Preventability and Causes of Readmissions in a National Cohort of General Medicine Patients

Importance  Readmission penalties have catalyzed efforts to improve care transitions, but few programs have incorporated viewpoints of patients and health care professionals to determine readmission preventability or to prioritize opportunities for care improvement.

Objectives  To determine preventability of readmissions and to use these estimates to prioritize areas for improvement.

Design, Setting, and Participants  An observational study was conducted of 1000 general medicine patients readmitted within 30 days of discharge to 12 US academic medical centers between April 1, 2012, and March 31, 2013. We surveyed patients and physicians, reviewed documentation, and performed 2-physician case review to determine preventability of and factors contributing to readmission. We used bivariable statistics to compare preventable and nonpreventable readmissions, multivariable models to identify factors associated with potential preventability, and baseline risk factor prevalence and adjusted odds ratios (aORs) to determine the proportion of readmissions affected by individual risk factors.

Main Outcome and Measure  Likelihood that a readmission could have been prevented.

Results  The study cohort comprised 1000 patients (median age was 55 years). Of these, 269 (26.9%) were considered potentially preventable. In multivariable models, factors most strongly associated with potential preventability included emergency department decision making regarding the readmission (aOR, 9.13; 95% CI, 5.23-15.95), failure to relay important information to outpatient health care professionals (aOR, 4.19; 95% CI, 2.17-8.09), discharge of patients too soon (aOR, 3.88; 95% CI, 2.44-6.17), and lack of discussions about care goals among patients with serious illnesses (aOR, 3.84; 95% CI, 1.39-10.64). The most common factors associated with potentially preventable readmissions included emergency department decision making (affecting 9.0%; 95% CI, 7.1%-10.3%), inability to keep appointments after discharge (affecting 8.3%; 95% CI, 4.1%-12.0%), premature discharge from the hospital (affecting 8.7%; 95% CI, 5.8%-11.3%), and patient lack of awareness of whom to contact after discharge (affecting 6.2%; 95% CI, 3.5%-8.7%).

Conclusions and Relevance  Approximately one-quarter of readmissions are potentially preventable when assessed using multiple perspectives. High-priority areas for improvement efforts include improved communication among health care teams and between health care professionals and patients, greater attention to patients’ readiness for discharge, enhanced disease monitoring, and better support for patient self-management.

Despite continuous and robust efforts, the ability of health systems to reduce hospital readmissions has been disappointing.¹ The discouraging progress in reducing readmissions across broad populations points to potential gaps in health systems and communities,^2-6 as well as to shortcomings of broad-based readmission reduction programs, few of which have fulfilled their initial promise.^7-9

Underlying readmission reduction programs are the concepts that some proportion of readmissions is preventable^2,3,10 and that identifying and addressing the drivers of “preventable” readmissions can improve the effectiveness of care transitions programs.¹¹ However, few nationally representative data exist to define the frequency of readmission preventability.^3,12 Moreover, national data are lacking on whether specific care processes, patients’ needs, or comorbidities are more associated or less associated with preventability. Finally, although small studies^7,8,13 have included viewpoints of patients in understanding readmission preventability, few large-scale studies have explicitly included their viewpoints and that of their physicians in determining preventability.¹⁴

To explore these questions, we performed an observational study of general medicine patients readmitted within 30 days of discharge to 12 academic medical centers in the United States. We collected data from patient and physician surveys and medical record review to identify factors contributing to readmissions. After aggregating information from these sources, we used a structured case review process to determine if a readmission was potentially preventable, whether clinical or health care delivery processes could have contributed to the readmission, and which of these processes were most commonly associated with preventable readmissions.

Our study took place in the Hospital Medicine Reengineering Network (HOMERuN), a national network of hospital medicine investigators at 12 academic medical centers.¹⁵ Patients in our study were discharged by general medicine services at HOMERuN sites and readmitted (also to a general medicine service) within 30 days of discharge between April 1, 2012, and March 31, 2013.

Eligible patients were 18 years or older and spoke English as their primary language. Patients who had a scheduled readmission (eg, for chemotherapy or a procedure) were excluded. Within the eligible sample, we used a random-digit generation schema to select up to 5 patients per week at each site for interview and study participation. If a patient declined an interview, was too sick to participate, was unavailable, or otherwise declined participation, the next randomly selected patient was approached for enrollment. Institutional review boards at the University of California, San Francisco (the data coordinating center) and all participating HOMERuN sites approved the study.

Data were collected from interviews with patients, from reviews of available inpatient and outpatient medical records, and from surveys of patients’ physicians (primary care physician when available, discharging inpatient physician from the index admission, and inpatient physician from the readmission). After obtaining written informed consent, trained research assistants administered patient interviews that included fixed-choice and open-ended questions to learn about patients’ perceptions of their care during their previous admission and their experience since discharge. Fixed-choice items included the following domains: social support, quality of communication with hospital physicians, whether a follow-up appointment was made and attended, and perceived ability to manage medications, symptoms, and appointments after discharge. Open-ended questions asked patients about any problems in recovery that they experienced after the index discharge, as well as what patients thought could have helped avoid a readmission to the hospital.

We then emailed or faxed up to 5 surveys to each patient’s primary care physician, physician from the index admission, and current attending physician. Physician surveys asked questions regarding their impression of factors contributing to the readmission and aspects of care that could have been improved, such as timely communication about discharge plans. Physicians were encouraged to read the medical record to better inform their answers to survey questions. Using this approach, we received 359 responses from primary care physicians, 683 responses from physicians from the index admission, and 743 responses from current attending physicians. All cases had at least 1 physician survey available at the time of case review.

Next, research assistants performed a structured medical record review, collecting information regarding patients’ comorbidities and medications. They also recorded medical record–based measures of care transitions processes (eg, receipt of a reconciled list of medications at the time of the index hospital discharge).

Our medical record–based measures were determined based on those criteria proposed by the National Quality Forum’s Care Coordination Measures¹⁶ and other published standards for discharge documentation completeness.¹⁷ Patient survey questions included modified items from the 3-item Care Transition Measure¹⁸ and the interpersonal processes of care measure.^19,20 Physician surveys were developed to include questions that paralleled those questions asked in our case review process (see the next subsection below), as well as impressions of key transitions processes (eg, the completeness of the discharge summary). Before use, all surveys were pretested among the investigator group and with physicians not associated with the study.

Our case review process was based in part on the approach used in other studies,^21,22 as well as approaches considered standard in defining preventability in adverse drug events and care transitions gaps.^22-29 We further refined past approaches to permit implementation across multiple sites, while adding processes to retain intersite and longitudinal intrareviewer consistency.

Our case review process had the following 2 key objectives: (1) to determine whether readmission was potentially preventable and (2) to identify factors that contributed to readmission, regardless of preventability. Case reviewers chose from a large set of potential factors that were identified and categorized using the framework of the Ideal Transitions in Care.³⁰ In assessing preventability, we trained case reviewers to consider patient illness but to primarily focus on system flaws and gaps in care that could have been avoided with reasonable patient or physician activities. As a framing example, we trained physician adjudicators to consider an “ideal health system” as a model for system and care assessment, even if all aspects of an ideal system did not exist at their site. For example, if a patient’s readmission appeared to be related to the inability to obtain a postdischarge appointment, we instructed case reviewers to consider it a preventable readmission because an ideal system would be able to accommodate these patients’ needs without requiring readmission.

We assigned preventability with a scale used in previous research regarding care transitions.^22,28 Within this scale, we further defined a threshold of “greater than 50-50, but close call” as a standard cutoff, also based on previous studies.^22,28 This approach is useful in that it links an approach that encourages reviewers to explicitly avoid a “neutral” response in assessing preventability and provides a valid cut point that can help direct intervention strategies.

Physician reviewers had access to completed patient interviews, physician surveys, data derived from abstracted medical records, and the complete medical record. At a minimum, each case review packet included the patient interview, a complete medical record review, and at least 1 physician survey. All physician adjudicators reviewed several reference cases during a series of weekly webinars and conference calls. As the case review work proceeded, each site presented at least 2 anonymized cases for group discussion during biweekly conference calls to foster consistency among physician adjudicators.

The Hospital Medicine Reengineering Network did not calculate interrater reliability as part of its methods and instead used a 2-physician case review process to assign preventability. We provided substantial structure and support for the dual-physician reviews. All reviews were performed by physicians who were initially trained via our physician review guides, and then by having all reviewers perform “test” reviews and by regularly discussing reviews at biweekly conference calls. In addition, we maintained an “FAQ” document for how to adjudicate various situations, with an email of all updates as they became available, and maintained a resource for teaching points and clarifications using a HOMERuN wiki webpage.

Each site had a pool of 3 to 10 physician adjudicators coordinated by a physician lead, who oversaw the process and resolved difficult cases. A pair of physician adjudicators reviewed all available information for each case and developed the initial assessments, after which the pair made a final assessment of the case jointly. Site physician leads were responsible for resolving any challenging cases, and these cases were also reviewed at regular telephone conference calls.

We first characterized study patients using univariable methods. Readmissions were categorized as preventable if physician adjudicators rated the likelihood of preventability as 50% or more (≥4 on a 6-point scale), as done in previous studies.^22,28 Using bivariable methods, we then compared patients whose readmissions were judged to be preventable vs those whose readmissions were judged to be nonpreventable in terms of factors that contributed to the readmission.

We selected potential contributing factors after initially screening for those variables with an unadjusted P value for association with preventability of P ≤ .20. Using these initial variables, we then constructed hierarchical multivariable models, including clustering at the hospital level to predict preventability of readmissions. If covariates had high bivariable correlation, we considered only one for model inclusion by excluding variables with lower face validity. We next used a backward stepwise approach to develop our final model by removing variables until the final covariates were associated with the outcome at P < .05. We then used our final model to calculate the percentage of the preventable readmissions that were potentially affected by each identified risk factor. Specifically, we calculated an adjusted risk difference of preventable readmission for the model between cases with and without the factor, and then multiplied this value by the prevalence in our data of the factor and divided by the overall proportion of preventable readmissions.³¹ All analyses were performed using statistical software (SAS, version 9.4; SAS Institute, Inc).

One thousand patients were readmitted to study hospitals, were randomly selected for our study, and gave written informed consent to participate. Their median age was 55 years. Other characteristics of the cohort are listed in Table 1.

Of readmitted patients, 26.9% (269 of 1000) had a readmission that was considered potentially preventable after case review (Table 2). Among preventable readmissions, 52.0% (140 of 269) were thought to have been potentially preventable with efforts made during the index admission.

Patients whose readmission was deemed preventable reported experiences similar to those of patients whose readmission was deemed nonpreventable in terms of inpatient care processes (eg, having enough time to say what they thought was important or perceiving that their physician took their preferences into account) and in terms of their ability to manage their care after discharge. However, patients who reported problems with drugs or alcohol were less likely to have their readmission considered preventable (4.5% [12 of 269] vs 8.1% [59 of 731]; P = .048) (Table 3), while patients who did not know how to reach their physician after discharge were more likely to have their readmission considered preventable (18.6% [50 of 269] vs 12.6% [92 of 731]; P = .02).

Multiple potential underlying factors were noted when we compared preventable and nonpreventable readmissions in the domains of medication safety, care coordination, discharge planning, advance care planning, promotion of self-management, enlisting of help and social supports, diagnostic and therapeutic problems, and monitoring and managing of symptoms after discharge. Of potential underlying factors, those variables with the largest absolute differences in prevalence between preventable and nonpreventable readmissions were the following: inadequate treatment of symptoms other than pain (20.8% [56 of 269] vs 6.4% [47 of 731]), inadequate monitoring for medication adverse effects or nonadherence (14.9% [40 of 269] vs 4.4% [32 of 731]), follow-up appointments not scheduled sufficiently soon after discharge (16.0% [43 of 269] vs 5.7% [42 of 731]), patient lack of awareness of whom to contact after discharge or when to go (or not to go) to the emergency department (18.6% [50 of 269] vs 5.7% [42 of 731]), patient need for additional or different home services than those services included in discharge plans (17.8% [48 of 269] vs 7.8% [57 of 731]), discharge of patients too soon (eg, symptoms such as inability to eat or dyspnea not completely managed) from the index hospitalization (19.3% [52 of 269] vs 4.0% [29 of 731]), and issues related to the decision to admit the patient made in the emergency department (eg, the patient may not have required an inpatient stay, or useful information from the primary care physician was not available or reviewed) (12.6% [34 of 269] vs 2.6% [19 of 731]) (Table 4).

In multivariable models, 4 factors were most strongly associated with potentially preventable readmissions. These included premature discharge from the index hospitalization (adjusted odds ratio [aOR], 3.88; 95% CI, 2.44-6.17), failure to relay important information to outpatient health care professionals (aOR, 4.19; 95% CI, 2.17-8.09), lack of discussions about care goals among patients with serious illnesses (aOR, 3.84; 95% CI, 1.39-10.64), and emergency department decision making to admit a patient who may not have required an inpatient stay (aOR, 9.13; 95% CI, 5.23-15.95). The most common factors associated with potentially preventable readmissions included emergency department decision making (affecting 9.0%, 95% CI, 7.1%-10.3%), inability to keep appointments after discharge (affecting 8.3%; 95% CI, 4.1%-12.0%), premature discharge from the hospital (affecting 8.7%; 95% CI, 5.8%-11.3%), and patient lack of awareness of whom to contact after discharge (affecting 6.2%; 95% CI, 3.5%-8.7%) (Table 5).

In sensitivity analyses, we performed multivariable models that excluded data from sites with fewer than 50 patients, and these results were similar to those findings already presented. We also performed 2 additional analyses that excluded sites whose aggregate estimates of preventability were in the top or lower 2 of sites. Results from these analyses also did not reveal any significant changes in the factors identified.

In this multicenter, multiperspective study of readmitted patients, 26.9% (269 of 1000) of readmissions were considered potentially preventable, with half of these readmissions thought to represent gaps in care during the initial inpatient stay. Structured case review with multiple viewpoints, including perspectives of patients, identified a prioritized list of targets for refined care transitions programs.

Our estimates of readmission preventability are within the ranges suggested by other researchers³ but extend previous work in important ways. Our review process linked a comprehensive picture of clinical care, one that included viewpoints of patients, to a rigorous case review process that sought to identify not only readmission preventability but also opportunities for improvement. The process whereby we identified potential improvement targets also represents an important feature of our work. That is, our focus on an ideal health system lens for determining preventability provides a safeguard against fatalistic interpretations of readmissions as “nonpreventable” or solely owing to advancing illnesses, while also allowing us to identify factors that should be addressed so that improvement leads toward an “ideal.”

An ideal transition of care^1,8 can include a dauntingly wide range of potential programs³⁰ for health systems to implement and manage. Our calculations providing estimates of the proportion of potentially preventable readmissions affected by each risk factor can help prioritize efforts by weighting the odds of individual associations using the prevalence of the risk factor in our population. While the effectiveness of individual programs addressing individual gaps in care likely varies across issues, our study adds substantially to previous work by providing a prioritization schema that is useful in the beginning of program development. Perhaps not surprisingly, the use of population-based estimation produced a ranked list of important underlying causes for readmission that differed slightly from the list of factors ranked by adjusted odds. The list of factors that overlap in terms of risk and potential effect is shorter still, providing a potential approach to prioritizing readmission reduction efforts.

One key observation in our cohort related to improving decision making for patients arriving in the emergency department, a factor that represents not a shortcoming of emergency medicine or emergency departments, but a limitation of the health system itself. Overcoming gaps in care in the attempt to avoid potentially unnecessary admissions from the emergency department may need to involve improved communication among primary care health care professionals, hospital-based physicians, and emergency medicine physicians about criteria for admission and resources available in the community, in addition to providing greater access to urgent care for patients who would otherwise seek care in an emergency department and improving patients’ understanding of how and when to seek emergency care.

Our research also adds to the existing literature on readmissions by identifying the possibility that premature discharge from the hospital may contribute to readmission risk. While secondary data analyses have not demonstrated a correlation between shortening lengths of stay and readmission rates nationally,³² our data suggest that in the current era some proportion of readmissions may be prevented with better attention to patients’ readiness for discharge³³ in terms of their ability to manage care after discharge or recover from (or develop an effective management plan for) symptoms, such as dyspnea, vomiting, and pain.

Our results were also notable for factors that were not found to be key underlying contributors. Functional status is a clear risk factor for readmission^34,35 but in our cohort of readmitted patients was not associated with potential preventability. Patient reports of care processes and satisfaction with care were not associated with readmission preventability in our data, suggesting that patient satisfaction with care, while valuable for other reasons, may not be a valid approach to identifying readmission program priorities. Disconnect between most patients’ perceptions of care and readmission preventability may also represent gaps in the ability of satisfaction measures to detect patients’ actual ability to carry out the discharge plan.

Our study has some limitations. While our case review process has strengths, it was limited by the subjective nature of determining preventability of readmissions. For example, we cannot rule out biases of our reviewers regarding which factors may have contributed to readmission preventability. However, our results are similar to other estimates of preventability, and our training and quality assurance processes sought to maintain consistency of our approach across sites. Also, no patient factors were retained in our final models, but we cannot rule out the possibility of confounding by patient factors that were associated with both the identified risk factors and potential preventability. In addition, it is possible that our medical record tools may have led to instrument bias that may have limited our ability to detect factors outside of our tool’s list. That said, the list of factors we collected from patients, physicians, and medical records was large and is based on existing frameworks.³⁰ In addition, the large number of factors that were found to be significant makes the threat of this bias less likely. While our study included patients from a variety of hospitals, most were large academic medical centers, potentially limiting generalizability. Also limiting generalizability are our criteria that excluded non–English-speaking patients and patients unable to provide informed consent. We also did not track reasons for refusal among potentially eligible patients. That said, our cohort is similar to previous studies^36,37 of readmitted patients from our sites that did not use exclusion criteria. Our approach was associated with variation in rates of preventability across sites, which could represent true variation in care processes but also possible inconsistency of case review across sites. However, despite potential variation in case review processes, the factors we identified were robust in sensitivity analyses that excluded patients from the sites with the highest and lowest rates of preventability. Finally, population-attributable estimates can be used to prioritize potential benefits but do not take into account the effectiveness or cost of those programs. These estimates are best-case scenarios in terms of the proportion of readmissions that could be prevented, assuming 100% preventability owing to that factor and 100% efficacy of an intervention designed to address it.

Multicomponent care transitions programs are a desired approach to improving patient outcomes in the period after acute care. Because our study cannot ascribe causality to the factors we have identified, our results cannot support the conclusion that eliminating the factors we identified will surely reduce readmissions. The answer to that question will require further studies. Our study formulates a potential approach for prioritizing local efforts, as well as monitoring the effectiveness of programs in place. Finally, our results suggest a potential approach to focus interventions in ways that span the continuum of care, prioritize efforts to prepare patients more effectively for discharge, and provide better ability for patients, caregivers, and health care professionals to support patients and improve outcomes during the period after hospitalization.

Correction: This article was corrected on August 15, 2016, to fix 4 P values in Table 1.

Accepted for Publication: November 30, 2015.

Corresponding Author: Andrew D. Auerbach, MD, MPH, Division of Hospital Medicine, Department of Medicine, University of California, San Francisco, 505 Parnassus Ave, PO Box 0131, San Francisco, CA 94143 (andrew.auerbach@ucsf.edu).

Published Online: March 7, 2016. doi:10.1001/jamainternmed.2015.7863.

Author Contributions: Dr Auerbach had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Auerbach, Kripalani, Sehgal, Lindenauer, Metlay, Fletcher, Ruhnke, Flanders, Williams, Giang, Herzig, Robinson, Schnipper.

Acquisition, analysis, or interpretation of data: Auerbach, Kripalani, Vasilevskis, Sehgal, Lindenauer, Fletcher, Ruhnke, Kim, Williams, Thomas, Herzig, Patel, Boscardin, Robinson, Schnipper.

Drafting of the manuscript: Auerbach, Fletcher, Giang, Patel.

Critical revision of the manuscript for important intellectual content: Auerbach, Kripalani, Vasilevskis, Sehgal, Lindenauer, Metlay, Fletcher, Ruhnke, Flanders, Kim, Williams, Thomas, Herzig, Boscardin, Robinson, Schnipper.

Statistical analysis: Auerbach, Sehgal, Patel, Boscardin.

Obtained funding: Auerbach, Vasilevskis, Lindenauer, Metlay, Ruhnke, Kim, Robinson, Schnipper.

Administrative, technical, or material support: Auerbach, Vasilevskis, Lindenauer, Fletcher, Ruhnke, Flanders, Kim, Williams, Thomas, Giang, Robinson.

Study supervision: Kripalani, Vasilevskis, Sehgal, Fletcher, Ruhnke, Kim, Thomas, Herzig, Schnipper.

Funding/Support: Dr Auerbach is supported by grant K24HL098372 from the National Heart, Lung, and Blood Institute. This work was supported by an unrestricted research grant from the American Association of Medical Colleges and in part by grant 2 UL1 TR000445-06 from the National Institute on Aging.

Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Contributions: Association of American Medical Colleges employees Anne Bonham, MD, PhD, Mildred Solomon, PhD, and Ellen Sakeld, PhD, supported this program. We also acknowledge the Hospital Medicine Reengineering Network and the Association of American Medical Colleges. We would also like to acknowledge the support and assistance of the dozens of physicians, unit teams, and patients who helped carry out this project.