Long-term health outcomes and cost-effectiveness of a computer-tailored physical activity intervention among people aged over fifty: modelling the results of a randomized controlled trial

Four different intervention-versions and a waiting-list control group (who did not receive the intervention until the study was ended) were studied using a clustered RCT with evaluation assessments at the start, after three, six and 12 months. Results on self-reported PA after 12 months were extrapolated to long-term outcomes (i.e. 5-year, 10-year, and lifetime horizon) using the Chronic Disease Model. The current study was registered at the Dutch Trial Register (NTR2297) and approved by the Medical Ethics Committee of Atrium–Orbis–Zuyd (MEC10-N-36).

Participants were recruited via direct mailing in communities of the Municipal Health Council (MHC) regions participating in this study (2010-2011) [21]. Communities were matched on their urbanity, percentage of people with a low SES, percentage of people with a high SES, percentage of immigrants, and the percentage of people aged over 50. Each MHC provided a random sample of eligible participants. Using a random number generator, regions were randomly assigned to one of five research arms. Participants had to be aged over 50 (no maximum age), and needed a sufficient understanding of the Dutch language. No other in- or exclusion criteria were set.

A power calculation (effect size = 0.4, power = 80%, intracluster-correlation coefficient = 0 · 1) showed that at baseline 420 participants were needed per intervention condition (considering a dropout rate of 40% during the 1-year follow-up based on a previous Active Plus study) [21]. Figure 1 provides an overview of the participant numbers included at baseline and 12-months assessment. To reach an equal number of participants per intervention condition, a higher number of invitations had to be distributed in the online intervention regions. Participants provided informed consent before enrolment.

The Chronic Disease Model is a population-based, Markov-type state-transition model that forecasts the development over time in demographics, prevalence of risk factors, disease incidences, and mortality [19]. Markov-type state-transition models are often used to estimate the health effects of PA behaviour [22‐24]. The Chronic Disease Model describes the changes in physical (in)activity over time (e.g. the model projects a decrease in PA behavior (i.e. an increased risk of less PA) over time as a result of an increase in age) and the resulting changes of morbidity and mortality for several PA-related chronic diseases (i.e. diabetes, colon cancer, breast cancer, acute myocardial infarctions, stroke) in the Dutch population. The Chronic Disease Model has been described extensively by Hoogenveen and colleagues [19]. The model is based on data provided by national studies with regard to the risk factor distribution and transition, and on data from regional and national registries and registries in general practices with regard to disease incidence and prevalence [19].

Leisure-time PA of the Dutch population in the model’s base year (2007) is described as a continuous distribution expressed in MET-hours. MET (Metabolic Equivalents) presents the energy costs of PA as a multiple of resting metabolic rate. MET-hours thereby include both duration and intensity of PA, and are assumed to have a dose-response relation to PA-related disease incidences. The dose-response regression parameters were obtained by meta-regression on several studies (see Additional file 1). Since literature only provides sufficient evidence on the relation between PA-related diseases and leisure-time PA, work-related PA was excluded from the model.

Active Plus is a computer-tailored, theory- and evidence-based intervention to stimulate or maintain PA in people aged over fifty by targeting pre-motivational constructs (i.e. awareness, knowledge), motivational constructs (i.e. attitude, self-efficacy, social influence, intrinsic motivation, and intention) and post-motivational constructs (i.e. strategic planning, action planning and coping planning) [17]. The intervention was developed using the Intervention Mapping protocol [25], and based on several theoretical models such as the I-Change Model [26], transtheoretical model [27], the health action process approach [28], the precaution adoption model [29], the self-regulation theory [30], and the self-determination theory [31]. Whereas several models and studies have emphasized the importance the participants’ perceptions of their environment in explaining PA behaviour [32‐39], additional environmental information was added to the intervention.

In a previous phase of the Active Plus project (2005–2009), the printed intervention was proven to be effective in stimulating physical activity among the over-fifties until one year after the intervention started [40, 41]. In a follow-up study (started in 2010), the intervention was adapted based on the results of the evaluation of the original (print-delivered) intervention, translated into a Web-based version [18], and implemented in an RCT in other regions than the previous Active Plus intervention. The intervention was thus available in a printed and in a Web-based delivery mode, with and without additional environmental information, resulting in the afore mentioned four intervention conditions. Intervention participants received three times tailored advice based on the answers given in previous assessments [17, 18]: (1) after the baseline assessment; (2) two months after the baseline assessment; and (3) within four months after baseline assessment, based on a combination of the first and the second assessment. The specific content of the basic tailored advice depended on the participants’ personal and psychosocial characteristics, PA behaviour, and the extent to which they were planning to change their behaviour [17]. Participants received among others a graphic that provided insight in their own PA behaviour compared to the average PA behaviour of similar others (same age and gender), model stories of similar others, information on the consequences of physical inactivity specifically for their own age group and gender, and suggestions on how to deal with their perceived barriers for physical activity. The intervention with additional environmental components contained the same information as the basic tailored intervention, with additional tailored advice on local PA possibilities and initiatives [42]. Participants received walking and cycling routes tailored to their own neighbourhood, home exercises tailored to their gender, and information on sports opportunities tailored to the opportunities in the participants own neighbourhood, the presence of a chronic physical limitation, and to the participants preferences (e.g. costs, inside or outside, being physically active with others or alone).

The content of the Web-based intervention was identical to the content of the printed intervention; however, the possibility to use interactive components was optimally used (i.e. static modelling pictures and PA exercises were transferred into videos; the printed neighbourhood map was transferred into a Google neighbourhood map; several hyperlinks and a forum were added to the webpage). One day after receiving the online advice, participants in the Web-based intervention received an email with a copy (pdf format) of their advice, enabling them to print and save their tailored advice. The advice sent by email had the same format as the printed version. The tailored advice contained between five and 11 pages of text and illustrations, depending on (changes in) PA behaviour and determinant scores. The tailored advice texts, in the printed letter or on the website, formed the basis of the Active Plus intervention. Intervention components (mainly expressed in pictures/figures/videos or schema’s) were added to the tailored advice to increase the active participation in the intervention. The specific content of the theoretical methods, practical strategies, and intervention components used in the current intervention have been described extensively elsewhere [42].

Baseline characteristics of the study population (N = 2,140) are shown in Table 2. Participants in the control group and the printed environmental intervention were significantly older than participants in the Web-based basic (p = .001; p = .002) and the Web-based environmental intervention (p < .001; p < .001); printed basic intervention participants were significantly older than Web-based environmental participants (p = .001). After 12 months, outcome measures were available for 1,235 participants (58%; Figure 1).

Dropout analyses showed that participants who received additional environmental information (B = .398, p < .001), participants in Web-based conditions (B = .614, p < .001), and younger participants (B = −.016, p = .006) were more likely to dropout from the 12 month follow-up measurement. Demographic predictors of dropout did not differ between the intervention conditions.

Effects of the intervention conditions on PA after 12 months are presented in Table 3 for the complete case analyses and the analyses using imputed values for participants who dropped out. Only the results using imputation are implemented in the Chronic Disease Model.

Analyses on the data including imputations showed that the printed basic (∆MET-hours = 3.06; 95% CI = 0.38-5.75; p = .025) and printed environmental intervention (∆MET-hours = 2.94; 95% CI = 0.26-5.62; p = .031) resulted in a significant increase in MET-hours. Since the control group decrease PA, the effect of the printed basic intervention (p = .022) and the printed environmental intervention (p = .030) were even more significant when compared to the control group. The increase in MET-hours of both Web-based interventions was not statistically significant.

The printed and the Web-based intervention scenarios decreased incidences for PA-related diseases (Table 4). This indicates that although the effect of the Web-based intervention was not statistically significant, it can be clinically relevant. The effects on incidence numbers reduce at longer follow-up times. The printed scenario results in a larger disease incidence decrease than the Web-based scenario, since it was more effective and showed lower dropout rates [16]. Providing additional environmental information resulted in increased incidences compared to the basic intervention, since the effectiveness was lower and dropout was higher.

The printed scenario results in more QALYs gained (i.e. 312,000 QALYs on a lifetime horizon) than the Web-based scenario (i.e. 78,000 QALYs on a lifetime horizon). Compared to the basic intervention, the environmental intervention results in a decrease of 129,000 QALYs. These results are visualized in Figure 2, in which the lines present the QALYs gained for the printed and the Web-based intervention in contrast to care-as-usual (i.e. the zero line in the figure), and for the environmental intervention in contrast to the basic intervention (i.e. the zero line in the figure).

Considering a WTP of €20,000/QALY, on a 10-year and on a lifetime horizon, both the printed and Web-based scenarios were cost-effective (Table 5). On a 5-year horizon (when health-effects are not that noticeable yet), however, the Web-based scenario was only borderline cost-effective and the printed scenario was not cost-effective.

Comparing the Web-based to the printed scenario, on a 10-year and on a lifetime horizon monetary savings gained by implementing the Web-based intervention instead of the printed intervention do not outweigh its lower effects (i.e. the ICER of the Web-based intervention versus the printed intervention is lower than the WTP). Since both the cost and the effects are negative when comparing the Web-based intervention to the printed intervention, in this case the WTP can be seen as the minimum amount of money that the society would like to save by implementing a less effective intervention. Since the ICER of the Web-based intervention compared to the printed intervention is lower than €20.000, the monetary savings of implementing the Web-based intervention do not outweigh its lower effects, making the printed intervention the preferred intervention condition on a 10 year or a lifetime horizon. In contrast, on a 5-year horizon, the lower intervention costs of the Web-based intervention when compared to the printed intervention do outweigh the loss in effectiveness over this short time horizon. An ICER of €32,290/QALY indicates that for a decrease of one QALY in effect compared to the printed intervention, society would save €32,290, making the Web-based intervention the preferred intervention condition on a 5-year time horizon.

The environmental scenario results in higher intervention costs and lower effects than the basic scenario. However, total costs of the environmental scenario on lifetime horizon are lower than the basic scenario, because the basic scenario results in increased life years, and thereby increased healthcare costs. These monetary savings of the environmental scenario do not outweigh the loss in health gains (ICER < €20,000/QALY).

Figure 3 shows that on all time horizons, the outcomes of the printed and the Web-based scenarios are mainly located in the north-east quadrant of the cost-effectiveness plane implying that both scenarios are more effective and have higher costs than care-as-usual. On a 5-year horizon, the ICER is mainly located above the WTP (i.e. not cost-effective). On a 10-year and a lifetime horizon, the ICER is mainly below the WTP (i.e. cost-effective). The probability of the printed and Web-based interventions being cost-effective compared to care-as-usual is higher on a 10-year horizon than on a 5-year horizon (Figure 4), since the effects on QALYs and health-care costs become more noticeable over time. On a life-time horizon, probability of being cost-effective slightly declines as a result of a higher life expectancy (due to increased PA), and thereby an increase in healthcare costs.

The outcomes of the environmental scenario (in contrast to the basic scenario) move from the north-west quadrant to the south-west quadrant of the cost-effectiveness plane when the time horizon increases (Figure 3), implying that incremental costs of the environmental scenario decrease with time. At a WTP of €20,000/QALY, the environmental scenario had a probability of less than 20% of being cost-effective on all time horizons (compared to the basic intervention) (Figure 4).

Analyses varying discount and participation rates mainly resulted in comparable ICERs (Table 6). A remarkable difference is observed in the ICER of the Web-based scenario with minimal participation. Since intervention costs of the Web-based intervention mainly exist of fixed costs and less costs per participant when compared to the printed intervention, a lower participation rate results in higher intervention costs per participant than the printed intervention. All other sensitivity analyses resulted in an ICER < 20,000, proving cost-effectiveness of the intervention and confirming the robustness of the results.

A main strength of this study is that cost-effectiveness of the intervention is extrapolated beyond the time horizon of the RCT using a model-based approach, resulting in summary measures of population health [19]. However, modelling unavoidably involves making assumptions, which increases the outcomes’ uncertainty. One of the assumptions made was on the persistence of the intervention effect. Based on the results of another PA intervention in an older population, it was assumed that about 72% of the intervention effects on PA observed at 12 months persisted [50]. A study of McAuley et al that was performed among older adults showed a persistence of 85% of the PA behaviour from year 1 (M = 148.80) to year 5 (M = 126.72) after a six month intervention period [61]. A study on the Diabetes Prevention Program, showed a persistence of 76% of the PA behaviour from the end of a lifestyle intervention (where an increase 21.5 met h/week was observed) to year five (were 16.3 met h/week of the PA behaviour was maintained) [62]. Compared to both studies, and taken into account that the Chronic Disease Models additionally projects a decrease in PA behaviour over time as a result of an increase in age, our assumption on the persistence of the intervention effect is relatively conservative.

Other important assumptions made were on the population being reached by the intervention, the effect in participants who dropped-out, and on the immediate dose-response relationship between PA-levels and PA-related disease risks. In reality, a more complex relation might exist with all past PA-levels and other affecting risk behaviours (e.g. smoking). Furthermore, the reach of the intervention might differ due to the use of different recruitment strategies. However, the current approach is in line with estimates from literature, and in most cases assumptions were made as conservative as possible. Sensitivity analyses making variations in assumptions, proved the robustness of the results in this paper.

A disadvantage of the CDM is that it only keeps track of the probability of having a disease, but not includes the probability of having multiple diseases [19]. Another main aspect of the CDM is its Markov type 1 property, which means that conditional on the current demographics and risk factors, future probabilities are estimated independent of the past behaviours [19]. As it can be expected that, especially in an older population, past PA behaviours will determine ones future health states as well, outcomes might be different if past behaviours were included as well.

Another limitation of the current study is that effectiveness estimates were based on self-reported PA. Although self-reports may be less accurate than objective observations, self-administered questionnaires are valid, most commonly used, and most inexpensive method to use in large-scale studies. However, stronger validation of the effects on PA using objective measurements is recommended [63]. Furthermore, the baseline level of PA observed in the current study population was relatively high. This might be caused by the use of self-reported PA questionnaires. Increasing PA behaviour in a highly active population might not make much health differences. However, the relative risk ratios used in the Chronic Disease Model are based on a broad range of PA behaviours to prevent overestimating of the health effects in populations who were already highly active. Promoting PA in less active populations might result in higher health effects and consequently more positive cost-effectiveness ratios.

A final limitation is that the current modelling analyses were only performed from a preventive perspective (i.e. the effects of PA on a decreased risk on diabetes, colon cancer, breast cancer, acute myocardial infarctions, stroke were calculated), ignoring other potential health related effects of increased PA, like improved mobility and independence, improved muscle strength, cognitive functioning, and mental and emotional well-being, and a decreased risk of falling among older adults [64‐69]. Furthermore, by only including myocardial infarction and not including other potential heart diseases (due to a lack of valid data) the health effect of the intervention might be underestimated. If all those aspects were taken into account as well, the effect of the intervention on quality of life might have been even higher.