Overview

This chapter turns the JITAI model into a fit test across oral self-care, smoking-risk location, and fluid-intake examples. It asks what state is observed, which option targets what mechanism, whether the person can use it, and what proximal or distal outcome should change. The audit separates observation from inference and engagement, adherence, retention, and outcome. It also matches evidence claims to estimand, comparator, adaptation level, outcome importance, and missing-data limits.

Learning outcomes

A signal can support an observation about behavior, location, or self-report without proving cause, intent, or receptivity.
A just-right audit tests the benefit pathway and the cost-and-uncertainty pathway, including burden, missing data, and possible harm.
Effectiveness claims depend on the estimand, comparator, adaptation actually delivered, outcome, time horizon, and missing-data limits.

Keywords

JITAIsFitMechanismsEstimandsComparatorsOutcomes

From Timely Delivery to Timely Fit

A JITAI is ready for judgment only when its changing state, selected option, human usability, and intended outcome can be examined together. This checkpoint connects those elements through three familiar health-behavior situations and turns them into a practical fit test: what is directly observed, what is inferred, and what remains unknown. The aim is not to repeat definitions, but to see how the same reasoning changes when the signal, the option, or the person’s circumstances change.

One model, three different decisions

Start with oral self-care. An electric toothbrush can report whether brushing occurred in the previous 12 hours, and a rule can check that information before a person’s morning or evening brushing window. If no brushing is detected, the intervention may deliver an encouraging prompt; if brushing is detected, it may withhold that prompt. The observed fact is narrow: the sensor recorded a behavior, or did not record it, during a specified period. The rule then makes a delivery decision based on that fact.

Several further claims are embedded in the example. The absence of a recorded brushing episode may be treated as a vulnerability for another missed episode. A reminder may be expected to affect forgetfulness or motivation, which are mechanisms closer to the behavior than the long-term adherence goal. The person must notice and use the prompt. Repeated brushing over time is the distal concern. Only the first part is directly observed at the decision point. The rest belongs to a causal model that needs evidence.

Now change the scenario. A person trying to stop smoking approaches a location associated with prior use. Location can be a useful tailoring variable because it marks a possible high-risk context. But the same location may represent different situations: a brief passage, a workplace, a meeting with supportive people, or an actual opportunity to smoke. The signal indicates exposure to a context that may matter; it does not establish craving, intent, or receptivity. A warning message may be appropriate for one person and intrusive or unusable for another.

The third scenario involves fluid intake. Not reaching a prespecified intake level in the previous hour can be interpreted both as vulnerability—the person may be falling short of a health behavior—and as an opportunity to reinforce a repeated drinking habit. A reminder may be reasonable if it addresses the mechanism the intervention is meant to change. But the same condition does not tell us whether the person can drink now, whether a reminder is the option they can use, or whether a short-term response will persist without continued prompts.

The scenarios share a structure but not a conclusion. In each, a changing condition is used to decide whether an intervention option should be delivered. In each, the condition can carry more than one meaning. In each, the rule can be technically faithful while the intervention remains poorly fitted to the moment. That is why a JITAI cannot be audited by asking only whether it used real-time data.

The first question is what state is being detected? Is the system observing a behavior, inferring a context, or asking the person to describe an internal experience? The answer determines what can be claimed. A toothbrush signal can support a statement about detected brushing. A location signal can support a statement about where the device was. Neither alone proves why the person acted or what they can do next.

The second question is which option is selected, and what mechanism is it meant to affect? A prompt can address forgetfulness, a feedback summary can promote reflection, and a skill-oriented message can aim at a different short-term process. If the option has no plausible path to the proximal outcome, precision in timing cannot compensate.

Published by DeepMox

FAQ

What is the JITAI fit test?

The JITAI fit test asks whether a meaningful signal supports the decision, the selected option matches the target mechanism, the person can use it at acceptable burden, and the study measures an outcome that fits the claim.

Why must observation and inference be separated in a JITAI?

Observation and inference must be separated because a device recording no brushing, a location, or a self-report does not by itself prove the person’s cause, intent, or ability to respond. Keeping the levels distinct prevents a proxy from becoming an unsupported explanation.

Why does the comparator matter when evaluating a JITAI?

The comparator matters because improvement over a waitlist, usual care, no treatment, or another intervention answers a different practical question. A result against no support does not establish superiority to a credible alternative or isolate the adaptive rule’s contribution.

Previous13 minJITAIs: Where the Chain Breaks

Next13 minJITAIs: What the Evidence Shows

JITAIs: The Timely-Fit Test

From Timely Delivery to Timely Fit

One model, three different decisions

FAQ