What a JITAI Actually Decides

A just-in-time adaptive intervention becomes understandable when we stop treating it as a smart notification and follow one delivery decision from input to consequence. This chapter focuses on the anatomy of that decision: what changes, what the system observes, which intervention options are available, and why the person’s ability to use an option matters as much as the clock. The result is a practical way to inspect a JITAI without accepting “adaptive” as a sufficient explanation.

The anatomy of one repeated decision

A JITAI operates at decision points: moments when its protocol requires a determination about whether and how to intervene. The device may reach such a point every few hours, every day, or even more often, depending on the behavior and the intervention option. The point is not that something must be delivered each time. The point is that the system has a rule for deciding.

That distinction sounds small, but it changes how the intervention should be described. “The app sends a reminder every evening” describes a schedule. “The app checks whether the person has brushed in the previous 12 hours and sends a prompt only when the condition is met” describes an adaptive decision. In the second case, the absence of a prompt is part of the design, not a missed opportunity to run the program.

A useful JITAI rule has three essential parts. The first is the intervention option: what the person could receive or do. It might be a short motivational message, a breathing exercise, feedback on a recent behavior, a suggestion to walk, a monetary incentive, or no intervention at all. “Support” is too broad to be useful here. The option has a particular content, intensity, duration, delivery modality, and set of tasks. Those details determine what the person must engage with.

The second part is the tailoring variable: the information used to decide among options. It must be able to change during the intervention and may describe an internal state, a behavior, or a context. Examples include a self-reported mood, a recent lapse, a location associated with risk, a movement pattern, or data from a sensor. A tailoring variable is not necessarily the cause of the behavior. It is an observed signal the protocol treats as relevant to the decision.

The third part is the threshold or condition that links the signal to the option. The condition may be a simple level—no brushing in the previous 12 hours—or a combination of several variables. It can be a yes-or-no rule, a range, or a weighted relationship among signals. Without this link, a system may collect rich data and still not be a JITAI. Measurement alone does not constitute adaptation.

Consider the oral-self-care example used in the JITAI framework. Twice a day, one hour before a person’s prespecified brushing window, information from an electric toothbrush is checked. If the person has not brushed in the previous 12 hours, the mobile intervention delivers a prompt encouraging brushing. If they have brushed, it does not deliver that prompt. The rule is simple enough to state in one sentence, but it contains the complete logic: a repeated decision point, a behavior-derived tailoring variable, a threshold condition, two intervention options, and a specified action for each branch.

The example also shows why a rule should be read as a protocol rather than as a promise. The toothbrush supplies evidence about one behavior. It does not automatically reveal whether the person is rushing to work, driving, in a meeting, exhausted, annoyed by previous prompts, or ready to act. The rule answers one question—whether a prompt should be considered based on recent brushing—not every question relevant to whether that prompt will help.

A JITAI can use more than one option. A system might choose between a message emphasizing the benefits of a behavior, a message emphasizing the consequences of not doing it, and no message. It might vary intensity or modality rather than content. The choice among options should be tied to the mechanism the intervention is meant to affect. A reminder intended to address forgetfulness is a different option from a feedback summary intended to promote reflection, even if both arrive through the same phone.

The speed of the decision is another defining feature. A JITAI’s adaptation happens on a relatively fast timescale, such as days, hours, minutes, or seconds. This is different from a program that decides after several weeks whether to add coaching or change treatment intensity. A slower adaptive intervention can include a JITAI as one of its components, but the two operate at different rhythms. Keeping the timescale visible prevents a general idea of “personalization” from swallowing the specific design being studied.

The decision rule is normally the same protocol across people and across decision points in a standard JITAI, even though its inputs and selected options may vary. One person may meet the condition for a prompt at one decision point and not at another. Another person may show a different pattern. The rule remains fixed while the state-dependent decision changes. Keeping that distinction visible prevents the basic delivery rule from being mistaken for a complete account of how an intervention learns.

sequenceDiagram participant C as Person and context participant S as Sensor or assessment participant D as JITAI decision rule participant P as Person participant O as Proximal and distal outcomes C->>S: Changing state or context becomes observable S->>D: Provide tailoring variable at a decision point D->>D: Compare the signal with the protocol condition alt Condition supports delivery D->>P: Deliver selected intervention option P->>O: Engage with the option and attempt the behavior else Condition does not support delivery D-->>P: Withhold this option P-->>O: Continue without this delivery end O-->>O: Proximal response contributes to longer-term outcome

The sequence makes control ownership explicit. The person and context generate changing conditions; a sensor or assessment turns some part of those conditions into data; the decision rule determines the branch; and the person—not the device—must engage with the option for the intended mechanism to operate. The outcome is not simply “message delivered.” It is the next meaningful change the intervention was designed to influence.

This is also why a decision point should not be confused with a trigger in the everyday sense. A trigger is often understood as an event that directly causes an action. In a JITAI, the observed event or signal initiates a protocol decision, but the signal may be only a proxy for vulnerability or opportunity. The rule may then select an option, withhold it, or require further conditions. The causal work still has to occur through the person’s engagement and the mechanism targeted by the option.

From a delivery rule to a causal pathway

The most useful way to test a JITAI is to trace its intended causal pathway. Start with the distal outcome, the longer-term result that matters in practice. For an oral-self-care intervention, the framework gives an illustrative distal goal of reaching a specified level of adherence to recommended brushing over a defined period. In a mental-health intervention, the distal outcome might be lower symptoms or better functioning over time. The distal outcome gives the intervention a destination, but it is usually too slow and broad to explain what one prompt can change.

The next link is a proximal outcome: a shorter-term behavior or state that can respond within the timescale of the decision rule. In the brushing example, brushing during the next 12 hours is a proximal outcome. It is close enough to the prompt to be observed and tested, while repeated proximal successes may contribute to the longer-term adherence goal.

Between the proximal and distal outcomes are mechanisms. A mechanism is the process through which an intervention is expected to produce change. Forgetfulness may directly affect whether someone brushes. Understanding the relationship between oral and general health may first affect motivation and then indirectly affect brushing. A prompt, feedback message, or educational component should be selected because it can plausibly influence one of these pathways—not because it is easy to send.

This structure prevents a common error: treating the tailoring variable as if it were the mechanism or the outcome. “No brushing in the last 12 hours” can be the signal that activates a prompt, but it is not by itself proof that forgetfulness caused the omission. The prompt may influence attention or motivation, which may influence brushing, which may contribute to longer-term adherence. Each link is a claim that can be tested, and a failure at one link can make the final outcome unchanged even when the device followed its rule perfectly.

The framework distinguishes two kinds of conditions that a good rule must consider. A state of vulnerability represents a near-term risk for an adverse outcome. Stress can be a vulnerability state for a smoking lapse in the following hours. Craving can signal risk of later binge eating. Proximity to a frequently visited bar or liquor store can signal relapse risk for someone recovering from alcohol dependence. These signals matter because the intervention is trying to alter what happens next, not merely describe the person’s current state.

A state of opportunity is different but related. It is a condition under which delivering support is expected to promote a desirable proximal outcome. The period immediately after someone completes a self-monitoring task may be an opportunity to reinforce that behavior. Reporting a difficult emotion may be an opportunity to offer a strategy for regulating it. The same condition can be both vulnerability and opportunity. Not reaching a fluid-intake threshold in the previous hour can signal risk of falling short of a health goal and create an opportunity for a reminder that supports repetition.

Identifying vulnerability or opportunity is still not enough. A JITAI must consider receptivity, the conditions in which a person is likely to engage effectively with a specific option. Receptivity is not a general trait such as “this person is motivated.” It belongs to the relationship between a person, a situation, and an option. Someone may be receptive to a one-sentence reminder but not to a longer exercise. They may be willing to receive feedback but not to complete a task requiring sustained attention. They may be able to listen to a brief audio instruction but not look at a screen while driving.

That option-specific quality matters because an intervention option is usually more than a stimulus. A prompt encouraging a person to notice work-related emotions and take a short walk includes the message itself, the task of paying attention to emotion, and the task of walking. Receptivity therefore includes the ability to notice and process the prompt, understand what it requests, and perform the associated activity. Being in a meeting may make the walk impossible; being preoccupied may make emotional processing unlikely; being under acute stress may reduce the ability to attend to a message. The state that makes support urgent can also make the support difficult to use.

A rule that includes vulnerability but ignores receptivity may be accurate in a narrow sense and ineffective in practice. It can identify the person at the right risk level yet select the wrong moment for delivery. This is not an argument for waiting until the person is comfortable. It is an argument for asking whether a different option, a simpler format, or a different decision point can reach the relevant mechanism without adding avoidable burden.

Burden is part of the causal pathway, not an administrative footnote. Every prompt asks for some investment of physical, cognitive, or emotional energy. Repeated requests can produce mental fatigue or habituation, in which a person’s response diminishes after repeated stimulation. If a rule sends an option whenever a signal is present, it may convert a useful cue into an interruption. If it withholds every time the person appears busy, it may miss the opportunity it was designed to address. The decision rule is balancing the expected benefit of delivery against the cost and possible harm of asking for engagement.

This is the logic behind the JITAI rationale of resource efficiency. Avoiding unnecessary delivery can reduce effort, fatigue, and habituation. But “fewer interventions” is not the same as “more efficient intervention.” Efficiency means doing more good with fewer resources. A system that minimizes notifications but does not affect the intended proximal outcome has not demonstrated efficiency; it has only reduced exposure. Conversely, a system that sends many successful prompts may still impose too much burden to be sustainable. Both the benefit pathway and the cost pathway need to be considered.

A complete rule can therefore be written as a chain of questions rather than a technical slogan. What changing condition is relevant to the target mechanism? How well does the tailoring variable represent that condition? Which option can influence the mechanism? At what decision point can the option matter? What level or combination of signals selects it? Can the person engage with that option now? What would count as a proximal response, and how could repeated responses contribute to the distal outcome?

The answers do not need to be perfect before a JITAI is tested. They do need to be explicit enough to expose uncertainty. If the system cannot measure the relevant state directly, it should be clear that the signal is a proxy. If the option requires a task that cannot be observed easily, engagement may remain partly unknown. If the distal goal is ambitious, a change in a proximal behavior should not be presented as proof that the long-term goal has been achieved.

This model also clarifies what a JITAI can and cannot promise. It can make delivery more responsive to specified conditions. It cannot guarantee that the conditions were measured correctly, that the selected option is persuasive, or that a short-term response will persist. It can make the intervention protocol more transparent by stating its branches. It cannot remove the complexity of a person’s daily life or turn a correlational signal into a complete causal explanation.

The practical value of the design lies in making these choices visible. A reader evaluating a JITAI should be able to ask what the system decides, what evidence feeds that decision, and which outcome is supposed to change next. If the answer stops at “the app knows when to send a message,” the mechanism remains unfinished. The decisive question is whether the rule connects a meaningful state to an appropriate option under conditions in which the person can use it. Only then does a fast decision become a plausible route to just-right support.

References