risk and the risk difference that leverage external validation data to account for outcome misclassification. If misclassification is differential with respect to covariates that themselves are differentially distributed in the validation and study samples, the misclassification parameters are not immediately transportable. We introduce two ways to account for such covariates: (1) standardize by these covariates or (2) iteratively model the outcome. If conditioning on a covariate for transporting the misclassification parameters induces bias of the causal effect (e.g., M-bias), the former but not the latter approach is biased. We provide proof of identification, describe estimation using parametric models, and assess performance in simulations. We also illustrate implementation to estimate the risk of preterm birth and the effect of maternal HIV infection on preterm birth. Measurement error should not be ignored and it can be addressed using external validation data via transportability methods.

Mass Probation: Effects of Sentencing Severity on Mental Health for Black and White Individuals.

BACKGROUND: Incarceration is associated with negative impacts on mental health. Probation, a form of community supervision, has been lauded as an alternative. However, the effect of probation versus incarceration on mental health is unclear. Our objective was to estimate the impact on mental health of reducing sentencing severity at individuals' first adult criminal-legal encounter. METHODS: We used the US National Longitudinal Survey on Youth 1997, a nationally representative dataset of youth followed into their mid-thirties. Restricting to those with an adult encounter (arrest, charge alone or no sentence, probation, incarceration), we used parametric g-computation to estimate the difference in mental health at age 30 (Mental Health Inventory-5) if (1) everyone who received incarceration for their first encounter had received probation and (2) everyone who received probation had received no sentence. RESULTS: Among 1835 individuals with adult encounters, 19% were non-Hispanic Black and 65% were non-Hispanic White. Median age at first encounter was 20. Under hypothetical interventions to reduce sentencing, we did not see better mental health overall (Intervention 1, incarceration to probation: RD = -0.01; CI = -0.02, 0.01; Intervention 2, probation to no sentence: RD = 0.00; CI = -0.01, 0.01) or when stratified by race. CONCLUSION: Among those with criminal-legal encounters, hypothetical interventions to reduce sentencing, including incremental sentencing reductions, were not associated with improved mental health. Future work should consider the effects of preventing individuals' first criminal-legal encounter.

Accounting for nonmonotone missing data using inverse probability weighting.

Inverse probability weighting can be used to correct for missing data. New estimators for the weights in the nonmonotone setting were introduced in 2018. These estimators are the unconstrained maximum likelihood estimator (UMLE) and the constrained Bayesian estimator (CBE), an alternative if UMLE fails to converge. In this work we describe and illustrate these estimators, and examine performance in simulation and in an applied example estimating the effect of anemia on spontaneous preterm birth in the Zambia Preterm Birth Prevention Study. We compare performance with multiple imputation (MI) and focus on the setting of an observational study where inverse probability of treatment weights are used to address confounding. In simulation, weighting was less statistically efficient at the smallest sample size and lowest exposure prevalence examined (n = 1500, 15% respectively) but in other scenarios statistical performance of weighting and MI was similar. Weighting had improved computational efficiency taking, on average, 0.4 and 0.05 times the time for MI in R and SAS, respectively. UMLE was easy to implement in commonly used software and convergence failure occurred just twice in >200 000 simulated cohorts making implementation of CBE unnecessary. In conclusion, weighting is an alternative to MI for nonmonotone missingness, though MI performed as well as or better in terms of bias and statistical efficiency. Weighting's superior computational efficiency may be preferred with large sample sizes or when using resampling algorithms. As validity of weighting and MI rely on correct specification of different models, both approaches could be implemented to check agreement of results.

Adjunctive Diagnostic Studies Completed Following Detection of Candidemia in Children: Secondary Analysis of Observed Practice From a Multicenter Cohort Study Conducted by the Pediatric Fungal Network.

BACKGROUND: Adjunctive diagnostic studies (aDS) are recommended to identify occult dissemination in patients with candidemia. Patterns of evaluation with aDS across pediatric settings are unknown. METHODS: Candidemia episodes were included in a secondary analysis of a multicenter comparative effectiveness study that prospectively enrolled participants age 120 days to 17 years with invasive candidiasis (predominantly candidemia) from 2014 to 2017. Ophthalmologic examination (OE), abdominal imaging (AbdImg), echocardiogram, neuroimaging, and lumbar puncture (LP) were performed per clinician discretion. Adjunctive diagnostic studies performance and positive results were determined per episode, within 30 days from candidemia onset. Associations of aDS performance with episode characteristics were evaluated via mixed-effects logistic regression. RESULTS: In 662 pediatric candidemia episodes, 490 (74%) underwent AbdImg, 450 (68%) OE, 426 (64%) echocardiogram, 160 (24%) neuroimaging, and 76 (11%) LP; performance of each aDS per episode varied across sites up to 16-fold. Longer durations of candidemia were associated with undergoing OE, AbdImg, and echocardiogram. Immunocompromised status (58% of episodes) was associated with undergoing AbdImg (adjusted odds ratio [aOR] 2.38; 95% confidence intervals [95% CI] 1.51-3.74). Intensive care at candidemia onset (30% of episodes) was associated with undergoing echocardiogram (aOR 2.42; 95% CI 1.51-3.88). Among evaluated episodes, positive OE was reported in 15 (3%), AbdImg in 30 (6%), echocardiogram in 14 (3%), neuroimaging in 9 (6%), and LP in 3 (4%). CONCLUSIONS: Our findings show heterogeneity in practice, with some clinicians performing aDS selectively, potentially influenced by clinical factors. The low frequency of positive results suggests that targeted application of aDS is warranted.

Missing Outcome Data in Epidemiologic Studies.

Missing data are pandemic and a central problem for epidemiology. Missing data reduce precision and can cause notable bias. There remain too few simple published examples detailing types of missing data and illustrating their possible impact on results. Here we take an example randomized trial that was not subject to missing data and induce missing data to illustrate 4 scenarios in which outcomes are 1) missing completely at random, 2) missing at random with positivity, 3) missing at random without positivity, and 4) missing not at random. We demonstrate that accounting for missing data is generally a better strategy than ignoring missing data, which unfortunately remains a standard approach in epidemiology.

Different effects for different questions: An illustration using short cervix and the risk of preterm birth.

OBJECTIVE: To illustrate the difference between exposure effects and population attributable effects. METHODS: We examined the effect of mid-pregnancy short cervical length (<25 mm) on preterm birth using data from a prospective cohort of pregnant women in Lusaka, Zambia. Preterm birth was live birth or stillbirth before 37 weeks of pregnancy. For estimation, we used multivariable regression and parametric g-computation. RESULTS: Among 1409 women included in the analysis, short cervix was rare (2.4%); 13.6% of births were preterm. Exposure effect estimates were large (marginal risk ratio 2.86, 95% confidence interval [CI] 1.80-4.54), indicating that the preterm birth risk was substantially higher among women with a short cervix compared with women without a short cervix. However, the population attributable effect estimates were close to the null (risk ratio 1.06, 95% CI 1.02-1.10), indicating that an intervention to counteract the impact of short cervix on preterm birth would have minimal effect on the population risk of preterm birth. CONCLUSION: Although authors often refer to "the" effect, there are actually different types of effects, as we have illustrated here. In planning research, it is important to consider which effect to estimate to ensure that the estimate aligns with the research objective.

Nondifferential Treatment Misclassification Biases Toward the Null? Not a Safe Bet for Active Comparator Studies.

Active comparator studies are increasingly common, particularly in pharmacoepidemiology. In such studies, the parameter of interest is a contrast (difference or ratio) in the outcome risks between the treatment of interest and the selected active comparator. While it may appear treatment is dichotomous, treatment is actually polytomous as there are at least 3 levels: no treatment, the treatment of interest, and the active comparator. Because misclassification may occur between any of these groups, independent nondifferential treatment misclassification may not be toward the null (as expected with a dichotomous treatment). In this work, we describe bias from independent nondifferential treatment misclassification in active comparator studies with a focus on misclassification that occurs between each active treatment and no treatment. We derive equations for bias in the estimated outcome risks, risk difference, and risk ratio, and we provide bias correction equations that produce unbiased estimates, in expectation. Using data obtained from US insurance claims data, we present a hypothetical comparative safety study of antibiotic treatment to illustrate factors that influence bias and provide an example probabilistic bias analysis using our derived bias correction equations.

Reflection on modern methods: combining weights for confounding and missing data.

Inverse probability weights are increasingly used in epidemiological analysis, and estimation and application of weights to address a single bias are well discussed in the literature. Weights to address multiple biases simultaneously (i.e. a combination of weights) have almost exclusively been discussed related to marginal structural models in longitudinal settings where treatment weights (estimated first) are combined with censoring weights (estimated second). In this work, we examine two examples of combined weights for confounding and missingness in a time-fixed setting in which outcome or confounder data are missing, and the estimand is the marginal expectation of the outcome under a time-fixed treatment. We discuss the identification conditions, construction of combined weights and how assumptions of the missing data mechanisms affect this construction. We use a simulation to illustrate the estimation and application of the weights in the two examples. Notably, when only outcome data are missing, construction of combined weights is straightforward; however, when confounder data are missing, we show that in general we must follow a specific estimation procedure which entails first estimating missingness weights and then estimating treatment probabilities from data with missingness weights applied. However, if treatment and missingness are conditionally independent, then treatment probabilities can be estimated among the complete cases.

Comparative Effectiveness of Echinocandins vs Triazoles or Amphotericin B Formulations as Initial Directed Therapy for Invasive Candidiasis in Children and Adolescents.

BACKGROUND: Invasive candidiasis is the most common invasive fungal disease in children and adolescents, but there are limited pediatric-specific antifungal effectiveness data. We compared the effectiveness of echinocandins to triazoles or amphotericin B formulations (triazole/amphotericin B) as initial directed therapy for invasive candidiasis. METHODS: This multinational observational cohort study enrolled patients aged >120 days and <18 years with proven invasive candidiasis from January 1, 2014, to November 28, 2017, at 43 International Pediatric Fungal Network sites. Primary exposure was initial directed therapy administered at the time qualifying culture became positive for yeast. Exposure groups were categorized by receipt of an echinocandin vs receipt of triazole/amphotericin B. Primary outcome was global response at 14 days following invasive candidiasis onset, adjudicated by a centralized data review committee. Stratified Mantel-Haenszel analyses estimated risk difference between exposure groups. RESULTS: Seven-hundred and fifty invasive candidiasis episodes were identified. After exclusions, 541 participants (235 in the echinocandin group and 306 in the triazole/amphotericin B group) remained. Crude failure rates at 14 days for echinocandin and triazole/amphotericin B groups were 9.8% (95% confidence intervals [CI]: 6.0% to 13.6%) and 13.1% (95% CI: 9.3% to 16.8%), respectively. The adjusted 14-day risk difference between echinocandin and triazole/amphotericin B groups was -7.1% points (95% CI: -13.1% to -2.4%), favoring echinocandins. The risk difference was -0.4% (95% CI: -7.5% to 6.7%) at 30 days. CONCLUSIONS: In children with invasive candidiasis, initial directed therapy with an echinocandin was associated with reduced failure rate at 14 days but not 30 days. These results may support echinocandins as initial directed therapy for invasive candidiasis in children and adolescents. CLINICAL TRIALS REGISTRATION: NCT01869829.

Decreased Susceptibility of Marginal Odds Ratios to Finite-sample Bias.

Parameters representing adjusted treatment effects may be defined marginally or conditionally on covariates. The choice between a marginal or covariate-conditional parameter should be driven by the study question. However, an unappreciated benefit of marginal estimators is a reduction in susceptibility to finite-sample bias relative to the unpenalized maximum likelihood estimator of the covariate-conditional odds ratio (OR). Using simulation, we compare the finite-sample bias of different marginal and conditional estimators of the OR. We simulated a logistic model to have 15 events per parameter and two events per parameter. We estimated the covariate-conditional OR by maximum likelihood with and without Firth's penalization. We used three estimators of the marginal OR: g-computation, inverse probability of treatment weighting, and augmented inverse probability of treatment weighting. At 15 events per parameter, as expected, all estimators were effectively unbiased. At two events per parameter, the unpenalized covariate-conditional estimator was notably biased but penalized covariate-conditional and marginal estimators exhibited minimal bias.

Fluoroquinolone Antibiotics and Tendon Injury in Adolescents.

OBJECTIVES: To estimate the association between fluoroquinolone use and tendon injury in adolescents. METHODS: We conducted an active-comparator, new-user cohort study using population-based claims data from 2000 to 2018. We included adolescents (aged 12-18 years) with an outpatient prescription fill for an oral fluoroquinolone or comparator broad-spectrum antibiotic. The primary outcome was Achilles, quadricep, patellar, or tibial tendon rupture identified by diagnosis and procedure codes. Tendinitis was a secondary outcome. We used weighting to adjust for measured confounding and a negative control outcome to assess residual confounding. RESULTS: The cohort included 4.4 million adolescents with 7.6 million fills for fluoroquinolone (275 767 fills) or comparator (7 365 684) antibiotics. In the 90 days after the index antibiotic prescription, there were 842 tendon ruptures and 16 750 tendinitis diagnoses (crude rates 0.47 and 9.34 per 1000 person-years, respectively). The weighted 90-day tendon rupture risks were 13.6 per 100 000 fluoroquinolone-treated adolescents and 11.6 per 100 000 comparator-treated adolescents (fluoroquinolone-associated excess risk: 1.9 per 100 000 adolescents; 95% confidence interval -2.6 to 6.4); the corresponding number needed to treat to harm was 52 632. For tendinitis, the weighted 90-day risks were 200.8 per 100 000 fluoroquinolone-treated adolescents and 178.1 per 100 000 comparator-treated adolescents (excess risk: 22.7 per 100 000; 95% confidence interval 4.1 to 41.3); the number needed to treat to harm was 4405. CONCLUSIONS: The excess risk of tendon rupture associated with fluoroquinolone treatment was extremely small, and these events were rare. The excess risk of tendinitis associated with fluoroquinolone treatment was also small. Other more common potential adverse drug effects may be more important to consider for treatment decision-making, particularly in adolescents without other risk factors for tendon injury.

Outpatient Fluoroquinolone Use in Children, 2000-2018.

BACKGROUND: While fluoroquinolones are commonly used in adults, the use in children has been low. Since 2000, there were 3 US Food and Drug Administration (FDA) Boxed warnings regarding fluoroquinolones (2008, 2013, and 2016). Our objective was to describe the use of fluoroquinolones in children and assess the impact of 3 recent FDA warnings on fluoroquinolone use. METHODS: From 2000 to 2018, we assessed claims for all outpatient prescription fills to measure the use of systemic fluoroquinolones and other broad-spectrum antibiotics in children  less than 18 years old in the MarketScan Commercial Claims and Encounters database. We describe demographics, indication for antibiotic, and clinical characteristics. To assess the impact of FDA warnings on fill rates, we conducted an interrupted time-series analysis. RESULTS: The cohort included 34.6 million unique beneficiaries less than 18 years old with 441 062 fluoroquinolone fills (5.5 fills per 1000 person-years). The fluoroquinolone fill rate was highest among children > 11 years old. Urinary tract infection was the most common associated diagnosis (21.8%). Since 2008, the fluoroquinolone fill rate has declined. By the end of the study period in December 2018, in the (counterfactual) absence of the FDA warnings, fluoroquinolone fill rate would have been 7.5 (95% confidence interval [CI]: 5.2-9.7); however, the corresponding rate in observed data was 2.8 (95% CI: 1.7-3.9). CONCLUSIONS: Fluoroquinolone use was low compared with other common broad-spectrum antibiotics and declining trends over time were associated with FDA warnings, even though these warnings were not pediatric specific. Future work should assess the adverse events at issue in these warnings in children.

Initiator Types and the Causal Question of the Prevalent New-User Design: A Simulation Study.

New-user designs restricting to treatment initiators have become the preferred design for studying drug comparative safety and effectiveness using nonexperimental data. This design reduces confounding by indication and healthy-adherer bias at the cost of smaller study sizes and reduced external validity, particularly when assessing a newly approved treatment compared with standard treatment. The prevalent new-user design includes adopters of a new treatment who switched from or previously used standard treatment (i.e., the comparator), expanding study sample size and potentially broadening the study population for inference. Previous work has suggested the use of time-conditional propensity-score matching to mitigate prevalent user bias. In this study, we describe 3 "types" of initiators of a treatment: new users, direct switchers, and delayed switchers. Using these initiator types, we articulate the causal questions answered by the prevalent new-user design and compare them with those answered by the new-user design. We then show, using simulation, how conditioning on time since initiating the comparator (rather than full treatment history) can still result in a biased estimate of the treatment effect. When implemented properly, the prevalent new-user design estimates new and important causal effects distinct from the new-user design.

Meta-Analysis and Sparse-Data Bias.

Meta-analyses are undertaken to combine information from a set of studies, often in settings where some of the individual study-specific estimates are based on relatively small study samples. Finite sample bias may occur when maximum likelihood estimates of associations are obtained by fitting logistic regression models to sparse data sets. Here we show that combining information from small studies by undertaking a meta-analytical summary of logistic regression estimates can propagate such sparse-data bias. In simulations, we illustrate 2 challenges encountered in meta-analyses of logistic regression results in settings of sparse data: 1) bias in the summary meta-analytical result and 2) confidence interval coverage that can worsen rather than improve, in terms of being less than nominal, as the number of studies in the meta-analysis increases.