Five-Year Rehospitalization Outcomes in a Cohort of Patients With Acute Ischemic Stroke
Medicare Linkage Study
Background and Purpose—The purposes of this study were to track mortality and rehospitalizations over 5 years poststroke in a stroke cohort (SC) and compare long-term risks of complications to a matched nonstroke cohort (NSC).
Methods—A cohort design with a matched NSC comparison was used. The SC constituted a validated database of acute ischemic stroke patients, ≥65 years, hospitalized across 19 Minnesota hospitals in the year 2000. The NSC was constructed from the year 2000 General Medicare Population by matching SC members on age, race, and sex. Both cohorts were tracked across 5 years of Medicare claims data to identify dates and causes of rehospitalization and death dates. Kaplan–Meier survival curves estimated cumulative incidence rates. Cox models calculated adjusted hazard ratios.
Results—Event rates and adjusted hazard ratios were: mortality: 1 year SC=24%, NSC=4%; 5 years SC=49%, NSC=24% (hazard ratio, 4.4; 95% CI, 3.6 to 5.5). Overall rehospitalization rates were: 1 year SC=49%, NSC=20%; 5 years SC=83%; NSC=63% (hazard ratio, 2.6; 95% CI, 2.2 to 3.0). Cause-specific 5-year rehospitalization rates were significantly higher in SC versus NSC for recurrent ischemic stroke, heart failure, cardiac events, any vascular events, pneumonia, and hip fractures. The excess risk of mortality and rehospitalizations in the SC persisted beyond the initial aftermath of the acute stroke (ie, beyond 30 days poststroke) and persisted even after 1 year poststroke. Average acute care Medicare charges in SC were more than doubled those in NSC.
Conclusions—The high rates of acute care poststroke readmissions indicate a need for trials to prevent long-term complications in stroke survivors.
The estimated 2010 direct and indirect cost of stroke in the United States is $73.7 billion.1 The direct medical costs of stroke include acute hospitalization and rehabilitation costs of the index event and subsequent outpatient care and also the recurrent acute care costs due to rehospitalizations from stroke-related complications.1,–,4 Although there are established and accruing data on the poststroke survival and vascular events after stroke,3,–,9 information on other poststroke complications such as hip fractures, pulmonary embolism, and gastrointestinal hemorrhages over the long term are sparse.10 The goals of the work described here are to track events over a 5-year period in a cohort of patients hospitalized with acute ischemic stroke, ascertain short-term versus long-term risks of various complications leading to subsequent acute hospitalizations, and compare event rates of the stroke cohort with a matched nonstroke cohort.
Overall Study Design
A cohort study design with a stroke cohort and a matched nonstroke comparison cohort was used. Both cohorts were tracked for mortality and rehospitalization outcomes over a 5-year period by linkage with Medicare claims databases. Medicare is a health insurance system for people in the United States ≥65 years and also those with certain disabilities. Approximately 98% of Americans ≥65 years are Medicare beneficiaries. The Centers for Medicare and Medicaid Services compiles data on medical care received and billed for by healthcare providers. A description of Medicare data sets can be found at www.resdac.umn.edu/Medicare/Data_File_Descriptions_RIF.asp (accessed August 10, 2010).
Institutional Review Board Approval
This study was approved as secondary data analysis by the Institutional Review Boards of the University of Minnesota. Individual patients were not contacted.
The Project for Improvement of Stroke Care Management in Minnesota (PRISMM) was a clinical trial to improve stroke care quality in 19 Minnesota hospitals. The PRISMM patient population is described in Lakshminarayan et al.11 We used the baseline preintervention data compiled by PRISMM for the current Medicare Linkage Study. Subjects included in the PRISMM database were hospitalized patients aged 30 to 84 years, discharged with International Classification of Diseases, 9th Revision, codes 434.x and 436.x, and confirmed as having an acute ischemic stroke by physician review. PRISMM subjects included all patients with ischemic stroke discharged during July to December 2000 from the 19 participating hospitals. For this Medicare Linkage Study, we restricted PRISMM patients to those aged ≥65 years and Minnesota residents at the time of the index stroke event (83% of all ischemic strokes entered into PRISMM). These Medicare-eligible patients from the PRISMM database were identified in the year 2000 Medicare Denominator file through their Medicare Beneficiary ID. The Medicare Denominator file contains demographic and enrollment information about each beneficiary enrolled in Medicare during a calendar year. Patients from the PRISMM database who were identified in the Medicare Denominator file formed the stroke cohort for the Medicare Linkage Study.
A sampling frame was constructed using the Minnesota state segment of the year 2000 Medicare Denominator file. A 5:1 sampling ratio was used with 5 nonstroke members for each member of the stroke cohort. Matching variables included age (within ±2 years), sex, and race. Each sampled member of the nonstroke cohort had to be alive up until the index stroke hospitalization discharge date of their matched stroke cohort member.
We ensured the nonstroke status of nonstroke cohort members by checking each sampled nonstroke member against the year 2000 Medicare Provider Analysis and Review (MedPAR) file to ensure that they did not have an acute stroke hospitalization (International Classification of Diseases, 9th Revision 430 to 437) up until and including the discharge date of the index hospitalization of their matched stroke cohort member. The MedPAR file contains inpatient hospital “stay” records summarizing services rendered to a beneficiary during hospitalization. Each stay record has up to 10 diagnosis codes, which are the International Classification of Diseases, 9th Revision codes identifying the primary disease condition or coexisting conditions present in the medical records.
The occurrence of death and the death date for both cohorts was identified from the Denominator file and confirmed in the MedPAR file. Death during the index hospitalization was also confirmed from the PRISMM database. The unique Medicare Beneficiary ID was used to track individual stroke and nonstroke cohort members across the years through Denominator and MedPAR files from 2000 to 2005 to obtain 5-year mortality.
Acute care rehospitalization events and their dates were identified from the years 2000 to 2005 MedPAR files. Cohort members who were part of a managed care organization (Medicare Advantage) in the year 2000 were not tracked for rehospitalization events because their claims data would not be represented in MedPAR. (Unlike rehospitalizations claims, Medicare versus managed care enrollment did not affect mortality because death dates were recorded in the Denominator file regardless of entitlement status.) Rehospitalization events were ascertained using acute care hospitalizations. Rehabilitation and nursing home transfers were excluded. Outcome events were acute rehospitalizations for recurrent ischemic stroke, intracerebral hemorrhage, cardiac events including heart failure, myocardial infarction and arrhythmias, pneumonia, pulmonary embolism, gastrointestinal hemorrhages, urinary tract infections, and hip fractures. Events were identified using International Classification of Diseases, 9th Revision codes as listed in Supplemental Table 1 (http://stroke.ahajournals.org). Only the primary diagnosis code was used to enhance specificity.
Time to Event and Censoring
Time to event in person-years for mortality was calculated from discharge date of the index hospitalization (Day 0) to event occurrence. The Day 0 for a member of the stroke cohort was also assigned as Day 0 of their matched nonstroke cohort members. Time to event for rehospitalization outcomes was calculated from Day 1. Only the first event in each outcome category was counted for each person. Cohort members were censored when they died, went to managed care (Medicare Advantage), left Medicare enrollment, and when the event of interest occurred. Patients who died during the index stroke hospitalization and those who were in managed care at baseline did not contribute person-years to rehospitalization rates. There were no significant differences in the mortality outcomes of the baseline managed and nonmanaged care enrollees.
Pre-existing comorbidity conditions that could potentially influence outcomes were identified using 1 Part A institutional/inpatient claim or 2 outpatient/Part B physician/supplier claims from a 6-month period before the index hospitalization. The following International Classification of Diseases, 9th Revision codes were used: diabetes mellitus 250, 357.2, 362, 366.41; hypertension 404 to 405, 362.11, 437.2; myocardial infarction 410 to 411; chronic obstructive pulmonary disease 490 to 496; dyslipidemia 272; heart failure 428; and peptic ulcer disease 531 to 535.
Socioeconomic status is a strong determinant of health outcomes. We used the median income of cohort members' zip code (obtained from census record) as a surrogate variable indicating individual socioeconomic status.12,13
Charges to and Payments by Medicare for Hospital Admissions
We calculated the average Medicare-covered charges as well as the payments made by Medicare per person-year of follow-up time using all acute care admissions (not just the first event) for each cohort. Member-time was censored on death, when members left Medicare enrollment, and when they went to managed care.
Baseline characteristics of stroke cohort and nonstroke cohorts when not part of the matching criteria were compared using the χ2 test for categorical variables and the t test or Wilcoxon rank sum test for continuous variables.
Kaplan–Meier survival curves (PROC LIFETEST; SAS) were used to obtain cumulative incidence (%) with 95% confidence limits at 7 days, 30 days, 1 year, and 5 years for the outcome events described earlier in the stroke and nonstroke cohorts. Kaplan–Meier estimates for rehospitalization outcomes were compared with cumulative incidence estimates from survival models that accounted for the competing risk of mortality (PROC LIFETEST, Nelson Aalen estimates) to evaluate for possible bias.
Cox proportional hazards regression models were used to compare adjusted event rates in the 2 cohorts and calculate hazard ratios (HRs) with 95% CIs. The proportional hazards assumption was tested for each outcome and if this assumption was violated, we used the extended Cox regression modeling framework for that outcome.14 All Cox models were adjusted for relevant comorbid conditions and socioeconomic status. Cox models for mortality were also adjusted for Medicare Advantage status. Because age, sex, and race were matching variables, they were not used in further model adjustment.
We calculated the described Cox model HRs comparing the stroke and nonstroke cohorts with 3 starting points for event observation. The first compared 5-year event rates between the 2 cohorts starting on Day 0 for mortality and Day 1 for rehospitalization outcomes. In the second, events were ascertained starting on Day 31. The risk set for this model was all members alive on Day 31. Events between Day 1 and Day 30 were ignored. The outcome clock for everyone in the risk set was restarted and the first event in each outcome category (regardless of prior events) starting Day 31 was counted. Early events could presumably be due to the index hospitalization itself and could suggest opportunities to improve acute stroke care. In the third model, we started event observation at Day 366 (1 year). Again, only the first event in each category starting Day 366 was counted. We examined outcome ascertainment starting on Day 366 to examine if the risk of various outcomes in the stroke cohort returned to nonstroke levels after 1 year.
Figure 1 shows the study design.
Of 879 patients with acute ischemic stroke ≥65 years in the baseline PRISMM database, a total of 823 patients were linked to the year 2000 Minnesota state segment of the Medicare Denominator file for a linkage rate of 94%. Nonlinkage was commonly due to lack of a valid social security number (26 patients [3%]) or non-Minnesota residence (21 patients [2.4%]).
These 823 patients formed the stroke cohort. The nonstroke cohort, formed by the 5:1 sampling described previously, had 4115 patients. These 2 cohorts formed the denominator for tracking 5-year mortality.
Of 823 patients in the stroke cohort, 141 (17%) had Medicare Advantage at baseline. The remaining 682 stroke cohort members were linked to MedPAR and tracked for 5-year rehospitalization events. Of the 4115 nonstroke cohort members, 377 (9%) had Medicare Advantage at baseline. The remaining 3738 nonstroke cohort members were linked to MedPAR and tracked for 5-year rehospitalization events.
Table 1 shows comparisons of baseline (preindex date) comorbid characteristics. These comparisons show significantly higher rates of vascular risk factors in stroke cohort members versus the nonstroke cohort members, including hypertension 65% versus 19%; diabetes mellitus 26% versus 9%; dyslipidemia 24% versus 11%; myocardial infarction 5% versus 1%; and heart failure 9% versus 4%. Chronic obstructive pulmonary disease rates were also different: 14% in the stroke cohort versus 4% in the nonstroke cohort. Rates of gastrointestinal ulcers were not significantly different. The stroke cohort lived in higher median income neighborhoods compared with the nonstroke cohort.
Figure 2, Table 2, and Supplemental Table I show comparative outcomes in the stroke and nonstroke cohorts. Table 3 shows HRs and 95% CIs from the Cox models comparing adjusted event rates in the 2 cohorts. These are discussed subsequently.
In the stroke cohort, 24% of the patients were dead at 1 year and 49% were dead at 5 years. In the nonstroke cohort, 4% were dead at 1 year and 24% were dead at 5 years. These proportions were significantly different from each other. The adjusted HR for mortality from the Cox model when events were counted from Day 0 was 4.4 (95% CI, 3.6 to 5.5). When events were counted from Day 31, the HR was still significant though lower at 2.5 (95% CI, 2.0 to 3.2) and when events were counted from Day 366, the HR was further attenuated but still significant at 1.4 (95% CI, 1.2 to 1.7).
In the stroke cohort, 14% were rehospitalized within 30 days, 49% were rehospitalized at least once at 1 year, and 83% at 5 years. In the nonstroke cohort, 2% were hospitalized at 30 days, 20% at 1 year, and 63% at 5 years. The adjusted HR was 2.6 (95% CI, 2.2 to 3.0) when hospitalization events were counted from Day 1, still significant at 2.0 (95% CI, 1.7 to 2.3) when hospitalization events were counted starting at Day 31, and attenuated but still significant at 1.2 (95% CI, 1.1 to 1.4) when events were counted from Day 366.
Hospitalization for Specific Conditions
Salient findings include significantly higher 5-year hospitalization rates in the stroke cohort compared with the nonstroke cohort for the following conditions: ischemic stroke (18% versus 4%), heart failure (15% versus 7%), cardiac events (23% versus 15%), any vascular event (38% versus 19%), pneumonia (20% versus 10%), and hip fractures (10% versus 5%; Table 2).
Setting aside composite vascular events, recurrent ischemic stroke (2%), dysrhythmias (2%), and pneumonia (1%) were the most common causes of hospital admission in the stroke cohort in the early, 30-day, postdischarge period. Notably, 70% of these early dysrhythmias hospitalizations in the stroke cohort were due to atrial fibrillation or flutter. Atrial arrhythmias were the most frequent cause of dysrhythmia admissions in the nonstroke cohort as well, although overall event rates were much lower. Over 5 years, pneumonia (20%), recurrent ischemic stroke (18%), heart failure (15%), and hip fractures (10%) were the most frequent causes of hospitalization in the stroke cohort (Table 2). In the nonstroke cohort, pneumonia (10%) was the leading cause of hospitalization over the 5-year period, although the rate was half that of the stroke cohort.
Kaplan–Meier estimates of cumulative incidence rates for rehospitalization outcomes, shown in Table 2 and Supplemental Table I, were almost identical to estimates obtained from survival models that adjusted for the competing risk of mortality.
Charges and Payments Made by Medicare
The average acute care hospital admission Medicare-covered charges per person-year of follow-up time were $15 562 for the stroke cohort and $7016 for the nonstroke cohort. The average acute-care related payment made by Medicare per person-year of follow-up time was $5218 for the stroke cohort and $2712 for the nonstroke cohort.
Many patients with acute ischemic stroke who have survived their initial stroke hospitalization are rehospitalized over a 5-year period with recurrent ischemic stroke, pneumonia, cardiac events, and hip fracture. Our results pertain to Minnesota residents but are consistent with those reported from other geographical areas.5,6,10,15 For example, the strikingly high rehospitalization rates in our study are similar to those reported by Bravata et al10 who tracked patients with stroke hospitalized in Connecticut from 1995 to 2000. Feng et al reported on mortality and vascular events in South Carolina and those event rates are similar to those reported here.6 The unique features of our study are the wide range of outcomes as well as the comparison to a matched nonstroke cohort illustrating the additional cost and morbidity entailed due to stroke. The differential higher risks of mortality and hospitalization are seen in patients with stroke even after the tumultuous early period after hospitalization (ie, the first 30 days) and these risks persist long-term, as seen from the significant HRs even when event observation is started 1 year poststroke (Table 3).
The early postdischarge period is of special interest because it could be a marker for in-hospital care quality and the Centers for Medicare & Medicaid Services has targeted 30-day readmissions for reporting in heart failure. The overall 30-day readmission rate for our stroke cohort, 14%, is lower than reported rates among Medicare beneficiaries for other conditions such as heart failure (27%), pneumonia (20%), and all-cause hospitalizations (20%).16 The lower readmission rates for stroke could be due to patients with stroke being discharged to inpatient rehabilitation or nursing homes rather than to a home situation. Regardless, the specific high-frequency causes for 30-day poststroke readmissions, dysrhythmias, and recurrent ischemic stroke present potential intervention targets. For instance, more aggressive Holter monitoring in the peristroke period may prevent admissions for dysrhythmias and postdischarge nurse follow-up may improve medication adherence and reduce recurrent strokes.
The greater burden of vascular risk factors in the stroke cohort, including hypertension, diabetes mellitus, and dyslipidemia (Table 1), explains the poor long-term vascular outcomes. Although our data are not equipped to address the issue of long-term adherence to antithrombotic therapies or blood pressure management regimens in our cohorts, other studies have shown high rates of discontinuation and nonadherence to secondary prevention therapies in patients with stroke and this may contribute to the continued high rates of vascular events after stroke.17
We acknowledge the following weaknesses of our study. Although the nonstroke cohort was constructed by matching to the stroke cohort and eliminating anyone with a prior stroke hospitalization, some of these nonstroke members could have had a prior stroke that was not represented in the Medicare data. Nevertheless, our results show a wide gap between the outcomes of stroke and nonstroke cohorts, suggesting that these 2 cohorts differ from each other substantially. Furthermore, stroke prevalence in the US population is on the order of 2.9% in the general population.1 Hence, any stroke contamination of the nonstroke cohort, if it exists, is likely small. A different drawback is that our outcomes were ascertained from Medicare administrative data and not from clinical records. We argue, however, that outcome tracking such as that presented here actually represents a good use of Medicare data due to their ready availability and the substantially lower costs of using administrative data rather than following the patients in the community for 5 years. The Medicare payments reported by us do not reflect the total expense of stroke because we do not have information on patient or private insurance payments. We have, however, reported total Medicare-covered charges as well as the amounts paid out by Medicare per person-year to convey the implications of poststroke readmissions. We also acknowledge that although our results pertain to the Medicare-aged population, a substantial fraction of ischemic strokes (17% in our database) occur in those <65 years. Uncommon causes of stroke (vasculitis, dissection) predominate in this younger population who may be sicker and the total cost of stroke may be higher in this group.
The results we have presented illustrate an important aspect of the public health burden of stroke, namely the continued economic costs due to recurrent acute care hospitalizations in stroke survivors. There are 2 take-home messages. First, although the differential rate of cause-specific hospitalizations in the stroke cohort is unsurprising and reflects the differential risk factor burden and consequences of stroke-related disability, it does suggest considerable room for improvement in tertiary prevention (ie, prevention of complications) after stroke. Trials focusing on different models of healthcare delivery to stroke survivors in the community and interventions to improve the follow-up care to this population are needed.18 Strategies to reduce morbidity, mortality, and healthcare costs due to stroke have to continue over the long-term, beyond the first year after stroke. Second, current population-based surveillance efforts primarily track stroke incidence and prevalence.1,8,9 Based on our results, we argue that such surveillance should be expanded to include a more comprehensive tracking of stroke outcomes, including poststroke rehospitalizations.
Sources of Funding
This research was supported by 5K23NS051377 to K.L.
K.L. receives significant salary and research support from the National Institutes of Health (K23NS051377).
The online-only Data Supplement is available at http://stroke.ahajournals.org/cgi/content/full/STROKEAHA.110.605600/DC1.
- Received October 12, 2010.
- Revision received December 29, 2010.
- Accepted January 11, 2011.
- © 2011 American Heart Association, Inc.
American Heart Association Statistics Committee and Stroke Statistics Subcommittee. Heart disease and stroke statistics—2010 update. Circulation. 2010; 121: e46–e215.
- Cadilhac DA,
- Carter R,
- Thrift AG,
- Dewey HM
- Luengo-Fernandez R,
- Gray AM,
- Rothwell PM
- Taylor TN,
- Davis PH,
- Torner JC,
- Holmes J,
- Meyer JW,
- Jacobson MF
- Dhamoon MS,
- Tai W,
- Boden-Albala B,
- Rundek T,
- Paik MC,
- Sacco RL,
- Elkind MS
- Feng W,
- Hendry RM,
- Adams RJ
- Lakshminarayan K,
- Anderson DC,
- Jacobs DR Jr.,
- Barber CA,
- Luepker RV
- Shahar E,
- McGovern PG,
- Pankow JS,
- Doliszny KM,
- Smith MA,
- Blackburn H,
- Luepker RV
- Bravata DM,
- Ho SY,
- Meehan TP,
- Brass LM,
- Concato J
- Lakshminarayan K,
- Borbas C,
- McLaughlin B,
- Morris NE,
- Vazquez G,
- Luepker RV,
- Anderson DC
- Thomas AJ,
- Eberly LE,
- Davey Smith G,
- Neaton JD
- Fox J
- Bravata DM,
- Ho SY,
- Brass LM,
- Concato J,
- Scinto J,
- Meehan TP
- Bushnell CD,
- Zimmer LO,
- Pan W,
- Olson DM,
- Zhao X,
- Meteleva T,
- Schwamm L,
- Ovbiagele B,
- Williams L,
- Labresh KA,
- Peterson ED
- Saposnik G,
- Kapral MK