Short-term, intermediate-term, and long-term mortality in patients hospitalized for stroke

2.1. Population studied and data sources 

Patients discharged from VA inpatient facilities with a diagnosis of acute stroke were identified from the VA Patient Treatment File (PTF) for fiscal years 1994 through 1998. These administrative files contain all discharges from VA acute care facilities and include patient-level information on sociodemographic and diagnostic characteristics.

2.2. Identification of Strokes Based on International Classification of Diseases, 9th revision, Clinical Modification (ICD-9-CM) codes 

Acute stroke patients were classified, based on a first- or second-listed diagnosis, as either ischemic stroke (ICD-9-CM codes of 434 or 436) or hemorrhagic stroke (ICD-9-CM codes of 430, 431, or 432). For hemorrhagic stroke, the descriptive classifications for each ICD-9-CM code were as follows: 430—subarachnoid hemorrhage, 431—intracerebral hemorrhage, and 432—other intracerebral hemorrhage. The descriptive classifications for ischemic stroke are listed in Table 1.

Table 1. Sensitivity, specificity, and positive predictive value for diagnostic codes indicative of hospitalization for acute ischemic stroke17;19–21

	ICD-9-CM code and descriptive classification	Sensitivity %	Specificity %	Positive predictive value %
	433 (occlusion of precerebral arteries)	1.9 [20] 1.9 [19] 7.4 [17]	87.8 [20] 33.3 [19] 31.8 [17]	15 [20] 4.2 [19] 6.1 [17]
	434 (occlusion of cerebral arteries)	45.5 [20] 84.0 [19] 79.0 [17]	89.2 [20] 72.5 [19] 91.6 [17]	82.3 [20] 82.4 [19] 85.0 [17]
	436 (acute but ill-defined cerebrovascular disease)	27.9 [20] 14.2 [19] 9.9 [17]	93.5 [20] 94.2 [19] 27.9 [17]	82.7 [20] 78.9 [19] 82.8 [17]
	433 plus 434	47.4 [20] 85.8 [19] 86.4 [17]	77.0 [20] 5.8 [19] 23.4 [17]	69.5 [20] 58.3 [19] 40.3 [17]
	434 plus 436	73.4 [20] 98.1 [19] 88.9 [17]	82.7 [20] 66.7 [19] 90.4 [17]	82.5 [20] 81.9 [19] 84.7 [17]

Our choice of codes for both hemorrhagic and ischemic strokes was based on a review of the literature 16, 17, 18, 19, 20, 21. The classification of hemorrhagic strokes based on codes 430–432 was commonly used within prior work evaluating stroke mortality.

In contrast to those ICD-9-CM codes used to identify hemorrhagic stroke, more controversy has arisen over the choice of ICD-9-CM codes that most accurately capture hospitalization for ischemic stroke 17, 19, 20. Because our focus was that of hospitalization for acute stroke, we initially eliminated the following ischemic cerebrovascular disease codes: transient cerebral ischemia (code 435), other and ill-defined cerebrovascular disease (code 437), and late effects of cerebrovascular disease (code 438). We then examined the accuracy of individual ICD-9-CM codes 433, 434, and 436 for defining ischemic stroke by calculating sensitivity, specificity, and positive predictive value, based on data in prior reports [17].

We initially evaluated the sensitivity, specificity, and positive predictive value of individual ICD-9-CM codes for ischemic stroke (Table 1). Because of low sensitivity of identifying ischemic stroke based only on individual codes, we then explored combining codes to accurately capture those hospitalizations for acute ischemic stroke.

To decide which combination of codes would be most appropriate to obtain a more accurate rate of mortality for patients hospitalized for acute ischemic stroke with few false positives, we calculated the sensitivity, specificity, and positive predictive value of combinations of the remaining ICD-9-CM codes, by using the information available within the literature. Because ICD-9-CM code 434 was the most robust of all three codes, our calculations were based on combining ICD-9-CM codes 434 and 433, or 434 and 436.

2.3. Study population 

This initial hospitalization was termed the index stay. Patients were excluded from the study if they had an admission within the previous 5 years in which any of the 10 listed diagnoses were stroke or hemiplegia (ICD-9 codes 430–438, 342). Patients were also excluded if they were nonveterans. Patients whose initial hospitalization for stroke occurred in an extended care facility or in a non-VA setting paid for by the VA were not included among the cohort.

Vital status was determined using VA's Beneficiary Identification Records Locator Subsystem (BIRLS) file [22]. This file contains the date of death for both in-hospital and out-of hospital deaths of all veterans reported to the Veterans Benefits Administration. Previous studies have shown this file to be over 90% complete 22, 23. The patient's admission date to the hospital for the acute stroke was used as the starting point for counting days until death. Thirty-seven patients whose dates of death were incorrect were excluded from the analysis. Our ascertainment of the cause of death was based on discharge diagnoses and not death certificates.

The patient's age, gender, race, county of residence, and location of hospital care were obtained from their initial stroke hospitalization recorded in the PTF. The patient's county of residence was used to assign each patient to one of the four Census Bureau regions and to one of the nine Census Bureau divisions that comprise the regions [24]. Twenty patients who resided in Canada, Mexico, or the U.S. territories were excluded from analysis.

Coexisting illnesses for each patient were obtained from the PTF using all acute care stays with an admission date within 5 years prior to the initial stroke admission date, including the index stay. For each hospitalization, we examined all listed diagnoses to determine if a patient had any of several coexisting illnesses (Table 2) found in previous work to be important risk factors for stroke and stroke mortality [25].

Table 2. Characteristics of patients having a first stroke, 1994–1998

	Total number	Ischemic stroke 34,866	Hemorrhagic stroke 5,442
	Fiscal year of discharge
	1994	7,753	1,189
	1995	7,638	1,152
	1996	7,431	1,138
	1997	6,220	1,051
	1998	5,824	912
	Age in years (%)
	<45	2.4	6.4
	45–54	10.3	15.2
	55–64	21.9	20.1
	65–74	39.1	32.9
	75 +	26.3	25.3
	Income in dollars (%)
	Unknown	19.5	22.7
	$1–9,999	41.9	41.6
	$10,000–19,999	24.9	22.0
	$20,000+	13.8	13.7
	Sex (%)
	Male	98.3	98.3
	Female	1.7	1.7
	Ethnicity (%)
	Black/Hispanic Black	23.1	25.3
	Non-Hispanic White	69.5	65.0
	Hispanic White	4.5	6.2
	Other/unknown	2.9	3.5
	Marital Status (%)
	Currently married	51.9	47.1
	Not currently married	47.6	52.9
	Coexisting illnesses (%)
	Diabetes mellitus	33.8	20.6
	Hypertension	67.2	60.0
	Coronary artery disease	33.8	22.5
	Chronic obstructive pulmonary disease	19.6	16.9
	Atrial fibrillation / atrial flutter	13.6	10.1
	Other cardiac arrhythmias	9.5	9.0
	Obesity	5.0	3.5
	Heart failure	14.3	10.0
	Renal failure	3.7	3.9
	Liver disease	1.8	5.2
	Deaths within each time period
	0–30 Days	2,876	1,116
	0–90 Days	4,330	1,429
	0–365 Days	6,963	1,822

2.4. Statistical analysis 

Kaplan-Meier analyses were used to obtain the unadjusted 30-day, 90-day, and 1-year mortality rates. These analyses were done by age and race for ischemic and hemorrhagic stroke. Adjusted mortality rates were computed using Cox proportional hazards regression models for ischemic and for hemorrhagic strokes, separately. A stepwise selection procedure was used to enter variables at a P-value ⩽.15 and to retain variables at a P-value ⩽.05.

In addition, we tested whether risk factors differed significantly by time period for each type of stroke (e.g., the effect of hypertension on 30-day mortality following ischemic stroke vs. its effect on 31–90-day and 91–365-day mortality). Survival-dependent Cox regression models were each run with only patients who had survived to the start of each the two time periods—31–90 days and 91–365 days. Variables that were significant in the 30-day mortality model but were not retained in the survival-dependent models or had a P- value of >.05 were considered to vary in their relationship to mortality by time period.

Variables considered for inclusion in the models were age, gender, ethnicity, marital status, income, and coexisting illnesses. Ten of these coexisting illnesses are listed in Table 2. The other four illnesses that were considered included mental illness, atherosclerotic vascular disease, malignant neoplasms, and paralytic syndromes excluding hemiplegia. Age was grouped into five categories (under 45, 45–54, 55–64, 65–74, and 75 and older), with age under 45 as the reference group. Ethnicity was divided into four groups—Black and Hispanic Black; Hispanic White; non-Hispanic White; and other and unknown. Non-Hispanic White was used as the reference group. Two categories were used for marital status—those who were married and all others, which was the reference group. Because many veterans using the VA health care system are low income, income was categorized into four levels (unknown or no income, $1–$9,999, $10,000–$19,999, and $20,000 and higher) with the income of $20,000 or more as the reference group. For each of the coexisting illnesses, each patient was coded as either having the illness or not.

Because changes in the delivery of care may have occurred over the 5 years studied, each model was adjusted for fiscal year. Thus, we created dummy variables representing the patient's year of entry for their initial stroke hospitalization, using 1994 as the reference group.

We also tested whether selected interaction terms were significant. These included race by hypertension and age groups by race for each of the six models (three for ischemic and three for hemorrhagic). All variables were checked for correlation to minimize the effects of collinearity within each multivariate model.

3.1. Stroke mortality 

The unadjusted 30-day mortality rate for ischemic stroke from 1994 to 1998 was 8.2%. (Fig. 1) The unadjusted 30-day mortality rate for hemorrhagic stroke during this same time period was 20.5%. Unadjusted mortality for a first ischemic stroke was highest among those patients who were 75 years of age and older. For hemorrhagic stroke, unadjusted mortality was also highest among patients age 75 years and older.

View Large Image Download to PowerPoint

Fig. 1. Proportion of hemorrhagic and ischemic stroke patients dying over time.

The adjusted mortality rate for ischemic stroke was 7.4% for 30 days, 11.4% for 90 days, and 19.1% for 1 year. For hemorrhagic stroke, the adjusted mortality rate was 18.8% for 30 days, 24.6% for 90 days, and 31.8% for 1 year.

For the specific time periods, the unadjusted and adjusted 31–90-day mortality rates following an ischemic stroke were 4.5% and 4.0%, respectively; the unadjusted and adjusted 91–365-day mortality rates were 8.6 and 7.7%, respectively. For patients who survived 30 days following a hemorrhagic stroke, the 31–90-day unadjusted and adjusted mortality rates were 5.8 and 5.6%, respectively. The 91–365-day unadjusted and adjusted mortality rates following a hemorrhagic stroke were 7.2 and 7.3%.

We report the hazard ratios from the Cox regression models in Table 3, Table 4. The risk factor associated with the highest hazard ratio (HR) for 30-day mortality among patients who experienced an ischemic stroke was advanced age (75 +) [HR 4.47; 95% confidence interval (CI) 3.03, 6.60]. The risk factor associated with the highest statistically significant hazard ratio for 30-day mortality following hemorrhagic stroke was liver disease (HR 2.07; 95% CI 1.67, 2.55).

Table 3. Hazard ratios of mortality in patients hospitalized for an ischemic stroke

	Risk factor	30–day mortality (95% CI)a	31–90 day mortality (95% CI)a	91–365 day mortality (95% CI)a
	Age 45 – 54 years	b	0.60 (0.36, 1.01)	b
	Age 55 – 64 yearsc	1.60 (1.08, 2.38)	b	1.94 (1.27, 2.95)*
	Age 65 – 74 yearsc	2.58 (1.75, 3.82)**	2.12 (1.36, 3.32)**	3.19 (2.11, 4.82)**
	Age ⩾ 75 yearsc	4.47 (3.03, 6.60)**	3.47 (2.22, 5.43)**	5.39 (3.57, 8.16)**
	Black raced	0.81 (0.74, 0.89)**	b	b
	Hispanic raced	0.77 (0.64, 0.94)*	b	0.74 (0.60, 0.91)*
	Currently married	b	0.84 (0.76, 0.94)*	b
	No reported incomee	b	1.41 (1.16, 1.71)**	b
	Chronic heart failure	1.13 (1.02, 1.25)	b	1.14 (1.03, 1.27)*
	Other cardiac arrhythmias	1.14 (1.01, 1.28)	b	§
	Mental illness	b	b	1.15 (1.05, 1.25)*

a All variables with a 95% Confidence Interval were significant at P ⩽ .05. Highly significant variables are indicated by *P⩽ .01 and **P ⩽ .001. b This variable was not significant for increased mortality during this time period. c Reference group is age less than 45 years. d Reference group is non-Hispanic white race. e Reference group is income ⩾ $20,000. Models were adjusted for fiscal years 1995 through 1998 with 1994 as the reference group.

Table 4. Hazard ratios of mortality in patients hospitalized for a hemorrhagic stroke

	Risk factor	30–Day mortality (95% CI)a	31–90 day mortality (95% CI)a	91–365 day mortality (95% CI)a
	Age 65–74 yearsc	b	1.44 (1.08, 1.92)**	1.86 (1.43, 2.42)**
	Age ⩾ 75 yearsc	1.21 (1.06, 1.38)*	2.09 (1.58, 2.78)**	2.30 (1.76, 3.00)**
	Liver disease	2.07 (1.67, 2.55)**	b	b
	Other cardiac arrhythmias	1.89 (1.60, 2.23)**	1.50 (1.06, 2.06)**	b
	Renal failure	1.60 (1.26, 2.04)**	b	b
	Malignant neoplasm	1.39 (1.19, 1.61)**	2.18 (1.68, 2.81)**	2.50 (1.99, 3.13)**
	Chronic obstructive pulmonary disease	1.34 (1.16, 1.55)**	b	1.46 (1.15, 1.85)*
	Atrial fibrillation/ atrial flutter	1.29 (1.08, 1.54)*	b	b
	Chronic heart failure	1.21 (1.01, 1.45)	1.54 (1.13, 2.12)*	1.44 (1.07, 1.96)
	Mental illness	0.79 (0.68, 0.91)**	b	1.48 (1.20, 1.83)**

a All variables were significant at P ⩽ .05. Highly significant variables are indicated by * P⩽ .01 and **P ⩽ .001. b This variable was not significant for increased mortality during this time period. c Reference group is age less than 65 years. Models were adjusted for fiscal years 1995 through 1998, with 1994 as the reference group.

3.2. Intermediate and 1-year mortality in stroke survivors 

To determine the association of each variable with the probability of dying within 31–90 days and 91–365 days, the survival-dependent Cox regression models (31–90 days and 91–365 days) were run after removal of patients who died before each time period Table 3, Table 4. For patients who survived 90 days, a potentially treatable risk factor significantly associated with an increased 91–365-day mortality following an ischemic stroke was chronic heart failure. For patients surviving 90 days following an ischemic stroke, mental illness was associated with increased 91–365-day mortality. Black race and Hispanic ethnicity were each significantly associated with a lower risk for short-term mortality following hospitalization for an ischemic stroke. Similarly, marital status (i.e., currently married) was associated with a lower risk for intermediate-term mortality.

For patients who experienced a hemorrhagic stroke, variables other than age that were significant for 31–90-day and 91–365-day mortality included malignant neoplasm and chronic heart failure (Table 4).

There were no significant interaction terms or evidence of collinearity among risk factors for either stroke type within any of the three time periods.