Towards a patient journey perspective on causes of unplanned readmissions using a classification framework: results of a systematic review with narrative synthesis

The literature search was applied in Web of Science, Scopus, Pubmed and Embase in December 2016, using “readmission and avoidability or preventability” as key terms (Additional file 1). A medical information specialist was consulted for the search. All citations were imported into Endnote × 7.3.1TM.

The 77 studies which were included in the previous review [10] were checked for eligibility for this review by two researchers (CB and RS) (see Additional file 2). Studies were included if they fulfilled to the following criteria: original data, primary focus on preventable readmissions, clear description of preventable readmissions and potential causes of preventable readmissions (the cause classification) (see Additional file 3). We defined a cause classification as the description of at least three causes or synonyms for causes (e.g. (contributing) factors and reasons). Studies that described the causes of preventable readmissions (≥3) were further reviewed. Studies that made no distinction in readmissions and (primary) admissions, or preventable readmissions and non-preventable readmissions, were excluded from further analysis. In case of disagreement, a senior researcher was consulted to reach a final selection of studies.

We applied a validated approach to evaluate and compare the studies. This approach was developed by the Cochrane Consumers and Communication Review Group with the aim to perform a structured and transparent analysis for systematic reviews that rely primarily on the use of words and text to summarize and explain the findings of the synthesis. The approach can be described as ‘narrative’ analysis or synthesis [12]. We used this approach because of the large heterogeneity in study designs, and because other quality appraisal approaches and tools were less appropriate to apply. Prior to the narrative analysis a data-charting form was jointly developed by the authors to determine which variables to extract (including defining these variables). Subsequently, the authors independently charted the data, discussed the results and continuously updated the data-charting form in an iterative process. The data-charting form included variables such as: study design characteristics, sources of information to assess preventability, definition of preventability and percentage of preventable readmissions. After extracting the data the authors critically evaluated the applicability of the data and tabulated the results in order to identify patterns across the included studies.

The root causes were based on the PRISMA classification scheme of Schaaf & Habraken [9]. PRISMA is an abbreviation of Prevention and Recovery Information System for Monitoring and Analysis, and is often applied for root cause analyses in healthcare. Schaaf & Habraken defined four root causes: technical, organizational, human behavior and patient. We adapted these categories prior to the start of the data collection based on our experience with patient record reviewing in several Dutch hospitals [91]. We sub-divided the root causes “organization, human and patient” in organization (integrated care), organization (hospital department level), human (care provider), human (informal caregiver), patient (self-management) and patient (disease). In addition, we provided examples of causes for the settings: hospital care, ambulatory care and home (care). Hospital care was defined as the complete care process from admission until discharge, including outpatient visits after discharge. Ambulatory care was defined as medical care provided in an outpatient setting. Care at home was defined as informal care provided at home after discharge.

Two researchers (RS, MdB, CB, FK independently) allocated each cause to the root causes mentioned above. A third senior researcher performed a double check on the final cause classification. Disagreements between researchers were resolved during meetings (RS, CB, FK, MdB) until consensus was reached.

A summary of the study characteristics is shown in Table 1. The majority of the studies (64.4%) was conducted in the US and all studies were performed in a hospital setting (75.6% monocenter studies). Eight studies (17.8%) examined all-cause readmissions, meaning that these studies included readmissions from all departments, irrespective of whether this department was similar to that of the previous admission [46‐53]. The majority of the studies included readmissions of a specific specialty (31.1%) [54‐67] such as readmissions to a pediatric department. Twenty-seven percent of the studies examined readmissions of a specific group to a specific specialty (26.6%) [14, 68‐78] such as patients with diabetes readmitted to internal medicine. Twenty-four percent of the studies included readmissions of a specific group (24.4%), such as elderly [2, 3, 13, 79‐86]. Other studies included readmissions occurring to any department for a specific group (e.g. elderly to internal medicine).

Eight studies referred to an existing root-cause assessment tool with predefined causes such as STAAR, PRISMA or REDcap tool [2, 53, 65, 69, 76, 81, 84, 85]. Multiple studies referred to causes used in previous publications [3, 50, 55, 57, 59‐61, 64, 66, 67, 71, 74, 75, 79]. In particular, the studies of Clarke [59], Goldfield [87], Jiminez-Puente [49] and Oddone [74] were frequently referred to.

The frequencies of each root cause are listed in Additional file 5. A total of 381 causes were found of which 275 were reported as preventable and 44 as unpreventable (see Additional file 5). Twenty-six causes were reported as both preventable and unpreventable in a single study, and these causes were coded as both. Examples of causes that were considered differently among studies are listed in Table 2. Other causes that were not explicitly defined as preventable or unpreventable (n = 36) were coded as neutral. The final cause classification framework is listed in Table 3.

The main objective of this systematic review was to summarize the range of causes described for unplanned readmissions and to present a cause classification framework for causes of readmissions. The findings of this review show that many studies primarily focus on causes related to organization of care at the department level. The causes were mainly evaluated from a hospital perspective. None reported technical failures as one of the causes of (preventable) readmissions. We succeeded in classifying the majority of the causes according to the adapted PRISMA scheme and to distinguish between levels of care. We were not able to allocate causes that were ambiguous.

The results of this review indicate that the causes of readmissions depend on the context of the study, e.g. the inclusion of department/healthcare organization, case-mix, assessor of preventability, and definition of preventability. All studies were conducted in the hospital and several studies included causes beyond the setting of the hospital. The latter studies often identified other causes for readmissions in addition to hospital care (e.g. social support [13, 71, 77, 84‐86]). This finding is in line with the existing literature [6, 11, 88‐90]. Zhou et al. (2017) reviewed studies which used prediction models for (preventable) unplanned readmissions. They identified that health insurance and overall prognosis should be considered as contributors in addition to the clinical factors which are normally included in the statistical models. We also found that factors such as self-management, social support, and type of healthcare organization can play a role. Other factors beyond the scope of the hospital should be assessed in more detail to gain a better insight into the complexity behind unplanned readmission and to understand where interventions can have the most impact. In addition, gaining more understanding on these factors is necessary in order to optimize the applicability of unplanned readmissions as a quality indicator of integrated care [11, 88]. Decreasing unplanned readmissions needs a holistic approach, because health care is a complex dynamic system in which patients, professionals, health care organization are acting within the regulations of their national health care system.

Furthermore, this review highlights that studies differ in which causes are considered to be preventable. In particular, complications, adherence and disease progression were seen as both preventable and unpreventable. For example, ‘foreseen complications’ were seen as preventable [61, 79]. Complications of surgical procedures were also reported as preventable in many studies [48, 49, 81]. Greenberg et al. reported complications related to neurological impairment and immobility as an unpreventable cause [70]. Shah et al. considered complications as preventable and unpreventable. They defined the following complications: complications occurring in spite of best practices being followed, complications due to progression of natural history of certain chronic neurosurgical disease, post-procedure complications and complications related to medication use. The first two were reported as unpreventable, and the latter two as preventable [63]. These findings highlight that preventability can differ depending on many contextual factors and what the authors of studies regard as adequate care.

The proposed classification allows researchers and policymakers to further consider the complex nature of readmissions. Differences in cause classifications can be problematic for fair comparison of studies. Therefore, it is essential that studies describe the context of the research setting/research population and elaborate on what they regard as preventable.

In addition, this review indicates that the assessment of readmissions is mainly based on data that is inputted in the hospital information system. Some studies included interviews and were thereby able to capture the patients’ experience. However, the transcripts were not available and therefore not included in this review. Furthermore, studies indicated that interviews with patients and/or caregivers can provide additional information on factors beyond the hospital setting which cannot be found in medical records [85]. For instance, Sutherland et al. (2016) found that the combination of patient interviews and chart review revealed additional gaps in care [75]. In addition, Toomey et al. (2014) found in 31.2% of cases that interviews with patients and primary care providers provided new information. These findings support other studies stating that increased involvement of patients and other stakeholders are crucial for assessing the cause of the readmission and may contribute to the prevention of readmissions and to better quality of care [89].

This review provides insight into the range of causes reported in unplanned readmission studies. The insights gained from this review may help others in conducting readmission studies. However, this review has some limitations. Firstly, the studies included were based on a set of articles which were collected for a previous review [10]. As a consequence, we may have missed studies which also examined the causes of unplanned readmissions, but did not provide information on the rate of preventable readmissions. This might have limited the number of causes reported in this study and consequently the number of root causes/categories in our cause classification framework. However, we performed a cross reference checking in Scopus and Web-of-Science and did not find additional studies. Secondly, we allocated a single cause to only one root cause. Thus, causes which consisted of multiple sub-causes were disentangled into separate causes. This might have influenced the frequency of the themes and subthemes that were found. Thirdly, studies which did not explicitly specify the context of a specific cause (ambiguous causes) were considered “Unclassifiable”. As a result, the number of causes related to other root causes may be underestimated. Lastly, all phases were either consensus based-driven and/or performed by at least two independent data extractors. However, using this procedure can not preclude that some amount of interpretation bias occurs during data collection, synthesis and interpretation.

Since 2016, unplanned readmissions have been used as a quality indicator for hospital care in the Netherlands. To increase the comparability of causes between readmission studies, we advise that the use of a standardized cause classification framework should be stimulated. The framework proposed in this review provides a broad range of causes which are most frequently observed. Regarding hospital care at the department and institutional level this overview is rather complete, although technical causes are lacking and are becoming more relevant due to development of eHealth, mHealth and ICT supported long distance care. Causes at the patient and health care system level are still of limited detail. For example, more information on the patient experience can be collected using interviews. Insight into these causes is crucial for developing improvement opportunities.

In addition, the proposed framework may contribute to a transparent development of prediction models. Information on how data sources are used to calculate the readmission indicator are often lacking [11]. Current prediction models depend on a limited range of data documented in the hospital information system. Information on what happens after discharge is (often) not documented. Hence, policy makers should be careful in holding hospitals accountable for unplanned readmissions as long as it is unclear which percentage of these readmissions is truly preventable by preventive actions of the hospital itself. Unplanned readmission should be viewed as adverse events occurring in a complex dynamic system, in which patients, professionals and organizations all play a crucial role, regulated by the healthcare system.

This review indicates that the causes of unplanned readmissions which are examined by the studies differ between studies and are often limited to causes focused on hospital care. As a result, the multifactorial nature of readmissions is not well understood and certain contributing causes might be overlooked. This limits the potential impact of (patient-centered) interventions to prevent readmissions [88]. To improve our understanding of readmissions, all actors involved in the patient journey should be considered. The suggested framework may provide direction on which causes to include in a readmission study and prediction models. In the future, we hope that with the use of prediction models, high-risk patients can be more easily identified and targeted for alternative management [90]. In addition, future research should take into account the interdependency of causes. For instance, when a patient gets readmitted due to an incorrect assessment by a physician at discharge, this readmission might be the result of how readmissions are handled at a department/organization. Physicians might not be effectively trained to ask the patient if they are ready to be discharged. Lack of skills can thus also be considered a result of suboptimal organization of care (organization department level care). Failure on organization level (e.g. missed opportunity to train physicians) or care-provider level (lack of skills) are important to consider. In addition, it is also possible that failure on one level might be mitigated by other levels. Each level must be considered in the study of readmissions. Multiple causes of readmissions - in and beyond the hospital - play a role. All these causes should be assessed to capture the complex nature of readmissions and should be considered interdependently.