Chapter 2: Coordinating community stroke services
2.8 Randomised Controlled Trials and complex interventions
As discussed in chapter one (section 1.7), a complex intervention in health services can be defined as ‘interventions that are not drugs or surgical procedures, but have many potential ‘active ingredients’. A complex intervention combines different components in a whole that is more than the sum of its parts’ (Oakley et al., 2006). The LoTS care system of care falls within this catergory, as do many other interventions e.g. the Stroke Liaison Worker. Randomised Controlled Trials of the Stroke Liaison Worker have resulted in inconclusive and sometimes contradictory findings (Forster et al., 2009, Forster and Young, 1996, Friedland, 1992, Allen et al., 2009, Dennis et al., 1997, Lincoln et al., 2003, Mant et al., 2000). More recently the results of the LoTS care trial found no significant difference between the intervention and control group in the outcomes of interest including the GHQ-12, the BI, the FAI and the LUNS. Explanations to account for these results include poorly defined interventions and the use of inappropriate outcome measures, the use of non-stroke specific outcomes, and the reliance on other community services (Ellis et al., 2010, Dennis et al., 1997, Boter and for the HESTIA Study Group, 2004, McKevitt et al., 2004, Tilling, 2005).
Explaining the results of clinical trials (success or failure) can be problematic, as the trial design prioritises outcome rather than process measures; therefore the intervention often remains a ‘black box’ (Rychetnik et al., 2002, Oakley et al., 2006, Stame, 2004, Pope and Mays, 1993). For this reason, if successful, the parts of the intervention that need to be replicated to produce a similar outcome remain unclear (Pope and Mays, 1993). An example often cited is that of the stroke unit, a complex intervention shown to be effective in reducing death and disability for all stroke survivors in clinical trials and subsequent meta-analyses (Stroke Unit Trialists' Collaboration, 2007). However, the active ingredients that worked to produce these outcomes have not been definitely established (Langhorne et al., 2002, Whyte and Hart, 2003). The literature on specific rehabilitative interventions, reviewed in chapter one, further emphasises this point. Most reviews highlighted a need to define the content, duration, frequency and timing of the intervention to understand what has produced the benefits identified. It is argued that the strengths of the trial design are particular to the evaluation of certain types of intervention, Berwick (2008) commented:
To study a linear, mechanical or natural tightly coupled causal relationship most efficiently an OXO design (such as an RCT) may be exactly correct. But with social changes, multicomponent interventions some of which are interpersonal all of which are non-linear, in complex social systems then other richer but equally disciplined ways are needed. (Berwick, 2008)
The ‘OXO design’, to which Berwick refers, describes the process of observe (O), introduce an intervention (X) and observe again (O) (Berwick, 2008), i.e. the ‘successionist’ view of causality described in chapter one (section 1.5) (Pawson, 2008). For example, in the Chinese Acute Stroke Trial (CAST), individuals were allocated to either the intervention (received aspirin) or to the control group (do not receive aspirin). The outcome observed (O) was recurrent stroke and this was significantly reduced in the intervention group i.e. aspirin worked to reduce the risk of recurrent stroke. Aspirin worked because of its affect on the blood platelets within the body. The mechanism of action was, therefore, physiological and more easily described than most at work in more complex rehabilitative interventions (Whyte and Hart, 2003). In comparison to aspirin, health care professionals were expected to interpret and implement the components of the system of care before it could impact on the patient outcomes observed in the LoTS care trial. The mechanisms of action assumed to be at work, as in many other complex interventions, were therefore social in nature and effectiveness depended on the ability to change human behaviour (Davidoff, 2009).
If the mechanisms of action are unknown, selecting an appropriate outcome to measure effectiveness becomes problematic. For example, in the meta-analyses of the Stroke Liaison Worker service, reviewed in chapter one, Ellis (2008) described that the interventions were developed on a pragmatic and intuitive basis and therefore lacked a clear underlying mechanism of action. For this reason, it was unclear which outcome, from the numerous targeted, would be effected most by the multifaceted intervention (Ellis, 2008). Defining a good outcome is also dependent on the perspective used e.g. commissioners might prioritise service costs, physiotherapists walking speed, whereas stroke survivors might prioritise their pre- stroke level of activities (McKevitt et al., 2004). Qualitative studies have indicated
that the benefits of receiving the Stroke Liaison Worker service might be less tangible i.e. not easily quantified using standardised measures prioritised by the trial design (Lilley et al., 2003, Dowswell et al., 1997).
In the evaluation of complex interventions the links between the intervention and the outcome of interest are not always apparent and usually form part of a much longer chain than those observed when evaluating less complex interventions. Further to this, the context in which complex interventions are implemented are also characterised as complex adaptive systems (Begun et al., 2003). Distinctive features of such systems include feedback loops that crucially shape how change occurs; behaviour, which emerges unpredictably from the interaction of the parts, and the system’s ability to adapt through learnt experience (Begun et al., 2003). Community health services (the context for the LoTS care trial) could be described as complex adaptive systems, they change and adapt over time in response to national and local drivers of change. Therefore, contextual circumstances might also impact on the outcomes observed and need to be considered in the evaluation process. Instead of imagining the trial design as a comparison of 1) a site with intervention on and 2) to an identical site with intervention off, it has been suggested that a more appropriate comparison might be to consider that of, 1) a complex adaptive system thrust into a complex adaptive system and 2) another complex adaptive system (Pawson et al., 2004).