The main goal of the PAC was to use the mHealth device as a coach to provide JIT support to prompt real-time changes in PA intensity to maintain MVPA. Participants received individualized walking prescriptions based on heart rate targets equivalent to MVPA obtained from submaximal graded exercise testing (GXT) at baseline. As such, each participant’s heart rate target for the MVPA prescription was different based on their GXT performance (described in the Measures section). The individualized exercise prescriptions were built to gradually increase the intensity of PA and the number of minutes spent in MVPA to 150 per week. Refer to Table 2 for an example of the exercise prescription. Although the prescriptions were based on walking, participants were encouraged to perform any purposeful activity that increased their heart rate above their individualized minimum target equivalent to MVPA. Participants who did not meet the weekly number of MVPA minutes in the prescription during a given week were encouraged to increase MVPA by 10 minutes the following week. The exercise prescription also included prompts for performing strength and flexibility exercises weekly (ankle circles, toe lifts, partial stands, knee flexion and extension, and wrist, shoulder, calf, and shin stretches) accompanied by pictures and descriptions to guide participants on how to perform these exercises safely at home. As our focus was on increasing MVPA, we did not track progress or engagement in the strength and flexibility exercises.

During the randomization visit, participants in the PAC received their individualized walking prescriptions, the mHealth device, and a folder containing readings about PA, safety, and troubleshooting the mHealth device, as well as important contact information and a detailed schedule of upcoming telephone calls and in-person appointments. Participants were also coached on how to use the mHealth device and respond to the JIT feedback during a walking session with study staff. Participants received a weekly personalized email summarizing the number of minutes spent in MVPA for that week (Figure 2), which also delineated their MVPA goal for the following week. Weekly telephone calls from study staff (ZZZ, the PI, who is a licensed clinical neuropsychologist) lasted 10 to 15 minutes and were conducted during the first month, followed by biweekly calls during months 2 and 3. The telephone calls focused on reviewing the minutes of MVPA achieved, identifying challenges to engaging in exercise, revising goals as needed, and checking on each participant’s safety.

At home, the mHealth device tracked participant heart rate during each exercise session, providing JIT haptic and visual feedback in real time (blue=below zone, green=in zone, and red=above zone) when they deviated from their prescribed heart rate target zone. This allowed them to correct their walking pace or exercise intensity in real time. Participants were encouraged to exercise above their minimum heart rate equivalent to MVPA, but they were not discouraged from going above their target zone if they felt it was safe to do so. Minute-by-minute heart rate data gathered from the mHealth device was saved to the device and later synced with the corresponding app and emailed to the study team on a regular basis. This helped with accountability and tracking by the study team to provide weekly feedback about progress via emails and telephone calls. Participants’ perspectives about using the mHealth device in this study have been described elsewhere [23].

Participants assigned to the HAEC received a reading prescription that equated the time spent reading study materials to the time those in the PAC spent engaging in PA (approximately 2.5 hours per week). The reading prescription consisted of 50 short or bulleted chapters to be completed over 12 weeks on topics that included retirement, quality of life, talking to your physician, bladder health, pain, preventing falls, legal and financial planning, heart health, nutrition, stress management, and humor. There were 4 to 5 chapters (the number of chapters varied based on chapter length) emailed to participants every week, which they could read at their own pace. Participants received a weekly quiz via REDCap that required them to answer 3 to 5 questions about each chapter to measure adherence to the reading prescription. Participants received weekly telephone calls from study staff during the first month, followed by biweekly calls during months 2 and 3. The purpose of the telephone calls, which lasted 5 to 10 minutes, was for the participants to receive equal amounts of attention from study staff between the conditions as well as for the study staff to answer any questions related to the materials and check on the participant’s health status. All the chapters were prepared by the investigative team using materials freely available from the National Institute on Aging and Alzheimer’s Association websites among other reputable sources. Textbox 1 lists the reading prescription topics assigned per month.

In a quiet room with a trained study staff member, participants completed the Mattis Dementia Rating Scale [31]; the National Institutes of Health (NIH) Toolbox Cognition Battery [32] (including the optional Oral Symbol Digit Test); the Rey Auditory Verbal Learning Test (RAVLT) [33]; the Golden version of the Stroop Color Word Interference Test [34]; the Wechsler Memory Scale-revised Logical Memory I and II [35]; Trail-Making Test parts A and B [36]; and verbal fluency tests (FAS Test [assessment of phonemic fluency by requesting an individual to orally produce as many words as possible that begin with the letters F, A, and S within 1 minute] and animal naming) [37]. As executive and memory functions are most responsive to exercise in intervention trials with older adults [14], we created executive and memory composite scores by converting raw scores into z scores based on the entire sample, and then averaging across z scores for the following tests:

We picked these measures because they have previously been reported in the PA intervention literature and have been shown to change as a function of exercise [14].

PA was measured using triaxial accelerometers (GT3X; ActiGraph, LLC) for 7 consecutive days [38,39]. Participants were instructed to wear the accelerometer on a belt on their hip during waking hours for a minimum of 12 hours per day. They were also instructed not to change their regular activities. To ensure compliance, all participants received 2 telephone calls from study staff (on days 2 and 5 of the monitoring period). Participant data were considered valid only if they had accumulated a minimum of 600 minutes of accelerometer wear per day and a minimum of 3000 total minutes of wear spread across at least 4 valid days. Data were processed using ActiLife software (version 6.0; ActiGraph, LLC). Using the low-frequency extension option thought to be most appropriate for older adults [40], data were aggregated to 60-second epochs so that published cut points could be applied. Consistent with standard practice, sedentary time was defined as time spent at <100 counts per minute and MVPA as ≥1952 counts per minute [41]. Minutes within each intensity level were then averaged across days worn, reflecting the average time in minutes per day spent at each intensity level.

Cardiorespiratory fitness was measured via submaximal GXT protocol at baseline and at 12 weeks. During the treadmill test, heart rate was plotted against the workload and its corresponding estimated VO₂ [42], and VO₂ max was estimated plotting the slope of the line to estimated maximal heart rate (calculated using the formula 220–age). A 1-minute warm-up at 2 mph (3.2 km/h) was used to familiarize participants with the treadmill and give them time to adjust to walking without holding the handrails. After the warm-up, the speed was increased to 2.5 mph (4 km/h) at 0% grade, equivalent to 2.9 metabolic equivalents (METs), which is just below the threshold for moderate-intensity exercise. This stage lasted 3 minutes to ensure that steady state was achieved. The speed was then increased to 3 mph (4.8 km/h) at 0% grade, equivalent to 3.3 METs, which is just above moderate-intensity exercise. This stage lasted 5 minutes. One-minute stages with grade increments of 1.5% (0.62 METs) per stage continued until the participant’s heart rate reached volitional fatigue. Heart rate was monitored continuously using a Polar heart rate monitor (Polar FT4 or similar; Polar Electro) throughout the test and for at least 5 minutes of recovery. Blood pressure was recorded for the first 3 stages of exercise and every minute during recovery, at which time participants continued walking at a very slow pace (1 mph; 1.6 km/h) for 2 minutes, followed by 3 minutes of sitting recovery with light foot tapping to prevent venous pooling.

The heart rates measured during the final 10 seconds of minutes 7 and 9 of the GXT were used to determine heart rate target zones for the exercise prescription.

These heart rate measurement times occurred during stage 3 (3 mph; 4.8 km/h) at which participants are estimated to be working at a level equivalent to 3.3 METS, a threshold slightly higher than the minimum level to be considered moderate activity. Minutes 7 and 9 of the total protocol were chosen because of the likelihood that participants would have reached steady state after ≥3 minutes at that workload. Steady state is the point at which the heart’s work is equal to the workload and is unlikely to change with increased exercise duration. For exercise prescription purposes, the 2 heart rate values gathered during minutes 7 and 9 of the total protocol were averaged if the difference was ≤3 beats per minute, and the greater of the 2 heart rates was used if the difference was >3 beats per minute. Participants were asked to refrain from consuming caffeine or other stimulants on the day of their test.

For this pilot RCT, we were interested in examining the plausible mechanisms by which changes in MVPA may affect cognition, including changes in cardiorespiratory fitness. We defined changes in cardiorespiratory fitness as the total time to 85% of estimated maximal heart rate (with an exception made for participants taking β-blockers, as described later in this paragraph) and changes in rate-pressure product (RPP), which is the product of heart rate and systolic blood pressure divided by 100 measured toward the end of each of the first 3 stages of exercise. RPP is an indirect assessment of cardiac workload. A decrease in RPP for a given amount of aerobic work often signifies an improvement in cardiac function [43]. Individuals taking β-blockers were asked to exercise to a workload corresponding to 80% of their age-predicted maximal heart rate. This precaution accounts for the approximately 5% reduction in heart rate observed with taking β-blockers and allowed the research team to provide scaled exercise prescriptions.

MRI data were acquired on a GE Discovery MR750 3.0T (General Electric Company) whole body system with a body transmit coil and an 8-channel receive-only head coil at the University of California, San Diego Center for Functional MRI. The structural brain sequence consisted of a high-resolution T1-weighted fast spoiled gradient echo (3DFSPGR) scan for anatomy and registration purposes: 172 1-mm contiguous sagittal slices, field of view=25 cm, repetition time (TR)=8 ms, echo time (TE)=3.1 ms, flip angle=12°, inversion time=600 ms, 256 × 192 matrix, bandwidth=31.25 kHz, frequency direction=S-I, NEX=1, and scan time=8:13 minutes. CBF was quantified with a 2D pseudocontinuous arterial spin labeling (ASL) MRI sequence: TR=4500 ms, TE=3.2 ms, field of view=24 cm, labeling duration=1800 ms, postlabeling delay=2000 ms, with a single-shot spiral acquisition and a total scan time of 4:30 minutes plus a 40.5-second calibration scan. The calibration scan was acquired immediately after the ASL scan using a spiral readout with TR=4.5 seconds and TE=3.2 ms with 8 dummy radiofrequency pulses (amplitude set to 0) to generate a 36-second delay followed by a 90-degree radiofrequency pulse in the last repetition interval to generate proton density–weighted contrast. Field map scans were collected for offline field map correction to minimize signal bunching and dropouts in the frontal or medial temporal lobes.

T1-weighted anatomical images were processed using FreeSurfer software (version 6.0). Briefly, images underwent skull stripping, B1 bias field correction, segmentation of gray matter and white matter, reconstruction of cortical surface models, and parcellation and labeling of regions on the cortical surface, as well as segmentation and labeling of subcortical structures [44]. FreeSurfer was also used to generate intracranial volume. ASL data were processed using the Cerebral Blood Flow Biomedical Informatics Research Network (CBFBIRN) [45] pipeline established by the University of California, San Diego Center for Functional MRI. CBFBIRN uses a combination of custom MATLAB (The MathWorks, Inc) [46] routines and various Analysis of Functional NeuroImages [47] and FMRIB Software Library [48] functions to quantify CBF and adjust for partial volume effects. MATLAB was used to form a mean ASL image from the average difference of the control and tag images. Voxelwise CBF calibration was performed using the proton-density image to convert the ASL difference signal into physiological units (mL/100 g/min). Slice-timing delays were also accounted for, making the postlabeling delay slice specific. Skull stripping of the high-resolution T1-weighted image was performed using the Analysis of Functional NeuroImages 3dSkullStrip function. Tissue segmentation was performed using the FMRIB Software Library Automated Segmentation Tool algorithm to define cerebrospinal fluid, gray matter, and white matter regions. The high-resolution T1-weighted image and partial volume segmentations were registered to ASL space. A linear regression method was used to correct for partial volume effects and ensure that CBF values were not influenced by decreased perfusion in the white matter or increased volume of cerebrospinal fluid [49], with a 5 × 5 regression kernel to obtain corrected gray matter CBF measurements. Each participant’s partial volume–corrected quantified CBF map (in units of mL/100 g tissue/min) was downloaded to a local server. Voxels with negative intensities were replaced with 0. FreeSurfer was used to generate a priori anatomical regions of interest for the CBF data. Briefly, for each participant, the FreeSurfer-formatted T1-weighted brain volume was registered to the ASL CBF-aligned T1-weighted anatomical image (the latter was derived as part of the CBFBIRN pipeline). The resulting coregistration matrix was used to align the FreeSurfer aparc+aseg segmentation volume to the ASL CBF-aligned T1-weighted image. The CBF-aligned FreeSurfer volumes were visually inspected to ensure proper alignment and were then downsampled to the resolution of the CBF ASL image. Mean CBF was then extracted for FreeSurfer regions of interest. For this pilot RCT, we were interested in examining the plausible mechanisms by which changes in MVPA may affect cognition, including changes in total gray matter CBF, frontal lobe CBF (mean of superior frontal, rostral and caudal middle frontal, pars opercularis, pars triangularis, pars orbitalis, lateral and medial orbitofrontal, precentral, paracentral, and frontal pole CBF), and medial temporal lobe CBF (mean of entorhinal, parahippocampal, and hippocampal CBF).

Saliva samples were collected using the Isohelix SK-2S DNA buccal swab kit (Cell Projects Ltd) to provide information regarding genetic risk for Alzheimer's disease. Genotype for apolipoprotein E was determined by real-time analysis of single nucleotide polymorphism using TaqMan technology (Life Technologies). The real-time analysis was performed using validated TaqMan assays on a CFX96 Touch Real-Time Polymerase Chain Reaction Detection System (Bio-Rad Laboratories, Inc) according to the manufacturer’s instructions. Analysis was performed using Bio-Rad CFX Manager software.