Background
Methods
Data source and searches
Study selection
-
Step 1: Two researchers (CB, EK) independently screened all abstracts using the major inclusion and exclusion criteria, i.e. English language, manual assessment using, at least, the medical record and a clear method description regarding preventability assessment in the aim, method or result section, see Additional file 2. Cohen’s kappa for interrater agreement (CB and EK) was good (k = 0.70).
-
Step 2: References of included articles were assessed and a cited reference search in Web of Science and Scopus (CB and EK) was performed additionally for all full text articles included in step 1 (n = 77).
-
Step 3: Detailed inclusion and exclusion criteria (Additional file 2) were applied to all 77 articles by two researchers independently (equally divided over CB, EK, RS). This additional step was conducted to ensure that the finally selected articles were able to help us reach our study objective; 1. Full text article in English; 2. The article should be based on original patient data; in case of ≥2 or more papers used the same, or partly the same, patient sample only the paper with the most thoroughly described methodology of preventability assessment was included; 3. Studying hospital readmissions should be clearly stated in the aim/ primary objective; 4. Duration between index and readmission should be ≤6 months; 5. Assessment of preventability should be performed via manual medical record review or at least, it should be clear that the preventability assessment was performed on an individual patient level by a care provider and/or trained researcher which cannot be performed without a review; 6. The methodology of the preventability assessment of readmissions should be described clearly in order to perform data-synthesis; this includes a description of criteria of preventability and/or a cause classification (≥3 cause categories) of preventable readmissions and the reviewer process (at least 2 independent reviewers and disagreement should have been solved by reaching consensus and/ or a third independent reviewer OR, in case not performed/ nor reported (NR) > 50 medical files of readmitted patients should have been reviewed).
Critical appraisal of individual sources of evidence
Data synthesis
Data extraction and analysis
Results
Study design and characteristics
Study characteristics (n = 48) | No. or percentage of studies |
---|---|
Year of publication, range | 1988–2016 |
Country, n (%) | |
USA | 32 (67%) |
Other | 16 (33%) |
Study design, n (%) | |
Retrospective | 30 (63%) |
Cross-sectional | 10 (21%) |
Prospective | 8 (16%) |
Setting, n (%) | |
Single center | 37 (77%) |
Multicenter | 11 (23%) |
Number of readmissions reviewed, n ± SD | 226 ± 208 |
Planned readmission excluded, n (%) | |
Yes | 30 (63%) |
No | 11 (23%) |
Not reported | 7 (14%) |
All-cause readmission, n (%) | |
Yes | 9 (19%) |
No | 39 (81%) |
Percentage preventable readmissions, mean, ± SD | 27,8 ± 16,7% |
Scoring of preventability, n (%) | |
Binary | 22 (46%) |
Scale | 4 (8%) |
Categorical | 17 (35%) |
Not applicable (a priori studies) | 5 (11%) |
A priori preventable causes determined, n (%) | |
Yes | 32 (67%) |
No | 16 (33%) |
Training of reviewers, n (%) | |
Yes | 16 (33%) |
No | 2 (4%) |
Not reported | 30 (63%) |
Number of reviewers, n (%) | |
Individual | 8 (16%) |
Duo | 23 (48%) |
Duo + team | 2 (4%) |
Individual + team | 2 (4%) |
Team | 5 (11%) |
Individual or duo + panel | 3 (6%) |
Other | 5 (11%) |
Double check, n (%) | |
All cases | 28 (58%) |
Partially | 7 (15%) |
No | 3 (6%) |
Not reported | 10 (21%) |
Additional sources, n (%) | |
Interview or survey | 13 (27%) |
None | 35 (73%) |
Sources of information
Preventability
‘Providers were given no specific guidelines for deciding whether a readmission was preventable. This allowed use of their different backgrounds in choosing which elements of the clinical record to focus on.’ [33]
Cause classification
Author | Planned read-missions excluded?a | No. read-missions reviewedb | No. of preventable unplanned readmissions | % preventable unplanned readmissionsc | Scoring of preventability | A priori preventable causes determined | Training of reviewers | Reviewersd | Double check of preventability | Additional sources used for the review |
---|---|---|---|---|---|---|---|---|---|---|
Agrawal | yes | 30 | 11 | 36,7 | categorical | yes | no | individual | no | – |
Auerbach | yes | 1000 | 269 | 26,9 | scale | yes | yes | duo | all cases | Interviewf |
Balla | yes | 271 | 90 | 33,2 | binary | no | NR | duo | all cases | Interviewe |
Bianco | no | 229 | 100 | 43,7 | binary | yes | yes | duo | all cases | Interviewe |
Burke | yes | 335 | 78 | 23,3 | categorical | yes | yes | duo | all cases | Interviewe |
Cakir | NR | 85 | 4 | 4,7 | categorical | yes | NR | individual | NR | – |
Clarke | yes | 74 | 18,9 | 25,5 | categorical | yes | NR | duo or team | all cases | – |
Dawes | yes | 258 | 55 | 21,3 | categorical | yes | yes | duo + team | All cases | – |
Epstein | no | 50 | 1 | 2,0 | categorical | yes | NR | duo + team | All cases | – |
Feigenbaum | no | 537 | 250 | 46,6 | categorical | yes | yes | duo | all cases | Interviewf |
Fluitman | yes | 50 | 26 | 52,0 | binary | yes | NR | duo | all cases | – |
Frankl | yes | 318 | 28 | 8,8 | categorical | yes | NR | individual + team | partially | – |
Gautam | yes | 109 | 16 | 14,7 | binary | yes | NR | individual + team | NR | Interviewf |
Glass | NR | 96 | 25 | 26,0 | binary | no | NR | NR | NR | – |
Greenberg | yes | 97 | 22 | 22,7 | categorical | yes | NR | duo | NR | – |
Hain | no | 200 | 40 | 20,0 | scale | yes | yes | panel | all cases | – |
Halfon | yes | 429 | 40 | 9,3 | NA | yes | NR | duo | partially | – |
Harhay | yes | 201 | 19 | 9,5 | binary | yes | yes | duo | all cases | |
Jiminez-Puente | no | 185 | 44 | 23,9 | binary | yes | NR | duo | all cases | – |
Jonas | no | 248 | 15 | 6,0 | binary | yes | NR | individual + panel | partially | – |
Kelly | yes | 32 | 22 | 68,8 | binary | yes | NR | duo | all cases | |
Koekkoek | no | 298 | 45 | 15,1 | categorical | no | yes | individual | NR | – |
Maurer | yes | 32 | 3 | 9,4 | binary | yes | NR | duo | partially | – |
Meisenberg | yes | 72 | 22 | 30,6 | binary | yes | NR | duo | all cases | – |
Miles | yes | 437 | 24 | 5,5 | categorical | no | NR | duo | partially | – |
Mittal | yes | 35 | 15 | 42,9 | binary | no | NR | duo | all cases | |
Nahab | no | 174 | 92 | 52,9 | NA | yes | NR | duo | all cases | – |
Nijhawan | yes | 130 | 62 | 47,7 | NA | yes | NR | duo + panel | all cases | – |
Njeim | NR | 161 | 51 | 31,7 | binary | no | yes | individual | no | – |
Oddone | NR | 514 | 183 | 34,2* | categorical | no | yes | duo | partially | – |
Pace | yes | 140 | 19 | 13,9 | categorical | yes | NR | duo | all cases | – |
Ryan | yes | 40 | NR | 26,7 | categorical | no | yes | team | all cases | – |
Saunders | yes | 282 | 51 | 18,1 | binary | yes | NR | team | all cases | – |
Shah | no | 407 | 149 | 36,6 | NA | yes | NR | duo | all cases | – |
Shalchi | NR | 63 | 45 | 71,4 | binary | no | NR | team | all cases | – |
Shimizu | no | 153 | 50 | 32,7 | binary | yes | NR | panel | all cases | Interviewe |
Stein | yes | 213 | 64 | 29,5 | binary | no | NR | individual | NA | Interviewf |
Sutherland | yes | 47 | 11 | 23,4 | NA | no | NR | individual | NR | Interviewe |
Tejedor-Sojo | no | 147 | 62 | 42,2 | categorical | yes | yes | team | NR | – |
Toomey | yes | 305 | 90 | 29,5 | scale | no | yes | team | all cases | Interviewf |
Vachon | yes | 98 | 14 | 14,3 | binary | yes | NR | individual | NR | – |
Van Walraven | yes | 317 | 70 | 22,1 | scale | no | yes | duo | partially | Interviewe |
Vinson | NR | 66 | 10 | 15,2 | categorical | no | NR | duo | NR | Interviewe |
Wallace | yes | 204 | 41 | 20,1 | binary | yes | yes | duo | all cases | – |
Wasfy | NR | 893 | 380 | 42,6 | categorical | yes | yes | duo | all cases | – |
Weinberg | yes | 50 | 3 | 6,0 | binary | yes | NR | panel | all cases | – |
Williams | yes | 133 | 78 | 58,6 | binary | no | NR | individual | no | Interviewf |
Yam | yes | 603 | 246 | 40,8 | binary | no | no | Duo + panel | all cases | – |
Reproducibility/reviewer process
Discussion
Advantage | Limitation | Recommendations | |
---|---|---|---|
Single center versus multicenter | Single center studies provide information on one’s own performance which is needed to induce a quality improvement cycle | For scientific purposes it is easier to identify which results can be extrapolated to other institutes when the results are obtained via a multicenter study. Furthermore, in a multicenter study benchmarking between the centers is possible. | Compare the results with the current literature on the preventability of readmissions, and be aware of (inter)national and regional differences in organization of care. |
Population (Focus on a specific population versus a broad population) | Manual review is easier to perform on a specific group (e.g. diagnosis heart failure or department). | Focus on single group can cause underestimation of the preventability readmission rate and/or underreporting of certain causes. | Consider a multidisciplinary panel or team to review the readmissions to reduce blind spots. |
Relatedness (focus on readmissions that are related to the index readmission versus all-cause readmissions) | Readmissions related to the index hospitalization will generally identify causes that are related to hospital care. | All-cause readmissions are easier to identify based on administrative data, provide a broad scope and will identify other causes; for example causes related to care in the primary care setting. | Determine the scope of the quality improvement cycle; to identify causes related to hospital care or to care of a region |
Type of readmissions (unplanned versus planned readmissions) | Selecting only unplanned readmissions resembles the readmissions that are used to calculate the readmission quality indicator | Planned readmission might also have preventable causes which will be missed if planned readmissions are excluded | Determine whether you consider unplanned readmissions preventable prior to starting a readmission study |
Setting and sources (focus on hospital versus an integrated care network) | Assessment based on a hospital’s perspective only requires the medical record as single source. | Fragmented and incomplete description of the patient’s journey can result in underreporting causes related to integrated care, patient and social factors. | Interview, questionnaire or survey a (subset) of patients and or primary care providers. |
Information and sources (which sources and information to include; and in which order) | Including the full medical record, outpatient data and even additional sources (e.g. interviews) can change the perspective on preventability and its causes. | Reviewers might use a different approach of obtaining/using the (additional) information which can create unwanted differences in the perspective on preventability. Note that for an interview of stakeholders a cross-sectional or prospective study design is needed to reduce recall bias. | A strict protocol and logbook as well as training prior to start of the study. Consider to provide additional information stepwise to assess its added value on the preventability assessment. |
A priori (preventability) cause classification | Easier to perform and probably better agreement between reviewers. | Does not invite reviewer to look beyond this list of predefined (potentially preventable) causes and can therefore narrow the reviewer’s view. | Usa a multidisciplinary approach with more than one reviewer. The use of a strict protocol and logbook as well as training prior to start of the study, and case discussion during the study, can increase uniformity |
Reviewers (single reviewer versus duo/team) | Using a single reviewer to perform the preventability assessment is less time-consuming. | Due to the poor reproducibility some kind of double check is needed. | Double (partial) review can increase uniformity. If a double check is not possible, consider a team or panel discussion (of a subset) of cases. Moreover, case discussion adds to the learning and awareness component of the medical record review process. |
Experience | Residents as reviewer can contribute to the learning environment. | Some studies suggest that years of experience can influence the preventability assessment. | Approach seniors to be available for supervision, double check by a senior and/or training, strict protocol or discussion meetings. |
Complete or partial double check | A partial double check is less time consuming. | This can influence the agreement calculation. | In case of partial double check use the appropriate analysis. |
Final preventability judgment (binary score versus scale or category) | Using a binary score for preventability is straightforward and easy to interpret | Since the majority of readmissions have multifactorial causes a binary preventability score does not resemble reality; a scale of category offers the option of making a thoughtful decision | Use a scale or category which includes intermediate scores on preventability. Be clear on which categories are used/combined to calculate the preventability percentage. |