Interventions to reduce readmissions: can complex adaptive system theory explain the heterogeneity in effectiveness? A systematic review

Successfully transitioning patients from the hospital to home is a complex task. As hospital readmission rates have been adopted as markers of quality with financial penalties for poor performance, care transition improvement efforts have multiplied.

Over the last 30 years, a variety of care transition interventions have been implemented, including those with differing types and numbers of components; some have proven effective but many have not [1‐3]. There is a lack of information about which components or combinations of components are critical to reducing early (30 day) hospital readmissions [4, 5]. We believe that one source of this variability in effectiveness is due to a mismatch between the inherent complexity of care transitions and the health care organizations in which they occur, and the types of interventions being attempted.

Health care organizations have been studied as complex adaptive systems (CAS) for many years [6‐8], contributing insights about the importance of interdependencies among different parts of the system, the role of learning, and the impact of local interactions in self-organization (stable but dynamic patterns of interaction among system components), emergence (system level behaviors that cannot be explained by examining the individual components of the system), and co-evolution (patterned changes that take place as a system and its environment react to each other over time) [9]. Using the lens of CAS provides a theoretical approach for describing and understanding care transition interventions.

Effective care transitions from inpatient to outpatient require multiple diverse individuals (providers [inpatient and outpatient, multiple professions, multiple specialties], patients and caregivers) to interact, share their expertise, learn together, and depend on each other to make sense of what is happening in real time to prevent rehospitalization. CAS theory helps us understand both the complexity of care transitions and the difficulty in predicting which processes or approaches will most likely contribute to success. There is a high degree of uncertainty in predicting a patient’s trajectory of recovery after hospitalization, and therefore their specific transitional needs [10].

CAS theory brings focus to the inherent uncertainty in many clinical tasks, and in clinical systems [10]. CAS theory also suggests that for interventions to be successful, they must be congruent with the uncertainties and contextual differences that people face when they deliver care. For example, interventions that rely on a strict application of a new process from one clinical setting to another may not be as successful as one that allows for local adaptation that reflects differences in the ways that people organize themselves across systems. Interventions that foster interconnections between individuals may be more likely to improve communication and support local potential for co-evolution [11], thus increasing implementation success. Moreover, in complex interventions like those required for improving care transitions, the line between intervention and implementation is often blurred [12], as activities aimed at increasing coordination between inpatient and outpatient care teams (intervention strategy) may also function to educate providers about local care transitions resources (implementation strategy) [13].

This paper is the third in a series of articles examining the association between the degree to which interventions to improve complex care delivery activities are consistent with CAS and their degree of effectiveness. In the first of these papers [14], we found type II diabetes intervention effectiveness was positively associated with (a) the number of CAS characteristics inherent in intervention designs and (b) the degree to which the intervention impacted CAS characteristics of interconnections between individuals, and the degree to which the intervention allowed for evolution in its implementation over time. In the second paper [15], congestive heart failure intervention effectiveness was related to (a) the number of CAS characteristics present in the intervention design and (b) the CAS characteristics of learning, self-organization, and co-evolution, again suggesting the need for interventions to adapt and evolve with local patterns of care organization. We hypothesized that observed differences across our two analyses were due to differences in the ways that diabetes and heart failure lead to uncertainty for patients in their daily lives. These differences require emphasis on different CAS characteristics for successful interventions.

This paper extends this work by examining care transition interventions that move patients across complex microsystems from hospital to home. Such transitions are characterized by uncertainty and dynamic change, and frequently involve changes in patients’ functional capacities, requiring them to learn new acute and chronic self-care practices. These transitions also require successful transfer of information between medical providers and between medical providers and informal caregivers.

Our objective is to explore whether care transition intervention designs that reflect the complexity of processes and conditions they are trying to improve are associated with better outcomes. We also aim to identify key characteristics of interventions that are related to better hospital readmissions outcomes. We hypothesize that care transition interventions with elements that are congruent with the nature of CAS will be more likely to have improved readmission outcomes. We do not suggest that interventions were developed with CAS in mind, but that they vary in their degree of congruence with important CAS principles.

Because of the extensive literature on care transitions, we used Kansagara et al.’s report [4], “Transitions of Care from Hospital to Home” as our starting point. This report was prepared for the Department of Veterans Affairs Evidence-based Synthesis Program and represented a comprehensive assessment of the literature on care transitions.

Kansagara et al. identified systematic reviews of care transitions, screening them for quality (e.g., the review clearly reported their search criteria, evaluated the internal validity of included trials). Eighty-three systematic reviews met their initial criteria. They further narrowed this group to 17 of the most recent and broadly scoped systematic reviews that fit pre-identified categories. Ten of those systematic reviews focused on different types of care transition interventions (e.g., enhanced access to primary care, telephone-based follow-up) and 7 focused on specific patient populations. All included reviews contained hospital readmissions as a reported outcome, though it may not have been the primary outcome and not all studies included in the reviews reported hospital readmissions as an outcome [4]. This approach provided us a high quality, diverse set of care transition intervention types upon which to test our research question.

We assessed whether care transition interventions whose design was congruent with CAS characteristics were more likely to report reduction in early readmissions. We found that nearly all the interventions assessed contained elements reflecting the 3 CAS characteristics of interconnections, affordances and processes, and none of the interventions displayed coevolution or emergence. The CAS characteristics that were most inconsistently reflected in our group of included studies were learning and self-organization. When both of these characteristics were present, interventions were associated with significantly fewer readmissions.

These results complement those reported by Leppin et al.’s work, in which early hospital readmission interventions were analyzed using the cumulative complexity model framework [3]. Their review included 42 trials compared to our 46, nine of which we shared in common. At least five of their included trials would have been excluded from ours because they were conducted on surgical patients. Leppin et al.’s analysis identified that interventions that were more complex (defined as involving more people or components), and those that increased patient capacity for self-care, were more likely to be effective. While that framework appears to share some similarities to CAS (e.g. attention to feedback loops, emergence), it focuses on workload-capacity imbalances and patient complexity, and considers intervention complexity in terms of component parts rather than their impact on characteristics of the complex systems in which the interventions are applied [72]. Whereas Leppin et al’s work was directed toward the intervention components, our CAS-grounded work was more focused on the interdependencies among the components in the systems of intervention. It is possible that their findings regarding self-care are related to learning, and that interventions with more components might be more likely to change interdependencies or self-organization. Future work might look at how the two approaches could be synthesized to enhance our understanding of applying complex interventions in complex adaptive systems.

Our findings have some important differences from our previous work examining the relationships between interventions’ congruence with CAS characteristics and outcomes in two complex chronic diseases: diabetes and chronic heart failure. In the case of type 2 diabetes interventions [14], interconnections and co-evolution were positively associated with outcomes. In our present analysis, 96% of the interventions changed interconnections, typically through approaches such as post-discharge calls or novel handoff practices. The fact that most of the interventions were designed to change the way people interacted with each other but were associated with varied results suggests that interconnections alone were not sufficient for improving outcomes. Additionally, in this group of studies, most of the changes in interaction patterns focused on providers (e.g., patient-provider, provider-provider). Changing other types of interconnections (e.g. between patients and their family caregivers) might influence different aspects of care delivery that is highly dependent on patient and family actions.

Similar to our current findings, in our review of chronic heart failure interventions [15], learning, self-organization, and co-evolution were significantly related to intervention effectiveness. In our analysis of care transition interventions, learning alone was not significantly associated with effective outcomes; when combined with self-organization, its association on readmissions was significant. This finding suggests that learning might be necessary to consider when trying to decrease readmissions, but perhaps is not sufficient. Returning home after a hospital stay involves many life changes, not all which can be anticipated and prepared for through patient education. As in the case of interconnections, different types of learning might also need to be distinguished to determine which approaches are more effective.

Care transitions may lead to a higher degree of uncertainty for patients and providers than other aspects of care delivery, including chronic disease management. After hospitalization, patients may experience medical, functional, social, and spiritual changes. While the interventions we studied made changes to the clinical or healthcare system, they frequently did not address the home context to which patients were returning. Learning and self-organization may be particularly important in this situation as they may help patients effectively respond to problems that might arise after they have returned home, reducing their likelihood of readmission. For example, interventions that include an aspect of patient education (like teaching congestive heart failure patients to manage weight and adjust diuretics) may help patients to know what to expect and respond.

One commonality between our studies of diabetes, congestive heart failure, and care transitions is that, in each study, interventions that had an adaptive element were more successful, whether it be through allowing local self-organization among individuals in the system, or recognizing that implementation is an evolutionary process that requires change over time. We found few descriptions of how interventions were modified or adapted over time, and no examples of co-evolution in the interventions we reviewed. This may spring from a desire for intervention fidelity, or typical word count limitations in the medical literature. Our findings across our three studies speak to the importance of the need for adaptive approaches. The complex and dynamic nature of clinical systems, and of patients’ trajectories over time, speaks to the need for adaptive approaches in which individuals can better tailor the intervention and implementation strategy to the ways they have organized care over time. More details of these adaptive approaches (such as detailed protocol papers) should be reported in the literature regarding these interventions. Increased interest in adaptation in the implementation science literature (e.g., [73, 74]) is helping expose the importance of both carefully documenting modifications as they happen and understanding adaptations as important for intervention success.

Transitional care interventions with characteristics of complex adaptive systems, particularly learning and self-organization, are associated with a greater reduction in early readmissions. New study designs allowing for adaptation are needed to better address interventions in the complex clinical systems in which they are implemented.