User:Snsphd/sandbox

An adaptive intervention is a set of well-operationalized guidelines for choosing effective treatments or interventions. The treatment or intervention choices are made adaptively over time, based on the changing characteristics of an individual. From the perspective of the individual, an adaptive intervention leads to a sequence of treatments or interventions. From the perspective of health-care providers, researchers, and other stakeholders, an adaptive intervention is a guide to the type of sequential treatment decision making that is typical of clinical or public health practice. More formally, an adaptive intervention (AI) is a sequence of decision rules that specify whether, how, when and based on which measures to alter the intensity, dosage, type or delivery of treatment(s) at critical points in the course of care.

Adaptive interventions are used in a variety of settings, ranging from medicine (e.g., to choose effective sequences of medications for patients), to schools (e.g., to choose effective sequences of educational and behavioral interventions for students). Adaptive interventions have also been used in organizational management or implementation science where the unit of intervention is a community-health organization, a hospital, or a school rather than an individual (e.g., to choose effective sequences of interventions to improve the uptake of evidence-based care in hospitals).

Examples of adaptive interventions include:

• An intervention for methamphetamine users that ‘steps down’ or decreases treatment dosage for individuals who indicate clinically significant reductions in methamphetamine use following initial treatment, but maintains treatment dosage for those with continued problematic methamphetamine use .
• An intervention to reduce conduct disorders in high-risk children that individualizes the number of home-based counseling sessions based on the level of parental functioning.
• A drug court that tailors frequency of court hearings and consequences of non-compliance to offender risk.

History and Terminology

Although clinicians frequently individualize the type and/or dose of treatment in the course of care, clinical practice has historically relied on clinical judgment for treatment decisions, or the type, dose and course of care for a given diagnosis. As clinicians have moved from the acute care to the long-term care model for treating chronic disease and disorders, however, concerns about the reliability and validity of clinical judgment have come to the fore. Research on adaptive interventions allows for systematic evaluation of clinical treatment sequences. In the statistical literature, adaptive interventions are commonly referred to as dynamic treatment regimes. , , Within the mental health and substance use literatures, adaptive interventions are commonly known as adaptive treatment strategies. The ‘stepped-care’ model, for example, is an adaptive intervention developed by substance use scholars to bridge evidence-based treatment and clinical practice with respect to substance abuse. Stepped-care interventions provide users initially with the minimal intervention (in terms of scope, intensity, burden and/or cost) considered clinically likely to be effective. Response to this first stage intervention is evaluated at a pre-determined follow-up. Non-responders may receive ‘stepped-up’ (or more intensive) treatment and/or responders may receive ‘stepped-down’ (or less intensive) treatment, including but not limited to a cessation of treatment. Response to this second-stage intervention is again evaluated at a pre-determined follow-up and treatment may be stepped up or down accordingly.

The Adaptive Intervention Perspective

Traditionally research on interventions aimed at treatment and prevention has focused on the effects of interventions with fixed composition and dosage. For example, all high school students are exposed to the same intervention related to the dangers of substance abuse; individuals looking to lose weight followed the same diet plan; smokers looking to quit smoking received the same amount of telephone counseling. Although traditional intervention research recognizes that individuals may have different intervention needs, and that different intervention components may be needed to meet these different needs, there is little recognition of the individual-level dilution or contravention of intervention effectiveness that may result when intervention components that are of little to no utility for a particular individual are combined with those components that are relevant. Adaptive intervention treatment design builds on the idea that individuals have varying needs for treatment or prevention that may not be best met by a single intervention that offers all participants the same components and dosage. Adaptive interventions assign different dosages and/or different components to different individuals, and/or within individuals at different time points. These individualized dosages—which may include dosages of zero for certain components—are assigned through decision rules that link individual characteristics with levels and types of program components. As such, the perspective of adaptive intervention mirrors the personalization elements of clinical practice; however adaptive interventions also enhance replicability by individualizing interventions through explicit decision rules.

Components of an Adaptive Intervention

An adaptive intervention consists of a package of four key components: (1) a sequence of critical decision points, (2) a set of intervention options, (3) tailoring variables, and (4) decision rules for each decision point. Note, however, that due to their different positions, clinicians and participants will have different perspectives as to what comprises the adaptive intervention:

• For the participant, an adaptive intervention is the particular sequence of treatments s/he experiences;
• For the clinician, an adaptive intervention is the sequence of decision rules recommending treatment pathways for individual participants.

1. A sequence of critical decision points: Decision points represent points in time at which important clinical decisions take place, for example: at program entry; every four weeks; at month X follow up.
2. A set of intervention options for each decision point: At each decision point, a set of possible intervention (or treatment) options must be decided. The set of intervention options could include options that relate to (1) the type or dose of treatment, for example, increasing, decreasing or discontinuing one or more pre-existing intervention components, or adding a new intervention component; (2) changes in intervention tactic, for example, stepping up treatment vs. continuing treatment; (3) the modality of the treatment, for example, face-to-face vs. web-based delivery of the same intervention; or (4) intervention timing—for example, offering a rescue intervention now or delaying until later.
3. Tailoring variables: Tailoring variables provide information from and/or about the individual that is useful for making treatment decisions, and specifically identifying which of a set of treatment options is best for a particular individual. Tailoring variables can be time-varying or baseline, and may both mediate the effect of prior treatment and moderate effects of subsequent treatment.
   1. Time-varying tailoring variables are measured in the course of treatment. Time-varying tailoring variables may provide information about the causal pathway of the intervention (e.g., mediators of intervention effects), or may serve as a surrogate indicator for a longer-term outcome (for example, cigarettes smoked in the prior week in a smoking cessation intervention, or weight loss to-date in a weight-loss intervention).
   2. Baseline tailoring variables are measured only once, prior to or at the beginning of an adaptive intervention. Information related to patient history or ascriptive characteristics could be used as baseline tailoring variables. A weight loss adaptive intervention, for example, might offer a different intervention option to individuals with a history of emotional or binge-eating than to those without.
4. Decision rule(s) for each critical decision point: Decision rules explicitly link the intervention options with one or more tailoring variable at specific decision points. Decision rules explicate the individualization in an adaptive intervention, and hence improve the ability of others to apply the same adaptive intervention in future studies or real-world applications, and achieve similar results (replicability).

There should be one decision rule per decision, and decision rules should specify treatment decisions for all study participants. This includes decisions that specify new or changing treatment for all study participants (e.g., the choice of first-stage treatment), as well as decisions that may modify treatment for only a small proportion or subset treatment participants. For example, adaptive interventions should include decision rules that address treatment changes for individuals who do not experience a sufficient response, even a sufficient response is expected of most participants. Adaptive interventions should also include decision rules that dictate ‘common contingencies’ of the AI, including treatment drop-out or treatment refusal.

Example: An adaptive intervention for minimally verbal children with autism spectrum disorder

Kasari et al. (2014) discuss developing an adaptive intervention for minimally verbal children with autism spectrum disorder (ASD). Prior research indicates that 25% to 30% of children with ASD remain minimally verbal even after years of interventions. Additionally, research indicates that children with ASD who fail to develop spoken language by five years of age have a much higher probability of poor long-term functioning. This high failure rate of past interventions and short window of time for efficacious intervention motivates the use of an adaptive strategy.

Intervention strategies

Kasari et al. (2014) were interested in augmenting well-established strategies for improving verbal skills amongst children with ASD with a promising, but untested and costly new strategy.

The study offered two interventions: first, a blend of two evidence-based interventions: Joint Attention Symbolic Play Engagement and Regulation (JASPER) and Enhanced Milieu Teaching (EMT); and second, treatment with a speech-generating device (SGD). Authors were interested in evaluating the efficacy of the SGD intervention within the context of JASPER+EMT, knowing that not all children would respond to JASPER+EMT.

Example adaptive intervention design

Figure 1 shows an example adaptive intervention design that could be used to improve verbal skills in children with ASD, using the JASPER+EMT treatment augmented with SGD for some participants.

This adaptive intervention has two stages:

• In Stage One, all children are treated with JASPER+EMT for 12 weeks.
• At the end of Stage One (or 12 weeks of treatment), children are evaluated for treatment response. Children who demonstrate 25% or greater change from baseline on at least half of the assessment variables are deemed ‘Early responders.’ Children who fall short of these benchmarks are deemed ‘Slow responders.’
• In Stage Two, children who were ‘early responders’ continue with JASPER+EMT for an additional twelve weeks. Children deemed ‘slow responders’ receive JASPER+EMT augmented with SGD (JASPER+EMT+SGD) for twelve weeks.

Figure 1: A decision rule under an adaptive intervention treatment model. At the end of week 12, a patient's treatment is or is not supplemented with SGD based on whether he or she is an "early" or "slow" responder

Decision rule specification

The adaptive intervention presented in Figure 1 corresponds to the following decision rule:

At end of week 12,

If responder status = early responder

Then, IO = [JASPER + EMT]

Else if responder status = slow responder

Then, IO = [JASPER + EMT + SGD]

In this decision rule:

• The decision point occurs at end of week 12.
• Two intervention options (IOs) are available for participants: JASPER + EMT or JASPER + EMT + SGD.
• Responder status (early responder or slow responder) is the tailoring variable used to define appropriate intervention option.

Types of Adaptive Interventions

Adaptive interventions can take on various forms, for example:

• Point-treatment/one-time matching adaptive interventions consist of only one decision point, typically at the beginning of the study, with decision rules deciding first-stage treatment. Only baseline tailoring variables are used. Point-treatment AI’s are also known as ‘individualized decision rules.’

Treatment decisions in personalized medicine, wherein medical treatment decisions are tailored to individuals’ genomic makeup, may be thought of as point-treatment AI’s.

• Sequential or dynamic adaptive interventions include multiple decision points, and decision rules that relate to each decision point. As such, sequential adaptive interventions can utilize information from tailoring variables that are time-varying—for example, using first-stage response surrogates to inform second-stage treatment.

Rationale for AIs

The sequence of treatment defined by an adaptive intervention may be necessary for several reasons, including the nature of the disorder; high patient heterogeneity in response to treatment; possible comorbidities; and considerations related to intervention adherence and/or the costs/burden of intervention or treatment options.

Nature of disorders

Disorders targeted by interventions are frequently long-term or chronic disorders, including substance use, mental health disorders, cancer, and diabetes. Chronic disorders follow courses of waxing and waning. Substance users may experience multiple periods of abstinence and abuse. An individual with depression may experience a recurrence of severe symptoms after several months with none. AIs can accommodate these fluctuations by building in decision points that can capture the temporal dynamics of the disorder and adapt the treatment accordingly. Relapse is particularly common for many disorders. In some forms of substance abuse disorders, patients are at high risk for relapse even after achieving an acute response to treatment. Adaptive interventions can allow treatment to be tailored both to reduce the chance for relapse , as well as to help the patient overcome relapse were it to occur.

Patient heterogeneity

The adaptive intervention approach is based on the knowledge that patients may differ in their responses to interventions, and that in order for an intervention to be most effective, the intervention may need to be adapted. Patients may respond differently to treatment. For example, four clinical trials of psychosocial interventions aimed at cocaine dependence found that between 35 and 60 percent of patients were still using cocaine at the end of the six-month intervention. In an adaptive intervention, subsequent intervention options—including increasing the intervention dose/intensity or switching to an alternative intervention—may be provided for individuals who failed to respond to the initial (first-stage) treatment. For any particular individual, the effectiveness of an intervention may also vary over time. In addition to waxing and waning of the disorder, dynamically-evolving notions of risk and resiliency may impede or enhance intervention effectiveness (Angst et al. 2009, Hser et al. 2007). It is important to determine if and when an intervention has stopped working, as well as what types of treatment should follow.

Co-morbidities

Interventions may target disorders that are co-morbid with others, for example, substance use and comorbid HIV infection; alcohol abuse and depression; or post-traumatic stress disorder and insomnia. Comorbid conditions require decisions about the order in which conditions should be treated, or whether the multiple disorders should be treated simultaneously. AIs also allow for treatment to be modified as certain disorders intensify or abate in relation to others.

Adherence

Adherence to interventions is difficult to maintain. Adherence to treatment of anti-psychotic medications amongst patients with schizophrenia ranges from 45 to 63 percent (Liu-Seifert et al. 2012). Low adherence or difficulties in maintaining adherence motivate the design of interventions that assess adherence and tailor treatment type or intensity to encourage adherence.

Intervention burden

Interventions, and especially certain intervention options, are costly to implement. Adaptive interventions allow for provision of the costliest elements only when and for as long as necessary. Stepped care interventions, for example, provide participants with the minimal intervention considered clinically likely to be effective, thus reserving the costliest/most intensive intervention options only for patients who fail to respond to less costly options. Participants in interventions also bear treatment burden, including potential treatment side effects. Adaptive interventions allow for reductions in treatment intensity whenever possible.