Targeted maximum likelihood estimation for a binary treatment: A tutorial

Propensity Weighted federated learning for treatment effect estimation in distributed imbalanced environments

Estimating treatment effects from observational data in medicine using causal inference is a very relevant task due to the abundance of observational data and the ethical and cost implications of conducting randomized experiments or experimental interventions. However, how could we estimate the effect of a treatment in a hospital that has very restricted access to treatment? In this paper, we want to address the problem of distributed causal inference, where hospitals not only have different distributions of patients, but also different treatment assignment criteria. Furthermore, it is necessary to take into account that due to privacy restrictions, personal patient data cannot be shared between hospitals. To address this problem, we propose an adaptation of the federated learning algorithm FederatedAveraging to one of the most advanced models for the prediction of treatment effects based on neural networks, TEDVAE. Our algorithm adaptation takes into account the shift in the treatment distribution between hospitals and is therefore called Propensity WeightedFederatedAveraging (PW FedAvg). As the distributions of the assignment of treatments become more unbalanced between the nodes, the estimation of causal effects becomes more challenging. The experiments show that PW FedAvg manages to reduce errors in the estimation of individual causal effects when imbalances are large, compared to VanillaFedAvg and other federated learning-based causal inference algorithms based on the application of federated learning to linear parametric models, Gaussian Processes and Random Fourier Features.

Statistical methods to control for confounders in rare disease settings that use external control

In the drug development for rare disease, the number of treated subjects in the clinical trial is often very small, whereas the number of external controls can be relatively large. There is no clear guidance on choosing an appropriate statistical method to control baseline confounding in this situation. To fill this gap, we conduct extensive simulations to evaluate the performance of commonly used matching and weighting methods as well as the more recently developed targeted maximum likelihood estimation (TMLE) and cardinality matching in small sample settings, mimicking the motivating data from a pediatric rare disease. Among the methods examined, the performance of coarsened exact matching (CEM) and TMLE are relatively robust under various model specifications. CEM is only feasible when the number of controls far exceeds the number of treated, whereas TMLE has better performance with less extreme treatment allocation ratios. Our simulations suggest bootstrap is useful for variance estimation in small samples after matching.

Targeted-BEHRT: Deep Learning for Observational Causal Inference on Longitudinal Electronic Health Records

Observational causal inference is useful for decision-making in medicine when randomized clinical trials (RCTs) are infeasible or nongeneralizable. However, traditional approaches do not always deliver unconfounded causal conclusions in practice. The rise of "doubly robust" nonparametric tools coupled with the growth of deep learning for capturing rich representations of multimodal data offers a unique opportunity to develop and test such models for causal inference on comprehensive electronic health records (EHRs). In this article, we investigate causal modeling of an RCT-established causal association: the effect of classes of antihypertensive on incident cancer risk. We develop a transformer-based model, targeted bidirectional EHR transformer (T-BEHRT) coupled with doubly robust estimation to estimate average risk ratio (RR). We compare our model to benchmark statistical and deep learning models for causal inference in multiple experiments on semi-synthetic derivations of our dataset with various types and intensities of confounding. In order to further test the reliability of our approach, we test our model on situations of limited data. We find that our model provides more accurate estimates of relative risk [least sum absolute error (SAE) from ground truth] compared with benchmark estimations. Finally, our model provides an estimate of class-wise antihypertensive effect on cancer risk that is consistent with results derived from RCTs.

Two-stage targeted maximum likelihood estimation for mixed aggregate and individual participant data analysis with an application to multidrug resistant tuberculosis

In this study, we develop a new method for the meta-analysis of mixed aggregate data (AD) and individual participant data (IPD). The method is an adaptation of inverse probability weighted targeted maximum likelihood estimation (IPW-TMLE), which was initially proposed for two-stage sampled data. Our methods are motivated by a systematic review investigating treatment effectiveness for multidrug resistant tuberculosis (MDR-TB) where the available data include IPD from some studies but only AD from others. One complication in this application is that participants with MDR-TB are typically treated with multiple antimicrobial agents where many such medications were not observed in all studies considered in the meta-analysis. We focus here on the estimation of the expected potential outcome while intervening on a specific medication but not intervening on any others. Our method involves the implementation of a TMLE that transports the estimation from studies where the treatment is observed to the full target population. A second weighting component adjusts for the studies with missing (inaccessible) IPD. We demonstrate the properties of the proposed method and contrast it with alternative approaches in a simulation study. We finally apply this method to estimate treatment effectiveness in the MDR-TB case study.

Longitudinal change of gut microbiota in hypertensive disorders in pregnancy: a nested case-control and Mendelian randomization study

Mounting evidence has shown that gut microbiota (GM) is related to hypertensive disorders in pregnancy (HDP), however, most studies only focused on one time point in pregnancy. In this study, we conducted a nested case-control study utilizing a follow-up cohort, resulting in the collection of 47 HDP patients and 30 healthy controls. The GM profiles were explored using 16S rRNA sequencing at three time points during pregnancy. The diversity analysis of GM showed no significant difference between HDP patients and controls, however, we found 21 differential GM during pregnancy. Trend analysis showed that there are statistical differences in the relative abundance of Thermomonas, Xanthomonas, and Phenylobacteriumat during pregnancy in the gestational hypertension group, and of Xanthomonas, Polycyclovorans, and Phenylobacterium in the control group. The correlation study found that six genera of GM are related to blood pressure. Furthermore, the MR analysis identified the causal relationship between Methanobrevibacter and pre-eclampsia (PE). This study first explored the longitudinal change of GM in HDP patients during pregnancy, found the differential GM, and detected the causal association. Our findings may promote the prevention and treatment of HDP from the perspective of GM and provide valuable insights into the pathogenesis of HDP.

Application of targeted maximum likelihood estimation in public health and epidemiological studies: a systematic review

PURPOSE

The targeted maximum likelihood estimation (TMLE) statistical data analysis framework integrates machine learning, statistical theory, and statistical inference to provide a least biased, efficient, and robust strategy for estimation and inference of a variety of statistical and causal parameters. We describe and evaluate the epidemiological applications that have benefited from recent methodological developments.

METHODS

We conducted a systematic literature review in PubMed for articles that applied any form of TMLE in observational studies. We summarized the epidemiological discipline, geographical location, expertize of the authors, and TMLE methods over time. We used the Roadmap of Targeted Learning and Causal Inference to extract key methodological aspects of the publications. We showcase the contributions to the literature of these TMLE results.

RESULTS

Of the 89 publications included, 33% originated from the University of California at Berkeley, where the framework was first developed by Professor Mark van der Laan. By 2022, 59% of the publications originated from outside the United States and explored up to seven different epidemiological disciplines in 2021-2022. Double-robustness, bias reduction, and model misspecification were the main motivations that drew researchers toward the TMLE framework. Through time, a wide variety of methodological, tutorial, and software-specific articles were cited, owing to the constant growth of methodological developments around TMLE.

CONCLUSIONS

There is a clear dissemination trend of the TMLE framework to various epidemiological disciplines and to increasing numbers of geographical areas. The availability of R packages, publication of tutorial papers, and involvement of methodological experts in applied publications have contributed to an exponential increase in the number of studies that understood the benefits and adoption of TMLE.

Gut microbiota and hypertensive disorders in pregnancy: evidence from the Mendelian randomization study

BACKGROUND

Recent studies have shown that gut microbiota (GM) is related to hypertensive disorders in pregnancy (HDP). However, the causal relationship needs to be treated with caution due to confounding factors and reverse causation.

METHODS

We obtained genetic variants from genome-wide association studies including GM (N = 18,340) in MiBioGen Consortium as well as HDP (7,686 cases/115,893 controls) and specific subtypes in FinnGen Consortium. Then, Inverse variance weighted, maximum likelihood, weighted median, MR-Egger, and MR.RAPS methods were applied to examine the causal association. Reverse Mendelian randomization (RMR) and multivariable MR were performed to confirm the causal direction and adjust the potential confounders, respectively. Furthermore, sensitivity analyses including Cochran's Q statistics, MR-Egger intercept, MR-PRESSO global test, and the leave-one-out analysis were conducted to detect the potential heterogeneity and horizontal pleiotropy.

RESULTS

The present study found causalities between eight gut microbial genera and HDP. The HDP-associated gut microbial genera identified by MR analyses varied in different subtypes. Specifically, our study found causal associations of LachnospiraceaeUCG010, Olsenella, RuminococcaceaeUCG009, Ruminococcus2, Anaerotruncus, Bifidobacterium, and Intestinibacter with GH, of Eubacterium (ruminantium group), Eubacterium (ventriosum group), Methanobrevibacter, RuminococcaceaeUCG002, and Tyzzerella3 with PE, and of Dorea and RuminococcaceaeUCG010 with eclampsia, respectively.

CONCLUSIONS

This study first applied the MR approach to detect the causal relationships between GM and specific HDP subtypes. Our findings may promote the prevention and treatment of HDP targeted on GM and provide valuable insights to understand the mechanism of HDP in different subtypes from the perspective of GM.

Toward a causal link between attachment styles and mental health during the COVID-19 pandemic

BACKGROUND

Recent research has shown that insecure attachment, especially attachment anxiety, is associated with poor mental health outcomes, especially during the COVID-19 pandemic. Other research suggests that insecure attachment may be linked to nonadherence to social distancing behaviours during the pandemic.

AIMS

The present study aims to examine the causal links between attachment styles (secure, anxious, avoidant), mental health outcomes (depression, anxiety, loneliness) and adherence to social distancing behaviours during the first several months of the UK lockdown (between April and August 2020).

MATERIALS & METHODS

We used a nationally representative UK sample (cross-sectional n = 1325; longitudinal n = 950). The data were analysed using state-of-the-art causal discovery and targeted learning algorithms to identify causal processes.

RESULTS

The results showed that insecure attachment styles were causally linked to poorer mental health outcomes, mediated by loneliness. Only attachment avoidance was causally linked to nonadherence to social distancing guidelines.

DISCUSSION

Future interventions to improve mental health outcomes should focus on mitigating feelings of loneliness. Limitations include no access to pre-pandemic data and the use of categorical attachment measure.

CONCLUSION

Insecure attachment is a risk factor for poorer mental health outcomes.

Investigating causal associations among gut microbiota, gut microbiota-derived metabolites, and gestational diabetes mellitus: a bidirectional Mendelian randomization study

BACKGROUND

Previous studies have shown that gut microbiota (GM) and gut microbiota-derived metabolites are associated with gestational diabetes mellitus (GDM). However, the causal associations need to be treated with caution due to confounding factors and reverse causation.

METHODS

This study obtained genetic variants from genome-wide association study including GM (N = 18,340), GM-derived metabolites (N = 7,824), and GDM (5,687 cases and 117,89 controls). To examine the causal association, several methods were utilized, including inverse variance weighted, maximum likelihood, weighted median, MR-Egger, and MR.RAPS. Additionally, reverse Mendelian Randomization (MR) analysis and multivariable MR were conducted to confirm the causal direction and account for potential confounders, respectively. Furthermore, sensitivity analyses were performed to identify any potential heterogeneity and horizontal pleiotropy.

RESULTS

Greater abundance of Collinsella was detected to increase the risk of GDM. Our study also found suggestive associations among Coprobacter, Olsenella, Lachnoclostridium, Prevotella9, Ruminococcus2, Oscillibacte, and Methanobrevibacter with GDM. Besides, eight GM-derived metabolites were found to be causally associated with GDM. For the phenylalanine metabolism pathway, phenylacetic acid was found to be related to the risk of GDM.

CONCLUSIONS

The study first used the MR approach to explore the causal associations among GM, GM-derived metabolites, and GDM. Our findings may contribute to the prevention and treatment strategies for GDM by targeting GM and metabolites, and offer novel insights into the underlying mechanism of the disease.

Association of pulmonary artery catheter with in-hospital outcomes after cardiac surgery in the United States: National Inpatient Sample 1999-2019

To examine associations of pulmonary artery catheter (PAC) use with in-hospital death and hospital length of stay (days) overall and within subgroups of hospitalized cardiac surgery patients. Secondary analyses of 1999-2019 National Inpatient Sample data were performed using 969,034 records (68% male, mean age: 65 years) representing adult cardiac surgery patients in the United States. A subgroup of 323,929 records corresponded to patients with congestive heart failure, pulmonary hypertension, mitral/tricuspid valve disease and/or combined surgeries. We evaluated PAC in relation to clinical outcomes using regression and targeted maximum likelihood estimation (TMLE). Hospitalized cardiac surgery patients experienced more in-hospital deaths and longer stays if they had ≥ 1 subgroup characteristics. For risk-adjusted models, in-hospital deaths were similar among recipients and non-recipients of PAC (odds ratio [OR] 1.04, 95% confidence interval [CI] 0.96, 1.12), although PAC was associated with more in-hospital deaths among the subgroup with congestive heart failure (OR 1.14, 95% CI 1.03, 1.26). PAC recipients experienced shorter stays than non-recipients (β =  - 0.40, 95% CI - 0.64, - 0.15), with variations by subgroup. We obtained comparable results using TMLE. In this retrospective cohort study, PAC was associated with shorter stays and similar in-hospital death rates among cardiac surgery patients. Worse clinical outcomes associated with PAC were observed only among patients with congestive heart failure. Prospective cohort studies and randomized controlled trials are needed to confirm and extend these preliminary findings.

Mendelian randomization analyses support causal relationship between gut microbiota and childhood obesity

Background

Childhood obesity (CO) is an increasing public health issue. Mounting evidence has shown that gut microbiota (GM) is closely related to CO. However, the causal association needs to be treated with caution due to confounding factors and reverse causation.

Methods

Data were obtained from the Microbiome Genome Consortium for GM as well as the Early Growth Genetics Consortium for childhood obesity and childhood body mass index (CBMI). Inverse variance weighted, maximum likelihood, weighted median, and MR.RAPS methods were applied to examine the causal association. Then replication dataset was used to validate the results and reverse Mendelian randomization analysis was performed to confirm the causal direction. Additionally, sensitivity analyses including Cochran's Q statistics, MR-Egger intercept, MR-PRESSO global test, and the leave-one-out analysis were conducted to detect the potential heterogeneity and horizontal pleiotropy.

Results

Our study found suggestive causal relationships between eight bacterial genera and the risk of childhood obesity (five for CO and four for CBMI). After validating the results in the replication dataset, we finally identified three childhood obesity-related GM including the genera Akkermansia, Intestinibacter, and Butyricimonas. Amongst these, the genus Akkermansia was both negatively associated with the risk of CO (OR = 0.574; 95% CI: 0.417, 0.789) and CBMI (β = -0.172; 95% CI: -0.306, -0.039).

Conclusions

In this study, we employed the MR approach to investigate the causal relationship between GM and CO, and discovered that the genus Akkermansia has a protective effect on both childhood obesity and BMI. Our findings may provide a potential strategy for preventing and intervening in CO, while also offering novel insights into the pathogenesis of CO from the perspective of GM.

Emotion dynamic patterns between intimate relationship partners predict their separation two years later: A machine learning approach

Contemporary emotion theories predict that how partners' emotions are coupled together across an interaction can inform on how well the relationship functions. However, few studies have compared how individual (i.e., mean, variability) and dyadic aspects of emotions (i.e., coupling) during interactions predict future relationship separation. In this exploratory study, we utilized machine learning methods to evaluate whether emotions during a positive and a negative interaction from 101 couples (N = 202 participants) predict relationship stability two years later (17 breakups). Although the negative interaction was not predictive, the positive was: Intra-individual variability of emotions as well as the coupling between partners' emotions predicted relationship separation. The present findings demonstrate that utilizing machine learning methods enables us to improve our theoretical understanding of complex patterns.

Causal Evaluation of Post-Marketing Drugs for Drug-induced Liver Injury from Electronic Health Records

Drug-induced liver injury (DILI) is one of the most common and serious adverse drug reactions that can lead to acute liver failure and death. Detection of DILI and causal estimation of drug-hepatotoxicity association are of great importance for patient safety. This paper proposes a framework for causal estimation of post-marketing drugs for DILI from real-world electronic health record (EHR) data. Randomized clinical trials were replicated at scale by automatically generating different user and non-user cohorts for each potential drug, and average treatment effects (ATEs) of drugs were estimated using targeted maximum likelihood estimation. Ten years of real-world EHRs were used to validate the framework. Of all 1199 single-ingredient drugs analyzed, 7 novel and 7 known drug-hepatotoxicity associations were found to be causal.

Frameworks for estimating causal effects in observational settings: comparing confounder adjustment and instrumental variables

To estimate causal effects, analysts performing observational studies in health settings utilize several strategies to mitigate bias due to confounding by indication. There are two broad classes of approaches for these purposes: use of confounders and instrumental variables (IVs). Because such approaches are largely characterized by untestable assumptions, analysts must operate under an indefinite paradigm that these methods will work imperfectly. In this tutorial, we formalize a set of general principles and heuristics for estimating causal effects in the two approaches when the assumptions are potentially violated. This crucially requires reframing the process of observational studies as hypothesizing potential scenarios where the estimates from one approach are less inconsistent than the other. While most of our discussion of methodology centers around the linear setting, we touch upon complexities in non-linear settings and flexible procedures such as target minimum loss-based estimation and double machine learning. To demonstrate the application of our principles, we investigate the use of donepezil off-label for mild cognitive impairment. We compare and contrast results from confounder and IV methods, traditional and flexible, within our analysis and to a similar observational study and clinical trial.

On logistic regression with right censored data, with or without competing risks, and its use for estimating treatment effects

Simple logistic regression can be adapted to deal with right-censoring by inverse probability of censoring weighting (IPCW). We here compare two such IPCW approaches, one based on weighting the outcome, the other based on weighting the estimating equations. We study the large sample properties of the two approaches and show that which of the two weighting methods is the most efficient depends on the censoring distribution. We show by theoretical computations that the methods can be surprisingly different in realistic settings. We further show how to use the two weighting approaches for logistic regression to estimate causal treatment effects, for both observational studies and randomized clinical trials (RCT). Several estimators for observational studies are compared and we present an application to registry data. We also revisit interesting robustness properties of logistic regression in the context of RCTs, with a particular focus on the IPCW weighting. We find that these robustness properties still hold when the censoring weights are correctly specified, but not necessarily otherwise.

Dealing with confounding in observational studies: A scoping review of methods evaluated in simulation studies with single-point exposure

The aim of this article was to perform a scoping review of methods available for dealing with confounding when analyzing the effect of health care treatments with single-point exposure in observational data. We aim to provide an overview of methods and their performance assessed by simulation studies indexed in PubMed. We searched PubMed for simulation studies published until January 2021. Our search was restricted to studies evaluating binary treatments and binary and/or continuous outcomes. Information was extracted on the methods' assumptions, performance, and technical properties. Of 28,548 identified references, 127 studies were eligible for inclusion. Of them, 84 assessed 14 different methods (ie, groups of estimators that share assumptions and implementation) for dealing with measured confounding, and 43 assessed 10 different methods for dealing with unmeasured confounding. Results suggest that there are large differences in performance between methods and that the performance of a specific method is highly dependent on the estimator. Furthermore, the methods' assumptions regarding the specific data features also substantially influence the methods' performance. Finally, the methods result in different estimands (ie, target of inference), which can even vary within methods. In conclusion, when choosing a method to adjust for measured or unmeasured confounding it is important to choose the most appropriate estimand, while considering the population of interest, data structure, and whether the plausibility of the methods' required assumptions hold.

A cautionary note on the use of G-computation in population adjustment

In a recent issue of the Journal; Remiro-Azócar et al. introduce a new method to adjust for population difference between two trials; when the individual patient data (IPD) are only accessible for one study. The proposed method generates the covariate data for the trial without IPD; then using a G-computation approach to transport information about the treatment effect from the other study with IPD to this trial. The authors advocate the use of G-computation over matching-adjusted indirect comparison because (i) the former allows for "useful extrapolation" when there is poor case-mix overlap between populations; and (ii) nonparametric; data-adaptive methods can be used to reduce the risk of (outcome) model misspecification. In this commentary; we provide a different perspective from these arguments. Despite certain disagreements; we believe that the proposed data generation approaches can open new and interesting research directions for population adjustment methodology in the future.

The impact of disease changes and mental health illness on readapted return to work after repeated sick leaves among Brazilian public university employees

Introduction

Health affects work absenteeism and productivity of workers, making it a relevant marker of an individual's professional development.

Objectives

The aims of this article were to investigate whether changes in the main cause of the sick leaves and the presence of mental health illnesses are associated with return to work with readaptation.

Materials and methods

A historical cohort study was carried out with non-work-related illnesses suffered by statutory workers of university campuses in a medium-sized city in the state of São Paulo, Brazil. Two exposures were measured: (a) changes, throughout medical examinations, in the International Classification of Diseases (ICD-10) chapter regarding the main condition for the sick leave; and (b) having at least one episode of sick leave due to mental illness, with or without change in the ICD-10 chapter over the follow-up period. The outcome was defined as return to work with adapted conditions. The causal model was established a priori and tested using a multiple logistic regression (MLR) model considering the effects of several confounding factors, and then compared with the same estimators obtained using Targeted Machine Learning.

Results

Among workers in adapted conditions, 64% were health professionals, 34% had had changes in the ICD-10 chapter throughout the series of sick leaves, and 62% had diagnoses of mental health issues. In addition, they worked for less time at the university and were absent for longer periods. Having had a change in the illness condition reduced the chance of returning to work in another function by more than 30%, whereas having had at least one absence because of a cause related to mental and behavioral disorders more than doubled the chance of not returning to work in the same activity as before.

Conclusion

These results were independent of the analysis technique used, which allows concluding that there were no advantages in the use of targeted maximum likelihood estimation (TMLE), given its difficulties in access, use, and assumptions.

A comparison of covariate adjustment approaches under model misspecification in individually randomized trials

Adjustment for baseline covariates in randomized trials has been shown to lead to gains in power and can protect against chance imbalances in covariates. For continuous covariates, there is a risk that the the form of the relationship between the covariate and outcome is misspecified when taking an adjusted approach. Using a simulation study focusing on individually randomized trials with small sample sizes, we explore whether a range of adjustment methods are robust to misspecification, either in the covariate-outcome relationship or through an omitted covariate-treatment interaction. Specifically, we aim to identify potential settings where G-computation, inverse probability of treatment weighting (IPTW), augmented inverse probability of treatment weighting (AIPTW) and targeted maximum likelihood estimation (TMLE) offer improvement over the commonly used analysis of covariance (ANCOVA). Our simulations show that all adjustment methods are generally robust to model misspecification if adjusting for a few covariates, sample size is 100 or larger, and there are no covariate-treatment interactions. When there is a non-linear interaction of treatment with a skewed covariate and sample size is small, all adjustment methods can suffer from bias; however, methods that allow for interactions (such as G-computation with interaction and IPTW) show improved results compared to ANCOVA. When there are a high number of covariates to adjust for, ANCOVA retains good properties while other methods suffer from under- or over-coverage. An outstanding issue for G-computation, IPTW and AIPTW in small samples is that standard errors are underestimated; they should be used with caution without the availability of small-sample corrections, development of which is needed. These findings are relevant for covariate adjustment in interim analyses of larger trials.

An Alternative Perspective on the Robust Poisson Method for Estimating Risk or Prevalence Ratios

The robust Poisson method is becoming increasingly popular when estimating the association of exposures with a binary outcome. Unlike the logistic regression model, the robust Poisson method yields results that can be interpreted as risk or prevalence ratios. In addition, it does not suffer from frequent nonconvergence problems such as the most common implementations of maximum likelihood estimators of the log-binomial model. However, using a Poisson distribution to model a binary outcome may seem counterintuitive. Methodologic papers have often presented this as a good approximation to the more natural binomial distribution. In this article, we provide an alternative perspective to the robust Poisson method based on the semiparametric theory. This perspective highlights that the robust Poisson method does not require assuming a Poisson distribution for the outcome. In fact, the method only assumes a log-linear relation between the risk or prevalence of the outcome and the explanatory variables. This assumption and the consequences of its violation are discussed. We also provide suggestions to reduce the risk of violating the modeling assumption. Additionally, we discuss and contrast the robust Poisson method with other approaches for estimating exposure risk or prevalence ratios. See video abstract at, http://links.lww.com/EDE/B987 .

Causal evidence in health decision making: methodological approaches of causal inference and health decision science

Objectives

Public health decision making is a complex process based on thorough and comprehensive health technology assessments involving the comparison of different strategies, values and tradeoffs under uncertainty. This process must be based on best available evidence and plausible assumptions. Causal inference and health decision science are two methodological approaches providing information to help guide decision making in health care. Both approaches are quantitative methods that use statistical and modeling techniques and simplifying assumptions to mimic the complexity of the real world. We intend to review and lay out both disciplines with their aims, strengths and limitations based on a combination of textbook knowledge and expert experience.

Methods

To help understanding and differentiating the methodological approaches of causal inference and health decision science, we reviewed both methods with the focus on aims, research questions, methods, assumptions, limitations and challenges, and software. For each methodological approach, we established a group of four experts from our own working group to carefully review and summarize each method, followed by structured discussion rounds and written reviews, in which the experts from all disciplines including HTA and medicine were involved. The entire expert group discussed objectives, strengths and limitations of both methodological areas, and potential synergies. Finally, we derived recommendations for further research and provide a brief outlook on future trends.

Results

Causal inference methods aim for drawing causal conclusions from empirical data on the relationship of pre-specified interventions on a specific target outcome and apply a counterfactual framework and statistical techniques to derive causal effects of exposures or interventions from these data. Causal inference is based on a causal diagram, more specifically, a directed acyclic graph (DAG), which encodes the assumptions regarding the causal relations between variables. Depending on the type of confounding and selection bias, traditional statistical methods or more complex g-methods are needed to derive valid causal effects. Besides the correct specification of the DAG and the statistical model, assumptions such as consistency, positivity, and exchangeability must be checked when aiming at causal inference. Health decision science aims for guiding policy decision making regarding health interventions considering and balancing multiple competing objectives of a decision based on data from multiple sources and studies, for example prevalence studies, clinical trials and long-term observational routine effectiveness studies, and studies on preferences and costs. It involves decision analysis, a systematic, explicit and quantitative framework to guide decisions under uncertainty. Decision analyses are based on decision-analytic models to mimic the course of disease as well as aspects and consequences of the intervention in order to quantitatively optimize the decision. Depending on the type of decision problem, decision trees, state-transition models, discrete event simulation models, dynamic transmission models, or other model types are applied. Models must be validated against observed data, and comprehensive sensitivity analyses must be performed to assess uncertainty. Besides the appropriate choice of the model type and the valid specification of the model structure, it must be checked if input parameters of effects can be interpreted as causal parameters in the model. Otherwise results will be biased.

Conclusions

Both causal inference and health decision science aim for providing best causal evidence for informed health decision making. The strengths and limitations of both methods differ and a good understanding of both methods is essential for correct application but also for correct interpretation of findings from the described methods. Importantly, decision-analytic modeling should be combined with causal inference when developing guidance and recommendations regarding decisions on health care interventions.

Causal inference in case of near-violation of positivity: comparison of methods

In causal studies, the near-violation of the positivity may occur by chance, because of sample-to-sample fluctuation despite the theoretical veracity of the positivity assumption in the population. It may mostly happen when the exposure prevalence is low or when the sample size is small. We aimed to compare the robustness of g-computation (GC), inverse probability weighting (IPW), truncated IPW, targeted maximum likelihood estimation (TMLE), and truncated TMLE in this situation, using simulations and one real application. We also tested different extrapolation situations for the sub-group with a positivity violation. The results illustrated that the near-violation of the positivity impacted all methods. We demonstrated the robustness of GC and TMLE-based methods. Truncation helped in limiting the bias in near-violation situations, but at the cost of bias in normal conditions. The application illustrated the variability of the results between the methods and the importance of choosing the most appropriate one. In conclusion, compared to propensity score-based methods, methods based on outcome regression should be preferred when suspecting near-violation of the positivity assumption.

Assessment of Glucose Lowering Medications' Effectiveness for Cardiovascular Clinical Risk Management of Real-World Patients with Type 2 Diabetes: Targeted Maximum Likelihood Estimation under Model Misspecification and Missing Outcomes

The results from many cardiovascular (CV) outcome trials suggest that glucose lowering medications (GLMs) are effective for the CV clinical risk management of type 2 diabetes (T2D) patients. The aim of this study is to compare the effectiveness of two GLMs (SGLT2i and GLP-1RA) for the CV clinical risk management of T2D patients in a real-world setting, by simultaneously reducing glycated hemoglobin, body weight, and systolic blood pressure. Data from the real-world Italian multicenter retrospective study Dapagliflozin Real World evideNce in Type 2 Diabetes (DARWINT 2D) are analyzed. Different statistical approaches are compared to deal with the real-world-associated issues, which can arise from model misspecification, nonrandomized treatment assignment, and a high percentage of missingness in the outcome, and can potentially bias the marginal treatment effect (MTE) estimate and thus have an influence on the clinical risk management of patients. We compare the logistic regression (LR), propensity score (PS)-based methods, and the targeted maximum likelihood estimator (TMLE), which allows for the use of machine learning (ML) models. Furthermore, a simulation study is performed, resembling the structure of the conditional dependencies among the main variables in DARWIN-T2D. LR and PS methods do not underline any difference in the effectiveness regarding the attainment of combined CV risk factor goals between the two treatments. TMLE suggests instead that dapagliflozin is significantly more effective than GLP-1RA for the CV risk management of T2D patients. The results from the simulation study suggest that TMLE has the lowest bias and SE for the estimate of the MTE.

A Machine Learning Approach to Predicting Perceived Partner Support From Relational and Individual Variables

Perceiving one’s partner as supportive is considered essential for relationships, but we know little about which factors are central to predicting perceived partner support. Traditional statistical techniques are ill-equipped to compare a large number of potential predictor variables and cannot answer this question. This research used machine learning analysis (random forest with Shapley values) to identify the most salient self-report predictors of perceived partner support cross-sectionally and 6 months later. We analyzed data from five dyadic data sets ( N = 550 couples) enabling us to have greater confidence in the findings and ensure generalizability. Our novel results advance the literature by showing that relationship variables and attachment avoidance are central to perceived partner support, whereas partner similarity, other individual differences, individual well-being, and demographics explain little variance in perceiving partners as supportive. The findings are crucial in constraining and further developing our theories on perceived partner support.

Cumming S, Diop A, Talbot D. Confounding adjustment methods for multi-level treatment comparisons under lack of positivity and unknown model specification

Imbalances in covariates between treatment groups are frequent in observational studies and can lead to biased comparisons. Various adjustment methods can be employed to correct these biases in the context of multi-level treatments (> 2). Analytical challenges, such as positivity violations and incorrect model specification due to unknown functional relationships between covariates and treatment or outcome, may affect their ability to yield unbiased results. Such challenges were expected in a comparison of fire-suppression interventions for preventing fire growth. We identified the overlap weights, augmented overlap weights, bias-corrected matching and targeted maximum likelihood as methods with the best potential to address those challenges. A simple variance estimator for the overlap weight estimators that can naturally be combined with machine learning is proposed. In a simulation study, we investigated the performance of these methods as well as those of simpler alternatives. Adjustment methods that included an outcome modeling component performed better than those that focused on the treatment mechanism in our simulations. Additionally, machine learning implementation was observed to efficiently compensate for the unknown model specification for the former methods, but not the latter. Based on these results, we compared the effectiveness of fire-suppression interventions using the augmented overlap weight estimator.

Association Between Telomere Length and Skin Cancer and Aging: A Mendelian Randomization Analysis

Background: Telomere shortening is a hallmark of cellular senescence. However, telomere length (TL)-related cellular senescence has varying effects in different cancers, resulting in a paradoxical relationship between senescence and cancer. Therefore, we used observational epidemiological studies to investigate the association between TL and skin cancer and aging, and to explore whether such a paradoxical relationship exists in skin tissue.

Methods: This study employed two-sample Mendelian randomization (MR) to analyze the causal relationship between TL and skin cancer [melanoma and non-melanoma skin cancers (NMSCs)] and aging. We studied single nucleotide polymorphisms (SNPs) obtained from pooled data belonging to genome-wide association studies (GWAS) in the literature and biobanks. Quality control was performed using pleiotropy, heterogeneity, and sensitivity analyses.

Results: We used five algorithms to analyze the causal relationship between TL and skin aging, melanoma, and NMSCs, and obtained consistent results. TL shortening reduced NMSC and melanoma susceptibility risk with specific odds ratios (ORs) of 1.0344 [95% confidence interval (CI): 1.0168–1.0524, p = 0.01] and 1.0127 (95% CI: 1.0046–1.0209, p = 6.36E-07), respectively. Conversely, TL shortening was validated to increase the odds of skin aging (OR = 0.96, 95% CI: 0.9332–0.9956, p = 0.03). Moreover, the MR-Egger, maximum likelihood, and inverse variance weighted (IVW) methods found significant heterogeneity among instrumental variable (IV) estimates (identified as MR-Egger skin aging Q = 76.72, p = 1.36E-04; melanoma Q = 97.10, p = 1.62E-07; NMSCsQ = 82.02, p = 1.90E-05). The leave-one-out analysis also showed that the SNP sensitivity was robust to each result.

Conclusion: This study found that TL shortening may promote skin aging development and reduce the risk of cutaneous melanoma and NMSCs. The results provide a reference for future research on the causal relationship between skin aging and cancer in clinical practice.

Evaluating the robustness of targeted maximum likelihood estimators via realistic simulations in nutrition intervention trials

Several recently developed methods have the potential to harness machine learning in the pursuit of target quantities inspired by causal inference, including inverse weighting, doubly robust estimating equations and substitution estimators like targeted maximum likelihood estimation. There are even more recent augmentations of these procedures that can increase robustness, by adding a layer of cross-validation (cross-validated targeted maximum likelihood estimation and double machine learning, as applied to substitution and estimating equation approaches, respectively). While these methods have been evaluated individually on simulated and experimental data sets, a comprehensive analysis of their performance across real data based simulations have yet to be conducted. In this work, we benchmark multiple widely used methods for estimation of the average treatment effect using ten different nutrition intervention studies data. A nonparametric regression method, undersmoothed highly adaptive lasso, is used to generate the simulated distribution which preserves important features from the observed data and reproduces a set of true target parameters. For each simulated data, we apply the methods above to estimate the average treatment effects as well as their standard errors and resulting confidence intervals. Based on the analytic results, a general recommendation is put forth for use of the cross-validated variants of both substitution and estimating equation estimators. We conclude that the additional layer of cross-validation helps in avoiding unintentional over-fitting of nuisance parameter functionals and leads to more robust inferences.

Does Amount of Information Support Aesthetic Values?

Obtaining information from the world is important for survival. The brain, therefore, has special mechanisms to extract as much information as possible from sensory stimuli. Hence, given its importance, the amount of available information may underlie aesthetic values. Such information-based aesthetic values would be significant because they would compete with others to drive decision-making. In this article, we ask, "What is the evidence that amount of information support aesthetic values?" An important concept in the measurement of informational volume is entropy. Research on aesthetic values has thus used Shannon entropy to evaluate the contribution of quantity of information. We review here the concepts of information and aesthetic values, and research on the visual and auditory systems to probe whether the brain uses entropy or other relevant measures, specially, Fisher information, in aesthetic decisions. We conclude that information measures contribute to these decisions in two ways: first, the absolute quantity of information can modulate aesthetic preferences for certain sensory patterns. However, the preference for volume of information is highly individualized, with information-measures competing with organizing principles, such as rhythm and symmetry. In addition, people tend to be resistant to too much entropy, but not necessarily, high amounts of Fisher information. We show that this resistance may stem in part from the distribution of amount of information in natural sensory stimuli. Second, the measurement of entropic-like quantities over time reveal that they can modulate aesthetic decisions by varying degrees of surprise given temporally integrated expectations. We propose that amount of information underpins complex aesthetic values, possibly informing the brain on the allocation of resources or the situational appropriateness of some cognitive models.

State-Level Masking Mandates and COVID-19 Outcomes in the United States: A Demonstration of the Causal Roadmap

BACKGROUND

We sought to investigate the effect of public masking mandates in US states on COVID-19 at the national level in Fall 2020. Specifically, we aimed to evaluate how the relative growth of COVID-19 cases and deaths would have differed if all states had issued a mandate to mask in public by 1 September 2020 versus if all states had delayed issuing such a mandate.

METHODS

We applied the Causal Roadmap, a formal framework for causal and statistical inference. We defined the outcome as the state-specific relative increase in cumulative cases and in cumulative deaths 21, 30, 45, and 60 days after 1 September. Despite the natural experiment occurring at the state-level, the causal effect of masking policies on COVID-19 outcomes was not identifiable. Nonetheless, we specified the target statistical parameter as the adjusted rate ratio (aRR): the expected outcome with early implementation divided by the expected outcome with delayed implementation, after adjusting for state-level confounders. To minimize strong estimation assumptions, primary analyses used targeted maximum likelihood estimation with Super Learner.

RESULTS

After 60 days and at a national level, early implementation was associated with a 9% reduction in new COVID-19 cases (aRR = 0.91 [95% CI = 0.88, 0.95]) and a 16% reduction in new COVID-19 deaths (aRR = 0.84 [95% CI = 0.76, 0.93]).

CONCLUSIONS

Although lack of identifiability prohibited causal interpretations, application of the Causal Roadmap facilitated estimation and inference of statistical associations, providing timely answers to pressing questions in the COVID-19 response.

Introduction to computational causal inference using reproducible Stata, R, and Python code: A tutorial

The main purpose of many medical studies is to estimate the effects of a treatment or exposure on an outcome. However, it is not always possible to randomize the study participants to a particular treatment, therefore observational study designs may be used. There are major challenges with observational studies; one of which is confounding. Controlling for confounding is commonly performed by direct adjustment of measured confounders; although, sometimes this approach is suboptimal due to modeling assumptions and misspecification. Recent advances in the field of causal inference have dealt with confounding by building on classical standardization methods. However, these recent advances have progressed quickly with a relative paucity of computational-oriented applied tutorials contributing to some confusion in the use of these methods among applied researchers. In this tutorial, we show the computational implementation of different causal inference estimators from a historical perspective where new estimators were developed to overcome the limitations of the previous estimators (ie, nonparametric and parametric g-formula, inverse probability weighting, double-robust, and data-adaptive estimators). We illustrate the implementation of different methods using an empirical example from the Connors study based on intensive care medicine, and most importantly, we provide reproducible and commented code in Stata, R, and Python for researchers to adapt in their own observational study. The code can be accessed at https://github.com/migariane/Tutorial_Computational_Causal_Inference_Estimators.

Recombinant hepatitis B vaccine uptake and multiple sclerosis risk: A marginal structural modeling approach

BACKGROUND

There has been an ongoing debate regarding the association between uptake of recombinant vaccine against hepatitis B virus (HBV) and multiple sclerosis (MS) risk. This case-control study tested the hypothesis whether recombinant HBV vaccination status is causally associated with MS risk using targeted maximum likelihood estimation (TMLE) technique.

METHODS

Confirmed 110 MS cases and age (± 5 years), sex and nativity matched (1:1) 110 controls were enrolled. Data were collected on sociodemographics, environmental factors, history of vaccinations and past morbidities through face-to-face interview both from cases and controls. To estimate the causal parameters including marginal odds ratio (OR), causal relative risk (RR), causal risk difference (RD) and their 95% confidence intervals (CIs), we implemented case-control-weighted TMLE for a matched design that uses data-adaptive flexible stacked ensemble-based machine learning system namely Super Learner. Additionally, population preventable fraction (PPF) of MS risk was computed.

RESULTS

This study demonstrated a significant nonspecific protective effect of HBV vaccination against MS risk (marginal OR 0.44; 95% CI: 0.19-0.68; p = 0.006; causal RR 0.64, 95% CI: 0.46-0.89; p = 0.004). The significant causal RD showed that among the vaccinated 19% fewer MS cases occurred owing to their HBV vaccination (causal RD -0.19; 95% CI: -0.32 - -0.06; p = 0.014). In the source population, vaccination against HBV led to 17.4% reduced MS risk (PPF = 17.4%; 95% CI: 3.8%, 36.3%).

CONCLUSION

The results suggest a significant nonspecific protective effect of recombinant HBV vaccine against MS risk. Future studies may contemplate to confirm these results.

A Comparison of Confounder Selection and Adjustment Methods for Estimating Causal Effects Using Large Healthcare Databases

PURPOSE

Confounding adjustment is required to estimate the effect of an exposure on an outcome in observational studies. However, variable selection and unmeasured confounding are particularly challenging when analyzing large healthcare data. Machine learning methods may help address these challenges. The objective was to evaluate the capacity of such methods to select confounders and reduce unmeasured confounding bias.

METHODS

A simulation study with known true effects was conducted. Completely synthetic and partially synthetic data incorporating real large healthcare data were generated. We compared Bayesian Adjustment for Confounding, Generalized Bayesian Causal Effect Estimation, Group Lasso and Doubly Robust Estimation, high-dimensional propensity score, and scalable collaborative targeted maximum likelihood algorithms. For the high-dimensional propensity score, two adjustment approaches targeting the effect in the whole population were considered: full matching and inverse probability weighting.

RESULTS

In scenarios without hidden confounders, most methods were essentially unbiased. The bias and variance of the high-dimensional propensity score varied considerably according to the number of variables selected by the algorithm. In scenarios with hidden confounders, substantial bias reduction was achieved by using machine learning methods to identify proxies as compared to adjusting only by observed confounders. High-dimensional propensity score and Group Lasso performed poorly in the partially synthetic simulation. Bayesian Adjustment for Confounding, Generalized Bayesian Causal Effect Estimation, and scalable collaborative targeted maximum likelihood algorithms performed particularly well.

CONCLUSIONS

Machine learning can help to identify measured confounders in large healthcare databases. They can also capitalize on proxies of unmeasured confounders to substantially reduce residual confounding bias. This article is protected by copyright. All rights reserved.

G-computation and doubly robust standardisation for continuous-time data: A comparison with inverse probability weighting

In time-to-event settings, g-computation and doubly robust estimators are based on discrete-time data. However, many biological processes are evolving continuously over time. In this paper, we extend the g-computation and the doubly robust standardisation procedures to a continuous-time context. We compare their performance to the well-known inverse-probability-weighting estimator for the estimation of the hazard ratio and restricted mean survival times difference, using a simulation study. Under a correct model specification, all methods are unbiased, but g-computation and the doubly robust standardisation are more efficient than inverse-probability-weighting. We also analyse two real-world datasets to illustrate the practical implementation of these approaches. We have updated the R package RISCA to facilitate the use of these methods and their dissemination.

Gluten-free Diet Reduces the Risk of Irritable Bowel Syndrome: A Mendelian Randomization Analysis

Background: Whether a gluten-free diet (GFD) is a cause of irritable bowel syndrome (IBS) remains controversial. We aim at exploring the causal relationship between gluten intake and IBS within Mendelian randomization (MR) design. Methods: We conducted a two-sample MR and selected single-nucleotide polymorphisms (SNPs) associated with GFD as instrumental variables (IVs). SNPs and genetic associations with GFD and IBS were obtained from the latest genome-wide association studies (GWAS) in Europeans (GFD: cases: 1,376; controls: 63,573; IBS: cases:1,121; controls: 360,073). We performed inverse variance weighting (IVW) as the primary method with several sensitivity analyses like MR-Egger and MR-PRESSO for quality control. The above analyses were re-run using another large dataset of IBS, as well as changing the p-value threshold when screening IVs, to verify the stability of the results. Results: The final estimate indicated significant causal association [per one copy of effect allele predicted log odds ratio (OR) change in GFD intake: OR = 0.97, 95% confidence interval (CI) 0.96 to 0.99, p < 0.01] without heterogeneity statistically (Q = 2.48, p = 0.78) nor horizontal pleiotropy biasing the causality (p = 0.92). Consistent results were found in validation analyses. Results of MR Steiger directionality test indicated the accuracy of our estimate of the causal direction (Steiger p < 0.001). Conclusion: GFD might be a protective factor of IBS. Therefore, we suggest taking a diet of lower gluten intake into account in IBS prevention and clinical practice.

Hospitalization outcomes among brain metastasis patients receiving radiation therapy with or without stereotactic radiosurgery from the 2005-2014 Nationwide Inpatient Sample

The purpose of this study was to compare hospitalization outcomes among US inpatients with brain metastases who received stereotactic radiosurgery (SRS) and/or non-SRS radiation therapies without neurosurgical intervention. A cross-sectional study was conducted whereby existing data on 35,199 hospitalization records (non-SRS alone: 32,981; SRS alone: 1035; SRS + non-SRS: 1183) from 2005 to 2014 Nationwide Inpatient Sample were analyzed. Targeted maximum likelihood estimation and Super Learner algorithms were applied to estimate average treatment effects (ATE), marginal odds ratios (MOR) and causal risk ratio (CRR) for three distinct types of radiation therapy in relation to hospitalization outcomes, including length of stay (' ≥ 7 days' vs. ' < 7 days') and discharge destination ('non-routine' vs. 'routine'), controlling for patient and hospital characteristics. Recipients of SRS alone (ATE = - 0.071, CRR = 0.88, MOR = 0.75) or SRS + non-SRS (ATE = - 0.17, CRR = 0.70, MOR = 0.50) had shorter hospitalizations as compared to recipients of non-SRS alone. Recipients of SRS alone (ATE = - 0.13, CRR = 0.78, MOR = 0.59) or SRS + non-SRS (ATE = - 0.17, CRR = 0.72, MOR = 0.51) had reduced risks of non-routine discharge as compared to recipients of non-SRS alone. Similar analyses suggested recipients of SRS alone had shorter hospitalizations and similar risk of non-routine discharge when compared to recipients of SRS + non-SRS radiation therapies. SRS alone or in combination with non-SRS therapies may reduce the risks of prolonged hospitalization and non-routine discharge among hospitalized US patients with brain metastases who underwent radiation therapy without neurosurgical intervention.

The change in estimate method for selecting confounders: A simulation study

BACKGROUND

The change in estimate is a popular approach for selecting confounders in epidemiology. It is recommended in epidemiologic textbooks and articles over significance test of coefficients, but concerns have been raised concerning its validity. Few simulation studies have been conducted to investigate its performance.

METHODS

An extensive simulation study was realized to compare different implementations of the change in estimate method. The implementations were also compared when estimating the association of body mass index with diastolic blood pressure in the PROspective Québec Study on Work and Health.

RESULTS

All methods were susceptible to introduce important bias and to produce confidence intervals that included the true effect much less often than expected in at least some scenarios. Overall mixed results were obtained regarding the accuracy of estimators, as measured by the mean squared error. No implementation adequately differentiated confounders from non-confounders. In the real data analysis, none of the implementation decreased the estimated standard error.

CONCLUSION

Based on these results, it is questionable whether change in estimate methods are beneficial in general, considering their low ability to improve the precision of estimates without introducing bias and inability to yield valid confidence intervals or to identify true confounders.

The Impact of Same-Day Antiretroviral Therapy Initiation Under the World Health Organization Treat-All Policy

Rapid initiation of antiretroviral therapy (ART) is recommended for people living with human immunodeficiency virus (HIV), with the option to start treatment on the day of diagnosis (same-day ART). However, the effect of same-day ART remains unknown in realistic public sector settings. We established a cohort of ≥16-year-old patients who initiated first-line ART under a treat-all policy in Nhlangano (Eswatini) during 2014-2016, either on the day of HIV care enrollment (same-day ART) or 1-14 days thereafter (early ART). Directed acyclic graphs, flexible parametric survival analysis, and targeted maximum likelihood estimation (TMLE) were used to estimate the effect of same-day-ART initiation on a composite unfavorable treatment outcome (loss to follow-up, death, viral failure, treatment switch). Of 1,328 patients, 839 (63.2%) initiated same-day ART. The adjusted hazard ratio of the unfavorable outcome was higher, 1.48 (95% confidence interval: 1.16, 1.89), for same-day ART compared with early ART. TMLE suggested that after 1 year, 28.9% of patients would experience the unfavorable outcome under same-day ART compared with 21.2% under early ART (difference: 7.7%; 1.3%-14.1%). This estimate was driven by loss to follow-up and varied over time, with a higher hazard during the first year after HIV care enrollment and a similar hazard thereafter. We found an increased risk with same-day ART. A limitation was that possible silent transfers that were not captured.

Uncovering the Most Important Factors for Predicting Sexual Desire Using Explainable Machine Learning

BACKGROUND

Low sexual desire is the most common sexual problem reported with 34% of women and 15% of men reporting lack of desire for at least 3 months in a 12-month period. Sexual desire has previously been associated with both relationship and individual well-being highlighting the importance of understanding factors that contribute to sexual desire as improving sexual desire difficulties can help improve an individual's overall quality of life.

AIM

The purpose of the present study was to identify the most salient individual (eg, attachment style, attitudes toward sexuality, gender) and relational (eg, relationship satisfaction, sexual satisfaction, romantic love) predictors of dyadic and solitary sexual desire from a large number of predictor variables.

METHODS

Previous research has relied primarily on traditional statistical models which are limited in their ability to estimate a large number of predictors, non-linear associations, and complex interactions. We used a machine learning algorithm, random forest (a type of highly non-linear decision tree), to circumvent these issues to predict dyadic and solitary sexual desire from a large number of predictors across 2 online samples (N = 1,846; includes 754 individuals forming 377 couples). We also used a Shapley value technique to estimate the size and direction of the effect of each predictor variable on the model outcome.

OUTCOMES

The outcomes included total, dyadic, and solitary sexual desire measured using the Sexual Desire Inventory.

RESULTS

The models predicted around 40% of variance in dyadic and solitary desire with women's desire being more predictable than men's overall. Several variables consistently predicted dyadic sexual desire such as sexual satisfaction and romantic love, and solitary desire such as masturbation and attitudes toward sexuality. These predictors were similar for both men and women and gender was not an important predictor of sexual desire.

CLINICAL TRANSLATION

The results highlight the importance of addressing overall relationship satisfaction when sexual desire difficulties are presented in couples therapy. It is also important to understand clients' attitudes toward sexuality.

STRENGTHS & LIMITATIONS

The study improves on existing methodologies in the field and compares a large number of predictors of sexual desire. However, the data were cross-sectional and there may have been variables that are important for desire but were not present in the datasets.

CONCLUSION

Higher sexual satisfaction and feelings of romantic love toward one's partner are important predictors of dyadic sexual desire whereas regular masturbation and more permissive attitudes toward sexuality predicted solitary sexual desire. Vowels LM, Vowels MJ, Mark KP. Uncovering the Most Important Factors for Predicting Sexual Desire Using Explainable Machine Learning. J Sex Med 2021;18:1198-1216.

Identification of microenvironment related potential biomarkers of biochemical recurrence at 3 years after prostatectomy in prostate adenocarcinoma

Prostate adenocarcinoma is one of the leading adult malignancies. Identification of multiple causative biomarkers is necessary and helpful for determining the occurrence and prognosis of prostate adenocarcinoma. We aimed to identify the potential prognostic genes in the prostate adenocarcinoma microenvironment and to estimate the causal effects simultaneously. We obtained the gene expression data of prostate adenocarcinoma from TCGA project and identified the differentially expressed genes based on immune-stromal components. Among these genes, 68 were associated with biochemical recurrence at 3 years after prostatectomy in prostate adenocarcinoma. After adjusting for the minimal sets of confounding covariates, 14 genes (TNFRSF4, ZAP70, ERMN, CXCL5, SPINK6, SLC6A18, CHRM2, TG, CLLU1OS, POSTN, CTSG, NETO1, CEACAM7, and IGLV3-22) related to the microenvironment were identified as prognostic biomarkers using the targeted maximum likelihood estimation. Both the average and individual causal effects were obtained to measure the magnitude of the effect. CIBERSORT and gene set enrichment analyses showed that these prognostic genes were mainly associated with immune responses. POSTN and NETO1 were correlated with androgen receptor expression, a main driver of prostate adenocarcinoma progression. Finally, five genes were validated in another prostate adenocarcinoma cohort (GEO: GSE70770). These findings might lead to the improved prognosis of prostate adenocarcinoma.

Identification of Tumor Microenvironment-Related Prognostic Biomarkers for Ovarian Serous Cancer 3-Year Mortality Using Targeted Maximum Likelihood Estimation: A TCGA Data Mining Study

Ovarian serous cancer (OSC) is one of the leading causes of death across the world. The role of the tumor microenvironment (TME) in OSC has received increasing attention. Targeted maximum likelihood estimation (TMLE) is developed under a counterfactual framework to produce effect estimation for both the population level and individual level. In this study, we aim to identify TME-related genes and using the TMLE method to estimate their effects on the 3-year mortality of OSC. In total, 285 OSC patients from the TCGA database constituted the studying population. ESTIMATE algorithm was implemented to evaluate immune and stromal components in TME. Differential analysis between high-score and low-score groups regarding ImmuneScore and StromalScore was performed to select shared differential expressed genes (DEGs). Univariate logistic regression analysis was followed to evaluate associations between DEGs and clinical pathologic factors with 3-year mortality. TMLE analysis was conducted to estimate the average effect (AE), individual effect (IE), and marginal odds ratio (MOR). The validation was performed using three datasets from Gene Expression Omnibus (GEO) database. Additionally, 355 DEGs were selected after differential analysis, and 12 genes from DEGs were significant after univariate logistic regression. Four genes remained significant after TMLE analysis. In specific, ARID3C and FREM2 were negatively correlated with OSC 3-year mortality. CROCC2 and PTF1A were positively correlated with OSC 3-year mortality. Combining of ESTIMATE algorithm and TMLE algorithm, we identified four TME-related genes in OSC. AEs were estimated to provide averaged effects based on the population level, while IEs were estimated to provide individualized effects and may be helpful for precision medicine.

Reflection on modern methods: when worlds collide-prediction, machine learning and causal inference

Causal inference requires theory and prior knowledge to structure analyses, and is not usually thought of as an arena for the application of prediction modelling. However, contemporary causal inference methods, premised on counterfactual or potential outcomes approaches, often include processing steps before the final estimation step. The purposes of this paper are: (i) to overview the recent emergence of prediction underpinning steps in contemporary causal inference methods as a useful perspective on contemporary causal inference methods, and (ii) explore the role of machine learning (as one approach to 'best prediction') in causal inference. Causal inference methods covered include propensity scores, inverse probability of treatment weights (IPTWs), G computation and targeted maximum likelihood estimation (TMLE). Machine learning has been used more for propensity scores and TMLE, and there is potential for increased use in G computation and estimation of IPTWs.

Using computable knowledge mined from the literature to elucidate confounders for EHR-based pharmacovigilance

INTRODUCTION

Drug safety research asks causal questions but relies on observational data. Confounding bias threatens the reliability of studies using such data. The successful control of confounding requires knowledge of variables called confounders affecting both the exposure and outcome of interest. However, causal knowledge of dynamic biological systems is complex and challenging. Fortunately, computable knowledge mined from the literature may hold clues about confounders. In this paper, we tested the hypothesis that incorporating literature-derived confounders can improve causal inference from observational data.

METHODS

We introduce two methods (semantic vector-based and string-based confounder search) that query literature-derived information for confounder candidates to control, using SemMedDB, a database of computable knowledge mined from the biomedical literature. These methods search SemMedDB for confounders by applying semantic constraint search for indications treated by the drug (exposure) and that are also known to cause the adverse event (outcome). We then include the literature-derived confounder candidates in statistical and causal models derived from free-text clinical notes. For evaluation, we use a reference dataset widely used in drug safety containing labeled pairwise relationships between drugs and adverse events and attempt to rediscover these relationships from a corpus of 2.2 M NLP-processed free-text clinical notes. We employ standard adjustment and causal inference procedures to predict and estimate causal effects by informing the models with varying numbers of literature-derived confounders and instantiating the exposure, outcome, and confounder variables in the models with dichotomous EHR-derived data. Finally, we compare the results from applying these procedures with naive measures of association (χ2 and reporting odds ratio) and with each other.

RESULTS AND CONCLUSIONS

We found semantic vector-based search to be superior to string-based search at reducing confounding bias. However, the effect of including more rather than fewer literature-derived confounders was inconclusive. We recommend using targeted learning estimation methods that can address treatment-confounder feedback, where confounders also behave as intermediate variables, and engaging subject-matter experts to adjudicate the handling of problematic covariates.

Effect of perinatal depression on risk of adverse infant health outcomes in mother-infant dyads in Gondar town: a causal analysis

BACKGROUND

Approximately one-third of pregnant and postnatal women in Ethiopia experience depression posing a substantial health burden for these women and their families. Although associations between postnatal depression and worse infant health have been observed, there have been no studies to date assessing the causal effects of perinatal depression on infant health in Ethiopia. We applied longitudinal data and recently developed causal inference methods that reduce the risk of bias to estimate associations between perinatal depression and infant diarrhea, Acute Respiratory Infection (ARI), and malnutrition in Gondar Town, Ethiopia.

METHODS

A cohort of 866 mother-infant dyads were followed from infant birth for 6 months and the cumulative incidence of ARI, diarrhea, and malnutrition were assessed. The Edinburgh Postnatal Depression Scale (EPDS) was used to assess the presence of maternal depression, the Integrated Management of Newborn and Childhood Illnesses (IMNCI) guidelines were used to identify infant ARI and diarrhea, and the mid upper arm circumference (MUAC) was used to identify infant malnutrition. The risk difference (RD) due to maternal depression for each outcome was estimated using targeted maximum likelihood estimation (TMLE), a doubly robust causal inference method used to reduce bias in observational studies.

RESULTS

The cumulative incidence of diarrhea, ARI and malnutrition during 6-month follow-up was 17.0% (95%CI: 14.5, 19.6), 21.6% (95%CI: 18.89, 24.49), and 14.4% (95%CI: 12.2, 16.9), respectively. There was no association between antenatal depression and ARI (RD = - 1.3%; 95%CI: - 21.0, 18.5), diarrhea (RD = 0.8%; 95%CI: - 9.2, 10.9), or malnutrition (RD = -7.3%; 95%CI: - 22.0, 21.8). Similarly, postnatal depression was not associated with diarrhea (RD = -2.4%; 95%CI: - 9.6, 4.9), ARI (RD = - 3.2%; 95%CI: - 12.4, 5.9), or malnutrition (RD = 0.9%; 95%CI: - 7.6, 9.5).

CONCLUSION

There was no evidence for an association between perinatal depression and the risk of infant diarrhea, ARI, and malnutrition amongst women in Gondar Town. Previous reports suggesting increased risks resulting from maternal depression may be due to unobserved confounding.

Far from MCAR: Obtaining Population-level Estimates of HIV Viral Suppression

BACKGROUND

Population-level estimates of disease prevalence and control are needed to assess prevention and treatment strategies. However, available data often suffer from differential missingness. For example, population-level HIV viral suppression is the proportion of all HIV-positive persons with suppressed viral replication. Individuals with measured HIV status, and among HIV-positive individuals those with measured viral suppression, likely differ from those without such measurements.

METHODS

We discuss three sets of assumptions to identify population-level suppression in the intervention arm of the SEARCH Study (NCT01864603), a community randomized trial in rural Kenya and Uganda (2013-2017). Using data on nearly 100,000 participants, we compare estimates from (1) an unadjusted approach assuming data are missing-completely-at-random (MCAR); (2) stratification on age group, sex, and community; and (3) targeted maximum likelihood estimation to adjust for a larger set of baseline and time-updated variables.

RESULTS

Despite high measurement coverage, estimates of population-level viral suppression varied by identification assumption. Unadjusted estimates were most optimistic: 50% (95% confidence interval [CI] = 46%, 54%) of HIV-positive persons suppressed at baseline, 80% (95% CI = 78%, 82%) at year 1, 85% (95% CI = 83%, 86%) at year 2, and 85% (95% CI = 83%, 87%) at year 3. Stratifying on baseline predictors yielded slightly lower estimates, and full adjustment reduced estimates meaningfully: 42% (95% CI = 37%, 46%) of HIV-positive persons suppressed at baseline, 71% (95% CI = 69%, 73%) at year 1, 76% (95% CI = 74%, 78%) at year 2, and 79% (95% CI = 77%, 81%) at year 3.

CONCLUSIONS

Estimation of population-level disease burden and control requires appropriate adjustment for missing data. Even in large studies with limited missingness, estimates relying on the MCAR assumption or baseline stratification should be interpreted cautiously.

Maternal folate levels during pregnancy and offspring brain development in late childhood

BACKGROUND

Cumulative evidence shows that low maternal folate levels during pregnancy are associated with offspring neuropsychiatric disorders even in the absence of neural tube defects. However, the relationship between prenatal exposure to folate and brain development in late childhood has been rarely investigated.

METHODS

In 2095 children from a prospective population-based cohort in Rotterdam, the Netherlands, we examined the association of maternal folate levels during pregnancy with downstream brain development in offspring. Maternal folate concentrations were measured from venous blood in early gestation. Child structural neuroimaging data were measured at age 9-11 years. In addition, measures of child head circumference using fetal ultrasound in the third trimester and total brain volume using magnetic resonance imaging at age 6-8 years were used for analyses with repeated assessments of brain development.

RESULTS

Maternal folate deficiency (i.e., <7 nmol/L) during pregnancy was associated with smaller total brain volume (B = -18.7 cm³, 95% CI -37.2 to -0.2) and smaller cerebral white matter (B = -7.2 cm³, 95% CI -11.8 to -2.6) in children aged 9-11 years. No differences in cortical thickness or surface area were observed. Analysis of the repeated brain assessments showed that children exposed to deficient folate concentrations in utero had persistently smaller brains compared to controls from the third trimester to childhood (β = -0.4, 95% CI -0.6 to -0.1).

CONCLUSIONS

Low maternal folate levels during pregnancy are associated with altered offspring brain development in childhood, suggesting the importance of essential folate concentrations in early pregnancy.

A Systematic Review of Methods Used for Confounding Adjustment in Observational Economic Evaluations in Cardiology Conducted between 2013 and 2017

Background. Observational economic evaluations (i.e., economic evaluations in which treatment allocation is not randomized) are prone to confounding bias. Prior reviews published in 2013 have shown that adjusting for confounding is poorly done, if done at all. Although these reviews raised awareness on the issues, it is unclear if their results improved the methodological quality of future work. We therefore aimed to investigate whether and how confounding was accounted for in recently published observational economic evaluations in the field of cardiology. Methods. We performed a systematic review of PubMed, Embase, Cochrane Library, Web of Science, and PsycInfo databases using a set of Medical Subject Headings and keywords covering topics in "observational economic evaluations in health within humans" and "cardiovascular diseases." Any study published in either English or French between January 1, 2013, and December 31, 2017, addressing our search criteria was eligible for inclusion in our review. Our protocol was registered with PROSPERO (CRD42018112391). Results. Forty-two (0.6%) out of 7523 unique citations met our inclusion criteria. Fewer than half of the selected studies adjusted for confounding (n = 19 [45.2%]). Of those that adjusted for confounding, propensity score matching (n = 8 [42.1%]) and other matching-based approaches were favored (n = 8 [42.1%]). Our results also highlighted that most authors who adjusted for confounding rarely justified their methodological choices. Conclusion. Our results indicate that adjustment for confounding is often ignored when conducting an observational economic evaluation. Continued knowledge translation efforts aimed at improving researchers' knowledge regarding confounding bias and methods aimed at addressing this issue are required and should be supported by journal editors.

G-computation, propensity score-based methods, and targeted maximum likelihood estimator for causal inference with different covariates sets: a comparative simulation study

Controlling for confounding bias is crucial in causal inference. Distinct methods are currently employed to mitigate the effects of confounding bias. Each requires the introduction of a set of covariates, which remains difficult to choose, especially regarding the different methods. We conduct a simulation study to compare the relative performance results obtained by using four different sets of covariates (those causing the outcome, those causing the treatment allocation, those causing both the outcome and the treatment allocation, and all the covariates) and four methods: g-computation, inverse probability of treatment weighting, full matching and targeted maximum likelihood estimator. Our simulations are in the context of a binary treatment, a binary outcome and baseline confounders. The simulations suggest that considering all the covariates causing the outcome led to the lowest bias and variance, particularly for g-computation. The consideration of all the covariates did not decrease the bias but significantly reduced the power. We apply these methods to two real-world examples that have clinical relevance, thereby illustrating the real-world importance of using these methods. We propose an R package RISCA to encourage the use of g-computation in causal inference.

Association of Overweight and Obesity Development Between Pregnancies With Stillbirth and Infant Mortality in a Cohort of Multiparous Women

OBJECTIVE

To identify the association of newly developed prepregnancy overweight and obesity with stillbirth and infant mortality.

METHODS

We studied subsequent pregnancies of mothers who were normal weight at fertilization of their first identified pregnancy, from a population-based cohort that linked birth registry with death records in Pennsylvania, 2003-2013. Women with newly developed prepregnancy overweight and obesity were defined as those whose body mass index (BMI) before second pregnancy was between 25 and 29.9 or 30 or higher, respectively. Our main outcomes of interest were stillbirth (intrauterine death at 20 weeks of gestation or greater), infant mortality (less than 365 days after birth), neonatal death (less than 28 days after birth) and postneonatal death (29-365 days after birth). Associations of both prepregnancy BMI categories and continuous BMI with each outcome were estimated by nonparametric targeted minimum loss-based estimation and inverse-probability weighted dose-response curves, respectively, adjusting for race-ethnicity, smoking, and other confounders (eg, age, education).

RESULTS

A cohort of 212,889 women were included for infant mortality analysis (192,941 women for stillbirth analysis). The crude rate of stillbirth and infant mortality in these final analytic cohorts were 3.3 per 1,000 pregnancies and 2.9 per 1,000 live births, respectively. Compared with women who stayed at a normal weight in their second pregnancies, those becoming overweight had 1.4 (95% CI 0.6-2.1) excess stillbirths per 1,000 pregnancies. Those becoming obese had 3.6 (95% CI 1.3-5.9) excess stillbirths per 1,000 pregnancies and 2.4 (95% CI 0.4-4.4) excess neonatal deaths per 1,000 live births. There was a dose-response relationship between prepregnancy BMI increases of more than 2 units and increased risk of stillbirth and infant mortality. In addition, BMI increases were associated with higher risks of infant mortality among women with shorter interpregnancy intervals (less than 18 months) compared with longer intervals.

CONCLUSION

Transitioning from normal weight to overweight or obese between pregnancies was associated with an increased risk of stillbirth and neonatal mortality.