Unwanted bias or promising intervention effect? Dealing with measurement reactivity in controlled trials of physical activity interventions

This invited commentary refers to ‘Examining control group changes in physical activity levels in people with coronary heart disease: a systematic review and meta-analyses’, by J. Seymour et al. https://doi.org/10.1093/eurjcn/zvaf115. Physical activity (PA) is a crucial health behaviour in the secondary prevention of cardiovascular disease (CVD). Accordingly, PA measurements constitute important endpoints in controlled trials of secondary prevention interventions. However, the act of measuring PA itself may induce behavioural, emotional, and/or cognitive changes in study participants, including participants in control groups. This phenomenon is known as measurement reactivity.1 The systematic review and meta-analysis by Seymour et al.2 examined this phenomenon in controlled trials of PA interventions for people with coronary heart disease. The authors synthesized PA changes over time in the control groups of 26 studies and found that in 42% the control groups showed noteworthy improvements in PA levels (≥10%) from baseline to end-intervention. These improvements were partly associated with hospitalization and revascularization and could therefore to some extent reflect recovery from acute episodes. But some influence of measurement reactivity appears likely. On the one hand, measurement reactivity introduces an element of unwanted bias to study results. On the other hand, Seymour et al. suggest that measurement reactivity could be leveraged to promote desired health behaviour. This commentary expands on how researchers might deal with measurement reactivity under either approach. In principle, researchers strive to eliminate measurement error and bias as much as possible, but error and bias are inherent to most measurement procedures. Measurement bias describes a systematic distortion of variables through measurement error, for example systematically higher PA levels due to measurement reactivity. Such bias does not necessarily pose a threat to the internal validity of a controlled trial, if, firstly, the measurements are conducted equally in intervention and control groups (meaning that the bias is equally represented across study groups); and secondly, if the magnitude of the bias is small in relation to the expected intervention effect. But it is recommended that trialists of PA interventions carefully consider to what extent measurement reactivity might introduce relevant bias to their study. The Measurement Reactions In Trials (MERIT) collaborative group have developed helpful guidance for assessing the likelihood of measurement reactivity.3 For example, the likelihood of measurement reactivity is high when study participants are aware of the outcome of interest, the outcome is a health-promoting behaviour, and the socially desirable direction of the outcome is well-recognized. All these points generally apply to the measurement of PA. Moreover, self-report measures are more prone to inducing measurement reactivity. Of note, while in the study by Seymore et al., all studies that showed noteworthy improvements in control group PA levels employed self-report measures of PA, other literature describes reactivity also to automated PA measurement devices such as accelerometry.4 If trialists identify ‘red flags’ for measurement reactivity, they can consider collecting data to examine the likelihood and magnitude of this bias specific to the context of their research. For example, qualitative data can elucidate study participants’ thinking and emotional and behavioural responses to the measurement. More complex longitudinal study designs, such as the Solomon four-group design cited by Seymore et al., offer possibilities to quantify bias related to measurement reactivity. Lastly, if the expected likelihood and magnitude of bias threatens the validity of a study, trialists should incorporate design features for minimizing measurement reactivity. Masking or concealment of PA measurements are strong design choices to guard against measurement reactivity, but these are only possible for automated tracking devices such as accelerometers, pedometers, or smartphones, and not for self-report measures of PA. In contrast to unwanted bias, measurement reactivity may also present a welcome opportunity, when it leads to a desirable health behaviour change such as increased PA levels. This could occur because the individual who is being measured reduces or resolves the discrepancy between actual and desired behaviour through self-regulation.1 Physical activity measurements could therefore be intentionally leveraged to nudge patients towards healthier behaviour by aligning with social norms and the human desire to be viewed positively by others—such as being seen as responsible and competent—especially among individuals who are less active and sensitive to the opinions of others.5 This effect can be further enhanced by healthcare professionals through the use of contextual factors—such as fostering positive expectations, providing support, and offering reassurance—to positively influence health outcomes.6 The review by Seymore et al. further highlights the value of prompts and cues, showing that printed PA information and brief PA advice were associated with improvements in PA in the control groups of up to half of the reviewed studies. This measurement reactivity is likely driven by increased self-awareness and reflection7 and could be integrated into healthcare through regular reminders or feedback, even in a digital format and independent of clinical appointments. This may help shape patients’ attitudes towards PA, increase the salience of health behaviours, and facilitate behaviour together with intention formation.8,9 Encouraging patients to report PA data frequently, and explaining the benefits of doing so, may improve both data quality and patient outcomes. Regular self-monitoring not only increases awareness but can also enhance self-efficacy, a core component of perceived behavioural control that can be crucial for maintaining behavioural intentions over time and for behaviour change.10 This may be further enhanced by shifting from solely tracking PA to empowering patients to understand and effectively use these data, while making patients accountable by sharing the data with healthcare professionals. Such external monitoring can create teachable moments, helping patients to learn how to track their activity and to understand the relevance of these data.11 Together, these psychological mechanisms that centre around measurement reactivity may serve as a behaviour change intervention. It is recommended that researchers make transparent the conceptualization and the proposed mechanisms of action of such interventions, in particular through the description of logic models.12 Reactivity to PA measurements in the secondary prevention of CVD might lead to unwanted measurement bias, or it may be leveraged to support positive behaviour change. The systematic review by Seymore et al. provides a helpful evidence base for researchers to reflect on either scenario in the context of their own research. Stefan Tino Kulnik (Conceptualization [equal], Writing—original draft [equal], Writing—review & editing [equal]), Dirk Leysen (Conceptualization [equal], Writing—original draft [supporting], Writing—review & editing [equal]), and Franziska Pfannerstill (Conceptualization [equal], Writing—original draft [equal], Writing—review & editing [equal]) None. Not applicable.