Minimum effective dose of a multicomponent behaviour change intervention to increase the physical activity of individuals on primary statin therapy: an adaptive study using the time-to-event continual reassessment method (TiTE-CRM)

Abstract

Objectives

To identify the minimum effective dose of a multi-behaviour change technique (BCT) intervention to increase physical activity among individuals on primary statin therapy using the time-to-event continual reassessment method (TiTE-CRM).

Setting

A large New York metropolitan area healthcare system comprising approximately 85 000 employees and 5.5 million patient encounters annually.

Participants

42 participants enrolled in 13 cohorts of 3 participants, 1 cohort of 2 participants and 1 cohort of 1 participant. The sample was composed of 16.7% individuals aged 66 and older (n=7), 64.3% women (n=27), 69.0% white individuals (n=29) and 7.1% Hispanic individuals (n=3).

Interventions

A variable-duration, four-BCT text message intervention and a 2-week follow-up. Dose assignment relied on TiTE-CRM to adjust the duration of the intervention based on adherence of participants in prior cohorts. Five mechanisms of action (MoAs) were assessed: self-efficacy, intrinsic regulation, discrepancy in behaviour, motivation and barriers to activity.

Primary and secondary outcome measures

The primary outcome measure was the proportion of participants who achieved a 2000 step/day increase between baseline and follow-up. The secondary outcomes were within-participant changes in daily steps (examined as a continuous variable at the daily level) and potential MoAs for increased physical activity.

Results

Of the 40 participants who completed follow-up, 7 (17.5%) achieved the goal of 2000 or more steps per day during their follow-up period. Though participants did increase the number of steps they walked during the intervention (B(SE)=373.1 (154.7) steps; p=0.016), there was no association between increased intervention duration and increased daily average steps. The intervention was also associated with increases in self-efficacy (p=0.002), intrinsic regulation (p=0.037), discrepancy in behaviour (p<0.001) and motivation (p=0.039).

Conclusions

The results of this trial did not show a traditional dose-response curve to increasing the length of a multicomponent BCT intervention. Results did show that the intervention successfully increased steps during the intervention period and that the benefit of the intervention dwindled during follow-up. Further, potential MoAs for the intervention were confirmed.

Trial registration number

NCT05273723.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• The current study uses a state-of-the-art dose finding methodology in time-to-event continual reassessment method and is one of the first trials to apply this method for a behavioural intervention.

• The study uses behavioural theory to design the intervention and measure potential mechanisms of behaviour change.

• The study has a small sample size of 42 participants.

Introduction

Cardiovascular diseases (CVDs) are among the most prevalent causes of mortality and morbidity in the USA.¹,³ Insufficient physical activity (PA) has been recognised as a leading contributor to poor cardiovascular health, as those who engage in regular PA show lower levels of CVD risk factors.⁴ In fact, even if the minimum WHO recommendations for PA are not met, people can still improve their cardiovascular health and reduce risk of all-cause mortality by becoming more active.⁴ An increase of even 500–1000 steps per day for sedentary individuals can reduce the risk of chronic disease and all-cause mortality.⁵,⁷

Dyslipidaemia, characterised by heightened plasma cholesterol, increased triglyceride levels and/or reduced high-density lipoprotein cholesterol levels, is a major risk factor for CVD.⁸ HMG-CoA reductase inhibitors (commonly known as statins) reduce the effects of dyslipidaemia and assist in CVD prevention and treatment. Statin therapy is extremely effective for reducing CVD risk, but the addition of other interventions to improve cardiovascular health (such as increasing PA) has been shown to produce benefits beyond statin therapy alone.⁹,¹³

While statin therapy has been shown to be more effective for lowering blood triglyceride levels compared with PA among those already prescribed statin medication,¹⁴ a combination of statin medication and PA lowers CVD mortality risk more than either therapy alone.^{9 10} Combining statin therapy and PA also improves cognitive functioning and reduces statin therapy side effects (including myalgia, myopathy and new onset diabetes).¹⁵,¹⁷ Side effects experienced by patients taking statins often cause them to discontinue statin therapy.¹⁸ However, PA may reduce the likelihood of medication non-adherence by addressing potential side effects. The implication is that increasing PA among individuals who take statins may independently reduce their risk of CVD while enhancing the effectiveness of their statin therapy.

One approach for successfully increasing PA is to use a multicomponent behavioural intervention,¹⁹,²³ such as those that incorporate behaviour change techniques (BCTs). BCTs are observable, replicable, irreducible intervention components that are thought to be the ‘active ingredients’ of behaviour change interventions.²⁴ The duration of these interventions and frequency of BCT delivery varies significantly between studies,¹⁹ meaning the intervention duration needed to meaningfully increase PA is not known. As intervention length has important implications for intervention cost and effectiveness, identifying the appropriate duration for a behavioural intervention to produce meaningful results is particularly important. Dose-finding methods such as those used in pharmaceutical studies have the potential to identify an effective minimal duration of a behavioural intervention.^{25 26} Applying these methods to multi-BCT interventions for PA would allow behavioural researchers to determine the duration of the intervention (or intervention dose) required to elicit a meaningful change in PA.

In the current trial, we adopted a state-of-the-art dose-finding methodology to determine the minimum effective dose (MED) of a multicomponent BCT intervention needed to increase PA among participants on primary prevention statin therapy. BCTs included in the current intervention—namely goal setting, action planning, self-monitoring of behaviour and prompts/cues—have all been shown to improve health behaviours, including PA.²⁷,²⁹ The study employed a modified version of the time-to-event continual reassessment method (TiTE-CRM)³⁰ to identify the MED of the BCT intervention. Although the TiTE-CRM model has traditionally been used to determine a maximum tolerated dose in pharmacologic agents, it can also be applied to determine a minimum dose. The MED is defined as the smallest dosage of a particular drug or intervention that will elicit a clinically discernable response.³¹

The primary aim was to identify the MED of a multi-BCT intervention (ie, duration (in weeks) of intervention) required to increase PA by an average of 2000 steps per day between run-in and follow-up periods in at least 80% of participants on primary prevention statin therapy at elevated risk for CVD. The secondary aim was to identify effect sizes for mediating mechanisms of action (MoAs) of the multi-BCT intervention on PA.

Methods

Study design

This dose-finding adaptive trial aimed to identify the MED of a PA intervention among individuals on primary prevention statin medication, defined as the intervention duration that increased walking by >2000 steps per day between run-in and follow-up periods with 80% probability. The goal was to enrol 42 participants in 14 cohorts of 3 participants each (figure 1). However, we enrolled 15 cohorts: 13 cohorts of 3 participants, 1 cohort of 2 participants and 1 cohort of 1 participant. This enrolment scheme was used to correct an implementation error in which one participant was incorrectly assigned to the fourth recruited cohort and withdrawn by the study team due to ineligibility, resulting in a cohort of two participants. A final participant was assigned to a cohort of one to reach a sample size of 42 (online supplemental table 1). Based on our prior experience with TiTE-CRM, small variations in cohort sizes have minimal impact on the accuracy and statistical power of the analysis, minimising effects on our study power. Notably, while the TiTE-CRM was originally designed to update dose assignment after every single participant (ie, cohort size of 1), we designed to update doses in cohorts of 3 in this study because of logistical considerations. As a model-based method, the TiTE-CRM uses all available data to make the best prediction for subsequent dose assignments, thus offering the flexibility to accommodate different dose assignment scenarios.

Figure 1. Participant enrolment diagram.

All participants underwent a 2-week run-in period, a variable intervention duration (‘dose’), and a 2-week follow-up period. The first cohort of three participants received a 5-week multicomponent BCT intervention delivered via text messages. Subsequent cohorts received varying BCT intervention durations (from 5 weeks to 10 weeks) where the duration was determined by the MED estimate according to the modified TiTE-CRM model and data from previously enrolled cohorts in the study. Though durations from 1 to 10 weeks in length were possible, the TiTE-CRM estimates did not identify a duration shorter than 5 weeks. Online supplemental table 1 shows the dose assignment for all 15 enrolled cohorts. After completion of the intervention, participants continued for a 2-week follow-up period during which they did not receive the BCT intervention but continued to have their data collected continuously. It should be noted that the objective of the study was not to compare across doses (for which having the same observation length would have been critical) but rather to see if BCTs would have a short-term effect on PA (ie, step counts). The MED determines how long the intervention period is for each cohort, but the run-in and follow-up periods were the same length for all participants. The trial can be considered as an adaptive design trial that used participant data for intervention duration assignment but did not randomise treatment assignment.

Participants were provided with a commercially available Fitbit Charge 5 device to measure PA levels. All enrolled participants received a Fitbit study account (with no identifying information about the participant) with a unique identifier created by the research team.

Measures of PA were collected by the Fitbit continuously throughout the run-in, intervention and follow-up periods. In addition to PA, all participants provided demographic data (eg, age, sex) and assessments of MoAs by which the BCT intervention may have influenced PA (ie, walking) behaviour: ‘beliefs about capabilities/self-efficacy’, ‘behavioural regulation/intrinsic regulation’, ‘feedback processes/discrepancy in behaviour’, ‘motivation’ and ‘barriers to physical activity’. These MoAs have been previously identified as important mechanisms of the BCT interventions used in this trial and/or have been linked with PA in prior research.³²,³⁶ The study protocol was approved by the Northwell Institutional Review Board (IRB) and has been published elsewhere.³⁷ Study recruitment began in March 2022, and study completion occurred in July 2023.

Study population and recruitment

We recruited individuals from within Northwell, a large healthcare system in the New York metropolitan area; it is composed of approximately 85 000 employees and 5.5 million patient encounters annually. Potential participants included both Northwell Health employees and patients who were currently 18 years of age or older, prescribed a statin, ambulatory without limitations and had access to and were capable of using a smart phone. Exclusion criteria were history of CVD, inability to speak English, cognitive impairment, severe mental illness, pregnancy and unavailability for follow-up. Participants were primarily recruited via advertising and postings across established Northwell communication channels, including email, employee engagement platforms, Northwell community groups and flyers distributed at primary care and cardiology offices. In addition, potentially eligible participants were recruited using the Northwell electronic health record. We also conducted outreach through social media (including Facebook, Reddit, Craigslist and other social media outlets) in forums associated with statins, cardiovascular health and overall health. Recruitment links directed interested individuals to an online screening survey in REDCap (a secure, web-based electronic data capture tool)^{38 39} containing questions regarding study inclusion and exclusion criteria and provided study contact information. If the participant was deemed eligible, they were automatically directed to complete detailed demographic questions, view an educational study video and review the electronic consent form. A four-question assessment measured participant understanding of the protocol and consent process. Consent was electronically obtained,⁴⁰ and a copy of the consent was emailed to the participant for their personal records. Prior to being provided study devices for the run-in period, participants were asked to electronically sign a device contract outlining use and return of the devices.

Run-in period

The first 2 weeks constituted a baseline assessment period (run-in). Participants were mailed a Fitbit and an eCAP bottle after completing a brief onboarding survey, where they provided basic demographic information and indicated their preferred time of day for walking. Participants were encouraged to wear their Fitbit day and night, sync it with the Fitbit application on their phone at least every 2 days, and charge it at least every 4 days. On receiving the eCAP bottle, participants were asked to fill it using their existing statin prescription and continue to take their statin as prescribed.

Assignment of interventions

Participants found to be adherent to the study protocol during the run-in received the four-BCT intervention, the duration of which varied between 5 and 10 weeks depending on the assigned dose. Assignment to doses used a modified TiTE-CRM methodology for each cohort that was based on the results from the previous cohort.⁴¹ The dose was adjusted by the duration that increased walking by at least 2000 steps per day with 80% probability. The first cohort received 5 weeks of the multi-BCT intervention and subsequent cohorts were assigned to a dose level calculated according to the TiTE-CRM model using all available observations from prior participants enrolled to the study. Interim calculations were performed with a web app developed by the study statistician using methods from a previous trial.⁴² Briefly, the interim calculations would use the most recent 2-week average daily steps as a proxy for the final average steps of all currently enrolled participants at the time of the calculations, as well as their baseline average, to estimate the dose-response curve. As the trial progressed, the interim outcomes used in the calculations would be updated; and at the end of the trial, the dose-response curve would be estimated using the primary outcomes of all participants during the follow-up period. The general methodology of the TiTE-CRM has been documented in prior publications.³¹ Specifically, for the present study, the dose-response model was calibrated such that the TiTE-CRM would eventually select a dose that improves daily step counts with 75%–85% probability—that is, within 5 percentage points of our target MED.⁴¹ To ensure sufficient information accrued prior to dose calculation, we also imposed a 1-week waiting window between cohorts (ie, we did not enrol participants to a new cohort unless all the participants in the prior cohort had completed at least 1 week of the intervention period).

Interventions

During the BCT intervention, participants received daily text messages containing all four of the BCTs: goal setting, action planning, self-monitoring of behaviour and prompts/cues.²⁷,²⁹ Timing of text messages was based on the participants’ preferred time for walking specified in survey data collected during trial onboarding. Generally, the morning message addressed prompt/cues, goal setting and action planning, and the evening message addressed self-monitoring. Goal setting entails setting a goal defined in terms of the behaviour to be achieved. For this study, individuals received a personalised text message based on their average step count during the run-in. Action planning involves forming a plan about where, for how long and at what time PA is going to be performed. This BCT was intended to encourage the participant to decide to act or make a behavioural resolution by forming detailed plans that link the behaviour to specific situational cues. Self-monitoring of behaviour refers to monitoring and recording behaviour. Lastly, prompts/cues were used to cue the behaviour (ie, PA). For this study, participants were sent a text at a scheduled time reminding them to walk a set number of steps. The text message content for each BCT can be found in online supplemental table 2.

As part of the planned intervention, participants were also supposed to receive another daily BCT: feedback on behaviour. Unfortunately, due to an error in the automation of intervention delivery, participants were not provided with evaluative feedback on their daily step count related to their goal. Due to this automation error, no participants enrolled in the trial received the feedback on behaviour, which was therefore not a part of the multi-BCT intervention package.

During the intervention period, participants completed one questionnaire assessing potential MoAs every 2 weeks, with the total number completed varying according to the duration of their multi-BCT intervention. For example, a participant with a total intervention duration of 6 weeks would have completed this questionnaire three times.

Patient and public involvement

Pilot data with participants was used to help determine which interventions were selected for the current trial. We did not directly involve the public or participants in any other elements of the design or conduct, or reporting, or dissemination plans of our research.

Outcomes and analysis

Sample size calculation

The sample size of 42 participants was identified to allow for preliminary assessment of the MED for the multi-BCT intervention to increase walking between baseline and follow-up periods. The dose-efficacy model is calibrated such that the modified TiTE-CRM will eventually select a BCT duration associated with 75%–85% a successful increase in PA (defined as an increase of 2000 or more per day between baseline and follow-up).⁴³ The sample size of 42 participants is determined to achieve 60% probability of correct selection under logistic dose-efficacy curves with an OR of 2, conforming with prior estimates in the literature.⁴⁴

Primary outcome

The primary outcome was a successful increase in walking between the run-in and follow-up periods as measured by a Fitbit device. A successful increase in walking was defined as an average increase of 2000 steps of walking per day between run-in and follow-up periods, determined by calculating the mean daily step totals for the run-in and follow-up periods. Steps were measured continuously and aggregated by day. The MED was identified as the dose of the intervention (in weeks), which led to a successful increase in at least 80% of participants who received it. Participant step counts over time are visualised in figure 2. For follow-up comparisons, all cohorts were compared during their respective follow-up period after dose initiation, but as intervention period length varied according to the study design, the point in time for participants varied. In other words, the 2-week follow-up period immediately followed all cohorts regardless of intervention dose. The shortest dose for any cohort was 5 weeks (cohort 1); this cohort had a total of 9 weeks of monitoring, whereas a cohort with a 6-week dose (cohort 2 only) had a total of 10 weeks of monitoring, and so on for longer doses. The intervention length is treated as the varying ‘dose’ across cohorts, but the run-in and follow-up windows remain the same for the purposes of comparison.

Figure 2. Visualisations of participant step counts over time. Participant daily step counts at baseline seemed to be strongly associated with step counts during follow-up (upper left panel). In addition, baseline step counts did not appear to be strongly associated with change in steps between baseline and follow-up (upper right panel). Examining the association between intervention duration (ie, intervention dose) and change in steps between baseline and follow-up did not appear to show a dose-response increase (lower left and lower right panels; green indicates baseline and follow-up and red indicates intervention).

Secondary outcomes

In addition to this primary outcome, we conducted analyses to identify whether the intervention had an influence on within-participant changes in daily steps (examined as a continuous variable at the daily level) and potential MoAs for PA (including self-efficacy, intrinsic regulation, discrepancy in behaviour, motivation and barriers to PA).

To examine changes in step counts, we used the Fitbit to identify day-level averages of step counts and then examined changes in these averages over time. This provides a more granular examination of how the intervention may have influenced step count rather than the binary primary outcome analysis. The effect of treatment on steps was assessed using generalised linear mixed models (GLMM). We specified fixed effects for the intervention, time and interaction of the intervention and time. A random effect was specified for participant. We used an autoregressive model (AR (1)) to account for possible autocorrelation and linear trends between daily steps across time.

All MoAs were assessed at the completion of the run-in and every 2 weeks until the end of the follow-up period via questionnaires sent by text message. Self-efficacy was operationalised with the Self-Efficacy for Walking scale,⁴⁵ a 10-item measure assessing a participant’s beliefs about their physical capability to walk for durations of 5–50 min. Intrinsic regulation was assessed using a four-item measure assessing intrinsic regulation, a subscale of the Behavioural Regulations in Exercise Questionnaire Version 2.³⁵ Within-person change in discrepancy in behaviour was queried with a single item adapted from one examining discrepancy in behaviour for alcohol use.⁴⁶ The text of the measure was ‘How large is the difference between your current walking behaviour and your goal concerning your walking?’ Within-person change in motivation to walk was assessed with an item stating, ‘I feel motivated to walk each day’. Participants rated this item on a scale of 1 (‘not true at all’) to 7 (‘very true’), with higher scores indicating higher levels of motivation to walk. Within-person changes in barriers to PA were assessed with a checklist of seven potential barriers to walking. This list was adapted from a prior study identifying barriers for walking.³⁶ Barriers were coded on a 1 (‘not often at all’) to 5 (‘very often’) scale and averaged to create a total score, with higher scores indicating that the listed barriers had greater effects on walking. To analyse the effect of the intervention on each MoA, we conducted GLMM models with fixed effects for the intervention, time and the intervention by time interaction with a random effect for participant. We then examined the effect of intervention and each MoA on daily steps. As with the analyses of daily steps, autoregressive models (AR (1)) were used to account for serial autocorrelation in MoAs and steps over time.

Results

Demographics

There were 42 participants enrolled in 13 cohorts of 3 participants, 1 cohort of 2 participants and 1 cohort of 1 participant. The sample comprised 16.7% individuals aged 66 and older (n=7), 64.3% women (n=27), 69.0% white individuals (n=29) and 7.1% Hispanic individuals (n=3). Full demographics can be found in table 1. Participants in cohort 1 received a 5-week dose of the multi-BCT intervention (online supplemental table 1). The dose for cohort 2 was increased to 6 weeks. Cohorts 3 and 4 had 8-week doses while cohort 5 had a 9-week dose. For the remainder of the trial, cohorts 6 through 15 were assigned a 10-week dose based on the TiTE-CRM algorithm.

Table 1. Participant characteristics.

Participant characteristic		N (%)
Total N		42
Age	<35	1 (2.4)
	36–45	4 (9.5)
	46–55	12 (28.6)
	56–65	18 (42.9)
	≥66	7 (16.7)
Sex; N (%)	Female	27 (64.3)
	Male	15 (35.7)
Race; N (%)	American Indian or Alaskan Native	0 (0.0)
	Asian	5 (11.9)
	Black	3 (7.1)
	Mixed/more than one race	2 (4.8)
	Other/unknown/not reported	3 (7.1)
	White	29 (69.0)
Ethnicity; N (%)	Hispanic	3 (7.1)
	Non-Hispanic	37 (88.1)
	Unknown	2 (4.8)

Primary outcome

Of the 42 participants who received the intervention, 40 (95.2%) had sufficient data during the baseline and follow-up periods to generate estimates of the MED. Of this sample of 40 individuals, 7 participants (17.5%) achieved the goal of an increase of 2000 steps (or more) per day between baseline and follow-up. Based on this initial data, there were not enough participants who achieved the goal to calculate the MED for this BCT intervention to increase walking. The probability of a successful increase in daily walking based on the TiTE-CRM algorithm was 0.49 (95% CI: 0.37 to 0.67) for the longest duration of the intervention (ie, 10 weeks).

Secondary outcomes

Analyses examining the association between intervention and steps showed that during the intervention period, participants increased their daily steps (B(SE)=428.40 (196.03); p=0.029; table 2). There was also a small drop in participant steps during follow-up that did not reach the 0.05 significance level of alpha (B(SE)=−229.91 (256.90); p=0.371; table 2). The association between intervention dose (ie, length of the intervention in weeks) and average daily step counts was not statistically significant but negative (B(SE)=−34.56 (30.66); p=0.260; table 2). There was also a non-significant but positive association between intervention dose and follow-up (B(SE)=137.82 (123.99); p=0.266; table 2). However, neither of these interaction effects was statistically significant and should be interpreted cautiously.

Table 2. Regression parameters for step count by intervention and follow-up period.

Parameter	B (SE)	P value
Run-in period/intercept	6627.91 (479.42)	<0.001
Intervention period	428.40 (196.03)	0.029
Follow-up period	−229.91 (256.90)	0.371
Treatment period by dose interaction effects
Run-in period/intercept	6633.13 (480.01)	<0.001
Intervention period	588.98 (238.09)	0.013
Follow-up period	−1476.66 (1139.67)	0.195
Intervention by treatment length interaction	−34.56 (30.66)	0.260
Follow-up by treatment length interaction	137.82 (123.99)	0.266

Parameter estimates are represented as step counts.

Examinations of the associations between the five MoAs and step counts showed that only changes in barriers to walking were associated with daily average step counts (B(SE)=−1216.13 (470.66); p=0.011; table 3). Changes in the other four potential MoAs were not associated with changes in average daily steps. However, the intervention was found to be significantly associated with four of the five potential MoAs. The intervention was associated with increases in self-efficacy (B(SE)=0.08 (0.03); p=0.002; table 4), intrinsic regulation (B(SE)=0.20 (0.10); p=0.037; table 4), discrepancy in behaviour (B(SE)=−1216.13 (470.66); p=0.011; table 4), and motivation (B(SE)=−1216.13 (470.66); p=0.011; table 4). Intervention dose was found to be associated with increases in self-efficacy (B(SE)=0.01 (0.01); p=0.015; table 4) and decreases in intrinsic regulation (B(SE)=−0.04 (0.02); p=0.045; table 4).

Table 3. Regression parameters for step count by intervention, follow-up period and mechanisms of action (MoAs).

Parameter		B (SE)	P value
Run-in period/intercept		5833.01 (1974.61)	0.0037
Intervention period		681.26 (615.68)	0.270
Follow-up period		−1913.40 (849.03)	0.025
MoA	Self-efficacy	3001.87 (1557.88)	0.056
	Intrinsic regulation	383.84 (352.86)	0.279
	Discrepancy	−129.62 (204.04)	0.526
	Motivation	299.93 (215.27)	0.166
	Barriers	−1216.13 (470.66)	0.011

Note: Parameter estimates are represented as step counts.

Table 4. Associations between the intervention and mechanisms of action.

Parameter	B (SE)	P value
Self-efficacy
Run-in period/intercept	0.66 (0.04)	<0.001
Intervention period	0.08 (0.03)	0.0017
Follow-up period	0.13 (0.03)	<0.001
Treatment period by dose interaction effects
Run-in period/intercept	0.64 (0.04)	<0.001
Intervention period	0.03 (0.03)	0.390
Follow-up period	0.01 (0.15)	0.958
Intervention by treatment length interaction	0.01 (0.01)	0.015
Follow-up by treatment length interaction	0.02 (0.02)	0.270
Intrinsic regulation
Run-in period/intercept	2.13 (0.17)	<0.001
Intervention period	0.20 (0.10)	0.037
Follow-up period	0.34 (0.13)	0.011
Treatment period by dose interaction effects
Run-in period/intercept	2.17 (0.17)	<0.001
Intervention period	0.38 (0.13)	0.004
Follow-up period	0.93 (0.57)	0.106
Intervention by treatment length interaction	−0.04 (0.02)	0.045
Follow-up by treatment length interaction	−0.08 (0.06)	0.231
Discrepancy in behaviour
Run-in period/intercept	4.90 (0.26)	<0.001
Intervention period	−0.73 (0.21)	<0.001
Follow-up period	−0.93 (0.28)	<0.001
Treatment period by dose interaction effects
Run-in period/intercept	4.90 (0.26)	<0.001
Intervention period	−0.71 (0.30)	0.018
Follow-up period	−1.78 (1.24)	0.152
Intervention by treatment length interaction	−0.00 (0.04)	0.935
Follow-up by treatment length interaction	0.10 (0.13)	0.482
Motivation
Run-in period/intercept	4.44 (0.26)	<0.001
Intervention period	0.38 (0.18)	0.039
Follow-up period	0.11 (0.25)	0.671
Treatment period by dose interaction effects
Run-in period/intercept	4.46 (0.26)	<0.001
Intervention period	0.51 (0.26)	0.049
Follow-up period	0.72 (1.10)	0.517
Intervention by treatment length interaction	−0.03 (0.04)	0.474
Follow-up by treatment length interaction	−0.07 (0.12)	0.552
Barriers
Run-in period/intercept	2.28 (0.12)	<0.001
Intervention period	0.04 (0.08)	0.613
Follow-up period	0.09 (0.11)	0.390
Treatment period by dose interaction effects
Run-in period/intercept	2.27 (0.12)	<0.001
Intervention period	0.13 (0.12)	0.270
Follow-up period	−0.26 (0.48)	0.592
Intervention by treatment length interaction	−0.02 (0.02)	0.308
Follow-up by treatment length interaction	0.04 (0.05)	0.453

Discussion

The current study was not able to identify the MED for a multi-BCT intervention to increase walking by an average of 2000 daily steps in individuals on primary statin therapy. Though the current trial uses an adaptive trial design that is recommended for behavioural interventions,²⁶ our findings suggest that the MED is either greater than the 10-week maximum used in the current trial or that there is no MED for a multi-BCT intervention for walking when the dose is operationalised as duration. We did find that the intervention was associated with increases in daily walking and in four potential MoAs for walking. However, only changes in the barriers to walking MoA were significantly associated with average daily walking.

Our finding that there was not sufficient information to calculate the MED for this multi-BCT intervention to increase walking was unexpected. The behavioural intervention dose to find an MED can be characterised by duration, frequency or amount.⁴⁷ In this study, duration was selected as the dimension of dose we chose to vary between participant cohorts because (1) duration is most often reported in the literature, while frequency and particularly amount are inconsistently reported⁴⁷ and (2) because of the effects of duration on participant burden and engagement.²⁶ However, varying the frequency (ie, how many times the participants receive the BCTs) and/or the amount (ie, number and combination of BCTs) may better operationalise the dose of a multi-BCT intervention and allow for identifying the MED. There are also interaction effects between dose parameters, where modifying, for example, duration and frequency may more clearly elucidate the effective dose⁴⁷; this should be a consideration for future trials. Implicitly, when an MED is established, it should be significantly better than the response expected for no intervention at all (in this case, no increase in average daily walking). Our finding that the probability of a successful increase in daily walking was just under 50% for the longest duration of the intervention (ie, 10 weeks) suggests that even with the longest duration intervention, we would only expect approximately half of the sample to achieve a successful increase in daily walking. More research is needed to examine if a longer duration multi-BCT intervention to increase walking enables determination of the MED, or if varying the behavioural dose by frequency or amount would lead to identification of the MED.

Of the five potential MoAs examined, only changes in barriers to walking were associated with daily average step counts. Lower average daily step counts were associated with higher scores on potential barriers to walking. None of the other MoAs were significantly associated with changes in steps. The intervention itself was associated with increases in four of the five MoAs, including self-efficacy, and the intervention dose was associated with increases in self-efficacy but decreases in intrinsic regulation. These findings suggest that the benefits of the intervention on self-efficacy increase as the duration of the intervention also increases, while the benefits of the intervention on intrinsic regulation are attenuated over time. In prior work, self-monitoring (included as one of the BCTs in our intervention) has been promising when used in interventions targeting walking (ie, because it increases self-efficacy and reduces perceived barriers),^{48 49} a finding supported by a systematic review of BCTs used to promote walking.⁵⁰ Alternatively, some researchers have suggested employing different BCT combinations based on the population of interest (statin users in this case).⁵¹ Additional research examining variations in BCT combinations in a larger sample size may help to further identify MoAs for PA increases among statin users.

Examining the effect of the intervention by treatment lengths suggests that individuals with longer treatment lengths have lowered benefits during the intervention period but may have increased benefits during follow-up; though we were not able to identify the MED, longer interventions may help to sustain benefits after BCT messages have stopped. However, these findings should be interpreted cautiously as both intervention-by-duration interaction effects were not statistically significant at the 0.05 level of alpha.

Limitations

One of the primary limitations in the current trial is the 10-week intervention duration. The length of the trial was chosen based on prior trials and the availability of resources. However, the results suggest that a longer research duration may be needed to identify the MED. A relatively short follow-up period may also have limited our ability to fully capture persistent (or delayed) effects. Another primary limitation for the current trial is that the feedback on behaviour BCT was not included in the multi-BCT intervention. With feedback on behaviour added to the current intervention, we may have successfully identified the MED for the intervention. Though feedback is a useful BCT for influencing rates of PA,^{52 53} it is one of many available options that may successfully increase walking. Other BCTs (individually or in combination) may have successfully increased walking and helped to identify the MED for a multi-BCT intervention for walking. Notably, as we used multiple repeated measures over time to examine the association between intervention dose and outcomes using the adaptive method, it is possible that confounding may have occurred and affected our results. Additional research is required to identify which other BCTs or combinations of BCTs may help to identify the MED. Additionally, the current trial had a small sample size of 42 participants; a larger sample size may have helped to identify the dose-response for the intervention. Finally, while our sample was from a diverse metropolitan area, our demographics would not be considered ‘representative’, were not balanced by cohort/dose and only included individuals on primary prevention statins, limiting generalisability of our results.

Conclusions

As CVD is the leading cause of mortality in the USA, high rates of non-adherence to statin therapy (a mainstay of CVD prevention) are concerning. Multi-BCT interventions can increase medication adherence, but the ‘dose’ (ie, intervention duration in weeks) needed to achieve a clinically meaningful improvement in statin adherence is unknown. This study was unable to determine the minimum effective dosage of a multi-BCT intervention to increase walking by 2000 steps each day among individuals on primary prevention statin therapy. Knowing what the minimum effective dosage is would provide valuable public health information (PHI) for adults at risk for CVD, and additional research is warranted to determine the minimal length of an intervention (ie, intervention dose) necessary to increase PA among statin users.

HARMS

Treatment adverse events

This study poses a minimal risk of physical harm to subjects. Risk of loss of confidentiality or privacy was minimised by securely storing data including PHI in a Northwell-approved database and minimising the use of PHI. Only study team members approved by the Northwell Health IRB were able to access this data. Participants were informed they could withdraw from the study at any time without consequences. Participants may experience mild skin irritation from using the Fitbit activity monitor but were instructed on methods to reduce irritation and were told they could remove the Fitbit for a brief period of time.

The funder had no role in the design and conduct of the study; collection, management, analysis and interpretation of the data; preparation, review or approval of the manuscript; or decision to submit the manuscript for publication.

Data availability statement

Data are available on reasonable request.