Dynamic learning of individual-level suicidal ideation trajectories to enhance mental health care.

There has recently been an increase in ongoing patient-report routine outcome monitoring for individuals within clinical care, which has corresponded to increased longitudinal information about an individual. However, many models that are aimed at clinical practice have difficulty fully incorporating this information. This is in part due to the difficulty in dealing with the irregularly time-spaced observations that are common in clinical data. Consequently, we built individual-level continuous-time trajectory models of suicidal ideation for a clinical population (N = 585) with data collected via a digital platform. We demonstrate how such models predict an individual's level and variability of future suicide ideation, with implications for the frequency that individuals may need to be observed. These individual-level predictions provide a more personalised understanding than other predictive methods and have implications for enhanced measurement-based care.

Panel stacking is a threat to consensus statement validity.

Consensus statements can be very influential in medicine and public health. Some of these statements use systematic evidence synthesis but others fail on this front. Many consensus statements use panels of experts to deduce perceived consensus through Delphi processes. We argue that stacking of panel members towards one particular position or narrative is a major threat, especially in absence of systematic evidence review. Stacking may involve financial conflicts of interest, but non-financial conflicts of strong advocacy can also cause major bias. Given their emerging importance, we describe here how such consensus statements may be misleading, by analysing in depth a recent high-impact Delphi consensus statement on COVID-19 recommendations as a case example. We demonstrate that many of the selected panel members and at least 35% of the core panel members had advocated towards COVID-19 elimination (zero-COVID) during the pandemic and were leading members of aggressive advocacy groups. These advocacy conflicts were not declared in the Delphi consensus publication, with rare exceptions. Therefore, we propose that consensus statements should always require rigorous evidence synthesis and maximal transparency on potential biases towards advocacy or lobbyist groups to be valid. While advocacy can have many important functions, its biased impact on consensus panels should be carefully avoided.

Plug-and-Play Model-Agnostic Counterfactual Policy Synthesis for Deep Reinforcement Learning-Based Recommendation.

Recent advances in recommender systems have proved the potential of reinforcement learning (RL) to handle the dynamic evolution processes between users and recommender systems. However, learning to train an optimal RL agent is generally impractical with commonly sparse user feedback data in the context of recommender systems. To circumvent the lack of interaction of current RL-based recommender systems, we propose to learn a general model-agnostic counterfactual synthesis (MACS) policy for counterfactual user interaction data augmentation. The counterfactual synthesis policy aims to synthesize counterfactual states while preserving significant information in the original state relevant to the user's interests, building upon two different training approaches we designed: learning with expert demonstrations and joint training. As a result, the synthesis of each counterfactual data is based on the current recommendation agent's interaction with the environment to adapt to users' dynamic interests. We integrate the proposed policy deep deterministic policy gradient (DDPG), soft actor critic (SAC), and twin delayed DDPG (TD3) in an adaptive pipeline with a recommendation agent that can generate counterfactual data to improve the performance of recommendation. The empirical results on both online simulation and offline datasets demonstrate the effectiveness and generalization of our counterfactual synthesis policy and verify that it improves the performance of RL recommendation agents.

The temporal dependencies between social, emotional and physical health factors in young people receiving mental healthcare: a dynamic Bayesian network analysis.

AIMS: The needs of young people attending mental healthcare can be complex and often span multiple domains (e.g., social, emotional and physical health factors). These factors often complicate treatment approaches and contribute to poorer outcomes in youth mental health. We aimed to identify how these factors interact over time by modelling the temporal dependencies between these transdiagnostic social, emotional and physical health factors among young people presenting for youth mental healthcare. METHODS: Dynamic Bayesian networks were used to examine the relationship between mental health factors across multiple domains (social and occupational function, self-harm and suicidality, alcohol and substance use, physical health and psychiatric syndromes) in a longitudinal cohort of 2663 young people accessing youth mental health services. Two networks were developed: (1) 'initial network', that shows the conditional dependencies between factors at first presentation, and a (2) 'transition network', how factors are dependent longitudinally. RESULTS: The 'initial network' identified that childhood disorders tend to precede adolescent depression which itself was associated with three distinct pathways or illness trajectories; (1) anxiety disorder; (2) bipolar disorder, manic-like experiences, circadian disturbances and psychosis-like experiences; (3) self-harm and suicidality to alcohol and substance use or functioning. The 'transition network' identified that over time social and occupational function had the largest effect on self-harm and suicidality, with direct effects on ideation (relative risk [RR], 1.79; CI, 1.59-1.99) and self-harm (RR, 1.32; CI, 1.22-1.41), and an indirect effect on attempts (RR, 2.10; CI, 1.69-2.50). Suicide ideation had a direct effect on future suicide attempts (RR, 4.37; CI, 3.28-5.43) and self-harm (RR, 2.78; CI, 2.55-3.01). Alcohol and substance use, physical health and psychiatric syndromes (e.g., depression and anxiety, at-risk mental states) were independent domains whereby all direct effects remained within each domain over time. CONCLUSIONS: This study identified probable temporal dependencies between domains, which has causal interpretations, and therefore can provide insight into their differential role over the course of illness. This work identified social, emotional and physical health factors that may be important early intervention and prevention targets. Improving social and occupational function may be a critical target due to its impacts longitudinally on self-harm and suicidality. The conditional independence of alcohol and substance use supports the need for specific interventions to target these comorbidities.

Bayesian network modelling to identify on-ramps to childhood obesity.

BACKGROUND: When tackling complex public health challenges such as childhood obesity, interventions focused on immediate causes, such as poor diet and physical inactivity, have had limited success, largely because upstream root causes remain unresolved. A priority is to develop new modelling frameworks to infer the causal structure of complex chronic disease networks, allowing disease "on-ramps" to be identified and targeted. METHODS: The system surrounding childhood obesity was modelled as a Bayesian network, using data from The Longitudinal Study of Australian Children. The existence and directions of the dependencies between factors represent possible causal pathways for childhood obesity and were encoded in directed acyclic graphs (DAGs). The posterior distribution of the DAGs was estimated using the Partition Markov chain Monte Carlo. RESULTS: We have implemented structure learning for each dataset at a single time point. For each wave and cohort, socio-economic status was central to the DAGs, implying that socio-economic status drives the system regarding childhood obesity. Furthermore, the causal pathway socio-economic status and/or parental high school levels → parental body mass index (BMI) → child's BMI existed in over 99.99% of posterior DAG samples across all waves and cohorts. For children under the age of 8 years, the most influential proximate causal factors explaining child BMI were birth weight and parents' BMI. After age 8 years, free time activity became an important driver of obesity, while the upstream factors influencing free time activity for boys compared with girls were different. CONCLUSIONS: Childhood obesity is largely a function of socio-economic status, which is manifest through numerous downstream factors. Parental high school levels entangle with socio-economic status, and hence, are on-ramp to childhood obesity. The strong and independent causal relationship between birth weight and childhood BMI suggests a biological link. Our study implies that interventions that improve the socio-economic status, including through increasing high school completion rates, may be effective in reducing childhood obesity prevalence.

AdaptSPEC-X: Covariate-Dependent Spectral Modeling of Multiple Nonstationary Time Series.

We present the AdaptSPEC-X method for the joint analysis of a panel of possibly nonstationary time series. The approach is Bayesian and uses a covariate-dependent infinite mixture model to incorporate multiple time series, with mixture components parameterized by a time-varying mean and log spectrum. The mixture components are based on AdaptSPEC, a nonparametric model which adaptively divides the time series into an unknown number of segments and estimates the local log spectra by smoothing splines. AdaptSPEC-X extends AdaptSPEC in three ways. First, through the infinite mixture, it applies to multiple time series linked by covariates. Second, it can handle missing values, a common feature of time series which can cause difficulties for nonparametric spectral methods. Third, it allows for a time-varying mean. Through these extensions, AdaptSPEC-X can estimate time-varying means and spectra at observed and unobserved covariate values, allowing for predictive inference. Estimation is performed by Markov chain Monte Carlo (MCMC) methods, combining data augmentation, reversible jump, and Riemann manifold Hamiltonian Monte Carlo techniques. We evaluate the methodology using simulated data, and describe applications to Australian rainfall data and measles incidence in the US. Software implementing the method proposed in this paper is available in the R package BayesSpec.

Forecasting for COVID-19 has failed.

Epidemic forecasting has a dubious track-record, and its failures became more prominent with COVID-19. Poor data input, wrong modeling assumptions, high sensitivity of estimates, lack of incorporation of epidemiological features, poor past evidence on effects of available interventions, lack of transparency, errors, lack of determinacy, consideration of only one or a few dimensions of the problem at hand, lack of expertise in crucial disciplines, groupthink and bandwagon effects, and selective reporting are some of the causes of these failures. Nevertheless, epidemic forecasting is unlikely to be abandoned. Some (but not all) of these problems can be fixed. Careful modeling of predictive distributions rather than focusing on point estimates, considering multiple dimensions of impact, and continuously reappraising models based on their validated performance may help. If extreme values are considered, extremes should be considered for the consequences of multiple dimensions of impact so as to continuously calibrate predictive insights and decision-making. When major decisions (e.g. draconian lockdowns) are based on forecasts, the harms (in terms of health, economy, and society at large) and the asymmetry of risks need to be approached in a holistic fashion, considering the totality of the evidence.

Effect estimates of COVID-19 non-pharmaceutical interventions are non-robust and highly model-dependent.

OBJECTIVE: To compare the inference regarding the effectiveness of the various non-pharmaceutical interventions (NPIs) for COVID-19 obtained from different SIR models. STUDY DESIGN AND SETTING: We explored two models developed by Imperial College that considered only NPIs without accounting for mobility (model 1) or only mobility (model 2), and a model accounting for the combination of mobility and NPIs (model 3). Imperial College applied models 1 and 2 to 11 European countries and to the USA, respectively. We applied these models to 14 European countries (original 11 plus another 3), over two different time horizons. RESULTS: While model 1 found that lockdown was the most effective measure in the original 11 countries, model 2 showed that lockdown had little or no benefit as it was typically introduced at a point when the time-varying reproduction number was already very low. Model 3 found that the simple banning of public events was beneficial, while lockdown had no consistent impact. Based on Bayesian metrics, model 2 was better supported by the data than either model 1 or model 3 for both time horizons. CONCLUSION: Inferences on effects of NPIs are non-robust and highly sensitive to model specification. In the SIR modeling framework, the impacts of lockdown are uncertain and highly model-dependent.

A case study in model failure? COVID-19 daily deaths and ICU bed utilisation predictions in New York state.

Forecasting models have been influential in shaping decision-making in the COVID-19 pandemic. However, there is concern that their predictions may have been misleading. Here, we dissect the predictions made by four models for the daily COVID-19 death counts between March 25 and June 5 in New York state, as well as the predictions of ICU bed utilisation made by the influential IHME model. We evaluated the accuracy of the point estimates and the accuracy of the uncertainty estimates of the model predictions. First, we compared the "ground truth" data sources on daily deaths against which these models were trained. Three different data sources were used by these models, and these had substantial differences in recorded daily death counts. Two additional data sources that we examined also provided different death counts per day. For accuracy of prediction, all models fared very poorly. Only 10.2% of the predictions fell within 10% of their training ground truth, irrespective of distance into the future. For accurate assessment of uncertainty, only one model matched relatively well the nominal 95% coverage, but that model did not start predictions until April 16, thus had no impact on early, major decisions. For ICU bed utilisation, the IHME model was highly inaccurate; the point estimates only started to match ground truth after the pandemic wave had started to wane. We conclude that trustworthy models require trustworthy input data to be trained upon. Moreover, models need to be subjected to prespecified real time performance tests, before their results are provided to policy makers and public health officials.

The differential impact of major life events on cognitive and affective wellbeing.

Major life events affect our wellbeing. However the comparative impact of different events, which often co-occur, has not been systematically evaluated, or studies assumed that the impacts are equivalent in both amplitude and duration, that different wellbeing domains are equally affected, and that individuals exhibit hedonic adaptation. We evaluated the individual and conditional impact of eighteen major life-events, and compared their effects on affective and cognitive wellbeing in a large population-based cohort using fixed-effect regression models assessing within person change. Several commonly cited events had little, if any, independent effect on wellbeing (promotion, being fired, friends passing), whilst others had profound impacts regardless of co-occurring events (e.g., financial loss, death of partner, childbirth). No life events had overall positive effects on both types of wellbeing, but separation, injury/illnesses and monetary losses caused negative impacts on both, which did not display hedonic adaptation. Affective hedonic adaptation to all positive events occurred by two years but monetary gains and retirement had ongoing benefits on cognitive wellbeing. Marriage, retirement and childbirth had positive effects on cognitive wellbeing but no overall effect on affective wellbeing, whilst moving home was associated with a negative effect on cognitive wellbeing but no affective wellbeing response. Describing the independent impact of different life events, and, for some, the differential affective and life satisfaction responses, and lack of hedonic adaptation people display, may help clinicians, economists and policy-makers, but individual's hopes for happiness from positive events appears misplaced.

Bayesian Analysis of Individual Level Personality Dynamics.

A Bayesian technique with analyses of within-person processes at the level of the individual is presented. The approach is used to examine whether the patterns of within-person responses on a 12-trial simulation task are consistent with the predictions of ITA theory (Dweck, 1999). ITA theory states that the performance of an individual with an entity theory of ability is more likely to spiral down following a failure experience than the performance of an individual with an incremental theory of ability. This is because entity theorists interpret failure experiences as evidence of a lack of ability which they believe is largely innate and therefore relatively fixed; whilst incremental theorists believe in the malleability of abilities and interpret failure experiences as evidence of more controllable factors such as poor strategy or lack of effort. The results of our analyses support ITA theory at both the within- and between-person levels of analyses and demonstrate the benefits of Bayesian techniques for the analysis of within-person processes. These include more formal specification of the theory and the ability to draw inferences about each individual, which allows for more nuanced interpretations of individuals within a personality category, such as differences in the individual probabilities of spiraling. While Bayesian techniques have many potential advantages for the analyses of processes at the level of the individual, ease of use is not one of them for psychologists trained in traditional frequentist statistical techniques.