Potentially preventable hospital readmissions after patients’ first stroke in Taiwan

Readmission is an important indicator of the quality of care. The purpose of this study was to explore the probabilities and predictors of 30-day and 1-year potentially preventable hospital readmission (PPR) after a patient’s first stroke. We used claims data from the National Health Insurance (NHI) from 2010 to 2018. Multinomial logistic regression was used to assess the predictors of 30-day and 1-year PPR. A total of 41,921 discharged stroke patients was identified. We found that hospital readmission rates were 15.48% within 30-days and 47.25% within 1-year. The PPR and non-PPR were 9.84% (4123) and 5.65% (2367) within 30-days, and 30.65% (12,849) and 16.60% (6959) within 1-year, respectively. The factors of older patients, type of stroke, shorter length of stay, higher Charlson Comorbidity Index (CCI), higher stroke severity index (SSI), regional hospital, public and private hospital, and hospital in the lower urbanized area were associated significantly with the 30-day PPR. In addition, the factors of male, hospitalization year, and monthly income were associated significantly with 1-year PPR. The ORs of long-term PPR showed a decreasing trend since implementing the national health insurance post-acute care (PAC) program in 2014 and a dramatic drop in 2018 after the government expanded the long-term care plan-LTC 2.0 in 2017. The results showed that better discharge planning, implementing post-acute care programs and long-term care plan-LTC 2.0 may benefit the care of stroke patients and help reduce long-term readmission in Taiwan.

www.nature.com/scientificreports/ after initial admission with a stroke diagnosis where the aforementioned diseases were not diagnosed at the time of readmission. Patients who were not readmitted to the hospital were defined as having no readmission.
Covariates and subgroups. The covariates in this study included sex, age group, year of first admission for stroke, length of stay (LOS), stroke type (ISC: ischaemic stroke (codes 433-434), ICH: intracerebral hemorrhage (codes 431-432), SAH: subarachnoid hemorrhage (codes 430), and other (codes 435-437), monthly income based upon the NHI premium each patient paid, which was used as a proxy for income, hospital level, ownership, and region, and urbanization level, in which level 1 represents the most urbanized area and level 5 the least 26 . A modified version of the Charlson Comorbidity Index (CCI) was used to summarize comorbidities 27,28 .
In the 1-year follow-up, the LOS included the initial hospitalization and hospital stays later during the year. The comorbidities with PPR and non-PPR after discharge post-stroke were calculated according to 29 diagnosed diseases (Supplementary Table S2), which include any primary and secondary diagnosed conditions in outpatient or inpatient data during the period between the first admission for stroke and readmission post-discharge. The items included in the Stroke Severity Index (SSI) essentially reflect the management of stroke-related complications, and are generally correlated with stroke severity and other accompanying neurological deficits. We extracted the above claims information from the inpatient claims database at first admission for stroke and then computed each patient's SSI. Following a previous study, patients were categorized as having mild (SSI ≤ 5), moderate (SSI 5 to ≤ 12), or severe (SSI > 12) stroke 29 . Data analysis. Descriptive statistics were used to summarize all of the covariates considered in this study, in which categorical variables were analyzed using Pearson's Chi-squared test, and continuous variables (LOS) were analyzed using the t-test. In this study, separate models were built to examine the covariates associated with readmission status within 30-days and 1-year. Multinomial logistic regression (MLR) was performed to determine the association between related factors and readmissions. Three levels were defined for the dependent variable, readmission status: PPR patients, non-PPR patients, and no-readmission patients. Among them, no readmission was set as the reference level. The MLR results were presented as odds ratios (ORs), 95% confidence intervals (95% CIs), and p-values. Statistical significance was set at p < 0.05. All statistical analyses were performed with SAS software v. 9.4 (SAS Institute Inc., Cary, NC, U.S.A.).
The comparison results showed the significant covariates associated with readmission status within 30-days: age, monthly income, year, stroke type, LOS, CCI, SSI, hospital level, hospital ownership, and urbanization. For both 30-day and 1-year readmissions, the PPR rate of these significant factors was more likely to be higher than non-PPR, and the highest PPR rate of subgroups among the significant factors were one patient 80+, patients with a monthly income of NT$19,048-21,900, those treated in 2010 and 2012, those with ICH, those with CCI 7+, those with severe SSI, and those treated at district, private, and the least urbanized area hospitals.
The mean LOS within 30-days was 11.51 (SD = 9.59) for PPR and 11.25 (SD = 9.75) for non-PPR and within 1-year were 18.56 (SD = 22.97) for PPR and 19.24 (SD = 25.08) for non-PPR. The standard deviation of LOS within 1-year appeared to be large, which indicates that the LOS values were distributed over a broader range.
The results of the multinomial logistic regression are shown in www.nature.com/scientificreports/ Table 2. Factors associated with PPR and non-PPR vs. non-readmission after stroke (multinomial logistic regression). *p < 0.05, **p < 0.01, ***p < 0.001.  Fig. 2 shows the forest plot of the odds ratios and 95% confidence intervals for factors associated with 30-day and 1-year PPR.

Discussion
Studies have found that the rates of hospital readmissions after a stroke ranged from 6.5 to 24.3% within 30-days 5 and 31 to 49% within 1-year 11 . However, not all readmissions are considered "potentially preventable" 30 . A review paper reported that preventable readmissions ranged from 14 to 23% within 30-days and from 48 to 59% within 1-year based upon older patients or general medical patients 15,31 . Another study estimated that the 1-year cumulative risks of readmission for ischemic stroke patients in Taiwan were 34.1%, 44.7%, and 62.9% for patients with mild, moderate, or severe stroke, respectively 32 . In this study, we determined that hospital readmission rates were 15.48% within 30-days and 47.25% within 1-year; the PPRs based upon the PQI definition were 9.84% within 30-days and 30.65% within 1-year using population-based data in Taiwan. Because our study included only patients older than 18, the readmission rates shown should be the upper bound. Mittal et al. found that 41 (7.6%) of 537 acute ischemic stroke (AIS) patients were readmitted within 30-days post-stroke, and 2.8% among them were PPR 23 . Based upon 79 unplanned readmissions, an investigation at a Hong Kong geriatric center found that only 15 cases (19%) were avoidable 33 . These variations in PPR rates may be associated with age, patient diagnosis, duration of follow-up, methodology, and factors related to the mixtures of case diagnoses 15 . Nakagawa et al. found in a multiethnic population in Hawaii that 840 (8.4%) of 10,050 patients with any type of stroke-related hospitalization had 30-day PPR. Some studies have demonstrated that a higher readmission rate may be attributable to language barriers that affect receiving hospital care and/ or accessing post-hospital care 30,34 .
The extant literature has reported that certain patient characteristics, such as age and socioeconomic status, were potential factors associated with readmission after stroke 5,8,15 . Our study found the same effect of age, but patients with the highest and lowest monthly income had a significantly higher rate of readmissions than those with the median income. This may indicate inequalities in healthcare and additional investigation is necessary to determine the reasons 35 . Regarding the age effect, the results showed that patients in the other age groups were at higher risk for short-term and long-term preventable readmissions than patients in the youngest age group. In contrast, the risk of non-preventable readmission was lower than in the youngest age group. It indicated that most readmissions in the younger age group are due to unpreventable factors.
The severity of stroke upon the first admission was also a significant predictor of 28-day readmission in Australia 5 . Further, CCI was found to be associated with the 30-day PPR after stroke discharge 30 . In this study, we identified a direct, positive relation between age, CCI, SSI, and long-term PPR. In cases of long-term PPR, the increase in these factors was associated with increasing readmission. These findings were similar to a 234 hospital-based study in Florida, which found that PPR was related to the severity of illness and older age. In addition, their results showed that increased severity of the disease and time between admission and readmission increased readmission rates 14 .
After adjusting for other variables, regional hospitals showed a higher risk of PPR compared to medical centers and district hospitals. The effect of hospital-level on short-and long-term readmissions was consistent with those of previous studies 32, 36 . We assume that medical centers provide a better quality of inpatient care 32 , and suggest that regional hospitals' policymakers give more attention to the quality of patient care. In addition, the fact that district hospitals had lower PPR than regional hospitals may be attributable to the implementation of the Post-Acute Care (PAC) program in Taiwan described in the next paragraph. The district hospitals received more PAC patients, which led to a decreasing readmission rate.
Our results showed that the hospitals' urbanization level was related significantly to both short-and longterm PPR; the most urbanized area had the lowest readmission rate compared to the least urbanized area. One study suggested that this may be related to the poor quality of care in rural areas 37 . Most discharged stroke patients still need to receive follow-up healthcare at home or in a skilled nursing or inpatient rehabilitation facility; however, those resources may not be allocated sufficiently in rural areas compared to urban areas 38 . As a result, the quality of post-discharge care in rural areas may be poorer than that in urban areas and have led to a higher readmission rate.
Our study demonstrated further that, compared to no-readmission patients, a 1-day increase in LOS was associated significantly with 0.97 times the risk of 30-day PPR. However, a 1-day increase in LOS was associated significantly with 1.01 times the risk of 1-year PPR. Hence, LOS may have different implications for short-and long-term PPR. This finding is consistent with that in Bjerkreim's study 11 . LOS' short-term effect on PPR may be explained by incomplete treatment during the index hospitalization 14 , and suggests the need for a better quality of care and discharge planning. On the other hand, LOS' long-term effect on PPR may be related to the severity of the stroke or comorbidities 10,39,40 , and suggests the need to improve the continuity of follow-up care.
The comorbidities associated with PPR diagnosed most frequently in our study were hypertension without complications, diabetes, and congestive heart failure. Previous reports have indicated that patients who were readmitted either early or later seemed to have higher frequencies of hypertension, atrial fibrillation, cerebrovascular disease, and diabetes as prior comorbidity conditions 10,11 . To decrease the risk of short-term PPR after discharge, our results showed that older patients, stroke type (ICH), CCI level of 4-6 and 7+, either moderate or severe SSI, and patients treated at regional, public or private, and hospitals in less urbanized areas are the groups most likely to experience a first-ever stroke, which suggest that adequate discharge planning must be provided for the first month after these patients are discharged. Although a previous study indicated that readmission reduction initiatives might not be highly effective for patients who are socioeconomically disadvantaged 41 , we found that more attention should be given to median-income patients to decrease readmission rates. www.nature.com/scientificreports/ An important finding was that the ORs of long-term PPR vs. no readmission showed a decreasing trend. We believe that this is attributable to Taiwan's implementation in 2014 of the national health insurance postacute care (PAC) program for first-ever stroke patients. Patients who qualify for the PAC can receive intensive rehabilitation and integrated care within the treatment period. The PAC plan proposes to improve the incentives and review of discharge care for stroke patients in these hospitals. We suspect that the sudden drop in PPR and non-PPR readmission in 2018 might be due to the government expansion of the long-term care plan-LTC 2.0 in 2017, which encourages hospitals to provide more discharge plan services and offer more home-based medical care to the patients. The LTC-2.0 may benefit the care of stroke patients and decrease readmission. We suggest further research to verify this relationship.
PPR events may be avoided and healthcare costs reduced by improving the quality of care during the index inpatient stay and the period immediately following discharge. As a consequence, our study suggests that specific groups of patients should be targeted for PPR intervention.

Limitations
This study has certain limitations. First, it was a retrospective cohort study with data derived from Taiwan's National Health Insurance (NHI) claims database. Although we adopted rigorous definition of stroke to selection patients, overdiagnosis or underdiagnosis could not be overall excluded. In addition, data on certain important factors, such as patient behavioral characteristics, the process of care, and health-related quality of life could not be collected 23 . Nonetheless, compared to a hospital chart review database, the NHI claims database provides a population-based sample from a wide range of hospitals, as well as longitudinal follow-up information on readmission post-stroke. Second, the diagnosis codes' accuracy was uncertain. Therefore, the results may be limited to patients who are hospitalized with a primary discharge diagnosis of stroke in Taiwan 36 . Third, identifying the risk of readmission may help with interventions to reduce readmissions. However, they cannot always be considered preventable, and readmission conditions such as the risk of deterioration of the chronic pulmonary disease, diabetes complications, heart failure, new stroke events, etc., cannot be eliminated entirely. Forth, Due to the data availability, we couldn't include post-discharge death in our analysis. The results for the separate predictive model of death or potentially avoidable readmission may differ, and it is valuable for future researchers to explore it.

Conclusion
We suggest that hospital managers provide better discharge planning and post-discharge follow-up programs for these patients before and after discharge, as the combination is likely to reduce the number of PPR substantially. www.nature.com/scientificreports/