Health Psychology

Testing Self-Regulation Interventions to Increase Walking Using Factorial Randomized N-of-1 Trials
Falko F. Sniehotta, Justin Presseau, Nicola Hobbs, and Vera Arajo-Soares Online First Publication, February 20, 2012. doi: 10.1037/a0027337

CITATION Sniehotta, F. F., Presseau, J., Hobbs, N., & Arajo-Soares, V. (2012, February 20). Testing Self-Regulation Interventions to Increase Walking Using Factorial Randomized N-of-1 Trials. Health Psychology. Advance online publication. doi: 10.1037/a0027337

Health Psychology 2012, Vol. , No. , 000 000

2012 American Psychological Association 0278-6133/12/$12.00 DOI: 10.1037/a0027337

BRIEF REPORT

Testing Self-Regulation Interventions to Increase Walking Using Factorial Randomized N-of-1 Trials
Falko F. Sniehotta, Justin Presseau, Nicola Hobbs, and Vera Araujo-Soares
Newcastle University
Objective: To investigate the suitability of N-of-1 randomized controlled trials (RCTs) as a means of testing the effectiveness of behavior change techniques based on self-regulation theory (goal setting and self-monitoring) for promoting walking in healthy adult volunteers. Method: A series of N-of-1 RCTs in 10 normal and overweight adults ages 19 67 (M 36.9 years). We randomly allocated 60 days within each individual to text message-prompted daily goal-setting and/or self-monitoring interventions in accordance with a 2 (step-count goal prompt vs. alternative goal prompt) 2 (self-monitoring: open vs. blinded Omron-HJ-113-E pedometer) factorial design. Aggregated data were analyzed using random intercept multilevel models. Single cases were analyzed individually. The primary outcome was daily pedometer step counts over 60 days. Results: Single-case analyses showed that 4 participants significantly increased walking: 2 on self-monitoring days and 2 on goal-setting days, compared with control days. Six participants did not benefit from the interventions. In aggregated analyses, mean step counts were higher on goal-setting days (8,499.9 vs. 7,956.3) and on self-monitoring days (8,630.3 vs. 7,825.9). Multilevel analyses showed a significant effect of the self-monitoring condition (p .01), the goalsetting condition approached significance (p .08), and there was a small linear increase in walking over time (p .03). Conclusion: N-of-1 randomized trials are a suitable means to test behavioral interventions in individual participants. Keywords: N-of-1 designs, randomized controlled trials, physical activity, self-regulation

Randomized controlled trials (RCTs) are regularly used in health psychological science to evaluate the effects of behavior change techniques and interventions and to test theories of behavior change. Between-subjects RCTs contrast postintervention data obtained from two or more groups of participants randomly allocated to different conditions (Altman & Bland, 1996). Even if a between-subjects analysis identifies a significant group effect in favor of one experimental condition, it is possible that many participants within the group do not benefit from the condition and that some participants may even incur negative effects. Therefore, the conclusions from between-subjects designs cannot be generalized to the individual and the pooling of data in between-subjects statistical analyses may mask the intraindividual effects of an intervention (Molenaar & Campbell, 2009; Ottenbacher, 1990). An alternative to between-subjects RCTs is an N-of-1 RCT design, which randomly allocates time periods within each partic-

Falko F. Sniehotta, Justin Presseau, Nicola Hobbs, and Vera Araujo Soares, Institute of Health and Society, Newcastle University, Newcastle, England. Correspondence concerning this article should be addressed to Falko F. Sniehotta, Centre for Translational Research in Public Health, Institute of Health and Society, Medical Faculty, Newcastle University, BaddileyClark Building, Richardson Road, Newcastle Upon Tyne NE2 4AX, England. E-mail: falko.sniehotta@ncl.ac.uk 1

ipant to conditions (Barlow & Hersen, 1984; Edgington, 1984; Kazdin, 1982). In clinical settings, N-of-1 RCTs are used to determine optimal treatment for individual patients. Typically, an N-of-1 RCT has various treatment phases and a different treatment is administered at each phase. Outcomes across each phase are then compared (Avins, Bent, & Neuhaus, 2005; Guyatt et al., 1986; March et al., 1994). N-of-1 trials are recognized as a suitable methodology to test the effectiveness and theoretical mechanisms of health interventions in individuals (Craig et al., 2008). Moreover, theory in health psychology formulates hypotheses about individual cases. Thus, evaluating theory using idiographic tests within individuals is epistemologically more appropriate than nomothetic between-subjects tests (Westmeyer, 2003). We report the first, to our knowledge, N-of-1 factorial RCT to test the effects of interventions to increase walking among adults based on behavior change techniques inherent to self-regulation theory. Self-regulation theory suggests that people control their behavior by setting a specific goal (e.g., I want to walk at least 10,000 steps today), which serves as a reference value against which self-monitored ongoing behavior is compared. Depending on progress, individuals adjust their efforts to self-regulate their behavior (Carver & Scheier, 1982; Kanfer, 1986; Webb, Sniehotta, & Michie, 2010). Several systematic reviews of between-subjects RCTs have shown that intervention effectiveness in changing physical activity is associated with the use of behavior change techniques that are congruent with self-regulation theory, that is, goal-setting and self-monitoring (Conn, Hafdahl, Brown, &

SNIEHOTTA, PRESSEAU, HOBBS, AND ARAUJO-SOARES

Brown, 2008; Dombrowski et al., in press; Michie, Whittington, Abraham, McAteer, & Gupta, 2009). Walking was selected as a target health behavior in the present study for two reasons. First, walking is a key target for public health. It is the safest, most acceptable form of physical activity for sedentary people; it does not require costly equipment or setup; and it has a solid evidence base (Ogilvie et al., 2007). Second, there is a minimal relationship between self-reports of walking and objectively measured step counts, possibly because walking is often a byproduct of pursuing other objectives and therefore less memorable (Scott, Eves, French, & Hoppe, 2007). Thus, self monitoring of walking can be easily manipulated by providing feedback and external memory through pedometers.

Procedure
Each day at 6 a.m., participants received a text message prompting the type of goal and the corresponding pedometer required for the day. Daily records on the open and blinded pedometers were compared with allocation to verify the fidelity of the selfmonitoring condition. Sample size. As this was the first N-of-1 RCT of behavioral interventions, we had no prior evidence on which to base a priori power analyses. Thus, the rule of thumb that at least 30 participants per comparison arm should provide at least 80% power (Cohen, 1988) was used. The decision was made not to power for interaction effects as the acceptability of delivering the protocol over a long time period had not yet been established. The factorial design was used to reduce study duration and to test the feasibility of this design. Primary outcome. The primary outcome was daily step counts over 60 days. Participants were seen weekly by a researcher to collect the step count data. Baseline questionnaire. Participants were asked to complete a questionnaire prior to randomization containing information relating to sex and self-reported weight and height. Randomization: Sequence generation. A computergenerated list of random numbers generated by a member of the research team without contact to participants was used to allocate equal numbers of days to conditions (30 days per condition). Allocation concealment. Daily text messages prompting interventions were sent by an independent researcher. Blinding. Intervention provision was blinded by prompting interventions via daily text messages. The outcome assessor was blinded to the goal-setting allocations, but could not be blinded to the self-monitoring condition because outcome (daily steps) was recorded on two pedometers with different displays; one visible and one sealed. Participants were not blinded.

Method Design
Ten 2 (goal setting) 2 (self-monitoring) factorial randomized controlled N-of-1 trials were conducted in Scotland.

Participants and Study Setting

Six female and four male normal or overweight adults (MBMI 25.4, SD 1.99) ages 19 67 years (Mage 36.9 years, SD 17.46) were recruited from a university psychology department research volunteer register. Ethical approval was granted by the University of Aberdeen, School of Psychology ethical review board.

Interventions
At the beginning of the study, all participants received the following instructions for each of the goal setting and selfmonitoring interventions. Goal setting for walking. If prompted to set a step count goal for the day, please think through your commitments and activities planned for the day ahead. What do you have to do and what opportunities for walking might arise? Also, we are in Scotland, have a look at the weather. Please make an achievable but challenging goal specific to the day ahead. Goal setting (control condition). In the active control condition, participants received the U.K. Department of Health leaflet 5 a Day Made Easy: Just Eat More (Fruit and Veg) and the same instructions as the walking goal condition rephrased for fruit and vegetable consumption.

Statistical Method
In an initial overall analysis, the main effects of each condition between and within participants were tested using a multilevel random intercepts model with MLwiN2.2 (Rasbash, Charlton, Browne, Healy, & Cameron, 2010). Then 10 separate single-case ordinary least squares regressions were run to investigate for whom goal setting and self-monitoring were effective. The study was not powered to detect interaction effects. For each individual separately, we examined autocorrelation in the steps data series. All models also tested a linear time trend as an indicator of possible cumulative carryover effects. The linear time trend and both main effects were tested simultaneously.

Self-Monitoring
Participants were provided with two Omron HJ-113-E piezoelectronic pedometers: The display on one pedometer was visible (i.e., open) to enable self-monitoring (self-monitoring condition); the other was sealed (i.e., blinded) to prevent self-monitoring (control condition). The equivalence of both devices in recording steps was tested in advance. The Omron HJ-113-E has been validated against triaxial accelerometry and direct observation (Sugden et al., 2008).

Results
Descriptive statistics are presented in Table 1 by condition for each participant and aggregated over all 10 participants. Overall, step counts in the walking goal-setting condition were 8,500 (SD 5,102) compared with 7,956 (SD 5,105) in the goalsetting control condition. Participants walked 8,630 steps (SD 5,238) on self-monitoring days and 7,826 (SD 4,947) on days with the blinded step counter. Autocorrelation in the steps data series for each participant did not differ significantly from zero;

THEORY-BASED RANDOMIZED N-OF-1 TRIALS FOR WALKING

Table 1 Descriptive Statistics Between and Within Participants

Mean (SD) steps Goal-setting condition Participant Overall 1 2 3 4 5 6 7 8 9 10 Note. M male; F Age, sex, BMI 24, M, 28.1 43, F, 23.9 50, F, 24.7 67, M, 26.0 19, F, 22.8 20, M, 28.7 24, F, 23.3 53, M, 24.0 20, F, 26.3 49, F, 26.3 female; BMI Intervention 8,500 (5,102) 8,498 (3,972) 11,181 (3,226) 6,361 (2,203) 4,863 (1,902) 14,420 (7,044) 8,158 (3,763) 5,090 (1,695) 12,350 (5,577) 9,806 (4,324) 4,273 (2,724) body mass index. Control 7,956 (5,105) 7,376 (3,148) 9,209 (3,618) 6,279 (1,958) 4,566 (1,646) 14,321 (7,048) 8,159 (4,165) 4,810 (1,935) 12,397 (5,075) 9,194 (4,853) 3,251 (2,274) Self-monitoring condition Intervention 8,630 (5,238) 8,350 (3,647) 9,844 (3,319) 6,441 (2,083) 5,441 (1,851) 16,398 (6,982) 7,908 (3,814) 5,103 (1,723) 12,477 (5,177) 10,292 (4,227) 4,050 (2,652) Control 7,826 (4,947) 7,525 (3,562) 10,546 (3,774) 6,198 (2,079) 3,988 (1,361) 12,345 (6,484) 8,409 (4,103) 4,797 (1,908) 12,270 (5,480) 8,708 (4,825) 3,474 (2,434)

therefore, analyses were conducted without controlling for autocorrelation.

Overall Analysis
Decomposing the variability in step counts between and within individuals demonstrated that the majority (59.4%) of the variability in step counts was observed within individuals. Overall, goal setting approached significance (B 565.86, SE 318.55, p .08, 95% CI [ 58.49, 1190.21]), and self-monitoring led to increased step counts overall (B 793.72, SE 318.43, p .01, 95% CI [169.61, 1417.83]).1 In addition, a small linear time effect was found, indicating an increase of step records over the study period (B 20.35, SE 9.20, p .03, 95% CI [2.33, 38.38]), indicating possible small cumulative carryover effects.

Single-Case Analyses
Table 2 presents a series of single-case analyses testing for main effects of goal setting and self-monitoring and linear time trends. Participants 2 and 10 had significantly higher step counts on days that they set a step count goal, whereas Participants 4 and 5 showed a significant self-monitoring effect. Participants 8, 9, and 10 showed a slight increase in walking over the study period, and Participant 1 showed a small decrease.

Discussion
This study was the first to test behavior change techniques in a series of N-of-1 RCTs. We found that most of the 10 participants showed higher step counts on days allocated to the goal-setting and self-monitoring interventions compared with control days, and there was a small linear increase over time. In the aggregated analyses, a main effect for self-monitoring was found, whereas the main effect for goal setting only approached significance (p .08). The observed self-monitoring effect was smaller than the effect found in a systematic review of the use of pedometers in between-participants RCTs (Bravata et al., 2007). However, this review also included interventions that involved additional behav-

ior change techniques to the delivery/use of pedometers. The single-case analyses revealed that the self-monitoring intervention was effective for two participants and the goal-setting intervention was effective for two participants. For the remaining six participants, we did not find significant differences between the intervention and control conditions. However, the finding that differences between intervention and control days as high as 1,122 steps for goal setting (Participant 1) and 1,584 for self-monitoring (Participant 9) were not significant indicates limited statistical power to detect relevant differences in some cases because of considerable variability in day-to-day walking. Power differs between individuals in N-of-1 RCTs. If N-of-1 RCTs are to be used as a basis for individual treatment decisions, dynamic algorithms could be developed and used to determine the minimally sufficient number of observations needed per case to base a treatment decision on a statistically significant difference between conditions. Overall, goal setting and self-monitoring were effective for some but not all participants. Based on 10 participants, it is not possible to predict the type of participant for which each intervention would be effective. Future research should combine withinsubjects and between-subjects approaches to identify moderators of effectiveness within individuals. These results suggest that N-of-1 RCTs can be particularly valuable for health psychology in selecting appropriate theory for intervention development and delivery and to tailor intervention content to participants by pretesting to which type of behavior change technique or combination thereof an individual best responds. For complex long-term interventions, single intervention components can be randomly varied over time and be optimized by an iterative process of adjusting interventions to participants. As such, N-of-1 RCTs may offer a solution to some of the key challenges to the science of behavior change. The N-of-1 RCT methodology has several limitations. First, the range of suitable interventions may be limited to those that can be
These results were invariant for sensitivity analyses controlling for autocorrelation.
1

SNIEHOTTA, PRESSEAU, HOBBS, AND ARAUJO-SOARES

Table 2 Main Effects by Single Case With Linear Time Trend

Goal-setting condition Participant 1 2 3 4 5 6 7 8 9 10 Note. B 1157 1947 66 314 66 24 297 839 82 1392 SE 873 859 527 413 1721 985 459 1371 1135 613 p .19 .02 .90 .45 .97 .98 .52 .54 .94 .02 95% CI 554, 2867 263, 3632 967, 1098 494, 1123 3439, 3306 1906, 1954 621, 1182 1848, 3526 2142, 2307 191, 2593 B 699 544 159 1434 4183 574 326 483 1317 257 Self-monitoring condition SE 875 875 541 413 1714 986 459 1312 1115 609 p .42 .53 .77 .01 .02 .56 .48 .71 .24 .67 95% CI 1016, 2413 2259, 1171 901, 1219 625, 2243 824, 7542 2506, 1357 596, 1207 2088, 3054 867, 3502 936, 1450 B 53 24 11 5 35 46 9 84 70 49 Time trend (linear) SE 25 25 15 12 50 29 13 40 33 18 p .04 .35 .50 .69 .49 .11 .50 .03 .03 .01 95% CI 102, 3 26, 73 20, 40 28, 19 132, 63 10, 102 17, 35 6, 162 6, 135 13, 84

Polynomial time trends were also tested, but linear time trends, where observed, were consistently stronger than polynomial trends.

provided and withheld as part of experimental procedures and that are unlikely to create carryover effects from periods allocated to the intervention to those allocated to other conditions. For example, N-of-1 RCTs might be less applicable for testing social cognitive theories, as most interventions targeting beliefs such as self-efficacy or risk perceptions would have carryover effects beyond the time in which they are delivered. The methodology may be best suited to techniques such as planning, goal setting, contingent reinforcement, self-monitoring, and feedback, where carryover effects can be minimized. However, unlike in pharmacological trials, there is no conclusive test to rule out the possibility that previous interventions maintain their effect on behavior, and careful tests need to be developed to establish the wash-out periods required to prevent carryover effects of previous interventions. In this study, no autocorrelation and a small increase over time may serve as weak indicators for limited risk of carryover effects. Eventually, more sophisticated N-of-1 RCT designs will be needed to understand carryover effects, for example, by allocating time periods after intervention delivery with no intervention and measuring whether effects exceed the prior period of intervention delivery. In addition, walking varies considerably within individuals. There is scope to extend N-of-1 RCTs by using established methods from single-case research to control for covariates, confounders, and mediators over time to enhance the explanatory value of the study and to account for intraindividual variability not explained by the interventions. This study provided the first evidence that N-of-1 RCTs can be used to robustly test behavioral interventions. It showed that selfregulation techniques are effective in increasing walking in some but not all participants. More research is needed to test what works for whom in behavior change and N-of-1 RCTs might help to understand these processes.

References
Altman, D. G., & Bland, J. M. (1996). Statistics notes: Comparing several groups using analysis of variance. British Medical Journal, 312, 1472. doi:10.1136/bmj.312.7044.1472 Avins, A. L., Bent, S., & Neuhaus, J. M. (2005). Use of an embedded N-of-1 trial to improve adherence and increase information from a

clinical study. Contemporary Clinical Trials, 26, 397 401. doi:10.1016/ j.cct.2005.02.004 Barlow, D. H., & Hersen, M. (1984). Single case experiment designs: Strategies for evaluating change (2nd ed.). New York, NY: Pergamon Press. Bravata, D. M., Smith-Spangler, C., Sundaram, V., Gienger, A. L., Lin, N., Lewis, R., . . . Sirard, J. R. (2007). Using pedometers to increase physical activity and improve health: A systematic review. Journal of the American Medical Association, 298, 2296 2304. doi:10.1001/ jama.298.19.2296 Carver, C. S., & Scheier, M. F. (1982). Control theory: A useful conceptual framework for personality-social, clinical, and health psychology. Psychological Bulletin, 92, 111135. doi:10.1037/0033-2909.92.1.111 Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ: Lawrence Earlbaum Associates. Conn, V. S., Hafdahl, A. R., Brown, S. A., & Brown, L. M. (2008). Meta-analysis of patient education interventions to increase physical activity among chronically ill adults. Patient Education and Counseling, 70, 157172. doi:10.1016/j.pec.2007.10.004 Craig, P., Dieppe, P., Macintyre, S., Michie, S., Nazareth, I., & Petticrew, M. (2008). Developing and evaluating complex interventions: The new Medical Research Council guidance. British Medical Journal, 337, 979 983. doi:10.1136/bmj.a1655 Dombrowski, S. U., Sniehotta, F. F., Avenell, A., Johnston, M., MacLennan, G., & Araujo-Soares, V. (in press). Identifying active ingredients in complex behavioural interventions for obese adults with obesity-related co-morbidities or additional risk factors for co-morbidities: A systematic review. Health Psychology Review. Edgington, E. S. (1984). Statistics in single case analysis. Progress in Behavior Modification, 16, 83119. Guyatt, G. H., Sackett, D., Taylor, D. W., Chong, J., Roberts, R., & Pugsley, S. (1986). Determining optimal therapyRandomized trials in individual patients. New England Journal of Medicine, 314, 889 892. doi:10.1056/NEJM198604033141406 Kanfer, F. H. (Ed.). (1986). Helping people change: A textbook of methods (3rd ed.). New York, NY: Pergamon Press. Kazdin, A. E. (1982). Single-case research designs: Methods for clinical and applied settings. New York, NY: Oxford University Press. March, L., Irwig, L., Schwarz, J., Simpson, J., Chock, C., & Brooks, P. (1994). N of 1 trials comparing a non-steroidal anti-inflammatory drug with paracetamol in osteoarthritis. British Medical Journal, 309, 1041 1046. doi:10.1136/bmj.309.6961.1041 Michie, S., Whittington, C., Abraham, C., McAteer, J., & Gupta, S. (2009).

THEORY-BASED RANDOMIZED N-OF-1 TRIALS FOR WALKING Effective techniques in healthy eating and physical activity interventions: A meta-regression. Health Psychology, 28, 690701. doi:10.1037/a0016136 Molenaar, P. C. M., & Campbell, C. G. (2009). The new person-specific paradigm in psychology. Current Directions in Psychological Science, 18, 112117. doi:10.1111/j.1467-8721.2009.01619.x Ogilvie, D., Foster, C., Rothnie, H., Cavill, N., Hamilton, V., Fitzsimons, C. F., & Mutrie, N. (2007). Interventions to promote walking: Systematic review. British Medical Journal, 334, 12042114. doi:10.1136/bmj.39198.722720.BE Ottenbacher, K. J. (1990). Clinically relevant designs for rehabilitation research: The idiographic model. American Journal of Physical Medicine and Rehabilitation, 69, 286292. doi:10.1097/00002060-199012000-00002 Rasbash, J., Charlton, C., Browne, W. J., Healy, M., & Cameron, B. (2010). MLwiN Version 2.3. Centre for Multilevel Modelling, University of Bristol, Bristol, England.

Scott, E. J., Eves, F. F., French, D. P., & Hoppe, R. (2007). The theory of planned behaviour predicts self-reports of walking, but does not predict step count. British Journal of Health Psychology, 12, 601 620. doi: 10.1348/135910706X160335 Sugden, J. A., Sniehotta, F. F., Donnan, P. T, Boyle, P., Johnston, D. W., & McMurdo, M. E. T. (2008). The feasibility of using pedometers and brief advice to increase activity in sedentary older women - A pilot study. BMC Health Services Research, 8, 169. Webb, T. L., Sniehotta, F. F., & Michie, S. (2010). Using theories of behaviour change to inform interventions for addictive behaviours. Addiction, 105, 1879 1892. doi:10.1111/j.1360-0443.2010.03028.x Westmeyer, H. (2003). On the structure of case formulations. European Journal of Psychological Assessment, 19, 210 216. doi:10.1027//10155759.19.3.210