Study design, setting and population
This study on the effects of the Active Plus intervention on CF was a clustered two-group RCT with a waiting list control group and measurements at baseline, six and 12 months. Ethical approval for the study was obtained from the Research Ethics Committee of the Open University and the trial is registered in the Dutch Trial Register, protocol number NL6005. The study was conducted following the Declaration of Helsinki. A comprehensive rationale and description of the study protocol is published elsewhere [34].
Two socioeconomic comparable neighborhoods or residential areas from seven municipalities were randomly allocated [35] to either the intervention group or the waiting list control group. Due to the nature of the study, blinding was not possible. Municipalities selected between 250 and 2000 independently living adults aged 65 years or older per neighborhood. Potential participants were invited for study participation by their municipalities with an invitation letter via post containing information about the study and an informed consent which could be returned to the researchers from February 2018 until July 2018.
Inclusion criteria were: 65 years or older, fluent in the Dutch language, and suffering from at least one self-reported chronic disease that affects mobility or other physical problems that may affect mobility.Footnote 1 Participants were excluded if they reported severe cognitive problems or were wheelchair users. Participants had to be able to walk at least 100 m, possibly with the help of a walker or walking stick. All participants provided written informed consent.
Procedure
Figure 1 displays the design of the study [34]. To assess PA, participants in both conditions were asked to wear an accelerometer (ActiGraph GT3X-BT) on their right hip for seven consecutive days prior to the CF tests at baseline. The CF tests were conducted by a trained researcher or student at the participants’ home. Inquisit 5 software [36] was used on a tablet (iPad Air 2) to execute the CF tests. To become familiar with the iPad, participants were asked to draw a house and a tree in the Notes application. The CF tests started with the first part of the Verbal Learning Test (VLT), followed by the Trail Making Test (TMT) part A and B, the Stop-signal Task (SST), the Letter Digit Substitution Test (LDST), and the second part of the VLT. Completion time in total was around 45 min. Any occurring disturbances or difficulties in completing the tests were administered to clarify whether the test data were valid. After completing the CF tests, participants received both a paper-based (with a prepaid return envelope) and an online questionnaire (integrated into the project website: www.actief-plus.nl) with the choice to fill out their preferred version (e.g., written or online) within two weeks. In total, 33% chose to fill in this questionnaire online. The questionnaire was used to gather information on demographic variables, but also on concepts that are outside the scope of this manuscript (e.g., self-reported PA, self-reliance, health related quality of life). Thereafter, the four month lasting intervention commenced for the experimental group. Six months (± 3 weeks) and 12 months (± 3 weeks) after the first accelerometer measurement, participants wore that device again, were visited at home for the CF tests and completed a questionnaire following the same procedure as the baseline measurement. After the final assessment (i.e., after 12 months) participants in the waiting list control group received access to the Active Plus intervention.
Intervention group
The Active Plus intervention is a computer-tailored intervention aimed at awareness, initiation, and maintenance of PA behavior, especially in daily life, and was originally developed for the general older adult population aged 50 years or over [27]. The content is structured in line with behavioral change theories such as the theory of planned behavior [37], precaution adoption process model [38], integrated model for change (I-Change Model) [39], and self-regulation theory [40]. The intervention was found to be effective in increasing PA behavior in general population of adults over 50 [28]. At a later stage the intervention was adapted with the intervention mapping protocol to an older population of 65 years or older who suffered from chronic diseases [30]. Information was gathered trough literature study, focus groups and expert panels. Although general determinants for PA in older adults with chronic diseases did not seem to differ from the general older adult population, some determinants had a larger influence on PA in this specific population. For example, pain, fear of injury, and lack of energy turned out to be more important barriers. Therefore, the tailored messages were rewritten for the older adults with chronic diseases population to fit more to their needs and requirements. In the present study we tested this adapted intervention. An overview of the adjustments to the adapted intervention made for this study can be found in the design paper [34]. The intervention was tailored to two extra common chronic diseases (neuromuscular and vascular disorders) in addition to the existing 13 common chronic diseasesFootnote 2 to which Active Plus was already tailored. Furthermore, information on the risks of sedentary behavior and benefits of PA for CF were extended.
As can be seen in Fig. 1, participants in the intervention group received three times advice, both online on a secured website (if they provided an e-mail address) and on paper (via a letter by mail). The tailored text was the same in both versions, but the online version contained more interactive content (e.g., videos). The first and second personal advice were tailored to the answers that the participants gave in the baseline questionnaire and were received respectively within two weeks and two months after filling in the baseline questionnaire. Three months after the baseline questionnaire, a follow-up questionnaire was conducted that was only used to compose the third advice (i.e. not used to assess any effects). Participants received their third advice within two weeks after completing the follow-up questionnaire. Thus, the intervention period in total lasted four months.
Each advice gave tailored information on PA especially focused on older adults with chronic diseases. Among other things, in the first and second advice participants obtained information on their PA level and whether this was sufficient to gain health effects. Furthermore, they were urged to think of reasons that could motivate them to be sufficiently active and how to overcome barriers in achieving this goal. In addition, suggestions on how to implement PA in daily life and avert fallbacks were given. The third tailored advice provided feedback about the progress in behavior and determinant scores in the previous months. The exact content of all three advice depended on the participants’ characteristics (e.g., age, gender, and presence of chronic disease), psychosocial characteristics/motivational constructs (e.g., awareness, intention, self-efficacy and action planning), their current PA behavior, and to what extent they were willing to alter their behavior. The website and advice also included additional information on local PA possibilities (e.g., walking or cycling routes in their neighborhood or local sports clubs), as well as a user forum, and examples of PA exercises.
Waiting list control group
Waiting list control group participants received usual care and had no access to the intervention until the 12-month study period ended. Hereafter they gained access to the Active Plus intervention and received their personalized PA advice on paper and online. But the waiting list control group participants were still visited at home by a researcher or student for the assessment of the CF tests.
Outcome measures
Table 1 provides an overview of all outcome measures and other assessed variables. The concepts of CF (e.g., verbal memory, shifting, inhibition, processing speed) assessed in this study are chosen because they are known to deteriorate with age and can possibly improve with increased PA behavior (Table 1) [1, 11, 16, 49,50,51]. There are no normative data available, as these tests, administered with Inquisit 5 software, cannot be compared to pencil and paper versions of the tests [52, 53].
In the Verbal Learning Test (VLT) [49, 54], which assesses verbal memory, 15 common monosyllabic words representing concrete objects were presented one by one on an iPad screen in fixed order, with a presentation time of one second and an interstimulus interval of one second. Afterwards participants were asked to verbally recall the words they had remembered. The first trial was followed by four more trials in which the words were presented in identical order and each followed by an immediate free recall procedure. After a delay of 15–25 min in which the remaining CF tests were assessed, and unexpectedly for the participants, the instruction was given to recall the 15 words learned once more. Finally, a recognition trial was administered where participants had to recognize the 15 learned words out of 30 words. Outcome measures for the VLT were the learning curve ratio over trials 1–5, the mean number of recalled words in trial 1–5, and the number of words recalled in delayed trial (Table 1).
During the Trail Making Test (TMT) part A and B [55], which can be used to asses shifting, participants had to draw lines with their fingers connecting 25 randomly placed numbers in the correct order (part A) or numbers and letters alternatively (part B). The time in seconds required to complete the task was noted for each task. The outcome measure for the TMT was the time to complete part B minus the time to complete part A.
In the Stop-signal Task (SST) [56], which is an inhibition task, participants had to quickly press the left-hand button if the arrow on the iPad screen pointed to the left and press the right-hand button if the arrow pointed to the right. However, when a signal beep was played after the presentation of the arrow, participants should have inhibited their reaction and withheld from pressing either of the buttons. These beeps occurred in 25% of the trials. Firstly, participants could practice the task in a block of 32 trials. Afterward, three blocks of 64 trials were completed with 10 s of rest in between blocks. The stop-signal delay between presentation of the arrow and signal beep was varied and depended on participants’ performance. The delay, which started at 250 milliseconds (ms), was increased with 50 ms if the previous inhibition was successful. The delay got smaller with 50 ms if the previous inhibition was unsuccessful. This stop-signal delay staircase design ensured that participants were able to inhibit their response on approximately half of all trials. The stop-signal reaction time (SSRT) was the outcome measure.
During the Letter Digit Substitution Test (LDST) [50], which is a processing speed task, participants were presented with a matrix. Odd rows contained letters; even rows contained empty answer boxes. The task was to translate the letters by clicking the corresponding digits with the help of a provided key. After a practice round of 10 letters, the participant had 60 s to replace as many randomized letters with the appropriate digit indicated by the key. The outcome measure for the LDST was the number of correct substitutions made in 60 s.
Demographic and health characteristics
As age, gender, educational level, marital status (living together with a spouse or living single), body mass index (BMI), and physical impairment, are known to influence PA behavior [57] and some also CF [58], these factors were assessed at baseline. Educational level is categorized into low (i.e., primary, basic vocational, or lower general school), moderate (i.e., medium vocational school, higher general secondary education, and preparatory academic education), or high (i.e., higher vocational school or university level) according to the Dutch educational system.
BMI is defined as the body mass (in kg) divided by the square of body height (in cm). The degree of physical impairment is measured with a self-report questionnaire [30]. The participant stated for 14 common chronic diseases (e.g., cardiovascular, osteoarthritis) and physical conditions (e.g., hearing or visually impaired) to what degree he/she was limited in his/her PA behavior by one of the diseases mentioned or by another disease not mentioned. For each chronic disease, the participant scored the degree of impairment on a 5-point scale ranging from 0 = not applicable, 1 = not at all/hardly, 2 = a little, 3 = very, to 4 = extremely. Consequently, degree of impairment was computed into three categories following the next rules: (1) Little impaired: a maximum score of one on at least one question, (2) Medium impaired: a maximum score of two on at least one question, (3) Very impaired: at least a score of three or four on at least one question. PA was objectively measured using the ActiGraph GT3X-BT (ActiGraph, Pensacola, FL, USA). The accelerometer was placed on the right hip with an elastic belt. Participants were asked to wear the accelerometer for seven consecutive days. However, during the night participants were not obliged to wear the device. While showering or swimming, the meter had to be removed.
Sample size and statistical power
Based on the general effect size in a review by Northey et al. [9] of earlier PA intervention studies to improve CF, we used an estimated effect size (ES) of 0.3. Because of the multilevel design, the sample size had to be inflated. Therefore, based on the intra-cluster correlation (ICC) of previous Active Plus projects (ICC < 0.01) an estimate of ICC of 0.01 was used. Statistical power analysis using G*Power [59] (ES = 0.30; power = 0.80; ICC = 0.01) showed that 190 participants per group were required. We expected a 30% dropout rate at 12 months based on our previous study [28]. Therefore, 270 participants needed to be enrolled at baseline in both the intervention group and the waiting list control group.
Statistical analyses
Baseline differences between both groups were tested with a χ2 test for categorical variables, a Mann–Whitney U-test for continuous variables with skewed distributions, and an independent two-sample t-test for normally distributed continuous variables. For further analyses, we log transformed the non-normally distributed TMT outcome measure, and for the SST outcome measure SSRT, we applied a square root transformation. To assess predictors of dropout at six and 12 months, logistic regressions with condition, baseline outcome measures, demographics, degree of impairment regarding chronic diseases and amount of moderate-to-vigorous PA and light PA were performed.
Linear mixed-effects models were used to assess intervention effects on CF. With participants originating from different municipalities, it was expected that their data was clustered. Therefore, we applied multilevel analyses with participants nested in municipalities, with level one being the different time points, level two the participant, and level three the municipality. The analyses revealed that the ICC values for all CF outcomes were smaller than 0.01. Consequently, two-level analyses were performed for all outcomes to assess intervention effects. Time, group, and the interaction between time and group were added to the models as fixed effects to assess intervention effects over time. Intervention effects between intervention group and control group were compared between baseline and six months follow-up and between baseline and 12 months follow-up. All models were fitted using the maximum likelihood procedure. For all analyses age, gender, educational level, marital status, BMI, degree of impairment, and objectively measured weekly minutes of moderate-to-vigorous PA and light PA were added as covariates, as all variables except BMI and degree of physical impairment contributed significantly to the multilevel models. In addition, BMI and degree of physical impairment were significant covariates for PA behavior [32]. Continuous variables were standardized. Furthermore, confidence intervals (CI) and ES were calculated for all outcomes. ES were calculated by standardizing all variables [60]. Analyses were conducted on all available and valid data without any ad hoc imputation [61].
For exploratory purposes, differences regarding intervention efficacy were assessed for degree of impairment, age, gender, educational level, marital status, BMI, moderate-to-vigorous PA and light PA. Three-way interaction terms (time × group × covariate) of each significant covariate were separately added to the model. When a three-way interaction term was significant, subgroup effects were examined by repeating the primary analysis. In these multilevel analyses, the two-level data structure was applied again. Subgroups were defined by ‘logical’ cut-off points. For categorical variables the different categories of the covariate were used. For the continuous variable BMI, the groups were split by normal weight (< 25 kg/m2), overweight (25–29.9 kg/m2) or obese (≥30 kg/m2), based on cut-off points used by the WHO. For age, the limit was at 80 years or older. The WHO calls this group the oldest-old [33]. For minutes of moderate-to-vigorous PA the groups were split into meeting the moderate-to-vigorous PA minutes guidelines (≥ 150 min) and not meeting the guidelines. Since interaction terms have less power, the significance levels were set at p < 0.10 for the interaction term [61]. Significance levels for all other analyses were set at p < 0.05. All analyses were conducted using R [62].