Validity of Self-Reported Stroke
The Tromsø Study
Background and Purpose—The aim of this study was to validate the diagnosis of self-reported stroke.
Methods—During 1994–1995, 27 159 people attended a population health survey in the community of Tromsø, Norway, a response rate of 77%. A total of 418 attenders reported a history of stroke. In 1997, all individuals with a self-reported stroke who were still living in the community (n=362) were invited to a clinical reexamination. For each of the 269 people who were reexamined, a person who reported no history of stroke was selected and was reexamined in the same way.
Results—On the basis of the reexamination, 213 (79.2%) of the self-reported strokes were confirmed. Thirteen individuals (4.8%) had a possible stroke. The remaining 43 individuals had either transient ischemic attack (TIA; n=18), traumatic head injuries (n=16), or perinatal cerebral damage, complicated migraine, syncope, possible TIA, or cerebral aneurysm without bleeding (n=9). Among the confirmed strokes, 30 (14.1%) were hemorrhagic and 118 (55.4%) were thromboembolic. Of the 30 hemorrhages, 16 were subarachnoidal bleedings, 10 due to ruptured aneurysms. The histories of stroke, including both the symptoms and the signs, often had a paucity of details and precision, making it impossible to classify 65 stroke victims (30.5%) into stroke subtypes. The positive predictive value (PPV) of a self-reported stroke was 0.79. The PPV was significantly (P=0.016) greater in men (0.88) than in women (0.73). Individuals older than 60 years had a significantly greater PPV than those younger than 60 years (PPV 0.83 and 0.73, respectively; P=0.05). Hypertension was associated with a greater PPV, whereas a history of either ischemic heart disease, diabetes mellitus, lung disease, or depression had no impact on the PPV. The estimated sensitivity of self-reported stroke in the survey population was ≈80% and the specificity was 99%.
Conclusions—We conclude that a self-administered questionnaire can be used to assess the prevalence of stroke in epidemiological research.
Self-administered questionnaires are often used in population surveys to obtain information about a previous history of stroke. However, information on the validity of a self-reported stroke is varied.1 2 3 4 5 In general, self-reports on medical conditions that are well defined and relatively easy to diagnose often have a high positive predictive value (PPV), in contrast to conditions characterized by complex symptoms.6 Stroke is often associated with sensory and memory problems and sometimes additional speech and communicative disorders. These are likely to increase the problems of verifying self-reported stroke. The importance of being aware of these difficulties, particularly in epidemiological studies among the elderly, has been emphasized.7 However, it is not known whether aging itself has an effect on the accuracy of self-reported diseases.8 9 Also, the influence of sex, personality traits, or specific medical conditions on the validity of self-reported stroke has not been clarified.
The purpose of this article was to examine the validity of self-reported stroke by calculating positive predictive value (PPV), sensitivity, and specificity. We also studied whether the validity of self-reported stroke differed depending on the specific characteristics of the participants.
Subjects and Methods
The Tromsø Study, a population-based study of inhabitants in the municipality of Tromsø, Norway, was begun in 1974. The study design includes repeated population health surveys to which total birth cohorts and random samples are invited. The fourth survey of the Tromsø population began in September 1994 and was completed in June 1995. The study was conducted by the University of Tromsø in cooperation with the National Health Screening Service and the Community Health Services in Tromsø. All inhabitants older than 24 years were invited, and 27 159 subjects, or 77% of the eligible population, attended. A protocol similar to previous studies in this population was followed.10 The letter of invitation contained questions on medical history, including the following questions: “Do you have or have you ever had a stroke/cerebral hemorrhage?” “Have you felt happy or optimistic during the last 2 weeks?” “Have you felt down or depressed during the last 2 weeks?” Additional questions were asked about use of antihypertensive drugs, smoking habits, physical activity, and length of education. The examination included standardized measurements of height and weight. Body mass index was calculated as weight in kilograms divided by the square of height in meters. Total cholesterol level was determined by an enzymatic colorimetric method with a commercial kit (CHOD-PAP, Boehringer Mannheim). Blood pressure was measured by specially trained assistants using an automatic device (Dinamap Vital Signs Monitor). After the participants had been seated for 2 minutes, 3 recordings were made at 2-minute intervals. The average of the second and third measurements was used. In this report, subjects with systolic blood pressure >159 mm Hg or diastolic blood pressure >89 mm Hg or those being treated with antihypertensive drugs were defined as hypertensive.
A total of 418 participants, 190 women and 228 men, answered “yes” to the question about a history of previous stroke. During the subsequent 3 years, 51 (12.2%) of the 418 participants with a self-reported history of stroke died, and 5 moved from the community. The remaining 362 persons were invited to a clinical reexamination, which began in August 1997 (Figure⇓). A control subject who did not report a history of stroke was randomly drawn from the study population and matched to each of the 362 stroke subjects, with age (±2 years) and sex used as matching criteria. After 1 written reminder was sent, 269 individuals with a history of self-reported stroke and 262 control subjects underwent identical reexaminations. Survey data from the screening in 1994–1995 were compared among subjects who died during the 3 years between survey and reexamination and those who did and did not attend the reexamination. The regional ethical committee approved the study.
Validation of Self-Reported Stroke
A specialist in internal and geriatric medicine at Tromsø University hospital (T.E.) validated the medical history of stroke of all participants. The diagnosis of stroke was based on a semistructured interview of the participants, a clinical examination, and a critical evaluation of all available information from the hospital medical records. All hospital records but 1 were retrieved. Special attention was paid to any symptoms or signs that had been present initially but had later resolved, as well as the duration of such symptoms. During the interview, all participants were encouraged to comment on or extend the original report in the hospital records. This often resulted in a more detailed history of stroke. Paresis in the extremities was tested with a modified Scandinavian Stroke Scale11 by selecting that part dealing with muscle strength in the upper and lower extremities. The stroke diagnosis was supported by anatomic cerebral changes on cerebral CT scans. If the diagnosis was uncertain and if cerebral CT scans were inconclusive or lacking, then a new CT scan was obtained. A new CT scan was also performed if the time interval between the clinical event and actual CT scan was <3 weeks. Cerebral CTs were taken in a clinical setting and assessed by radiologists in a routine manner. Hemorrhages were identified as areas of hyperdensity on CT scans. Ischemic strokes were identified as areas of hypodensity, with or without mass effect, when the shape of the lesion indicated a vascular origin.
The diagnosis was considered definite when the history of stroke fulfilled the diagnostic criteria of stroke. Cerebral lesions verified by neuroimaging, matching symptoms and clinical signs with regard to localization and time of onset of symptoms, also made the stroke diagnosis definite, even if the history was incomplete. The diagnosis was considered possible if the history of stroke did not fulfill all the diagnostic criteria and auxiliary examination or findings were lacking. A medical doctor with clinical and scientific practice from a stroke unit evaluated the hospital records of those patients with possible stroke and confirmed the classification. Those individuals with a history of self-reported stroke and their controls were classified, based on all available data, as having had a stroke or a possible stroke or not having had a stroke.
Stroke was defined according to World Health Organization criteria as a clinical syndrome of rapidly developing symptoms or signs of focal or global disturbance of cerebral function, with no apparent cause other than that of vascular origin, and that lasted >24 hours unless interrupted by surgery or death.12 This definition includes patients with ischemic brain infarction and intracerebral and subarachnoidal hemorrhage.
Differences between groups were tested by ANOVA, Student’s t test, χ2 test, or Fisher exact test. Values for continuous variables were age adjusted by ANOVA, and proportions were age adjusted by the Mantel-Haenszel method. The PPV was calculated by dividing the number of confirmed strokes by the total number of self-reported strokes. The significance of differences in PPV between groups was calculated by use of χ2. The variables that were significantly associated with the PPV in univariate analysis were included in a multivariate logistic regression model to assess independent relationships. To estimate the sensitivity and specificity of self-reported stroke in the total survey population, we used the prevalence of unreported stroke (false-negative rate) among the control subjects and extrapolated this to the survey population. The confidence interval for the proportion of unreported strokes was estimated by use of the Poisson distribution. All calculations were performed with SAS software.13 Two-sided probability values <0.05 were considered statistically significant.
Classification of Stroke Survivors and Control Subjects at the Reexamination
Among the 269 subjects with a self-reported history of stroke, diagnosis was confirmed at reexamination in 213 subjects (79.2%; 91 women and 122 men; Table 1⇓). Thirteen subjects had suffered a possible stroke. The proportion of stroke among those who were reexamined was significantly higher (P=0.02) among men than among women (Table 1⇓). A total of 43 individuals had not had a stroke but instead had either a transient ischemic attack (TIA; n=18), traumatic head injury (n=16); or perinatal cerebral damage, complicated migraine, syncope, possible TIA, or cerebral aneurysms without bleeding (n=7). Two subjects reported a previous history of stroke without any suspicious symptoms. Men who had suffered a stroke were older than men with a possible stroke or with no history of stroke (Table 2⇓). The cholesterol level among women who had a stroke or a possible stroke was higher than among women with no stroke (P=0.02). Otherwise there were no statistically significant differences either between the groups or between the sexes (Table 2⇓).
Of the 361 invited control subjects, 262 attended the reexamination (72.6%; Figure⇑). Eight of these subjects were found to have a history of stroke at reexamination. However, only 2 had had their stroke before the 1994–1995 survey and had not reported it on the questionnaire (Table 3⇓). Of the 262 attending control subjects, a total of 243 had never experienced symptoms or signs indicative of stroke or TIA.
Subtypes and Characteristics of Stroke Survivors
A total of 244 (91%) of the 269 individuals with self-reported stroke were scanned with cerebral CT. CT scanning was done in 194 of 213 subjects with definite stroke, 12 of 13 subjects with possible stroke, and 38 of 43 subjects with no stroke. Among the 194 subjects who were scanned in the stroke group, 118 (60.8%) had ischemic stroke and 30 (15.5%) had hemorrhagic stroke. Stroke subtype was impossible to determine in the remaining 46 cases (23.7%) based on cerebral CT findings. Among the 46 undetermined cases, the CT scan was normal in 39 cases, whereas 7 had nonspecific CT findings. Subjects with ischemic stroke or stroke due to intracerebral bleedings were older at the screening and at the onset of stroke than those with subarachnoidal bleedings (data not shown). When adjusted for age at the screening, there were no statistically significant differences between stroke subtypes regarding systolic or diastolic blood pressure, use of antihypertensive drugs, comorbidity, body mass index, or daily smoking (data not shown).
Validity of Self-Reported Stroke
In a pooled analysis of both sexes, the PPV of a self-reported stroke was 0.79. Table 4⇓ presents factors influencing the PPV. The factors that were significantly associated with a high PPV in univariate analysis were sex (P=0.016), age over 60 years (P=0.05), hypertension (P=0.001), and a confirming answer to the question whether the subjects had felt happy or optimistic during the last 14 days (P=0.04). The difference between older and younger participants was statistically significant only for men (P=0.02) (data not shown). Length of education and a history of either ischemic heart disease, diabetes mellitus, or depression had no statistically significant influence on the PPV. Of the factors that were significantly associated with a high PPV in the univariate analysis, only sex (P=0.015) and hypertension (P=0.013) turned out to be independent significant predictors of PPV when included in a multivariate logistic regression model (data not shown).
Sensitivity and Specificity
When 2 control subjects who did not report a history of stroke at the screening in 1994–1995 were taken into account (Figure⇑), the estimated number of unreported strokes in the survey population ranged from 39 (95% CI 0 to 191) to 92 (95% CI 0 to 389) depending on what model was chosen (Table 5⇓). The corresponding sensitivity estimates of self-reported stroke for the total survey population ranged from 84.5% to 69.8% (Table 5⇓), whereas specificity remained at 99.7% in all models.
Validity of Self-Reported Stroke
We found that a self-reported history of stroke had a PPV of 0.79. This value is probably a low estimate considering that some of the 13 possible strokes could be definite stroke. However, even a PPV of 0.79 is higher than reported in other studies that used similar questions about stroke in self-administered questionnaires.4 8 14 15 In the Copenhagen stroke study,3 the true PPV of self-reported stroke via questionnaire was 0.50. An Italian longitudinal study on aging,15 the American prospective cohort study of nurses,6 the American National Health Survey,8 and a population-based study from Rotterdam14 all found similar positive rates of self-reported strokes (PPV 0.66 and 0.67). Unlike the present study, none of these studies (with the exception of the one from Rotterdam14 ) used cerebral CT to assess uncertain medical histories. The failure to separate TIA from stroke is a common contributor to a false-positive rate of self-reported stroke and was shown in the present study to be the most important factor other than traumatic head injury. False-positive rates of stroke ranging from 5% to 15% have even been reported from specialized stroke units.16 In contrast, the labeling of minor stroke as TIA increases the false-negative rate. Although stroke is mainly a clinical diagnosis, neuroanatomic data facilitate the validation of stroke, particularly in those cases characterized by ambiguous symptoms and signs. In the Rotterdam study, 10.2% of self-reported strokes turned out to be TIAs, compared with 6.7% in the present study. TIA was mistaken for stroke in 14.4% of self-reported strokes in a cross-sectional study from Newcastle-on-Tyne4 and in 21% of self-reported strokes in the American nurse study.6
Different prevalence rates of stroke in the study populations provide one explanation for the various false-positive rates among studies. The posttest probability (PPV) of a disease normally rises with increasing prevalence (pretest probability), whereas the concomitant negative predictive value or the posttest probability of not having a disease falls with increasing prevalence of the disease.17 The PPV is simply the rate of the disease among the participants with positive test results. The prevalence of stroke increases almost exponentially with age after the age of 60 years. This should be taken into consideration when PPVs of stroke from survey data are compared. The participants in the American nurse study were 30 to 55 years old, whereas the mean age of participants in the present study was 63 years. On the other hand, the age-specific prevalence of self-reported stroke among participants in the Rotterdam study was similar to what we found in the present study, and the percentages of TIA did not differ much.
Little is known about factors that may influence the predictive value of self-reported stroke. In the present study, we identified several factors that were associated with the PPV. The PPV value for men was significantly higher than for women, mainly because of the greater PPV in men older than 60 years. This finding is in agreement with a national study from the United States that assessed the validity of self-reported diagnosis.8 However, it is in contrast to what was found in a study from Amsterdam,1 owing to more overreporting of stroke by men. In the present study, the greater PPV in men may be due in part to a higher prevalence17 of self-reported stroke among men than among women (11.2/1000 and 8.8/1000, respectively; P=0.04). The greater PPV among men than among women persisted even after the group of possible strokes (10 women and 3 men) was included in the group of confirmed strokes (0.87 and 0.81, respectively), although the difference was no longer statistically significant (P=0.18).
The positive association between the PPV and hypertension was statistically significant for both sexes. It is likely that hypertensive subjects had previous contacts with health services or physicians at which hypertension as a risk factor for stroke was discussed, thus making this group more aware of symptoms and signs of stroke. The inverse relation between the PPV and a feeling of happiness in men was not statistically significant when other variables in the multivariate analysis were controlled for in the analysis. Depressive symptoms, anxiety, or educational level did not influence the accuracy of self-reported stroke in the present study, which is in contrast to what has been reported previously from other studies.1 9 18
The proportion of false-negative stroke among the controls in the present study (2/262) was similar to the findings among controls (3/220; mean age 68 years) in a study of carotid stenosis in the same survey population.18A The false-negative rate is also in accordance with that from a population-based study from Newcastle-on-Tyne (6/1338).4
When estimating the sensitivity of self-reported stroke in a survey population based on a smaller sample, one has to make approximations. Only 269 of the 418 people who reported a stroke at the survey were validated, excluding 149 individuals from the estimates of sensitivity. However, these estimates are not likely to be changed considerably either by the dropouts (n=98) or by the individuals who died (n=51) before the reexamination. The dropouts were similar to the attenders with respect to age, blood pressure, body mass index, and smoking habits (data not shown), whereas those who died before the reexamination were older than the attenders. The mean age of those who were reexamined was considerably higher (62.9 years) than that of the total survey population (47.1 years). The prevalence of stroke, both reported and unreported, will therefore be different in the 2 groups. It is likely that unreported strokes in the total survey population occurred mainly among elderly participants,5 19 who are also the victims of most strokes. Approximately 85% of all strokes occur in individuals older than 60 years,20 21 and 75% occur in persons >65 years of age.22 Memory problems also contribute to a higher false-negative rate of self-reported stroke in the elderly compared with younger persons. The elderly are probably more apt to experience a minor stroke as less dramatic and threatening than young people. This would reinforce recall problems in the elderly when asked about prior events. In the present study, the 2 participants with unreported strokes were 72 and 89 years old, respectively. On the bases of these considerations, a model that limits the unreported strokes to those >60 or even >70 years of age is probably more likely to be “true” than if one expects the number of unreported strokes to be distributed evenly in all age groups (Table 5⇑). The sensitivity in the present study is probably closer to 80% than 70%, corresponding to alternative 2 or alternative 3 in Table 5⇑. The specificity will be high (99.7%) regardless of the age group to which the false-negative strokes belong (Table 5⇑).
In conclusion, this population-based study shows that a self-reported history of stroke has a PPV of 0.79, with male sex and hypertension predicting greater PPV. The estimated sensitivity was ≈80% and the specificity was 99%. We conclude that questionnaires can be used to assess a history of stroke in epidemiological research.
This study was supported by the research program “Research on the Elderly in Tromsø” and financed by The Ministry of Health and Social affairs and the Norwegian Research Council.
- Received February 25, 2000.
- Revision received April 21, 2000.
- Accepted April 28, 2000.
- Copyright © 2000 by American Heart Association
Tretli S, Lund-Larsen PG, Foss OP. Reliability of questionnaire information on cardiovascular disease and diabetes: cardiovascular disease study in Finnmark county. J Epidemiol Commun Health. 1982;36:269–273.
O’Mahony PG, Dobson R, Rodgers H, James OFW, Thomson MD. Validation of a population screening questionnaire to assess prevalence of stroke. Stroke. 1995;26:1334–1337.
Mittelmark MB, Psaty BM, Rautaharju PM, Fried LP, Borhani NO, Tracy RP, Gardin JM, O’Leary DH. Prevalence of cardiovascular diseases among older adults: the Cardiovascular Health Study. Am J Epidemiol. 1991;137:311–317.
Colditz GA, Martin P, Stampfer MJ, Willett WC, Sampson L, Rosner B, Hennekens CH, Speizer FE. Validation of questionnaire information on risk factors and disease outcomes in a prospective cohort study on women. Am J Epidemiol. 1986;123:894–900.
Kelsey JL, O’Brian LA, Grisso JA, Hoffman S. Issues in carrying out epidemiologic research in the elderly. Am J Epidemiol. 1989;130:857–866.
Bergmann MM, Byers T, Freedman D, Mokdad A. Validity of self-reported diagnoses leading to hospitalization: a comparison of self-reports with hospital records in a prospective study of American adults. Am J Epidemiol. 1998;147:969–977.
Bønaa KH, Thelle DS. Association between blood pressure and serum lipids in a population. Circulation. 1991;83:1305–1314.
SAS/STAT User’s Guide, Version 6. 4th ed. Cary, NC: SAS Institute Inc, 1990.
Bots ML, Looman SJ, Koudstaal PJ, Hofman A, Hoes AW, Grobbe DE. Prevalence of stroke in the general population: the Rotterdam study. Stroke. 1996;27:1499–1501.
Di Carlo A, Candelise L, Gandolfo C, Grigoletto F, Volonnino G, Baldereschi M, Maggi S, Scarlato G, Amaducci L, Inzitari D. Influence of different screening procedures on the stroke prevalence estimates: the Italian Longitudinal Study on Aging. Cerebrovasc Dis. 1999;4:231–237.
The Members of the Lille Stroke Program. Misdiagnosis in 1250 consecutive patients admitted to an acute stroke unit. Cerebrovasc Dis. 2000;7:284–288.
Sackett DL, Haynes RB, Guyatt GH, Tugwell P. Clinical Epidemiology. Boston, Mass: Little Brown & Co; 1991:69–152.
Rodgers WL, Herzog AR. Interviewing older adults: the accuracy of factual information. J Gerontol. 1987;42:387–394.
Joakimsen O. Thesis: Skriftserie No. 52. Tromsø, Norway: University of Tromsø; 2000.
Wyller TB, Bautz-Holter E, Holmen J. Prevalence of stroke and stroke-related disability in North-Trøndelag County, Norway. Cerebrovasc Dis. 1994;4:421–427.
Reddy MP, Reddy V. After a stroke: strategies to restore function and prevent complications. Geriatrics. 2000;52:59–75.