Article Text

The association between patient experience factors and likelihood of 30-day readmission: a prospective cohort study
1. Jocelyn Carter1,
2. Charlotte Ward2,3,
3. Deborah Wexler4,
4. Karen Donelan1,5
1. 1 Department of Medicine, Massachussetts General Hospital, Boston, Massachusetts, USA
2. 2 Center for Healthcare Studies, Northwestern University, Bridgeview, Illinois, USA
3. 3 Center for Health Statistics, University of Chicago, Chicago, Illinois, USA
4. 4 Diabetes Center, Massachusetts General Hospital, Boston, Massachusetts, USA
5. 5 Mongan Institute for Health Policy Centre, Massachusetts General Hospital, Boston, Massachusetts, USA
1. Correspondence to Dr. Jocelyn Carter, Department of Medicine, Massachussetts General Hospital, 50 Staniford Street, Suite 503A, Boston 02114, MA, USA; jcarter0{at}partners.org

## Abstract

Objective Hospital care comprises nearly a third of US healthcare expenditures. Fifteen to 20 per cent of this spending is considered to be potentially preventable. Risk prediction models have suboptimal accuracy and typically exclude patient experience data. No studies have explored patient perceptions of the likelihood of readmission during index admission. Our objective was to examine associations between patient perceptions of care during index hospital admission and 30-day readmission.

Design Prospective cohort study.

Setting Two inpatient adult medicine units at Massachusetts General Hospital, Boston, Massachusetts.

Participants Eight hundred and forty-six patients admitted to study units between January 2012 and January 2016 who met eligibility criteria and consented to enrolment.

Main outcome Odds of 30-day readmission.

Conclusion Participants reporting high satisfaction and good provider communication were less likely to be readmitted. Rates of readmission were increased among participants stating they were very likely to be readmitted though this association was not statistically significant. Incorporating patient-reported measures during index hospitalisations may improve readmission prediction.

• patient satisfaction
• patient-centred care
• hospital medicine

## Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

## Introduction

In 2011, there were approximately 3.3 million adult 30-day all-cause hospital readmissions in the USA, generating $41.3 billion in hospital costs.1 These readmissions accounted for 32.1% of the total healthcare expenditures in the USA.2 Conservatively, up to$8.26 billion (15–20%) of readmissions are considered to be potentially preventable.3 Reasons for readmissions are multifactorial. Numerous studies have demonstrated correlations between hospital readmissions and specific disease comorbidities/conditions,4–7 older age,3 8 general poor health9 or high previous healthcare utilisation.6 10 11 Racial/ethnic disparities,9 12–13 low educational attainment,14–15 socioeconomic status16–18 and lack of social support19 have all been cited as core contributors to readmission rates and/or poor care transitions.

In order to better understand the relationship between inpatient patient-reported measures and 30-day readmissions, we interviewed patients during index hospital admissions and subsequent readmission over a 4-year period. We hypothesised that patient-reported measures would inform our understanding of the factors driving hospital readmission, generate novel predictors of readmission and identify appropriate targets for interventions during the hospital stay.

## Methods

### Study population and setting

The population consisted of inpatients admitted to one of two inpatient medical units at the Massachusetts General Hospital between January 2012 and January 2016. Patients eligible for the study were greater than 18 years of age, had the capacity to complete the questionnaire and spoke English (figure 1). Patients were excluded if their readmission was planned or scheduled, or if admission and discharge occurred in a time interval that made patient consent impossible. The two units included medical patients with similar distributions of insurance coverage and medical conditions. Eligible study participants were identified by daily surveillance of the census on both units. Efforts were made to enroll all eligible patients admitted to the study units, but for practical reasons (staffing, shift changes, etc) not all potentially eligible patients were able to be approached and consented prior to discharge. There were no significant differences in the demographic characteristics, proportion of disease categories or baseline readmission rates of study participants and those that were eligible for the study but not enrolled.

Figure 1

Flow diagram for study participation and survey January 2012–January 2016.

### Sources of data

#### Patient surveys

The patient experience data reported here are drawn from interviewer-administered surveys conducted with eligible and consenting patients during index admissions. The patient survey was developed by study investigators for interviewer-assisted administration with hospitalised patients. We used a qualitative process to identify core domains through key informant interviews with patients, community-based primary care physicians and facility providers. This was coupled with review of the literature on hospital patient experience of care surveys and consultations with experts in survey and health services research. For benchmarking, we used some standard established measures of the patient experience from Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS).28 Draft surveys were pretested with 30 patients and revised prior to study administration. The final index admission questionnaire included 20 items: patient perceptions of physical and mental health, confidence in ability to perform self-care after discharge, satisfaction with inpatient care received, patient-predicted likelihood of readmission within 30 days, understanding of the care plan, basic demographic information (language preference, highest level of education, racial/ethnic identity) and presence of a caregiver at home. Open-ended questions queried patients regarding anything that would help them manage their health outside the hospital. Text and response scales are detailed in the tables and in the text of the Results section; the full questionnaire is available in an online appendix

### Supplementary file 1

Trained research coordinators interviewed patients and verbally administered questionnaires on the day of or day prior to discharge after obtaining verbal consent. Patient index admission questionnaires took between 10 min to 12 min to complete. Patient responses were recorded and stored in a secure online database (REDCap V.7.0.14, Vanderbilt University).29

### Clinical data repository

In addition to patient-reported data, research coordinators completed a structured medical record review to obtain clinical history using the electronic medical record with data captured in the same REDCap database. Abstracted data included demographic information, insurance status, education, primary language spoken, homelessness, primary diagnosis associated with admission, and major medical and psychiatric comorbidities including substance use disorder. Both chart review and patient interview responses were used to construct clinical or system-based underlying causes for admission.

### Statistical analysis

Thirty-day readmission was the primary outcome. The primary covariates of interest were participant responses to items on the admissions questionnaire. Items which had more than two response categories were dichotomised based on the most positive response category. For example, for self-rated overall health on a 5-point scale from ‘poor’ to ‘excellent,’ ‘excellent’ is the most positive response category and would be coded as ‘1’, while all other responses would be recoded as ‘0’. Due to the small amount of missing data from survey items (ranging from 0% to 4.7% with an average of 1.1%), missing data were excluded from the analysis. Demographic covariates included gender, age category, race, ethnicity, marital status, education, insurance status, primary language spoken, history of homelessness or a substance use disorder.

Bivariate analyses, using Pearson’s χ2 tests, were performed to assess differences between readmission status groups on all demographic and survey item variables. Associations between readmission status and survey item responses were assessed using separate multivariable logistic regression models for each survey item controlling for demographic covariates which had a bivariate P value <0.25. A P value of <0.05 was considered statistically significant. The proportion of missing data (as noted in the unadjusted results tables) was assessed for random missingness at values <5%; the rate of missing data for all variables had negligible missingness and was 1.1% on average. All analyses were performed using SAS software V.9.4 (SAS Institute, Cary, North Carolina, USA).

## Results

Eight hundred and forty-six patients were enrolled and completed admission interviews (figure 1) during their index admission. These patients represent 55.4% of unique eligible patients admitted for index admissions during the study period who were screened and invited to participate in the interview. Of the 846 enrolled, 201 of these patients were readmitted to the index hospital within 30 days of their hospital discharge. Participant baseline characteristics are listed in table 1. Participants who were readmitted were less likely to have been educated beyond a high school degree or equivalent compared with those who were not readmitted (44.3% vs 53.5%, P=0.02).

Table 1

Patient characteristics

The most common primary reasons for admission were infectious disease, respiratory, gastrointestinal bleeding, cardiac or psychiatric diagnoses. Those who were readmitted were less likely than those not readmitted to have been diagnosed with a respiratory condition on index admission (18.9% vs 28.8%, P=0.02).

Table 2

Table 3

## Discussion

In this survey of 846 hospitalised general medical patients, we found that certain patient-reported measures predicted readmissions after adjusting for demographic characteristics. In particular, respondents who were very satisfied with overall care received were 39% less likely to be readmitted. We also found that respondents who indicated that their doctors always listened to them carefully also had lower rates of readmission. This group of respondents was 32% less likely to be readmitted than those who indicated that doctors did not always listen to them carefully.

A handful of studies have attempted to predict readmission using patient-reported measures,27 30–31 but most have used postdischarge data. None of these studies have considered patient-perceived likelihood of readmission at the time of index admission. In this cohort, patient prediction of their own likelihood of readmission was not statistically significant. However, responses among larger or different study populations may have produced different results.

Though not statistically significant, we noted that respondents who indicated that doctors, nurses or other hospital staff always talked to them about whether they will have help after the hospital stay were more likely to be readmitted. While seemingly counterintuitive, this association likely stems from allocation bias driven by systematic and routine hospital-based initiatives to improve patient care for those who clearly have increased needs for support.35

Some of the correlates of readmission that we identified are similar to prior reports, while others are unique. Increased likelihood of readmission among patients with low educational attainment underlines ongoing disparities of populations with socioeconomic challenges. This is consistent with what has been seen in other cohorts, as well as with the well known relationship between socioeconomic status and health outcomes.24 While we did not explore the domain of preventability here since we think that this is best presented in the context of identified reasons for readmission, we believe that both clinical and socioeconomic factors are central in this sphere. Although respondents over the age of 65 years were more likely to have less than a high school education, this finding persisted after controlling for age. We identified many factors that have not been reported previously. There was no increased likelihood of readmission associated with specific levels of insurance, although many prior studies have demonstrated that readmission rates are higher among Medicare or Medicaid/self pay populations.6 This lack of association may be related to the fact that universal health insurance coverage in the Commonwealth of Massachusetts was in place during the study period. We also found in our sample that respondents over the age of 45 years were more likely to be readmitted relative to younger patients. This is different from other studies that have demonstrated vulnerability with regard to readmission in terms of populations older than age 65 years.3 10 14 16 We also found that marital status was not particularly discriminating with regard to determining risk for readmission, in contrast to some prior reports.14 23 It seems possible that the study participant sample, which was drawn from a medicine service designed to care for complex patients at a tertiary care centre, drove clinical complexity across participants from each age category, and that the risk factors for readmission may differ in the setting of greater burden of illness.

There were a number of limitations of this study. It was not possible to track readmissions to other health systems which may have resulted in an underestimation of readmissions. Also, ‘healthier’ user bias, with patients that were sickest being unable to complete the survey, may have resulted in under-representation of patients with even higher rates of medical complexity. The fact that this study was conducted on a single medical service among English-speaking patients also limits the generalisability of our findings.

## Conclusions

The impact of patient-reported measures with respect to hospital discharge and care transitions has yet to be fully realised in terms of predicting and preventing readmission. This is the first study to evaluate the relationship between the patient-reported outcomes and patient experience data during index admission and 30-day hospital readmission rates. We found that lower rates of readmission were associated with higher rates of patient satisfaction and higher level of patient perception of providers listening to them. We were not able to find a statistically significant association between those who stated that they were more likely to be readmitted within 30 days after discharge and those that were actually readmitted. However, the observation that the estimate suggested an increased risk warrants further exploration in different populations and should be explored in larger studies. Additional research is needed to study the integration of patient experience data into models to identify patients at risk for readmission prospectively.

## Acknowledgments

The authors thank all of the research coordinators that contributed to this work including Lindsey Lebel, Courtney Kaiser,Thomas Cunningham, Catherine Caughlin, Rebecca Kogan, Aradhna Agarwal, Noemia Nau, Layne Keating, Rachel Gallimore, Chelsea Hicks, Keneolisa Ogamba, Yolaine N Kamga, Rachel Weinstock, Montane Silverman, Muhammad Farhan Asghar. The authors also thank Yuchiao Chang, PhD of the Division of General Internal Medicine at Massachusetts General Hospital for senior statistical support.

## Footnotes

• Handling editor Kaveh G Shojania

• Contributors JAC, DW and KD designed the study and outcomes of interest. JAC wrote the original manuscript. DW, KD and CW edited the manuscript. CW performed all statistical analysis.

• Funding 2012 Clinician Teacher Development Award, MGPO, Massachusetts General Hospital

• Competing interests DW reports serving as a consultant to Novartis on the management of hyperglycaemia associated with a novel oncology drug in May 2016.

• Ethics approval The institutional review board at the Partners Human Research Committee.

• Provenance and peer review Not commissioned; externally peer reviewed.

• Data sharing statement Anonymised full data set and statistical codes are available upon request.

• Correction notice A change has been made to the original version of this article. The following sentence in the abstract was updated “Hospital readmissions comprise nearly a third of US healthcare expenditures” to be changed to “Hospital care comprises nearly a third of US healthcare expenditures.”