Association between early outpatient visits and readmissions after ischemic stroke

Abstract

Background

Reducing hospital readmission is an important goal to optimize post-stroke care and reduce costs. Early outpatient follow-up may represent one important strategy to reduce readmissions. We examined the association between time to first outpatient contact and readmission to inform post-discharge transitions.

Methods and Results

We performed a retrospective cohort study of all Medicare fee-for-service patients discharged home after an acute ischemic stroke in 2012 identified by ICD-9-CM codes. Our primary predictor variable was whether patients had a primary care or neurology visit within 30 days of discharge. Our primary outcome variable was all-cause 30-day hospital readmission. We used separate multivariable Cox models with primary care and neurology visits specified as time-dependent covariates, adjusted for numerous patient- and systems-level factors. The cohort included 78,345 patients. Sixty-one percent and 16% of patients, respectively, had a primary care and neurology visit within 30 days of discharge. Visits occurred a median (IQR) 7 (4–13) and 15 (5–22) days after discharge for primary care and neurology, respectively. Thirty-day readmission occurred in 9.4% of patients. Readmissions occurred a median 14 (IQR 7–21) days after discharge. Patients who had a primary care visit within 30 days of discharge had a slightly lower adjusted hazard of readmission than those who did not (HR 0.98, 95% CI 0.97–0.98). The association was nearly identical for 30-day neurology visits (HR 0.98, 95% CI 0.97–0.98).

Conclusions

Thirty-day outpatient follow-up was associated with a small reduction in hospital readmission among elderly stroke patients discharged home. Further work should assess how outpatient care may be improved to further reduce readmissions.

Introduction

Reducing hospital readmission is a focus of national quality initiatives designed to minimize costs of care and the burden of disease.¹ The annual cost of unplanned rehospitalizations has been estimated at over $17 billion for Medicare alone²; thus, this focus is likely to sharpen as public reporting of readmission data becomes more common and as financial incentives for reducing readmissions increase. Stroke readmissions are a priority because they are common (estimated 17% at 30-days and 42% at 1-year) and associated with high mortality, morbidity, and cost.^{3, 4} Studies have identified many potential factors associated with hospital readmissions after a stroke including but not limited to demographics (i.e. age, sex, race), comorbidities (i.e. cardiovascular disease, past stroke), hospital and hospitalization characteristics (i.e. length of stay, discharge location, inpatient complications, rehabilitation intensity), and stroke characteristics (i.e. NIH stroke scale).^5–13 However, patient- and system-level factors associated with readmission have been inconsistent across studies¹; thus, our understanding of inpatient and outpatient practices which might prevent readmissions is incomplete.

One strategy to reduce readmissions focuses on optimizing transitional care during the immediate post-discharge period. The transition between inpatient and outpatient care represents an especially vulnerable period for post-stroke patients who are often elderly with multiple complex chronic conditions, have extensive care needs, and are accommodating new disability. The set of actions necessary to ensure coordination and continuity of care as patients transfer between locations and levels of care is complex and little is known about how to best facilitate transitions.^{14, 15} Interventions designed to optimize transitional care might reduce undesirable outcomes such as readmission by ensuring rapid follow-up of diagnostic testing, improving secondary prevention, coordinating rehabilitative therapies, detecting and treating new symptoms after discharge, and addressing social needs of patient and caregivers. Current guidelines provide little information regarding follow-up time and more broadly the optimal set of interventions to reduce readmissions.¹⁶

Clinical studies in patients recently hospitalized for heart failure, general medical conditions, or various surgeries have suggested that early outpatient provider contact after hospital discharge may improve outcomes and reduce readmissions.^17–19 However, comparing across disorders is challenging. We therefore sought to assess the association between early outpatient post-discharge follow-up and hospital readmissions to determine whether early outpatient physician contact with a primary care physician and/or neurologist is associated with lower 30-day readmissions after an acute ischemic stroke. These data may inform future readmission-reduction initiatives and improve our understanding of the relative contributions of inpatient and outpatient care on readmissions.

Methods

The analytic methods including Stata code have been made available to other researchers for purposes of reproducing our results or replicating our procedures. ²⁰ Because of the sensitive nature of the data collected for this study, requests to access the dataset from qualified researchers trained in human subject confidentiality protocols may be sent to the Research Data Assistance Center at https://www.resdac.org/about-resdac/contact-us.

Study Design and Patient Selection

We performed a retrospective cohort study of all Medicare fee-for-service beneficiaries 65 years or older discharged directly home after an acute ischemic stroke between January and November 2012. Patients were included only if they were admitted within the first 11 months of 2012 to ensure that at least 30 days were available to identify readmissions. Potential subjects were identified if they had an acute inpatient hospitalization, with a primary discharge diagnosis of ischemic stroke (ICD-9-CM 433.x1, 434.x1, 436).^21–24 Subjects were included if the index hospitalization occurred in a short stay hospital (identified using Medicare provider numbers) and if patients were discharged directly to home. We excluded patients discharged to post-acute care facilities (i.e. acute and sub-acute rehabilitation facility, or long-term hospital) for many reasons: 1. There is considerably more variation in stroke severity and other markers of disease severity in patients discharged to institutional rehabilitation setting than to patients discharged home.²⁵ Given that severity is a likely predictor of which patients have early outpatient visits, our approach of focusing on discharge home is a strategy to mitigate this bias. 2. Conceptually, the role of outpatient visits on readmission is substantially attenuated in patients discharged to institutional rehabilitation. The average length of stay in a skilled nursing facility or acute rehabilitation facility averages over 17 days and for more than ¼ of cases is over 30 days so there is a very limited window for outpatient visits prior to the 30-day outcome. 3. There is marked variation in the quality of care provided in these facilities^{26, 27} and, without good measures of quality of care in rehabilitation facilities, it would be impossible to disentangle the effects of outpatient visits from the quality of rehabilitation care. 4. There is marked variation in the utilization of different rehabilitation settings in different regions and without robust clinical data it would be difficult to account for this selection in our analyses. Of the 78,345 eligible patients, 64,712 patients (83% of original cohort) had complete data capture included in our two fully adjusted models.

The study protocol, which does not rely on human subjects, was deemed not regulated by the University of Michigan Institutional Review Board.

Data and Variables

We used Medicare Inpatient files to identify the cohort (acute ischemic stroke patients hospitalized January–November 2012), patient-level and systems-level variables, and identify readmissions. The Medicare Carrier file was used to identify and characterize outpatient visits. Transfers of an index stroke case to another acute care hospital were excluded. Readmission dates were assigned based on the discharge date from the receiving hospital.

Our two binary primary predictor variables consisted of whether patients had outpatient visit within 30 days to 1) primary care and 2) neurology. Primary care was defined as general practice, family medicine, internal medicine excluding subspecialist visits, and geriatric medicine. Outpatient visits were defined as visits that occurred after the date of discharge in any non-inpatient or emergency department setting. The provider specialty of the outpatient visits (primary care and neurology) was identified from carrier claim billing data. We performed multivariable models assessing what baseline factors were associated with 30-day outpatient visits.

Our binary primary outcome was all-cause 30-day hospital readmission. This was defined as any inpatient visit in a short stay hospital with an admission date greater than the discharge date of the index hospitalization, regardless of length of stay. Visits to the Emergency Room which did not result in readmission were identified based on place of service codes and were not counted as a readmission. We documented primary readmission diagnoses.

We accounted for a large array of potential confounders including patient demographics, comorbidities, clinical characteristics including life-sustaining measures (percutaneous gastrostomy tube, intubation), regional socioeconomic factors, and hospital characteristics and quality measures. These are listed in Table 1. Patient demographics including age, sex, and race were abstracted from the Inpatient and Medicare Beneficiary Summary Files. Comorbidities were identified using modified Charlson definitions based on data from the index admission.²⁸ Hospitalization characteristics, such as length of stay and whether a patient received thrombolysis, were defined using previously described methods.⁸ Regional socioeconomic factors (segregation index, household income, high school graduation rate) were based on county level census data and were obtained from the Robert Wood Johnson County Health Rankings and Roadmaps project.²⁹ Hospital characteristics included Medicare stroke volume and hospital-level quality measures that were obtained from Medicare’s Hospital Compare website and included eight stroke process measures included in Hospital Compare in 2012 (i.e. STK1-STK6, STK8, STK10).³⁰

Table 1.

Baseline characteristics

	30-day Readmission
	No (N = 70,973)	Yes (N = 7,372)	p*
Demographics
Age in years, mean (SD)	77.8 (8.0)	77.8 (8.3)	0.76
Female sex, No. (%)	32,990 (46.5%)	3,413 (46.3%)	0.76
Race			<0.01
Unknown	1,698 (2.4%)	165 (2.2%)
White	58,816 (82.9%)	5,916 (80.2%)
Black	7,979 (11.2%)	1,010 (13.7%)
Asian	1,151 (1.6%)	115 (1.6%)
Hispanic	1,329 (1.9%)	166 (2.3%)
Comorbidities, No. (%)
Myocardial infarction	6,179 (8.7%)	764 (10.4%)	<0.01
Peripheral vascular disease	7,375 (10.4%)	1,013 (13.7%)	<0.01
Congestive heart failure	9,284 (13.1%)	1,390 (18.9%)	<0.01
Dementia	3,006 (4.2%)	378 (5.1%)	<0.01
Chronic obstructive pulmonary disease	12,724 (17.9%)	1,689 (22.9%)	<0.01
Rheumatological condition	2,180 (3.1%)	260 (3.5%)	0.03
Peptic ulcer disease	522 (0.7%)	85 (1.2%)	<0.01
Mild liver disease	532 (0.7%)	70 (0.9%)	0.06
Diabetes (uncomplicated)	21,099 (29.7%)	2,359 (32.0%)	<0.01
Diabetes (complicated)	3,226 (4.5%)	452 (6.1%)	<0.01
Hemiplegia	9,694 (12.5%)	1,117 (14.2%)	<0.01
Renal disease	10,698 (15.1%)	1,591 (21.6%)	<0.01
Cancer	2,488 (3.5%)	471 (6.4%)	<0.01
Moderate-severe liver disease	57 (0.1%)	14 (0.2%)	<0.01
Metastases	659 (0.9%)	198 (2.7%)	<0.01
AIDS	21 (0.0%)	3 (0.0%)	0.60
Life-sustaining treatment, No. (%)
Gastrostomy tube	353 (0.5%)	84 (1.1%)	<0.01
Intubation	277 (0.4%)	49 (0.7%)	<0.01
Hemodialysis	756 (1.1%)	209 (2.8%)	<0.01
Hospital stroke quality measures, Mean (SD)
Venous thromboembolism prophylaxis	93.5% (8.1%)	93.4% (8.4%)	0.43
Discharged on antithrombotics	98.8% (2.6%)	98.8% (2.6%)	0.74
Anticoagulation for atrial fibrillation/flutter	95.5% (5.9%)	95.5% (5.8%)	0.48
Thrombolytic therapy	78.2% (16.9%)	78.2% (17.1%)	0.74
Antithrombotics by hospital day two	97.9% (3.1%)	97.7% (3.5%)	0.03
Discharged on statin	94.2% (7.3%)	94.1% (7.4%)	0.83
Stroke education	87.2% (16.5%)	87.5% (16.7%)	<0.01
Assessed for rehabilitation	97.2% (4.1%)	97.2% (4.0%)	0.39
Hospitalization characteristics, No. (%) unless listed otherwise
Stroke volume†	61.2 (43.3)	61.8 (44.8)	0.46
LOS in days, mean (SD)	5.2 (5.6)	5.0 (4.6)	<0.01
tPA	2,650 (3.7%)	249 (3.4%)	0.12
Home Health	25,881 (36.5%)	3,192 (43.3%)	<0.01
Transfer	15,596 (19.2%)	1,169 (15.9%)	<0.01
Regional factors, mean (SD)‡
Graduation rate	82.7% (7.9%)	82.4% (8.0%)	0.01
Household income	$54,023 ($14,812)	$53,035 ($14,222)	<0.01
Segregation index	52.6% (14.0%)	52.0% (14.2%)	<0.01
Outpatient visit within 30 days, No. (%)
Primary care	44,737 (63.0%)	3,159 (42.9%)	<0.01
Neurology	11,978 (16.9%)	558 (7.6%)	<0.01

^*P-values represent t-tests (or Wilcoxon rank sum tests when non-normally distributed) for continuous variables and chi-squared tests for categorical variables. Specifically, continuous variables include age, stroke volume, length of stay, household income, and segregation index. All other variables are analyzed via chi-squared tests.

^†Stroke volume refers to the number of stroke hospitalizations per hospital per year

^‡Selected regional socioeconomic factors are defined as follows. Segregation index measures the degree to which the minority group is distributed differently than whites across census tracts. It ranges from 0 (complete integration) to 100% (complete segregation) where the value indicates the percentage of the minority group that would need to move to create an equal population distributed exactly like whites. Graduation rate refers to high school graduation rate.

AIDS: Acquired immunodeficiency deficiency syndrome. LOS: Length of stay. tPA: Tissue Plasminogen Activator

Statistical Analysis

Descriptive statistics were used to summarize the patient cohort. We evaluated the relationship between outpatient visits and readmissions using Cox models with time of follow-up visits specified as a time-dependent variable while accounting for clustering at the hospital level. If a patient had a second stroke within 30 days, they were still only counted once in the cohort and the second stroke was counted as a readmission assuming they were re-hospitalized for the second stroke. Death of any cause prior to 30 days (or 30-day readmission) was censored in these Cox models. We included outpatient office visits as a time dependent variable given the possibility of survival bias,^{31, 32} i.e. the time to first office visit could be impacted by a readmission. Separate models were constructed using a variable for primary care and neurology visits individually, and another model with variables included to assess their independent effects. Both unadjusted and adjusted models were generated. We adjusted for patient demographics, comorbidities, hospital level performance on all Hospital Compare performance measures, stroke volume, and all regional socioeconomic (SES) variables. All variables were included regardless of statistical significance; we did not employ any model-based variable selection procedures.

We performed a variety of sensitivity analyses. First, given missing data in our SES variables and to assess the effect of each type of variable, we performed serial model building. To this end, we conducted Cox proportional hazard models including only the primary predictor plus age, then added comorbidities and hospitalization characteristics, then added quality measures, then finally the fully adjusted model adding SES variables. Second, we performed competing risk analysis. This analysis was conducted because of the possibility that mortality within 30 days of discharge could prevent 30-day outpatient follow-up and readmission, and thus early post-discharge mortality could be important relative to our small primary effect size. Whereas Cox regressions employ the survival function to calculate the probability of surviving beyond a given time and does not account for the competing nature of multiple causes of the same event, competing risk analysis employs the cumulative incidence function to estimate the marginal probability of each competing event such as mortality. Given this alternative approach to events that would have been censored in Cox regressions, competing risk analysis may more accurately assess the impact of outpatient visits on readmissions. Competing risk models were adjusted for all of the same variables as the fully adjusted Cox models. Third, we performed a model assessing ‘preventable’ readmission as the outcome using definitions described by Lichtman et al according to the AHRQ Prevention Quality Indicators.³³ These include conditions relevant to elderly patients including chronic lung disease, diabetes, cardiovascular indications, and certain acute conditions such as infections and dehydration. This analysis was conducted in order to assess whether primary care and neurology would have a differential effect on preventable readmissions rather than overall readmissions. For example, it is possible that primary care might be more effective at reducing ambulatory-sensitive conditions as defined by the AHRQ (heart failure, pneumonia, diabetes, etc) which tend to be more likely managed by primary care physicians.

All analyses were performed using Stata statistical software, release 14 (StataCorp LP, College Station, TX).

Results

Summary of Study Population and Key Variables

We display a flowchart of our cohort construction in Figure 1. There were 270,725 primary short-stay stroke discharges. After applying exclusions for discharge to care facilities, unavailability of adequate Medicare data, and discharge in the final month of the year (given that they did not have the chance for a 30-day readmission within 2012), 78,345 patients remained. Of these, 64,712 had complete data available for our final fully adjusted models. Nearly all missing data was due to missing regional SES data comprising household income, segregation index, and high school graduation rate. Missing SES data were due to either true missing values, or else imperfect linkages between the measured county level measures and patient-level zip measures.

Figure 1.

Patient flowchart. Exclusions are displayed.

Patient demographics are displayed in Table 1. The overall 30-day readmission risk was 7,372/78,345 = 9.4%. For the 7,372 patients who were readmitted within 30 days of discharge, the median number of days until readmission was 14 (IQR 7–21). For the 47,896 (61%) of patients which had a primary care visit within 30 days of discharge, the median number of days until the first primary care visit was 7 (IQR 4–13). For the 12,536 (16%) of patients which had a neurology visit within 30 days of discharge, the median days until first neurology visit was 15 (IQR 5–22). The full distribution for each of these variables is depicted in Figure 2.

Figure 2.

Time distribution of primary covariates and outcome. The denominator of these graphs is the entire cohort.

A) Distribution of days to first neurology visit

B) Distribution of days to first primary care visit

C) Distribution of days to first readmission

A variety of factors were associated with readmission as listed in Table 1. These included most comorbidities, life-sustaining treatments such as requiring a PEG tube and intubation, hospitalization characteristics (i.e. length of stay and transfer status), and several regional factors. A variety of factors which might have been plausibly related to the outcome were found to be not significantly related to readmissions, including age, sex, hospital stroke volume, and most stroke quality measures.

Table 1 also includes the percentage of patients with and without a 30-day readmission who had 30-day primary care follow-up (42.9% and 63%, p<0.01). Such proportions are listed for 30-day neurology follow-up as well (7.6% and 16.9%, p<0.01). Both results suggest that readmitted patients had lower proportions of outpatient follow-up.

We tabulated the most common causes for readmission in Table 2. We list those diagnoses comprising at least 1% of readmissions for any of the 3 listed groups (overall, those with 30-day primary care visits, and those with 30-day neurology visits) for display purposes. The most frequent types of readmissions included cerebrovascular, cardiovascular, and infectious primary diagnoses. Diagnoses appeared roughly similar between groups.

Table 2.

Readmission diagnoses

	Overall	Primary care	Neurology
Cerebral artery occlusion NOS with infarction	869 (11.8%)	363 (11.5%)	58 (10.4%)
Carotid artery occlusion without infarction	763 (10.4%)	309 (9.8%)	68 (12.2%)
Missing	455 (6.2%)	190 (6.0%)	25 (4.5%)
Transient cerebral ischemia NOS	279 (2.8%)	120 (3.8%)	29 (5.2%)
Urinary tract infection NOS	205 (2.3%)	87 (2.3%)	12 (2.2%)
Cerebral embolism with infarction	195 (2.7%)	94 (3.0%)	22 (3.9%)
Acute renal failure NOS	188 (2.6)	85 (2.7%)	15 (2.7%)
Atrial fibrillation	175 (2.4%)	76 (2.4%)	16 (2.9%)
Carotid artery occlusion with cerebral infarction	173 (2.4%)	61 (1.9%)	9 (3.4%)
Pneumonia, organism NOS	158 (2.1%)	71 (2.3%)	12 (2.2%)
Subendocardial infarction	92 (1.3%)	40 (1.3%)	6 (1.1%)
HIV caused CNS disease	88 (1.2%)	40 (1.3%)	13 (2.3%)
Syncope and collapse	87 (1.2%)	39 (1.2%)	14 (2.5%)
Coronary atherosclerotic native vessel disease	80 (1.1%)	31 (1.0%)	4 (07%)
Food/vomit pneumonitis	71 (1.0%)	32 (1.0%)	1 (0.2%)
Late effects of CVD	70 (1.0%)	28 (0.9%)	8 (1.4%)
Gastrointestinal hemorrhage NOS	63 (0.9%)	23 (0.7%)	7 (1.3%)
Sinoatrial node dysfunction	51 (0.7%)	21 (0.7%)	7 (1.3%)

Association between Patient, Hospital and Regional Factors with Outpatient Visits

We performed separate models to identify factors associated with 30-day primary care and neurology visits. Results are found in Supplemental Table 1. These odds ratios were produced from fully adjusted models including all variables in Table 1. Older age, female sex, uncomplicated diabetes, and discharge with home health were associated with increased 30-day primary care visits but decreased neurology visits. Black race, numerous comorbidities (i.e. myocardial infarction, congestive heart failure, chronic obstructive lung disease, hemiplegia, metastatic malignancy), gastrostomy tube placement, and length of stay were associated with reduced visits for both provider types. Of note per the residual intraclass correlation for each model, 6% of the variance for primary care visits and 15% of the variance for neurology visits was at the hospital level. This suggests that hospital correlates with other factors which may predict readmission.

Association between 30-day Outpatient Follow-up Visit and 30-day Readmissions

Tables 3 and 4 describe readmission risk in more detail according to whether patients received 30-day outpatient physician follow-up for primary care and neurology, respectively. These tables include Cox proportional hazards regressions adjusted only for age and also models fully adjusted for all variables noted in the table. We included 78,294 (99.9% of original cohort) patients in the model adjusted only for age, and 64,712 patients (83% of original cohort) who had complete data capture in our two fully adjusted main models. Patients who had 30-day primary care follow-up had a slightly lower fully adjusted hazard of readmission than those who did not have 30-day primary care follow-up (HR 0.98, 95% 0.97–0.98). In other words, patients who had 30-day primary care follow-up had a 2% reduced readmission rate per unit time compared with patients who did not have 30-day primary care follow-up. Similarly, patients who had 30-day neurology follow-up had a lower fully adjusted hazard of readmission than those who did not (HR 0.98, 95% CI 0.97–0.98). We included variables for 30-day primary care and neurology follow-up in the same model to assess their independent effects, and found that both variables remained independently significant with similar effect sizes (adjusted HR, 95% CI: neurology 0.98 (0.98–0.98), primary care 0.98 (0.97–0.98)).

Table 3.

All-cause 30-day readmission stratified by 30-day primary care follow-up

	No readmission	Readmission	Total	HR adjusted only for age (95% CI) (n=78,294)	Fully adjusted HR* (95% CI) (n=64,712)
No primary care in < 30 days	26,236	4,213 (13.8%)	30,449	0.98 (0.97–0.98)	0.98 (0.97–0.98)
Primary care in < 30 days	44,737	3,159 (6.6%)	47,896
Total	70,973	7,372 (9.4%)	78,345

^*Adjusted for demographics (age, sex, race), comorbidities (myocardial infarction, congestive heart failure, dementia, chronic obstructive pulmonary disease, rheumatological disorders, peptic ulcer disease, mild liver disease, moderate-severe liver disease, uncomplicated diabetes, complicated diabetes, hemiplegia, renal disease, cancer, AIDS), life-sustaining treatment (percutaneous gastrostomy tube placement, tracheostomy, intubation, hemodialysis), hospital characteristics (length of stay, tPA administration, discharge with home health, transfer, stroke volume), all hospital stroke quality measures, and additional regional socioeconomic factors (graduation rate, household income, segregation index).

Table 4.

All-cause 30-day readmission stratified by 30-day neurology follow-up

	No readmission	Readmission	Total	HR adjusted only for age (95% CI) (n=78,294)	Fully adjusted HR* (95% CI) (n=64,712)
No neurologist in < 30 days	58,995	6,814 (10.4%)	65,809	0.97 (0.97–0.98)	0.98 (0.97–0.98)
Neurologist in < 30 days	11,978	558 (4.5%)	12,536
Total	70,973	7,372 (9.4%)	78,345

^*Adjusted for the same variables as in Table 3.

Sensitivity Analyses

Given 17% of patients had at least one missing variable thus excluded from the final fully adjusted models, we performed serial models to better understand the effect of missing data. Results are displayed in Supplemental Table 2. The N values for each model describes the number of patients with complete data capture for the listed variables. Nearly all missing values were due to regional socioeconomic variables (segregation index, household income, graduation rate), and very little data was missing due to any other variables. Our main adjusted hazard ratio was essentially unchanged when performing the displayed serial model building, including the final step when regional variables were included or excluded.

To overcome potential issues related to censorship and competing risk as mentioned in the Methods section, we performed competing risk analysis. The following provides a description of patients who died prior to the opportunity for outpatient follow-up or readmission in order to consider the potential effect of censorship due to the competing risk of death within 30 days. Of the 1,365 (1.7% of total sample of 78,345) who died within 30 days of discharge, 682 had neither 30-day primary care nor 30-day readmission prior to death, 166 had both, 312 were readmitted but did not have a primary care visit, and 205 had a primary care visit but no readmission. Examined in a different way, of the 70,973 patients who were not readmitted within 30 days (i.e. could have been censored in a Cox regression had they not been readmitted due to early death), 1.3% (N=887) died within 30 days and thus may not have had the opportunity for readmission. Using competing risk analysis, despite the above counts of patients who died prior to 30 days, 30-day primary care (HR 0.98, 95% CI 0.97–0.98) and neurology (HR 0.98, 95% CI 0.97–0.98) visits had nearly identical effects on 30-day readmission as in the Cox regressions.

We next performed analysis similar to that presented in Lichtman et al 2013³³ regarding potentially preventable readmissions. Of the 7,370 patients with 30-day readmissions in our sample whose ‘preventability’ could be assessed (out of 7,372 total readmissions), 1,189 (16%) were considered ‘preventable’ by the criteria used in Lichtman et al. Table 5 reports models similar to our main analyses (shown in Tables 3 and 4), except in Table 5 the regression outcome is whether a patient had a ‘preventable readmission’ according to definitions used in Licthman et al 2013.

Table 5.

‘Preventable’ readmission stratified by 30-day provider follow-up

	HR adjusted only for age (95% CI) (n=78,294)	Fully adjusted HR* (95% CI) (n=64,712)
Primary care	0.97 (0.97–0.98)	0.97 (0.97–0.98)
Neurology	0.96 (0.95–0.98)	0.97 (0.96–0.99)

^*Adjusted for the same variables as in Table 3.

Discussion

Effective strategies to achieve the important goal of reducing post-stroke readmissions are largely unknown. Prior research in acute stroke has not identified consistent modifiable patient- or system-level factors associated with readmissions,¹ and thus it has proven challenging to design systems of care to meaningfully impact readmissions.^34–36 While we found that early outpatient visits are associated with a reduction in readmission rates, the magnitude of this effect is small. Thus, novel approaches will be needed to reduce readmission rates. Given that only 61% and 16% of patients visit a primary care provider or neurologist respectively within 30 days, though, our data suggest that there may be a small opportunity to reduce readmission, and potentially improve other outcomes, by improving access to early outpatient care.

Still, even a small reduction in readmission rates by a relative low-cost strategy such as this could a have non-trivial economic impact. Jencks et al 2009² estimates the cost to Medicare of all unplanned readmissions in 2004 was $17.4 billion. Kind et al³ estimated the ‘bounce-back’ cost for acute stroke patients. They calculated for zero bounce-backs adjusted predicted payments ranged from $1,667 at the 10th percentile to $35,854 at the 90^th, patients with one bounce-back had payments ranging from $2,726 to $45,404, and patients with two or more bounce-backs had payments ranging from $3,753 to $53,766 at the 10th and 90th percentiles respectively.

Numerous possible explanations for the small effect size of early outpatient follow-up on readmission exist. One possible explanation is that only a small percentage of readmissions after stroke may be truly preventable, in which case it may not be worth large-scale implementation of rapid outpatient visits.^{13, 33} Many non-modifiable factors likely exist in an elderly post-stroke population with high baseline risk of vascular and non-vascular events even with guideline-based secondary prevention. It should also be mentioned that the optimal readmission rate is unknown but unlikely to be 0. Even though readmissions are costly and often undesirable, it is preferable to appropriately re-hospitalize a patient with additional inpatient needs rather than provide continued inadequate outpatient care, and at least one study documented increased readmissions with increased outpatient access. ³⁷ Furthermore, even if a physician provides an effective treatment plan, systems barriers and socioeconomic factors may prevent a patient from executing a physician’s recommendations. Another possible explanation for our small effect size is that a single early contact with an outpatient provider may be inadequate to meaningfully alter readmission rates given the focus of other studies on longitudinal multidisciplinary interventions.^38–45 A final explanation for our effect size could be that the true effect is larger than detected in this study, but the association was distorted by unmeasured confounding. Sicker patients may be more likely to attend early outpatient visits and be at higher risk for readmission, which would bias the measured association towards the null. Interestingly, in our study, length of stay was slightly shorter for those patients who were readmitted. One may have hypothesized that patients with longer lengths of stay due to higher acuity or additional complications may have been more likely to be readmitted, but this was not the case in our analysis. It is possible that premature hospital discharge could be a driver of readmission.

Some of the more successful care models in stroke and other conditions that aim to reduce readmissions have evaluated multidisciplinary outpatient teams with early or longitudinal interventions.^38–45 For example, a recent observational study found that attending a nursing-led Transitional Stroke Clinic with added interim patient phone contact was associated with significantly lower 30-day (though not 90-day) readmission,⁴⁶ Expedited outpatient care would make intuitive sense as an optimizable target to reduce readmissions through a variety of plausible mechanisms such as encouraging adherence to secondary prevention measures, addressing existing or new symptoms developing after discharge, coordinating therapies, and ensuring rapid follow-up of issues identified by inpatient providers such as pending diagnostic testing. The main drivers of post-stroke readmission include recurrent stroke, coronary artery disease, and infection,⁴ all of which would seem plausibly reducible with optimal medical care including early outpatient follow-up.

Our study has numerous strengths. We addressed survival bias through our statistical analysis. Survival bias (if not addressed) could make an intervention appear falsely effective. This overestimation of effect would occur by counting person-time prior to an outpatient visit as ‘intervention’ person-time thus inflating the denominator and artificially lowering the early follow-up group’s event rate, despite the fact that office visits could have only possibly been effective during or after a visit. Thus, the seemingly large difference between our sample’s risk of readmission for those who did versus did not have a 30-day primary care (6.6% versus 13.8%) or neurology (4.5% versus 10.4%) visit very likely overestimates the effect size if survival bias is not considered. Our time-dependent covariate model addressed this issue by contributing time prior to the office visit as ‘non-intervention’ person-time.^{31, 32} We also had a large sample size of a nationally representative Medicare sample. Finally, we addressed confounding by incorporating a wide variety of potential confounders at numerous levels to account for additional important confounders regarding the patient’s discharge environment and proxy severity measures (i.e. length of stay, intubation, tPA administration, need for post-discharge home health, etc).

Our study has several limitations. First, large administrative data do not capture the content of a given outpatient visit, so the mechanism underlying our studied relationship cannot be definitively determined. Additional research may focus on what elements of a brief outpatient visit may be considered highest value in reducing readmissions and thus how to make outpatient visits more effective. Second, the potential role of unmeasured confounding as discussed above may bias our findings in either direction. Our dataset does not contain the NIH Stroke Scale for individual patients, which is an important marker of stroke severity. Regional measures likely do not adjust for all potential effects of socioeconomic factors at the individual level and unmeasured differences in baseline risk almost certainly still exist. Third, although our study has excellent generalizability in the sense that our nationwide population was drawn from patients across SES and geographic strata, the Medicare population may not generalize to those under 65 years. Limited studies including stroke survivors have documented differing readmission rates according to discharge location,^{7, 47, 48} and our results do not inform the effect of outpatient visits on readmission for the large and important patient population discharged to post-acute care facilities. Additional studies of the population discharged to post-acute care facilities are thus likely needed. Fourth, our fully adjusted models excluded 17% of the initial population (initial population 78,345; analyzed population 64,712) due to missing data primarily in socioeconomic variables, though serial model building such as removing these variables from our model (with resultant very little missing data) produced very similar results. Finally, although our time-dependent analysis reduced bias, we do acknowledge that measuring the outcome and exposure on the same timescale adds some conceptual complexity. Our goal was to understand how much real-world variation in outpatient care may influence readmission rates on the timescale (30 days) measured by current quality measures. Our findings, then, do not preclude the possibility that outpatient visits may have a larger association with readmissions if timing (and other features) were optimized.

Conclusions

In conclusion, 30-day outpatient follow-up was associated with a small but significant reduction in hospital readmission among elderly stroke patients who were discharged from the hospital to home. Our data identifies a potentially underutilized target for future interventions aimed at reducing the cost and burden of stroke. However, the magnitude of effect was quite small, and thus future work should focus on mechanisms to enhance outpatient care. Additional directions of inquiry could include identifying what provider types may have the largest impacts on specific readmission diagnoses at various time points in different patient populations, clarifying the mechanism through which outpatient care might ultimately affects health maintenance and how to make such visits most useful to patients, and studying patients discharged to rehabilitation facilities.

What is known

• Readmissions after stroke are common, costly and burdensome

• A variety of patient and health system factors predict readmission, but little is known about how to prevent readmission

What the study adds

• Early outpatient physician follow-up visits after a hospitalization for an acute stroke has a small but significant association with lower readmission rates

• Both primary care and neurology providers had a similar effect on readmissions