Interpersonal touch interventions for patients in intensive care: A design‐oriented realist review

Abstract Aim To develop a theoretical framework to inform the design of interpersonal touch interventions intended to reduce stress in adult intensive care unit patients. Design Realist review with an intervention design‐oriented approach. Methods We searched CINAHL, MEDLINE, EMBASE, CENTRAL, Web of Science and grey literature sources without date restrictions. Subject experts suggested additional articles. Evidence synthesis drew on diverse sources of literature and was conducted iteratively with theory testing. We consulted stakeholders to focus the review. We performed systematic searches to corroborate our developing theoretical framework. Results We present a theoretical framework based around six intervention construction principles. Theory testing provided some evidence in favour of treatment repetition, dynamic over static touch and lightening sedation. A lack of empirical evidence was identified for construction principles relating to intensity and positive/negative evaluation of emotional experience, moderate pressure touch for sedated patients and intervention delivery by relatives versus healthcare practitioners.

Routine nursing and medical interventions often involve procedural touch, which patients may find unpleasant (Samuelson, 2011).
In contrast, interpersonal touch interventions, such as affective touch and massage, are aimed at improving a patient's psychological state. Further, in the current context of light sedation targets for ICU patients, the role of human presence in reducing fear and anxiety is increasingly recognized as important (Baumgarten & Poulsen, 2015).
Such variations may result from differences in intervention characteristics or study methodology. Importantly, outcomes may also vary depending on context because contextual factors can either activate or block the underlying mechanisms (Ellingsen, Leknes, Løseth, Wessberg, & Olausson, 2016). Thus, while meta-analyses are useful for estimating aggregate effectiveness, the more important task is to understand how interventions work (Chen & Rossi, 1983).
Recent reviews (Boitor et al., 2017;Hu et al., 2015;Miozzo et al., 2016;Papathanassoglou & Mpouzika, 2012) have described studies of interpersonal touch interventions in adult ICU as being generally small-scale randomized trials (group sizes <50), of variable methodological quality. To date, attention has focused more on massage interventions compared with acupressure, reflexology or handholding interventions. Study comparability remains problematic because of inadequate reporting for intervention characteristics such as pressure and velocity (Lindgren et al., 2013). In terms of outcome parameters, few ICU studies have investigated neuroendocrine effects and none have employed neuroimaging techniques.
The aim of this review was to understand how interpersonal touch interventions modulate stress and related outcomes in ICU patients and to develop a theoretical framework to inform the design, implementation and evaluation of interpersonal touch interventions.

| DE S I G N
We used realist review methodology (Pawson, 2002;Saul, Willis, Bitz, & Best, 2013; and followed RAMESES reporting standards (Wong, Greenhalgh, Westhorp, Buckingham, & Pawson, 2013). Realist review is an approach to building and testing conceptual frameworks that can inform intervention design (Fletcher et al., 2016;Pawson, Greenhalgh, Harvey, & Walshe, 2005). In contrast to outcome-driven approaches, which conceptualize interventions as black boxes (Astbury & Leeuw, 2010), realism recognizes the complexity of interventions (Pawson, 2013). Thus, realists seek to answer all or part of the question "What works, how, why, for whom, to what extent and in what circumstances, in what respect and over what duration?" , p. 1011. Unlike conventional systematic reviews, realist reviews follow a more iterative and idiosyncratic route. Thus, while the theory testing process should be systematic and transparent, creativity and judgement are largely prioritized over reproducibility and uniformity (Pawson, Greenhalgh, Harvey, & Walshe, 2004).
Additionally, we took a design-oriented approach, using "context-intervention-mechanism-outcome logic" (CIMO-logic) (Denyer, Tranfield, & van Aken, 2008;Pawson & Tilley, 1997) to gain an understanding of how different types of interventions might work best in different contexts. We considered that an intervention design approach was appropriate to the subject area because the paucity of empirical evidence from ICU studies suggested a need to transcend existing systems to create new "design propositions" (Romme, 2003), that is CIMO configurations (CIMOCs). Design propositions are depicted as follows: in context C, use intervention type I, to activate mechanism M, to achieve outcome O. Further, we created "construction principles" comprising interlinked sets of CIMOCs (Romme & Endenburg, 2006). Construction principles were framed as broad solutions to the problem of reducing patient stress.
In line with the sensory and social dimensions of interpersonal touch, our definition of "mechanism" was broader than definitions employed for exclusively social interventions (cf. Lacouture, Breton, Guichard, & Ridde, 2015); we defined mechanism as a generally hidden, context-sensitive, physiological or psychological response of an individual to the intervention that leads directly or indirectly to an outcome of interest. We defined "context" as any feature distinct from the intervention per se that acts on a mechanism to influence outcomes.

| Focusing the review
We focused the review based on: data limitations suggested by a preliminary scoping review; issues relating to intervention feasibility, acceptability and safety; and insights gained during the theory testing process. We consulted local stakeholders (four ICU nurses (colleagues of SJH) and two patient representative groups) to ensure that the focus of the review took into account the concerns of potential knowledge users. Stakeholder consultations took place by email and/or in person. Further, to ensure that relevant theories were identified and given appropriate consideration and that data interpretation was congruent with current knowledge in the field, we established correspondence with a panel of four external experts.
Experts were selected on the basis of recent publications (since 2012) on interpersonal touch in critical care or complex critical care interventions (one existing contact of SJH).
While the benefits of touch extend to touch interventionists such as family members (Prichard & Newcomb, 2015), who may be well placed to deliver touch interventions (Hill, 1993), we restricted the review to patient outcomes following stakeholder opinion that relatives would be more motivated to use the intervention on the basis of evidence supporting patient benefits. We chose not to focus on intervention duration because we considered that short interventions (e.g., 10 min) would be preferable to minimize clinical interruptions (Martorella, Boitor, Michaud, & Gélinas, 2014).
Additionally, while recognizing that light and moderate pressure touch activate different neurophysiological pathways , because of stakeholder concerns about safety, we chose not focus on pressure because forceful massage techniques are associated more frequently with serious, although rare, adverse events (Ernst, 2003;Posadzki & Ernst, 2013). To a large extent however, the review's focus was restricted by the data limitations suggested by the scoping review. For example, we identified limited reporting of interventionists' use of eye contact and facial expression, which may constitute influential contextual cues (Ellingsen et al., 2016;Kerr, Wiechula, Feo, Schultz, & Kitson, 2016).

| Search methods
We employed a two-stage search process consisting of a broad, scoping search and a systematic search to corroborate our developing theoretical framework . Searches were completed by SJH.

| Document selection
The first author (SJH) screened titles and where indicated, abstracts or full texts of documents against eligibility criteria (Supporting information Appendix S2). We included a broad range of outcomes, including pain, in recognition of the close interactions that may exist between stress and other distinct outcomes (McCracken, Zayfert, & Gross, 1992). Records for the same study were identified as "sibling papers" to create "study clusters" (Booth et al., 2013). Resources were not available for translation; therefore, we included translations only where provided. Additionally, we included English language abstracts of non-English language sibling papers as sources of contextual information.

| Quality appraisal
All eligible studies were appraised for quality and relevance by SJH.
The Mixed Methods Appraisal Tool (MMAT; Pluye et al., 2011;Souto et al., 2015) was used to assess study quality. In accordance with realist methodology, studies were not excluded on the basis of MMAT scores; rather, case-by-case decisions were taken as to whether data were of sufficient quality and relevance to warrant some contribution towards theory development (Pawson, 2006). Extracted data were judged of insufficient quality if they were considered to be seriously untrustworthy.

| Data abstraction
Data abstraction was undertaken by SJH using a data extraction form (template provided in Supporting information Appendix S3).

Data extraction form design was informed by Cochrane Skin Group
(2009) and Higgins and Deeks (2008) and adapted for the review following a trial using five studies. Outcome data were extracted only if reported in full-text English language papers.

| Analysis
Through a process of scoping and focusing with stakeholder consulta- within-study and between-study comparisons. Due to high study heterogeneity, study selection criteria were developed iteratively, rather than being set in advance. Where studies compared identical interventions with and without essential oils, essential oil comparators were excluded. The selected data were interpreted by SJH, EDEP and LL.
Insights gained during theory testing directed our continued search for explanatory theory and led to two further construction principles that we considered untestable within the limitations of our matrices.

| Ethics
Patient consent and ethical approval were not required.

| Search outcome
Document flow processes for our main and supplementary systematic searches are presented in Figure 1 and Supporting information

| Theoretical Framework
Our theoretical framework ( i.e. touch involving motion (e.g., massage) or no motion; and moderate vs. light pressure); three contextual factors (sedation level, the patient-interventionist relationship and the patient's previous experience of the intervention); and one mechanism (subjective intensity and positive/negative evaluation of emotional experience or affect).
Intervention mechanisms and emergent processes encompassed the neurophysiology of touch and pain, psychological factors and interpersonal factors.

| Evidence for intervention Construction Principles
The following sections present evidence for construction principles 1-4 (Table 1) using data from studies identified in our systematic search. Additionally, where possible, we consider evidence relating to interactions between specific CIMO components. However, a consideration of all details covered by our theoretical framework was beyond the scope of this review.

| Construction Principle 1: Dynamic touch may reduce stress and pain more effectively than static touch
Two studies investigated the effects of dynamic touch versus predominantly static touch ( Table 2). As can be seen from Table 2, reported results for these studies were entirely (Tsay, Wang, Lin, & Chung, 2005) or predominantly (Boitor, Martorella, Arbour, Michaud, & Gélinas, 2015;Martorella et al., 2014) in favour of dynamic touch, thus supporting our hypothesis.
In considering the mechanisms underlying the effects of dynamic touch, we now focus on quantitative and qualitative data reported by Boitor et al. (2015) and Martorella et al. (2014) in their study of hand massage versus handholding for postoperative cardiac surgery patients. To explain the potential analgesic effects of hand massage, Boitor and colleagues refer to the mechanism of ascending inhibition of nociceptive signalling in the spinal cord via the stimulation of large-diameter Aβ mechanoreceptive afferents (Melzack & Wall, 1965). An important condition, however, for activation of the ascending spinal gating mechanism, is that a connection must exist between the spinal nerve transmitting the nociceptive input and the spinal nerve transmitting the tactile input. Furthermore, contrary to R. Melzack, personal communication, November 20, 2012, cited in Hogan et al. (2014, recent work by Mancini, Nash, Iannetti, and Haggard (2014) suggests that within the sensory territory innervated by the relevant spinal nerve, the proximity of tactile input to the site of injury is an important factor.
Thus, activation of the ascending spinal gating mechanism in Boitor et al.'s study is likely to have been limited because the massage was not directed towards the patients' painful thoracic surgical site.
Additionally, mechanism activation is likely to have been restricted further by the limited overlap between dermatomes stimulated by the hand massage (C6-C8 and T1) and dermatomes proximal to the patients' sternal incision (e.g., C4 and T2-T8) (Ladak, Tubbs, & Spinner, 2014;Lee, McPhee, & Stringer, 2008). We are, however, unable to fully exclude this mechanism because of the possibility of interneural communication between peripheral nerve territories (Ladak et al., 2014), variation in specific nerve territories between individuals and major discrepancies between dermatome maps (Downs & Laporte, 2011). Interestingly, some study participants stated they would have preferred the massage had targeted areas they identified as painful (Martorella et al.), which would then have activated the ascending inhibitory pathway.
As an alternative to the ascending spinal gating mechanism described above, we suggest that supraspinal mechanisms

| Construction Principle 2: Lightening sedation may promote touch-mediated reductions in stress
To investigate the effect of lightening sedation on the effects of touch, we compared outcomes from two studies specifying that

TA B L E 1 Construction principles and theoretical framework for interpersonal touch interventions
Modifying contexts

Emergent processes
Construction Principle 1. Dynamic touch may reduce stress and pain more effectively than static touch Proximity to nociceptive input Ascending inhibition of pain signals at the neural gate in the spinal cord. Inhibition  (2015), Melzack and Wall (1965), Watanabe, Piché, and Hotta (2015) Reduced pain signalling to the brain, inhibition of somatocardiac reflexes Positive interactions between pain perception, health-promoting behaviours and stress regulation Eye contact Supraspinal mechanisms, including sensory, cognitive & affective processes, modulate pain transmission (Melzack & Wall, 1965) & subjective pain experience (Bushnell et al., 2013;Melzack, 2001;Melzack & Katz, 2013) Reduced pain signalling, reduced perceptions of physical and psychological pain. Improved stress regulation and action programmes Neural activity and connectivity in the reward system The reward system, which comprises cortical & subcortical brain regions, is activated more strongly by gentle stroking movements versus static touch. Lindgren et al.. (2012) Increased reward processing Calmness of environment Reward reduces stress reactivity via endogenous opioid release (Creswell, Pacillio, Denson, & Satyshur, 2013;Kaada & Torsteinbø, 1989 (Diener, Colvin, et al., 1991;Human et al., 2015;Parducci, 1968Parducci, , 1984Pressman & Cohen, 2005;Solomon, 1980) & higher relative frequency of PA versus NA (Blevins et al., 2017;Diener, Sandvik, et al., 1991) Reduced stress response, improved stress resilience and recovery. Positive effects on social support and health behaviours TA B L E 1 (Continued) (Continues) patients received sedatives with six studies where sedation was specified as restricted (Table 3). Both sedation condition studies investigated non-coma patients who predominantly received mechanical ventilation; we therefore restricted our comparison to studies investigating similar populations. The pattern of outcomes obtained from the between-study comparison ( Pre-stimulus heart rate Mechanical pressure stimulation of mechanoreceptors in skeletal muscle elicits a reflexive autonomic nervous system response. Watanabe and Hotta (2017) Modulation of heart rate and blood pressure Interoceptive sensitivity Arterial baroreceptors transmit information about cardiovascular arousal to brain regions implicated in affective & cognitive processing. Garfinkel and Critchley (2016) Reduced fear and anxiety, generally enhanced perception and cognition Note. OT: oxytocin; HPA: hypothalamic-pituitary-adrenal; PA: positive affect; NA: negative affect. Linkages between context-intervention-mechanism-outcome components occur both within and across construction principles. For simplicity, we consider only positive effects. The exclusion of mechanisms relating to negative effects does not imply any hypotheses regarding the importance of negative effects.
a Mechanisms may also be applicable to repetition of massage strokes.

TA B L E 1 (Continued) numbers of interventionists per group (one nurse vs. 30 companions)
may have contributed to the higher post-treatment variability in cortisol levels for the companion group.
In terms of potential underlying mechanisms, given that the beneficial effects of touch may depend on relationship quality (Coan, Schaefer, & Davidson, 2006), it is possible that the inter- F I G U R E 2 Logic models for interpersonal touch interventions in ICU. The models are based on our interpretation and synthesis of the evidence sources informing our theoretical framework. For simplicity, we consider only positive effects and do not present all contextintervention-mechanism-outcome configurations. The exclusion of mechanisms relating to negative effects does not imply any hypotheses regarding the importance of negative effects. (a) Dynamic touch. (b) The figure illustrates the direct effects of moderate pressure on the autonomic nervous system versus the more indirect effects of light pressure CT optimal touch that are more reliant on cortical processes. CT: C-tactile afferents (present in hairy skin only); PA: positive affect. CT optimal stimuli: indentation force 0.3-2.5 mN, velocity 1-10 cm/s, warm (typical skin) temperature (Vallbo, Löken, & Wessberg, 2016

| Construction Principle 4: Treatment repetition may provide cumulative benefits
To investigate the effects of treatment repetition, we present quantitative evidence from 11 study groups (Table 4). As can be seen from We now focus on the proposed mechanism of pleasure, using qualitative findings from two contrasting studies: Henricson et al. Percentages are based on results reported for each treatment, with the exception of Korhan et al. (2014),  and Yousefi, Naderi, Daryabeigi, and Tajmiri (2015) who report mean results for twice daily treatments.
a Differences between intervention and control groups either not reported or not statistically significant (p>0.05). b In favour of the intervention group for ≤70% of treatments (p <0.05 influence emotional processing and expression (Hofstede, 2011), we tentatively suggest that the more intense levels of PA reported by Henricson et al. may, at  Theoretical and empirical evidence suggests that infrequent intense PA may incur emotional costs to ICU patients for several reasons (Diener, Colvin, Pavot, & Allman, 1991;Diener, Sandvik, & Pavot, 1991). Firstly, as identified by Henricson et al. (2009), affective contrast may cause an unpleasant affective state on returning to normality; secondly, since infrequent events are likely to produce less hedonic habituation, extremes of emotion are likely to persist (Solomon, 1980); and thirdly, since the value of an event depends on comparisons with other events (Parducci, 1968(Parducci, , 1984, an ICU patient's negative situation may enhance both the pleasure and associated psychological costs of intense PA. Additionally, high PA states can trigger short-term increases in physiological arousal that may be potentially harmful, particularly in individuals at risk of acute health events (Pressman & Cohen, 2005 (Human et al., 2015) and may result in less positive psychological and health-related outcomes.
Additionally, given that ICU patients experience high levels of uncertainty (Egerod et al., 2015) and subjective uncertainty is a defining characteristic of stress (Peters, McEwen, & Friston, 2017), the provision of frequent positive events might reduce stress by virtue of reducing environmental uncertainty. Moreover, more frequent and predicable interventions might promote anticipatory pleasure.
Finally, it is possible that while moderate PA mitigates against the distress associated with high-intensity PA, it may elicit a more positive, adaptive stress response (Selye, 1974) than low-intensity PA. Thus, as suggested by Pressman and Cohen (2005), moderate PA may provide greater long-term health benefits than low-or highintensity PA. We therefore speculate that a polynomial relationship Repetition effects suggested, supported by statistically significant differences (p < 0.05) between intervention and comparator groups for ≤70% of treatments. c Repetition effects supported by statistically significant differences (p < 0.05) between intervention and comparator groups for>70% of treatments.
TA B L E 4 (Continued) may exist between the immediate effects of treatment on PA and the long-term effects of multiple treatments.

| Additional Insights
In the light of evidence that cortical processing may be a key mechanism underlying the benefits of gentle, medium-velocity touch (Table 1, Figure 2b), gentle touch may have limited effectiveness for sedated patients due to reduced corticocortical and subcorticocortical connectivity (MacDonald, Naci, MacDonald, & Owen, 2015). In contrast, moderate pressure touch is suggested to elicit a relaxation response by increasing parasympathetic activity and/or reducing sympathetic activity (Field, 2016b;Watanabe & Hotta, 2017

| D ISCUSS I ON
To our knowledge, this is the first realist inquiry into interpersonal touch interventions in ICU. Unsurprisingly, given that the current state of knowledge remains in its infancy, empirical evidence for our construction principles was weak (Principles 1, 2 and 4), unsupportive (Principle 3) or unavailable (Principles 5 and 6). Furthermore, we were unable to link outcomes to specific mechanisms. Nevertheless, we believe our review has produced insights into how interpersonal touch interventions might work in the ICU context. These insights would not have been possible within the confines of a traditional outcome-driven systematic review.
We found some evidence that for dynamic touch, mechanisms other than pleasure, such as distraction, may be more important in achieving supraspinal pain inhibition. We also highlighted the importance of considering the proximity of tactile input to nociceptive input in activating the ascending inhibitory pathway proposed by the gate control theory of pain. We found some evidence that sedation inhibits the effects of touch. We found weak evidence supporting the role of OT in treatment repetition. By comparing qualitative findings of two contrasting studies, we gained insights into the potential emotional costs patients might incur from infrequent episodes of high-intensity PA. Further, we speculated that a polynomial relationship might exist between the immediate treatment effects and the long-term effects of multiple treatments. Finally, we hypothesized that sedated patients might benefit preferentially from moderate rather than light pressure touch.
In conjunction with our intervention design approach, CIMOlogic provided a useful, albeit circuitous route to theory development; having initially chosen not to focus on touch pressure because of stakeholder concerns about safety, later insights led us to reappraise the potential benefits of moderate pressure touch in the context of sedation. Additionally, our broad definition of mechanism usefully enabled us to envision reality acting across multiple levels, from biophysical to social (Bhaskar, 1986).
One of the main strengths of this review is that our theoretical framework is built on relevant research evidence (Fildes, 1985), as well as transferable mid-range theories and neurophysiological mechanisms (cf. McConnell & Porter, 2016). Consultation with stakeholders ensured that the concerns of potential knowledge users were influential in focusing the review. Additionally, we attempted to minimize publication bias by employing a wide systematic search strategy, which encompassed grey literature.
We acknowledge that this review has presented only a partial description of how interpersonal touch interventions might work in an ICU setting. For example, we did not consider structural factors (McConnell & Porter, 2016) or benefits to touch interventionists (Prichard & Newcomb, 2015;Wilson, Gettel, Walsh, & Esquenazi, 2016). Moreover, by focusing on the positive effects of interpersonal touch, we have elided potentially important negative effects.
For example, touch may exacerbate the symptoms of patients who have experienced trauma or abuse (Benjamin & Sohnen-Moe, 2014;Phelan, 2009), light touch may elicit a sympathetic nervous system (i.e. pro-stress) response (Diego & Field, 2009) and, in certain contexts, OT can elicit antisocial rather than prosocial effects (Piva & Chang, 2018). Also, there may be alternative physiological or psychological explanations for the positive effects we investigated. For example, Moyer, Rounds, and Hannum (2004) suggest that the delayed analgesic effects of multidose massage interventions may result indirectly via the facilitation of deep sleep, which is proposed to inhibit release of the pain promoting peptide, substance P (Field, 2016a).
However, given the differences between analgosedation and normal sleep (Delaney, Van Haren, & Lopez, 2015), this effect may be less important in an ICU context. In considering our systematic review, this was restricted by the small number of ICU studies, high study heterogeneity, limited reporting of contextual and intervention characteristics, uncertain validity of surrogate outcome measures (Everly & Lating, 2013) and variable study quality. Few studies used qualitative designs and none used mixed method designs. Rather, most studies were RCTs that included a standard care control condition, which may have biased results due to the control groups' potential disappointment at their allocation status (Lindström, Sundberg-Petersson, Adami, & Tönnesen, 2010;Stevensen, 1994).
Due to resource constraints, a single individual completed the search and quality appraisal processes, which may have increased the risk of bias. Generalizability of results may be limited by the historical context where the two sedation condition studies were conducted (pre-2007); given recent trends towards lighter sedation targets (Shehabi, Bellomo, Mehta, Riker, & Takala, 2013), study participants may have been over-sedated relative to current sedation practices.

| CON CLUS ION
Interpersonal touch interventions have the potential to reduce psychological and physiological stress in ICU patients. This review provides insights into how interpersonal touch interventions may more likely achieve their aims if intervention designs are informed by an understanding of the underlying generative mechanisms and the key contextual factors that activate those mechanisms. Moreover, we have described how specific types of touch interventions may be more effective in specific contexts. We have also highlighted the potential complexities of temporal effects associated with treatment repetition by identified that polynomial relationships might exist between short-term and long-term outcomes. While substantial gaps in the ICU literature limited our ability to fully evaluate our theoretical framework, we have outlined novel construction principles and design propositions that can be tested and refined in future studies.
In addition, our theoretical framework provides guidance for nurses and other members of the multidisciplinary team wishing to support the use of interpersonal touch in practice.

CO N FLI C T O F I NTE R E S T
No conflict of interest has been declared by the authors.