Sequential Set Research Articles

Sequential monitoring of time-to-event safety endpoints in clinical trials

Background/aims Safety monitoring is a crucial requirement for Phase II and III clinical trials. To protect patients from toxicity risk, stopping rules may be implemented that will halt the study if an unexpectedly high number of events occur. These rules are constructed using statistical procedures that typically treat the toxicity data as binary occurrences. Because the exact dates of toxicities are often available, a strategy that handles these as time-to-event data may offer higher power and require less calendar time to identify excess risk. This work investigates several statistical methods for monitoring safety events as time-to-event endpoints and illustrates our R software package for designing and evaluating these procedures. Methods The performance metrics of safety stopping rules derived from Wang-Tsiatis tests, Bayesian Gamma–Poisson models, and sequential probability ratio tests are evaluated and contrasted in Phase II and III trial scenarios. We developed a publicly available R package “stoppingrule” for designing and assessing these stopping rules whose utility is illustrated through the design of a stopping rule for Blood and Marrow Transplant Clinical Trials Network 1204 (National Clinical Trial number NCT01998633), a multicenter, Phase II, single-arm trial that assessed the efficacy and safety of bone marrow transplant for the treatment of hemophagocytic lymphohistiocytosis and primary immune deficiencies. Results As seen previously in group sequential testing settings, rules with strict stopping criteria early in a study tend to have more lenient stopping criteria late in the trial. Consequently, methods with aggressive early monitoring, such as Gamma–Poisson models with weak priors and certain choices of truncated sequential probability ratio tests, usually yield a smaller number of toxicities and lower power than ones that are more permissive at early stages, such as Gamma–Poisson models with strong priors and the O’Brien–Fleming test. The Pocock test and maximized sequential probability ratio test performed contrary to these trends, however, exhibiting both diminished power and higher numbers of toxicities than other methods due to their extremely aggressive early stopping criteria, failing to reserve adequate power to identify safety issues beyond the start of the study. In contrast to binary toxicity approaches, our time-to-event methods offer meaningful reductions in expected toxicities of up to 20% across scenarios considered. Conclusion Safety monitoring procedures aim to guard study participants from being exposed to and suffering toxicity from unsafe treatments. Toward this end, we recommend considering the time-to-event-oriented Gamma–Poisson model—weak prior model or truncated sequential probability ratio test for constructing safety stopping rules, as they performed the best in minimizing the number of toxicities in our investigations. Our R package “stoppingrule” offers procedures for creating and assessing stopping rules to aid trial design.

The use of sequential multiple assignment randomized trials (SMARTs) in physical activity interventions: a systematic review

BackgroundPhysical activity (PA) is often the cornerstone in risk-reduction interventions for the prevention and treatment of many chronic health conditions. PA interventions are inherently multi-dimensional and complex in nature. Thus, study designs used in the evaluation of PA interventions must be adaptive to intervention components and individual capacities. A Sequential Multiple Assignment Randomised Trial (SMART) is a factorial design in a sequential setting used to build effective adaptive interventions. SMARTs represent a relatively new design for PA intervention research. This systematic review aims to examine the state-of-the-art of SMARTs used to develop PA interventions, with a focus on study characteristics, design, and analyses.MethodsPubMed, Embase, PsychINFO, CENTRAL, and CinAHL were systematically searched through May 2023 for studies wherein PA SMARTs were conducted. Methodological quality was assessed using the Cochrane Risk of Bias 2 Tool.ResultsTwenty studies across a variety of populations – e.g., obesity, chronic pain, and cardiovascular conditions, were included. All PA SMARTs involved two decision stages, with the majority including two initial treatment options. PA interventions most commonly consisted of individual aerobic exercise with strategies such as goal setting, wearable technology, and motivational interviewing also used to promote PA. Variation was observed across tailoring variables and timing of tailoring variables. Non-response strategies primarily involved augmenting and switching treatment options, and for responders to continue with initial treatment options. For analyses, most sample size estimations and outcome analyses accounted for the SMART aims specified. Techniques such as linear mixed models, weighted regressions, and Q-learning regression were frequently used. Risk of bias was high across the majority of included studies.ConclusionsIndividual-based aerobic exercise interventions supported by behaviour change techniques and wearable sensing technology may play a key role in the future development of SMARTs addressing PA intervention development. Clearer rationale for the selection of tailoring variables, timing of tailoring variables, and included measures is essential to advance PA SMART designs. Collaborative efforts from researchers, clinicians, and patients are needed in order to bridge the gap between adaptive research designs and personalised treatment pathways observed in clinical practice.

A new bandit setting balancing information from state evolution and corrupted context

We propose a new sequential decision-making setting, combining key aspects of two established online learning problems with bandit feedback. The optimal action to play at any given moment is contingent on an underlying changing state that is not directly observable by the agent. Each state is associated with a context distribution, possibly corrupted, allowing the agent to identify the state. Furthermore, states evolve in a Markovian fashion, providing useful information to estimate the current state via state history. In the proposed problem setting, we tackle the challenge of deciding on which of the two sources of information the agent should base its action selection. We present an algorithm that uses a referee to dynamically combine the policies of a contextual bandit and a multi-armed bandit. We capture the time-correlation of states through iteratively learning the action-reward transition model, allowing for efficient exploration of actions. Our setting is motivated by adaptive mobile health (mHealth) interventions. Users transition through different, time-correlated, but only partially observable internal states, determining their current needs. The side information associated with each internal state might not always be reliable, and standard approaches solely rely on the context risk of incurring high regret. Similarly, some users might exhibit weaker correlations between subsequent states, leading to approaches that solely rely on state transitions risking the same. We analyze our setting and algorithm in terms of regret lower bound and upper bounds and evaluate our method on simulated medication adherence intervention data and several real-world data sets, showing improved empirical performance compared to several popular algorithms.

Rollout designs for lump-sum data

This work studies rollout design problems with a focus of suitable choices of rollout rate under the standard Type I and Type II error probabilities control framework. The main challenge of rollout design is that data is often observed in a lump-sum manner from a spatio-temporal point of view: (1) temporally, only the sum of data in a given sliding time window can be observed; (2) spatially, there are two subgroups for the data at each time step: control and treatment, but one can only observe the total values instead of individual values from each subgroup. We develop rollout tests of lump-sum data under both fixed-sample-size and sequential settings, subject to the constraints on Type I and Type II error probabilities. Numerical studies are conducted to validate our theoretical results.

Sequential rough set: a conservative extension of Pawlak’s classical rough set

Rough set theory is an important approach to deal with uncertainty in data mining. However, Pawlak’s classical rough set has low fault-tolerance on concept approximation based on knowledge granules, which may influence the classification accuracy in practical application. To address this problem, the present paper proposes a novel sequential rough-set model that is proved to be a conservative extension of Pawlak’s classical rough set. As a result, it effectively improves the fault-tolerance ability, classification accuracy and concept approximation accuracy of the latter without any additional assumption. Based on the properties and theoretical analysis of the proposed model, an algorithm is presented to automatically determine the sequential thresholds and compute the three regions for the given concept. Experiments on real data verify the validity of the algorithm, and also show the stable improvement on the two types of accuracy.

Increasing the efficiency of operation and management of railroad transport infrastructure based on maximum levels of fault tolerance

This paper considers the optimization of parameters for a railroad transport system. The maximum level of technological reliability and the average time spent by trains on the route are used as optimization criteria. The purpose of the study is to establish the optimal parameters for the operational process of railroad transport systems according to the criterion of the maximum level of technological reliability and the minimum time of trains on the route. Methods of technological reliability research have been proposed. Taking into account that the entire technological process is a sequential set of technological elements, a simulation model of the technological process of the transit transport-technological line along a route direction has been built. A population of agents that simulates the operation of railroad sections of the rotation of train locomotives and is a key subsystem of the simulation model has been developed and configured. The simulation model makes it possible to optimize the parameters of multi-section railroad lines. This approach is provided owing to the agent approach. As a result of the experiments, the optimal parameters of the functioning of railroad lines were established when organizing the passage of transit trains. The coefficient of utilization of the locomotive fleet fluctuates within the optimal range (0.55–0.65), which indicates the sufficiency of traction resources in the railroad system. The optimal parameters of the railroad transport system were established experimentally using the example of a train flow of 85 pairs of trains on a two-track route with five sections. The problem of "abandoned trains" has a solution but, to this end, it is necessary to increase the fleet of train locomotives by 150–200 % relative to existing standards. At the same time, even with an unlimited fleet of train locomotives, there is a fairly high probability (up to 30–50 %) of technological failures

Immunotherapy after progression to double immunotherapy: pembrolizumab and lenvatinib versus conventional chemotherapy for patients with metastatic melanoma after failure of PD-1/CTLA-4 inhibition.

Programmed cell death 1 receptor (PD-1) inhibition as monotherapy followed by Cytotoxic T-lymphocyte associated protein 4 (CTLA-4) inhibition in case of progression or as upfront double co-inhibition has drastically improved the survival outcomes of metastatic melanoma. Still, many patients develop primary or acquired resistance to both agents, relapse soon, and survive less. For these patients, the therapeutic options are very limited, and for many years, conventional chemotherapy (CC) was the standard of care. Recently, the phase II LEAP-004 trial supported that pembrolizumab/lenvatinib could potentially overcome anti-PD-1/anti-CTLA-4 immunotherapy refractoriness. In the absence of any prospective comparative study and to evaluate in a real-world context the clinical benefit of re-administering a PD-1 inhibitor (pembrolizumab 200 mg i.v. every 3 weeks, Q3W) with a multi-kinase inhibitor (lenvatinib, but at a reduced dose 10 mg p.o. daily due to its known toxicity) in this frail population of unmet need, we conducted here a retrospective comparison of LEAP-004-proposed combination with CC (carboplatin 4 AUC and dacarbazine 850 mg/m2 i.v. Q3W) in melanoma patients who relapsed to both checkpoint inhibitors, either in combinatorial or in sequential setting, between July 2022 and January 2024. Baseline demographics, disease characteristics, and treatment outcomes (objective response rate (ORR), progression-free survival (PFS), and overall survival (OS)) were recorded. Survival analyses were performed using the Kaplan-Meier method. All patients were also considered for safety analysis. A total of 84 patients were included in the effectiveness and safety analysis (pembrolizumab/lenvatinib, n=39 and CC, n=45). The median age was 67 (45-87) years and 64 (34-87) years, and men were 33.3% and 46.7%, respectively. The distribution of their metastatic sites was comparable, including 12.8% and 20% with brain involvement. Most patients had a good PS<2 (69.9% and 56.5%), increased lactate dehydrogenase (LDH) (71.8% and 84.4%), BRAF-wild status (82.1% and 84.8%), and received ≥2 previous systemic therapies (61.5% and 53.3%). The median follow-up was 18 months. The ORR was 23.1% and 11.1% (p<0.0001), the median PFS was 4.8 months and 3.8 months [HR (95%CI), 0.57 (0.36-0.92); p=0.017], and the median OS was 14.2 months and 7.8 months [HR (95%CI), 0.39 (0.22-0.69), p=0.0009] in pembrolizumab/lenvatinib and CC arms, respectively. Grade 3-5 treatment-related adverse events were documented in 48.7% (pembrolizumab/lenvatinib) and 75.6% (CC) of patients (p=0.034), which led to treatment discontinuation in 10.3% and 17.8% of cases, respectively. This is the first comparative study in patients with metastatic melanoma refractory to PD-1/CTLA-4 inhibition and showed significantly longer outcomes in cases treated with pembrolizumab/lenvatinib versus CC.

Accelerating large-scale DEA computation using sequential categorization and dynamic reference set selection

Data envelopment analysis (DEA) is a well-known data-enabled analytic tool for evaluating relative efficiency of units with multiple inputs and multiple outputs. The DEA computation increases substantially in the presence of large samples. In this study, we first recall two lemmas to distinguish efficient units using arithmetic operations without solving linear programming (LP). Using the used lemmas, the total sample of units is partitioned into several sequential blocks, where units in the preceding blocks are relatively efficient to those in the subsequent blocks. A novel reference set selection procedure is then formulated. We implement the proposed approach into one of the fastest existing methods and demonstrate a significant improvement in elapsed time. We conduct simulation experiments and illustrate the outcomes across varying dimensions, cardinality, and density.

Novel Picture of Topological Spaces in the Frame of Fuzzy Soft Sequential Sets

Novel Picture of Topological Spaces in the Frame of Fuzzy Soft Sequential Sets

Towards reconciling usability and usefulness of policy explanations for sequential decision-making systems.

Safefy-critical domains often employ autonomous agents which follow a sequential decision-making setup, whereby the agent follows a policy to dictate the appropriate action at each step. AI-practitioners often employ reinforcement learning algorithms to allow an agent to find the best policy. However, sequential systems often lack clear and immediate signs of wrong actions, with consequences visible only in hindsight, making it difficult to humans to understand system failure. In reinforcement learning, this is referred to as the credit assignment problem. To effectively collaborate with an autonomous system, particularly in a safety-critical setting, explanations should enable a user to better understand the policy of the agent and predict system behavior so that users are cognizant of potential failures and these failures can be diagnosed and mitigated. However, humans are diverse and have innate biases or preferences which may enhance or impair the utility of a policy explanation of a sequential agent. Therefore, in this paper, we designed and conducted human-subjects experiment to identify the factors which influence the perceived usability with the objective usefulness of policy explanations for reinforcement learning agents in a sequential setting. Our study had two factors: the modality of policy explanation shown to the user (Tree, Text, Modified Text, and Programs) and the "first impression" of the agent, i.e., whether the user saw the agent succeed or fail in the introductory calibration video. Our findings characterize a preference-performance tradeoff wherein participants perceived language-based policy explanations to be significantly more useable; however, participants were better able to objectively predict the agent's behavior when provided an explanation in the form of a decision tree. Our results demonstrate that user-specific factors, such as computer science experience (p 0.05), and situational factors, such as watching agent crash (p 0.05), can significantly impact the perception and usefulness of the explanation. This research provides key insights to alleviate prevalent issues regarding innapropriate compliance and reliance, which are exponentially more detrimental in safety-critical settings, providing a path forward for XAI developers for future work on policy-explanations.

Comparison of two bundles for reducing surgical site infection in colorectal surgery: multicentre cohort study.

There is controversy regarding the maximum number of elements that can be included in a surgical site infection prevention bundle. In addition, it is unclear whether a bundle of this type can be implemented at a multicentre level. A pragmatic, multicentre cohort study was designed to analyse surgical site infection rates in elective colorectal surgery after the sequential implementation of two preventive bundle protocols. Secondary outcomes were to determine compliance with individual measures and to establish their effectiveness, duration of stay, microbiology and 30-day mortality rate. A total of 32 205 patients were included. A 50% reduction in surgical site infection was achieved after the implementation of two sequential sets of bundles: from 18.16% in the Baseline group to 10.03% with Bundle-1 and 8.19% with Bundle-2. Bundle-2 reduced superficial-surgical site infection (OR 0.74 (95% c.i. 0.58 to 0.95); P = 0.018) and deep-surgical site infection (OR 0.66 (95% c.i. 0.46 to 0.93); P = 0.018) but not organ/space-surgical site infection (OR 0.88 (95% c.i. 0.74 to 1.06); P = 0.172). Compliance increased after the addition of four measures to Bundle-2. In the multivariable analysis, for organ/space-surgical site infection, laparoscopy, oral antibiotic prophylaxis and mechanical bowel preparation were protective factors in colonic procedures, while no protective factors were found in rectal surgery. Duration of stay fell significantly over time, from 7 in the Baseline group to 6 and 5 days for Bundle-1 and Bundle-2 respectively (P < 0.001). The mortality rate fell from 1.4% in the Baseline group to 0.59% and 0.6% for Bundle-1 and Bundle-2 respectively (P < 0.001). There was an increase in Gram-positive bacteria and yeast isolation, and reduction in Gram-negative bacteria and anaerobes in organ/space-surgical site infection. The addition of measures to create a final 10-measure protocol had a cumulative protective effect on reducing surgical site infection. However, organ/space-surgical site infection did not benefit from the addition. No protective measures were found for organ/space-surgical site infection in rectal surgery. Compliance with preventive measures increased from Bundle-1 to Bundle-2.

A Novel generalization of sequential decision-theoretic rough set model and its application

This paper introduces a refined and broadened version of decision-theoretic rough sets (DTRSs) named Generalized Sequential Decision-Theoretic Rough Set (GSeq-DTRS), which integrates the three-way decision (3WD) methodology. Operating through multiple levels, this iterative approach aims to comprehensively explore the boundary region. It introduces the concept of generalized granulation, departing from conventional equivalence-relation-based structures to incorporate similarity/tolerance relations. GSeq-DTRS addresses the limitations encountered by its predecessor, Seq-DTRS, particularly in managing sequential procedures and integrating new attributes. A notable advancement is its seamless handling of continuous-scale datasets through a defined Generalized Granular Structure (GGS), enabling classification across multiple levels without necessitating reduction of attributes. A refined version of conditional probability (CP), aligned with tolerance classes, enhances the approach, supported by a meticulously designed algorithm. Extensive experimental analysis conducted on two datasets sourced from https://www.kaggle.com demonstrates the efficacy of the procedure for both continuous and discrete datasets, effectively classifying probable elements into POS or NEG regions based on their membership. Unlike traditional Seq-DTRS, it does not require reduction of attributes at each new level. Additionally, the algorithm exhibits lower sensitivity to parametric values and yields results in fewer iterations. Thus, its potential impact on decision-making processes is readily apparent.

Sequential good lattice point sets for computer experiments

Sequential Latin hypercube designs have recently received great attention for computer experiments. Much of the work has been restricted to invariant spaces. The related systematic construction methods are inflexible while algorithmic methods are ineffective for large designs. For such designs in space contraction, systematic construction methods have not been investigated yet. This paper proposes a new method for constructing sequential Latin hypercube designs via good lattice point sets in a variety of experimental spaces. These designs are called sequential good lattice point sets. Moreover, we provide fast and efficient approaches for identifying the (nearly) optimal sequential good lattice point sets under a given criterion. Combining with the linear level permutation technique, we obtain a class of asymptotically optimal sequential Latin hypercube designs in invariant spaces where the $L_1$-distance in each stage is either optimal or asymptotically optimal. Numerical results demonstrate that the sequential good lattice point set has a better space-filling property than the existing sequential Latin hypercube designs in the invariant space. It is also shown that the sequential good lattice point sets have less computational complexity and more adaptability.

SPLITTING THE DIFFERENCE IN INCOMPLETE‐INFORMATION BARGAINING: THEORY AND WIDESPREAD EVIDENCE FROM THE FIELD

AbstractThis article documents a robust pattern from diverse sequential bargaining settings: agents favor offers that split the difference between the previous two offers. Our empirical settings include used cars, insurance claims, home sale, trade tariffs, a TV game show, eBay, and auto‐rickshaws. These even‐split offers are more likely to be accepted, less likely to spur exit by the opponent, and more likely to be followed by subsequent split‐the‐difference offers if bargaining continues. We propose several theoretical frameworks to explain this behavior, including an inference argument under which split‐the‐difference offers can be viewed as an equal split of the potential surplus.

A rank-based sequential test of independence

Summary We consider the problem of independence testing for two univariate random variables in a sequential setting. By leveraging recent developments on safe, anytime-valid inference, we propose a test with time-uniform Type-I error control and derive explicit bounds on the finite-sample performance of the test. We demonstrate the empirical performance of the procedure in comparison to existing sequential and nonsequential independence tests. Furthermore, since the proposed test is distribution-free under the null hypothesis, we empirically simulate the gap due to Ville’s inequality, the supermartingale analogue of Markov’s inequality, that is commonly applied to control Type-I error in anytime-valid inference, and apply this to construct a truncated sequential test.

Sequential multichannel allocation in cache-enabled multicasting small-cell networks

This paper investigates the optimization of multichannel allocation in cache-enabled multicasting small-cell networks (CMSNs). In the traditional fixed channel allocation scheme (FCA), small-cell base stations (SBSs) allocate a fix channel to each local cache popular content. In order to further increasing the utilization of allocated channels, we first propose a sequential multichannel allocation scheme (SMA), in which SBSs allocate a sequential channel set to their cached files dynamically. We formulate the problem of sequential multichannel allocation as two NP-hard non-linear programming problems with two optimization objectives of minimizing the global collision level (GCL) and maximizing the global file-downloading level (GFL), and demonstrate the advantages of the proposed DSCA over FCA. Then, two channel allocation potential games are proposed to achieve GCL and GFL optimization respectively. To achieve Nash equilibria (NEs) of the proposed games, we present a better response based SMA algorithm (BR-SMA) for both two games, and propose an improved better response based SMA algorithm (IBR-SMA) for the game for GCL maximization which have two exploration modes for each decision-maker.

Explainable Reinforcement Learning: A Survey and Comparative Review

Explainable reinforcement learning (XRL) is an emerging subfield of explainable machine learning that has attracted considerable attention in recent years. The goal of XRL is to elucidate the decision-making process of reinforcement learning (RL) agents in sequential decision-making settings. Equipped with this information, practitioners can better understand important questions about RL agents (especially those deployed in the real world), such as what the agents will do and why. Despite increased interest, there exists a gap in the literature for organizing the plethora of papers—especially in a way that centers the sequential decision-making nature of the problem. In this survey, we propose a novel taxonomy for organizing the XRL literature that prioritizes the RL setting. We propose three high-level categories: feature importance, learning process and Markov decision process, and policy-level. We overview techniques according to this taxonomy, highlighting challenges and opportunities for future work. We conclude by using these gaps to motivate and outline a roadmap for future work.

Generator Assisted Mixture of Experts for Feature Acquisition in Batch

Given a set of observations, feature acquisition is about finding the subset of unobserved features which would enhance accuracy. Such problems has been explored in a sequential setting in prior work. Here, the model receives feedback from every new feature acquireed and chooses to explore more features or to predict. However, sequential acquisition is not feasible in some settings where time is of essence. We consider the problem of feature acquisition in batch, where the subset of features to be queried in batch is chosen based on the currently observed features, and then acquired as a batch, followed by prediction. We solve this problem using several technical innovations. First, we use a feature generator to draw a subset of the synthetic features for some examples, which reduces the cost of oracle queries. Second, to make the feature acquisition problem tractable for the large heterogeneous observed features, we partition the data into buckets, by borrowing tools from locality sensitive hashing and then train a mixture of experts model. Third, we design a tractable lower bound of the original objective. We use a greedy algorithm combined with model training to solve the underlying problem. Experiments with four datasets shows that our approach outperforms these methods in terms of trade off between accuracy and feature acquisition cost.

Fair Participation via Sequential Policies

Leading approaches to algorithmic fairness and policy-induced distribution shift are often misaligned with long-term objectives in sequential settings. We aim to correct these shortcomings by ensuring that both the objective and fairness constraints account for policy-induced distribution shift. First, we motivate this problem using an example in which individuals subject to algorithmic predictions modulate their willingness to participate with the policy maker. Fairness in this example is measured by the variance of group participation rates. Next, we develop a method for solving the resulting constrained, non-linear optimization problem and prove that this method converges to a fair, locally optimal policy given first-order information. Finally, we experimentally validate our claims in a semi-synthetic setting.

Premonition-driven deep learning model for short-term ship violent roll motion prediction based on the hull attitude premonition mechanism

Ship transit forecasting can provide advanced movement information to a multidimensional wave motion compensation system. However, traditional deep learning models based on data-driven or physical data have difficulty in accurately predicting the rapid variations in ship motion attitudes that frequently occur in complex ocean environments. Therefore, a premonition-driven deep learning model is proposed in this paper to significantly improve the prediction accuracy of ship violent roll motion. This model applies precursor information to predict premonitory values of ship prospective motion attitudes based on an integrated statistical regression model and then employs these premonitory values to revise the outputs of the deep learning time series model. First, a hull attitude premonition mechanism is presented based on biological premonition cognitive behaviour, which captures the correlation relationships between extreme values by downgrading the globally continuous time series prediction problem to a time-independent regression problem. The corresponding methods for predicting local extreme values and deep feedforward neural network cluster are established based on these correlation relationships between the extreme values to achieve precise premonitory values. Second, the time and amplitude variables of the extreme angle values are transformed to countdown time variables and sequential sets of premonitory angle values. Two types of premonition units and their respective premonitory long short-term memory neural networks are constructed to upgrade the regression problem to a globally continuous time series prediction problem. Finally, a single-step output recurrent strategy is applied to forecast the short-term ship motion attitude for fifteen seconds by considering eleven initially predicted points across the two datasets. In sixty-six simulation experiments, the mean absolute error, root mean squared error, and mean absolute percentage error of the premonition-driven long short-term memory neural networks are considerably lower than those of traditional deep learning models. These simulation experiments verify that the premonition-driven deep learning model can effectively forecast short-term violent ship roll motion and provide safeguards for cargo transfer.